diff --git "a/SplitEinsum-Resources/Unet.mlmodelc/model.mil" "b/SplitEinsum-Resources/Unet.mlmodelc/model.mil" new file mode 100644--- /dev/null +++ "b/SplitEinsum-Resources/Unet.mlmodelc/model.mil" @@ -0,0 +1,57934 @@ +program(1.0) +[buildInfo = dict, tensor>({{"coremlc-component-MIL", "3304.5.2"}, {"coremlc-version", "3304.6.2"}, {"coremltools-component-torch", "2.2.0"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "8.0b1"}})] +{ + func main(tensor encoder_hidden_states, tensor sample, tensor text_embeds, tensor time_ids, tensor timestep) { + tensor var_24 = const()[name = tensor("op_24"), val = tensor(-1)]; + tensor var_41_axes_0 = const()[name = tensor("op_41_axes_0"), val = tensor([1])]; + tensor var_41_cast_fp16 = expand_dims(axes = var_41_axes_0, x = timestep)[name = tensor("op_41_cast_fp16")]; + tensor var_43_to_fp16 = const()[name = tensor("op_43_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(64)))]; + tensor emb_3_cast_fp16 = mul(x = var_41_cast_fp16, y = var_43_to_fp16)[name = tensor("emb_3_cast_fp16")]; + tensor var_48_cast_fp16 = sin(x = emb_3_cast_fp16)[name = tensor("op_48_cast_fp16")]; + tensor var_49_cast_fp16 = cos(x = emb_3_cast_fp16)[name = tensor("op_49_cast_fp16")]; + tensor emb_7_interleave_0 = const()[name = tensor("emb_7_interleave_0"), val = tensor(false)]; + tensor emb_7_cast_fp16 = concat(axis = var_24, interleave = emb_7_interleave_0, values = (var_48_cast_fp16, var_49_cast_fp16))[name = tensor("emb_7_cast_fp16")]; + tensor var_53_begin_0 = const()[name = tensor("op_53_begin_0"), val = tensor([0, 160])]; + tensor var_53_end_0 = const()[name = tensor("op_53_end_0"), val = tensor([2, 320])]; + tensor var_53_end_mask_0 = const()[name = tensor("op_53_end_mask_0"), val = tensor([true, true])]; + tensor var_53_cast_fp16 = slice_by_index(begin = var_53_begin_0, end = var_53_end_0, end_mask = var_53_end_mask_0, x = emb_7_cast_fp16)[name = tensor("op_53_cast_fp16")]; + tensor var_55_begin_0 = const()[name = tensor("op_55_begin_0"), val = tensor([0, 0])]; + tensor var_55_end_0 = const()[name = tensor("op_55_end_0"), val = tensor([2, 160])]; + tensor var_55_end_mask_0 = const()[name = tensor("op_55_end_mask_0"), val = tensor([true, false])]; + tensor var_55_cast_fp16 = slice_by_index(begin = var_55_begin_0, end = var_55_end_0, end_mask = var_55_end_mask_0, x = emb_7_cast_fp16)[name = tensor("op_55_cast_fp16")]; + tensor sample_3_interleave_0 = const()[name = tensor("sample_3_interleave_0"), val = tensor(false)]; + tensor sample_3_cast_fp16 = concat(axis = var_24, interleave = sample_3_interleave_0, values = (var_53_cast_fp16, var_55_cast_fp16))[name = tensor("sample_3_cast_fp16")]; + tensor var_58 = const()[name = tensor("op_58"), val = tensor(1)]; + tensor var_65_axes_0 = const()[name = tensor("op_65_axes_0"), val = tensor([-1])]; + tensor var_65_cast_fp16 = expand_dims(axes = var_65_axes_0, x = sample_3_cast_fp16)[name = tensor("op_65_cast_fp16")]; + tensor input_1_axes_0 = const()[name = tensor("input_1_axes_0"), val = tensor([-1])]; + tensor input_1_cast_fp16 = expand_dims(axes = input_1_axes_0, x = var_65_cast_fp16)[name = tensor("input_1_cast_fp16")]; + tensor var_69 = const()[name = tensor("op_69"), val = tensor([1, 1])]; + tensor var_71 = const()[name = tensor("op_71"), val = tensor([1, 1])]; + tensor input_3_pad_type_0 = const()[name = tensor("input_3_pad_type_0"), val = tensor("custom")]; + tensor input_3_pad_0 = const()[name = tensor("input_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor time_embedding_linear_1_weight_to_fp16 = const()[name = tensor("time_embedding_linear_1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(448)))]; + tensor time_embedding_linear_1_bias_to_fp16 = const()[name = tensor("time_embedding_linear_1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(819712)))]; + tensor input_3_cast_fp16 = conv(bias = time_embedding_linear_1_bias_to_fp16, dilations = var_71, groups = var_58, pad = input_3_pad_0, pad_type = input_3_pad_type_0, strides = var_69, weight = time_embedding_linear_1_weight_to_fp16, x = input_1_cast_fp16)[name = tensor("input_3_cast_fp16")]; + tensor input_5_cast_fp16 = silu(x = input_3_cast_fp16)[name = tensor("input_5_cast_fp16")]; + tensor var_77 = const()[name = tensor("op_77"), val = tensor([1, 1])]; + tensor var_79 = const()[name = tensor("op_79"), val = tensor([1, 1])]; + tensor emb_pad_type_0 = const()[name = tensor("emb_pad_type_0"), val = tensor("custom")]; + tensor emb_pad_0 = const()[name = tensor("emb_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor time_embedding_linear_2_weight_to_fp16 = const()[name = tensor("time_embedding_linear_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(822336)))]; + tensor time_embedding_linear_2_bias_to_fp16 = const()[name = tensor("time_embedding_linear_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4099200)))]; + tensor emb_cast_fp16 = conv(bias = time_embedding_linear_2_bias_to_fp16, dilations = var_79, groups = var_58, pad = emb_pad_0, pad_type = emb_pad_type_0, strides = var_77, weight = time_embedding_linear_2_weight_to_fp16, x = input_5_cast_fp16)[name = tensor("emb_cast_fp16")]; + tensor concat_0 = const()[name = tensor("concat_0"), val = tensor([12])]; + tensor timesteps_cast_fp16 = reshape(shape = concat_0, x = time_ids)[name = tensor("timesteps_cast_fp16")]; + tensor var_85 = const()[name = tensor("op_85"), val = tensor(-1)]; + tensor var_102_axes_0 = const()[name = tensor("op_102_axes_0"), val = tensor([1])]; + tensor var_102_cast_fp16 = expand_dims(axes = var_102_axes_0, x = timesteps_cast_fp16)[name = tensor("op_102_cast_fp16")]; + tensor var_104_to_fp16 = const()[name = tensor("op_104_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4101824)))]; + tensor emb_11_cast_fp16 = mul(x = var_102_cast_fp16, y = var_104_to_fp16)[name = tensor("emb_11_cast_fp16")]; + tensor var_109_cast_fp16 = sin(x = emb_11_cast_fp16)[name = tensor("op_109_cast_fp16")]; + tensor var_110_cast_fp16 = cos(x = emb_11_cast_fp16)[name = tensor("op_110_cast_fp16")]; + tensor emb_15_interleave_0 = const()[name = tensor("emb_15_interleave_0"), val = tensor(false)]; + tensor emb_15_cast_fp16 = concat(axis = var_85, interleave = emb_15_interleave_0, values = (var_109_cast_fp16, var_110_cast_fp16))[name = tensor("emb_15_cast_fp16")]; + tensor var_114_begin_0 = const()[name = tensor("op_114_begin_0"), val = tensor([0, 128])]; + tensor var_114_end_0 = const()[name = tensor("op_114_end_0"), val = tensor([12, 256])]; + tensor var_114_end_mask_0 = const()[name = tensor("op_114_end_mask_0"), val = tensor([true, true])]; + tensor var_114_cast_fp16 = slice_by_index(begin = var_114_begin_0, end = var_114_end_0, end_mask = var_114_end_mask_0, x = emb_15_cast_fp16)[name = tensor("op_114_cast_fp16")]; + tensor var_116_begin_0 = const()[name = tensor("op_116_begin_0"), val = tensor([0, 0])]; + tensor var_116_end_0 = const()[name = tensor("op_116_end_0"), val = tensor([12, 128])]; + tensor var_116_end_mask_0 = const()[name = tensor("op_116_end_mask_0"), val = tensor([true, false])]; + tensor var_116_cast_fp16 = slice_by_index(begin = var_116_begin_0, end = var_116_end_0, end_mask = var_116_end_mask_0, x = emb_15_cast_fp16)[name = tensor("op_116_cast_fp16")]; + tensor time_embeds_1_interleave_0 = const()[name = tensor("time_embeds_1_interleave_0"), val = tensor(false)]; + tensor time_embeds_1_cast_fp16 = concat(axis = var_85, interleave = time_embeds_1_interleave_0, values = (var_114_cast_fp16, var_116_cast_fp16))[name = tensor("time_embeds_1_cast_fp16")]; + tensor var_124 = const()[name = tensor("op_124"), val = tensor([2, -1])]; + tensor time_embeds_cast_fp16 = reshape(shape = var_124, x = time_embeds_1_cast_fp16)[name = tensor("time_embeds_cast_fp16")]; + tensor var_127 = const()[name = tensor("op_127"), val = tensor(-1)]; + tensor sample_interleave_0 = const()[name = tensor("sample_interleave_0"), val = tensor(false)]; + tensor sample_cast_fp16 = concat(axis = var_127, interleave = sample_interleave_0, values = (text_embeds, time_embeds_cast_fp16))[name = tensor("sample_cast_fp16")]; + tensor var_129 = const()[name = tensor("op_129"), val = tensor(1)]; + tensor var_136_axes_0 = const()[name = tensor("op_136_axes_0"), val = tensor([-1])]; + tensor var_136_cast_fp16 = expand_dims(axes = var_136_axes_0, x = sample_cast_fp16)[name = tensor("op_136_cast_fp16")]; + tensor input_7_axes_0 = const()[name = tensor("input_7_axes_0"), val = tensor([-1])]; + tensor input_7_cast_fp16 = expand_dims(axes = input_7_axes_0, x = var_136_cast_fp16)[name = tensor("input_7_cast_fp16")]; + tensor var_140 = const()[name = tensor("op_140"), val = tensor([1, 1])]; + tensor var_142 = const()[name = tensor("op_142"), val = tensor([1, 1])]; + tensor input_9_pad_type_0 = const()[name = tensor("input_9_pad_type_0"), val = tensor("custom")]; + tensor input_9_pad_0 = const()[name = tensor("input_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor add_embedding_linear_1_weight_to_fp16 = const()[name = tensor("add_embedding_linear_1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4102144)))]; + tensor add_embedding_linear_1_bias_to_fp16 = const()[name = tensor("add_embedding_linear_1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11311168)))]; + tensor input_9_cast_fp16 = conv(bias = add_embedding_linear_1_bias_to_fp16, dilations = var_142, groups = var_129, pad = input_9_pad_0, pad_type = input_9_pad_type_0, strides = var_140, weight = add_embedding_linear_1_weight_to_fp16, x = input_7_cast_fp16)[name = tensor("input_9_cast_fp16")]; + tensor input_11_cast_fp16 = silu(x = input_9_cast_fp16)[name = tensor("input_11_cast_fp16")]; + tensor var_148 = const()[name = tensor("op_148"), val = tensor([1, 1])]; + tensor var_150 = const()[name = tensor("op_150"), val = tensor([1, 1])]; + tensor aug_emb_pad_type_0 = const()[name = tensor("aug_emb_pad_type_0"), val = tensor("custom")]; + tensor aug_emb_pad_0 = const()[name = tensor("aug_emb_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor add_embedding_linear_2_weight_to_fp16 = const()[name = tensor("add_embedding_linear_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11313792)))]; + tensor add_embedding_linear_2_bias_to_fp16 = const()[name = tensor("add_embedding_linear_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14590656)))]; + tensor aug_emb_cast_fp16 = conv(bias = add_embedding_linear_2_bias_to_fp16, dilations = var_150, groups = var_129, pad = aug_emb_pad_0, pad_type = aug_emb_pad_type_0, strides = var_148, weight = add_embedding_linear_2_weight_to_fp16, x = input_11_cast_fp16)[name = tensor("aug_emb_cast_fp16")]; + tensor input_19_cast_fp16 = add(x = emb_cast_fp16, y = aug_emb_cast_fp16)[name = tensor("input_19_cast_fp16")]; + tensor var_158 = const()[name = tensor("op_158"), val = tensor(1)]; + tensor var_161 = const()[name = tensor("op_161"), val = tensor([1, 1])]; + tensor var_163 = const()[name = tensor("op_163"), val = tensor([1, 1])]; + tensor input_13_pad_type_0 = const()[name = tensor("input_13_pad_type_0"), val = tensor("custom")]; + tensor input_13_pad_0 = const()[name = tensor("input_13_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor conv_in_weight_to_fp16 = const()[name = tensor("conv_in_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14593280)))]; + tensor conv_in_bias_to_fp16 = const()[name = tensor("conv_in_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14616384)))]; + tensor input_13_cast_fp16 = conv(bias = conv_in_bias_to_fp16, dilations = var_163, groups = var_158, pad = input_13_pad_0, pad_type = input_13_pad_type_0, strides = var_161, weight = conv_in_weight_to_fp16, x = sample)[name = tensor("input_13_cast_fp16")]; + tensor var_172 = const()[name = tensor("op_172"), val = tensor(1)]; + tensor reshape_0_shape_0 = const()[name = tensor("reshape_0_shape_0"), val = tensor([2, 32, 10, 128, 128])]; + tensor reshape_0_cast_fp16 = reshape(shape = reshape_0_shape_0, x = input_13_cast_fp16)[name = tensor("reshape_0_cast_fp16")]; + tensor reduce_mean_0_axes_0 = const()[name = tensor("reduce_mean_0_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_0_keep_dims_0 = const()[name = tensor("reduce_mean_0_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_0_cast_fp16 = reduce_mean(axes = reduce_mean_0_axes_0, keep_dims = reduce_mean_0_keep_dims_0, x = reshape_0_cast_fp16)[name = tensor("reduce_mean_0_cast_fp16")]; + tensor sub_0_cast_fp16 = sub(x = reshape_0_cast_fp16, y = reduce_mean_0_cast_fp16)[name = tensor("sub_0_cast_fp16")]; + tensor square_0_cast_fp16 = square(x = sub_0_cast_fp16)[name = tensor("square_0_cast_fp16")]; + tensor reduce_mean_2_axes_0 = const()[name = tensor("reduce_mean_2_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_2_keep_dims_0 = const()[name = tensor("reduce_mean_2_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_2_cast_fp16 = reduce_mean(axes = reduce_mean_2_axes_0, keep_dims = reduce_mean_2_keep_dims_0, x = square_0_cast_fp16)[name = tensor("reduce_mean_2_cast_fp16")]; + tensor add_0_y_0_to_fp16 = const()[name = tensor("add_0_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_0_cast_fp16 = add(x = reduce_mean_2_cast_fp16, y = add_0_y_0_to_fp16)[name = tensor("add_0_cast_fp16")]; + tensor sqrt_0_cast_fp16 = sqrt(x = add_0_cast_fp16)[name = tensor("sqrt_0_cast_fp16")]; + tensor real_div_0_cast_fp16 = real_div(x = sub_0_cast_fp16, y = sqrt_0_cast_fp16)[name = tensor("real_div_0_cast_fp16")]; + tensor reshape_1_shape_0 = const()[name = tensor("reshape_1_shape_0"), val = tensor([2, 320, 128, 128])]; + tensor reshape_1_cast_fp16 = reshape(shape = reshape_1_shape_0, x = real_div_0_cast_fp16)[name = tensor("reshape_1_cast_fp16")]; + tensor add_1_mean_0_to_fp16 = const()[name = tensor("add_1_mean_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14617088)))]; + tensor add_1_variance_0_to_fp16 = const()[name = tensor("add_1_variance_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14617792)))]; + tensor add_1_gamma_0_to_fp16 = const()[name = tensor("add_1_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14618496)))]; + tensor add_1_beta_0_to_fp16 = const()[name = tensor("add_1_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14619200)))]; + tensor add_1_epsilon_0_to_fp16 = const()[name = tensor("add_1_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_1_cast_fp16 = batch_norm(beta = add_1_beta_0_to_fp16, epsilon = add_1_epsilon_0_to_fp16, gamma = add_1_gamma_0_to_fp16, mean = add_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_1_cast_fp16)[name = tensor("add_1_cast_fp16")]; + tensor input_17_cast_fp16 = silu(x = add_1_cast_fp16)[name = tensor("input_17_cast_fp16")]; + tensor var_190 = const()[name = tensor("op_190"), val = tensor([1, 1])]; + tensor var_192 = const()[name = tensor("op_192"), val = tensor([1, 1])]; + tensor hidden_states_1_pad_type_0 = const()[name = tensor("hidden_states_1_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_1_pad_0 = const()[name = tensor("hidden_states_1_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor down_blocks_0_resnets_0_conv1_weight_to_fp16 = const()[name = tensor("down_blocks_0_resnets_0_conv1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14619904)))]; + tensor down_blocks_0_resnets_0_conv1_bias_to_fp16 = const()[name = tensor("down_blocks_0_resnets_0_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16463168)))]; + tensor hidden_states_1_cast_fp16 = conv(bias = down_blocks_0_resnets_0_conv1_bias_to_fp16, dilations = var_192, groups = var_172, pad = hidden_states_1_pad_0, pad_type = hidden_states_1_pad_type_0, strides = var_190, weight = down_blocks_0_resnets_0_conv1_weight_to_fp16, x = input_17_cast_fp16)[name = tensor("hidden_states_1_cast_fp16")]; + tensor input_21_cast_fp16 = silu(x = input_19_cast_fp16)[name = tensor("input_21_cast_fp16")]; + tensor var_198 = const()[name = tensor("op_198"), val = tensor([1, 1])]; + tensor var_200 = const()[name = tensor("op_200"), val = tensor([1, 1])]; + tensor temb_1_pad_type_0 = const()[name = tensor("temb_1_pad_type_0"), val = tensor("custom")]; + tensor temb_1_pad_0 = const()[name = tensor("temb_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_0_resnets_0_time_emb_proj_weight_to_fp16 = const()[name = tensor("down_blocks_0_resnets_0_time_emb_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16463872)))]; + tensor down_blocks_0_resnets_0_time_emb_proj_bias_to_fp16 = const()[name = tensor("down_blocks_0_resnets_0_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(17283136)))]; + tensor temb_1_cast_fp16 = conv(bias = down_blocks_0_resnets_0_time_emb_proj_bias_to_fp16, dilations = var_200, groups = var_172, pad = temb_1_pad_0, pad_type = temb_1_pad_type_0, strides = var_198, weight = down_blocks_0_resnets_0_time_emb_proj_weight_to_fp16, x = input_21_cast_fp16)[name = tensor("temb_1_cast_fp16")]; + tensor input_23_cast_fp16 = add(x = hidden_states_1_cast_fp16, y = temb_1_cast_fp16)[name = tensor("input_23_cast_fp16")]; + tensor reshape_4_shape_0 = const()[name = tensor("reshape_4_shape_0"), val = tensor([2, 32, 10, 128, 128])]; + tensor reshape_4_cast_fp16 = reshape(shape = reshape_4_shape_0, x = input_23_cast_fp16)[name = tensor("reshape_4_cast_fp16")]; + tensor reduce_mean_3_axes_0 = const()[name = tensor("reduce_mean_3_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_3_keep_dims_0 = const()[name = tensor("reduce_mean_3_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_3_cast_fp16 = reduce_mean(axes = reduce_mean_3_axes_0, keep_dims = reduce_mean_3_keep_dims_0, x = reshape_4_cast_fp16)[name = tensor("reduce_mean_3_cast_fp16")]; + tensor sub_2_cast_fp16 = sub(x = reshape_4_cast_fp16, y = reduce_mean_3_cast_fp16)[name = tensor("sub_2_cast_fp16")]; + tensor square_1_cast_fp16 = square(x = sub_2_cast_fp16)[name = tensor("square_1_cast_fp16")]; + tensor reduce_mean_5_axes_0 = const()[name = tensor("reduce_mean_5_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_5_keep_dims_0 = const()[name = tensor("reduce_mean_5_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_5_cast_fp16 = reduce_mean(axes = reduce_mean_5_axes_0, keep_dims = reduce_mean_5_keep_dims_0, x = square_1_cast_fp16)[name = tensor("reduce_mean_5_cast_fp16")]; + tensor add_2_y_0_to_fp16 = const()[name = tensor("add_2_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_2_cast_fp16 = add(x = reduce_mean_5_cast_fp16, y = add_2_y_0_to_fp16)[name = tensor("add_2_cast_fp16")]; + tensor sqrt_1_cast_fp16 = sqrt(x = add_2_cast_fp16)[name = tensor("sqrt_1_cast_fp16")]; + tensor real_div_1_cast_fp16 = real_div(x = sub_2_cast_fp16, y = sqrt_1_cast_fp16)[name = tensor("real_div_1_cast_fp16")]; + tensor reshape_5_shape_0 = const()[name = tensor("reshape_5_shape_0"), val = tensor([2, 320, 128, 128])]; + tensor reshape_5_cast_fp16 = reshape(shape = reshape_5_shape_0, x = real_div_1_cast_fp16)[name = tensor("reshape_5_cast_fp16")]; + tensor add_3_gamma_0_to_fp16 = const()[name = tensor("add_3_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(17283840)))]; + tensor add_3_beta_0_to_fp16 = const()[name = tensor("add_3_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(17284544)))]; + tensor add_3_epsilon_0_to_fp16 = const()[name = tensor("add_3_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_3_cast_fp16 = batch_norm(beta = add_3_beta_0_to_fp16, epsilon = add_3_epsilon_0_to_fp16, gamma = add_3_gamma_0_to_fp16, mean = add_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_5_cast_fp16)[name = tensor("add_3_cast_fp16")]; + tensor input_27_cast_fp16 = silu(x = add_3_cast_fp16)[name = tensor("input_27_cast_fp16")]; + tensor var_210 = const()[name = tensor("op_210"), val = tensor([1, 1])]; + tensor var_212 = const()[name = tensor("op_212"), val = tensor([1, 1])]; + tensor hidden_states_3_pad_type_0 = const()[name = tensor("hidden_states_3_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_3_pad_0 = const()[name = tensor("hidden_states_3_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor down_blocks_0_resnets_0_conv2_weight_to_fp16 = const()[name = tensor("down_blocks_0_resnets_0_conv2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(17285248)))]; + tensor down_blocks_0_resnets_0_conv2_bias_to_fp16 = const()[name = tensor("down_blocks_0_resnets_0_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(19128512)))]; + tensor hidden_states_3_cast_fp16 = conv(bias = down_blocks_0_resnets_0_conv2_bias_to_fp16, dilations = var_212, groups = var_172, pad = hidden_states_3_pad_0, pad_type = hidden_states_3_pad_type_0, strides = var_210, weight = down_blocks_0_resnets_0_conv2_weight_to_fp16, x = input_27_cast_fp16)[name = tensor("hidden_states_3_cast_fp16")]; + tensor input_29_cast_fp16 = add(x = input_13_cast_fp16, y = hidden_states_3_cast_fp16)[name = tensor("input_29_cast_fp16")]; + tensor reshape_8_shape_0 = const()[name = tensor("reshape_8_shape_0"), val = tensor([2, 32, 10, 128, 128])]; + tensor reshape_8_cast_fp16 = reshape(shape = reshape_8_shape_0, x = input_29_cast_fp16)[name = tensor("reshape_8_cast_fp16")]; + tensor reduce_mean_6_axes_0 = const()[name = tensor("reduce_mean_6_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_6_keep_dims_0 = const()[name = tensor("reduce_mean_6_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_6_cast_fp16 = reduce_mean(axes = reduce_mean_6_axes_0, keep_dims = reduce_mean_6_keep_dims_0, x = reshape_8_cast_fp16)[name = tensor("reduce_mean_6_cast_fp16")]; + tensor sub_4_cast_fp16 = sub(x = reshape_8_cast_fp16, y = reduce_mean_6_cast_fp16)[name = tensor("sub_4_cast_fp16")]; + tensor square_2_cast_fp16 = square(x = sub_4_cast_fp16)[name = tensor("square_2_cast_fp16")]; + tensor reduce_mean_8_axes_0 = const()[name = tensor("reduce_mean_8_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_8_keep_dims_0 = const()[name = tensor("reduce_mean_8_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_8_cast_fp16 = reduce_mean(axes = reduce_mean_8_axes_0, keep_dims = reduce_mean_8_keep_dims_0, x = square_2_cast_fp16)[name = tensor("reduce_mean_8_cast_fp16")]; + tensor add_4_y_0_to_fp16 = const()[name = tensor("add_4_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_4_cast_fp16 = add(x = reduce_mean_8_cast_fp16, y = add_4_y_0_to_fp16)[name = tensor("add_4_cast_fp16")]; + tensor sqrt_2_cast_fp16 = sqrt(x = add_4_cast_fp16)[name = tensor("sqrt_2_cast_fp16")]; + tensor real_div_2_cast_fp16 = real_div(x = sub_4_cast_fp16, y = sqrt_2_cast_fp16)[name = tensor("real_div_2_cast_fp16")]; + tensor reshape_9_shape_0 = const()[name = tensor("reshape_9_shape_0"), val = tensor([2, 320, 128, 128])]; + tensor reshape_9_cast_fp16 = reshape(shape = reshape_9_shape_0, x = real_div_2_cast_fp16)[name = tensor("reshape_9_cast_fp16")]; + tensor add_5_gamma_0_to_fp16 = const()[name = tensor("add_5_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(19129216)))]; + tensor add_5_beta_0_to_fp16 = const()[name = tensor("add_5_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(19129920)))]; + tensor add_5_epsilon_0_to_fp16 = const()[name = tensor("add_5_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_5_cast_fp16 = batch_norm(beta = add_5_beta_0_to_fp16, epsilon = add_5_epsilon_0_to_fp16, gamma = add_5_gamma_0_to_fp16, mean = add_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_9_cast_fp16)[name = tensor("add_5_cast_fp16")]; + tensor input_33_cast_fp16 = silu(x = add_5_cast_fp16)[name = tensor("input_33_cast_fp16")]; + tensor var_227 = const()[name = tensor("op_227"), val = tensor([1, 1])]; + tensor var_229 = const()[name = tensor("op_229"), val = tensor([1, 1])]; + tensor hidden_states_5_pad_type_0 = const()[name = tensor("hidden_states_5_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_5_pad_0 = const()[name = tensor("hidden_states_5_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor down_blocks_0_resnets_1_conv1_weight_to_fp16 = const()[name = tensor("down_blocks_0_resnets_1_conv1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(19130624)))]; + tensor down_blocks_0_resnets_1_conv1_bias_to_fp16 = const()[name = tensor("down_blocks_0_resnets_1_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(20973888)))]; + tensor hidden_states_5_cast_fp16 = conv(bias = down_blocks_0_resnets_1_conv1_bias_to_fp16, dilations = var_229, groups = var_172, pad = hidden_states_5_pad_0, pad_type = hidden_states_5_pad_type_0, strides = var_227, weight = down_blocks_0_resnets_1_conv1_weight_to_fp16, x = input_33_cast_fp16)[name = tensor("hidden_states_5_cast_fp16")]; + tensor var_235 = const()[name = tensor("op_235"), val = tensor([1, 1])]; + tensor var_237 = const()[name = tensor("op_237"), val = tensor([1, 1])]; + tensor temb_3_pad_type_0 = const()[name = tensor("temb_3_pad_type_0"), val = tensor("custom")]; + tensor temb_3_pad_0 = const()[name = tensor("temb_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_0_resnets_1_time_emb_proj_weight_to_fp16 = const()[name = tensor("down_blocks_0_resnets_1_time_emb_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(20974592)))]; + tensor down_blocks_0_resnets_1_time_emb_proj_bias_to_fp16 = const()[name = tensor("down_blocks_0_resnets_1_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(21793856)))]; + tensor temb_3_cast_fp16 = conv(bias = down_blocks_0_resnets_1_time_emb_proj_bias_to_fp16, dilations = var_237, groups = var_172, pad = temb_3_pad_0, pad_type = temb_3_pad_type_0, strides = var_235, weight = down_blocks_0_resnets_1_time_emb_proj_weight_to_fp16, x = input_21_cast_fp16)[name = tensor("temb_3_cast_fp16")]; + tensor input_37_cast_fp16 = add(x = hidden_states_5_cast_fp16, y = temb_3_cast_fp16)[name = tensor("input_37_cast_fp16")]; + tensor reshape_12_shape_0 = const()[name = tensor("reshape_12_shape_0"), val = tensor([2, 32, 10, 128, 128])]; + tensor reshape_12_cast_fp16 = reshape(shape = reshape_12_shape_0, x = input_37_cast_fp16)[name = tensor("reshape_12_cast_fp16")]; + tensor reduce_mean_9_axes_0 = const()[name = tensor("reduce_mean_9_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_9_keep_dims_0 = const()[name = tensor("reduce_mean_9_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_9_cast_fp16 = reduce_mean(axes = reduce_mean_9_axes_0, keep_dims = reduce_mean_9_keep_dims_0, x = reshape_12_cast_fp16)[name = tensor("reduce_mean_9_cast_fp16")]; + tensor sub_6_cast_fp16 = sub(x = reshape_12_cast_fp16, y = reduce_mean_9_cast_fp16)[name = tensor("sub_6_cast_fp16")]; + tensor square_3_cast_fp16 = square(x = sub_6_cast_fp16)[name = tensor("square_3_cast_fp16")]; + tensor reduce_mean_11_axes_0 = const()[name = tensor("reduce_mean_11_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_11_keep_dims_0 = const()[name = tensor("reduce_mean_11_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_11_cast_fp16 = reduce_mean(axes = reduce_mean_11_axes_0, keep_dims = reduce_mean_11_keep_dims_0, x = square_3_cast_fp16)[name = tensor("reduce_mean_11_cast_fp16")]; + tensor add_6_y_0_to_fp16 = const()[name = tensor("add_6_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_6_cast_fp16 = add(x = reduce_mean_11_cast_fp16, y = add_6_y_0_to_fp16)[name = tensor("add_6_cast_fp16")]; + tensor sqrt_3_cast_fp16 = sqrt(x = add_6_cast_fp16)[name = tensor("sqrt_3_cast_fp16")]; + tensor real_div_3_cast_fp16 = real_div(x = sub_6_cast_fp16, y = sqrt_3_cast_fp16)[name = tensor("real_div_3_cast_fp16")]; + tensor reshape_13_shape_0 = const()[name = tensor("reshape_13_shape_0"), val = tensor([2, 320, 128, 128])]; + tensor reshape_13_cast_fp16 = reshape(shape = reshape_13_shape_0, x = real_div_3_cast_fp16)[name = tensor("reshape_13_cast_fp16")]; + tensor add_7_gamma_0_to_fp16 = const()[name = tensor("add_7_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(21794560)))]; + tensor add_7_beta_0_to_fp16 = const()[name = tensor("add_7_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(21795264)))]; + tensor add_7_epsilon_0_to_fp16 = const()[name = tensor("add_7_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_7_cast_fp16 = batch_norm(beta = add_7_beta_0_to_fp16, epsilon = add_7_epsilon_0_to_fp16, gamma = add_7_gamma_0_to_fp16, mean = add_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_13_cast_fp16)[name = tensor("add_7_cast_fp16")]; + tensor input_41_cast_fp16 = silu(x = add_7_cast_fp16)[name = tensor("input_41_cast_fp16")]; + tensor var_247 = const()[name = tensor("op_247"), val = tensor([1, 1])]; + tensor var_249 = const()[name = tensor("op_249"), val = tensor([1, 1])]; + tensor hidden_states_7_pad_type_0 = const()[name = tensor("hidden_states_7_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_7_pad_0 = const()[name = tensor("hidden_states_7_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor down_blocks_0_resnets_1_conv2_weight_to_fp16 = const()[name = tensor("down_blocks_0_resnets_1_conv2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(21795968)))]; + tensor down_blocks_0_resnets_1_conv2_bias_to_fp16 = const()[name = tensor("down_blocks_0_resnets_1_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(23639232)))]; + tensor hidden_states_7_cast_fp16 = conv(bias = down_blocks_0_resnets_1_conv2_bias_to_fp16, dilations = var_249, groups = var_172, pad = hidden_states_7_pad_0, pad_type = hidden_states_7_pad_type_0, strides = var_247, weight = down_blocks_0_resnets_1_conv2_weight_to_fp16, x = input_41_cast_fp16)[name = tensor("hidden_states_7_cast_fp16")]; + tensor input_43_cast_fp16 = add(x = input_29_cast_fp16, y = hidden_states_7_cast_fp16)[name = tensor("input_43_cast_fp16")]; + tensor var_256 = const()[name = tensor("op_256"), val = tensor([2, 2])]; + tensor var_258 = const()[name = tensor("op_258"), val = tensor([1, 1])]; + tensor input_45_pad_type_0 = const()[name = tensor("input_45_pad_type_0"), val = tensor("custom")]; + tensor input_45_pad_0 = const()[name = tensor("input_45_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor down_blocks_0_downsamplers_0_conv_weight_to_fp16 = const()[name = tensor("down_blocks_0_downsamplers_0_conv_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(23639936)))]; + tensor down_blocks_0_downsamplers_0_conv_bias_to_fp16 = const()[name = tensor("down_blocks_0_downsamplers_0_conv_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(25483200)))]; + tensor input_45_cast_fp16 = conv(bias = down_blocks_0_downsamplers_0_conv_bias_to_fp16, dilations = var_258, groups = var_172, pad = input_45_pad_0, pad_type = input_45_pad_type_0, strides = var_256, weight = down_blocks_0_downsamplers_0_conv_weight_to_fp16, x = input_43_cast_fp16)[name = tensor("input_45_cast_fp16")]; + tensor var_288 = const()[name = tensor("op_288"), val = tensor(1)]; + tensor reshape_16_shape_0 = const()[name = tensor("reshape_16_shape_0"), val = tensor([2, 32, 10, 64, 64])]; + tensor reshape_16_cast_fp16 = reshape(shape = reshape_16_shape_0, x = input_45_cast_fp16)[name = tensor("reshape_16_cast_fp16")]; + tensor reduce_mean_12_axes_0 = const()[name = tensor("reduce_mean_12_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_12_keep_dims_0 = const()[name = tensor("reduce_mean_12_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_12_cast_fp16 = reduce_mean(axes = reduce_mean_12_axes_0, keep_dims = reduce_mean_12_keep_dims_0, x = reshape_16_cast_fp16)[name = tensor("reduce_mean_12_cast_fp16")]; + tensor sub_8_cast_fp16 = sub(x = reshape_16_cast_fp16, y = reduce_mean_12_cast_fp16)[name = tensor("sub_8_cast_fp16")]; + tensor square_4_cast_fp16 = square(x = sub_8_cast_fp16)[name = tensor("square_4_cast_fp16")]; + tensor reduce_mean_14_axes_0 = const()[name = tensor("reduce_mean_14_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_14_keep_dims_0 = const()[name = tensor("reduce_mean_14_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_14_cast_fp16 = reduce_mean(axes = reduce_mean_14_axes_0, keep_dims = reduce_mean_14_keep_dims_0, x = square_4_cast_fp16)[name = tensor("reduce_mean_14_cast_fp16")]; + tensor add_8_y_0_to_fp16 = const()[name = tensor("add_8_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_8_cast_fp16 = add(x = reduce_mean_14_cast_fp16, y = add_8_y_0_to_fp16)[name = tensor("add_8_cast_fp16")]; + tensor sqrt_4_cast_fp16 = sqrt(x = add_8_cast_fp16)[name = tensor("sqrt_4_cast_fp16")]; + tensor real_div_4_cast_fp16 = real_div(x = sub_8_cast_fp16, y = sqrt_4_cast_fp16)[name = tensor("real_div_4_cast_fp16")]; + tensor reshape_17_shape_0 = const()[name = tensor("reshape_17_shape_0"), val = tensor([2, 320, 64, 64])]; + tensor reshape_17_cast_fp16 = reshape(shape = reshape_17_shape_0, x = real_div_4_cast_fp16)[name = tensor("reshape_17_cast_fp16")]; + tensor add_9_gamma_0_to_fp16 = const()[name = tensor("add_9_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(25483904)))]; + tensor add_9_beta_0_to_fp16 = const()[name = tensor("add_9_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(25484608)))]; + tensor add_9_epsilon_0_to_fp16 = const()[name = tensor("add_9_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_9_cast_fp16 = batch_norm(beta = add_9_beta_0_to_fp16, epsilon = add_9_epsilon_0_to_fp16, gamma = add_9_gamma_0_to_fp16, mean = add_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_17_cast_fp16)[name = tensor("add_9_cast_fp16")]; + tensor input_49_cast_fp16 = silu(x = add_9_cast_fp16)[name = tensor("input_49_cast_fp16")]; + tensor var_311 = const()[name = tensor("op_311"), val = tensor([1, 1])]; + tensor var_313 = const()[name = tensor("op_313"), val = tensor([1, 1])]; + tensor hidden_states_9_pad_type_0 = const()[name = tensor("hidden_states_9_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_9_pad_0 = const()[name = tensor("hidden_states_9_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor down_blocks_1_resnets_0_conv1_weight_to_fp16 = const()[name = tensor("down_blocks_1_resnets_0_conv1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(25485312)))]; + tensor down_blocks_1_resnets_0_conv1_bias_to_fp16 = const()[name = tensor("down_blocks_1_resnets_0_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(29171776)))]; + tensor hidden_states_9_cast_fp16 = conv(bias = down_blocks_1_resnets_0_conv1_bias_to_fp16, dilations = var_313, groups = var_288, pad = hidden_states_9_pad_0, pad_type = hidden_states_9_pad_type_0, strides = var_311, weight = down_blocks_1_resnets_0_conv1_weight_to_fp16, x = input_49_cast_fp16)[name = tensor("hidden_states_9_cast_fp16")]; + tensor var_319 = const()[name = tensor("op_319"), val = tensor([1, 1])]; + tensor var_321 = const()[name = tensor("op_321"), val = tensor([1, 1])]; + tensor temb_5_pad_type_0 = const()[name = tensor("temb_5_pad_type_0"), val = tensor("custom")]; + tensor temb_5_pad_0 = const()[name = tensor("temb_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_resnets_0_time_emb_proj_weight_to_fp16 = const()[name = tensor("down_blocks_1_resnets_0_time_emb_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(29173120)))]; + tensor down_blocks_1_resnets_0_time_emb_proj_bias_to_fp16 = const()[name = tensor("down_blocks_1_resnets_0_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(30811584)))]; + tensor temb_5_cast_fp16 = conv(bias = down_blocks_1_resnets_0_time_emb_proj_bias_to_fp16, dilations = var_321, groups = var_288, pad = temb_5_pad_0, pad_type = temb_5_pad_type_0, strides = var_319, weight = down_blocks_1_resnets_0_time_emb_proj_weight_to_fp16, x = input_21_cast_fp16)[name = tensor("temb_5_cast_fp16")]; + tensor input_53_cast_fp16 = add(x = hidden_states_9_cast_fp16, y = temb_5_cast_fp16)[name = tensor("input_53_cast_fp16")]; + tensor reshape_20_shape_0 = const()[name = tensor("reshape_20_shape_0"), val = tensor([2, 32, 20, 64, 64])]; + tensor reshape_20_cast_fp16 = reshape(shape = reshape_20_shape_0, x = input_53_cast_fp16)[name = tensor("reshape_20_cast_fp16")]; + tensor reduce_mean_15_axes_0 = const()[name = tensor("reduce_mean_15_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_15_keep_dims_0 = const()[name = tensor("reduce_mean_15_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_15_cast_fp16 = reduce_mean(axes = reduce_mean_15_axes_0, keep_dims = reduce_mean_15_keep_dims_0, x = reshape_20_cast_fp16)[name = tensor("reduce_mean_15_cast_fp16")]; + tensor sub_10_cast_fp16 = sub(x = reshape_20_cast_fp16, y = reduce_mean_15_cast_fp16)[name = tensor("sub_10_cast_fp16")]; + tensor square_5_cast_fp16 = square(x = sub_10_cast_fp16)[name = tensor("square_5_cast_fp16")]; + tensor reduce_mean_17_axes_0 = const()[name = tensor("reduce_mean_17_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_17_keep_dims_0 = const()[name = tensor("reduce_mean_17_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_17_cast_fp16 = reduce_mean(axes = reduce_mean_17_axes_0, keep_dims = reduce_mean_17_keep_dims_0, x = square_5_cast_fp16)[name = tensor("reduce_mean_17_cast_fp16")]; + tensor add_10_y_0_to_fp16 = const()[name = tensor("add_10_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_10_cast_fp16 = add(x = reduce_mean_17_cast_fp16, y = add_10_y_0_to_fp16)[name = tensor("add_10_cast_fp16")]; + tensor sqrt_5_cast_fp16 = sqrt(x = add_10_cast_fp16)[name = tensor("sqrt_5_cast_fp16")]; + tensor real_div_5_cast_fp16 = real_div(x = sub_10_cast_fp16, y = sqrt_5_cast_fp16)[name = tensor("real_div_5_cast_fp16")]; + tensor reshape_21_shape_0 = const()[name = tensor("reshape_21_shape_0"), val = tensor([2, 640, 64, 64])]; + tensor reshape_21_cast_fp16 = reshape(shape = reshape_21_shape_0, x = real_div_5_cast_fp16)[name = tensor("reshape_21_cast_fp16")]; + tensor add_11_mean_0_to_fp16 = const()[name = tensor("add_11_mean_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(30812928)))]; + tensor add_11_variance_0_to_fp16 = const()[name = tensor("add_11_variance_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(30814272)))]; + tensor add_11_gamma_0_to_fp16 = const()[name = tensor("add_11_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(30815616)))]; + tensor add_11_beta_0_to_fp16 = const()[name = tensor("add_11_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(30816960)))]; + tensor add_11_epsilon_0_to_fp16 = const()[name = tensor("add_11_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_11_cast_fp16 = batch_norm(beta = add_11_beta_0_to_fp16, epsilon = add_11_epsilon_0_to_fp16, gamma = add_11_gamma_0_to_fp16, mean = add_11_mean_0_to_fp16, variance = add_11_variance_0_to_fp16, x = reshape_21_cast_fp16)[name = tensor("add_11_cast_fp16")]; + tensor input_57_cast_fp16 = silu(x = add_11_cast_fp16)[name = tensor("input_57_cast_fp16")]; + tensor var_331 = const()[name = tensor("op_331"), val = tensor([1, 1])]; + tensor var_333 = const()[name = tensor("op_333"), val = tensor([1, 1])]; + tensor hidden_states_11_pad_type_0 = const()[name = tensor("hidden_states_11_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_11_pad_0 = const()[name = tensor("hidden_states_11_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor down_blocks_1_resnets_0_conv2_weight_to_fp16 = const()[name = tensor("down_blocks_1_resnets_0_conv2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(30818304)))]; + tensor down_blocks_1_resnets_0_conv2_bias_to_fp16 = const()[name = tensor("down_blocks_1_resnets_0_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(38191168)))]; + tensor hidden_states_11_cast_fp16 = conv(bias = down_blocks_1_resnets_0_conv2_bias_to_fp16, dilations = var_333, groups = var_288, pad = hidden_states_11_pad_0, pad_type = hidden_states_11_pad_type_0, strides = var_331, weight = down_blocks_1_resnets_0_conv2_weight_to_fp16, x = input_57_cast_fp16)[name = tensor("hidden_states_11_cast_fp16")]; + tensor var_338 = const()[name = tensor("op_338"), val = tensor([1, 1])]; + tensor var_340 = const()[name = tensor("op_340"), val = tensor([1, 1])]; + tensor x_1_pad_type_0 = const()[name = tensor("x_1_pad_type_0"), val = tensor("custom")]; + tensor x_1_pad_0 = const()[name = tensor("x_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_resnets_0_conv_shortcut_weight_to_fp16 = const()[name = tensor("down_blocks_1_resnets_0_conv_shortcut_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(38192512)))]; + tensor down_blocks_1_resnets_0_conv_shortcut_bias_to_fp16 = const()[name = tensor("down_blocks_1_resnets_0_conv_shortcut_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(38602176)))]; + tensor x_1_cast_fp16 = conv(bias = down_blocks_1_resnets_0_conv_shortcut_bias_to_fp16, dilations = var_340, groups = var_288, pad = x_1_pad_0, pad_type = x_1_pad_type_0, strides = var_338, weight = down_blocks_1_resnets_0_conv_shortcut_weight_to_fp16, x = input_45_cast_fp16)[name = tensor("x_1_cast_fp16")]; + tensor hidden_states_13_cast_fp16 = add(x = x_1_cast_fp16, y = hidden_states_11_cast_fp16)[name = tensor("hidden_states_13_cast_fp16")]; + tensor reshape_24_shape_0 = const()[name = tensor("reshape_24_shape_0"), val = tensor([2, 32, 20, 64, 64])]; + tensor reshape_24_cast_fp16 = reshape(shape = reshape_24_shape_0, x = hidden_states_13_cast_fp16)[name = tensor("reshape_24_cast_fp16")]; + tensor reduce_mean_18_axes_0 = const()[name = tensor("reduce_mean_18_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_18_keep_dims_0 = const()[name = tensor("reduce_mean_18_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_18_cast_fp16 = reduce_mean(axes = reduce_mean_18_axes_0, keep_dims = reduce_mean_18_keep_dims_0, x = reshape_24_cast_fp16)[name = tensor("reduce_mean_18_cast_fp16")]; + tensor sub_12_cast_fp16 = sub(x = reshape_24_cast_fp16, y = reduce_mean_18_cast_fp16)[name = tensor("sub_12_cast_fp16")]; + tensor square_6_cast_fp16 = square(x = sub_12_cast_fp16)[name = tensor("square_6_cast_fp16")]; + tensor reduce_mean_20_axes_0 = const()[name = tensor("reduce_mean_20_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_20_keep_dims_0 = const()[name = tensor("reduce_mean_20_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_20_cast_fp16 = reduce_mean(axes = reduce_mean_20_axes_0, keep_dims = reduce_mean_20_keep_dims_0, x = square_6_cast_fp16)[name = tensor("reduce_mean_20_cast_fp16")]; + tensor add_12_y_0_to_fp16 = const()[name = tensor("add_12_y_0_to_fp16"), val = tensor(0x1.1p-20)]; + tensor add_12_cast_fp16 = add(x = reduce_mean_20_cast_fp16, y = add_12_y_0_to_fp16)[name = tensor("add_12_cast_fp16")]; + tensor sqrt_6_cast_fp16 = sqrt(x = add_12_cast_fp16)[name = tensor("sqrt_6_cast_fp16")]; + tensor real_div_6_cast_fp16 = real_div(x = sub_12_cast_fp16, y = sqrt_6_cast_fp16)[name = tensor("real_div_6_cast_fp16")]; + tensor reshape_25_shape_0 = const()[name = tensor("reshape_25_shape_0"), val = tensor([2, 640, 64, 64])]; + tensor reshape_25_cast_fp16 = reshape(shape = reshape_25_shape_0, x = real_div_6_cast_fp16)[name = tensor("reshape_25_cast_fp16")]; + tensor add_13_gamma_0_to_fp16 = const()[name = tensor("add_13_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(38603520)))]; + tensor add_13_beta_0_to_fp16 = const()[name = tensor("add_13_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(38604864)))]; + tensor add_13_epsilon_0_to_fp16 = const()[name = tensor("add_13_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_13_cast_fp16 = batch_norm(beta = add_13_beta_0_to_fp16, epsilon = add_13_epsilon_0_to_fp16, gamma = add_13_gamma_0_to_fp16, mean = add_11_mean_0_to_fp16, variance = add_11_variance_0_to_fp16, x = reshape_25_cast_fp16)[name = tensor("add_13_cast_fp16")]; + tensor var_362 = const()[name = tensor("op_362"), val = tensor([1, 1])]; + tensor var_364 = const()[name = tensor("op_364"), val = tensor([1, 1])]; + tensor hidden_states_15_pad_type_0 = const()[name = tensor("hidden_states_15_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_15_pad_0 = const()[name = tensor("hidden_states_15_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_0_proj_in_weight_to_fp16 = const()[name = tensor("down_blocks_1_attentions_0_proj_in_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(38606208)))]; + tensor down_blocks_1_attentions_0_proj_in_bias_to_fp16 = const()[name = tensor("down_blocks_1_attentions_0_proj_in_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(39425472)))]; + tensor hidden_states_15_cast_fp16 = conv(bias = down_blocks_1_attentions_0_proj_in_bias_to_fp16, dilations = var_364, groups = var_288, pad = hidden_states_15_pad_0, pad_type = hidden_states_15_pad_type_0, strides = var_362, weight = down_blocks_1_attentions_0_proj_in_weight_to_fp16, x = add_13_cast_fp16)[name = tensor("hidden_states_15_cast_fp16")]; + tensor var_369 = const()[name = tensor("op_369"), val = tensor([2, 640, 1, 4096])]; + tensor inputs_1_cast_fp16 = reshape(shape = var_369, x = hidden_states_15_cast_fp16)[name = tensor("inputs_1_cast_fp16")]; + tensor hidden_states_17_axes_0 = const()[name = tensor("hidden_states_17_axes_0"), val = tensor([1])]; + tensor hidden_states_17_gamma_0_to_fp16 = const()[name = tensor("hidden_states_17_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(39426816)))]; + tensor hidden_states_17_beta_0_to_fp16 = const()[name = tensor("hidden_states_17_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(39428160)))]; + tensor var_385_to_fp16 = const()[name = tensor("op_385_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_17_cast_fp16 = layer_norm(axes = hidden_states_17_axes_0, beta = hidden_states_17_beta_0_to_fp16, epsilon = var_385_to_fp16, gamma = hidden_states_17_gamma_0_to_fp16, x = inputs_1_cast_fp16)[name = tensor("hidden_states_17_cast_fp16")]; + tensor var_400 = const()[name = tensor("op_400"), val = tensor([1, 1])]; + tensor var_402 = const()[name = tensor("op_402"), val = tensor([1, 1])]; + tensor q_1_pad_type_0 = const()[name = tensor("q_1_pad_type_0"), val = tensor("custom")]; + tensor q_1_pad_0 = const()[name = tensor("q_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16 = const()[name = tensor("down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(39429504)))]; + tensor q_1_cast_fp16 = conv(dilations = var_402, groups = var_288, pad = q_1_pad_0, pad_type = q_1_pad_type_0, strides = var_400, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16, x = hidden_states_17_cast_fp16)[name = tensor("q_1_cast_fp16")]; + tensor var_406 = const()[name = tensor("op_406"), val = tensor([1, 1])]; + tensor var_408 = const()[name = tensor("op_408"), val = tensor([1, 1])]; + tensor k_1_pad_type_0 = const()[name = tensor("k_1_pad_type_0"), val = tensor("custom")]; + tensor k_1_pad_0 = const()[name = tensor("k_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16 = const()[name = tensor("down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(40248768)))]; + tensor k_1_cast_fp16 = conv(dilations = var_408, groups = var_288, pad = k_1_pad_0, pad_type = k_1_pad_type_0, strides = var_406, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16, x = hidden_states_17_cast_fp16)[name = tensor("k_1_cast_fp16")]; + tensor var_412 = const()[name = tensor("op_412"), val = tensor([1, 1])]; + tensor var_414 = const()[name = tensor("op_414"), val = tensor([1, 1])]; + tensor v_1_pad_type_0 = const()[name = tensor("v_1_pad_type_0"), val = tensor("custom")]; + tensor v_1_pad_0 = const()[name = tensor("v_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16 = const()[name = tensor("down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(41068032)))]; + tensor v_1_cast_fp16 = conv(dilations = var_414, groups = var_288, pad = v_1_pad_0, pad_type = v_1_pad_type_0, strides = var_412, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16, x = hidden_states_17_cast_fp16)[name = tensor("v_1_cast_fp16")]; + tensor var_418_begin_0 = const()[name = tensor("op_418_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_418_end_0 = const()[name = tensor("op_418_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_418_end_mask_0 = const()[name = tensor("op_418_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_418_cast_fp16 = slice_by_index(begin = var_418_begin_0, end = var_418_end_0, end_mask = var_418_end_mask_0, x = q_1_cast_fp16)[name = tensor("op_418_cast_fp16")]; + tensor var_422_begin_0 = const()[name = tensor("op_422_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_422_end_0 = const()[name = tensor("op_422_end_0"), val = tensor([2, 128, 1, 4096])]; + tensor var_422_end_mask_0 = const()[name = tensor("op_422_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_422_cast_fp16 = slice_by_index(begin = var_422_begin_0, end = var_422_end_0, end_mask = var_422_end_mask_0, x = q_1_cast_fp16)[name = tensor("op_422_cast_fp16")]; + tensor var_426_begin_0 = const()[name = tensor("op_426_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_426_end_0 = const()[name = tensor("op_426_end_0"), val = tensor([2, 192, 1, 4096])]; + tensor var_426_end_mask_0 = const()[name = tensor("op_426_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_426_cast_fp16 = slice_by_index(begin = var_426_begin_0, end = var_426_end_0, end_mask = var_426_end_mask_0, x = q_1_cast_fp16)[name = tensor("op_426_cast_fp16")]; + tensor var_430_begin_0 = const()[name = tensor("op_430_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_430_end_0 = const()[name = tensor("op_430_end_0"), val = tensor([2, 256, 1, 4096])]; + tensor var_430_end_mask_0 = const()[name = tensor("op_430_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_430_cast_fp16 = slice_by_index(begin = var_430_begin_0, end = var_430_end_0, end_mask = var_430_end_mask_0, x = q_1_cast_fp16)[name = tensor("op_430_cast_fp16")]; + tensor var_434_begin_0 = const()[name = tensor("op_434_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_434_end_0 = const()[name = tensor("op_434_end_0"), val = tensor([2, 320, 1, 4096])]; + tensor var_434_end_mask_0 = const()[name = tensor("op_434_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_434_cast_fp16 = slice_by_index(begin = var_434_begin_0, end = var_434_end_0, end_mask = var_434_end_mask_0, x = q_1_cast_fp16)[name = tensor("op_434_cast_fp16")]; + tensor var_438_begin_0 = const()[name = tensor("op_438_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_438_end_0 = const()[name = tensor("op_438_end_0"), val = tensor([2, 384, 1, 4096])]; + tensor var_438_end_mask_0 = const()[name = tensor("op_438_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_438_cast_fp16 = slice_by_index(begin = var_438_begin_0, end = var_438_end_0, end_mask = var_438_end_mask_0, x = q_1_cast_fp16)[name = tensor("op_438_cast_fp16")]; + tensor var_442_begin_0 = const()[name = tensor("op_442_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_442_end_0 = const()[name = tensor("op_442_end_0"), val = tensor([2, 448, 1, 4096])]; + tensor var_442_end_mask_0 = const()[name = tensor("op_442_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_442_cast_fp16 = slice_by_index(begin = var_442_begin_0, end = var_442_end_0, end_mask = var_442_end_mask_0, x = q_1_cast_fp16)[name = tensor("op_442_cast_fp16")]; + tensor var_446_begin_0 = const()[name = tensor("op_446_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_446_end_0 = const()[name = tensor("op_446_end_0"), val = tensor([2, 512, 1, 4096])]; + tensor var_446_end_mask_0 = const()[name = tensor("op_446_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_446_cast_fp16 = slice_by_index(begin = var_446_begin_0, end = var_446_end_0, end_mask = var_446_end_mask_0, x = q_1_cast_fp16)[name = tensor("op_446_cast_fp16")]; + tensor var_450_begin_0 = const()[name = tensor("op_450_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_450_end_0 = const()[name = tensor("op_450_end_0"), val = tensor([2, 576, 1, 4096])]; + tensor var_450_end_mask_0 = const()[name = tensor("op_450_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_450_cast_fp16 = slice_by_index(begin = var_450_begin_0, end = var_450_end_0, end_mask = var_450_end_mask_0, x = q_1_cast_fp16)[name = tensor("op_450_cast_fp16")]; + tensor var_454_begin_0 = const()[name = tensor("op_454_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_454_end_0 = const()[name = tensor("op_454_end_0"), val = tensor([2, 640, 1, 4096])]; + tensor var_454_end_mask_0 = const()[name = tensor("op_454_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_454_cast_fp16 = slice_by_index(begin = var_454_begin_0, end = var_454_end_0, end_mask = var_454_end_mask_0, x = q_1_cast_fp16)[name = tensor("op_454_cast_fp16")]; + tensor k_3_perm_0 = const()[name = tensor("k_3_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_461_begin_0 = const()[name = tensor("op_461_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_461_end_0 = const()[name = tensor("op_461_end_0"), val = tensor([2, 4096, 1, 64])]; + tensor var_461_end_mask_0 = const()[name = tensor("op_461_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_3_cast_fp16 = transpose(perm = k_3_perm_0, x = k_1_cast_fp16)[name = tensor("transpose_139")]; + tensor var_461_cast_fp16 = slice_by_index(begin = var_461_begin_0, end = var_461_end_0, end_mask = var_461_end_mask_0, x = k_3_cast_fp16)[name = tensor("op_461_cast_fp16")]; + tensor var_465_begin_0 = const()[name = tensor("op_465_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_465_end_0 = const()[name = tensor("op_465_end_0"), val = tensor([2, 4096, 1, 128])]; + tensor var_465_end_mask_0 = const()[name = tensor("op_465_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_465_cast_fp16 = slice_by_index(begin = var_465_begin_0, end = var_465_end_0, end_mask = var_465_end_mask_0, x = k_3_cast_fp16)[name = tensor("op_465_cast_fp16")]; + tensor var_469_begin_0 = const()[name = tensor("op_469_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_469_end_0 = const()[name = tensor("op_469_end_0"), val = tensor([2, 4096, 1, 192])]; + tensor var_469_end_mask_0 = const()[name = tensor("op_469_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_469_cast_fp16 = slice_by_index(begin = var_469_begin_0, end = var_469_end_0, end_mask = var_469_end_mask_0, x = k_3_cast_fp16)[name = tensor("op_469_cast_fp16")]; + tensor var_473_begin_0 = const()[name = tensor("op_473_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_473_end_0 = const()[name = tensor("op_473_end_0"), val = tensor([2, 4096, 1, 256])]; + tensor var_473_end_mask_0 = const()[name = tensor("op_473_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_473_cast_fp16 = slice_by_index(begin = var_473_begin_0, end = var_473_end_0, end_mask = var_473_end_mask_0, x = k_3_cast_fp16)[name = tensor("op_473_cast_fp16")]; + tensor var_477_begin_0 = const()[name = tensor("op_477_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_477_end_0 = const()[name = tensor("op_477_end_0"), val = tensor([2, 4096, 1, 320])]; + tensor var_477_end_mask_0 = const()[name = tensor("op_477_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_477_cast_fp16 = slice_by_index(begin = var_477_begin_0, end = var_477_end_0, end_mask = var_477_end_mask_0, x = k_3_cast_fp16)[name = tensor("op_477_cast_fp16")]; + tensor var_481_begin_0 = const()[name = tensor("op_481_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_481_end_0 = const()[name = tensor("op_481_end_0"), val = tensor([2, 4096, 1, 384])]; + tensor var_481_end_mask_0 = const()[name = tensor("op_481_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_481_cast_fp16 = slice_by_index(begin = var_481_begin_0, end = var_481_end_0, end_mask = var_481_end_mask_0, x = k_3_cast_fp16)[name = tensor("op_481_cast_fp16")]; + tensor var_485_begin_0 = const()[name = tensor("op_485_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_485_end_0 = const()[name = tensor("op_485_end_0"), val = tensor([2, 4096, 1, 448])]; + tensor var_485_end_mask_0 = const()[name = tensor("op_485_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_485_cast_fp16 = slice_by_index(begin = var_485_begin_0, end = var_485_end_0, end_mask = var_485_end_mask_0, x = k_3_cast_fp16)[name = tensor("op_485_cast_fp16")]; + tensor var_489_begin_0 = const()[name = tensor("op_489_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_489_end_0 = const()[name = tensor("op_489_end_0"), val = tensor([2, 4096, 1, 512])]; + tensor var_489_end_mask_0 = const()[name = tensor("op_489_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_489_cast_fp16 = slice_by_index(begin = var_489_begin_0, end = var_489_end_0, end_mask = var_489_end_mask_0, x = k_3_cast_fp16)[name = tensor("op_489_cast_fp16")]; + tensor var_493_begin_0 = const()[name = tensor("op_493_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_493_end_0 = const()[name = tensor("op_493_end_0"), val = tensor([2, 4096, 1, 576])]; + tensor var_493_end_mask_0 = const()[name = tensor("op_493_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_493_cast_fp16 = slice_by_index(begin = var_493_begin_0, end = var_493_end_0, end_mask = var_493_end_mask_0, x = k_3_cast_fp16)[name = tensor("op_493_cast_fp16")]; + tensor var_497_begin_0 = const()[name = tensor("op_497_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_497_end_0 = const()[name = tensor("op_497_end_0"), val = tensor([2, 4096, 1, 640])]; + tensor var_497_end_mask_0 = const()[name = tensor("op_497_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_497_cast_fp16 = slice_by_index(begin = var_497_begin_0, end = var_497_end_0, end_mask = var_497_end_mask_0, x = k_3_cast_fp16)[name = tensor("op_497_cast_fp16")]; + tensor var_499_begin_0 = const()[name = tensor("op_499_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_499_end_0 = const()[name = tensor("op_499_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_499_end_mask_0 = const()[name = tensor("op_499_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_499_cast_fp16 = slice_by_index(begin = var_499_begin_0, end = var_499_end_0, end_mask = var_499_end_mask_0, x = v_1_cast_fp16)[name = tensor("op_499_cast_fp16")]; + tensor var_503_begin_0 = const()[name = tensor("op_503_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_503_end_0 = const()[name = tensor("op_503_end_0"), val = tensor([2, 128, 1, 4096])]; + tensor var_503_end_mask_0 = const()[name = tensor("op_503_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_503_cast_fp16 = slice_by_index(begin = var_503_begin_0, end = var_503_end_0, end_mask = var_503_end_mask_0, x = v_1_cast_fp16)[name = tensor("op_503_cast_fp16")]; + tensor var_507_begin_0 = const()[name = tensor("op_507_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_507_end_0 = const()[name = tensor("op_507_end_0"), val = tensor([2, 192, 1, 4096])]; + tensor var_507_end_mask_0 = const()[name = tensor("op_507_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_507_cast_fp16 = slice_by_index(begin = var_507_begin_0, end = var_507_end_0, end_mask = var_507_end_mask_0, x = v_1_cast_fp16)[name = tensor("op_507_cast_fp16")]; + tensor var_511_begin_0 = const()[name = tensor("op_511_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_511_end_0 = const()[name = tensor("op_511_end_0"), val = tensor([2, 256, 1, 4096])]; + tensor var_511_end_mask_0 = const()[name = tensor("op_511_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_511_cast_fp16 = slice_by_index(begin = var_511_begin_0, end = var_511_end_0, end_mask = var_511_end_mask_0, x = v_1_cast_fp16)[name = tensor("op_511_cast_fp16")]; + tensor var_515_begin_0 = const()[name = tensor("op_515_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_515_end_0 = const()[name = tensor("op_515_end_0"), val = tensor([2, 320, 1, 4096])]; + tensor var_515_end_mask_0 = const()[name = tensor("op_515_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_515_cast_fp16 = slice_by_index(begin = var_515_begin_0, end = var_515_end_0, end_mask = var_515_end_mask_0, x = v_1_cast_fp16)[name = tensor("op_515_cast_fp16")]; + tensor var_519_begin_0 = const()[name = tensor("op_519_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_519_end_0 = const()[name = tensor("op_519_end_0"), val = tensor([2, 384, 1, 4096])]; + tensor var_519_end_mask_0 = const()[name = tensor("op_519_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_519_cast_fp16 = slice_by_index(begin = var_519_begin_0, end = var_519_end_0, end_mask = var_519_end_mask_0, x = v_1_cast_fp16)[name = tensor("op_519_cast_fp16")]; + tensor var_523_begin_0 = const()[name = tensor("op_523_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_523_end_0 = const()[name = tensor("op_523_end_0"), val = tensor([2, 448, 1, 4096])]; + tensor var_523_end_mask_0 = const()[name = tensor("op_523_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_523_cast_fp16 = slice_by_index(begin = var_523_begin_0, end = var_523_end_0, end_mask = var_523_end_mask_0, x = v_1_cast_fp16)[name = tensor("op_523_cast_fp16")]; + tensor var_527_begin_0 = const()[name = tensor("op_527_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_527_end_0 = const()[name = tensor("op_527_end_0"), val = tensor([2, 512, 1, 4096])]; + tensor var_527_end_mask_0 = const()[name = tensor("op_527_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_527_cast_fp16 = slice_by_index(begin = var_527_begin_0, end = var_527_end_0, end_mask = var_527_end_mask_0, x = v_1_cast_fp16)[name = tensor("op_527_cast_fp16")]; + tensor var_531_begin_0 = const()[name = tensor("op_531_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_531_end_0 = const()[name = tensor("op_531_end_0"), val = tensor([2, 576, 1, 4096])]; + tensor var_531_end_mask_0 = const()[name = tensor("op_531_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_531_cast_fp16 = slice_by_index(begin = var_531_begin_0, end = var_531_end_0, end_mask = var_531_end_mask_0, x = v_1_cast_fp16)[name = tensor("op_531_cast_fp16")]; + tensor var_535_begin_0 = const()[name = tensor("op_535_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_535_end_0 = const()[name = tensor("op_535_end_0"), val = tensor([2, 640, 1, 4096])]; + tensor var_535_end_mask_0 = const()[name = tensor("op_535_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_535_cast_fp16 = slice_by_index(begin = var_535_begin_0, end = var_535_end_0, end_mask = var_535_end_mask_0, x = v_1_cast_fp16)[name = tensor("op_535_cast_fp16")]; + tensor var_539_equation_0 = const()[name = tensor("op_539_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_539_cast_fp16 = einsum(equation = var_539_equation_0, values = (var_461_cast_fp16, var_418_cast_fp16))[name = tensor("op_539_cast_fp16")]; + tensor var_540_to_fp16 = const()[name = tensor("op_540_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1_cast_fp16 = mul(x = var_539_cast_fp16, y = var_540_to_fp16)[name = tensor("aw_1_cast_fp16")]; + tensor var_543_equation_0 = const()[name = tensor("op_543_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_543_cast_fp16 = einsum(equation = var_543_equation_0, values = (var_465_cast_fp16, var_422_cast_fp16))[name = tensor("op_543_cast_fp16")]; + tensor var_544_to_fp16 = const()[name = tensor("op_544_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3_cast_fp16 = mul(x = var_543_cast_fp16, y = var_544_to_fp16)[name = tensor("aw_3_cast_fp16")]; + tensor var_547_equation_0 = const()[name = tensor("op_547_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_547_cast_fp16 = einsum(equation = var_547_equation_0, values = (var_469_cast_fp16, var_426_cast_fp16))[name = tensor("op_547_cast_fp16")]; + tensor var_548_to_fp16 = const()[name = tensor("op_548_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5_cast_fp16 = mul(x = var_547_cast_fp16, y = var_548_to_fp16)[name = tensor("aw_5_cast_fp16")]; + tensor var_551_equation_0 = const()[name = tensor("op_551_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_551_cast_fp16 = einsum(equation = var_551_equation_0, values = (var_473_cast_fp16, var_430_cast_fp16))[name = tensor("op_551_cast_fp16")]; + tensor var_552_to_fp16 = const()[name = tensor("op_552_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_7_cast_fp16 = mul(x = var_551_cast_fp16, y = var_552_to_fp16)[name = tensor("aw_7_cast_fp16")]; + tensor var_555_equation_0 = const()[name = tensor("op_555_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_555_cast_fp16 = einsum(equation = var_555_equation_0, values = (var_477_cast_fp16, var_434_cast_fp16))[name = tensor("op_555_cast_fp16")]; + tensor var_556_to_fp16 = const()[name = tensor("op_556_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_9_cast_fp16 = mul(x = var_555_cast_fp16, y = var_556_to_fp16)[name = tensor("aw_9_cast_fp16")]; + tensor var_559_equation_0 = const()[name = tensor("op_559_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_559_cast_fp16 = einsum(equation = var_559_equation_0, values = (var_481_cast_fp16, var_438_cast_fp16))[name = tensor("op_559_cast_fp16")]; + tensor var_560_to_fp16 = const()[name = tensor("op_560_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_11_cast_fp16 = mul(x = var_559_cast_fp16, y = var_560_to_fp16)[name = tensor("aw_11_cast_fp16")]; + tensor var_563_equation_0 = const()[name = tensor("op_563_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_563_cast_fp16 = einsum(equation = var_563_equation_0, values = (var_485_cast_fp16, var_442_cast_fp16))[name = tensor("op_563_cast_fp16")]; + tensor var_564_to_fp16 = const()[name = tensor("op_564_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_13_cast_fp16 = mul(x = var_563_cast_fp16, y = var_564_to_fp16)[name = tensor("aw_13_cast_fp16")]; + tensor var_567_equation_0 = const()[name = tensor("op_567_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_567_cast_fp16 = einsum(equation = var_567_equation_0, values = (var_489_cast_fp16, var_446_cast_fp16))[name = tensor("op_567_cast_fp16")]; + tensor var_568_to_fp16 = const()[name = tensor("op_568_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_15_cast_fp16 = mul(x = var_567_cast_fp16, y = var_568_to_fp16)[name = tensor("aw_15_cast_fp16")]; + tensor var_571_equation_0 = const()[name = tensor("op_571_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_571_cast_fp16 = einsum(equation = var_571_equation_0, values = (var_493_cast_fp16, var_450_cast_fp16))[name = tensor("op_571_cast_fp16")]; + tensor var_572_to_fp16 = const()[name = tensor("op_572_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_17_cast_fp16 = mul(x = var_571_cast_fp16, y = var_572_to_fp16)[name = tensor("aw_17_cast_fp16")]; + tensor var_575_equation_0 = const()[name = tensor("op_575_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_575_cast_fp16 = einsum(equation = var_575_equation_0, values = (var_497_cast_fp16, var_454_cast_fp16))[name = tensor("op_575_cast_fp16")]; + tensor var_576_to_fp16 = const()[name = tensor("op_576_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_19_cast_fp16 = mul(x = var_575_cast_fp16, y = var_576_to_fp16)[name = tensor("aw_19_cast_fp16")]; + tensor var_578_cast_fp16 = softmax(axis = var_288, x = aw_1_cast_fp16)[name = tensor("op_578_cast_fp16")]; + tensor var_579_cast_fp16 = softmax(axis = var_288, x = aw_3_cast_fp16)[name = tensor("op_579_cast_fp16")]; + tensor var_580_cast_fp16 = softmax(axis = var_288, x = aw_5_cast_fp16)[name = tensor("op_580_cast_fp16")]; + tensor var_581_cast_fp16 = softmax(axis = var_288, x = aw_7_cast_fp16)[name = tensor("op_581_cast_fp16")]; + tensor var_582_cast_fp16 = softmax(axis = var_288, x = aw_9_cast_fp16)[name = tensor("op_582_cast_fp16")]; + tensor var_583_cast_fp16 = softmax(axis = var_288, x = aw_11_cast_fp16)[name = tensor("op_583_cast_fp16")]; + tensor var_584_cast_fp16 = softmax(axis = var_288, x = aw_13_cast_fp16)[name = tensor("op_584_cast_fp16")]; + tensor var_585_cast_fp16 = softmax(axis = var_288, x = aw_15_cast_fp16)[name = tensor("op_585_cast_fp16")]; + tensor var_586_cast_fp16 = softmax(axis = var_288, x = aw_17_cast_fp16)[name = tensor("op_586_cast_fp16")]; + tensor var_587_cast_fp16 = softmax(axis = var_288, x = aw_19_cast_fp16)[name = tensor("op_587_cast_fp16")]; + tensor var_589_equation_0 = const()[name = tensor("op_589_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_589_cast_fp16 = einsum(equation = var_589_equation_0, values = (var_499_cast_fp16, var_578_cast_fp16))[name = tensor("op_589_cast_fp16")]; + tensor var_591_equation_0 = const()[name = tensor("op_591_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_591_cast_fp16 = einsum(equation = var_591_equation_0, values = (var_503_cast_fp16, var_579_cast_fp16))[name = tensor("op_591_cast_fp16")]; + tensor var_593_equation_0 = const()[name = tensor("op_593_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_593_cast_fp16 = einsum(equation = var_593_equation_0, values = (var_507_cast_fp16, var_580_cast_fp16))[name = tensor("op_593_cast_fp16")]; + tensor var_595_equation_0 = const()[name = tensor("op_595_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_595_cast_fp16 = einsum(equation = var_595_equation_0, values = (var_511_cast_fp16, var_581_cast_fp16))[name = tensor("op_595_cast_fp16")]; + tensor var_597_equation_0 = const()[name = tensor("op_597_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_597_cast_fp16 = einsum(equation = var_597_equation_0, values = (var_515_cast_fp16, var_582_cast_fp16))[name = tensor("op_597_cast_fp16")]; + tensor var_599_equation_0 = const()[name = tensor("op_599_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_599_cast_fp16 = einsum(equation = var_599_equation_0, values = (var_519_cast_fp16, var_583_cast_fp16))[name = tensor("op_599_cast_fp16")]; + tensor var_601_equation_0 = const()[name = tensor("op_601_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_601_cast_fp16 = einsum(equation = var_601_equation_0, values = (var_523_cast_fp16, var_584_cast_fp16))[name = tensor("op_601_cast_fp16")]; + tensor var_603_equation_0 = const()[name = tensor("op_603_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_603_cast_fp16 = einsum(equation = var_603_equation_0, values = (var_527_cast_fp16, var_585_cast_fp16))[name = tensor("op_603_cast_fp16")]; + tensor var_605_equation_0 = const()[name = tensor("op_605_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_605_cast_fp16 = einsum(equation = var_605_equation_0, values = (var_531_cast_fp16, var_586_cast_fp16))[name = tensor("op_605_cast_fp16")]; + tensor var_607_equation_0 = const()[name = tensor("op_607_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_607_cast_fp16 = einsum(equation = var_607_equation_0, values = (var_535_cast_fp16, var_587_cast_fp16))[name = tensor("op_607_cast_fp16")]; + tensor input_61_interleave_0 = const()[name = tensor("input_61_interleave_0"), val = tensor(false)]; + tensor input_61_cast_fp16 = concat(axis = var_288, interleave = input_61_interleave_0, values = (var_589_cast_fp16, var_591_cast_fp16, var_593_cast_fp16, var_595_cast_fp16, var_597_cast_fp16, var_599_cast_fp16, var_601_cast_fp16, var_603_cast_fp16, var_605_cast_fp16, var_607_cast_fp16))[name = tensor("input_61_cast_fp16")]; + tensor var_613 = const()[name = tensor("op_613"), val = tensor([1, 1])]; + tensor var_615 = const()[name = tensor("op_615"), val = tensor([1, 1])]; + tensor var_617_pad_type_0 = const()[name = tensor("op_617_pad_type_0"), val = tensor("custom")]; + tensor var_617_pad_0 = const()[name = tensor("op_617_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(41887296)))]; + tensor down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(42706560)))]; + tensor var_617_cast_fp16 = conv(bias = down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_615, groups = var_288, pad = var_617_pad_0, pad_type = var_617_pad_type_0, strides = var_613, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16, x = input_61_cast_fp16)[name = tensor("op_617_cast_fp16")]; + tensor inputs_3_cast_fp16 = add(x = var_617_cast_fp16, y = inputs_1_cast_fp16)[name = tensor("inputs_3_cast_fp16")]; + tensor hidden_states_19_axes_0 = const()[name = tensor("hidden_states_19_axes_0"), val = tensor([1])]; + tensor hidden_states_19_gamma_0_to_fp16 = const()[name = tensor("hidden_states_19_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(42707904)))]; + tensor hidden_states_19_beta_0_to_fp16 = const()[name = tensor("hidden_states_19_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(42709248)))]; + tensor var_627_to_fp16 = const()[name = tensor("op_627_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_19_cast_fp16 = layer_norm(axes = hidden_states_19_axes_0, beta = hidden_states_19_beta_0_to_fp16, epsilon = var_627_to_fp16, gamma = hidden_states_19_gamma_0_to_fp16, x = inputs_3_cast_fp16)[name = tensor("hidden_states_19_cast_fp16")]; + tensor var_642 = const()[name = tensor("op_642"), val = tensor([1, 1])]; + tensor var_644 = const()[name = tensor("op_644"), val = tensor([1, 1])]; + tensor q_3_pad_type_0 = const()[name = tensor("q_3_pad_type_0"), val = tensor("custom")]; + tensor q_3_pad_0 = const()[name = tensor("q_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16 = const()[name = tensor("down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(42710592)))]; + tensor q_3_cast_fp16 = conv(dilations = var_644, groups = var_288, pad = q_3_pad_0, pad_type = q_3_pad_type_0, strides = var_642, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16, x = hidden_states_19_cast_fp16)[name = tensor("q_3_cast_fp16")]; + tensor var_648 = const()[name = tensor("op_648"), val = tensor([1, 1])]; + tensor var_650 = const()[name = tensor("op_650"), val = tensor([1, 1])]; + tensor k_5_pad_type_0 = const()[name = tensor("k_5_pad_type_0"), val = tensor("custom")]; + tensor k_5_pad_0 = const()[name = tensor("k_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16 = const()[name = tensor("down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(43529856)))]; + tensor k_5_cast_fp16 = conv(dilations = var_650, groups = var_288, pad = k_5_pad_0, pad_type = k_5_pad_type_0, strides = var_648, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_5_cast_fp16")]; + tensor var_654 = const()[name = tensor("op_654"), val = tensor([1, 1])]; + tensor var_656 = const()[name = tensor("op_656"), val = tensor([1, 1])]; + tensor v_3_pad_type_0 = const()[name = tensor("v_3_pad_type_0"), val = tensor("custom")]; + tensor v_3_pad_0 = const()[name = tensor("v_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16 = const()[name = tensor("down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(46151360)))]; + tensor v_3_cast_fp16 = conv(dilations = var_656, groups = var_288, pad = v_3_pad_0, pad_type = v_3_pad_type_0, strides = var_654, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_3_cast_fp16")]; + tensor var_660_begin_0 = const()[name = tensor("op_660_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_660_end_0 = const()[name = tensor("op_660_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_660_end_mask_0 = const()[name = tensor("op_660_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_660_cast_fp16 = slice_by_index(begin = var_660_begin_0, end = var_660_end_0, end_mask = var_660_end_mask_0, x = q_3_cast_fp16)[name = tensor("op_660_cast_fp16")]; + tensor var_664_begin_0 = const()[name = tensor("op_664_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_664_end_0 = const()[name = tensor("op_664_end_0"), val = tensor([2, 128, 1, 4096])]; + tensor var_664_end_mask_0 = const()[name = tensor("op_664_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_664_cast_fp16 = slice_by_index(begin = var_664_begin_0, end = var_664_end_0, end_mask = var_664_end_mask_0, x = q_3_cast_fp16)[name = tensor("op_664_cast_fp16")]; + tensor var_668_begin_0 = const()[name = tensor("op_668_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_668_end_0 = const()[name = tensor("op_668_end_0"), val = tensor([2, 192, 1, 4096])]; + tensor var_668_end_mask_0 = const()[name = tensor("op_668_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_668_cast_fp16 = slice_by_index(begin = var_668_begin_0, end = var_668_end_0, end_mask = var_668_end_mask_0, x = q_3_cast_fp16)[name = tensor("op_668_cast_fp16")]; + tensor var_672_begin_0 = const()[name = tensor("op_672_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_672_end_0 = const()[name = tensor("op_672_end_0"), val = tensor([2, 256, 1, 4096])]; + tensor var_672_end_mask_0 = const()[name = tensor("op_672_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_672_cast_fp16 = slice_by_index(begin = var_672_begin_0, end = var_672_end_0, end_mask = var_672_end_mask_0, x = q_3_cast_fp16)[name = tensor("op_672_cast_fp16")]; + tensor var_676_begin_0 = const()[name = tensor("op_676_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_676_end_0 = const()[name = tensor("op_676_end_0"), val = tensor([2, 320, 1, 4096])]; + tensor var_676_end_mask_0 = const()[name = tensor("op_676_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_676_cast_fp16 = slice_by_index(begin = var_676_begin_0, end = var_676_end_0, end_mask = var_676_end_mask_0, x = q_3_cast_fp16)[name = tensor("op_676_cast_fp16")]; + tensor var_680_begin_0 = const()[name = tensor("op_680_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_680_end_0 = const()[name = tensor("op_680_end_0"), val = tensor([2, 384, 1, 4096])]; + tensor var_680_end_mask_0 = const()[name = tensor("op_680_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_680_cast_fp16 = slice_by_index(begin = var_680_begin_0, end = var_680_end_0, end_mask = var_680_end_mask_0, x = q_3_cast_fp16)[name = tensor("op_680_cast_fp16")]; + tensor var_684_begin_0 = const()[name = tensor("op_684_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_684_end_0 = const()[name = tensor("op_684_end_0"), val = tensor([2, 448, 1, 4096])]; + tensor var_684_end_mask_0 = const()[name = tensor("op_684_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_684_cast_fp16 = slice_by_index(begin = var_684_begin_0, end = var_684_end_0, end_mask = var_684_end_mask_0, x = q_3_cast_fp16)[name = tensor("op_684_cast_fp16")]; + tensor var_688_begin_0 = const()[name = tensor("op_688_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_688_end_0 = const()[name = tensor("op_688_end_0"), val = tensor([2, 512, 1, 4096])]; + tensor var_688_end_mask_0 = const()[name = tensor("op_688_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_688_cast_fp16 = slice_by_index(begin = var_688_begin_0, end = var_688_end_0, end_mask = var_688_end_mask_0, x = q_3_cast_fp16)[name = tensor("op_688_cast_fp16")]; + tensor var_692_begin_0 = const()[name = tensor("op_692_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_692_end_0 = const()[name = tensor("op_692_end_0"), val = tensor([2, 576, 1, 4096])]; + tensor var_692_end_mask_0 = const()[name = tensor("op_692_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_692_cast_fp16 = slice_by_index(begin = var_692_begin_0, end = var_692_end_0, end_mask = var_692_end_mask_0, x = q_3_cast_fp16)[name = tensor("op_692_cast_fp16")]; + tensor var_696_begin_0 = const()[name = tensor("op_696_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_696_end_0 = const()[name = tensor("op_696_end_0"), val = tensor([2, 640, 1, 4096])]; + tensor var_696_end_mask_0 = const()[name = tensor("op_696_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_696_cast_fp16 = slice_by_index(begin = var_696_begin_0, end = var_696_end_0, end_mask = var_696_end_mask_0, x = q_3_cast_fp16)[name = tensor("op_696_cast_fp16")]; + tensor k_7_perm_0 = const()[name = tensor("k_7_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_703_begin_0 = const()[name = tensor("op_703_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_703_end_0 = const()[name = tensor("op_703_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_703_end_mask_0 = const()[name = tensor("op_703_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_7_cast_fp16 = transpose(perm = k_7_perm_0, x = k_5_cast_fp16)[name = tensor("transpose_138")]; + tensor var_703_cast_fp16 = slice_by_index(begin = var_703_begin_0, end = var_703_end_0, end_mask = var_703_end_mask_0, x = k_7_cast_fp16)[name = tensor("op_703_cast_fp16")]; + tensor var_707_begin_0 = const()[name = tensor("op_707_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_707_end_0 = const()[name = tensor("op_707_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_707_end_mask_0 = const()[name = tensor("op_707_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_707_cast_fp16 = slice_by_index(begin = var_707_begin_0, end = var_707_end_0, end_mask = var_707_end_mask_0, x = k_7_cast_fp16)[name = tensor("op_707_cast_fp16")]; + tensor var_711_begin_0 = const()[name = tensor("op_711_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_711_end_0 = const()[name = tensor("op_711_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_711_end_mask_0 = const()[name = tensor("op_711_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_711_cast_fp16 = slice_by_index(begin = var_711_begin_0, end = var_711_end_0, end_mask = var_711_end_mask_0, x = k_7_cast_fp16)[name = tensor("op_711_cast_fp16")]; + tensor var_715_begin_0 = const()[name = tensor("op_715_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_715_end_0 = const()[name = tensor("op_715_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_715_end_mask_0 = const()[name = tensor("op_715_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_715_cast_fp16 = slice_by_index(begin = var_715_begin_0, end = var_715_end_0, end_mask = var_715_end_mask_0, x = k_7_cast_fp16)[name = tensor("op_715_cast_fp16")]; + tensor var_719_begin_0 = const()[name = tensor("op_719_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_719_end_0 = const()[name = tensor("op_719_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_719_end_mask_0 = const()[name = tensor("op_719_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_719_cast_fp16 = slice_by_index(begin = var_719_begin_0, end = var_719_end_0, end_mask = var_719_end_mask_0, x = k_7_cast_fp16)[name = tensor("op_719_cast_fp16")]; + tensor var_723_begin_0 = const()[name = tensor("op_723_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_723_end_0 = const()[name = tensor("op_723_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_723_end_mask_0 = const()[name = tensor("op_723_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_723_cast_fp16 = slice_by_index(begin = var_723_begin_0, end = var_723_end_0, end_mask = var_723_end_mask_0, x = k_7_cast_fp16)[name = tensor("op_723_cast_fp16")]; + tensor var_727_begin_0 = const()[name = tensor("op_727_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_727_end_0 = const()[name = tensor("op_727_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_727_end_mask_0 = const()[name = tensor("op_727_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_727_cast_fp16 = slice_by_index(begin = var_727_begin_0, end = var_727_end_0, end_mask = var_727_end_mask_0, x = k_7_cast_fp16)[name = tensor("op_727_cast_fp16")]; + tensor var_731_begin_0 = const()[name = tensor("op_731_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_731_end_0 = const()[name = tensor("op_731_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_731_end_mask_0 = const()[name = tensor("op_731_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_731_cast_fp16 = slice_by_index(begin = var_731_begin_0, end = var_731_end_0, end_mask = var_731_end_mask_0, x = k_7_cast_fp16)[name = tensor("op_731_cast_fp16")]; + tensor var_735_begin_0 = const()[name = tensor("op_735_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_735_end_0 = const()[name = tensor("op_735_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_735_end_mask_0 = const()[name = tensor("op_735_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_735_cast_fp16 = slice_by_index(begin = var_735_begin_0, end = var_735_end_0, end_mask = var_735_end_mask_0, x = k_7_cast_fp16)[name = tensor("op_735_cast_fp16")]; + tensor var_739_begin_0 = const()[name = tensor("op_739_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_739_end_0 = const()[name = tensor("op_739_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_739_end_mask_0 = const()[name = tensor("op_739_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_739_cast_fp16 = slice_by_index(begin = var_739_begin_0, end = var_739_end_0, end_mask = var_739_end_mask_0, x = k_7_cast_fp16)[name = tensor("op_739_cast_fp16")]; + tensor var_741_begin_0 = const()[name = tensor("op_741_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_741_end_0 = const()[name = tensor("op_741_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_741_end_mask_0 = const()[name = tensor("op_741_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_741_cast_fp16 = slice_by_index(begin = var_741_begin_0, end = var_741_end_0, end_mask = var_741_end_mask_0, x = v_3_cast_fp16)[name = tensor("op_741_cast_fp16")]; + tensor var_745_begin_0 = const()[name = tensor("op_745_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_745_end_0 = const()[name = tensor("op_745_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_745_end_mask_0 = const()[name = tensor("op_745_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_745_cast_fp16 = slice_by_index(begin = var_745_begin_0, end = var_745_end_0, end_mask = var_745_end_mask_0, x = v_3_cast_fp16)[name = tensor("op_745_cast_fp16")]; + tensor var_749_begin_0 = const()[name = tensor("op_749_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_749_end_0 = const()[name = tensor("op_749_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_749_end_mask_0 = const()[name = tensor("op_749_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_749_cast_fp16 = slice_by_index(begin = var_749_begin_0, end = var_749_end_0, end_mask = var_749_end_mask_0, x = v_3_cast_fp16)[name = tensor("op_749_cast_fp16")]; + tensor var_753_begin_0 = const()[name = tensor("op_753_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_753_end_0 = const()[name = tensor("op_753_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_753_end_mask_0 = const()[name = tensor("op_753_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_753_cast_fp16 = slice_by_index(begin = var_753_begin_0, end = var_753_end_0, end_mask = var_753_end_mask_0, x = v_3_cast_fp16)[name = tensor("op_753_cast_fp16")]; + tensor var_757_begin_0 = const()[name = tensor("op_757_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_757_end_0 = const()[name = tensor("op_757_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_757_end_mask_0 = const()[name = tensor("op_757_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_757_cast_fp16 = slice_by_index(begin = var_757_begin_0, end = var_757_end_0, end_mask = var_757_end_mask_0, x = v_3_cast_fp16)[name = tensor("op_757_cast_fp16")]; + tensor var_761_begin_0 = const()[name = tensor("op_761_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_761_end_0 = const()[name = tensor("op_761_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_761_end_mask_0 = const()[name = tensor("op_761_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_761_cast_fp16 = slice_by_index(begin = var_761_begin_0, end = var_761_end_0, end_mask = var_761_end_mask_0, x = v_3_cast_fp16)[name = tensor("op_761_cast_fp16")]; + tensor var_765_begin_0 = const()[name = tensor("op_765_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_765_end_0 = const()[name = tensor("op_765_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_765_end_mask_0 = const()[name = tensor("op_765_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_765_cast_fp16 = slice_by_index(begin = var_765_begin_0, end = var_765_end_0, end_mask = var_765_end_mask_0, x = v_3_cast_fp16)[name = tensor("op_765_cast_fp16")]; + tensor var_769_begin_0 = const()[name = tensor("op_769_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_769_end_0 = const()[name = tensor("op_769_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_769_end_mask_0 = const()[name = tensor("op_769_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_769_cast_fp16 = slice_by_index(begin = var_769_begin_0, end = var_769_end_0, end_mask = var_769_end_mask_0, x = v_3_cast_fp16)[name = tensor("op_769_cast_fp16")]; + tensor var_773_begin_0 = const()[name = tensor("op_773_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_773_end_0 = const()[name = tensor("op_773_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_773_end_mask_0 = const()[name = tensor("op_773_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_773_cast_fp16 = slice_by_index(begin = var_773_begin_0, end = var_773_end_0, end_mask = var_773_end_mask_0, x = v_3_cast_fp16)[name = tensor("op_773_cast_fp16")]; + tensor var_777_begin_0 = const()[name = tensor("op_777_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_777_end_0 = const()[name = tensor("op_777_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_777_end_mask_0 = const()[name = tensor("op_777_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_777_cast_fp16 = slice_by_index(begin = var_777_begin_0, end = var_777_end_0, end_mask = var_777_end_mask_0, x = v_3_cast_fp16)[name = tensor("op_777_cast_fp16")]; + tensor var_781_equation_0 = const()[name = tensor("op_781_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_781_cast_fp16 = einsum(equation = var_781_equation_0, values = (var_703_cast_fp16, var_660_cast_fp16))[name = tensor("op_781_cast_fp16")]; + tensor var_782_to_fp16 = const()[name = tensor("op_782_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_21_cast_fp16 = mul(x = var_781_cast_fp16, y = var_782_to_fp16)[name = tensor("aw_21_cast_fp16")]; + tensor var_785_equation_0 = const()[name = tensor("op_785_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_785_cast_fp16 = einsum(equation = var_785_equation_0, values = (var_707_cast_fp16, var_664_cast_fp16))[name = tensor("op_785_cast_fp16")]; + tensor var_786_to_fp16 = const()[name = tensor("op_786_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_23_cast_fp16 = mul(x = var_785_cast_fp16, y = var_786_to_fp16)[name = tensor("aw_23_cast_fp16")]; + tensor var_789_equation_0 = const()[name = tensor("op_789_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_789_cast_fp16 = einsum(equation = var_789_equation_0, values = (var_711_cast_fp16, var_668_cast_fp16))[name = tensor("op_789_cast_fp16")]; + tensor var_790_to_fp16 = const()[name = tensor("op_790_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_25_cast_fp16 = mul(x = var_789_cast_fp16, y = var_790_to_fp16)[name = tensor("aw_25_cast_fp16")]; + tensor var_793_equation_0 = const()[name = tensor("op_793_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_793_cast_fp16 = einsum(equation = var_793_equation_0, values = (var_715_cast_fp16, var_672_cast_fp16))[name = tensor("op_793_cast_fp16")]; + tensor var_794_to_fp16 = const()[name = tensor("op_794_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_27_cast_fp16 = mul(x = var_793_cast_fp16, y = var_794_to_fp16)[name = tensor("aw_27_cast_fp16")]; + tensor var_797_equation_0 = const()[name = tensor("op_797_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_797_cast_fp16 = einsum(equation = var_797_equation_0, values = (var_719_cast_fp16, var_676_cast_fp16))[name = tensor("op_797_cast_fp16")]; + tensor var_798_to_fp16 = const()[name = tensor("op_798_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_29_cast_fp16 = mul(x = var_797_cast_fp16, y = var_798_to_fp16)[name = tensor("aw_29_cast_fp16")]; + tensor var_801_equation_0 = const()[name = tensor("op_801_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_801_cast_fp16 = einsum(equation = var_801_equation_0, values = (var_723_cast_fp16, var_680_cast_fp16))[name = tensor("op_801_cast_fp16")]; + tensor var_802_to_fp16 = const()[name = tensor("op_802_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_31_cast_fp16 = mul(x = var_801_cast_fp16, y = var_802_to_fp16)[name = tensor("aw_31_cast_fp16")]; + tensor var_805_equation_0 = const()[name = tensor("op_805_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_805_cast_fp16 = einsum(equation = var_805_equation_0, values = (var_727_cast_fp16, var_684_cast_fp16))[name = tensor("op_805_cast_fp16")]; + tensor var_806_to_fp16 = const()[name = tensor("op_806_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_33_cast_fp16 = mul(x = var_805_cast_fp16, y = var_806_to_fp16)[name = tensor("aw_33_cast_fp16")]; + tensor var_809_equation_0 = const()[name = tensor("op_809_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_809_cast_fp16 = einsum(equation = var_809_equation_0, values = (var_731_cast_fp16, var_688_cast_fp16))[name = tensor("op_809_cast_fp16")]; + tensor var_810_to_fp16 = const()[name = tensor("op_810_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_35_cast_fp16 = mul(x = var_809_cast_fp16, y = var_810_to_fp16)[name = tensor("aw_35_cast_fp16")]; + tensor var_813_equation_0 = const()[name = tensor("op_813_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_813_cast_fp16 = einsum(equation = var_813_equation_0, values = (var_735_cast_fp16, var_692_cast_fp16))[name = tensor("op_813_cast_fp16")]; + tensor var_814_to_fp16 = const()[name = tensor("op_814_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_37_cast_fp16 = mul(x = var_813_cast_fp16, y = var_814_to_fp16)[name = tensor("aw_37_cast_fp16")]; + tensor var_817_equation_0 = const()[name = tensor("op_817_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_817_cast_fp16 = einsum(equation = var_817_equation_0, values = (var_739_cast_fp16, var_696_cast_fp16))[name = tensor("op_817_cast_fp16")]; + tensor var_818_to_fp16 = const()[name = tensor("op_818_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_39_cast_fp16 = mul(x = var_817_cast_fp16, y = var_818_to_fp16)[name = tensor("aw_39_cast_fp16")]; + tensor var_820_cast_fp16 = softmax(axis = var_288, x = aw_21_cast_fp16)[name = tensor("op_820_cast_fp16")]; + tensor var_821_cast_fp16 = softmax(axis = var_288, x = aw_23_cast_fp16)[name = tensor("op_821_cast_fp16")]; + tensor var_822_cast_fp16 = softmax(axis = var_288, x = aw_25_cast_fp16)[name = tensor("op_822_cast_fp16")]; + tensor var_823_cast_fp16 = softmax(axis = var_288, x = aw_27_cast_fp16)[name = tensor("op_823_cast_fp16")]; + tensor var_824_cast_fp16 = softmax(axis = var_288, x = aw_29_cast_fp16)[name = tensor("op_824_cast_fp16")]; + tensor var_825_cast_fp16 = softmax(axis = var_288, x = aw_31_cast_fp16)[name = tensor("op_825_cast_fp16")]; + tensor var_826_cast_fp16 = softmax(axis = var_288, x = aw_33_cast_fp16)[name = tensor("op_826_cast_fp16")]; + tensor var_827_cast_fp16 = softmax(axis = var_288, x = aw_35_cast_fp16)[name = tensor("op_827_cast_fp16")]; + tensor var_828_cast_fp16 = softmax(axis = var_288, x = aw_37_cast_fp16)[name = tensor("op_828_cast_fp16")]; + tensor var_829_cast_fp16 = softmax(axis = var_288, x = aw_39_cast_fp16)[name = tensor("op_829_cast_fp16")]; + tensor var_831_equation_0 = const()[name = tensor("op_831_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_831_cast_fp16 = einsum(equation = var_831_equation_0, values = (var_741_cast_fp16, var_820_cast_fp16))[name = tensor("op_831_cast_fp16")]; + tensor var_833_equation_0 = const()[name = tensor("op_833_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_833_cast_fp16 = einsum(equation = var_833_equation_0, values = (var_745_cast_fp16, var_821_cast_fp16))[name = tensor("op_833_cast_fp16")]; + tensor var_835_equation_0 = const()[name = tensor("op_835_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_835_cast_fp16 = einsum(equation = var_835_equation_0, values = (var_749_cast_fp16, var_822_cast_fp16))[name = tensor("op_835_cast_fp16")]; + tensor var_837_equation_0 = const()[name = tensor("op_837_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_837_cast_fp16 = einsum(equation = var_837_equation_0, values = (var_753_cast_fp16, var_823_cast_fp16))[name = tensor("op_837_cast_fp16")]; + tensor var_839_equation_0 = const()[name = tensor("op_839_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_839_cast_fp16 = einsum(equation = var_839_equation_0, values = (var_757_cast_fp16, var_824_cast_fp16))[name = tensor("op_839_cast_fp16")]; + tensor var_841_equation_0 = const()[name = tensor("op_841_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_841_cast_fp16 = einsum(equation = var_841_equation_0, values = (var_761_cast_fp16, var_825_cast_fp16))[name = tensor("op_841_cast_fp16")]; + tensor var_843_equation_0 = const()[name = tensor("op_843_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_843_cast_fp16 = einsum(equation = var_843_equation_0, values = (var_765_cast_fp16, var_826_cast_fp16))[name = tensor("op_843_cast_fp16")]; + tensor var_845_equation_0 = const()[name = tensor("op_845_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_845_cast_fp16 = einsum(equation = var_845_equation_0, values = (var_769_cast_fp16, var_827_cast_fp16))[name = tensor("op_845_cast_fp16")]; + tensor var_847_equation_0 = const()[name = tensor("op_847_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_847_cast_fp16 = einsum(equation = var_847_equation_0, values = (var_773_cast_fp16, var_828_cast_fp16))[name = tensor("op_847_cast_fp16")]; + tensor var_849_equation_0 = const()[name = tensor("op_849_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_849_cast_fp16 = einsum(equation = var_849_equation_0, values = (var_777_cast_fp16, var_829_cast_fp16))[name = tensor("op_849_cast_fp16")]; + tensor input_63_interleave_0 = const()[name = tensor("input_63_interleave_0"), val = tensor(false)]; + tensor input_63_cast_fp16 = concat(axis = var_288, interleave = input_63_interleave_0, values = (var_831_cast_fp16, var_833_cast_fp16, var_835_cast_fp16, var_837_cast_fp16, var_839_cast_fp16, var_841_cast_fp16, var_843_cast_fp16, var_845_cast_fp16, var_847_cast_fp16, var_849_cast_fp16))[name = tensor("input_63_cast_fp16")]; + tensor var_855 = const()[name = tensor("op_855"), val = tensor([1, 1])]; + tensor var_857 = const()[name = tensor("op_857"), val = tensor([1, 1])]; + tensor var_859_pad_type_0 = const()[name = tensor("op_859_pad_type_0"), val = tensor("custom")]; + tensor var_859_pad_0 = const()[name = tensor("op_859_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(48772864)))]; + tensor down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(49592128)))]; + tensor var_859_cast_fp16 = conv(bias = down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_857, groups = var_288, pad = var_859_pad_0, pad_type = var_859_pad_type_0, strides = var_855, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16, x = input_63_cast_fp16)[name = tensor("op_859_cast_fp16")]; + tensor inputs_5_cast_fp16 = add(x = var_859_cast_fp16, y = inputs_3_cast_fp16)[name = tensor("inputs_5_cast_fp16")]; + tensor input_65_axes_0 = const()[name = tensor("input_65_axes_0"), val = tensor([1])]; + tensor input_65_gamma_0_to_fp16 = const()[name = tensor("input_65_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(49593472)))]; + tensor input_65_beta_0_to_fp16 = const()[name = tensor("input_65_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(49594816)))]; + tensor var_869_to_fp16 = const()[name = tensor("op_869_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_65_cast_fp16 = layer_norm(axes = input_65_axes_0, beta = input_65_beta_0_to_fp16, epsilon = var_869_to_fp16, gamma = input_65_gamma_0_to_fp16, x = inputs_5_cast_fp16)[name = tensor("input_65_cast_fp16")]; + tensor var_885 = const()[name = tensor("op_885"), val = tensor([1, 1])]; + tensor var_887 = const()[name = tensor("op_887"), val = tensor([1, 1])]; + tensor var_889_pad_type_0 = const()[name = tensor("op_889_pad_type_0"), val = tensor("custom")]; + tensor var_889_pad_0 = const()[name = tensor("op_889_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("down_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(49596160)))]; + tensor down_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("down_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(56149824)))]; + tensor var_889_cast_fp16 = conv(bias = down_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_887, groups = var_288, pad = var_889_pad_0, pad_type = var_889_pad_type_0, strides = var_885, weight = down_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16, x = input_65_cast_fp16)[name = tensor("op_889_cast_fp16")]; + tensor var_890_split_sizes_0 = const()[name = tensor("op_890_split_sizes_0"), val = tensor([2560, 2560])]; + tensor var_890_axis_0 = const()[name = tensor("op_890_axis_0"), val = tensor(1)]; + tensor var_890_cast_fp16_0, tensor var_890_cast_fp16_1 = split(axis = var_890_axis_0, split_sizes = var_890_split_sizes_0, x = var_889_cast_fp16)[name = tensor("op_890_cast_fp16")]; + tensor var_892_mode_0 = const()[name = tensor("op_892_mode_0"), val = tensor("EXACT")]; + tensor var_892_cast_fp16 = gelu(mode = var_892_mode_0, x = var_890_cast_fp16_1)[name = tensor("op_892_cast_fp16")]; + tensor input_67_cast_fp16 = mul(x = var_890_cast_fp16_0, y = var_892_cast_fp16)[name = tensor("input_67_cast_fp16")]; + tensor var_896 = const()[name = tensor("op_896"), val = tensor([1, 1])]; + tensor var_898 = const()[name = tensor("op_898"), val = tensor([1, 1])]; + tensor var_900_pad_type_0 = const()[name = tensor("op_900_pad_type_0"), val = tensor("custom")]; + tensor var_900_pad_0 = const()[name = tensor("op_900_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16 = const()[name = tensor("down_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(56160128)))]; + tensor down_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor("down_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(59436992)))]; + tensor var_900_cast_fp16 = conv(bias = down_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_898, groups = var_288, pad = var_900_pad_0, pad_type = var_900_pad_type_0, strides = var_896, weight = down_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16, x = input_67_cast_fp16)[name = tensor("op_900_cast_fp16")]; + tensor inputs_7_cast_fp16 = add(x = var_900_cast_fp16, y = inputs_5_cast_fp16)[name = tensor("inputs_7_cast_fp16")]; + tensor hidden_states_23_axes_0 = const()[name = tensor("hidden_states_23_axes_0"), val = tensor([1])]; + tensor hidden_states_23_gamma_0_to_fp16 = const()[name = tensor("hidden_states_23_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(59438336)))]; + tensor hidden_states_23_beta_0_to_fp16 = const()[name = tensor("hidden_states_23_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(59439680)))]; + tensor var_916_to_fp16 = const()[name = tensor("op_916_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_23_cast_fp16 = layer_norm(axes = hidden_states_23_axes_0, beta = hidden_states_23_beta_0_to_fp16, epsilon = var_916_to_fp16, gamma = hidden_states_23_gamma_0_to_fp16, x = inputs_7_cast_fp16)[name = tensor("hidden_states_23_cast_fp16")]; + tensor var_931 = const()[name = tensor("op_931"), val = tensor([1, 1])]; + tensor var_933 = const()[name = tensor("op_933"), val = tensor([1, 1])]; + tensor q_5_pad_type_0 = const()[name = tensor("q_5_pad_type_0"), val = tensor("custom")]; + tensor q_5_pad_0 = const()[name = tensor("q_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_0_transformer_blocks_1_attn1_to_q_weight_to_fp16 = const()[name = tensor("down_blocks_1_attentions_0_transformer_blocks_1_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(59441024)))]; + tensor q_5_cast_fp16 = conv(dilations = var_933, groups = var_288, pad = q_5_pad_0, pad_type = q_5_pad_type_0, strides = var_931, weight = down_blocks_1_attentions_0_transformer_blocks_1_attn1_to_q_weight_to_fp16, x = hidden_states_23_cast_fp16)[name = tensor("q_5_cast_fp16")]; + tensor var_937 = const()[name = tensor("op_937"), val = tensor([1, 1])]; + tensor var_939 = const()[name = tensor("op_939"), val = tensor([1, 1])]; + tensor k_9_pad_type_0 = const()[name = tensor("k_9_pad_type_0"), val = tensor("custom")]; + tensor k_9_pad_0 = const()[name = tensor("k_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_0_transformer_blocks_1_attn1_to_k_weight_to_fp16 = const()[name = tensor("down_blocks_1_attentions_0_transformer_blocks_1_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(60260288)))]; + tensor k_9_cast_fp16 = conv(dilations = var_939, groups = var_288, pad = k_9_pad_0, pad_type = k_9_pad_type_0, strides = var_937, weight = down_blocks_1_attentions_0_transformer_blocks_1_attn1_to_k_weight_to_fp16, x = hidden_states_23_cast_fp16)[name = tensor("k_9_cast_fp16")]; + tensor var_943 = const()[name = tensor("op_943"), val = tensor([1, 1])]; + tensor var_945 = const()[name = tensor("op_945"), val = tensor([1, 1])]; + tensor v_5_pad_type_0 = const()[name = tensor("v_5_pad_type_0"), val = tensor("custom")]; + tensor v_5_pad_0 = const()[name = tensor("v_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_0_transformer_blocks_1_attn1_to_v_weight_to_fp16 = const()[name = tensor("down_blocks_1_attentions_0_transformer_blocks_1_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(61079552)))]; + tensor v_5_cast_fp16 = conv(dilations = var_945, groups = var_288, pad = v_5_pad_0, pad_type = v_5_pad_type_0, strides = var_943, weight = down_blocks_1_attentions_0_transformer_blocks_1_attn1_to_v_weight_to_fp16, x = hidden_states_23_cast_fp16)[name = tensor("v_5_cast_fp16")]; + tensor var_949_begin_0 = const()[name = tensor("op_949_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_949_end_0 = const()[name = tensor("op_949_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_949_end_mask_0 = const()[name = tensor("op_949_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_949_cast_fp16 = slice_by_index(begin = var_949_begin_0, end = var_949_end_0, end_mask = var_949_end_mask_0, x = q_5_cast_fp16)[name = tensor("op_949_cast_fp16")]; + tensor var_953_begin_0 = const()[name = tensor("op_953_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_953_end_0 = const()[name = tensor("op_953_end_0"), val = tensor([2, 128, 1, 4096])]; + tensor var_953_end_mask_0 = const()[name = tensor("op_953_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_953_cast_fp16 = slice_by_index(begin = var_953_begin_0, end = var_953_end_0, end_mask = var_953_end_mask_0, x = q_5_cast_fp16)[name = tensor("op_953_cast_fp16")]; + tensor var_957_begin_0 = const()[name = tensor("op_957_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_957_end_0 = const()[name = tensor("op_957_end_0"), val = tensor([2, 192, 1, 4096])]; + tensor var_957_end_mask_0 = const()[name = tensor("op_957_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_957_cast_fp16 = slice_by_index(begin = var_957_begin_0, end = var_957_end_0, end_mask = var_957_end_mask_0, x = q_5_cast_fp16)[name = tensor("op_957_cast_fp16")]; + tensor var_961_begin_0 = const()[name = tensor("op_961_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_961_end_0 = const()[name = tensor("op_961_end_0"), val = tensor([2, 256, 1, 4096])]; + tensor var_961_end_mask_0 = const()[name = tensor("op_961_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_961_cast_fp16 = slice_by_index(begin = var_961_begin_0, end = var_961_end_0, end_mask = var_961_end_mask_0, x = q_5_cast_fp16)[name = tensor("op_961_cast_fp16")]; + tensor var_965_begin_0 = const()[name = tensor("op_965_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_965_end_0 = const()[name = tensor("op_965_end_0"), val = tensor([2, 320, 1, 4096])]; + tensor var_965_end_mask_0 = const()[name = tensor("op_965_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_965_cast_fp16 = slice_by_index(begin = var_965_begin_0, end = var_965_end_0, end_mask = var_965_end_mask_0, x = q_5_cast_fp16)[name = tensor("op_965_cast_fp16")]; + tensor var_969_begin_0 = const()[name = tensor("op_969_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_969_end_0 = const()[name = tensor("op_969_end_0"), val = tensor([2, 384, 1, 4096])]; + tensor var_969_end_mask_0 = const()[name = tensor("op_969_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_969_cast_fp16 = slice_by_index(begin = var_969_begin_0, end = var_969_end_0, end_mask = var_969_end_mask_0, x = q_5_cast_fp16)[name = tensor("op_969_cast_fp16")]; + tensor var_973_begin_0 = const()[name = tensor("op_973_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_973_end_0 = const()[name = tensor("op_973_end_0"), val = tensor([2, 448, 1, 4096])]; + tensor var_973_end_mask_0 = const()[name = tensor("op_973_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_973_cast_fp16 = slice_by_index(begin = var_973_begin_0, end = var_973_end_0, end_mask = var_973_end_mask_0, x = q_5_cast_fp16)[name = tensor("op_973_cast_fp16")]; + tensor var_977_begin_0 = const()[name = tensor("op_977_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_977_end_0 = const()[name = tensor("op_977_end_0"), val = tensor([2, 512, 1, 4096])]; + tensor var_977_end_mask_0 = const()[name = tensor("op_977_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_977_cast_fp16 = slice_by_index(begin = var_977_begin_0, end = var_977_end_0, end_mask = var_977_end_mask_0, x = q_5_cast_fp16)[name = tensor("op_977_cast_fp16")]; + tensor var_981_begin_0 = const()[name = tensor("op_981_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_981_end_0 = const()[name = tensor("op_981_end_0"), val = tensor([2, 576, 1, 4096])]; + tensor var_981_end_mask_0 = const()[name = tensor("op_981_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_981_cast_fp16 = slice_by_index(begin = var_981_begin_0, end = var_981_end_0, end_mask = var_981_end_mask_0, x = q_5_cast_fp16)[name = tensor("op_981_cast_fp16")]; + tensor var_985_begin_0 = const()[name = tensor("op_985_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_985_end_0 = const()[name = tensor("op_985_end_0"), val = tensor([2, 640, 1, 4096])]; + tensor var_985_end_mask_0 = const()[name = tensor("op_985_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_985_cast_fp16 = slice_by_index(begin = var_985_begin_0, end = var_985_end_0, end_mask = var_985_end_mask_0, x = q_5_cast_fp16)[name = tensor("op_985_cast_fp16")]; + tensor k_11_perm_0 = const()[name = tensor("k_11_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_992_begin_0 = const()[name = tensor("op_992_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_992_end_0 = const()[name = tensor("op_992_end_0"), val = tensor([2, 4096, 1, 64])]; + tensor var_992_end_mask_0 = const()[name = tensor("op_992_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_11_cast_fp16 = transpose(perm = k_11_perm_0, x = k_9_cast_fp16)[name = tensor("transpose_137")]; + tensor var_992_cast_fp16 = slice_by_index(begin = var_992_begin_0, end = var_992_end_0, end_mask = var_992_end_mask_0, x = k_11_cast_fp16)[name = tensor("op_992_cast_fp16")]; + tensor var_996_begin_0 = const()[name = tensor("op_996_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_996_end_0 = const()[name = tensor("op_996_end_0"), val = tensor([2, 4096, 1, 128])]; + tensor var_996_end_mask_0 = const()[name = tensor("op_996_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_996_cast_fp16 = slice_by_index(begin = var_996_begin_0, end = var_996_end_0, end_mask = var_996_end_mask_0, x = k_11_cast_fp16)[name = tensor("op_996_cast_fp16")]; + tensor var_1000_begin_0 = const()[name = tensor("op_1000_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_1000_end_0 = const()[name = tensor("op_1000_end_0"), val = tensor([2, 4096, 1, 192])]; + tensor var_1000_end_mask_0 = const()[name = tensor("op_1000_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1000_cast_fp16 = slice_by_index(begin = var_1000_begin_0, end = var_1000_end_0, end_mask = var_1000_end_mask_0, x = k_11_cast_fp16)[name = tensor("op_1000_cast_fp16")]; + tensor var_1004_begin_0 = const()[name = tensor("op_1004_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_1004_end_0 = const()[name = tensor("op_1004_end_0"), val = tensor([2, 4096, 1, 256])]; + tensor var_1004_end_mask_0 = const()[name = tensor("op_1004_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1004_cast_fp16 = slice_by_index(begin = var_1004_begin_0, end = var_1004_end_0, end_mask = var_1004_end_mask_0, x = k_11_cast_fp16)[name = tensor("op_1004_cast_fp16")]; + tensor var_1008_begin_0 = const()[name = tensor("op_1008_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_1008_end_0 = const()[name = tensor("op_1008_end_0"), val = tensor([2, 4096, 1, 320])]; + tensor var_1008_end_mask_0 = const()[name = tensor("op_1008_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1008_cast_fp16 = slice_by_index(begin = var_1008_begin_0, end = var_1008_end_0, end_mask = var_1008_end_mask_0, x = k_11_cast_fp16)[name = tensor("op_1008_cast_fp16")]; + tensor var_1012_begin_0 = const()[name = tensor("op_1012_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_1012_end_0 = const()[name = tensor("op_1012_end_0"), val = tensor([2, 4096, 1, 384])]; + tensor var_1012_end_mask_0 = const()[name = tensor("op_1012_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1012_cast_fp16 = slice_by_index(begin = var_1012_begin_0, end = var_1012_end_0, end_mask = var_1012_end_mask_0, x = k_11_cast_fp16)[name = tensor("op_1012_cast_fp16")]; + tensor var_1016_begin_0 = const()[name = tensor("op_1016_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_1016_end_0 = const()[name = tensor("op_1016_end_0"), val = tensor([2, 4096, 1, 448])]; + tensor var_1016_end_mask_0 = const()[name = tensor("op_1016_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1016_cast_fp16 = slice_by_index(begin = var_1016_begin_0, end = var_1016_end_0, end_mask = var_1016_end_mask_0, x = k_11_cast_fp16)[name = tensor("op_1016_cast_fp16")]; + tensor var_1020_begin_0 = const()[name = tensor("op_1020_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_1020_end_0 = const()[name = tensor("op_1020_end_0"), val = tensor([2, 4096, 1, 512])]; + tensor var_1020_end_mask_0 = const()[name = tensor("op_1020_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1020_cast_fp16 = slice_by_index(begin = var_1020_begin_0, end = var_1020_end_0, end_mask = var_1020_end_mask_0, x = k_11_cast_fp16)[name = tensor("op_1020_cast_fp16")]; + tensor var_1024_begin_0 = const()[name = tensor("op_1024_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_1024_end_0 = const()[name = tensor("op_1024_end_0"), val = tensor([2, 4096, 1, 576])]; + tensor var_1024_end_mask_0 = const()[name = tensor("op_1024_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1024_cast_fp16 = slice_by_index(begin = var_1024_begin_0, end = var_1024_end_0, end_mask = var_1024_end_mask_0, x = k_11_cast_fp16)[name = tensor("op_1024_cast_fp16")]; + tensor var_1028_begin_0 = const()[name = tensor("op_1028_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_1028_end_0 = const()[name = tensor("op_1028_end_0"), val = tensor([2, 4096, 1, 640])]; + tensor var_1028_end_mask_0 = const()[name = tensor("op_1028_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1028_cast_fp16 = slice_by_index(begin = var_1028_begin_0, end = var_1028_end_0, end_mask = var_1028_end_mask_0, x = k_11_cast_fp16)[name = tensor("op_1028_cast_fp16")]; + tensor var_1030_begin_0 = const()[name = tensor("op_1030_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1030_end_0 = const()[name = tensor("op_1030_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_1030_end_mask_0 = const()[name = tensor("op_1030_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1030_cast_fp16 = slice_by_index(begin = var_1030_begin_0, end = var_1030_end_0, end_mask = var_1030_end_mask_0, x = v_5_cast_fp16)[name = tensor("op_1030_cast_fp16")]; + tensor var_1034_begin_0 = const()[name = tensor("op_1034_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_1034_end_0 = const()[name = tensor("op_1034_end_0"), val = tensor([2, 128, 1, 4096])]; + tensor var_1034_end_mask_0 = const()[name = tensor("op_1034_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1034_cast_fp16 = slice_by_index(begin = var_1034_begin_0, end = var_1034_end_0, end_mask = var_1034_end_mask_0, x = v_5_cast_fp16)[name = tensor("op_1034_cast_fp16")]; + tensor var_1038_begin_0 = const()[name = tensor("op_1038_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_1038_end_0 = const()[name = tensor("op_1038_end_0"), val = tensor([2, 192, 1, 4096])]; + tensor var_1038_end_mask_0 = const()[name = tensor("op_1038_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1038_cast_fp16 = slice_by_index(begin = var_1038_begin_0, end = var_1038_end_0, end_mask = var_1038_end_mask_0, x = v_5_cast_fp16)[name = tensor("op_1038_cast_fp16")]; + tensor var_1042_begin_0 = const()[name = tensor("op_1042_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_1042_end_0 = const()[name = tensor("op_1042_end_0"), val = tensor([2, 256, 1, 4096])]; + tensor var_1042_end_mask_0 = const()[name = tensor("op_1042_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1042_cast_fp16 = slice_by_index(begin = var_1042_begin_0, end = var_1042_end_0, end_mask = var_1042_end_mask_0, x = v_5_cast_fp16)[name = tensor("op_1042_cast_fp16")]; + tensor var_1046_begin_0 = const()[name = tensor("op_1046_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_1046_end_0 = const()[name = tensor("op_1046_end_0"), val = tensor([2, 320, 1, 4096])]; + tensor var_1046_end_mask_0 = const()[name = tensor("op_1046_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1046_cast_fp16 = slice_by_index(begin = var_1046_begin_0, end = var_1046_end_0, end_mask = var_1046_end_mask_0, x = v_5_cast_fp16)[name = tensor("op_1046_cast_fp16")]; + tensor var_1050_begin_0 = const()[name = tensor("op_1050_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_1050_end_0 = const()[name = tensor("op_1050_end_0"), val = tensor([2, 384, 1, 4096])]; + tensor var_1050_end_mask_0 = const()[name = tensor("op_1050_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1050_cast_fp16 = slice_by_index(begin = var_1050_begin_0, end = var_1050_end_0, end_mask = var_1050_end_mask_0, x = v_5_cast_fp16)[name = tensor("op_1050_cast_fp16")]; + tensor var_1054_begin_0 = const()[name = tensor("op_1054_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_1054_end_0 = const()[name = tensor("op_1054_end_0"), val = tensor([2, 448, 1, 4096])]; + tensor var_1054_end_mask_0 = const()[name = tensor("op_1054_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1054_cast_fp16 = slice_by_index(begin = var_1054_begin_0, end = var_1054_end_0, end_mask = var_1054_end_mask_0, x = v_5_cast_fp16)[name = tensor("op_1054_cast_fp16")]; + tensor var_1058_begin_0 = const()[name = tensor("op_1058_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_1058_end_0 = const()[name = tensor("op_1058_end_0"), val = tensor([2, 512, 1, 4096])]; + tensor var_1058_end_mask_0 = const()[name = tensor("op_1058_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1058_cast_fp16 = slice_by_index(begin = var_1058_begin_0, end = var_1058_end_0, end_mask = var_1058_end_mask_0, x = v_5_cast_fp16)[name = tensor("op_1058_cast_fp16")]; + tensor var_1062_begin_0 = const()[name = tensor("op_1062_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_1062_end_0 = const()[name = tensor("op_1062_end_0"), val = tensor([2, 576, 1, 4096])]; + tensor var_1062_end_mask_0 = const()[name = tensor("op_1062_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1062_cast_fp16 = slice_by_index(begin = var_1062_begin_0, end = var_1062_end_0, end_mask = var_1062_end_mask_0, x = v_5_cast_fp16)[name = tensor("op_1062_cast_fp16")]; + tensor var_1066_begin_0 = const()[name = tensor("op_1066_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_1066_end_0 = const()[name = tensor("op_1066_end_0"), val = tensor([2, 640, 1, 4096])]; + tensor var_1066_end_mask_0 = const()[name = tensor("op_1066_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1066_cast_fp16 = slice_by_index(begin = var_1066_begin_0, end = var_1066_end_0, end_mask = var_1066_end_mask_0, x = v_5_cast_fp16)[name = tensor("op_1066_cast_fp16")]; + tensor var_1070_equation_0 = const()[name = tensor("op_1070_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1070_cast_fp16 = einsum(equation = var_1070_equation_0, values = (var_992_cast_fp16, var_949_cast_fp16))[name = tensor("op_1070_cast_fp16")]; + tensor var_1071_to_fp16 = const()[name = tensor("op_1071_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_41_cast_fp16 = mul(x = var_1070_cast_fp16, y = var_1071_to_fp16)[name = tensor("aw_41_cast_fp16")]; + tensor var_1074_equation_0 = const()[name = tensor("op_1074_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1074_cast_fp16 = einsum(equation = var_1074_equation_0, values = (var_996_cast_fp16, var_953_cast_fp16))[name = tensor("op_1074_cast_fp16")]; + tensor var_1075_to_fp16 = const()[name = tensor("op_1075_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_43_cast_fp16 = mul(x = var_1074_cast_fp16, y = var_1075_to_fp16)[name = tensor("aw_43_cast_fp16")]; + tensor var_1078_equation_0 = const()[name = tensor("op_1078_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1078_cast_fp16 = einsum(equation = var_1078_equation_0, values = (var_1000_cast_fp16, var_957_cast_fp16))[name = tensor("op_1078_cast_fp16")]; + tensor var_1079_to_fp16 = const()[name = tensor("op_1079_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_45_cast_fp16 = mul(x = var_1078_cast_fp16, y = var_1079_to_fp16)[name = tensor("aw_45_cast_fp16")]; + tensor var_1082_equation_0 = const()[name = tensor("op_1082_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1082_cast_fp16 = einsum(equation = var_1082_equation_0, values = (var_1004_cast_fp16, var_961_cast_fp16))[name = tensor("op_1082_cast_fp16")]; + tensor var_1083_to_fp16 = const()[name = tensor("op_1083_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_47_cast_fp16 = mul(x = var_1082_cast_fp16, y = var_1083_to_fp16)[name = tensor("aw_47_cast_fp16")]; + tensor var_1086_equation_0 = const()[name = tensor("op_1086_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1086_cast_fp16 = einsum(equation = var_1086_equation_0, values = (var_1008_cast_fp16, var_965_cast_fp16))[name = tensor("op_1086_cast_fp16")]; + tensor var_1087_to_fp16 = const()[name = tensor("op_1087_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_49_cast_fp16 = mul(x = var_1086_cast_fp16, y = var_1087_to_fp16)[name = tensor("aw_49_cast_fp16")]; + tensor var_1090_equation_0 = const()[name = tensor("op_1090_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1090_cast_fp16 = einsum(equation = var_1090_equation_0, values = (var_1012_cast_fp16, var_969_cast_fp16))[name = tensor("op_1090_cast_fp16")]; + tensor var_1091_to_fp16 = const()[name = tensor("op_1091_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_51_cast_fp16 = mul(x = var_1090_cast_fp16, y = var_1091_to_fp16)[name = tensor("aw_51_cast_fp16")]; + tensor var_1094_equation_0 = const()[name = tensor("op_1094_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1094_cast_fp16 = einsum(equation = var_1094_equation_0, values = (var_1016_cast_fp16, var_973_cast_fp16))[name = tensor("op_1094_cast_fp16")]; + tensor var_1095_to_fp16 = const()[name = tensor("op_1095_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_53_cast_fp16 = mul(x = var_1094_cast_fp16, y = var_1095_to_fp16)[name = tensor("aw_53_cast_fp16")]; + tensor var_1098_equation_0 = const()[name = tensor("op_1098_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1098_cast_fp16 = einsum(equation = var_1098_equation_0, values = (var_1020_cast_fp16, var_977_cast_fp16))[name = tensor("op_1098_cast_fp16")]; + tensor var_1099_to_fp16 = const()[name = tensor("op_1099_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_55_cast_fp16 = mul(x = var_1098_cast_fp16, y = var_1099_to_fp16)[name = tensor("aw_55_cast_fp16")]; + tensor var_1102_equation_0 = const()[name = tensor("op_1102_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1102_cast_fp16 = einsum(equation = var_1102_equation_0, values = (var_1024_cast_fp16, var_981_cast_fp16))[name = tensor("op_1102_cast_fp16")]; + tensor var_1103_to_fp16 = const()[name = tensor("op_1103_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_57_cast_fp16 = mul(x = var_1102_cast_fp16, y = var_1103_to_fp16)[name = tensor("aw_57_cast_fp16")]; + tensor var_1106_equation_0 = const()[name = tensor("op_1106_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1106_cast_fp16 = einsum(equation = var_1106_equation_0, values = (var_1028_cast_fp16, var_985_cast_fp16))[name = tensor("op_1106_cast_fp16")]; + tensor var_1107_to_fp16 = const()[name = tensor("op_1107_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_59_cast_fp16 = mul(x = var_1106_cast_fp16, y = var_1107_to_fp16)[name = tensor("aw_59_cast_fp16")]; + tensor var_1109_cast_fp16 = softmax(axis = var_288, x = aw_41_cast_fp16)[name = tensor("op_1109_cast_fp16")]; + tensor var_1110_cast_fp16 = softmax(axis = var_288, x = aw_43_cast_fp16)[name = tensor("op_1110_cast_fp16")]; + tensor var_1111_cast_fp16 = softmax(axis = var_288, x = aw_45_cast_fp16)[name = tensor("op_1111_cast_fp16")]; + tensor var_1112_cast_fp16 = softmax(axis = var_288, x = aw_47_cast_fp16)[name = tensor("op_1112_cast_fp16")]; + tensor var_1113_cast_fp16 = softmax(axis = var_288, x = aw_49_cast_fp16)[name = tensor("op_1113_cast_fp16")]; + tensor var_1114_cast_fp16 = softmax(axis = var_288, x = aw_51_cast_fp16)[name = tensor("op_1114_cast_fp16")]; + tensor var_1115_cast_fp16 = softmax(axis = var_288, x = aw_53_cast_fp16)[name = tensor("op_1115_cast_fp16")]; + tensor var_1116_cast_fp16 = softmax(axis = var_288, x = aw_55_cast_fp16)[name = tensor("op_1116_cast_fp16")]; + tensor var_1117_cast_fp16 = softmax(axis = var_288, x = aw_57_cast_fp16)[name = tensor("op_1117_cast_fp16")]; + tensor var_1118_cast_fp16 = softmax(axis = var_288, x = aw_59_cast_fp16)[name = tensor("op_1118_cast_fp16")]; + tensor var_1120_equation_0 = const()[name = tensor("op_1120_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1120_cast_fp16 = einsum(equation = var_1120_equation_0, values = (var_1030_cast_fp16, var_1109_cast_fp16))[name = tensor("op_1120_cast_fp16")]; + tensor var_1122_equation_0 = const()[name = tensor("op_1122_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1122_cast_fp16 = einsum(equation = var_1122_equation_0, values = (var_1034_cast_fp16, var_1110_cast_fp16))[name = tensor("op_1122_cast_fp16")]; + tensor var_1124_equation_0 = const()[name = tensor("op_1124_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1124_cast_fp16 = einsum(equation = var_1124_equation_0, values = (var_1038_cast_fp16, var_1111_cast_fp16))[name = tensor("op_1124_cast_fp16")]; + tensor var_1126_equation_0 = const()[name = tensor("op_1126_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1126_cast_fp16 = einsum(equation = var_1126_equation_0, values = (var_1042_cast_fp16, var_1112_cast_fp16))[name = tensor("op_1126_cast_fp16")]; + tensor var_1128_equation_0 = const()[name = tensor("op_1128_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1128_cast_fp16 = einsum(equation = var_1128_equation_0, values = (var_1046_cast_fp16, var_1113_cast_fp16))[name = tensor("op_1128_cast_fp16")]; + tensor var_1130_equation_0 = const()[name = tensor("op_1130_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1130_cast_fp16 = einsum(equation = var_1130_equation_0, values = (var_1050_cast_fp16, var_1114_cast_fp16))[name = tensor("op_1130_cast_fp16")]; + tensor var_1132_equation_0 = const()[name = tensor("op_1132_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1132_cast_fp16 = einsum(equation = var_1132_equation_0, values = (var_1054_cast_fp16, var_1115_cast_fp16))[name = tensor("op_1132_cast_fp16")]; + tensor var_1134_equation_0 = const()[name = tensor("op_1134_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1134_cast_fp16 = einsum(equation = var_1134_equation_0, values = (var_1058_cast_fp16, var_1116_cast_fp16))[name = tensor("op_1134_cast_fp16")]; + tensor var_1136_equation_0 = const()[name = tensor("op_1136_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1136_cast_fp16 = einsum(equation = var_1136_equation_0, values = (var_1062_cast_fp16, var_1117_cast_fp16))[name = tensor("op_1136_cast_fp16")]; + tensor var_1138_equation_0 = const()[name = tensor("op_1138_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1138_cast_fp16 = einsum(equation = var_1138_equation_0, values = (var_1066_cast_fp16, var_1118_cast_fp16))[name = tensor("op_1138_cast_fp16")]; + tensor input_69_interleave_0 = const()[name = tensor("input_69_interleave_0"), val = tensor(false)]; + tensor input_69_cast_fp16 = concat(axis = var_288, interleave = input_69_interleave_0, values = (var_1120_cast_fp16, var_1122_cast_fp16, var_1124_cast_fp16, var_1126_cast_fp16, var_1128_cast_fp16, var_1130_cast_fp16, var_1132_cast_fp16, var_1134_cast_fp16, var_1136_cast_fp16, var_1138_cast_fp16))[name = tensor("input_69_cast_fp16")]; + tensor var_1144 = const()[name = tensor("op_1144"), val = tensor([1, 1])]; + tensor var_1146 = const()[name = tensor("op_1146"), val = tensor([1, 1])]; + tensor var_1148_pad_type_0 = const()[name = tensor("op_1148_pad_type_0"), val = tensor("custom")]; + tensor var_1148_pad_0 = const()[name = tensor("op_1148_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_0_transformer_blocks_1_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("down_blocks_1_attentions_0_transformer_blocks_1_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(61898816)))]; + tensor down_blocks_1_attentions_0_transformer_blocks_1_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_1_attentions_0_transformer_blocks_1_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(62718080)))]; + tensor var_1148_cast_fp16 = conv(bias = down_blocks_1_attentions_0_transformer_blocks_1_attn1_to_out_0_bias_to_fp16, dilations = var_1146, groups = var_288, pad = var_1148_pad_0, pad_type = var_1148_pad_type_0, strides = var_1144, weight = down_blocks_1_attentions_0_transformer_blocks_1_attn1_to_out_0_weight_to_fp16, x = input_69_cast_fp16)[name = tensor("op_1148_cast_fp16")]; + tensor inputs_9_cast_fp16 = add(x = var_1148_cast_fp16, y = inputs_7_cast_fp16)[name = tensor("inputs_9_cast_fp16")]; + tensor hidden_states_25_axes_0 = const()[name = tensor("hidden_states_25_axes_0"), val = tensor([1])]; + tensor hidden_states_25_gamma_0_to_fp16 = const()[name = tensor("hidden_states_25_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(62719424)))]; + tensor hidden_states_25_beta_0_to_fp16 = const()[name = tensor("hidden_states_25_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(62720768)))]; + tensor var_1158_to_fp16 = const()[name = tensor("op_1158_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_25_cast_fp16 = layer_norm(axes = hidden_states_25_axes_0, beta = hidden_states_25_beta_0_to_fp16, epsilon = var_1158_to_fp16, gamma = hidden_states_25_gamma_0_to_fp16, x = inputs_9_cast_fp16)[name = tensor("hidden_states_25_cast_fp16")]; + tensor var_1173 = const()[name = tensor("op_1173"), val = tensor([1, 1])]; + tensor var_1175 = const()[name = tensor("op_1175"), val = tensor([1, 1])]; + tensor q_7_pad_type_0 = const()[name = tensor("q_7_pad_type_0"), val = tensor("custom")]; + tensor q_7_pad_0 = const()[name = tensor("q_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_0_transformer_blocks_1_attn2_to_q_weight_to_fp16 = const()[name = tensor("down_blocks_1_attentions_0_transformer_blocks_1_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(62722112)))]; + tensor q_7_cast_fp16 = conv(dilations = var_1175, groups = var_288, pad = q_7_pad_0, pad_type = q_7_pad_type_0, strides = var_1173, weight = down_blocks_1_attentions_0_transformer_blocks_1_attn2_to_q_weight_to_fp16, x = hidden_states_25_cast_fp16)[name = tensor("q_7_cast_fp16")]; + tensor var_1179 = const()[name = tensor("op_1179"), val = tensor([1, 1])]; + tensor var_1181 = const()[name = tensor("op_1181"), val = tensor([1, 1])]; + tensor k_13_pad_type_0 = const()[name = tensor("k_13_pad_type_0"), val = tensor("custom")]; + tensor k_13_pad_0 = const()[name = tensor("k_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_0_transformer_blocks_1_attn2_to_k_weight_to_fp16 = const()[name = tensor("down_blocks_1_attentions_0_transformer_blocks_1_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(63541376)))]; + tensor k_13_cast_fp16 = conv(dilations = var_1181, groups = var_288, pad = k_13_pad_0, pad_type = k_13_pad_type_0, strides = var_1179, weight = down_blocks_1_attentions_0_transformer_blocks_1_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_13_cast_fp16")]; + tensor var_1185 = const()[name = tensor("op_1185"), val = tensor([1, 1])]; + tensor var_1187 = const()[name = tensor("op_1187"), val = tensor([1, 1])]; + tensor v_7_pad_type_0 = const()[name = tensor("v_7_pad_type_0"), val = tensor("custom")]; + tensor v_7_pad_0 = const()[name = tensor("v_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_0_transformer_blocks_1_attn2_to_v_weight_to_fp16 = const()[name = tensor("down_blocks_1_attentions_0_transformer_blocks_1_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(66162880)))]; + tensor v_7_cast_fp16 = conv(dilations = var_1187, groups = var_288, pad = v_7_pad_0, pad_type = v_7_pad_type_0, strides = var_1185, weight = down_blocks_1_attentions_0_transformer_blocks_1_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_7_cast_fp16")]; + tensor var_1191_begin_0 = const()[name = tensor("op_1191_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1191_end_0 = const()[name = tensor("op_1191_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_1191_end_mask_0 = const()[name = tensor("op_1191_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1191_cast_fp16 = slice_by_index(begin = var_1191_begin_0, end = var_1191_end_0, end_mask = var_1191_end_mask_0, x = q_7_cast_fp16)[name = tensor("op_1191_cast_fp16")]; + tensor var_1195_begin_0 = const()[name = tensor("op_1195_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_1195_end_0 = const()[name = tensor("op_1195_end_0"), val = tensor([2, 128, 1, 4096])]; + tensor var_1195_end_mask_0 = const()[name = tensor("op_1195_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1195_cast_fp16 = slice_by_index(begin = var_1195_begin_0, end = var_1195_end_0, end_mask = var_1195_end_mask_0, x = q_7_cast_fp16)[name = tensor("op_1195_cast_fp16")]; + tensor var_1199_begin_0 = const()[name = tensor("op_1199_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_1199_end_0 = const()[name = tensor("op_1199_end_0"), val = tensor([2, 192, 1, 4096])]; + tensor var_1199_end_mask_0 = const()[name = tensor("op_1199_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1199_cast_fp16 = slice_by_index(begin = var_1199_begin_0, end = var_1199_end_0, end_mask = var_1199_end_mask_0, x = q_7_cast_fp16)[name = tensor("op_1199_cast_fp16")]; + tensor var_1203_begin_0 = const()[name = tensor("op_1203_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_1203_end_0 = const()[name = tensor("op_1203_end_0"), val = tensor([2, 256, 1, 4096])]; + tensor var_1203_end_mask_0 = const()[name = tensor("op_1203_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1203_cast_fp16 = slice_by_index(begin = var_1203_begin_0, end = var_1203_end_0, end_mask = var_1203_end_mask_0, x = q_7_cast_fp16)[name = tensor("op_1203_cast_fp16")]; + tensor var_1207_begin_0 = const()[name = tensor("op_1207_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_1207_end_0 = const()[name = tensor("op_1207_end_0"), val = tensor([2, 320, 1, 4096])]; + tensor var_1207_end_mask_0 = const()[name = tensor("op_1207_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1207_cast_fp16 = slice_by_index(begin = var_1207_begin_0, end = var_1207_end_0, end_mask = var_1207_end_mask_0, x = q_7_cast_fp16)[name = tensor("op_1207_cast_fp16")]; + tensor var_1211_begin_0 = const()[name = tensor("op_1211_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_1211_end_0 = const()[name = tensor("op_1211_end_0"), val = tensor([2, 384, 1, 4096])]; + tensor var_1211_end_mask_0 = const()[name = tensor("op_1211_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1211_cast_fp16 = slice_by_index(begin = var_1211_begin_0, end = var_1211_end_0, end_mask = var_1211_end_mask_0, x = q_7_cast_fp16)[name = tensor("op_1211_cast_fp16")]; + tensor var_1215_begin_0 = const()[name = tensor("op_1215_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_1215_end_0 = const()[name = tensor("op_1215_end_0"), val = tensor([2, 448, 1, 4096])]; + tensor var_1215_end_mask_0 = const()[name = tensor("op_1215_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1215_cast_fp16 = slice_by_index(begin = var_1215_begin_0, end = var_1215_end_0, end_mask = var_1215_end_mask_0, x = q_7_cast_fp16)[name = tensor("op_1215_cast_fp16")]; + tensor var_1219_begin_0 = const()[name = tensor("op_1219_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_1219_end_0 = const()[name = tensor("op_1219_end_0"), val = tensor([2, 512, 1, 4096])]; + tensor var_1219_end_mask_0 = const()[name = tensor("op_1219_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1219_cast_fp16 = slice_by_index(begin = var_1219_begin_0, end = var_1219_end_0, end_mask = var_1219_end_mask_0, x = q_7_cast_fp16)[name = tensor("op_1219_cast_fp16")]; + tensor var_1223_begin_0 = const()[name = tensor("op_1223_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_1223_end_0 = const()[name = tensor("op_1223_end_0"), val = tensor([2, 576, 1, 4096])]; + tensor var_1223_end_mask_0 = const()[name = tensor("op_1223_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1223_cast_fp16 = slice_by_index(begin = var_1223_begin_0, end = var_1223_end_0, end_mask = var_1223_end_mask_0, x = q_7_cast_fp16)[name = tensor("op_1223_cast_fp16")]; + tensor var_1227_begin_0 = const()[name = tensor("op_1227_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_1227_end_0 = const()[name = tensor("op_1227_end_0"), val = tensor([2, 640, 1, 4096])]; + tensor var_1227_end_mask_0 = const()[name = tensor("op_1227_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1227_cast_fp16 = slice_by_index(begin = var_1227_begin_0, end = var_1227_end_0, end_mask = var_1227_end_mask_0, x = q_7_cast_fp16)[name = tensor("op_1227_cast_fp16")]; + tensor k_15_perm_0 = const()[name = tensor("k_15_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_1234_begin_0 = const()[name = tensor("op_1234_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1234_end_0 = const()[name = tensor("op_1234_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_1234_end_mask_0 = const()[name = tensor("op_1234_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_15_cast_fp16 = transpose(perm = k_15_perm_0, x = k_13_cast_fp16)[name = tensor("transpose_136")]; + tensor var_1234_cast_fp16 = slice_by_index(begin = var_1234_begin_0, end = var_1234_end_0, end_mask = var_1234_end_mask_0, x = k_15_cast_fp16)[name = tensor("op_1234_cast_fp16")]; + tensor var_1238_begin_0 = const()[name = tensor("op_1238_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_1238_end_0 = const()[name = tensor("op_1238_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_1238_end_mask_0 = const()[name = tensor("op_1238_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1238_cast_fp16 = slice_by_index(begin = var_1238_begin_0, end = var_1238_end_0, end_mask = var_1238_end_mask_0, x = k_15_cast_fp16)[name = tensor("op_1238_cast_fp16")]; + tensor var_1242_begin_0 = const()[name = tensor("op_1242_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_1242_end_0 = const()[name = tensor("op_1242_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_1242_end_mask_0 = const()[name = tensor("op_1242_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1242_cast_fp16 = slice_by_index(begin = var_1242_begin_0, end = var_1242_end_0, end_mask = var_1242_end_mask_0, x = k_15_cast_fp16)[name = tensor("op_1242_cast_fp16")]; + tensor var_1246_begin_0 = const()[name = tensor("op_1246_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_1246_end_0 = const()[name = tensor("op_1246_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_1246_end_mask_0 = const()[name = tensor("op_1246_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1246_cast_fp16 = slice_by_index(begin = var_1246_begin_0, end = var_1246_end_0, end_mask = var_1246_end_mask_0, x = k_15_cast_fp16)[name = tensor("op_1246_cast_fp16")]; + tensor var_1250_begin_0 = const()[name = tensor("op_1250_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_1250_end_0 = const()[name = tensor("op_1250_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_1250_end_mask_0 = const()[name = tensor("op_1250_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1250_cast_fp16 = slice_by_index(begin = var_1250_begin_0, end = var_1250_end_0, end_mask = var_1250_end_mask_0, x = k_15_cast_fp16)[name = tensor("op_1250_cast_fp16")]; + tensor var_1254_begin_0 = const()[name = tensor("op_1254_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_1254_end_0 = const()[name = tensor("op_1254_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_1254_end_mask_0 = const()[name = tensor("op_1254_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1254_cast_fp16 = slice_by_index(begin = var_1254_begin_0, end = var_1254_end_0, end_mask = var_1254_end_mask_0, x = k_15_cast_fp16)[name = tensor("op_1254_cast_fp16")]; + tensor var_1258_begin_0 = const()[name = tensor("op_1258_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_1258_end_0 = const()[name = tensor("op_1258_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_1258_end_mask_0 = const()[name = tensor("op_1258_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1258_cast_fp16 = slice_by_index(begin = var_1258_begin_0, end = var_1258_end_0, end_mask = var_1258_end_mask_0, x = k_15_cast_fp16)[name = tensor("op_1258_cast_fp16")]; + tensor var_1262_begin_0 = const()[name = tensor("op_1262_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_1262_end_0 = const()[name = tensor("op_1262_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_1262_end_mask_0 = const()[name = tensor("op_1262_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1262_cast_fp16 = slice_by_index(begin = var_1262_begin_0, end = var_1262_end_0, end_mask = var_1262_end_mask_0, x = k_15_cast_fp16)[name = tensor("op_1262_cast_fp16")]; + tensor var_1266_begin_0 = const()[name = tensor("op_1266_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_1266_end_0 = const()[name = tensor("op_1266_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_1266_end_mask_0 = const()[name = tensor("op_1266_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1266_cast_fp16 = slice_by_index(begin = var_1266_begin_0, end = var_1266_end_0, end_mask = var_1266_end_mask_0, x = k_15_cast_fp16)[name = tensor("op_1266_cast_fp16")]; + tensor var_1270_begin_0 = const()[name = tensor("op_1270_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_1270_end_0 = const()[name = tensor("op_1270_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_1270_end_mask_0 = const()[name = tensor("op_1270_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1270_cast_fp16 = slice_by_index(begin = var_1270_begin_0, end = var_1270_end_0, end_mask = var_1270_end_mask_0, x = k_15_cast_fp16)[name = tensor("op_1270_cast_fp16")]; + tensor var_1272_begin_0 = const()[name = tensor("op_1272_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1272_end_0 = const()[name = tensor("op_1272_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_1272_end_mask_0 = const()[name = tensor("op_1272_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1272_cast_fp16 = slice_by_index(begin = var_1272_begin_0, end = var_1272_end_0, end_mask = var_1272_end_mask_0, x = v_7_cast_fp16)[name = tensor("op_1272_cast_fp16")]; + tensor var_1276_begin_0 = const()[name = tensor("op_1276_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_1276_end_0 = const()[name = tensor("op_1276_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_1276_end_mask_0 = const()[name = tensor("op_1276_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1276_cast_fp16 = slice_by_index(begin = var_1276_begin_0, end = var_1276_end_0, end_mask = var_1276_end_mask_0, x = v_7_cast_fp16)[name = tensor("op_1276_cast_fp16")]; + tensor var_1280_begin_0 = const()[name = tensor("op_1280_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_1280_end_0 = const()[name = tensor("op_1280_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_1280_end_mask_0 = const()[name = tensor("op_1280_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1280_cast_fp16 = slice_by_index(begin = var_1280_begin_0, end = var_1280_end_0, end_mask = var_1280_end_mask_0, x = v_7_cast_fp16)[name = tensor("op_1280_cast_fp16")]; + tensor var_1284_begin_0 = const()[name = tensor("op_1284_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_1284_end_0 = const()[name = tensor("op_1284_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_1284_end_mask_0 = const()[name = tensor("op_1284_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1284_cast_fp16 = slice_by_index(begin = var_1284_begin_0, end = var_1284_end_0, end_mask = var_1284_end_mask_0, x = v_7_cast_fp16)[name = tensor("op_1284_cast_fp16")]; + tensor var_1288_begin_0 = const()[name = tensor("op_1288_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_1288_end_0 = const()[name = tensor("op_1288_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_1288_end_mask_0 = const()[name = tensor("op_1288_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1288_cast_fp16 = slice_by_index(begin = var_1288_begin_0, end = var_1288_end_0, end_mask = var_1288_end_mask_0, x = v_7_cast_fp16)[name = tensor("op_1288_cast_fp16")]; + tensor var_1292_begin_0 = const()[name = tensor("op_1292_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_1292_end_0 = const()[name = tensor("op_1292_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_1292_end_mask_0 = const()[name = tensor("op_1292_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1292_cast_fp16 = slice_by_index(begin = var_1292_begin_0, end = var_1292_end_0, end_mask = var_1292_end_mask_0, x = v_7_cast_fp16)[name = tensor("op_1292_cast_fp16")]; + tensor var_1296_begin_0 = const()[name = tensor("op_1296_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_1296_end_0 = const()[name = tensor("op_1296_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_1296_end_mask_0 = const()[name = tensor("op_1296_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1296_cast_fp16 = slice_by_index(begin = var_1296_begin_0, end = var_1296_end_0, end_mask = var_1296_end_mask_0, x = v_7_cast_fp16)[name = tensor("op_1296_cast_fp16")]; + tensor var_1300_begin_0 = const()[name = tensor("op_1300_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_1300_end_0 = const()[name = tensor("op_1300_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_1300_end_mask_0 = const()[name = tensor("op_1300_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1300_cast_fp16 = slice_by_index(begin = var_1300_begin_0, end = var_1300_end_0, end_mask = var_1300_end_mask_0, x = v_7_cast_fp16)[name = tensor("op_1300_cast_fp16")]; + tensor var_1304_begin_0 = const()[name = tensor("op_1304_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_1304_end_0 = const()[name = tensor("op_1304_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_1304_end_mask_0 = const()[name = tensor("op_1304_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1304_cast_fp16 = slice_by_index(begin = var_1304_begin_0, end = var_1304_end_0, end_mask = var_1304_end_mask_0, x = v_7_cast_fp16)[name = tensor("op_1304_cast_fp16")]; + tensor var_1308_begin_0 = const()[name = tensor("op_1308_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_1308_end_0 = const()[name = tensor("op_1308_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_1308_end_mask_0 = const()[name = tensor("op_1308_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1308_cast_fp16 = slice_by_index(begin = var_1308_begin_0, end = var_1308_end_0, end_mask = var_1308_end_mask_0, x = v_7_cast_fp16)[name = tensor("op_1308_cast_fp16")]; + tensor var_1312_equation_0 = const()[name = tensor("op_1312_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1312_cast_fp16 = einsum(equation = var_1312_equation_0, values = (var_1234_cast_fp16, var_1191_cast_fp16))[name = tensor("op_1312_cast_fp16")]; + tensor var_1313_to_fp16 = const()[name = tensor("op_1313_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_61_cast_fp16 = mul(x = var_1312_cast_fp16, y = var_1313_to_fp16)[name = tensor("aw_61_cast_fp16")]; + tensor var_1316_equation_0 = const()[name = tensor("op_1316_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1316_cast_fp16 = einsum(equation = var_1316_equation_0, values = (var_1238_cast_fp16, var_1195_cast_fp16))[name = tensor("op_1316_cast_fp16")]; + tensor var_1317_to_fp16 = const()[name = tensor("op_1317_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_63_cast_fp16 = mul(x = var_1316_cast_fp16, y = var_1317_to_fp16)[name = tensor("aw_63_cast_fp16")]; + tensor var_1320_equation_0 = const()[name = tensor("op_1320_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1320_cast_fp16 = einsum(equation = var_1320_equation_0, values = (var_1242_cast_fp16, var_1199_cast_fp16))[name = tensor("op_1320_cast_fp16")]; + tensor var_1321_to_fp16 = const()[name = tensor("op_1321_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_65_cast_fp16 = mul(x = var_1320_cast_fp16, y = var_1321_to_fp16)[name = tensor("aw_65_cast_fp16")]; + tensor var_1324_equation_0 = const()[name = tensor("op_1324_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1324_cast_fp16 = einsum(equation = var_1324_equation_0, values = (var_1246_cast_fp16, var_1203_cast_fp16))[name = tensor("op_1324_cast_fp16")]; + tensor var_1325_to_fp16 = const()[name = tensor("op_1325_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_67_cast_fp16 = mul(x = var_1324_cast_fp16, y = var_1325_to_fp16)[name = tensor("aw_67_cast_fp16")]; + tensor var_1328_equation_0 = const()[name = tensor("op_1328_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1328_cast_fp16 = einsum(equation = var_1328_equation_0, values = (var_1250_cast_fp16, var_1207_cast_fp16))[name = tensor("op_1328_cast_fp16")]; + tensor var_1329_to_fp16 = const()[name = tensor("op_1329_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_69_cast_fp16 = mul(x = var_1328_cast_fp16, y = var_1329_to_fp16)[name = tensor("aw_69_cast_fp16")]; + tensor var_1332_equation_0 = const()[name = tensor("op_1332_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1332_cast_fp16 = einsum(equation = var_1332_equation_0, values = (var_1254_cast_fp16, var_1211_cast_fp16))[name = tensor("op_1332_cast_fp16")]; + tensor var_1333_to_fp16 = const()[name = tensor("op_1333_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_71_cast_fp16 = mul(x = var_1332_cast_fp16, y = var_1333_to_fp16)[name = tensor("aw_71_cast_fp16")]; + tensor var_1336_equation_0 = const()[name = tensor("op_1336_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1336_cast_fp16 = einsum(equation = var_1336_equation_0, values = (var_1258_cast_fp16, var_1215_cast_fp16))[name = tensor("op_1336_cast_fp16")]; + tensor var_1337_to_fp16 = const()[name = tensor("op_1337_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_73_cast_fp16 = mul(x = var_1336_cast_fp16, y = var_1337_to_fp16)[name = tensor("aw_73_cast_fp16")]; + tensor var_1340_equation_0 = const()[name = tensor("op_1340_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1340_cast_fp16 = einsum(equation = var_1340_equation_0, values = (var_1262_cast_fp16, var_1219_cast_fp16))[name = tensor("op_1340_cast_fp16")]; + tensor var_1341_to_fp16 = const()[name = tensor("op_1341_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_75_cast_fp16 = mul(x = var_1340_cast_fp16, y = var_1341_to_fp16)[name = tensor("aw_75_cast_fp16")]; + tensor var_1344_equation_0 = const()[name = tensor("op_1344_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1344_cast_fp16 = einsum(equation = var_1344_equation_0, values = (var_1266_cast_fp16, var_1223_cast_fp16))[name = tensor("op_1344_cast_fp16")]; + tensor var_1345_to_fp16 = const()[name = tensor("op_1345_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_77_cast_fp16 = mul(x = var_1344_cast_fp16, y = var_1345_to_fp16)[name = tensor("aw_77_cast_fp16")]; + tensor var_1348_equation_0 = const()[name = tensor("op_1348_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1348_cast_fp16 = einsum(equation = var_1348_equation_0, values = (var_1270_cast_fp16, var_1227_cast_fp16))[name = tensor("op_1348_cast_fp16")]; + tensor var_1349_to_fp16 = const()[name = tensor("op_1349_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_79_cast_fp16 = mul(x = var_1348_cast_fp16, y = var_1349_to_fp16)[name = tensor("aw_79_cast_fp16")]; + tensor var_1351_cast_fp16 = softmax(axis = var_288, x = aw_61_cast_fp16)[name = tensor("op_1351_cast_fp16")]; + tensor var_1352_cast_fp16 = softmax(axis = var_288, x = aw_63_cast_fp16)[name = tensor("op_1352_cast_fp16")]; + tensor var_1353_cast_fp16 = softmax(axis = var_288, x = aw_65_cast_fp16)[name = tensor("op_1353_cast_fp16")]; + tensor var_1354_cast_fp16 = softmax(axis = var_288, x = aw_67_cast_fp16)[name = tensor("op_1354_cast_fp16")]; + tensor var_1355_cast_fp16 = softmax(axis = var_288, x = aw_69_cast_fp16)[name = tensor("op_1355_cast_fp16")]; + tensor var_1356_cast_fp16 = softmax(axis = var_288, x = aw_71_cast_fp16)[name = tensor("op_1356_cast_fp16")]; + tensor var_1357_cast_fp16 = softmax(axis = var_288, x = aw_73_cast_fp16)[name = tensor("op_1357_cast_fp16")]; + tensor var_1358_cast_fp16 = softmax(axis = var_288, x = aw_75_cast_fp16)[name = tensor("op_1358_cast_fp16")]; + tensor var_1359_cast_fp16 = softmax(axis = var_288, x = aw_77_cast_fp16)[name = tensor("op_1359_cast_fp16")]; + tensor var_1360_cast_fp16 = softmax(axis = var_288, x = aw_79_cast_fp16)[name = tensor("op_1360_cast_fp16")]; + tensor var_1362_equation_0 = const()[name = tensor("op_1362_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1362_cast_fp16 = einsum(equation = var_1362_equation_0, values = (var_1272_cast_fp16, var_1351_cast_fp16))[name = tensor("op_1362_cast_fp16")]; + tensor var_1364_equation_0 = const()[name = tensor("op_1364_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1364_cast_fp16 = einsum(equation = var_1364_equation_0, values = (var_1276_cast_fp16, var_1352_cast_fp16))[name = tensor("op_1364_cast_fp16")]; + tensor var_1366_equation_0 = const()[name = tensor("op_1366_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1366_cast_fp16 = einsum(equation = var_1366_equation_0, values = (var_1280_cast_fp16, var_1353_cast_fp16))[name = tensor("op_1366_cast_fp16")]; + tensor var_1368_equation_0 = const()[name = tensor("op_1368_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1368_cast_fp16 = einsum(equation = var_1368_equation_0, values = (var_1284_cast_fp16, var_1354_cast_fp16))[name = tensor("op_1368_cast_fp16")]; + tensor var_1370_equation_0 = const()[name = tensor("op_1370_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1370_cast_fp16 = einsum(equation = var_1370_equation_0, values = (var_1288_cast_fp16, var_1355_cast_fp16))[name = tensor("op_1370_cast_fp16")]; + tensor var_1372_equation_0 = const()[name = tensor("op_1372_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1372_cast_fp16 = einsum(equation = var_1372_equation_0, values = (var_1292_cast_fp16, var_1356_cast_fp16))[name = tensor("op_1372_cast_fp16")]; + tensor var_1374_equation_0 = const()[name = tensor("op_1374_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1374_cast_fp16 = einsum(equation = var_1374_equation_0, values = (var_1296_cast_fp16, var_1357_cast_fp16))[name = tensor("op_1374_cast_fp16")]; + tensor var_1376_equation_0 = const()[name = tensor("op_1376_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1376_cast_fp16 = einsum(equation = var_1376_equation_0, values = (var_1300_cast_fp16, var_1358_cast_fp16))[name = tensor("op_1376_cast_fp16")]; + tensor var_1378_equation_0 = const()[name = tensor("op_1378_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1378_cast_fp16 = einsum(equation = var_1378_equation_0, values = (var_1304_cast_fp16, var_1359_cast_fp16))[name = tensor("op_1378_cast_fp16")]; + tensor var_1380_equation_0 = const()[name = tensor("op_1380_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1380_cast_fp16 = einsum(equation = var_1380_equation_0, values = (var_1308_cast_fp16, var_1360_cast_fp16))[name = tensor("op_1380_cast_fp16")]; + tensor input_71_interleave_0 = const()[name = tensor("input_71_interleave_0"), val = tensor(false)]; + tensor input_71_cast_fp16 = concat(axis = var_288, interleave = input_71_interleave_0, values = (var_1362_cast_fp16, var_1364_cast_fp16, var_1366_cast_fp16, var_1368_cast_fp16, var_1370_cast_fp16, var_1372_cast_fp16, var_1374_cast_fp16, var_1376_cast_fp16, var_1378_cast_fp16, var_1380_cast_fp16))[name = tensor("input_71_cast_fp16")]; + tensor var_1386 = const()[name = tensor("op_1386"), val = tensor([1, 1])]; + tensor var_1388 = const()[name = tensor("op_1388"), val = tensor([1, 1])]; + tensor var_1390_pad_type_0 = const()[name = tensor("op_1390_pad_type_0"), val = tensor("custom")]; + tensor var_1390_pad_0 = const()[name = tensor("op_1390_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_0_transformer_blocks_1_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("down_blocks_1_attentions_0_transformer_blocks_1_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(68784384)))]; + tensor down_blocks_1_attentions_0_transformer_blocks_1_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_1_attentions_0_transformer_blocks_1_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(69603648)))]; + tensor var_1390_cast_fp16 = conv(bias = down_blocks_1_attentions_0_transformer_blocks_1_attn2_to_out_0_bias_to_fp16, dilations = var_1388, groups = var_288, pad = var_1390_pad_0, pad_type = var_1390_pad_type_0, strides = var_1386, weight = down_blocks_1_attentions_0_transformer_blocks_1_attn2_to_out_0_weight_to_fp16, x = input_71_cast_fp16)[name = tensor("op_1390_cast_fp16")]; + tensor inputs_11_cast_fp16 = add(x = var_1390_cast_fp16, y = inputs_9_cast_fp16)[name = tensor("inputs_11_cast_fp16")]; + tensor input_73_axes_0 = const()[name = tensor("input_73_axes_0"), val = tensor([1])]; + tensor input_73_gamma_0_to_fp16 = const()[name = tensor("input_73_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(69604992)))]; + tensor input_73_beta_0_to_fp16 = const()[name = tensor("input_73_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(69606336)))]; + tensor var_1400_to_fp16 = const()[name = tensor("op_1400_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_73_cast_fp16 = layer_norm(axes = input_73_axes_0, beta = input_73_beta_0_to_fp16, epsilon = var_1400_to_fp16, gamma = input_73_gamma_0_to_fp16, x = inputs_11_cast_fp16)[name = tensor("input_73_cast_fp16")]; + tensor var_1416 = const()[name = tensor("op_1416"), val = tensor([1, 1])]; + tensor var_1418 = const()[name = tensor("op_1418"), val = tensor([1, 1])]; + tensor var_1420_pad_type_0 = const()[name = tensor("op_1420_pad_type_0"), val = tensor("custom")]; + tensor var_1420_pad_0 = const()[name = tensor("op_1420_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_0_transformer_blocks_1_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("down_blocks_1_attentions_0_transformer_blocks_1_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(69607680)))]; + tensor down_blocks_1_attentions_0_transformer_blocks_1_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("down_blocks_1_attentions_0_transformer_blocks_1_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(76161344)))]; + tensor var_1420_cast_fp16 = conv(bias = down_blocks_1_attentions_0_transformer_blocks_1_ff_net_0_proj_bias_to_fp16, dilations = var_1418, groups = var_288, pad = var_1420_pad_0, pad_type = var_1420_pad_type_0, strides = var_1416, weight = down_blocks_1_attentions_0_transformer_blocks_1_ff_net_0_proj_weight_to_fp16, x = input_73_cast_fp16)[name = tensor("op_1420_cast_fp16")]; + tensor var_1421_split_sizes_0 = const()[name = tensor("op_1421_split_sizes_0"), val = tensor([2560, 2560])]; + tensor var_1421_axis_0 = const()[name = tensor("op_1421_axis_0"), val = tensor(1)]; + tensor var_1421_cast_fp16_0, tensor var_1421_cast_fp16_1 = split(axis = var_1421_axis_0, split_sizes = var_1421_split_sizes_0, x = var_1420_cast_fp16)[name = tensor("op_1421_cast_fp16")]; + tensor var_1423_mode_0 = const()[name = tensor("op_1423_mode_0"), val = tensor("EXACT")]; + tensor var_1423_cast_fp16 = gelu(mode = var_1423_mode_0, x = var_1421_cast_fp16_1)[name = tensor("op_1423_cast_fp16")]; + tensor input_75_cast_fp16 = mul(x = var_1421_cast_fp16_0, y = var_1423_cast_fp16)[name = tensor("input_75_cast_fp16")]; + tensor var_1427 = const()[name = tensor("op_1427"), val = tensor([1, 1])]; + tensor var_1429 = const()[name = tensor("op_1429"), val = tensor([1, 1])]; + tensor var_1431_pad_type_0 = const()[name = tensor("op_1431_pad_type_0"), val = tensor("custom")]; + tensor var_1431_pad_0 = const()[name = tensor("op_1431_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_0_transformer_blocks_1_ff_net_2_weight_to_fp16 = const()[name = tensor("down_blocks_1_attentions_0_transformer_blocks_1_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(76171648)))]; + tensor down_blocks_1_attentions_0_transformer_blocks_1_ff_net_2_bias_to_fp16 = const()[name = tensor("down_blocks_1_attentions_0_transformer_blocks_1_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(79448512)))]; + tensor var_1431_cast_fp16 = conv(bias = down_blocks_1_attentions_0_transformer_blocks_1_ff_net_2_bias_to_fp16, dilations = var_1429, groups = var_288, pad = var_1431_pad_0, pad_type = var_1431_pad_type_0, strides = var_1427, weight = down_blocks_1_attentions_0_transformer_blocks_1_ff_net_2_weight_to_fp16, x = input_75_cast_fp16)[name = tensor("op_1431_cast_fp16")]; + tensor hidden_states_29_cast_fp16 = add(x = var_1431_cast_fp16, y = inputs_11_cast_fp16)[name = tensor("hidden_states_29_cast_fp16")]; + tensor var_1433 = const()[name = tensor("op_1433"), val = tensor([2, 640, 64, 64])]; + tensor input_77_cast_fp16 = reshape(shape = var_1433, x = hidden_states_29_cast_fp16)[name = tensor("input_77_cast_fp16")]; + tensor var_1437 = const()[name = tensor("op_1437"), val = tensor([1, 1])]; + tensor var_1439 = const()[name = tensor("op_1439"), val = tensor([1, 1])]; + tensor hidden_states_31_pad_type_0 = const()[name = tensor("hidden_states_31_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_31_pad_0 = const()[name = tensor("hidden_states_31_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_0_proj_out_weight_to_fp16 = const()[name = tensor("down_blocks_1_attentions_0_proj_out_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(79449856)))]; + tensor down_blocks_1_attentions_0_proj_out_bias_to_fp16 = const()[name = tensor("down_blocks_1_attentions_0_proj_out_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(80269120)))]; + tensor hidden_states_31_cast_fp16 = conv(bias = down_blocks_1_attentions_0_proj_out_bias_to_fp16, dilations = var_1439, groups = var_288, pad = hidden_states_31_pad_0, pad_type = hidden_states_31_pad_type_0, strides = var_1437, weight = down_blocks_1_attentions_0_proj_out_weight_to_fp16, x = input_77_cast_fp16)[name = tensor("hidden_states_31_cast_fp16")]; + tensor input_79_cast_fp16 = add(x = hidden_states_31_cast_fp16, y = hidden_states_13_cast_fp16)[name = tensor("input_79_cast_fp16")]; + tensor reshape_28_shape_0 = const()[name = tensor("reshape_28_shape_0"), val = tensor([2, 32, 20, 64, 64])]; + tensor reshape_28_cast_fp16 = reshape(shape = reshape_28_shape_0, x = input_79_cast_fp16)[name = tensor("reshape_28_cast_fp16")]; + tensor reduce_mean_21_axes_0 = const()[name = tensor("reduce_mean_21_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_21_keep_dims_0 = const()[name = tensor("reduce_mean_21_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_21_cast_fp16 = reduce_mean(axes = reduce_mean_21_axes_0, keep_dims = reduce_mean_21_keep_dims_0, x = reshape_28_cast_fp16)[name = tensor("reduce_mean_21_cast_fp16")]; + tensor sub_14_cast_fp16 = sub(x = reshape_28_cast_fp16, y = reduce_mean_21_cast_fp16)[name = tensor("sub_14_cast_fp16")]; + tensor square_7_cast_fp16 = square(x = sub_14_cast_fp16)[name = tensor("square_7_cast_fp16")]; + tensor reduce_mean_23_axes_0 = const()[name = tensor("reduce_mean_23_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_23_keep_dims_0 = const()[name = tensor("reduce_mean_23_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_23_cast_fp16 = reduce_mean(axes = reduce_mean_23_axes_0, keep_dims = reduce_mean_23_keep_dims_0, x = square_7_cast_fp16)[name = tensor("reduce_mean_23_cast_fp16")]; + tensor add_14_y_0_to_fp16 = const()[name = tensor("add_14_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_14_cast_fp16 = add(x = reduce_mean_23_cast_fp16, y = add_14_y_0_to_fp16)[name = tensor("add_14_cast_fp16")]; + tensor sqrt_7_cast_fp16 = sqrt(x = add_14_cast_fp16)[name = tensor("sqrt_7_cast_fp16")]; + tensor real_div_7_cast_fp16 = real_div(x = sub_14_cast_fp16, y = sqrt_7_cast_fp16)[name = tensor("real_div_7_cast_fp16")]; + tensor reshape_29_shape_0 = const()[name = tensor("reshape_29_shape_0"), val = tensor([2, 640, 64, 64])]; + tensor reshape_29_cast_fp16 = reshape(shape = reshape_29_shape_0, x = real_div_7_cast_fp16)[name = tensor("reshape_29_cast_fp16")]; + tensor add_15_gamma_0_to_fp16 = const()[name = tensor("add_15_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(80270464)))]; + tensor add_15_beta_0_to_fp16 = const()[name = tensor("add_15_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(80271808)))]; + tensor add_15_epsilon_0_to_fp16 = const()[name = tensor("add_15_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_15_cast_fp16 = batch_norm(beta = add_15_beta_0_to_fp16, epsilon = add_15_epsilon_0_to_fp16, gamma = add_15_gamma_0_to_fp16, mean = add_11_mean_0_to_fp16, variance = add_11_variance_0_to_fp16, x = reshape_29_cast_fp16)[name = tensor("add_15_cast_fp16")]; + tensor input_83_cast_fp16 = silu(x = add_15_cast_fp16)[name = tensor("input_83_cast_fp16")]; + tensor var_1454 = const()[name = tensor("op_1454"), val = tensor([1, 1])]; + tensor var_1456 = const()[name = tensor("op_1456"), val = tensor([1, 1])]; + tensor hidden_states_33_pad_type_0 = const()[name = tensor("hidden_states_33_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_33_pad_0 = const()[name = tensor("hidden_states_33_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor down_blocks_1_resnets_1_conv1_weight_to_fp16 = const()[name = tensor("down_blocks_1_resnets_1_conv1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(80273152)))]; + tensor down_blocks_1_resnets_1_conv1_bias_to_fp16 = const()[name = tensor("down_blocks_1_resnets_1_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(87646016)))]; + tensor hidden_states_33_cast_fp16 = conv(bias = down_blocks_1_resnets_1_conv1_bias_to_fp16, dilations = var_1456, groups = var_288, pad = hidden_states_33_pad_0, pad_type = hidden_states_33_pad_type_0, strides = var_1454, weight = down_blocks_1_resnets_1_conv1_weight_to_fp16, x = input_83_cast_fp16)[name = tensor("hidden_states_33_cast_fp16")]; + tensor var_1462 = const()[name = tensor("op_1462"), val = tensor([1, 1])]; + tensor var_1464 = const()[name = tensor("op_1464"), val = tensor([1, 1])]; + tensor temb_7_pad_type_0 = const()[name = tensor("temb_7_pad_type_0"), val = tensor("custom")]; + tensor temb_7_pad_0 = const()[name = tensor("temb_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_resnets_1_time_emb_proj_weight_to_fp16 = const()[name = tensor("down_blocks_1_resnets_1_time_emb_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(87647360)))]; + tensor down_blocks_1_resnets_1_time_emb_proj_bias_to_fp16 = const()[name = tensor("down_blocks_1_resnets_1_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(89285824)))]; + tensor temb_7_cast_fp16 = conv(bias = down_blocks_1_resnets_1_time_emb_proj_bias_to_fp16, dilations = var_1464, groups = var_288, pad = temb_7_pad_0, pad_type = temb_7_pad_type_0, strides = var_1462, weight = down_blocks_1_resnets_1_time_emb_proj_weight_to_fp16, x = input_21_cast_fp16)[name = tensor("temb_7_cast_fp16")]; + tensor input_87_cast_fp16 = add(x = hidden_states_33_cast_fp16, y = temb_7_cast_fp16)[name = tensor("input_87_cast_fp16")]; + tensor reshape_32_shape_0 = const()[name = tensor("reshape_32_shape_0"), val = tensor([2, 32, 20, 64, 64])]; + tensor reshape_32_cast_fp16 = reshape(shape = reshape_32_shape_0, x = input_87_cast_fp16)[name = tensor("reshape_32_cast_fp16")]; + tensor reduce_mean_24_axes_0 = const()[name = tensor("reduce_mean_24_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_24_keep_dims_0 = const()[name = tensor("reduce_mean_24_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_24_cast_fp16 = reduce_mean(axes = reduce_mean_24_axes_0, keep_dims = reduce_mean_24_keep_dims_0, x = reshape_32_cast_fp16)[name = tensor("reduce_mean_24_cast_fp16")]; + tensor sub_16_cast_fp16 = sub(x = reshape_32_cast_fp16, y = reduce_mean_24_cast_fp16)[name = tensor("sub_16_cast_fp16")]; + tensor square_8_cast_fp16 = square(x = sub_16_cast_fp16)[name = tensor("square_8_cast_fp16")]; + tensor reduce_mean_26_axes_0 = const()[name = tensor("reduce_mean_26_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_26_keep_dims_0 = const()[name = tensor("reduce_mean_26_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_26_cast_fp16 = reduce_mean(axes = reduce_mean_26_axes_0, keep_dims = reduce_mean_26_keep_dims_0, x = square_8_cast_fp16)[name = tensor("reduce_mean_26_cast_fp16")]; + tensor add_16_y_0_to_fp16 = const()[name = tensor("add_16_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_16_cast_fp16 = add(x = reduce_mean_26_cast_fp16, y = add_16_y_0_to_fp16)[name = tensor("add_16_cast_fp16")]; + tensor sqrt_8_cast_fp16 = sqrt(x = add_16_cast_fp16)[name = tensor("sqrt_8_cast_fp16")]; + tensor real_div_8_cast_fp16 = real_div(x = sub_16_cast_fp16, y = sqrt_8_cast_fp16)[name = tensor("real_div_8_cast_fp16")]; + tensor reshape_33_shape_0 = const()[name = tensor("reshape_33_shape_0"), val = tensor([2, 640, 64, 64])]; + tensor reshape_33_cast_fp16 = reshape(shape = reshape_33_shape_0, x = real_div_8_cast_fp16)[name = tensor("reshape_33_cast_fp16")]; + tensor add_17_gamma_0_to_fp16 = const()[name = tensor("add_17_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(89287168)))]; + tensor add_17_beta_0_to_fp16 = const()[name = tensor("add_17_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(89288512)))]; + tensor add_17_epsilon_0_to_fp16 = const()[name = tensor("add_17_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_17_cast_fp16 = batch_norm(beta = add_17_beta_0_to_fp16, epsilon = add_17_epsilon_0_to_fp16, gamma = add_17_gamma_0_to_fp16, mean = add_11_mean_0_to_fp16, variance = add_11_variance_0_to_fp16, x = reshape_33_cast_fp16)[name = tensor("add_17_cast_fp16")]; + tensor input_91_cast_fp16 = silu(x = add_17_cast_fp16)[name = tensor("input_91_cast_fp16")]; + tensor var_1474 = const()[name = tensor("op_1474"), val = tensor([1, 1])]; + tensor var_1476 = const()[name = tensor("op_1476"), val = tensor([1, 1])]; + tensor hidden_states_35_pad_type_0 = const()[name = tensor("hidden_states_35_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_35_pad_0 = const()[name = tensor("hidden_states_35_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor down_blocks_1_resnets_1_conv2_weight_to_fp16 = const()[name = tensor("down_blocks_1_resnets_1_conv2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(89289856)))]; + tensor down_blocks_1_resnets_1_conv2_bias_to_fp16 = const()[name = tensor("down_blocks_1_resnets_1_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(96662720)))]; + tensor hidden_states_35_cast_fp16 = conv(bias = down_blocks_1_resnets_1_conv2_bias_to_fp16, dilations = var_1476, groups = var_288, pad = hidden_states_35_pad_0, pad_type = hidden_states_35_pad_type_0, strides = var_1474, weight = down_blocks_1_resnets_1_conv2_weight_to_fp16, x = input_91_cast_fp16)[name = tensor("hidden_states_35_cast_fp16")]; + tensor hidden_states_37_cast_fp16 = add(x = input_79_cast_fp16, y = hidden_states_35_cast_fp16)[name = tensor("hidden_states_37_cast_fp16")]; + tensor reshape_36_shape_0 = const()[name = tensor("reshape_36_shape_0"), val = tensor([2, 32, 20, 64, 64])]; + tensor reshape_36_cast_fp16 = reshape(shape = reshape_36_shape_0, x = hidden_states_37_cast_fp16)[name = tensor("reshape_36_cast_fp16")]; + tensor reduce_mean_27_axes_0 = const()[name = tensor("reduce_mean_27_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_27_keep_dims_0 = const()[name = tensor("reduce_mean_27_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_27_cast_fp16 = reduce_mean(axes = reduce_mean_27_axes_0, keep_dims = reduce_mean_27_keep_dims_0, x = reshape_36_cast_fp16)[name = tensor("reduce_mean_27_cast_fp16")]; + tensor sub_18_cast_fp16 = sub(x = reshape_36_cast_fp16, y = reduce_mean_27_cast_fp16)[name = tensor("sub_18_cast_fp16")]; + tensor square_9_cast_fp16 = square(x = sub_18_cast_fp16)[name = tensor("square_9_cast_fp16")]; + tensor reduce_mean_29_axes_0 = const()[name = tensor("reduce_mean_29_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_29_keep_dims_0 = const()[name = tensor("reduce_mean_29_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_29_cast_fp16 = reduce_mean(axes = reduce_mean_29_axes_0, keep_dims = reduce_mean_29_keep_dims_0, x = square_9_cast_fp16)[name = tensor("reduce_mean_29_cast_fp16")]; + tensor add_18_y_0_to_fp16 = const()[name = tensor("add_18_y_0_to_fp16"), val = tensor(0x1.1p-20)]; + tensor add_18_cast_fp16 = add(x = reduce_mean_29_cast_fp16, y = add_18_y_0_to_fp16)[name = tensor("add_18_cast_fp16")]; + tensor sqrt_9_cast_fp16 = sqrt(x = add_18_cast_fp16)[name = tensor("sqrt_9_cast_fp16")]; + tensor real_div_9_cast_fp16 = real_div(x = sub_18_cast_fp16, y = sqrt_9_cast_fp16)[name = tensor("real_div_9_cast_fp16")]; + tensor reshape_37_shape_0 = const()[name = tensor("reshape_37_shape_0"), val = tensor([2, 640, 64, 64])]; + tensor reshape_37_cast_fp16 = reshape(shape = reshape_37_shape_0, x = real_div_9_cast_fp16)[name = tensor("reshape_37_cast_fp16")]; + tensor add_19_gamma_0_to_fp16 = const()[name = tensor("add_19_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(96664064)))]; + tensor add_19_beta_0_to_fp16 = const()[name = tensor("add_19_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(96665408)))]; + tensor add_19_epsilon_0_to_fp16 = const()[name = tensor("add_19_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_19_cast_fp16 = batch_norm(beta = add_19_beta_0_to_fp16, epsilon = add_19_epsilon_0_to_fp16, gamma = add_19_gamma_0_to_fp16, mean = add_11_mean_0_to_fp16, variance = add_11_variance_0_to_fp16, x = reshape_37_cast_fp16)[name = tensor("add_19_cast_fp16")]; + tensor var_1498 = const()[name = tensor("op_1498"), val = tensor([1, 1])]; + tensor var_1500 = const()[name = tensor("op_1500"), val = tensor([1, 1])]; + tensor hidden_states_39_pad_type_0 = const()[name = tensor("hidden_states_39_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_39_pad_0 = const()[name = tensor("hidden_states_39_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_1_proj_in_weight_to_fp16 = const()[name = tensor("down_blocks_1_attentions_1_proj_in_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(96666752)))]; + tensor down_blocks_1_attentions_1_proj_in_bias_to_fp16 = const()[name = tensor("down_blocks_1_attentions_1_proj_in_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(97486016)))]; + tensor hidden_states_39_cast_fp16 = conv(bias = down_blocks_1_attentions_1_proj_in_bias_to_fp16, dilations = var_1500, groups = var_288, pad = hidden_states_39_pad_0, pad_type = hidden_states_39_pad_type_0, strides = var_1498, weight = down_blocks_1_attentions_1_proj_in_weight_to_fp16, x = add_19_cast_fp16)[name = tensor("hidden_states_39_cast_fp16")]; + tensor var_1505 = const()[name = tensor("op_1505"), val = tensor([2, 640, 1, 4096])]; + tensor inputs_13_cast_fp16 = reshape(shape = var_1505, x = hidden_states_39_cast_fp16)[name = tensor("inputs_13_cast_fp16")]; + tensor hidden_states_41_axes_0 = const()[name = tensor("hidden_states_41_axes_0"), val = tensor([1])]; + tensor hidden_states_41_gamma_0_to_fp16 = const()[name = tensor("hidden_states_41_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(97487360)))]; + tensor hidden_states_41_beta_0_to_fp16 = const()[name = tensor("hidden_states_41_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(97488704)))]; + tensor var_1521_to_fp16 = const()[name = tensor("op_1521_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_41_cast_fp16 = layer_norm(axes = hidden_states_41_axes_0, beta = hidden_states_41_beta_0_to_fp16, epsilon = var_1521_to_fp16, gamma = hidden_states_41_gamma_0_to_fp16, x = inputs_13_cast_fp16)[name = tensor("hidden_states_41_cast_fp16")]; + tensor var_1536 = const()[name = tensor("op_1536"), val = tensor([1, 1])]; + tensor var_1538 = const()[name = tensor("op_1538"), val = tensor([1, 1])]; + tensor q_9_pad_type_0 = const()[name = tensor("q_9_pad_type_0"), val = tensor("custom")]; + tensor q_9_pad_0 = const()[name = tensor("q_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16 = const()[name = tensor("down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(97490048)))]; + tensor q_9_cast_fp16 = conv(dilations = var_1538, groups = var_288, pad = q_9_pad_0, pad_type = q_9_pad_type_0, strides = var_1536, weight = down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16, x = hidden_states_41_cast_fp16)[name = tensor("q_9_cast_fp16")]; + tensor var_1542 = const()[name = tensor("op_1542"), val = tensor([1, 1])]; + tensor var_1544 = const()[name = tensor("op_1544"), val = tensor([1, 1])]; + tensor k_17_pad_type_0 = const()[name = tensor("k_17_pad_type_0"), val = tensor("custom")]; + tensor k_17_pad_0 = const()[name = tensor("k_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16 = const()[name = tensor("down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(98309312)))]; + tensor k_17_cast_fp16 = conv(dilations = var_1544, groups = var_288, pad = k_17_pad_0, pad_type = k_17_pad_type_0, strides = var_1542, weight = down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16, x = hidden_states_41_cast_fp16)[name = tensor("k_17_cast_fp16")]; + tensor var_1548 = const()[name = tensor("op_1548"), val = tensor([1, 1])]; + tensor var_1550 = const()[name = tensor("op_1550"), val = tensor([1, 1])]; + tensor v_9_pad_type_0 = const()[name = tensor("v_9_pad_type_0"), val = tensor("custom")]; + tensor v_9_pad_0 = const()[name = tensor("v_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16 = const()[name = tensor("down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(99128576)))]; + tensor v_9_cast_fp16 = conv(dilations = var_1550, groups = var_288, pad = v_9_pad_0, pad_type = v_9_pad_type_0, strides = var_1548, weight = down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16, x = hidden_states_41_cast_fp16)[name = tensor("v_9_cast_fp16")]; + tensor var_1554_begin_0 = const()[name = tensor("op_1554_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1554_end_0 = const()[name = tensor("op_1554_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_1554_end_mask_0 = const()[name = tensor("op_1554_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1554_cast_fp16 = slice_by_index(begin = var_1554_begin_0, end = var_1554_end_0, end_mask = var_1554_end_mask_0, x = q_9_cast_fp16)[name = tensor("op_1554_cast_fp16")]; + tensor var_1558_begin_0 = const()[name = tensor("op_1558_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_1558_end_0 = const()[name = tensor("op_1558_end_0"), val = tensor([2, 128, 1, 4096])]; + tensor var_1558_end_mask_0 = const()[name = tensor("op_1558_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1558_cast_fp16 = slice_by_index(begin = var_1558_begin_0, end = var_1558_end_0, end_mask = var_1558_end_mask_0, x = q_9_cast_fp16)[name = tensor("op_1558_cast_fp16")]; + tensor var_1562_begin_0 = const()[name = tensor("op_1562_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_1562_end_0 = const()[name = tensor("op_1562_end_0"), val = tensor([2, 192, 1, 4096])]; + tensor var_1562_end_mask_0 = const()[name = tensor("op_1562_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1562_cast_fp16 = slice_by_index(begin = var_1562_begin_0, end = var_1562_end_0, end_mask = var_1562_end_mask_0, x = q_9_cast_fp16)[name = tensor("op_1562_cast_fp16")]; + tensor var_1566_begin_0 = const()[name = tensor("op_1566_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_1566_end_0 = const()[name = tensor("op_1566_end_0"), val = tensor([2, 256, 1, 4096])]; + tensor var_1566_end_mask_0 = const()[name = tensor("op_1566_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1566_cast_fp16 = slice_by_index(begin = var_1566_begin_0, end = var_1566_end_0, end_mask = var_1566_end_mask_0, x = q_9_cast_fp16)[name = tensor("op_1566_cast_fp16")]; + tensor var_1570_begin_0 = const()[name = tensor("op_1570_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_1570_end_0 = const()[name = tensor("op_1570_end_0"), val = tensor([2, 320, 1, 4096])]; + tensor var_1570_end_mask_0 = const()[name = tensor("op_1570_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1570_cast_fp16 = slice_by_index(begin = var_1570_begin_0, end = var_1570_end_0, end_mask = var_1570_end_mask_0, x = q_9_cast_fp16)[name = tensor("op_1570_cast_fp16")]; + tensor var_1574_begin_0 = const()[name = tensor("op_1574_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_1574_end_0 = const()[name = tensor("op_1574_end_0"), val = tensor([2, 384, 1, 4096])]; + tensor var_1574_end_mask_0 = const()[name = tensor("op_1574_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1574_cast_fp16 = slice_by_index(begin = var_1574_begin_0, end = var_1574_end_0, end_mask = var_1574_end_mask_0, x = q_9_cast_fp16)[name = tensor("op_1574_cast_fp16")]; + tensor var_1578_begin_0 = const()[name = tensor("op_1578_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_1578_end_0 = const()[name = tensor("op_1578_end_0"), val = tensor([2, 448, 1, 4096])]; + tensor var_1578_end_mask_0 = const()[name = tensor("op_1578_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1578_cast_fp16 = slice_by_index(begin = var_1578_begin_0, end = var_1578_end_0, end_mask = var_1578_end_mask_0, x = q_9_cast_fp16)[name = tensor("op_1578_cast_fp16")]; + tensor var_1582_begin_0 = const()[name = tensor("op_1582_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_1582_end_0 = const()[name = tensor("op_1582_end_0"), val = tensor([2, 512, 1, 4096])]; + tensor var_1582_end_mask_0 = const()[name = tensor("op_1582_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1582_cast_fp16 = slice_by_index(begin = var_1582_begin_0, end = var_1582_end_0, end_mask = var_1582_end_mask_0, x = q_9_cast_fp16)[name = tensor("op_1582_cast_fp16")]; + tensor var_1586_begin_0 = const()[name = tensor("op_1586_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_1586_end_0 = const()[name = tensor("op_1586_end_0"), val = tensor([2, 576, 1, 4096])]; + tensor var_1586_end_mask_0 = const()[name = tensor("op_1586_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1586_cast_fp16 = slice_by_index(begin = var_1586_begin_0, end = var_1586_end_0, end_mask = var_1586_end_mask_0, x = q_9_cast_fp16)[name = tensor("op_1586_cast_fp16")]; + tensor var_1590_begin_0 = const()[name = tensor("op_1590_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_1590_end_0 = const()[name = tensor("op_1590_end_0"), val = tensor([2, 640, 1, 4096])]; + tensor var_1590_end_mask_0 = const()[name = tensor("op_1590_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1590_cast_fp16 = slice_by_index(begin = var_1590_begin_0, end = var_1590_end_0, end_mask = var_1590_end_mask_0, x = q_9_cast_fp16)[name = tensor("op_1590_cast_fp16")]; + tensor k_19_perm_0 = const()[name = tensor("k_19_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_1597_begin_0 = const()[name = tensor("op_1597_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1597_end_0 = const()[name = tensor("op_1597_end_0"), val = tensor([2, 4096, 1, 64])]; + tensor var_1597_end_mask_0 = const()[name = tensor("op_1597_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_19_cast_fp16 = transpose(perm = k_19_perm_0, x = k_17_cast_fp16)[name = tensor("transpose_135")]; + tensor var_1597_cast_fp16 = slice_by_index(begin = var_1597_begin_0, end = var_1597_end_0, end_mask = var_1597_end_mask_0, x = k_19_cast_fp16)[name = tensor("op_1597_cast_fp16")]; + tensor var_1601_begin_0 = const()[name = tensor("op_1601_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_1601_end_0 = const()[name = tensor("op_1601_end_0"), val = tensor([2, 4096, 1, 128])]; + tensor var_1601_end_mask_0 = const()[name = tensor("op_1601_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1601_cast_fp16 = slice_by_index(begin = var_1601_begin_0, end = var_1601_end_0, end_mask = var_1601_end_mask_0, x = k_19_cast_fp16)[name = tensor("op_1601_cast_fp16")]; + tensor var_1605_begin_0 = const()[name = tensor("op_1605_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_1605_end_0 = const()[name = tensor("op_1605_end_0"), val = tensor([2, 4096, 1, 192])]; + tensor var_1605_end_mask_0 = const()[name = tensor("op_1605_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1605_cast_fp16 = slice_by_index(begin = var_1605_begin_0, end = var_1605_end_0, end_mask = var_1605_end_mask_0, x = k_19_cast_fp16)[name = tensor("op_1605_cast_fp16")]; + tensor var_1609_begin_0 = const()[name = tensor("op_1609_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_1609_end_0 = const()[name = tensor("op_1609_end_0"), val = tensor([2, 4096, 1, 256])]; + tensor var_1609_end_mask_0 = const()[name = tensor("op_1609_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1609_cast_fp16 = slice_by_index(begin = var_1609_begin_0, end = var_1609_end_0, end_mask = var_1609_end_mask_0, x = k_19_cast_fp16)[name = tensor("op_1609_cast_fp16")]; + tensor var_1613_begin_0 = const()[name = tensor("op_1613_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_1613_end_0 = const()[name = tensor("op_1613_end_0"), val = tensor([2, 4096, 1, 320])]; + tensor var_1613_end_mask_0 = const()[name = tensor("op_1613_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1613_cast_fp16 = slice_by_index(begin = var_1613_begin_0, end = var_1613_end_0, end_mask = var_1613_end_mask_0, x = k_19_cast_fp16)[name = tensor("op_1613_cast_fp16")]; + tensor var_1617_begin_0 = const()[name = tensor("op_1617_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_1617_end_0 = const()[name = tensor("op_1617_end_0"), val = tensor([2, 4096, 1, 384])]; + tensor var_1617_end_mask_0 = const()[name = tensor("op_1617_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1617_cast_fp16 = slice_by_index(begin = var_1617_begin_0, end = var_1617_end_0, end_mask = var_1617_end_mask_0, x = k_19_cast_fp16)[name = tensor("op_1617_cast_fp16")]; + tensor var_1621_begin_0 = const()[name = tensor("op_1621_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_1621_end_0 = const()[name = tensor("op_1621_end_0"), val = tensor([2, 4096, 1, 448])]; + tensor var_1621_end_mask_0 = const()[name = tensor("op_1621_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1621_cast_fp16 = slice_by_index(begin = var_1621_begin_0, end = var_1621_end_0, end_mask = var_1621_end_mask_0, x = k_19_cast_fp16)[name = tensor("op_1621_cast_fp16")]; + tensor var_1625_begin_0 = const()[name = tensor("op_1625_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_1625_end_0 = const()[name = tensor("op_1625_end_0"), val = tensor([2, 4096, 1, 512])]; + tensor var_1625_end_mask_0 = const()[name = tensor("op_1625_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1625_cast_fp16 = slice_by_index(begin = var_1625_begin_0, end = var_1625_end_0, end_mask = var_1625_end_mask_0, x = k_19_cast_fp16)[name = tensor("op_1625_cast_fp16")]; + tensor var_1629_begin_0 = const()[name = tensor("op_1629_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_1629_end_0 = const()[name = tensor("op_1629_end_0"), val = tensor([2, 4096, 1, 576])]; + tensor var_1629_end_mask_0 = const()[name = tensor("op_1629_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1629_cast_fp16 = slice_by_index(begin = var_1629_begin_0, end = var_1629_end_0, end_mask = var_1629_end_mask_0, x = k_19_cast_fp16)[name = tensor("op_1629_cast_fp16")]; + tensor var_1633_begin_0 = const()[name = tensor("op_1633_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_1633_end_0 = const()[name = tensor("op_1633_end_0"), val = tensor([2, 4096, 1, 640])]; + tensor var_1633_end_mask_0 = const()[name = tensor("op_1633_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1633_cast_fp16 = slice_by_index(begin = var_1633_begin_0, end = var_1633_end_0, end_mask = var_1633_end_mask_0, x = k_19_cast_fp16)[name = tensor("op_1633_cast_fp16")]; + tensor var_1635_begin_0 = const()[name = tensor("op_1635_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1635_end_0 = const()[name = tensor("op_1635_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_1635_end_mask_0 = const()[name = tensor("op_1635_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1635_cast_fp16 = slice_by_index(begin = var_1635_begin_0, end = var_1635_end_0, end_mask = var_1635_end_mask_0, x = v_9_cast_fp16)[name = tensor("op_1635_cast_fp16")]; + tensor var_1639_begin_0 = const()[name = tensor("op_1639_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_1639_end_0 = const()[name = tensor("op_1639_end_0"), val = tensor([2, 128, 1, 4096])]; + tensor var_1639_end_mask_0 = const()[name = tensor("op_1639_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1639_cast_fp16 = slice_by_index(begin = var_1639_begin_0, end = var_1639_end_0, end_mask = var_1639_end_mask_0, x = v_9_cast_fp16)[name = tensor("op_1639_cast_fp16")]; + tensor var_1643_begin_0 = const()[name = tensor("op_1643_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_1643_end_0 = const()[name = tensor("op_1643_end_0"), val = tensor([2, 192, 1, 4096])]; + tensor var_1643_end_mask_0 = const()[name = tensor("op_1643_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1643_cast_fp16 = slice_by_index(begin = var_1643_begin_0, end = var_1643_end_0, end_mask = var_1643_end_mask_0, x = v_9_cast_fp16)[name = tensor("op_1643_cast_fp16")]; + tensor var_1647_begin_0 = const()[name = tensor("op_1647_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_1647_end_0 = const()[name = tensor("op_1647_end_0"), val = tensor([2, 256, 1, 4096])]; + tensor var_1647_end_mask_0 = const()[name = tensor("op_1647_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1647_cast_fp16 = slice_by_index(begin = var_1647_begin_0, end = var_1647_end_0, end_mask = var_1647_end_mask_0, x = v_9_cast_fp16)[name = tensor("op_1647_cast_fp16")]; + tensor var_1651_begin_0 = const()[name = tensor("op_1651_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_1651_end_0 = const()[name = tensor("op_1651_end_0"), val = tensor([2, 320, 1, 4096])]; + tensor var_1651_end_mask_0 = const()[name = tensor("op_1651_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1651_cast_fp16 = slice_by_index(begin = var_1651_begin_0, end = var_1651_end_0, end_mask = var_1651_end_mask_0, x = v_9_cast_fp16)[name = tensor("op_1651_cast_fp16")]; + tensor var_1655_begin_0 = const()[name = tensor("op_1655_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_1655_end_0 = const()[name = tensor("op_1655_end_0"), val = tensor([2, 384, 1, 4096])]; + tensor var_1655_end_mask_0 = const()[name = tensor("op_1655_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1655_cast_fp16 = slice_by_index(begin = var_1655_begin_0, end = var_1655_end_0, end_mask = var_1655_end_mask_0, x = v_9_cast_fp16)[name = tensor("op_1655_cast_fp16")]; + tensor var_1659_begin_0 = const()[name = tensor("op_1659_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_1659_end_0 = const()[name = tensor("op_1659_end_0"), val = tensor([2, 448, 1, 4096])]; + tensor var_1659_end_mask_0 = const()[name = tensor("op_1659_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1659_cast_fp16 = slice_by_index(begin = var_1659_begin_0, end = var_1659_end_0, end_mask = var_1659_end_mask_0, x = v_9_cast_fp16)[name = tensor("op_1659_cast_fp16")]; + tensor var_1663_begin_0 = const()[name = tensor("op_1663_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_1663_end_0 = const()[name = tensor("op_1663_end_0"), val = tensor([2, 512, 1, 4096])]; + tensor var_1663_end_mask_0 = const()[name = tensor("op_1663_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1663_cast_fp16 = slice_by_index(begin = var_1663_begin_0, end = var_1663_end_0, end_mask = var_1663_end_mask_0, x = v_9_cast_fp16)[name = tensor("op_1663_cast_fp16")]; + tensor var_1667_begin_0 = const()[name = tensor("op_1667_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_1667_end_0 = const()[name = tensor("op_1667_end_0"), val = tensor([2, 576, 1, 4096])]; + tensor var_1667_end_mask_0 = const()[name = tensor("op_1667_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1667_cast_fp16 = slice_by_index(begin = var_1667_begin_0, end = var_1667_end_0, end_mask = var_1667_end_mask_0, x = v_9_cast_fp16)[name = tensor("op_1667_cast_fp16")]; + tensor var_1671_begin_0 = const()[name = tensor("op_1671_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_1671_end_0 = const()[name = tensor("op_1671_end_0"), val = tensor([2, 640, 1, 4096])]; + tensor var_1671_end_mask_0 = const()[name = tensor("op_1671_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1671_cast_fp16 = slice_by_index(begin = var_1671_begin_0, end = var_1671_end_0, end_mask = var_1671_end_mask_0, x = v_9_cast_fp16)[name = tensor("op_1671_cast_fp16")]; + tensor var_1675_equation_0 = const()[name = tensor("op_1675_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1675_cast_fp16 = einsum(equation = var_1675_equation_0, values = (var_1597_cast_fp16, var_1554_cast_fp16))[name = tensor("op_1675_cast_fp16")]; + tensor var_1676_to_fp16 = const()[name = tensor("op_1676_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_81_cast_fp16 = mul(x = var_1675_cast_fp16, y = var_1676_to_fp16)[name = tensor("aw_81_cast_fp16")]; + tensor var_1679_equation_0 = const()[name = tensor("op_1679_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1679_cast_fp16 = einsum(equation = var_1679_equation_0, values = (var_1601_cast_fp16, var_1558_cast_fp16))[name = tensor("op_1679_cast_fp16")]; + tensor var_1680_to_fp16 = const()[name = tensor("op_1680_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_83_cast_fp16 = mul(x = var_1679_cast_fp16, y = var_1680_to_fp16)[name = tensor("aw_83_cast_fp16")]; + tensor var_1683_equation_0 = const()[name = tensor("op_1683_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1683_cast_fp16 = einsum(equation = var_1683_equation_0, values = (var_1605_cast_fp16, var_1562_cast_fp16))[name = tensor("op_1683_cast_fp16")]; + tensor var_1684_to_fp16 = const()[name = tensor("op_1684_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_85_cast_fp16 = mul(x = var_1683_cast_fp16, y = var_1684_to_fp16)[name = tensor("aw_85_cast_fp16")]; + tensor var_1687_equation_0 = const()[name = tensor("op_1687_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1687_cast_fp16 = einsum(equation = var_1687_equation_0, values = (var_1609_cast_fp16, var_1566_cast_fp16))[name = tensor("op_1687_cast_fp16")]; + tensor var_1688_to_fp16 = const()[name = tensor("op_1688_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_87_cast_fp16 = mul(x = var_1687_cast_fp16, y = var_1688_to_fp16)[name = tensor("aw_87_cast_fp16")]; + tensor var_1691_equation_0 = const()[name = tensor("op_1691_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1691_cast_fp16 = einsum(equation = var_1691_equation_0, values = (var_1613_cast_fp16, var_1570_cast_fp16))[name = tensor("op_1691_cast_fp16")]; + tensor var_1692_to_fp16 = const()[name = tensor("op_1692_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_89_cast_fp16 = mul(x = var_1691_cast_fp16, y = var_1692_to_fp16)[name = tensor("aw_89_cast_fp16")]; + tensor var_1695_equation_0 = const()[name = tensor("op_1695_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1695_cast_fp16 = einsum(equation = var_1695_equation_0, values = (var_1617_cast_fp16, var_1574_cast_fp16))[name = tensor("op_1695_cast_fp16")]; + tensor var_1696_to_fp16 = const()[name = tensor("op_1696_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_91_cast_fp16 = mul(x = var_1695_cast_fp16, y = var_1696_to_fp16)[name = tensor("aw_91_cast_fp16")]; + tensor var_1699_equation_0 = const()[name = tensor("op_1699_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1699_cast_fp16 = einsum(equation = var_1699_equation_0, values = (var_1621_cast_fp16, var_1578_cast_fp16))[name = tensor("op_1699_cast_fp16")]; + tensor var_1700_to_fp16 = const()[name = tensor("op_1700_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_93_cast_fp16 = mul(x = var_1699_cast_fp16, y = var_1700_to_fp16)[name = tensor("aw_93_cast_fp16")]; + tensor var_1703_equation_0 = const()[name = tensor("op_1703_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1703_cast_fp16 = einsum(equation = var_1703_equation_0, values = (var_1625_cast_fp16, var_1582_cast_fp16))[name = tensor("op_1703_cast_fp16")]; + tensor var_1704_to_fp16 = const()[name = tensor("op_1704_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_95_cast_fp16 = mul(x = var_1703_cast_fp16, y = var_1704_to_fp16)[name = tensor("aw_95_cast_fp16")]; + tensor var_1707_equation_0 = const()[name = tensor("op_1707_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1707_cast_fp16 = einsum(equation = var_1707_equation_0, values = (var_1629_cast_fp16, var_1586_cast_fp16))[name = tensor("op_1707_cast_fp16")]; + tensor var_1708_to_fp16 = const()[name = tensor("op_1708_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_97_cast_fp16 = mul(x = var_1707_cast_fp16, y = var_1708_to_fp16)[name = tensor("aw_97_cast_fp16")]; + tensor var_1711_equation_0 = const()[name = tensor("op_1711_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1711_cast_fp16 = einsum(equation = var_1711_equation_0, values = (var_1633_cast_fp16, var_1590_cast_fp16))[name = tensor("op_1711_cast_fp16")]; + tensor var_1712_to_fp16 = const()[name = tensor("op_1712_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_99_cast_fp16 = mul(x = var_1711_cast_fp16, y = var_1712_to_fp16)[name = tensor("aw_99_cast_fp16")]; + tensor var_1714_cast_fp16 = softmax(axis = var_288, x = aw_81_cast_fp16)[name = tensor("op_1714_cast_fp16")]; + tensor var_1715_cast_fp16 = softmax(axis = var_288, x = aw_83_cast_fp16)[name = tensor("op_1715_cast_fp16")]; + tensor var_1716_cast_fp16 = softmax(axis = var_288, x = aw_85_cast_fp16)[name = tensor("op_1716_cast_fp16")]; + tensor var_1717_cast_fp16 = softmax(axis = var_288, x = aw_87_cast_fp16)[name = tensor("op_1717_cast_fp16")]; + tensor var_1718_cast_fp16 = softmax(axis = var_288, x = aw_89_cast_fp16)[name = tensor("op_1718_cast_fp16")]; + tensor var_1719_cast_fp16 = softmax(axis = var_288, x = aw_91_cast_fp16)[name = tensor("op_1719_cast_fp16")]; + tensor var_1720_cast_fp16 = softmax(axis = var_288, x = aw_93_cast_fp16)[name = tensor("op_1720_cast_fp16")]; + tensor var_1721_cast_fp16 = softmax(axis = var_288, x = aw_95_cast_fp16)[name = tensor("op_1721_cast_fp16")]; + tensor var_1722_cast_fp16 = softmax(axis = var_288, x = aw_97_cast_fp16)[name = tensor("op_1722_cast_fp16")]; + tensor var_1723_cast_fp16 = softmax(axis = var_288, x = aw_99_cast_fp16)[name = tensor("op_1723_cast_fp16")]; + tensor var_1725_equation_0 = const()[name = tensor("op_1725_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1725_cast_fp16 = einsum(equation = var_1725_equation_0, values = (var_1635_cast_fp16, var_1714_cast_fp16))[name = tensor("op_1725_cast_fp16")]; + tensor var_1727_equation_0 = const()[name = tensor("op_1727_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1727_cast_fp16 = einsum(equation = var_1727_equation_0, values = (var_1639_cast_fp16, var_1715_cast_fp16))[name = tensor("op_1727_cast_fp16")]; + tensor var_1729_equation_0 = const()[name = tensor("op_1729_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1729_cast_fp16 = einsum(equation = var_1729_equation_0, values = (var_1643_cast_fp16, var_1716_cast_fp16))[name = tensor("op_1729_cast_fp16")]; + tensor var_1731_equation_0 = const()[name = tensor("op_1731_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1731_cast_fp16 = einsum(equation = var_1731_equation_0, values = (var_1647_cast_fp16, var_1717_cast_fp16))[name = tensor("op_1731_cast_fp16")]; + tensor var_1733_equation_0 = const()[name = tensor("op_1733_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1733_cast_fp16 = einsum(equation = var_1733_equation_0, values = (var_1651_cast_fp16, var_1718_cast_fp16))[name = tensor("op_1733_cast_fp16")]; + tensor var_1735_equation_0 = const()[name = tensor("op_1735_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1735_cast_fp16 = einsum(equation = var_1735_equation_0, values = (var_1655_cast_fp16, var_1719_cast_fp16))[name = tensor("op_1735_cast_fp16")]; + tensor var_1737_equation_0 = const()[name = tensor("op_1737_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1737_cast_fp16 = einsum(equation = var_1737_equation_0, values = (var_1659_cast_fp16, var_1720_cast_fp16))[name = tensor("op_1737_cast_fp16")]; + tensor var_1739_equation_0 = const()[name = tensor("op_1739_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1739_cast_fp16 = einsum(equation = var_1739_equation_0, values = (var_1663_cast_fp16, var_1721_cast_fp16))[name = tensor("op_1739_cast_fp16")]; + tensor var_1741_equation_0 = const()[name = tensor("op_1741_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1741_cast_fp16 = einsum(equation = var_1741_equation_0, values = (var_1667_cast_fp16, var_1722_cast_fp16))[name = tensor("op_1741_cast_fp16")]; + tensor var_1743_equation_0 = const()[name = tensor("op_1743_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1743_cast_fp16 = einsum(equation = var_1743_equation_0, values = (var_1671_cast_fp16, var_1723_cast_fp16))[name = tensor("op_1743_cast_fp16")]; + tensor input_95_interleave_0 = const()[name = tensor("input_95_interleave_0"), val = tensor(false)]; + tensor input_95_cast_fp16 = concat(axis = var_288, interleave = input_95_interleave_0, values = (var_1725_cast_fp16, var_1727_cast_fp16, var_1729_cast_fp16, var_1731_cast_fp16, var_1733_cast_fp16, var_1735_cast_fp16, var_1737_cast_fp16, var_1739_cast_fp16, var_1741_cast_fp16, var_1743_cast_fp16))[name = tensor("input_95_cast_fp16")]; + tensor var_1749 = const()[name = tensor("op_1749"), val = tensor([1, 1])]; + tensor var_1751 = const()[name = tensor("op_1751"), val = tensor([1, 1])]; + tensor var_1753_pad_type_0 = const()[name = tensor("op_1753_pad_type_0"), val = tensor("custom")]; + tensor var_1753_pad_0 = const()[name = tensor("op_1753_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(99947840)))]; + tensor down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(100767104)))]; + tensor var_1753_cast_fp16 = conv(bias = down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_1751, groups = var_288, pad = var_1753_pad_0, pad_type = var_1753_pad_type_0, strides = var_1749, weight = down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16, x = input_95_cast_fp16)[name = tensor("op_1753_cast_fp16")]; + tensor inputs_15_cast_fp16 = add(x = var_1753_cast_fp16, y = inputs_13_cast_fp16)[name = tensor("inputs_15_cast_fp16")]; + tensor hidden_states_43_axes_0 = const()[name = tensor("hidden_states_43_axes_0"), val = tensor([1])]; + tensor hidden_states_43_gamma_0_to_fp16 = const()[name = tensor("hidden_states_43_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(100768448)))]; + tensor hidden_states_43_beta_0_to_fp16 = const()[name = tensor("hidden_states_43_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(100769792)))]; + tensor var_1763_to_fp16 = const()[name = tensor("op_1763_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_43_cast_fp16 = layer_norm(axes = hidden_states_43_axes_0, beta = hidden_states_43_beta_0_to_fp16, epsilon = var_1763_to_fp16, gamma = hidden_states_43_gamma_0_to_fp16, x = inputs_15_cast_fp16)[name = tensor("hidden_states_43_cast_fp16")]; + tensor var_1778 = const()[name = tensor("op_1778"), val = tensor([1, 1])]; + tensor var_1780 = const()[name = tensor("op_1780"), val = tensor([1, 1])]; + tensor q_11_pad_type_0 = const()[name = tensor("q_11_pad_type_0"), val = tensor("custom")]; + tensor q_11_pad_0 = const()[name = tensor("q_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16 = const()[name = tensor("down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(100771136)))]; + tensor q_11_cast_fp16 = conv(dilations = var_1780, groups = var_288, pad = q_11_pad_0, pad_type = q_11_pad_type_0, strides = var_1778, weight = down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16, x = hidden_states_43_cast_fp16)[name = tensor("q_11_cast_fp16")]; + tensor var_1784 = const()[name = tensor("op_1784"), val = tensor([1, 1])]; + tensor var_1786 = const()[name = tensor("op_1786"), val = tensor([1, 1])]; + tensor k_21_pad_type_0 = const()[name = tensor("k_21_pad_type_0"), val = tensor("custom")]; + tensor k_21_pad_0 = const()[name = tensor("k_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16 = const()[name = tensor("down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(101590400)))]; + tensor k_21_cast_fp16 = conv(dilations = var_1786, groups = var_288, pad = k_21_pad_0, pad_type = k_21_pad_type_0, strides = var_1784, weight = down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_21_cast_fp16")]; + tensor var_1790 = const()[name = tensor("op_1790"), val = tensor([1, 1])]; + tensor var_1792 = const()[name = tensor("op_1792"), val = tensor([1, 1])]; + tensor v_11_pad_type_0 = const()[name = tensor("v_11_pad_type_0"), val = tensor("custom")]; + tensor v_11_pad_0 = const()[name = tensor("v_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16 = const()[name = tensor("down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(104211904)))]; + tensor v_11_cast_fp16 = conv(dilations = var_1792, groups = var_288, pad = v_11_pad_0, pad_type = v_11_pad_type_0, strides = var_1790, weight = down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_11_cast_fp16")]; + tensor var_1796_begin_0 = const()[name = tensor("op_1796_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1796_end_0 = const()[name = tensor("op_1796_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_1796_end_mask_0 = const()[name = tensor("op_1796_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1796_cast_fp16 = slice_by_index(begin = var_1796_begin_0, end = var_1796_end_0, end_mask = var_1796_end_mask_0, x = q_11_cast_fp16)[name = tensor("op_1796_cast_fp16")]; + tensor var_1800_begin_0 = const()[name = tensor("op_1800_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_1800_end_0 = const()[name = tensor("op_1800_end_0"), val = tensor([2, 128, 1, 4096])]; + tensor var_1800_end_mask_0 = const()[name = tensor("op_1800_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1800_cast_fp16 = slice_by_index(begin = var_1800_begin_0, end = var_1800_end_0, end_mask = var_1800_end_mask_0, x = q_11_cast_fp16)[name = tensor("op_1800_cast_fp16")]; + tensor var_1804_begin_0 = const()[name = tensor("op_1804_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_1804_end_0 = const()[name = tensor("op_1804_end_0"), val = tensor([2, 192, 1, 4096])]; + tensor var_1804_end_mask_0 = const()[name = tensor("op_1804_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1804_cast_fp16 = slice_by_index(begin = var_1804_begin_0, end = var_1804_end_0, end_mask = var_1804_end_mask_0, x = q_11_cast_fp16)[name = tensor("op_1804_cast_fp16")]; + tensor var_1808_begin_0 = const()[name = tensor("op_1808_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_1808_end_0 = const()[name = tensor("op_1808_end_0"), val = tensor([2, 256, 1, 4096])]; + tensor var_1808_end_mask_0 = const()[name = tensor("op_1808_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1808_cast_fp16 = slice_by_index(begin = var_1808_begin_0, end = var_1808_end_0, end_mask = var_1808_end_mask_0, x = q_11_cast_fp16)[name = tensor("op_1808_cast_fp16")]; + tensor var_1812_begin_0 = const()[name = tensor("op_1812_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_1812_end_0 = const()[name = tensor("op_1812_end_0"), val = tensor([2, 320, 1, 4096])]; + tensor var_1812_end_mask_0 = const()[name = tensor("op_1812_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1812_cast_fp16 = slice_by_index(begin = var_1812_begin_0, end = var_1812_end_0, end_mask = var_1812_end_mask_0, x = q_11_cast_fp16)[name = tensor("op_1812_cast_fp16")]; + tensor var_1816_begin_0 = const()[name = tensor("op_1816_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_1816_end_0 = const()[name = tensor("op_1816_end_0"), val = tensor([2, 384, 1, 4096])]; + tensor var_1816_end_mask_0 = const()[name = tensor("op_1816_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1816_cast_fp16 = slice_by_index(begin = var_1816_begin_0, end = var_1816_end_0, end_mask = var_1816_end_mask_0, x = q_11_cast_fp16)[name = tensor("op_1816_cast_fp16")]; + tensor var_1820_begin_0 = const()[name = tensor("op_1820_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_1820_end_0 = const()[name = tensor("op_1820_end_0"), val = tensor([2, 448, 1, 4096])]; + tensor var_1820_end_mask_0 = const()[name = tensor("op_1820_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1820_cast_fp16 = slice_by_index(begin = var_1820_begin_0, end = var_1820_end_0, end_mask = var_1820_end_mask_0, x = q_11_cast_fp16)[name = tensor("op_1820_cast_fp16")]; + tensor var_1824_begin_0 = const()[name = tensor("op_1824_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_1824_end_0 = const()[name = tensor("op_1824_end_0"), val = tensor([2, 512, 1, 4096])]; + tensor var_1824_end_mask_0 = const()[name = tensor("op_1824_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1824_cast_fp16 = slice_by_index(begin = var_1824_begin_0, end = var_1824_end_0, end_mask = var_1824_end_mask_0, x = q_11_cast_fp16)[name = tensor("op_1824_cast_fp16")]; + tensor var_1828_begin_0 = const()[name = tensor("op_1828_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_1828_end_0 = const()[name = tensor("op_1828_end_0"), val = tensor([2, 576, 1, 4096])]; + tensor var_1828_end_mask_0 = const()[name = tensor("op_1828_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1828_cast_fp16 = slice_by_index(begin = var_1828_begin_0, end = var_1828_end_0, end_mask = var_1828_end_mask_0, x = q_11_cast_fp16)[name = tensor("op_1828_cast_fp16")]; + tensor var_1832_begin_0 = const()[name = tensor("op_1832_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_1832_end_0 = const()[name = tensor("op_1832_end_0"), val = tensor([2, 640, 1, 4096])]; + tensor var_1832_end_mask_0 = const()[name = tensor("op_1832_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1832_cast_fp16 = slice_by_index(begin = var_1832_begin_0, end = var_1832_end_0, end_mask = var_1832_end_mask_0, x = q_11_cast_fp16)[name = tensor("op_1832_cast_fp16")]; + tensor k_23_perm_0 = const()[name = tensor("k_23_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_1839_begin_0 = const()[name = tensor("op_1839_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1839_end_0 = const()[name = tensor("op_1839_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_1839_end_mask_0 = const()[name = tensor("op_1839_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_23_cast_fp16 = transpose(perm = k_23_perm_0, x = k_21_cast_fp16)[name = tensor("transpose_134")]; + tensor var_1839_cast_fp16 = slice_by_index(begin = var_1839_begin_0, end = var_1839_end_0, end_mask = var_1839_end_mask_0, x = k_23_cast_fp16)[name = tensor("op_1839_cast_fp16")]; + tensor var_1843_begin_0 = const()[name = tensor("op_1843_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_1843_end_0 = const()[name = tensor("op_1843_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_1843_end_mask_0 = const()[name = tensor("op_1843_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1843_cast_fp16 = slice_by_index(begin = var_1843_begin_0, end = var_1843_end_0, end_mask = var_1843_end_mask_0, x = k_23_cast_fp16)[name = tensor("op_1843_cast_fp16")]; + tensor var_1847_begin_0 = const()[name = tensor("op_1847_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_1847_end_0 = const()[name = tensor("op_1847_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_1847_end_mask_0 = const()[name = tensor("op_1847_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1847_cast_fp16 = slice_by_index(begin = var_1847_begin_0, end = var_1847_end_0, end_mask = var_1847_end_mask_0, x = k_23_cast_fp16)[name = tensor("op_1847_cast_fp16")]; + tensor var_1851_begin_0 = const()[name = tensor("op_1851_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_1851_end_0 = const()[name = tensor("op_1851_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_1851_end_mask_0 = const()[name = tensor("op_1851_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1851_cast_fp16 = slice_by_index(begin = var_1851_begin_0, end = var_1851_end_0, end_mask = var_1851_end_mask_0, x = k_23_cast_fp16)[name = tensor("op_1851_cast_fp16")]; + tensor var_1855_begin_0 = const()[name = tensor("op_1855_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_1855_end_0 = const()[name = tensor("op_1855_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_1855_end_mask_0 = const()[name = tensor("op_1855_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1855_cast_fp16 = slice_by_index(begin = var_1855_begin_0, end = var_1855_end_0, end_mask = var_1855_end_mask_0, x = k_23_cast_fp16)[name = tensor("op_1855_cast_fp16")]; + tensor var_1859_begin_0 = const()[name = tensor("op_1859_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_1859_end_0 = const()[name = tensor("op_1859_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_1859_end_mask_0 = const()[name = tensor("op_1859_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1859_cast_fp16 = slice_by_index(begin = var_1859_begin_0, end = var_1859_end_0, end_mask = var_1859_end_mask_0, x = k_23_cast_fp16)[name = tensor("op_1859_cast_fp16")]; + tensor var_1863_begin_0 = const()[name = tensor("op_1863_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_1863_end_0 = const()[name = tensor("op_1863_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_1863_end_mask_0 = const()[name = tensor("op_1863_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1863_cast_fp16 = slice_by_index(begin = var_1863_begin_0, end = var_1863_end_0, end_mask = var_1863_end_mask_0, x = k_23_cast_fp16)[name = tensor("op_1863_cast_fp16")]; + tensor var_1867_begin_0 = const()[name = tensor("op_1867_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_1867_end_0 = const()[name = tensor("op_1867_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_1867_end_mask_0 = const()[name = tensor("op_1867_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1867_cast_fp16 = slice_by_index(begin = var_1867_begin_0, end = var_1867_end_0, end_mask = var_1867_end_mask_0, x = k_23_cast_fp16)[name = tensor("op_1867_cast_fp16")]; + tensor var_1871_begin_0 = const()[name = tensor("op_1871_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_1871_end_0 = const()[name = tensor("op_1871_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_1871_end_mask_0 = const()[name = tensor("op_1871_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1871_cast_fp16 = slice_by_index(begin = var_1871_begin_0, end = var_1871_end_0, end_mask = var_1871_end_mask_0, x = k_23_cast_fp16)[name = tensor("op_1871_cast_fp16")]; + tensor var_1875_begin_0 = const()[name = tensor("op_1875_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_1875_end_0 = const()[name = tensor("op_1875_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_1875_end_mask_0 = const()[name = tensor("op_1875_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1875_cast_fp16 = slice_by_index(begin = var_1875_begin_0, end = var_1875_end_0, end_mask = var_1875_end_mask_0, x = k_23_cast_fp16)[name = tensor("op_1875_cast_fp16")]; + tensor var_1877_begin_0 = const()[name = tensor("op_1877_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1877_end_0 = const()[name = tensor("op_1877_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_1877_end_mask_0 = const()[name = tensor("op_1877_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1877_cast_fp16 = slice_by_index(begin = var_1877_begin_0, end = var_1877_end_0, end_mask = var_1877_end_mask_0, x = v_11_cast_fp16)[name = tensor("op_1877_cast_fp16")]; + tensor var_1881_begin_0 = const()[name = tensor("op_1881_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_1881_end_0 = const()[name = tensor("op_1881_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_1881_end_mask_0 = const()[name = tensor("op_1881_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1881_cast_fp16 = slice_by_index(begin = var_1881_begin_0, end = var_1881_end_0, end_mask = var_1881_end_mask_0, x = v_11_cast_fp16)[name = tensor("op_1881_cast_fp16")]; + tensor var_1885_begin_0 = const()[name = tensor("op_1885_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_1885_end_0 = const()[name = tensor("op_1885_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_1885_end_mask_0 = const()[name = tensor("op_1885_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1885_cast_fp16 = slice_by_index(begin = var_1885_begin_0, end = var_1885_end_0, end_mask = var_1885_end_mask_0, x = v_11_cast_fp16)[name = tensor("op_1885_cast_fp16")]; + tensor var_1889_begin_0 = const()[name = tensor("op_1889_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_1889_end_0 = const()[name = tensor("op_1889_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_1889_end_mask_0 = const()[name = tensor("op_1889_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1889_cast_fp16 = slice_by_index(begin = var_1889_begin_0, end = var_1889_end_0, end_mask = var_1889_end_mask_0, x = v_11_cast_fp16)[name = tensor("op_1889_cast_fp16")]; + tensor var_1893_begin_0 = const()[name = tensor("op_1893_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_1893_end_0 = const()[name = tensor("op_1893_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_1893_end_mask_0 = const()[name = tensor("op_1893_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1893_cast_fp16 = slice_by_index(begin = var_1893_begin_0, end = var_1893_end_0, end_mask = var_1893_end_mask_0, x = v_11_cast_fp16)[name = tensor("op_1893_cast_fp16")]; + tensor var_1897_begin_0 = const()[name = tensor("op_1897_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_1897_end_0 = const()[name = tensor("op_1897_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_1897_end_mask_0 = const()[name = tensor("op_1897_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1897_cast_fp16 = slice_by_index(begin = var_1897_begin_0, end = var_1897_end_0, end_mask = var_1897_end_mask_0, x = v_11_cast_fp16)[name = tensor("op_1897_cast_fp16")]; + tensor var_1901_begin_0 = const()[name = tensor("op_1901_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_1901_end_0 = const()[name = tensor("op_1901_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_1901_end_mask_0 = const()[name = tensor("op_1901_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1901_cast_fp16 = slice_by_index(begin = var_1901_begin_0, end = var_1901_end_0, end_mask = var_1901_end_mask_0, x = v_11_cast_fp16)[name = tensor("op_1901_cast_fp16")]; + tensor var_1905_begin_0 = const()[name = tensor("op_1905_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_1905_end_0 = const()[name = tensor("op_1905_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_1905_end_mask_0 = const()[name = tensor("op_1905_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1905_cast_fp16 = slice_by_index(begin = var_1905_begin_0, end = var_1905_end_0, end_mask = var_1905_end_mask_0, x = v_11_cast_fp16)[name = tensor("op_1905_cast_fp16")]; + tensor var_1909_begin_0 = const()[name = tensor("op_1909_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_1909_end_0 = const()[name = tensor("op_1909_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_1909_end_mask_0 = const()[name = tensor("op_1909_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1909_cast_fp16 = slice_by_index(begin = var_1909_begin_0, end = var_1909_end_0, end_mask = var_1909_end_mask_0, x = v_11_cast_fp16)[name = tensor("op_1909_cast_fp16")]; + tensor var_1913_begin_0 = const()[name = tensor("op_1913_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_1913_end_0 = const()[name = tensor("op_1913_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_1913_end_mask_0 = const()[name = tensor("op_1913_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1913_cast_fp16 = slice_by_index(begin = var_1913_begin_0, end = var_1913_end_0, end_mask = var_1913_end_mask_0, x = v_11_cast_fp16)[name = tensor("op_1913_cast_fp16")]; + tensor var_1917_equation_0 = const()[name = tensor("op_1917_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1917_cast_fp16 = einsum(equation = var_1917_equation_0, values = (var_1839_cast_fp16, var_1796_cast_fp16))[name = tensor("op_1917_cast_fp16")]; + tensor var_1918_to_fp16 = const()[name = tensor("op_1918_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_101_cast_fp16 = mul(x = var_1917_cast_fp16, y = var_1918_to_fp16)[name = tensor("aw_101_cast_fp16")]; + tensor var_1921_equation_0 = const()[name = tensor("op_1921_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1921_cast_fp16 = einsum(equation = var_1921_equation_0, values = (var_1843_cast_fp16, var_1800_cast_fp16))[name = tensor("op_1921_cast_fp16")]; + tensor var_1922_to_fp16 = const()[name = tensor("op_1922_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_103_cast_fp16 = mul(x = var_1921_cast_fp16, y = var_1922_to_fp16)[name = tensor("aw_103_cast_fp16")]; + tensor var_1925_equation_0 = const()[name = tensor("op_1925_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1925_cast_fp16 = einsum(equation = var_1925_equation_0, values = (var_1847_cast_fp16, var_1804_cast_fp16))[name = tensor("op_1925_cast_fp16")]; + tensor var_1926_to_fp16 = const()[name = tensor("op_1926_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_105_cast_fp16 = mul(x = var_1925_cast_fp16, y = var_1926_to_fp16)[name = tensor("aw_105_cast_fp16")]; + tensor var_1929_equation_0 = const()[name = tensor("op_1929_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1929_cast_fp16 = einsum(equation = var_1929_equation_0, values = (var_1851_cast_fp16, var_1808_cast_fp16))[name = tensor("op_1929_cast_fp16")]; + tensor var_1930_to_fp16 = const()[name = tensor("op_1930_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_107_cast_fp16 = mul(x = var_1929_cast_fp16, y = var_1930_to_fp16)[name = tensor("aw_107_cast_fp16")]; + tensor var_1933_equation_0 = const()[name = tensor("op_1933_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1933_cast_fp16 = einsum(equation = var_1933_equation_0, values = (var_1855_cast_fp16, var_1812_cast_fp16))[name = tensor("op_1933_cast_fp16")]; + tensor var_1934_to_fp16 = const()[name = tensor("op_1934_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_109_cast_fp16 = mul(x = var_1933_cast_fp16, y = var_1934_to_fp16)[name = tensor("aw_109_cast_fp16")]; + tensor var_1937_equation_0 = const()[name = tensor("op_1937_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1937_cast_fp16 = einsum(equation = var_1937_equation_0, values = (var_1859_cast_fp16, var_1816_cast_fp16))[name = tensor("op_1937_cast_fp16")]; + tensor var_1938_to_fp16 = const()[name = tensor("op_1938_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_111_cast_fp16 = mul(x = var_1937_cast_fp16, y = var_1938_to_fp16)[name = tensor("aw_111_cast_fp16")]; + tensor var_1941_equation_0 = const()[name = tensor("op_1941_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1941_cast_fp16 = einsum(equation = var_1941_equation_0, values = (var_1863_cast_fp16, var_1820_cast_fp16))[name = tensor("op_1941_cast_fp16")]; + tensor var_1942_to_fp16 = const()[name = tensor("op_1942_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_113_cast_fp16 = mul(x = var_1941_cast_fp16, y = var_1942_to_fp16)[name = tensor("aw_113_cast_fp16")]; + tensor var_1945_equation_0 = const()[name = tensor("op_1945_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1945_cast_fp16 = einsum(equation = var_1945_equation_0, values = (var_1867_cast_fp16, var_1824_cast_fp16))[name = tensor("op_1945_cast_fp16")]; + tensor var_1946_to_fp16 = const()[name = tensor("op_1946_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_115_cast_fp16 = mul(x = var_1945_cast_fp16, y = var_1946_to_fp16)[name = tensor("aw_115_cast_fp16")]; + tensor var_1949_equation_0 = const()[name = tensor("op_1949_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1949_cast_fp16 = einsum(equation = var_1949_equation_0, values = (var_1871_cast_fp16, var_1828_cast_fp16))[name = tensor("op_1949_cast_fp16")]; + tensor var_1950_to_fp16 = const()[name = tensor("op_1950_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_117_cast_fp16 = mul(x = var_1949_cast_fp16, y = var_1950_to_fp16)[name = tensor("aw_117_cast_fp16")]; + tensor var_1953_equation_0 = const()[name = tensor("op_1953_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1953_cast_fp16 = einsum(equation = var_1953_equation_0, values = (var_1875_cast_fp16, var_1832_cast_fp16))[name = tensor("op_1953_cast_fp16")]; + tensor var_1954_to_fp16 = const()[name = tensor("op_1954_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_119_cast_fp16 = mul(x = var_1953_cast_fp16, y = var_1954_to_fp16)[name = tensor("aw_119_cast_fp16")]; + tensor var_1956_cast_fp16 = softmax(axis = var_288, x = aw_101_cast_fp16)[name = tensor("op_1956_cast_fp16")]; + tensor var_1957_cast_fp16 = softmax(axis = var_288, x = aw_103_cast_fp16)[name = tensor("op_1957_cast_fp16")]; + tensor var_1958_cast_fp16 = softmax(axis = var_288, x = aw_105_cast_fp16)[name = tensor("op_1958_cast_fp16")]; + tensor var_1959_cast_fp16 = softmax(axis = var_288, x = aw_107_cast_fp16)[name = tensor("op_1959_cast_fp16")]; + tensor var_1960_cast_fp16 = softmax(axis = var_288, x = aw_109_cast_fp16)[name = tensor("op_1960_cast_fp16")]; + tensor var_1961_cast_fp16 = softmax(axis = var_288, x = aw_111_cast_fp16)[name = tensor("op_1961_cast_fp16")]; + tensor var_1962_cast_fp16 = softmax(axis = var_288, x = aw_113_cast_fp16)[name = tensor("op_1962_cast_fp16")]; + tensor var_1963_cast_fp16 = softmax(axis = var_288, x = aw_115_cast_fp16)[name = tensor("op_1963_cast_fp16")]; + tensor var_1964_cast_fp16 = softmax(axis = var_288, x = aw_117_cast_fp16)[name = tensor("op_1964_cast_fp16")]; + tensor var_1965_cast_fp16 = softmax(axis = var_288, x = aw_119_cast_fp16)[name = tensor("op_1965_cast_fp16")]; + tensor var_1967_equation_0 = const()[name = tensor("op_1967_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1967_cast_fp16 = einsum(equation = var_1967_equation_0, values = (var_1877_cast_fp16, var_1956_cast_fp16))[name = tensor("op_1967_cast_fp16")]; + tensor var_1969_equation_0 = const()[name = tensor("op_1969_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1969_cast_fp16 = einsum(equation = var_1969_equation_0, values = (var_1881_cast_fp16, var_1957_cast_fp16))[name = tensor("op_1969_cast_fp16")]; + tensor var_1971_equation_0 = const()[name = tensor("op_1971_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1971_cast_fp16 = einsum(equation = var_1971_equation_0, values = (var_1885_cast_fp16, var_1958_cast_fp16))[name = tensor("op_1971_cast_fp16")]; + tensor var_1973_equation_0 = const()[name = tensor("op_1973_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1973_cast_fp16 = einsum(equation = var_1973_equation_0, values = (var_1889_cast_fp16, var_1959_cast_fp16))[name = tensor("op_1973_cast_fp16")]; + tensor var_1975_equation_0 = const()[name = tensor("op_1975_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1975_cast_fp16 = einsum(equation = var_1975_equation_0, values = (var_1893_cast_fp16, var_1960_cast_fp16))[name = tensor("op_1975_cast_fp16")]; + tensor var_1977_equation_0 = const()[name = tensor("op_1977_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1977_cast_fp16 = einsum(equation = var_1977_equation_0, values = (var_1897_cast_fp16, var_1961_cast_fp16))[name = tensor("op_1977_cast_fp16")]; + tensor var_1979_equation_0 = const()[name = tensor("op_1979_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1979_cast_fp16 = einsum(equation = var_1979_equation_0, values = (var_1901_cast_fp16, var_1962_cast_fp16))[name = tensor("op_1979_cast_fp16")]; + tensor var_1981_equation_0 = const()[name = tensor("op_1981_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1981_cast_fp16 = einsum(equation = var_1981_equation_0, values = (var_1905_cast_fp16, var_1963_cast_fp16))[name = tensor("op_1981_cast_fp16")]; + tensor var_1983_equation_0 = const()[name = tensor("op_1983_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1983_cast_fp16 = einsum(equation = var_1983_equation_0, values = (var_1909_cast_fp16, var_1964_cast_fp16))[name = tensor("op_1983_cast_fp16")]; + tensor var_1985_equation_0 = const()[name = tensor("op_1985_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1985_cast_fp16 = einsum(equation = var_1985_equation_0, values = (var_1913_cast_fp16, var_1965_cast_fp16))[name = tensor("op_1985_cast_fp16")]; + tensor input_97_interleave_0 = const()[name = tensor("input_97_interleave_0"), val = tensor(false)]; + tensor input_97_cast_fp16 = concat(axis = var_288, interleave = input_97_interleave_0, values = (var_1967_cast_fp16, var_1969_cast_fp16, var_1971_cast_fp16, var_1973_cast_fp16, var_1975_cast_fp16, var_1977_cast_fp16, var_1979_cast_fp16, var_1981_cast_fp16, var_1983_cast_fp16, var_1985_cast_fp16))[name = tensor("input_97_cast_fp16")]; + tensor var_1991 = const()[name = tensor("op_1991"), val = tensor([1, 1])]; + tensor var_1993 = const()[name = tensor("op_1993"), val = tensor([1, 1])]; + tensor var_1995_pad_type_0 = const()[name = tensor("op_1995_pad_type_0"), val = tensor("custom")]; + tensor var_1995_pad_0 = const()[name = tensor("op_1995_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(106833408)))]; + tensor down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(107652672)))]; + tensor var_1995_cast_fp16 = conv(bias = down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_1993, groups = var_288, pad = var_1995_pad_0, pad_type = var_1995_pad_type_0, strides = var_1991, weight = down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16, x = input_97_cast_fp16)[name = tensor("op_1995_cast_fp16")]; + tensor inputs_17_cast_fp16 = add(x = var_1995_cast_fp16, y = inputs_15_cast_fp16)[name = tensor("inputs_17_cast_fp16")]; + tensor input_99_axes_0 = const()[name = tensor("input_99_axes_0"), val = tensor([1])]; + tensor input_99_gamma_0_to_fp16 = const()[name = tensor("input_99_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(107654016)))]; + tensor input_99_beta_0_to_fp16 = const()[name = tensor("input_99_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(107655360)))]; + tensor var_2005_to_fp16 = const()[name = tensor("op_2005_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_99_cast_fp16 = layer_norm(axes = input_99_axes_0, beta = input_99_beta_0_to_fp16, epsilon = var_2005_to_fp16, gamma = input_99_gamma_0_to_fp16, x = inputs_17_cast_fp16)[name = tensor("input_99_cast_fp16")]; + tensor var_2021 = const()[name = tensor("op_2021"), val = tensor([1, 1])]; + tensor var_2023 = const()[name = tensor("op_2023"), val = tensor([1, 1])]; + tensor var_2025_pad_type_0 = const()[name = tensor("op_2025_pad_type_0"), val = tensor("custom")]; + tensor var_2025_pad_0 = const()[name = tensor("op_2025_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("down_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(107656704)))]; + tensor down_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("down_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(114210368)))]; + tensor var_2025_cast_fp16 = conv(bias = down_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_2023, groups = var_288, pad = var_2025_pad_0, pad_type = var_2025_pad_type_0, strides = var_2021, weight = down_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16, x = input_99_cast_fp16)[name = tensor("op_2025_cast_fp16")]; + tensor var_2026_split_sizes_0 = const()[name = tensor("op_2026_split_sizes_0"), val = tensor([2560, 2560])]; + tensor var_2026_axis_0 = const()[name = tensor("op_2026_axis_0"), val = tensor(1)]; + tensor var_2026_cast_fp16_0, tensor var_2026_cast_fp16_1 = split(axis = var_2026_axis_0, split_sizes = var_2026_split_sizes_0, x = var_2025_cast_fp16)[name = tensor("op_2026_cast_fp16")]; + tensor var_2028_mode_0 = const()[name = tensor("op_2028_mode_0"), val = tensor("EXACT")]; + tensor var_2028_cast_fp16 = gelu(mode = var_2028_mode_0, x = var_2026_cast_fp16_1)[name = tensor("op_2028_cast_fp16")]; + tensor input_101_cast_fp16 = mul(x = var_2026_cast_fp16_0, y = var_2028_cast_fp16)[name = tensor("input_101_cast_fp16")]; + tensor var_2032 = const()[name = tensor("op_2032"), val = tensor([1, 1])]; + tensor var_2034 = const()[name = tensor("op_2034"), val = tensor([1, 1])]; + tensor var_2036_pad_type_0 = const()[name = tensor("op_2036_pad_type_0"), val = tensor("custom")]; + tensor var_2036_pad_0 = const()[name = tensor("op_2036_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16 = const()[name = tensor("down_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(114220672)))]; + tensor down_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor("down_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(117497536)))]; + tensor var_2036_cast_fp16 = conv(bias = down_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_2034, groups = var_288, pad = var_2036_pad_0, pad_type = var_2036_pad_type_0, strides = var_2032, weight = down_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16, x = input_101_cast_fp16)[name = tensor("op_2036_cast_fp16")]; + tensor inputs_19_cast_fp16 = add(x = var_2036_cast_fp16, y = inputs_17_cast_fp16)[name = tensor("inputs_19_cast_fp16")]; + tensor hidden_states_47_axes_0 = const()[name = tensor("hidden_states_47_axes_0"), val = tensor([1])]; + tensor hidden_states_47_gamma_0_to_fp16 = const()[name = tensor("hidden_states_47_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(117498880)))]; + tensor hidden_states_47_beta_0_to_fp16 = const()[name = tensor("hidden_states_47_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(117500224)))]; + tensor var_2052_to_fp16 = const()[name = tensor("op_2052_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_47_cast_fp16 = layer_norm(axes = hidden_states_47_axes_0, beta = hidden_states_47_beta_0_to_fp16, epsilon = var_2052_to_fp16, gamma = hidden_states_47_gamma_0_to_fp16, x = inputs_19_cast_fp16)[name = tensor("hidden_states_47_cast_fp16")]; + tensor var_2067 = const()[name = tensor("op_2067"), val = tensor([1, 1])]; + tensor var_2069 = const()[name = tensor("op_2069"), val = tensor([1, 1])]; + tensor q_13_pad_type_0 = const()[name = tensor("q_13_pad_type_0"), val = tensor("custom")]; + tensor q_13_pad_0 = const()[name = tensor("q_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_1_transformer_blocks_1_attn1_to_q_weight_to_fp16 = const()[name = tensor("down_blocks_1_attentions_1_transformer_blocks_1_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(117501568)))]; + tensor q_13_cast_fp16 = conv(dilations = var_2069, groups = var_288, pad = q_13_pad_0, pad_type = q_13_pad_type_0, strides = var_2067, weight = down_blocks_1_attentions_1_transformer_blocks_1_attn1_to_q_weight_to_fp16, x = hidden_states_47_cast_fp16)[name = tensor("q_13_cast_fp16")]; + tensor var_2073 = const()[name = tensor("op_2073"), val = tensor([1, 1])]; + tensor var_2075 = const()[name = tensor("op_2075"), val = tensor([1, 1])]; + tensor k_25_pad_type_0 = const()[name = tensor("k_25_pad_type_0"), val = tensor("custom")]; + tensor k_25_pad_0 = const()[name = tensor("k_25_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_1_transformer_blocks_1_attn1_to_k_weight_to_fp16 = const()[name = tensor("down_blocks_1_attentions_1_transformer_blocks_1_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(118320832)))]; + tensor k_25_cast_fp16 = conv(dilations = var_2075, groups = var_288, pad = k_25_pad_0, pad_type = k_25_pad_type_0, strides = var_2073, weight = down_blocks_1_attentions_1_transformer_blocks_1_attn1_to_k_weight_to_fp16, x = hidden_states_47_cast_fp16)[name = tensor("k_25_cast_fp16")]; + tensor var_2079 = const()[name = tensor("op_2079"), val = tensor([1, 1])]; + tensor var_2081 = const()[name = tensor("op_2081"), val = tensor([1, 1])]; + tensor v_13_pad_type_0 = const()[name = tensor("v_13_pad_type_0"), val = tensor("custom")]; + tensor v_13_pad_0 = const()[name = tensor("v_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_1_transformer_blocks_1_attn1_to_v_weight_to_fp16 = const()[name = tensor("down_blocks_1_attentions_1_transformer_blocks_1_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(119140096)))]; + tensor v_13_cast_fp16 = conv(dilations = var_2081, groups = var_288, pad = v_13_pad_0, pad_type = v_13_pad_type_0, strides = var_2079, weight = down_blocks_1_attentions_1_transformer_blocks_1_attn1_to_v_weight_to_fp16, x = hidden_states_47_cast_fp16)[name = tensor("v_13_cast_fp16")]; + tensor var_2085_begin_0 = const()[name = tensor("op_2085_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2085_end_0 = const()[name = tensor("op_2085_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_2085_end_mask_0 = const()[name = tensor("op_2085_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2085_cast_fp16 = slice_by_index(begin = var_2085_begin_0, end = var_2085_end_0, end_mask = var_2085_end_mask_0, x = q_13_cast_fp16)[name = tensor("op_2085_cast_fp16")]; + tensor var_2089_begin_0 = const()[name = tensor("op_2089_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_2089_end_0 = const()[name = tensor("op_2089_end_0"), val = tensor([2, 128, 1, 4096])]; + tensor var_2089_end_mask_0 = const()[name = tensor("op_2089_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2089_cast_fp16 = slice_by_index(begin = var_2089_begin_0, end = var_2089_end_0, end_mask = var_2089_end_mask_0, x = q_13_cast_fp16)[name = tensor("op_2089_cast_fp16")]; + tensor var_2093_begin_0 = const()[name = tensor("op_2093_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_2093_end_0 = const()[name = tensor("op_2093_end_0"), val = tensor([2, 192, 1, 4096])]; + tensor var_2093_end_mask_0 = const()[name = tensor("op_2093_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2093_cast_fp16 = slice_by_index(begin = var_2093_begin_0, end = var_2093_end_0, end_mask = var_2093_end_mask_0, x = q_13_cast_fp16)[name = tensor("op_2093_cast_fp16")]; + tensor var_2097_begin_0 = const()[name = tensor("op_2097_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_2097_end_0 = const()[name = tensor("op_2097_end_0"), val = tensor([2, 256, 1, 4096])]; + tensor var_2097_end_mask_0 = const()[name = tensor("op_2097_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2097_cast_fp16 = slice_by_index(begin = var_2097_begin_0, end = var_2097_end_0, end_mask = var_2097_end_mask_0, x = q_13_cast_fp16)[name = tensor("op_2097_cast_fp16")]; + tensor var_2101_begin_0 = const()[name = tensor("op_2101_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_2101_end_0 = const()[name = tensor("op_2101_end_0"), val = tensor([2, 320, 1, 4096])]; + tensor var_2101_end_mask_0 = const()[name = tensor("op_2101_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2101_cast_fp16 = slice_by_index(begin = var_2101_begin_0, end = var_2101_end_0, end_mask = var_2101_end_mask_0, x = q_13_cast_fp16)[name = tensor("op_2101_cast_fp16")]; + tensor var_2105_begin_0 = const()[name = tensor("op_2105_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_2105_end_0 = const()[name = tensor("op_2105_end_0"), val = tensor([2, 384, 1, 4096])]; + tensor var_2105_end_mask_0 = const()[name = tensor("op_2105_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2105_cast_fp16 = slice_by_index(begin = var_2105_begin_0, end = var_2105_end_0, end_mask = var_2105_end_mask_0, x = q_13_cast_fp16)[name = tensor("op_2105_cast_fp16")]; + tensor var_2109_begin_0 = const()[name = tensor("op_2109_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_2109_end_0 = const()[name = tensor("op_2109_end_0"), val = tensor([2, 448, 1, 4096])]; + tensor var_2109_end_mask_0 = const()[name = tensor("op_2109_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2109_cast_fp16 = slice_by_index(begin = var_2109_begin_0, end = var_2109_end_0, end_mask = var_2109_end_mask_0, x = q_13_cast_fp16)[name = tensor("op_2109_cast_fp16")]; + tensor var_2113_begin_0 = const()[name = tensor("op_2113_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_2113_end_0 = const()[name = tensor("op_2113_end_0"), val = tensor([2, 512, 1, 4096])]; + tensor var_2113_end_mask_0 = const()[name = tensor("op_2113_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2113_cast_fp16 = slice_by_index(begin = var_2113_begin_0, end = var_2113_end_0, end_mask = var_2113_end_mask_0, x = q_13_cast_fp16)[name = tensor("op_2113_cast_fp16")]; + tensor var_2117_begin_0 = const()[name = tensor("op_2117_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_2117_end_0 = const()[name = tensor("op_2117_end_0"), val = tensor([2, 576, 1, 4096])]; + tensor var_2117_end_mask_0 = const()[name = tensor("op_2117_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2117_cast_fp16 = slice_by_index(begin = var_2117_begin_0, end = var_2117_end_0, end_mask = var_2117_end_mask_0, x = q_13_cast_fp16)[name = tensor("op_2117_cast_fp16")]; + tensor var_2121_begin_0 = const()[name = tensor("op_2121_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_2121_end_0 = const()[name = tensor("op_2121_end_0"), val = tensor([2, 640, 1, 4096])]; + tensor var_2121_end_mask_0 = const()[name = tensor("op_2121_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2121_cast_fp16 = slice_by_index(begin = var_2121_begin_0, end = var_2121_end_0, end_mask = var_2121_end_mask_0, x = q_13_cast_fp16)[name = tensor("op_2121_cast_fp16")]; + tensor k_27_perm_0 = const()[name = tensor("k_27_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_2128_begin_0 = const()[name = tensor("op_2128_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2128_end_0 = const()[name = tensor("op_2128_end_0"), val = tensor([2, 4096, 1, 64])]; + tensor var_2128_end_mask_0 = const()[name = tensor("op_2128_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_27_cast_fp16 = transpose(perm = k_27_perm_0, x = k_25_cast_fp16)[name = tensor("transpose_133")]; + tensor var_2128_cast_fp16 = slice_by_index(begin = var_2128_begin_0, end = var_2128_end_0, end_mask = var_2128_end_mask_0, x = k_27_cast_fp16)[name = tensor("op_2128_cast_fp16")]; + tensor var_2132_begin_0 = const()[name = tensor("op_2132_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_2132_end_0 = const()[name = tensor("op_2132_end_0"), val = tensor([2, 4096, 1, 128])]; + tensor var_2132_end_mask_0 = const()[name = tensor("op_2132_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2132_cast_fp16 = slice_by_index(begin = var_2132_begin_0, end = var_2132_end_0, end_mask = var_2132_end_mask_0, x = k_27_cast_fp16)[name = tensor("op_2132_cast_fp16")]; + tensor var_2136_begin_0 = const()[name = tensor("op_2136_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_2136_end_0 = const()[name = tensor("op_2136_end_0"), val = tensor([2, 4096, 1, 192])]; + tensor var_2136_end_mask_0 = const()[name = tensor("op_2136_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2136_cast_fp16 = slice_by_index(begin = var_2136_begin_0, end = var_2136_end_0, end_mask = var_2136_end_mask_0, x = k_27_cast_fp16)[name = tensor("op_2136_cast_fp16")]; + tensor var_2140_begin_0 = const()[name = tensor("op_2140_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_2140_end_0 = const()[name = tensor("op_2140_end_0"), val = tensor([2, 4096, 1, 256])]; + tensor var_2140_end_mask_0 = const()[name = tensor("op_2140_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2140_cast_fp16 = slice_by_index(begin = var_2140_begin_0, end = var_2140_end_0, end_mask = var_2140_end_mask_0, x = k_27_cast_fp16)[name = tensor("op_2140_cast_fp16")]; + tensor var_2144_begin_0 = const()[name = tensor("op_2144_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_2144_end_0 = const()[name = tensor("op_2144_end_0"), val = tensor([2, 4096, 1, 320])]; + tensor var_2144_end_mask_0 = const()[name = tensor("op_2144_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2144_cast_fp16 = slice_by_index(begin = var_2144_begin_0, end = var_2144_end_0, end_mask = var_2144_end_mask_0, x = k_27_cast_fp16)[name = tensor("op_2144_cast_fp16")]; + tensor var_2148_begin_0 = const()[name = tensor("op_2148_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_2148_end_0 = const()[name = tensor("op_2148_end_0"), val = tensor([2, 4096, 1, 384])]; + tensor var_2148_end_mask_0 = const()[name = tensor("op_2148_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2148_cast_fp16 = slice_by_index(begin = var_2148_begin_0, end = var_2148_end_0, end_mask = var_2148_end_mask_0, x = k_27_cast_fp16)[name = tensor("op_2148_cast_fp16")]; + tensor var_2152_begin_0 = const()[name = tensor("op_2152_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_2152_end_0 = const()[name = tensor("op_2152_end_0"), val = tensor([2, 4096, 1, 448])]; + tensor var_2152_end_mask_0 = const()[name = tensor("op_2152_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2152_cast_fp16 = slice_by_index(begin = var_2152_begin_0, end = var_2152_end_0, end_mask = var_2152_end_mask_0, x = k_27_cast_fp16)[name = tensor("op_2152_cast_fp16")]; + tensor var_2156_begin_0 = const()[name = tensor("op_2156_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_2156_end_0 = const()[name = tensor("op_2156_end_0"), val = tensor([2, 4096, 1, 512])]; + tensor var_2156_end_mask_0 = const()[name = tensor("op_2156_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2156_cast_fp16 = slice_by_index(begin = var_2156_begin_0, end = var_2156_end_0, end_mask = var_2156_end_mask_0, x = k_27_cast_fp16)[name = tensor("op_2156_cast_fp16")]; + tensor var_2160_begin_0 = const()[name = tensor("op_2160_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_2160_end_0 = const()[name = tensor("op_2160_end_0"), val = tensor([2, 4096, 1, 576])]; + tensor var_2160_end_mask_0 = const()[name = tensor("op_2160_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2160_cast_fp16 = slice_by_index(begin = var_2160_begin_0, end = var_2160_end_0, end_mask = var_2160_end_mask_0, x = k_27_cast_fp16)[name = tensor("op_2160_cast_fp16")]; + tensor var_2164_begin_0 = const()[name = tensor("op_2164_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_2164_end_0 = const()[name = tensor("op_2164_end_0"), val = tensor([2, 4096, 1, 640])]; + tensor var_2164_end_mask_0 = const()[name = tensor("op_2164_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2164_cast_fp16 = slice_by_index(begin = var_2164_begin_0, end = var_2164_end_0, end_mask = var_2164_end_mask_0, x = k_27_cast_fp16)[name = tensor("op_2164_cast_fp16")]; + tensor var_2166_begin_0 = const()[name = tensor("op_2166_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2166_end_0 = const()[name = tensor("op_2166_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_2166_end_mask_0 = const()[name = tensor("op_2166_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2166_cast_fp16 = slice_by_index(begin = var_2166_begin_0, end = var_2166_end_0, end_mask = var_2166_end_mask_0, x = v_13_cast_fp16)[name = tensor("op_2166_cast_fp16")]; + tensor var_2170_begin_0 = const()[name = tensor("op_2170_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_2170_end_0 = const()[name = tensor("op_2170_end_0"), val = tensor([2, 128, 1, 4096])]; + tensor var_2170_end_mask_0 = const()[name = tensor("op_2170_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2170_cast_fp16 = slice_by_index(begin = var_2170_begin_0, end = var_2170_end_0, end_mask = var_2170_end_mask_0, x = v_13_cast_fp16)[name = tensor("op_2170_cast_fp16")]; + tensor var_2174_begin_0 = const()[name = tensor("op_2174_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_2174_end_0 = const()[name = tensor("op_2174_end_0"), val = tensor([2, 192, 1, 4096])]; + tensor var_2174_end_mask_0 = const()[name = tensor("op_2174_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2174_cast_fp16 = slice_by_index(begin = var_2174_begin_0, end = var_2174_end_0, end_mask = var_2174_end_mask_0, x = v_13_cast_fp16)[name = tensor("op_2174_cast_fp16")]; + tensor var_2178_begin_0 = const()[name = tensor("op_2178_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_2178_end_0 = const()[name = tensor("op_2178_end_0"), val = tensor([2, 256, 1, 4096])]; + tensor var_2178_end_mask_0 = const()[name = tensor("op_2178_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2178_cast_fp16 = slice_by_index(begin = var_2178_begin_0, end = var_2178_end_0, end_mask = var_2178_end_mask_0, x = v_13_cast_fp16)[name = tensor("op_2178_cast_fp16")]; + tensor var_2182_begin_0 = const()[name = tensor("op_2182_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_2182_end_0 = const()[name = tensor("op_2182_end_0"), val = tensor([2, 320, 1, 4096])]; + tensor var_2182_end_mask_0 = const()[name = tensor("op_2182_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2182_cast_fp16 = slice_by_index(begin = var_2182_begin_0, end = var_2182_end_0, end_mask = var_2182_end_mask_0, x = v_13_cast_fp16)[name = tensor("op_2182_cast_fp16")]; + tensor var_2186_begin_0 = const()[name = tensor("op_2186_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_2186_end_0 = const()[name = tensor("op_2186_end_0"), val = tensor([2, 384, 1, 4096])]; + tensor var_2186_end_mask_0 = const()[name = tensor("op_2186_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2186_cast_fp16 = slice_by_index(begin = var_2186_begin_0, end = var_2186_end_0, end_mask = var_2186_end_mask_0, x = v_13_cast_fp16)[name = tensor("op_2186_cast_fp16")]; + tensor var_2190_begin_0 = const()[name = tensor("op_2190_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_2190_end_0 = const()[name = tensor("op_2190_end_0"), val = tensor([2, 448, 1, 4096])]; + tensor var_2190_end_mask_0 = const()[name = tensor("op_2190_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2190_cast_fp16 = slice_by_index(begin = var_2190_begin_0, end = var_2190_end_0, end_mask = var_2190_end_mask_0, x = v_13_cast_fp16)[name = tensor("op_2190_cast_fp16")]; + tensor var_2194_begin_0 = const()[name = tensor("op_2194_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_2194_end_0 = const()[name = tensor("op_2194_end_0"), val = tensor([2, 512, 1, 4096])]; + tensor var_2194_end_mask_0 = const()[name = tensor("op_2194_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2194_cast_fp16 = slice_by_index(begin = var_2194_begin_0, end = var_2194_end_0, end_mask = var_2194_end_mask_0, x = v_13_cast_fp16)[name = tensor("op_2194_cast_fp16")]; + tensor var_2198_begin_0 = const()[name = tensor("op_2198_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_2198_end_0 = const()[name = tensor("op_2198_end_0"), val = tensor([2, 576, 1, 4096])]; + tensor var_2198_end_mask_0 = const()[name = tensor("op_2198_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2198_cast_fp16 = slice_by_index(begin = var_2198_begin_0, end = var_2198_end_0, end_mask = var_2198_end_mask_0, x = v_13_cast_fp16)[name = tensor("op_2198_cast_fp16")]; + tensor var_2202_begin_0 = const()[name = tensor("op_2202_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_2202_end_0 = const()[name = tensor("op_2202_end_0"), val = tensor([2, 640, 1, 4096])]; + tensor var_2202_end_mask_0 = const()[name = tensor("op_2202_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2202_cast_fp16 = slice_by_index(begin = var_2202_begin_0, end = var_2202_end_0, end_mask = var_2202_end_mask_0, x = v_13_cast_fp16)[name = tensor("op_2202_cast_fp16")]; + tensor var_2206_equation_0 = const()[name = tensor("op_2206_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2206_cast_fp16 = einsum(equation = var_2206_equation_0, values = (var_2128_cast_fp16, var_2085_cast_fp16))[name = tensor("op_2206_cast_fp16")]; + tensor var_2207_to_fp16 = const()[name = tensor("op_2207_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_121_cast_fp16 = mul(x = var_2206_cast_fp16, y = var_2207_to_fp16)[name = tensor("aw_121_cast_fp16")]; + tensor var_2210_equation_0 = const()[name = tensor("op_2210_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2210_cast_fp16 = einsum(equation = var_2210_equation_0, values = (var_2132_cast_fp16, var_2089_cast_fp16))[name = tensor("op_2210_cast_fp16")]; + tensor var_2211_to_fp16 = const()[name = tensor("op_2211_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_123_cast_fp16 = mul(x = var_2210_cast_fp16, y = var_2211_to_fp16)[name = tensor("aw_123_cast_fp16")]; + tensor var_2214_equation_0 = const()[name = tensor("op_2214_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2214_cast_fp16 = einsum(equation = var_2214_equation_0, values = (var_2136_cast_fp16, var_2093_cast_fp16))[name = tensor("op_2214_cast_fp16")]; + tensor var_2215_to_fp16 = const()[name = tensor("op_2215_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_125_cast_fp16 = mul(x = var_2214_cast_fp16, y = var_2215_to_fp16)[name = tensor("aw_125_cast_fp16")]; + tensor var_2218_equation_0 = const()[name = tensor("op_2218_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2218_cast_fp16 = einsum(equation = var_2218_equation_0, values = (var_2140_cast_fp16, var_2097_cast_fp16))[name = tensor("op_2218_cast_fp16")]; + tensor var_2219_to_fp16 = const()[name = tensor("op_2219_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_127_cast_fp16 = mul(x = var_2218_cast_fp16, y = var_2219_to_fp16)[name = tensor("aw_127_cast_fp16")]; + tensor var_2222_equation_0 = const()[name = tensor("op_2222_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2222_cast_fp16 = einsum(equation = var_2222_equation_0, values = (var_2144_cast_fp16, var_2101_cast_fp16))[name = tensor("op_2222_cast_fp16")]; + tensor var_2223_to_fp16 = const()[name = tensor("op_2223_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_129_cast_fp16 = mul(x = var_2222_cast_fp16, y = var_2223_to_fp16)[name = tensor("aw_129_cast_fp16")]; + tensor var_2226_equation_0 = const()[name = tensor("op_2226_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2226_cast_fp16 = einsum(equation = var_2226_equation_0, values = (var_2148_cast_fp16, var_2105_cast_fp16))[name = tensor("op_2226_cast_fp16")]; + tensor var_2227_to_fp16 = const()[name = tensor("op_2227_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_131_cast_fp16 = mul(x = var_2226_cast_fp16, y = var_2227_to_fp16)[name = tensor("aw_131_cast_fp16")]; + tensor var_2230_equation_0 = const()[name = tensor("op_2230_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2230_cast_fp16 = einsum(equation = var_2230_equation_0, values = (var_2152_cast_fp16, var_2109_cast_fp16))[name = tensor("op_2230_cast_fp16")]; + tensor var_2231_to_fp16 = const()[name = tensor("op_2231_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_133_cast_fp16 = mul(x = var_2230_cast_fp16, y = var_2231_to_fp16)[name = tensor("aw_133_cast_fp16")]; + tensor var_2234_equation_0 = const()[name = tensor("op_2234_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2234_cast_fp16 = einsum(equation = var_2234_equation_0, values = (var_2156_cast_fp16, var_2113_cast_fp16))[name = tensor("op_2234_cast_fp16")]; + tensor var_2235_to_fp16 = const()[name = tensor("op_2235_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_135_cast_fp16 = mul(x = var_2234_cast_fp16, y = var_2235_to_fp16)[name = tensor("aw_135_cast_fp16")]; + tensor var_2238_equation_0 = const()[name = tensor("op_2238_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2238_cast_fp16 = einsum(equation = var_2238_equation_0, values = (var_2160_cast_fp16, var_2117_cast_fp16))[name = tensor("op_2238_cast_fp16")]; + tensor var_2239_to_fp16 = const()[name = tensor("op_2239_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_137_cast_fp16 = mul(x = var_2238_cast_fp16, y = var_2239_to_fp16)[name = tensor("aw_137_cast_fp16")]; + tensor var_2242_equation_0 = const()[name = tensor("op_2242_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2242_cast_fp16 = einsum(equation = var_2242_equation_0, values = (var_2164_cast_fp16, var_2121_cast_fp16))[name = tensor("op_2242_cast_fp16")]; + tensor var_2243_to_fp16 = const()[name = tensor("op_2243_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_139_cast_fp16 = mul(x = var_2242_cast_fp16, y = var_2243_to_fp16)[name = tensor("aw_139_cast_fp16")]; + tensor var_2245_cast_fp16 = softmax(axis = var_288, x = aw_121_cast_fp16)[name = tensor("op_2245_cast_fp16")]; + tensor var_2246_cast_fp16 = softmax(axis = var_288, x = aw_123_cast_fp16)[name = tensor("op_2246_cast_fp16")]; + tensor var_2247_cast_fp16 = softmax(axis = var_288, x = aw_125_cast_fp16)[name = tensor("op_2247_cast_fp16")]; + tensor var_2248_cast_fp16 = softmax(axis = var_288, x = aw_127_cast_fp16)[name = tensor("op_2248_cast_fp16")]; + tensor var_2249_cast_fp16 = softmax(axis = var_288, x = aw_129_cast_fp16)[name = tensor("op_2249_cast_fp16")]; + tensor var_2250_cast_fp16 = softmax(axis = var_288, x = aw_131_cast_fp16)[name = tensor("op_2250_cast_fp16")]; + tensor var_2251_cast_fp16 = softmax(axis = var_288, x = aw_133_cast_fp16)[name = tensor("op_2251_cast_fp16")]; + tensor var_2252_cast_fp16 = softmax(axis = var_288, x = aw_135_cast_fp16)[name = tensor("op_2252_cast_fp16")]; + tensor var_2253_cast_fp16 = softmax(axis = var_288, x = aw_137_cast_fp16)[name = tensor("op_2253_cast_fp16")]; + tensor var_2254_cast_fp16 = softmax(axis = var_288, x = aw_139_cast_fp16)[name = tensor("op_2254_cast_fp16")]; + tensor var_2256_equation_0 = const()[name = tensor("op_2256_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2256_cast_fp16 = einsum(equation = var_2256_equation_0, values = (var_2166_cast_fp16, var_2245_cast_fp16))[name = tensor("op_2256_cast_fp16")]; + tensor var_2258_equation_0 = const()[name = tensor("op_2258_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2258_cast_fp16 = einsum(equation = var_2258_equation_0, values = (var_2170_cast_fp16, var_2246_cast_fp16))[name = tensor("op_2258_cast_fp16")]; + tensor var_2260_equation_0 = const()[name = tensor("op_2260_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2260_cast_fp16 = einsum(equation = var_2260_equation_0, values = (var_2174_cast_fp16, var_2247_cast_fp16))[name = tensor("op_2260_cast_fp16")]; + tensor var_2262_equation_0 = const()[name = tensor("op_2262_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2262_cast_fp16 = einsum(equation = var_2262_equation_0, values = (var_2178_cast_fp16, var_2248_cast_fp16))[name = tensor("op_2262_cast_fp16")]; + tensor var_2264_equation_0 = const()[name = tensor("op_2264_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2264_cast_fp16 = einsum(equation = var_2264_equation_0, values = (var_2182_cast_fp16, var_2249_cast_fp16))[name = tensor("op_2264_cast_fp16")]; + tensor var_2266_equation_0 = const()[name = tensor("op_2266_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2266_cast_fp16 = einsum(equation = var_2266_equation_0, values = (var_2186_cast_fp16, var_2250_cast_fp16))[name = tensor("op_2266_cast_fp16")]; + tensor var_2268_equation_0 = const()[name = tensor("op_2268_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2268_cast_fp16 = einsum(equation = var_2268_equation_0, values = (var_2190_cast_fp16, var_2251_cast_fp16))[name = tensor("op_2268_cast_fp16")]; + tensor var_2270_equation_0 = const()[name = tensor("op_2270_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2270_cast_fp16 = einsum(equation = var_2270_equation_0, values = (var_2194_cast_fp16, var_2252_cast_fp16))[name = tensor("op_2270_cast_fp16")]; + tensor var_2272_equation_0 = const()[name = tensor("op_2272_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2272_cast_fp16 = einsum(equation = var_2272_equation_0, values = (var_2198_cast_fp16, var_2253_cast_fp16))[name = tensor("op_2272_cast_fp16")]; + tensor var_2274_equation_0 = const()[name = tensor("op_2274_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2274_cast_fp16 = einsum(equation = var_2274_equation_0, values = (var_2202_cast_fp16, var_2254_cast_fp16))[name = tensor("op_2274_cast_fp16")]; + tensor input_103_interleave_0 = const()[name = tensor("input_103_interleave_0"), val = tensor(false)]; + tensor input_103_cast_fp16 = concat(axis = var_288, interleave = input_103_interleave_0, values = (var_2256_cast_fp16, var_2258_cast_fp16, var_2260_cast_fp16, var_2262_cast_fp16, var_2264_cast_fp16, var_2266_cast_fp16, var_2268_cast_fp16, var_2270_cast_fp16, var_2272_cast_fp16, var_2274_cast_fp16))[name = tensor("input_103_cast_fp16")]; + tensor var_2280 = const()[name = tensor("op_2280"), val = tensor([1, 1])]; + tensor var_2282 = const()[name = tensor("op_2282"), val = tensor([1, 1])]; + tensor var_2284_pad_type_0 = const()[name = tensor("op_2284_pad_type_0"), val = tensor("custom")]; + tensor var_2284_pad_0 = const()[name = tensor("op_2284_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_1_transformer_blocks_1_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("down_blocks_1_attentions_1_transformer_blocks_1_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(119959360)))]; + tensor down_blocks_1_attentions_1_transformer_blocks_1_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_1_attentions_1_transformer_blocks_1_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(120778624)))]; + tensor var_2284_cast_fp16 = conv(bias = down_blocks_1_attentions_1_transformer_blocks_1_attn1_to_out_0_bias_to_fp16, dilations = var_2282, groups = var_288, pad = var_2284_pad_0, pad_type = var_2284_pad_type_0, strides = var_2280, weight = down_blocks_1_attentions_1_transformer_blocks_1_attn1_to_out_0_weight_to_fp16, x = input_103_cast_fp16)[name = tensor("op_2284_cast_fp16")]; + tensor inputs_21_cast_fp16 = add(x = var_2284_cast_fp16, y = inputs_19_cast_fp16)[name = tensor("inputs_21_cast_fp16")]; + tensor hidden_states_49_axes_0 = const()[name = tensor("hidden_states_49_axes_0"), val = tensor([1])]; + tensor hidden_states_49_gamma_0_to_fp16 = const()[name = tensor("hidden_states_49_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(120779968)))]; + tensor hidden_states_49_beta_0_to_fp16 = const()[name = tensor("hidden_states_49_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(120781312)))]; + tensor var_2294_to_fp16 = const()[name = tensor("op_2294_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_49_cast_fp16 = layer_norm(axes = hidden_states_49_axes_0, beta = hidden_states_49_beta_0_to_fp16, epsilon = var_2294_to_fp16, gamma = hidden_states_49_gamma_0_to_fp16, x = inputs_21_cast_fp16)[name = tensor("hidden_states_49_cast_fp16")]; + tensor var_2309 = const()[name = tensor("op_2309"), val = tensor([1, 1])]; + tensor var_2311 = const()[name = tensor("op_2311"), val = tensor([1, 1])]; + tensor q_15_pad_type_0 = const()[name = tensor("q_15_pad_type_0"), val = tensor("custom")]; + tensor q_15_pad_0 = const()[name = tensor("q_15_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_1_transformer_blocks_1_attn2_to_q_weight_to_fp16 = const()[name = tensor("down_blocks_1_attentions_1_transformer_blocks_1_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(120782656)))]; + tensor q_15_cast_fp16 = conv(dilations = var_2311, groups = var_288, pad = q_15_pad_0, pad_type = q_15_pad_type_0, strides = var_2309, weight = down_blocks_1_attentions_1_transformer_blocks_1_attn2_to_q_weight_to_fp16, x = hidden_states_49_cast_fp16)[name = tensor("q_15_cast_fp16")]; + tensor var_2315 = const()[name = tensor("op_2315"), val = tensor([1, 1])]; + tensor var_2317 = const()[name = tensor("op_2317"), val = tensor([1, 1])]; + tensor k_29_pad_type_0 = const()[name = tensor("k_29_pad_type_0"), val = tensor("custom")]; + tensor k_29_pad_0 = const()[name = tensor("k_29_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_1_transformer_blocks_1_attn2_to_k_weight_to_fp16 = const()[name = tensor("down_blocks_1_attentions_1_transformer_blocks_1_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(121601920)))]; + tensor k_29_cast_fp16 = conv(dilations = var_2317, groups = var_288, pad = k_29_pad_0, pad_type = k_29_pad_type_0, strides = var_2315, weight = down_blocks_1_attentions_1_transformer_blocks_1_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_29_cast_fp16")]; + tensor var_2321 = const()[name = tensor("op_2321"), val = tensor([1, 1])]; + tensor var_2323 = const()[name = tensor("op_2323"), val = tensor([1, 1])]; + tensor v_15_pad_type_0 = const()[name = tensor("v_15_pad_type_0"), val = tensor("custom")]; + tensor v_15_pad_0 = const()[name = tensor("v_15_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_1_transformer_blocks_1_attn2_to_v_weight_to_fp16 = const()[name = tensor("down_blocks_1_attentions_1_transformer_blocks_1_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(124223424)))]; + tensor v_15_cast_fp16 = conv(dilations = var_2323, groups = var_288, pad = v_15_pad_0, pad_type = v_15_pad_type_0, strides = var_2321, weight = down_blocks_1_attentions_1_transformer_blocks_1_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_15_cast_fp16")]; + tensor var_2327_begin_0 = const()[name = tensor("op_2327_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2327_end_0 = const()[name = tensor("op_2327_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_2327_end_mask_0 = const()[name = tensor("op_2327_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2327_cast_fp16 = slice_by_index(begin = var_2327_begin_0, end = var_2327_end_0, end_mask = var_2327_end_mask_0, x = q_15_cast_fp16)[name = tensor("op_2327_cast_fp16")]; + tensor var_2331_begin_0 = const()[name = tensor("op_2331_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_2331_end_0 = const()[name = tensor("op_2331_end_0"), val = tensor([2, 128, 1, 4096])]; + tensor var_2331_end_mask_0 = const()[name = tensor("op_2331_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2331_cast_fp16 = slice_by_index(begin = var_2331_begin_0, end = var_2331_end_0, end_mask = var_2331_end_mask_0, x = q_15_cast_fp16)[name = tensor("op_2331_cast_fp16")]; + tensor var_2335_begin_0 = const()[name = tensor("op_2335_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_2335_end_0 = const()[name = tensor("op_2335_end_0"), val = tensor([2, 192, 1, 4096])]; + tensor var_2335_end_mask_0 = const()[name = tensor("op_2335_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2335_cast_fp16 = slice_by_index(begin = var_2335_begin_0, end = var_2335_end_0, end_mask = var_2335_end_mask_0, x = q_15_cast_fp16)[name = tensor("op_2335_cast_fp16")]; + tensor var_2339_begin_0 = const()[name = tensor("op_2339_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_2339_end_0 = const()[name = tensor("op_2339_end_0"), val = tensor([2, 256, 1, 4096])]; + tensor var_2339_end_mask_0 = const()[name = tensor("op_2339_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2339_cast_fp16 = slice_by_index(begin = var_2339_begin_0, end = var_2339_end_0, end_mask = var_2339_end_mask_0, x = q_15_cast_fp16)[name = tensor("op_2339_cast_fp16")]; + tensor var_2343_begin_0 = const()[name = tensor("op_2343_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_2343_end_0 = const()[name = tensor("op_2343_end_0"), val = tensor([2, 320, 1, 4096])]; + tensor var_2343_end_mask_0 = const()[name = tensor("op_2343_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2343_cast_fp16 = slice_by_index(begin = var_2343_begin_0, end = var_2343_end_0, end_mask = var_2343_end_mask_0, x = q_15_cast_fp16)[name = tensor("op_2343_cast_fp16")]; + tensor var_2347_begin_0 = const()[name = tensor("op_2347_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_2347_end_0 = const()[name = tensor("op_2347_end_0"), val = tensor([2, 384, 1, 4096])]; + tensor var_2347_end_mask_0 = const()[name = tensor("op_2347_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2347_cast_fp16 = slice_by_index(begin = var_2347_begin_0, end = var_2347_end_0, end_mask = var_2347_end_mask_0, x = q_15_cast_fp16)[name = tensor("op_2347_cast_fp16")]; + tensor var_2351_begin_0 = const()[name = tensor("op_2351_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_2351_end_0 = const()[name = tensor("op_2351_end_0"), val = tensor([2, 448, 1, 4096])]; + tensor var_2351_end_mask_0 = const()[name = tensor("op_2351_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2351_cast_fp16 = slice_by_index(begin = var_2351_begin_0, end = var_2351_end_0, end_mask = var_2351_end_mask_0, x = q_15_cast_fp16)[name = tensor("op_2351_cast_fp16")]; + tensor var_2355_begin_0 = const()[name = tensor("op_2355_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_2355_end_0 = const()[name = tensor("op_2355_end_0"), val = tensor([2, 512, 1, 4096])]; + tensor var_2355_end_mask_0 = const()[name = tensor("op_2355_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2355_cast_fp16 = slice_by_index(begin = var_2355_begin_0, end = var_2355_end_0, end_mask = var_2355_end_mask_0, x = q_15_cast_fp16)[name = tensor("op_2355_cast_fp16")]; + tensor var_2359_begin_0 = const()[name = tensor("op_2359_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_2359_end_0 = const()[name = tensor("op_2359_end_0"), val = tensor([2, 576, 1, 4096])]; + tensor var_2359_end_mask_0 = const()[name = tensor("op_2359_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2359_cast_fp16 = slice_by_index(begin = var_2359_begin_0, end = var_2359_end_0, end_mask = var_2359_end_mask_0, x = q_15_cast_fp16)[name = tensor("op_2359_cast_fp16")]; + tensor var_2363_begin_0 = const()[name = tensor("op_2363_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_2363_end_0 = const()[name = tensor("op_2363_end_0"), val = tensor([2, 640, 1, 4096])]; + tensor var_2363_end_mask_0 = const()[name = tensor("op_2363_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2363_cast_fp16 = slice_by_index(begin = var_2363_begin_0, end = var_2363_end_0, end_mask = var_2363_end_mask_0, x = q_15_cast_fp16)[name = tensor("op_2363_cast_fp16")]; + tensor k_31_perm_0 = const()[name = tensor("k_31_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_2370_begin_0 = const()[name = tensor("op_2370_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2370_end_0 = const()[name = tensor("op_2370_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_2370_end_mask_0 = const()[name = tensor("op_2370_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_31_cast_fp16 = transpose(perm = k_31_perm_0, x = k_29_cast_fp16)[name = tensor("transpose_132")]; + tensor var_2370_cast_fp16 = slice_by_index(begin = var_2370_begin_0, end = var_2370_end_0, end_mask = var_2370_end_mask_0, x = k_31_cast_fp16)[name = tensor("op_2370_cast_fp16")]; + tensor var_2374_begin_0 = const()[name = tensor("op_2374_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_2374_end_0 = const()[name = tensor("op_2374_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_2374_end_mask_0 = const()[name = tensor("op_2374_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2374_cast_fp16 = slice_by_index(begin = var_2374_begin_0, end = var_2374_end_0, end_mask = var_2374_end_mask_0, x = k_31_cast_fp16)[name = tensor("op_2374_cast_fp16")]; + tensor var_2378_begin_0 = const()[name = tensor("op_2378_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_2378_end_0 = const()[name = tensor("op_2378_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_2378_end_mask_0 = const()[name = tensor("op_2378_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2378_cast_fp16 = slice_by_index(begin = var_2378_begin_0, end = var_2378_end_0, end_mask = var_2378_end_mask_0, x = k_31_cast_fp16)[name = tensor("op_2378_cast_fp16")]; + tensor var_2382_begin_0 = const()[name = tensor("op_2382_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_2382_end_0 = const()[name = tensor("op_2382_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_2382_end_mask_0 = const()[name = tensor("op_2382_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2382_cast_fp16 = slice_by_index(begin = var_2382_begin_0, end = var_2382_end_0, end_mask = var_2382_end_mask_0, x = k_31_cast_fp16)[name = tensor("op_2382_cast_fp16")]; + tensor var_2386_begin_0 = const()[name = tensor("op_2386_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_2386_end_0 = const()[name = tensor("op_2386_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_2386_end_mask_0 = const()[name = tensor("op_2386_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2386_cast_fp16 = slice_by_index(begin = var_2386_begin_0, end = var_2386_end_0, end_mask = var_2386_end_mask_0, x = k_31_cast_fp16)[name = tensor("op_2386_cast_fp16")]; + tensor var_2390_begin_0 = const()[name = tensor("op_2390_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_2390_end_0 = const()[name = tensor("op_2390_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_2390_end_mask_0 = const()[name = tensor("op_2390_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2390_cast_fp16 = slice_by_index(begin = var_2390_begin_0, end = var_2390_end_0, end_mask = var_2390_end_mask_0, x = k_31_cast_fp16)[name = tensor("op_2390_cast_fp16")]; + tensor var_2394_begin_0 = const()[name = tensor("op_2394_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_2394_end_0 = const()[name = tensor("op_2394_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_2394_end_mask_0 = const()[name = tensor("op_2394_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2394_cast_fp16 = slice_by_index(begin = var_2394_begin_0, end = var_2394_end_0, end_mask = var_2394_end_mask_0, x = k_31_cast_fp16)[name = tensor("op_2394_cast_fp16")]; + tensor var_2398_begin_0 = const()[name = tensor("op_2398_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_2398_end_0 = const()[name = tensor("op_2398_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_2398_end_mask_0 = const()[name = tensor("op_2398_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2398_cast_fp16 = slice_by_index(begin = var_2398_begin_0, end = var_2398_end_0, end_mask = var_2398_end_mask_0, x = k_31_cast_fp16)[name = tensor("op_2398_cast_fp16")]; + tensor var_2402_begin_0 = const()[name = tensor("op_2402_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_2402_end_0 = const()[name = tensor("op_2402_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_2402_end_mask_0 = const()[name = tensor("op_2402_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2402_cast_fp16 = slice_by_index(begin = var_2402_begin_0, end = var_2402_end_0, end_mask = var_2402_end_mask_0, x = k_31_cast_fp16)[name = tensor("op_2402_cast_fp16")]; + tensor var_2406_begin_0 = const()[name = tensor("op_2406_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_2406_end_0 = const()[name = tensor("op_2406_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_2406_end_mask_0 = const()[name = tensor("op_2406_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2406_cast_fp16 = slice_by_index(begin = var_2406_begin_0, end = var_2406_end_0, end_mask = var_2406_end_mask_0, x = k_31_cast_fp16)[name = tensor("op_2406_cast_fp16")]; + tensor var_2408_begin_0 = const()[name = tensor("op_2408_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2408_end_0 = const()[name = tensor("op_2408_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_2408_end_mask_0 = const()[name = tensor("op_2408_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2408_cast_fp16 = slice_by_index(begin = var_2408_begin_0, end = var_2408_end_0, end_mask = var_2408_end_mask_0, x = v_15_cast_fp16)[name = tensor("op_2408_cast_fp16")]; + tensor var_2412_begin_0 = const()[name = tensor("op_2412_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_2412_end_0 = const()[name = tensor("op_2412_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_2412_end_mask_0 = const()[name = tensor("op_2412_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2412_cast_fp16 = slice_by_index(begin = var_2412_begin_0, end = var_2412_end_0, end_mask = var_2412_end_mask_0, x = v_15_cast_fp16)[name = tensor("op_2412_cast_fp16")]; + tensor var_2416_begin_0 = const()[name = tensor("op_2416_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_2416_end_0 = const()[name = tensor("op_2416_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_2416_end_mask_0 = const()[name = tensor("op_2416_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2416_cast_fp16 = slice_by_index(begin = var_2416_begin_0, end = var_2416_end_0, end_mask = var_2416_end_mask_0, x = v_15_cast_fp16)[name = tensor("op_2416_cast_fp16")]; + tensor var_2420_begin_0 = const()[name = tensor("op_2420_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_2420_end_0 = const()[name = tensor("op_2420_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_2420_end_mask_0 = const()[name = tensor("op_2420_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2420_cast_fp16 = slice_by_index(begin = var_2420_begin_0, end = var_2420_end_0, end_mask = var_2420_end_mask_0, x = v_15_cast_fp16)[name = tensor("op_2420_cast_fp16")]; + tensor var_2424_begin_0 = const()[name = tensor("op_2424_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_2424_end_0 = const()[name = tensor("op_2424_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_2424_end_mask_0 = const()[name = tensor("op_2424_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2424_cast_fp16 = slice_by_index(begin = var_2424_begin_0, end = var_2424_end_0, end_mask = var_2424_end_mask_0, x = v_15_cast_fp16)[name = tensor("op_2424_cast_fp16")]; + tensor var_2428_begin_0 = const()[name = tensor("op_2428_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_2428_end_0 = const()[name = tensor("op_2428_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_2428_end_mask_0 = const()[name = tensor("op_2428_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2428_cast_fp16 = slice_by_index(begin = var_2428_begin_0, end = var_2428_end_0, end_mask = var_2428_end_mask_0, x = v_15_cast_fp16)[name = tensor("op_2428_cast_fp16")]; + tensor var_2432_begin_0 = const()[name = tensor("op_2432_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_2432_end_0 = const()[name = tensor("op_2432_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_2432_end_mask_0 = const()[name = tensor("op_2432_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2432_cast_fp16 = slice_by_index(begin = var_2432_begin_0, end = var_2432_end_0, end_mask = var_2432_end_mask_0, x = v_15_cast_fp16)[name = tensor("op_2432_cast_fp16")]; + tensor var_2436_begin_0 = const()[name = tensor("op_2436_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_2436_end_0 = const()[name = tensor("op_2436_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_2436_end_mask_0 = const()[name = tensor("op_2436_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2436_cast_fp16 = slice_by_index(begin = var_2436_begin_0, end = var_2436_end_0, end_mask = var_2436_end_mask_0, x = v_15_cast_fp16)[name = tensor("op_2436_cast_fp16")]; + tensor var_2440_begin_0 = const()[name = tensor("op_2440_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_2440_end_0 = const()[name = tensor("op_2440_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_2440_end_mask_0 = const()[name = tensor("op_2440_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2440_cast_fp16 = slice_by_index(begin = var_2440_begin_0, end = var_2440_end_0, end_mask = var_2440_end_mask_0, x = v_15_cast_fp16)[name = tensor("op_2440_cast_fp16")]; + tensor var_2444_begin_0 = const()[name = tensor("op_2444_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_2444_end_0 = const()[name = tensor("op_2444_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_2444_end_mask_0 = const()[name = tensor("op_2444_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2444_cast_fp16 = slice_by_index(begin = var_2444_begin_0, end = var_2444_end_0, end_mask = var_2444_end_mask_0, x = v_15_cast_fp16)[name = tensor("op_2444_cast_fp16")]; + tensor var_2448_equation_0 = const()[name = tensor("op_2448_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2448_cast_fp16 = einsum(equation = var_2448_equation_0, values = (var_2370_cast_fp16, var_2327_cast_fp16))[name = tensor("op_2448_cast_fp16")]; + tensor var_2449_to_fp16 = const()[name = tensor("op_2449_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_141_cast_fp16 = mul(x = var_2448_cast_fp16, y = var_2449_to_fp16)[name = tensor("aw_141_cast_fp16")]; + tensor var_2452_equation_0 = const()[name = tensor("op_2452_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2452_cast_fp16 = einsum(equation = var_2452_equation_0, values = (var_2374_cast_fp16, var_2331_cast_fp16))[name = tensor("op_2452_cast_fp16")]; + tensor var_2453_to_fp16 = const()[name = tensor("op_2453_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_143_cast_fp16 = mul(x = var_2452_cast_fp16, y = var_2453_to_fp16)[name = tensor("aw_143_cast_fp16")]; + tensor var_2456_equation_0 = const()[name = tensor("op_2456_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2456_cast_fp16 = einsum(equation = var_2456_equation_0, values = (var_2378_cast_fp16, var_2335_cast_fp16))[name = tensor("op_2456_cast_fp16")]; + tensor var_2457_to_fp16 = const()[name = tensor("op_2457_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_145_cast_fp16 = mul(x = var_2456_cast_fp16, y = var_2457_to_fp16)[name = tensor("aw_145_cast_fp16")]; + tensor var_2460_equation_0 = const()[name = tensor("op_2460_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2460_cast_fp16 = einsum(equation = var_2460_equation_0, values = (var_2382_cast_fp16, var_2339_cast_fp16))[name = tensor("op_2460_cast_fp16")]; + tensor var_2461_to_fp16 = const()[name = tensor("op_2461_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_147_cast_fp16 = mul(x = var_2460_cast_fp16, y = var_2461_to_fp16)[name = tensor("aw_147_cast_fp16")]; + tensor var_2464_equation_0 = const()[name = tensor("op_2464_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2464_cast_fp16 = einsum(equation = var_2464_equation_0, values = (var_2386_cast_fp16, var_2343_cast_fp16))[name = tensor("op_2464_cast_fp16")]; + tensor var_2465_to_fp16 = const()[name = tensor("op_2465_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_149_cast_fp16 = mul(x = var_2464_cast_fp16, y = var_2465_to_fp16)[name = tensor("aw_149_cast_fp16")]; + tensor var_2468_equation_0 = const()[name = tensor("op_2468_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2468_cast_fp16 = einsum(equation = var_2468_equation_0, values = (var_2390_cast_fp16, var_2347_cast_fp16))[name = tensor("op_2468_cast_fp16")]; + tensor var_2469_to_fp16 = const()[name = tensor("op_2469_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_151_cast_fp16 = mul(x = var_2468_cast_fp16, y = var_2469_to_fp16)[name = tensor("aw_151_cast_fp16")]; + tensor var_2472_equation_0 = const()[name = tensor("op_2472_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2472_cast_fp16 = einsum(equation = var_2472_equation_0, values = (var_2394_cast_fp16, var_2351_cast_fp16))[name = tensor("op_2472_cast_fp16")]; + tensor var_2473_to_fp16 = const()[name = tensor("op_2473_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_153_cast_fp16 = mul(x = var_2472_cast_fp16, y = var_2473_to_fp16)[name = tensor("aw_153_cast_fp16")]; + tensor var_2476_equation_0 = const()[name = tensor("op_2476_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2476_cast_fp16 = einsum(equation = var_2476_equation_0, values = (var_2398_cast_fp16, var_2355_cast_fp16))[name = tensor("op_2476_cast_fp16")]; + tensor var_2477_to_fp16 = const()[name = tensor("op_2477_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_155_cast_fp16 = mul(x = var_2476_cast_fp16, y = var_2477_to_fp16)[name = tensor("aw_155_cast_fp16")]; + tensor var_2480_equation_0 = const()[name = tensor("op_2480_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2480_cast_fp16 = einsum(equation = var_2480_equation_0, values = (var_2402_cast_fp16, var_2359_cast_fp16))[name = tensor("op_2480_cast_fp16")]; + tensor var_2481_to_fp16 = const()[name = tensor("op_2481_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_157_cast_fp16 = mul(x = var_2480_cast_fp16, y = var_2481_to_fp16)[name = tensor("aw_157_cast_fp16")]; + tensor var_2484_equation_0 = const()[name = tensor("op_2484_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2484_cast_fp16 = einsum(equation = var_2484_equation_0, values = (var_2406_cast_fp16, var_2363_cast_fp16))[name = tensor("op_2484_cast_fp16")]; + tensor var_2485_to_fp16 = const()[name = tensor("op_2485_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_159_cast_fp16 = mul(x = var_2484_cast_fp16, y = var_2485_to_fp16)[name = tensor("aw_159_cast_fp16")]; + tensor var_2487_cast_fp16 = softmax(axis = var_288, x = aw_141_cast_fp16)[name = tensor("op_2487_cast_fp16")]; + tensor var_2488_cast_fp16 = softmax(axis = var_288, x = aw_143_cast_fp16)[name = tensor("op_2488_cast_fp16")]; + tensor var_2489_cast_fp16 = softmax(axis = var_288, x = aw_145_cast_fp16)[name = tensor("op_2489_cast_fp16")]; + tensor var_2490_cast_fp16 = softmax(axis = var_288, x = aw_147_cast_fp16)[name = tensor("op_2490_cast_fp16")]; + tensor var_2491_cast_fp16 = softmax(axis = var_288, x = aw_149_cast_fp16)[name = tensor("op_2491_cast_fp16")]; + tensor var_2492_cast_fp16 = softmax(axis = var_288, x = aw_151_cast_fp16)[name = tensor("op_2492_cast_fp16")]; + tensor var_2493_cast_fp16 = softmax(axis = var_288, x = aw_153_cast_fp16)[name = tensor("op_2493_cast_fp16")]; + tensor var_2494_cast_fp16 = softmax(axis = var_288, x = aw_155_cast_fp16)[name = tensor("op_2494_cast_fp16")]; + tensor var_2495_cast_fp16 = softmax(axis = var_288, x = aw_157_cast_fp16)[name = tensor("op_2495_cast_fp16")]; + tensor var_2496_cast_fp16 = softmax(axis = var_288, x = aw_159_cast_fp16)[name = tensor("op_2496_cast_fp16")]; + tensor var_2498_equation_0 = const()[name = tensor("op_2498_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2498_cast_fp16 = einsum(equation = var_2498_equation_0, values = (var_2408_cast_fp16, var_2487_cast_fp16))[name = tensor("op_2498_cast_fp16")]; + tensor var_2500_equation_0 = const()[name = tensor("op_2500_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2500_cast_fp16 = einsum(equation = var_2500_equation_0, values = (var_2412_cast_fp16, var_2488_cast_fp16))[name = tensor("op_2500_cast_fp16")]; + tensor var_2502_equation_0 = const()[name = tensor("op_2502_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2502_cast_fp16 = einsum(equation = var_2502_equation_0, values = (var_2416_cast_fp16, var_2489_cast_fp16))[name = tensor("op_2502_cast_fp16")]; + tensor var_2504_equation_0 = const()[name = tensor("op_2504_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2504_cast_fp16 = einsum(equation = var_2504_equation_0, values = (var_2420_cast_fp16, var_2490_cast_fp16))[name = tensor("op_2504_cast_fp16")]; + tensor var_2506_equation_0 = const()[name = tensor("op_2506_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2506_cast_fp16 = einsum(equation = var_2506_equation_0, values = (var_2424_cast_fp16, var_2491_cast_fp16))[name = tensor("op_2506_cast_fp16")]; + tensor var_2508_equation_0 = const()[name = tensor("op_2508_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2508_cast_fp16 = einsum(equation = var_2508_equation_0, values = (var_2428_cast_fp16, var_2492_cast_fp16))[name = tensor("op_2508_cast_fp16")]; + tensor var_2510_equation_0 = const()[name = tensor("op_2510_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2510_cast_fp16 = einsum(equation = var_2510_equation_0, values = (var_2432_cast_fp16, var_2493_cast_fp16))[name = tensor("op_2510_cast_fp16")]; + tensor var_2512_equation_0 = const()[name = tensor("op_2512_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2512_cast_fp16 = einsum(equation = var_2512_equation_0, values = (var_2436_cast_fp16, var_2494_cast_fp16))[name = tensor("op_2512_cast_fp16")]; + tensor var_2514_equation_0 = const()[name = tensor("op_2514_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2514_cast_fp16 = einsum(equation = var_2514_equation_0, values = (var_2440_cast_fp16, var_2495_cast_fp16))[name = tensor("op_2514_cast_fp16")]; + tensor var_2516_equation_0 = const()[name = tensor("op_2516_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2516_cast_fp16 = einsum(equation = var_2516_equation_0, values = (var_2444_cast_fp16, var_2496_cast_fp16))[name = tensor("op_2516_cast_fp16")]; + tensor input_105_interleave_0 = const()[name = tensor("input_105_interleave_0"), val = tensor(false)]; + tensor input_105_cast_fp16 = concat(axis = var_288, interleave = input_105_interleave_0, values = (var_2498_cast_fp16, var_2500_cast_fp16, var_2502_cast_fp16, var_2504_cast_fp16, var_2506_cast_fp16, var_2508_cast_fp16, var_2510_cast_fp16, var_2512_cast_fp16, var_2514_cast_fp16, var_2516_cast_fp16))[name = tensor("input_105_cast_fp16")]; + tensor var_2522 = const()[name = tensor("op_2522"), val = tensor([1, 1])]; + tensor var_2524 = const()[name = tensor("op_2524"), val = tensor([1, 1])]; + tensor var_2526_pad_type_0 = const()[name = tensor("op_2526_pad_type_0"), val = tensor("custom")]; + tensor var_2526_pad_0 = const()[name = tensor("op_2526_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_1_transformer_blocks_1_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("down_blocks_1_attentions_1_transformer_blocks_1_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(126844928)))]; + tensor down_blocks_1_attentions_1_transformer_blocks_1_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_1_attentions_1_transformer_blocks_1_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(127664192)))]; + tensor var_2526_cast_fp16 = conv(bias = down_blocks_1_attentions_1_transformer_blocks_1_attn2_to_out_0_bias_to_fp16, dilations = var_2524, groups = var_288, pad = var_2526_pad_0, pad_type = var_2526_pad_type_0, strides = var_2522, weight = down_blocks_1_attentions_1_transformer_blocks_1_attn2_to_out_0_weight_to_fp16, x = input_105_cast_fp16)[name = tensor("op_2526_cast_fp16")]; + tensor inputs_23_cast_fp16 = add(x = var_2526_cast_fp16, y = inputs_21_cast_fp16)[name = tensor("inputs_23_cast_fp16")]; + tensor input_107_axes_0 = const()[name = tensor("input_107_axes_0"), val = tensor([1])]; + tensor input_107_gamma_0_to_fp16 = const()[name = tensor("input_107_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(127665536)))]; + tensor input_107_beta_0_to_fp16 = const()[name = tensor("input_107_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(127666880)))]; + tensor var_2536_to_fp16 = const()[name = tensor("op_2536_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_107_cast_fp16 = layer_norm(axes = input_107_axes_0, beta = input_107_beta_0_to_fp16, epsilon = var_2536_to_fp16, gamma = input_107_gamma_0_to_fp16, x = inputs_23_cast_fp16)[name = tensor("input_107_cast_fp16")]; + tensor var_2552 = const()[name = tensor("op_2552"), val = tensor([1, 1])]; + tensor var_2554 = const()[name = tensor("op_2554"), val = tensor([1, 1])]; + tensor var_2556_pad_type_0 = const()[name = tensor("op_2556_pad_type_0"), val = tensor("custom")]; + tensor var_2556_pad_0 = const()[name = tensor("op_2556_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_1_transformer_blocks_1_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("down_blocks_1_attentions_1_transformer_blocks_1_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(127668224)))]; + tensor down_blocks_1_attentions_1_transformer_blocks_1_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("down_blocks_1_attentions_1_transformer_blocks_1_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(134221888)))]; + tensor var_2556_cast_fp16 = conv(bias = down_blocks_1_attentions_1_transformer_blocks_1_ff_net_0_proj_bias_to_fp16, dilations = var_2554, groups = var_288, pad = var_2556_pad_0, pad_type = var_2556_pad_type_0, strides = var_2552, weight = down_blocks_1_attentions_1_transformer_blocks_1_ff_net_0_proj_weight_to_fp16, x = input_107_cast_fp16)[name = tensor("op_2556_cast_fp16")]; + tensor var_2557_split_sizes_0 = const()[name = tensor("op_2557_split_sizes_0"), val = tensor([2560, 2560])]; + tensor var_2557_axis_0 = const()[name = tensor("op_2557_axis_0"), val = tensor(1)]; + tensor var_2557_cast_fp16_0, tensor var_2557_cast_fp16_1 = split(axis = var_2557_axis_0, split_sizes = var_2557_split_sizes_0, x = var_2556_cast_fp16)[name = tensor("op_2557_cast_fp16")]; + tensor var_2559_mode_0 = const()[name = tensor("op_2559_mode_0"), val = tensor("EXACT")]; + tensor var_2559_cast_fp16 = gelu(mode = var_2559_mode_0, x = var_2557_cast_fp16_1)[name = tensor("op_2559_cast_fp16")]; + tensor input_109_cast_fp16 = mul(x = var_2557_cast_fp16_0, y = var_2559_cast_fp16)[name = tensor("input_109_cast_fp16")]; + tensor var_2563 = const()[name = tensor("op_2563"), val = tensor([1, 1])]; + tensor var_2565 = const()[name = tensor("op_2565"), val = tensor([1, 1])]; + tensor var_2567_pad_type_0 = const()[name = tensor("op_2567_pad_type_0"), val = tensor("custom")]; + tensor var_2567_pad_0 = const()[name = tensor("op_2567_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_1_transformer_blocks_1_ff_net_2_weight_to_fp16 = const()[name = tensor("down_blocks_1_attentions_1_transformer_blocks_1_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(134232192)))]; + tensor down_blocks_1_attentions_1_transformer_blocks_1_ff_net_2_bias_to_fp16 = const()[name = tensor("down_blocks_1_attentions_1_transformer_blocks_1_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(137509056)))]; + tensor var_2567_cast_fp16 = conv(bias = down_blocks_1_attentions_1_transformer_blocks_1_ff_net_2_bias_to_fp16, dilations = var_2565, groups = var_288, pad = var_2567_pad_0, pad_type = var_2567_pad_type_0, strides = var_2563, weight = down_blocks_1_attentions_1_transformer_blocks_1_ff_net_2_weight_to_fp16, x = input_109_cast_fp16)[name = tensor("op_2567_cast_fp16")]; + tensor hidden_states_53_cast_fp16 = add(x = var_2567_cast_fp16, y = inputs_23_cast_fp16)[name = tensor("hidden_states_53_cast_fp16")]; + tensor var_2569 = const()[name = tensor("op_2569"), val = tensor([2, 640, 64, 64])]; + tensor input_111_cast_fp16 = reshape(shape = var_2569, x = hidden_states_53_cast_fp16)[name = tensor("input_111_cast_fp16")]; + tensor var_2573 = const()[name = tensor("op_2573"), val = tensor([1, 1])]; + tensor var_2575 = const()[name = tensor("op_2575"), val = tensor([1, 1])]; + tensor hidden_states_55_pad_type_0 = const()[name = tensor("hidden_states_55_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_55_pad_0 = const()[name = tensor("hidden_states_55_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_1_proj_out_weight_to_fp16 = const()[name = tensor("down_blocks_1_attentions_1_proj_out_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(137510400)))]; + tensor down_blocks_1_attentions_1_proj_out_bias_to_fp16 = const()[name = tensor("down_blocks_1_attentions_1_proj_out_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(138329664)))]; + tensor hidden_states_55_cast_fp16 = conv(bias = down_blocks_1_attentions_1_proj_out_bias_to_fp16, dilations = var_2575, groups = var_288, pad = hidden_states_55_pad_0, pad_type = hidden_states_55_pad_type_0, strides = var_2573, weight = down_blocks_1_attentions_1_proj_out_weight_to_fp16, x = input_111_cast_fp16)[name = tensor("hidden_states_55_cast_fp16")]; + tensor input_113_cast_fp16 = add(x = hidden_states_55_cast_fp16, y = hidden_states_37_cast_fp16)[name = tensor("input_113_cast_fp16")]; + tensor var_2582 = const()[name = tensor("op_2582"), val = tensor([2, 2])]; + tensor var_2584 = const()[name = tensor("op_2584"), val = tensor([1, 1])]; + tensor input_115_pad_type_0 = const()[name = tensor("input_115_pad_type_0"), val = tensor("custom")]; + tensor input_115_pad_0 = const()[name = tensor("input_115_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor down_blocks_1_downsamplers_0_conv_weight_to_fp16 = const()[name = tensor("down_blocks_1_downsamplers_0_conv_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(138331008)))]; + tensor down_blocks_1_downsamplers_0_conv_bias_to_fp16 = const()[name = tensor("down_blocks_1_downsamplers_0_conv_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(145703872)))]; + tensor input_115_cast_fp16 = conv(bias = down_blocks_1_downsamplers_0_conv_bias_to_fp16, dilations = var_2584, groups = var_288, pad = input_115_pad_0, pad_type = input_115_pad_type_0, strides = var_2582, weight = down_blocks_1_downsamplers_0_conv_weight_to_fp16, x = input_113_cast_fp16)[name = tensor("input_115_cast_fp16")]; + tensor var_2624 = const()[name = tensor("op_2624"), val = tensor(1)]; + tensor reshape_40_shape_0 = const()[name = tensor("reshape_40_shape_0"), val = tensor([2, 32, 20, 32, 32])]; + tensor reshape_40_cast_fp16 = reshape(shape = reshape_40_shape_0, x = input_115_cast_fp16)[name = tensor("reshape_40_cast_fp16")]; + tensor reduce_mean_30_axes_0 = const()[name = tensor("reduce_mean_30_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_30_keep_dims_0 = const()[name = tensor("reduce_mean_30_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_30_cast_fp16 = reduce_mean(axes = reduce_mean_30_axes_0, keep_dims = reduce_mean_30_keep_dims_0, x = reshape_40_cast_fp16)[name = tensor("reduce_mean_30_cast_fp16")]; + tensor sub_20_cast_fp16 = sub(x = reshape_40_cast_fp16, y = reduce_mean_30_cast_fp16)[name = tensor("sub_20_cast_fp16")]; + tensor square_10_cast_fp16 = square(x = sub_20_cast_fp16)[name = tensor("square_10_cast_fp16")]; + tensor reduce_mean_32_axes_0 = const()[name = tensor("reduce_mean_32_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_32_keep_dims_0 = const()[name = tensor("reduce_mean_32_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_32_cast_fp16 = reduce_mean(axes = reduce_mean_32_axes_0, keep_dims = reduce_mean_32_keep_dims_0, x = square_10_cast_fp16)[name = tensor("reduce_mean_32_cast_fp16")]; + tensor add_20_y_0_to_fp16 = const()[name = tensor("add_20_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_20_cast_fp16 = add(x = reduce_mean_32_cast_fp16, y = add_20_y_0_to_fp16)[name = tensor("add_20_cast_fp16")]; + tensor sqrt_10_cast_fp16 = sqrt(x = add_20_cast_fp16)[name = tensor("sqrt_10_cast_fp16")]; + tensor real_div_10_cast_fp16 = real_div(x = sub_20_cast_fp16, y = sqrt_10_cast_fp16)[name = tensor("real_div_10_cast_fp16")]; + tensor reshape_41_shape_0 = const()[name = tensor("reshape_41_shape_0"), val = tensor([2, 640, 32, 32])]; + tensor reshape_41_cast_fp16 = reshape(shape = reshape_41_shape_0, x = real_div_10_cast_fp16)[name = tensor("reshape_41_cast_fp16")]; + tensor add_21_gamma_0_to_fp16 = const()[name = tensor("add_21_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(145705216)))]; + tensor add_21_beta_0_to_fp16 = const()[name = tensor("add_21_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(145706560)))]; + tensor add_21_epsilon_0_to_fp16 = const()[name = tensor("add_21_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_21_cast_fp16 = batch_norm(beta = add_21_beta_0_to_fp16, epsilon = add_21_epsilon_0_to_fp16, gamma = add_21_gamma_0_to_fp16, mean = add_11_mean_0_to_fp16, variance = add_11_variance_0_to_fp16, x = reshape_41_cast_fp16)[name = tensor("add_21_cast_fp16")]; + tensor input_119_cast_fp16 = silu(x = add_21_cast_fp16)[name = tensor("input_119_cast_fp16")]; + tensor var_2645 = const()[name = tensor("op_2645"), val = tensor([1, 1])]; + tensor var_2647 = const()[name = tensor("op_2647"), val = tensor([1, 1])]; + tensor hidden_states_57_pad_type_0 = const()[name = tensor("hidden_states_57_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_57_pad_0 = const()[name = tensor("hidden_states_57_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor down_blocks_2_resnets_0_conv1_weight_to_fp16 = const()[name = tensor("down_blocks_2_resnets_0_conv1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(145707904)))]; + tensor down_blocks_2_resnets_0_conv1_bias_to_fp16 = const()[name = tensor("down_blocks_2_resnets_0_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(160453568)))]; + tensor hidden_states_57_cast_fp16 = conv(bias = down_blocks_2_resnets_0_conv1_bias_to_fp16, dilations = var_2647, groups = var_2624, pad = hidden_states_57_pad_0, pad_type = hidden_states_57_pad_type_0, strides = var_2645, weight = down_blocks_2_resnets_0_conv1_weight_to_fp16, x = input_119_cast_fp16)[name = tensor("hidden_states_57_cast_fp16")]; + tensor var_2653 = const()[name = tensor("op_2653"), val = tensor([1, 1])]; + tensor var_2655 = const()[name = tensor("op_2655"), val = tensor([1, 1])]; + tensor temb_9_pad_type_0 = const()[name = tensor("temb_9_pad_type_0"), val = tensor("custom")]; + tensor temb_9_pad_0 = const()[name = tensor("temb_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_resnets_0_time_emb_proj_weight_to_fp16 = const()[name = tensor("down_blocks_2_resnets_0_time_emb_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(160456192)))]; + tensor down_blocks_2_resnets_0_time_emb_proj_bias_to_fp16 = const()[name = tensor("down_blocks_2_resnets_0_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(163733056)))]; + tensor temb_9_cast_fp16 = conv(bias = down_blocks_2_resnets_0_time_emb_proj_bias_to_fp16, dilations = var_2655, groups = var_2624, pad = temb_9_pad_0, pad_type = temb_9_pad_type_0, strides = var_2653, weight = down_blocks_2_resnets_0_time_emb_proj_weight_to_fp16, x = input_21_cast_fp16)[name = tensor("temb_9_cast_fp16")]; + tensor input_123_cast_fp16 = add(x = hidden_states_57_cast_fp16, y = temb_9_cast_fp16)[name = tensor("input_123_cast_fp16")]; + tensor reshape_44_shape_0 = const()[name = tensor("reshape_44_shape_0"), val = tensor([2, 32, 40, 32, 32])]; + tensor reshape_44_cast_fp16 = reshape(shape = reshape_44_shape_0, x = input_123_cast_fp16)[name = tensor("reshape_44_cast_fp16")]; + tensor reduce_mean_33_axes_0 = const()[name = tensor("reduce_mean_33_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_33_keep_dims_0 = const()[name = tensor("reduce_mean_33_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_33_cast_fp16 = reduce_mean(axes = reduce_mean_33_axes_0, keep_dims = reduce_mean_33_keep_dims_0, x = reshape_44_cast_fp16)[name = tensor("reduce_mean_33_cast_fp16")]; + tensor sub_22_cast_fp16 = sub(x = reshape_44_cast_fp16, y = reduce_mean_33_cast_fp16)[name = tensor("sub_22_cast_fp16")]; + tensor square_11_cast_fp16 = square(x = sub_22_cast_fp16)[name = tensor("square_11_cast_fp16")]; + tensor reduce_mean_35_axes_0 = const()[name = tensor("reduce_mean_35_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_35_keep_dims_0 = const()[name = tensor("reduce_mean_35_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_35_cast_fp16 = reduce_mean(axes = reduce_mean_35_axes_0, keep_dims = reduce_mean_35_keep_dims_0, x = square_11_cast_fp16)[name = tensor("reduce_mean_35_cast_fp16")]; + tensor add_22_y_0_to_fp16 = const()[name = tensor("add_22_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_22_cast_fp16 = add(x = reduce_mean_35_cast_fp16, y = add_22_y_0_to_fp16)[name = tensor("add_22_cast_fp16")]; + tensor sqrt_11_cast_fp16 = sqrt(x = add_22_cast_fp16)[name = tensor("sqrt_11_cast_fp16")]; + tensor real_div_11_cast_fp16 = real_div(x = sub_22_cast_fp16, y = sqrt_11_cast_fp16)[name = tensor("real_div_11_cast_fp16")]; + tensor reshape_45_shape_0 = const()[name = tensor("reshape_45_shape_0"), val = tensor([2, 1280, 32, 32])]; + tensor reshape_45_cast_fp16 = reshape(shape = reshape_45_shape_0, x = real_div_11_cast_fp16)[name = tensor("reshape_45_cast_fp16")]; + tensor add_23_mean_0_to_fp16 = const()[name = tensor("add_23_mean_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(163735680)))]; + tensor add_23_variance_0_to_fp16 = const()[name = tensor("add_23_variance_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(163738304)))]; + tensor add_23_gamma_0_to_fp16 = const()[name = tensor("add_23_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(163740928)))]; + tensor add_23_beta_0_to_fp16 = const()[name = tensor("add_23_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(163743552)))]; + tensor add_23_epsilon_0_to_fp16 = const()[name = tensor("add_23_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_23_cast_fp16 = batch_norm(beta = add_23_beta_0_to_fp16, epsilon = add_23_epsilon_0_to_fp16, gamma = add_23_gamma_0_to_fp16, mean = add_23_mean_0_to_fp16, variance = add_23_variance_0_to_fp16, x = reshape_45_cast_fp16)[name = tensor("add_23_cast_fp16")]; + tensor input_127_cast_fp16 = silu(x = add_23_cast_fp16)[name = tensor("input_127_cast_fp16")]; + tensor var_2665 = const()[name = tensor("op_2665"), val = tensor([1, 1])]; + tensor var_2667 = const()[name = tensor("op_2667"), val = tensor([1, 1])]; + tensor hidden_states_59_pad_type_0 = const()[name = tensor("hidden_states_59_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_59_pad_0 = const()[name = tensor("hidden_states_59_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor down_blocks_2_resnets_0_conv2_weight_to_fp16 = const()[name = tensor("down_blocks_2_resnets_0_conv2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(163746176)))]; + tensor down_blocks_2_resnets_0_conv2_bias_to_fp16 = const()[name = tensor("down_blocks_2_resnets_0_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(193237440)))]; + tensor hidden_states_59_cast_fp16 = conv(bias = down_blocks_2_resnets_0_conv2_bias_to_fp16, dilations = var_2667, groups = var_2624, pad = hidden_states_59_pad_0, pad_type = hidden_states_59_pad_type_0, strides = var_2665, weight = down_blocks_2_resnets_0_conv2_weight_to_fp16, x = input_127_cast_fp16)[name = tensor("hidden_states_59_cast_fp16")]; + tensor var_2672 = const()[name = tensor("op_2672"), val = tensor([1, 1])]; + tensor var_2674 = const()[name = tensor("op_2674"), val = tensor([1, 1])]; + tensor x_3_pad_type_0 = const()[name = tensor("x_3_pad_type_0"), val = tensor("custom")]; + tensor x_3_pad_0 = const()[name = tensor("x_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_resnets_0_conv_shortcut_weight_to_fp16 = const()[name = tensor("down_blocks_2_resnets_0_conv_shortcut_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(193240064)))]; + tensor down_blocks_2_resnets_0_conv_shortcut_bias_to_fp16 = const()[name = tensor("down_blocks_2_resnets_0_conv_shortcut_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(194878528)))]; + tensor x_3_cast_fp16 = conv(bias = down_blocks_2_resnets_0_conv_shortcut_bias_to_fp16, dilations = var_2674, groups = var_2624, pad = x_3_pad_0, pad_type = x_3_pad_type_0, strides = var_2672, weight = down_blocks_2_resnets_0_conv_shortcut_weight_to_fp16, x = input_115_cast_fp16)[name = tensor("x_3_cast_fp16")]; + tensor hidden_states_61_cast_fp16 = add(x = x_3_cast_fp16, y = hidden_states_59_cast_fp16)[name = tensor("hidden_states_61_cast_fp16")]; + tensor reshape_48_shape_0 = const()[name = tensor("reshape_48_shape_0"), val = tensor([2, 32, 40, 32, 32])]; + tensor reshape_48_cast_fp16 = reshape(shape = reshape_48_shape_0, x = hidden_states_61_cast_fp16)[name = tensor("reshape_48_cast_fp16")]; + tensor reduce_mean_36_axes_0 = const()[name = tensor("reduce_mean_36_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_36_keep_dims_0 = const()[name = tensor("reduce_mean_36_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_36_cast_fp16 = reduce_mean(axes = reduce_mean_36_axes_0, keep_dims = reduce_mean_36_keep_dims_0, x = reshape_48_cast_fp16)[name = tensor("reduce_mean_36_cast_fp16")]; + tensor sub_24_cast_fp16 = sub(x = reshape_48_cast_fp16, y = reduce_mean_36_cast_fp16)[name = tensor("sub_24_cast_fp16")]; + tensor square_12_cast_fp16 = square(x = sub_24_cast_fp16)[name = tensor("square_12_cast_fp16")]; + tensor reduce_mean_38_axes_0 = const()[name = tensor("reduce_mean_38_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_38_keep_dims_0 = const()[name = tensor("reduce_mean_38_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_38_cast_fp16 = reduce_mean(axes = reduce_mean_38_axes_0, keep_dims = reduce_mean_38_keep_dims_0, x = square_12_cast_fp16)[name = tensor("reduce_mean_38_cast_fp16")]; + tensor add_24_y_0_to_fp16 = const()[name = tensor("add_24_y_0_to_fp16"), val = tensor(0x1.1p-20)]; + tensor add_24_cast_fp16 = add(x = reduce_mean_38_cast_fp16, y = add_24_y_0_to_fp16)[name = tensor("add_24_cast_fp16")]; + tensor sqrt_12_cast_fp16 = sqrt(x = add_24_cast_fp16)[name = tensor("sqrt_12_cast_fp16")]; + tensor real_div_12_cast_fp16 = real_div(x = sub_24_cast_fp16, y = sqrt_12_cast_fp16)[name = tensor("real_div_12_cast_fp16")]; + tensor reshape_49_shape_0 = const()[name = tensor("reshape_49_shape_0"), val = tensor([2, 1280, 32, 32])]; + tensor reshape_49_cast_fp16 = reshape(shape = reshape_49_shape_0, x = real_div_12_cast_fp16)[name = tensor("reshape_49_cast_fp16")]; + tensor add_25_gamma_0_to_fp16 = const()[name = tensor("add_25_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(194881152)))]; + tensor add_25_beta_0_to_fp16 = const()[name = tensor("add_25_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(194883776)))]; + tensor add_25_epsilon_0_to_fp16 = const()[name = tensor("add_25_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_25_cast_fp16 = batch_norm(beta = add_25_beta_0_to_fp16, epsilon = add_25_epsilon_0_to_fp16, gamma = add_25_gamma_0_to_fp16, mean = add_23_mean_0_to_fp16, variance = add_23_variance_0_to_fp16, x = reshape_49_cast_fp16)[name = tensor("add_25_cast_fp16")]; + tensor var_2712 = const()[name = tensor("op_2712"), val = tensor([1, 1])]; + tensor var_2714 = const()[name = tensor("op_2714"), val = tensor([1, 1])]; + tensor hidden_states_63_pad_type_0 = const()[name = tensor("hidden_states_63_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_63_pad_0 = const()[name = tensor("hidden_states_63_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_proj_in_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_proj_in_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(194886400)))]; + tensor down_blocks_2_attentions_0_proj_in_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_proj_in_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(198163264)))]; + tensor hidden_states_63_cast_fp16 = conv(bias = down_blocks_2_attentions_0_proj_in_bias_to_fp16, dilations = var_2714, groups = var_2624, pad = hidden_states_63_pad_0, pad_type = hidden_states_63_pad_type_0, strides = var_2712, weight = down_blocks_2_attentions_0_proj_in_weight_to_fp16, x = add_25_cast_fp16)[name = tensor("hidden_states_63_cast_fp16")]; + tensor var_2719 = const()[name = tensor("op_2719"), val = tensor([2, 1280, 1, 1024])]; + tensor inputs_25_cast_fp16 = reshape(shape = var_2719, x = hidden_states_63_cast_fp16)[name = tensor("inputs_25_cast_fp16")]; + tensor hidden_states_65_axes_0 = const()[name = tensor("hidden_states_65_axes_0"), val = tensor([1])]; + tensor hidden_states_65_gamma_0_to_fp16 = const()[name = tensor("hidden_states_65_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(198165888)))]; + tensor hidden_states_65_beta_0_to_fp16 = const()[name = tensor("hidden_states_65_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(198168512)))]; + tensor var_2735_to_fp16 = const()[name = tensor("op_2735_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_65_cast_fp16 = layer_norm(axes = hidden_states_65_axes_0, beta = hidden_states_65_beta_0_to_fp16, epsilon = var_2735_to_fp16, gamma = hidden_states_65_gamma_0_to_fp16, x = inputs_25_cast_fp16)[name = tensor("hidden_states_65_cast_fp16")]; + tensor var_2750 = const()[name = tensor("op_2750"), val = tensor([1, 1])]; + tensor var_2752 = const()[name = tensor("op_2752"), val = tensor([1, 1])]; + tensor q_17_pad_type_0 = const()[name = tensor("q_17_pad_type_0"), val = tensor("custom")]; + tensor q_17_pad_0 = const()[name = tensor("q_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(198171136)))]; + tensor q_17_cast_fp16 = conv(dilations = var_2752, groups = var_2624, pad = q_17_pad_0, pad_type = q_17_pad_type_0, strides = var_2750, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16, x = hidden_states_65_cast_fp16)[name = tensor("q_17_cast_fp16")]; + tensor var_2756 = const()[name = tensor("op_2756"), val = tensor([1, 1])]; + tensor var_2758 = const()[name = tensor("op_2758"), val = tensor([1, 1])]; + tensor k_33_pad_type_0 = const()[name = tensor("k_33_pad_type_0"), val = tensor("custom")]; + tensor k_33_pad_0 = const()[name = tensor("k_33_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(201448000)))]; + tensor k_33_cast_fp16 = conv(dilations = var_2758, groups = var_2624, pad = k_33_pad_0, pad_type = k_33_pad_type_0, strides = var_2756, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16, x = hidden_states_65_cast_fp16)[name = tensor("k_33_cast_fp16")]; + tensor var_2762 = const()[name = tensor("op_2762"), val = tensor([1, 1])]; + tensor var_2764 = const()[name = tensor("op_2764"), val = tensor([1, 1])]; + tensor v_17_pad_type_0 = const()[name = tensor("v_17_pad_type_0"), val = tensor("custom")]; + tensor v_17_pad_0 = const()[name = tensor("v_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(204724864)))]; + tensor v_17_cast_fp16 = conv(dilations = var_2764, groups = var_2624, pad = v_17_pad_0, pad_type = v_17_pad_type_0, strides = var_2762, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16, x = hidden_states_65_cast_fp16)[name = tensor("v_17_cast_fp16")]; + tensor var_2768_begin_0 = const()[name = tensor("op_2768_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2768_end_0 = const()[name = tensor("op_2768_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_2768_end_mask_0 = const()[name = tensor("op_2768_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2768_cast_fp16 = slice_by_index(begin = var_2768_begin_0, end = var_2768_end_0, end_mask = var_2768_end_mask_0, x = q_17_cast_fp16)[name = tensor("op_2768_cast_fp16")]; + tensor var_2772_begin_0 = const()[name = tensor("op_2772_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_2772_end_0 = const()[name = tensor("op_2772_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_2772_end_mask_0 = const()[name = tensor("op_2772_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2772_cast_fp16 = slice_by_index(begin = var_2772_begin_0, end = var_2772_end_0, end_mask = var_2772_end_mask_0, x = q_17_cast_fp16)[name = tensor("op_2772_cast_fp16")]; + tensor var_2776_begin_0 = const()[name = tensor("op_2776_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_2776_end_0 = const()[name = tensor("op_2776_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_2776_end_mask_0 = const()[name = tensor("op_2776_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2776_cast_fp16 = slice_by_index(begin = var_2776_begin_0, end = var_2776_end_0, end_mask = var_2776_end_mask_0, x = q_17_cast_fp16)[name = tensor("op_2776_cast_fp16")]; + tensor var_2780_begin_0 = const()[name = tensor("op_2780_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_2780_end_0 = const()[name = tensor("op_2780_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_2780_end_mask_0 = const()[name = tensor("op_2780_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2780_cast_fp16 = slice_by_index(begin = var_2780_begin_0, end = var_2780_end_0, end_mask = var_2780_end_mask_0, x = q_17_cast_fp16)[name = tensor("op_2780_cast_fp16")]; + tensor var_2784_begin_0 = const()[name = tensor("op_2784_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_2784_end_0 = const()[name = tensor("op_2784_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_2784_end_mask_0 = const()[name = tensor("op_2784_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2784_cast_fp16 = slice_by_index(begin = var_2784_begin_0, end = var_2784_end_0, end_mask = var_2784_end_mask_0, x = q_17_cast_fp16)[name = tensor("op_2784_cast_fp16")]; + tensor var_2788_begin_0 = const()[name = tensor("op_2788_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_2788_end_0 = const()[name = tensor("op_2788_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_2788_end_mask_0 = const()[name = tensor("op_2788_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2788_cast_fp16 = slice_by_index(begin = var_2788_begin_0, end = var_2788_end_0, end_mask = var_2788_end_mask_0, x = q_17_cast_fp16)[name = tensor("op_2788_cast_fp16")]; + tensor var_2792_begin_0 = const()[name = tensor("op_2792_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_2792_end_0 = const()[name = tensor("op_2792_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_2792_end_mask_0 = const()[name = tensor("op_2792_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2792_cast_fp16 = slice_by_index(begin = var_2792_begin_0, end = var_2792_end_0, end_mask = var_2792_end_mask_0, x = q_17_cast_fp16)[name = tensor("op_2792_cast_fp16")]; + tensor var_2796_begin_0 = const()[name = tensor("op_2796_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_2796_end_0 = const()[name = tensor("op_2796_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_2796_end_mask_0 = const()[name = tensor("op_2796_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2796_cast_fp16 = slice_by_index(begin = var_2796_begin_0, end = var_2796_end_0, end_mask = var_2796_end_mask_0, x = q_17_cast_fp16)[name = tensor("op_2796_cast_fp16")]; + tensor var_2800_begin_0 = const()[name = tensor("op_2800_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_2800_end_0 = const()[name = tensor("op_2800_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_2800_end_mask_0 = const()[name = tensor("op_2800_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2800_cast_fp16 = slice_by_index(begin = var_2800_begin_0, end = var_2800_end_0, end_mask = var_2800_end_mask_0, x = q_17_cast_fp16)[name = tensor("op_2800_cast_fp16")]; + tensor var_2804_begin_0 = const()[name = tensor("op_2804_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_2804_end_0 = const()[name = tensor("op_2804_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_2804_end_mask_0 = const()[name = tensor("op_2804_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2804_cast_fp16 = slice_by_index(begin = var_2804_begin_0, end = var_2804_end_0, end_mask = var_2804_end_mask_0, x = q_17_cast_fp16)[name = tensor("op_2804_cast_fp16")]; + tensor var_2808_begin_0 = const()[name = tensor("op_2808_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_2808_end_0 = const()[name = tensor("op_2808_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_2808_end_mask_0 = const()[name = tensor("op_2808_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2808_cast_fp16 = slice_by_index(begin = var_2808_begin_0, end = var_2808_end_0, end_mask = var_2808_end_mask_0, x = q_17_cast_fp16)[name = tensor("op_2808_cast_fp16")]; + tensor var_2812_begin_0 = const()[name = tensor("op_2812_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_2812_end_0 = const()[name = tensor("op_2812_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_2812_end_mask_0 = const()[name = tensor("op_2812_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2812_cast_fp16 = slice_by_index(begin = var_2812_begin_0, end = var_2812_end_0, end_mask = var_2812_end_mask_0, x = q_17_cast_fp16)[name = tensor("op_2812_cast_fp16")]; + tensor var_2816_begin_0 = const()[name = tensor("op_2816_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_2816_end_0 = const()[name = tensor("op_2816_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_2816_end_mask_0 = const()[name = tensor("op_2816_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2816_cast_fp16 = slice_by_index(begin = var_2816_begin_0, end = var_2816_end_0, end_mask = var_2816_end_mask_0, x = q_17_cast_fp16)[name = tensor("op_2816_cast_fp16")]; + tensor var_2820_begin_0 = const()[name = tensor("op_2820_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_2820_end_0 = const()[name = tensor("op_2820_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_2820_end_mask_0 = const()[name = tensor("op_2820_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2820_cast_fp16 = slice_by_index(begin = var_2820_begin_0, end = var_2820_end_0, end_mask = var_2820_end_mask_0, x = q_17_cast_fp16)[name = tensor("op_2820_cast_fp16")]; + tensor var_2824_begin_0 = const()[name = tensor("op_2824_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_2824_end_0 = const()[name = tensor("op_2824_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_2824_end_mask_0 = const()[name = tensor("op_2824_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2824_cast_fp16 = slice_by_index(begin = var_2824_begin_0, end = var_2824_end_0, end_mask = var_2824_end_mask_0, x = q_17_cast_fp16)[name = tensor("op_2824_cast_fp16")]; + tensor var_2828_begin_0 = const()[name = tensor("op_2828_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_2828_end_0 = const()[name = tensor("op_2828_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_2828_end_mask_0 = const()[name = tensor("op_2828_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2828_cast_fp16 = slice_by_index(begin = var_2828_begin_0, end = var_2828_end_0, end_mask = var_2828_end_mask_0, x = q_17_cast_fp16)[name = tensor("op_2828_cast_fp16")]; + tensor var_2832_begin_0 = const()[name = tensor("op_2832_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_2832_end_0 = const()[name = tensor("op_2832_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_2832_end_mask_0 = const()[name = tensor("op_2832_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2832_cast_fp16 = slice_by_index(begin = var_2832_begin_0, end = var_2832_end_0, end_mask = var_2832_end_mask_0, x = q_17_cast_fp16)[name = tensor("op_2832_cast_fp16")]; + tensor var_2836_begin_0 = const()[name = tensor("op_2836_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_2836_end_0 = const()[name = tensor("op_2836_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_2836_end_mask_0 = const()[name = tensor("op_2836_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2836_cast_fp16 = slice_by_index(begin = var_2836_begin_0, end = var_2836_end_0, end_mask = var_2836_end_mask_0, x = q_17_cast_fp16)[name = tensor("op_2836_cast_fp16")]; + tensor var_2840_begin_0 = const()[name = tensor("op_2840_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_2840_end_0 = const()[name = tensor("op_2840_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_2840_end_mask_0 = const()[name = tensor("op_2840_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2840_cast_fp16 = slice_by_index(begin = var_2840_begin_0, end = var_2840_end_0, end_mask = var_2840_end_mask_0, x = q_17_cast_fp16)[name = tensor("op_2840_cast_fp16")]; + tensor var_2844_begin_0 = const()[name = tensor("op_2844_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_2844_end_0 = const()[name = tensor("op_2844_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_2844_end_mask_0 = const()[name = tensor("op_2844_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2844_cast_fp16 = slice_by_index(begin = var_2844_begin_0, end = var_2844_end_0, end_mask = var_2844_end_mask_0, x = q_17_cast_fp16)[name = tensor("op_2844_cast_fp16")]; + tensor k_35_perm_0 = const()[name = tensor("k_35_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_2851_begin_0 = const()[name = tensor("op_2851_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2851_end_0 = const()[name = tensor("op_2851_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_2851_end_mask_0 = const()[name = tensor("op_2851_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_35_cast_fp16 = transpose(perm = k_35_perm_0, x = k_33_cast_fp16)[name = tensor("transpose_131")]; + tensor var_2851_cast_fp16 = slice_by_index(begin = var_2851_begin_0, end = var_2851_end_0, end_mask = var_2851_end_mask_0, x = k_35_cast_fp16)[name = tensor("op_2851_cast_fp16")]; + tensor var_2855_begin_0 = const()[name = tensor("op_2855_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_2855_end_0 = const()[name = tensor("op_2855_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_2855_end_mask_0 = const()[name = tensor("op_2855_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2855_cast_fp16 = slice_by_index(begin = var_2855_begin_0, end = var_2855_end_0, end_mask = var_2855_end_mask_0, x = k_35_cast_fp16)[name = tensor("op_2855_cast_fp16")]; + tensor var_2859_begin_0 = const()[name = tensor("op_2859_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_2859_end_0 = const()[name = tensor("op_2859_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_2859_end_mask_0 = const()[name = tensor("op_2859_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2859_cast_fp16 = slice_by_index(begin = var_2859_begin_0, end = var_2859_end_0, end_mask = var_2859_end_mask_0, x = k_35_cast_fp16)[name = tensor("op_2859_cast_fp16")]; + tensor var_2863_begin_0 = const()[name = tensor("op_2863_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_2863_end_0 = const()[name = tensor("op_2863_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_2863_end_mask_0 = const()[name = tensor("op_2863_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2863_cast_fp16 = slice_by_index(begin = var_2863_begin_0, end = var_2863_end_0, end_mask = var_2863_end_mask_0, x = k_35_cast_fp16)[name = tensor("op_2863_cast_fp16")]; + tensor var_2867_begin_0 = const()[name = tensor("op_2867_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_2867_end_0 = const()[name = tensor("op_2867_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_2867_end_mask_0 = const()[name = tensor("op_2867_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2867_cast_fp16 = slice_by_index(begin = var_2867_begin_0, end = var_2867_end_0, end_mask = var_2867_end_mask_0, x = k_35_cast_fp16)[name = tensor("op_2867_cast_fp16")]; + tensor var_2871_begin_0 = const()[name = tensor("op_2871_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_2871_end_0 = const()[name = tensor("op_2871_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_2871_end_mask_0 = const()[name = tensor("op_2871_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2871_cast_fp16 = slice_by_index(begin = var_2871_begin_0, end = var_2871_end_0, end_mask = var_2871_end_mask_0, x = k_35_cast_fp16)[name = tensor("op_2871_cast_fp16")]; + tensor var_2875_begin_0 = const()[name = tensor("op_2875_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_2875_end_0 = const()[name = tensor("op_2875_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_2875_end_mask_0 = const()[name = tensor("op_2875_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2875_cast_fp16 = slice_by_index(begin = var_2875_begin_0, end = var_2875_end_0, end_mask = var_2875_end_mask_0, x = k_35_cast_fp16)[name = tensor("op_2875_cast_fp16")]; + tensor var_2879_begin_0 = const()[name = tensor("op_2879_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_2879_end_0 = const()[name = tensor("op_2879_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_2879_end_mask_0 = const()[name = tensor("op_2879_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2879_cast_fp16 = slice_by_index(begin = var_2879_begin_0, end = var_2879_end_0, end_mask = var_2879_end_mask_0, x = k_35_cast_fp16)[name = tensor("op_2879_cast_fp16")]; + tensor var_2883_begin_0 = const()[name = tensor("op_2883_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_2883_end_0 = const()[name = tensor("op_2883_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_2883_end_mask_0 = const()[name = tensor("op_2883_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2883_cast_fp16 = slice_by_index(begin = var_2883_begin_0, end = var_2883_end_0, end_mask = var_2883_end_mask_0, x = k_35_cast_fp16)[name = tensor("op_2883_cast_fp16")]; + tensor var_2887_begin_0 = const()[name = tensor("op_2887_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_2887_end_0 = const()[name = tensor("op_2887_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_2887_end_mask_0 = const()[name = tensor("op_2887_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2887_cast_fp16 = slice_by_index(begin = var_2887_begin_0, end = var_2887_end_0, end_mask = var_2887_end_mask_0, x = k_35_cast_fp16)[name = tensor("op_2887_cast_fp16")]; + tensor var_2891_begin_0 = const()[name = tensor("op_2891_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_2891_end_0 = const()[name = tensor("op_2891_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_2891_end_mask_0 = const()[name = tensor("op_2891_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2891_cast_fp16 = slice_by_index(begin = var_2891_begin_0, end = var_2891_end_0, end_mask = var_2891_end_mask_0, x = k_35_cast_fp16)[name = tensor("op_2891_cast_fp16")]; + tensor var_2895_begin_0 = const()[name = tensor("op_2895_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_2895_end_0 = const()[name = tensor("op_2895_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_2895_end_mask_0 = const()[name = tensor("op_2895_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2895_cast_fp16 = slice_by_index(begin = var_2895_begin_0, end = var_2895_end_0, end_mask = var_2895_end_mask_0, x = k_35_cast_fp16)[name = tensor("op_2895_cast_fp16")]; + tensor var_2899_begin_0 = const()[name = tensor("op_2899_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_2899_end_0 = const()[name = tensor("op_2899_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_2899_end_mask_0 = const()[name = tensor("op_2899_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2899_cast_fp16 = slice_by_index(begin = var_2899_begin_0, end = var_2899_end_0, end_mask = var_2899_end_mask_0, x = k_35_cast_fp16)[name = tensor("op_2899_cast_fp16")]; + tensor var_2903_begin_0 = const()[name = tensor("op_2903_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_2903_end_0 = const()[name = tensor("op_2903_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_2903_end_mask_0 = const()[name = tensor("op_2903_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2903_cast_fp16 = slice_by_index(begin = var_2903_begin_0, end = var_2903_end_0, end_mask = var_2903_end_mask_0, x = k_35_cast_fp16)[name = tensor("op_2903_cast_fp16")]; + tensor var_2907_begin_0 = const()[name = tensor("op_2907_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_2907_end_0 = const()[name = tensor("op_2907_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_2907_end_mask_0 = const()[name = tensor("op_2907_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2907_cast_fp16 = slice_by_index(begin = var_2907_begin_0, end = var_2907_end_0, end_mask = var_2907_end_mask_0, x = k_35_cast_fp16)[name = tensor("op_2907_cast_fp16")]; + tensor var_2911_begin_0 = const()[name = tensor("op_2911_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_2911_end_0 = const()[name = tensor("op_2911_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_2911_end_mask_0 = const()[name = tensor("op_2911_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2911_cast_fp16 = slice_by_index(begin = var_2911_begin_0, end = var_2911_end_0, end_mask = var_2911_end_mask_0, x = k_35_cast_fp16)[name = tensor("op_2911_cast_fp16")]; + tensor var_2915_begin_0 = const()[name = tensor("op_2915_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_2915_end_0 = const()[name = tensor("op_2915_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_2915_end_mask_0 = const()[name = tensor("op_2915_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2915_cast_fp16 = slice_by_index(begin = var_2915_begin_0, end = var_2915_end_0, end_mask = var_2915_end_mask_0, x = k_35_cast_fp16)[name = tensor("op_2915_cast_fp16")]; + tensor var_2919_begin_0 = const()[name = tensor("op_2919_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_2919_end_0 = const()[name = tensor("op_2919_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_2919_end_mask_0 = const()[name = tensor("op_2919_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2919_cast_fp16 = slice_by_index(begin = var_2919_begin_0, end = var_2919_end_0, end_mask = var_2919_end_mask_0, x = k_35_cast_fp16)[name = tensor("op_2919_cast_fp16")]; + tensor var_2923_begin_0 = const()[name = tensor("op_2923_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_2923_end_0 = const()[name = tensor("op_2923_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_2923_end_mask_0 = const()[name = tensor("op_2923_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2923_cast_fp16 = slice_by_index(begin = var_2923_begin_0, end = var_2923_end_0, end_mask = var_2923_end_mask_0, x = k_35_cast_fp16)[name = tensor("op_2923_cast_fp16")]; + tensor var_2927_begin_0 = const()[name = tensor("op_2927_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_2927_end_0 = const()[name = tensor("op_2927_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_2927_end_mask_0 = const()[name = tensor("op_2927_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2927_cast_fp16 = slice_by_index(begin = var_2927_begin_0, end = var_2927_end_0, end_mask = var_2927_end_mask_0, x = k_35_cast_fp16)[name = tensor("op_2927_cast_fp16")]; + tensor var_2929_begin_0 = const()[name = tensor("op_2929_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2929_end_0 = const()[name = tensor("op_2929_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_2929_end_mask_0 = const()[name = tensor("op_2929_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2929_cast_fp16 = slice_by_index(begin = var_2929_begin_0, end = var_2929_end_0, end_mask = var_2929_end_mask_0, x = v_17_cast_fp16)[name = tensor("op_2929_cast_fp16")]; + tensor var_2933_begin_0 = const()[name = tensor("op_2933_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_2933_end_0 = const()[name = tensor("op_2933_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_2933_end_mask_0 = const()[name = tensor("op_2933_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2933_cast_fp16 = slice_by_index(begin = var_2933_begin_0, end = var_2933_end_0, end_mask = var_2933_end_mask_0, x = v_17_cast_fp16)[name = tensor("op_2933_cast_fp16")]; + tensor var_2937_begin_0 = const()[name = tensor("op_2937_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_2937_end_0 = const()[name = tensor("op_2937_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_2937_end_mask_0 = const()[name = tensor("op_2937_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2937_cast_fp16 = slice_by_index(begin = var_2937_begin_0, end = var_2937_end_0, end_mask = var_2937_end_mask_0, x = v_17_cast_fp16)[name = tensor("op_2937_cast_fp16")]; + tensor var_2941_begin_0 = const()[name = tensor("op_2941_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_2941_end_0 = const()[name = tensor("op_2941_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_2941_end_mask_0 = const()[name = tensor("op_2941_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2941_cast_fp16 = slice_by_index(begin = var_2941_begin_0, end = var_2941_end_0, end_mask = var_2941_end_mask_0, x = v_17_cast_fp16)[name = tensor("op_2941_cast_fp16")]; + tensor var_2945_begin_0 = const()[name = tensor("op_2945_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_2945_end_0 = const()[name = tensor("op_2945_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_2945_end_mask_0 = const()[name = tensor("op_2945_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2945_cast_fp16 = slice_by_index(begin = var_2945_begin_0, end = var_2945_end_0, end_mask = var_2945_end_mask_0, x = v_17_cast_fp16)[name = tensor("op_2945_cast_fp16")]; + tensor var_2949_begin_0 = const()[name = tensor("op_2949_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_2949_end_0 = const()[name = tensor("op_2949_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_2949_end_mask_0 = const()[name = tensor("op_2949_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2949_cast_fp16 = slice_by_index(begin = var_2949_begin_0, end = var_2949_end_0, end_mask = var_2949_end_mask_0, x = v_17_cast_fp16)[name = tensor("op_2949_cast_fp16")]; + tensor var_2953_begin_0 = const()[name = tensor("op_2953_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_2953_end_0 = const()[name = tensor("op_2953_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_2953_end_mask_0 = const()[name = tensor("op_2953_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2953_cast_fp16 = slice_by_index(begin = var_2953_begin_0, end = var_2953_end_0, end_mask = var_2953_end_mask_0, x = v_17_cast_fp16)[name = tensor("op_2953_cast_fp16")]; + tensor var_2957_begin_0 = const()[name = tensor("op_2957_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_2957_end_0 = const()[name = tensor("op_2957_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_2957_end_mask_0 = const()[name = tensor("op_2957_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2957_cast_fp16 = slice_by_index(begin = var_2957_begin_0, end = var_2957_end_0, end_mask = var_2957_end_mask_0, x = v_17_cast_fp16)[name = tensor("op_2957_cast_fp16")]; + tensor var_2961_begin_0 = const()[name = tensor("op_2961_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_2961_end_0 = const()[name = tensor("op_2961_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_2961_end_mask_0 = const()[name = tensor("op_2961_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2961_cast_fp16 = slice_by_index(begin = var_2961_begin_0, end = var_2961_end_0, end_mask = var_2961_end_mask_0, x = v_17_cast_fp16)[name = tensor("op_2961_cast_fp16")]; + tensor var_2965_begin_0 = const()[name = tensor("op_2965_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_2965_end_0 = const()[name = tensor("op_2965_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_2965_end_mask_0 = const()[name = tensor("op_2965_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2965_cast_fp16 = slice_by_index(begin = var_2965_begin_0, end = var_2965_end_0, end_mask = var_2965_end_mask_0, x = v_17_cast_fp16)[name = tensor("op_2965_cast_fp16")]; + tensor var_2969_begin_0 = const()[name = tensor("op_2969_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_2969_end_0 = const()[name = tensor("op_2969_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_2969_end_mask_0 = const()[name = tensor("op_2969_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2969_cast_fp16 = slice_by_index(begin = var_2969_begin_0, end = var_2969_end_0, end_mask = var_2969_end_mask_0, x = v_17_cast_fp16)[name = tensor("op_2969_cast_fp16")]; + tensor var_2973_begin_0 = const()[name = tensor("op_2973_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_2973_end_0 = const()[name = tensor("op_2973_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_2973_end_mask_0 = const()[name = tensor("op_2973_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2973_cast_fp16 = slice_by_index(begin = var_2973_begin_0, end = var_2973_end_0, end_mask = var_2973_end_mask_0, x = v_17_cast_fp16)[name = tensor("op_2973_cast_fp16")]; + tensor var_2977_begin_0 = const()[name = tensor("op_2977_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_2977_end_0 = const()[name = tensor("op_2977_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_2977_end_mask_0 = const()[name = tensor("op_2977_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2977_cast_fp16 = slice_by_index(begin = var_2977_begin_0, end = var_2977_end_0, end_mask = var_2977_end_mask_0, x = v_17_cast_fp16)[name = tensor("op_2977_cast_fp16")]; + tensor var_2981_begin_0 = const()[name = tensor("op_2981_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_2981_end_0 = const()[name = tensor("op_2981_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_2981_end_mask_0 = const()[name = tensor("op_2981_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2981_cast_fp16 = slice_by_index(begin = var_2981_begin_0, end = var_2981_end_0, end_mask = var_2981_end_mask_0, x = v_17_cast_fp16)[name = tensor("op_2981_cast_fp16")]; + tensor var_2985_begin_0 = const()[name = tensor("op_2985_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_2985_end_0 = const()[name = tensor("op_2985_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_2985_end_mask_0 = const()[name = tensor("op_2985_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2985_cast_fp16 = slice_by_index(begin = var_2985_begin_0, end = var_2985_end_0, end_mask = var_2985_end_mask_0, x = v_17_cast_fp16)[name = tensor("op_2985_cast_fp16")]; + tensor var_2989_begin_0 = const()[name = tensor("op_2989_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_2989_end_0 = const()[name = tensor("op_2989_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_2989_end_mask_0 = const()[name = tensor("op_2989_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2989_cast_fp16 = slice_by_index(begin = var_2989_begin_0, end = var_2989_end_0, end_mask = var_2989_end_mask_0, x = v_17_cast_fp16)[name = tensor("op_2989_cast_fp16")]; + tensor var_2993_begin_0 = const()[name = tensor("op_2993_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_2993_end_0 = const()[name = tensor("op_2993_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_2993_end_mask_0 = const()[name = tensor("op_2993_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2993_cast_fp16 = slice_by_index(begin = var_2993_begin_0, end = var_2993_end_0, end_mask = var_2993_end_mask_0, x = v_17_cast_fp16)[name = tensor("op_2993_cast_fp16")]; + tensor var_2997_begin_0 = const()[name = tensor("op_2997_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_2997_end_0 = const()[name = tensor("op_2997_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_2997_end_mask_0 = const()[name = tensor("op_2997_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2997_cast_fp16 = slice_by_index(begin = var_2997_begin_0, end = var_2997_end_0, end_mask = var_2997_end_mask_0, x = v_17_cast_fp16)[name = tensor("op_2997_cast_fp16")]; + tensor var_3001_begin_0 = const()[name = tensor("op_3001_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_3001_end_0 = const()[name = tensor("op_3001_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_3001_end_mask_0 = const()[name = tensor("op_3001_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3001_cast_fp16 = slice_by_index(begin = var_3001_begin_0, end = var_3001_end_0, end_mask = var_3001_end_mask_0, x = v_17_cast_fp16)[name = tensor("op_3001_cast_fp16")]; + tensor var_3005_begin_0 = const()[name = tensor("op_3005_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_3005_end_0 = const()[name = tensor("op_3005_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_3005_end_mask_0 = const()[name = tensor("op_3005_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3005_cast_fp16 = slice_by_index(begin = var_3005_begin_0, end = var_3005_end_0, end_mask = var_3005_end_mask_0, x = v_17_cast_fp16)[name = tensor("op_3005_cast_fp16")]; + tensor var_3009_equation_0 = const()[name = tensor("op_3009_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3009_cast_fp16 = einsum(equation = var_3009_equation_0, values = (var_2851_cast_fp16, var_2768_cast_fp16))[name = tensor("op_3009_cast_fp16")]; + tensor var_3010_to_fp16 = const()[name = tensor("op_3010_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_161_cast_fp16 = mul(x = var_3009_cast_fp16, y = var_3010_to_fp16)[name = tensor("aw_161_cast_fp16")]; + tensor var_3013_equation_0 = const()[name = tensor("op_3013_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3013_cast_fp16 = einsum(equation = var_3013_equation_0, values = (var_2855_cast_fp16, var_2772_cast_fp16))[name = tensor("op_3013_cast_fp16")]; + tensor var_3014_to_fp16 = const()[name = tensor("op_3014_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_163_cast_fp16 = mul(x = var_3013_cast_fp16, y = var_3014_to_fp16)[name = tensor("aw_163_cast_fp16")]; + tensor var_3017_equation_0 = const()[name = tensor("op_3017_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3017_cast_fp16 = einsum(equation = var_3017_equation_0, values = (var_2859_cast_fp16, var_2776_cast_fp16))[name = tensor("op_3017_cast_fp16")]; + tensor var_3018_to_fp16 = const()[name = tensor("op_3018_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_165_cast_fp16 = mul(x = var_3017_cast_fp16, y = var_3018_to_fp16)[name = tensor("aw_165_cast_fp16")]; + tensor var_3021_equation_0 = const()[name = tensor("op_3021_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3021_cast_fp16 = einsum(equation = var_3021_equation_0, values = (var_2863_cast_fp16, var_2780_cast_fp16))[name = tensor("op_3021_cast_fp16")]; + tensor var_3022_to_fp16 = const()[name = tensor("op_3022_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_167_cast_fp16 = mul(x = var_3021_cast_fp16, y = var_3022_to_fp16)[name = tensor("aw_167_cast_fp16")]; + tensor var_3025_equation_0 = const()[name = tensor("op_3025_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3025_cast_fp16 = einsum(equation = var_3025_equation_0, values = (var_2867_cast_fp16, var_2784_cast_fp16))[name = tensor("op_3025_cast_fp16")]; + tensor var_3026_to_fp16 = const()[name = tensor("op_3026_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_169_cast_fp16 = mul(x = var_3025_cast_fp16, y = var_3026_to_fp16)[name = tensor("aw_169_cast_fp16")]; + tensor var_3029_equation_0 = const()[name = tensor("op_3029_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3029_cast_fp16 = einsum(equation = var_3029_equation_0, values = (var_2871_cast_fp16, var_2788_cast_fp16))[name = tensor("op_3029_cast_fp16")]; + tensor var_3030_to_fp16 = const()[name = tensor("op_3030_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_171_cast_fp16 = mul(x = var_3029_cast_fp16, y = var_3030_to_fp16)[name = tensor("aw_171_cast_fp16")]; + tensor var_3033_equation_0 = const()[name = tensor("op_3033_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3033_cast_fp16 = einsum(equation = var_3033_equation_0, values = (var_2875_cast_fp16, var_2792_cast_fp16))[name = tensor("op_3033_cast_fp16")]; + tensor var_3034_to_fp16 = const()[name = tensor("op_3034_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_173_cast_fp16 = mul(x = var_3033_cast_fp16, y = var_3034_to_fp16)[name = tensor("aw_173_cast_fp16")]; + tensor var_3037_equation_0 = const()[name = tensor("op_3037_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3037_cast_fp16 = einsum(equation = var_3037_equation_0, values = (var_2879_cast_fp16, var_2796_cast_fp16))[name = tensor("op_3037_cast_fp16")]; + tensor var_3038_to_fp16 = const()[name = tensor("op_3038_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_175_cast_fp16 = mul(x = var_3037_cast_fp16, y = var_3038_to_fp16)[name = tensor("aw_175_cast_fp16")]; + tensor var_3041_equation_0 = const()[name = tensor("op_3041_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3041_cast_fp16 = einsum(equation = var_3041_equation_0, values = (var_2883_cast_fp16, var_2800_cast_fp16))[name = tensor("op_3041_cast_fp16")]; + tensor var_3042_to_fp16 = const()[name = tensor("op_3042_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_177_cast_fp16 = mul(x = var_3041_cast_fp16, y = var_3042_to_fp16)[name = tensor("aw_177_cast_fp16")]; + tensor var_3045_equation_0 = const()[name = tensor("op_3045_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3045_cast_fp16 = einsum(equation = var_3045_equation_0, values = (var_2887_cast_fp16, var_2804_cast_fp16))[name = tensor("op_3045_cast_fp16")]; + tensor var_3046_to_fp16 = const()[name = tensor("op_3046_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_179_cast_fp16 = mul(x = var_3045_cast_fp16, y = var_3046_to_fp16)[name = tensor("aw_179_cast_fp16")]; + tensor var_3049_equation_0 = const()[name = tensor("op_3049_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3049_cast_fp16 = einsum(equation = var_3049_equation_0, values = (var_2891_cast_fp16, var_2808_cast_fp16))[name = tensor("op_3049_cast_fp16")]; + tensor var_3050_to_fp16 = const()[name = tensor("op_3050_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_181_cast_fp16 = mul(x = var_3049_cast_fp16, y = var_3050_to_fp16)[name = tensor("aw_181_cast_fp16")]; + tensor var_3053_equation_0 = const()[name = tensor("op_3053_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3053_cast_fp16 = einsum(equation = var_3053_equation_0, values = (var_2895_cast_fp16, var_2812_cast_fp16))[name = tensor("op_3053_cast_fp16")]; + tensor var_3054_to_fp16 = const()[name = tensor("op_3054_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_183_cast_fp16 = mul(x = var_3053_cast_fp16, y = var_3054_to_fp16)[name = tensor("aw_183_cast_fp16")]; + tensor var_3057_equation_0 = const()[name = tensor("op_3057_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3057_cast_fp16 = einsum(equation = var_3057_equation_0, values = (var_2899_cast_fp16, var_2816_cast_fp16))[name = tensor("op_3057_cast_fp16")]; + tensor var_3058_to_fp16 = const()[name = tensor("op_3058_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_185_cast_fp16 = mul(x = var_3057_cast_fp16, y = var_3058_to_fp16)[name = tensor("aw_185_cast_fp16")]; + tensor var_3061_equation_0 = const()[name = tensor("op_3061_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3061_cast_fp16 = einsum(equation = var_3061_equation_0, values = (var_2903_cast_fp16, var_2820_cast_fp16))[name = tensor("op_3061_cast_fp16")]; + tensor var_3062_to_fp16 = const()[name = tensor("op_3062_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_187_cast_fp16 = mul(x = var_3061_cast_fp16, y = var_3062_to_fp16)[name = tensor("aw_187_cast_fp16")]; + tensor var_3065_equation_0 = const()[name = tensor("op_3065_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3065_cast_fp16 = einsum(equation = var_3065_equation_0, values = (var_2907_cast_fp16, var_2824_cast_fp16))[name = tensor("op_3065_cast_fp16")]; + tensor var_3066_to_fp16 = const()[name = tensor("op_3066_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_189_cast_fp16 = mul(x = var_3065_cast_fp16, y = var_3066_to_fp16)[name = tensor("aw_189_cast_fp16")]; + tensor var_3069_equation_0 = const()[name = tensor("op_3069_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3069_cast_fp16 = einsum(equation = var_3069_equation_0, values = (var_2911_cast_fp16, var_2828_cast_fp16))[name = tensor("op_3069_cast_fp16")]; + tensor var_3070_to_fp16 = const()[name = tensor("op_3070_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_191_cast_fp16 = mul(x = var_3069_cast_fp16, y = var_3070_to_fp16)[name = tensor("aw_191_cast_fp16")]; + tensor var_3073_equation_0 = const()[name = tensor("op_3073_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3073_cast_fp16 = einsum(equation = var_3073_equation_0, values = (var_2915_cast_fp16, var_2832_cast_fp16))[name = tensor("op_3073_cast_fp16")]; + tensor var_3074_to_fp16 = const()[name = tensor("op_3074_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_193_cast_fp16 = mul(x = var_3073_cast_fp16, y = var_3074_to_fp16)[name = tensor("aw_193_cast_fp16")]; + tensor var_3077_equation_0 = const()[name = tensor("op_3077_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3077_cast_fp16 = einsum(equation = var_3077_equation_0, values = (var_2919_cast_fp16, var_2836_cast_fp16))[name = tensor("op_3077_cast_fp16")]; + tensor var_3078_to_fp16 = const()[name = tensor("op_3078_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_195_cast_fp16 = mul(x = var_3077_cast_fp16, y = var_3078_to_fp16)[name = tensor("aw_195_cast_fp16")]; + tensor var_3081_equation_0 = const()[name = tensor("op_3081_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3081_cast_fp16 = einsum(equation = var_3081_equation_0, values = (var_2923_cast_fp16, var_2840_cast_fp16))[name = tensor("op_3081_cast_fp16")]; + tensor var_3082_to_fp16 = const()[name = tensor("op_3082_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_197_cast_fp16 = mul(x = var_3081_cast_fp16, y = var_3082_to_fp16)[name = tensor("aw_197_cast_fp16")]; + tensor var_3085_equation_0 = const()[name = tensor("op_3085_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3085_cast_fp16 = einsum(equation = var_3085_equation_0, values = (var_2927_cast_fp16, var_2844_cast_fp16))[name = tensor("op_3085_cast_fp16")]; + tensor var_3086_to_fp16 = const()[name = tensor("op_3086_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_199_cast_fp16 = mul(x = var_3085_cast_fp16, y = var_3086_to_fp16)[name = tensor("aw_199_cast_fp16")]; + tensor var_3088_cast_fp16 = softmax(axis = var_2624, x = aw_161_cast_fp16)[name = tensor("op_3088_cast_fp16")]; + tensor var_3089_cast_fp16 = softmax(axis = var_2624, x = aw_163_cast_fp16)[name = tensor("op_3089_cast_fp16")]; + tensor var_3090_cast_fp16 = softmax(axis = var_2624, x = aw_165_cast_fp16)[name = tensor("op_3090_cast_fp16")]; + tensor var_3091_cast_fp16 = softmax(axis = var_2624, x = aw_167_cast_fp16)[name = tensor("op_3091_cast_fp16")]; + tensor var_3092_cast_fp16 = softmax(axis = var_2624, x = aw_169_cast_fp16)[name = tensor("op_3092_cast_fp16")]; + tensor var_3093_cast_fp16 = softmax(axis = var_2624, x = aw_171_cast_fp16)[name = tensor("op_3093_cast_fp16")]; + tensor var_3094_cast_fp16 = softmax(axis = var_2624, x = aw_173_cast_fp16)[name = tensor("op_3094_cast_fp16")]; + tensor var_3095_cast_fp16 = softmax(axis = var_2624, x = aw_175_cast_fp16)[name = tensor("op_3095_cast_fp16")]; + tensor var_3096_cast_fp16 = softmax(axis = var_2624, x = aw_177_cast_fp16)[name = tensor("op_3096_cast_fp16")]; + tensor var_3097_cast_fp16 = softmax(axis = var_2624, x = aw_179_cast_fp16)[name = tensor("op_3097_cast_fp16")]; + tensor var_3098_cast_fp16 = softmax(axis = var_2624, x = aw_181_cast_fp16)[name = tensor("op_3098_cast_fp16")]; + tensor var_3099_cast_fp16 = softmax(axis = var_2624, x = aw_183_cast_fp16)[name = tensor("op_3099_cast_fp16")]; + tensor var_3100_cast_fp16 = softmax(axis = var_2624, x = aw_185_cast_fp16)[name = tensor("op_3100_cast_fp16")]; + tensor var_3101_cast_fp16 = softmax(axis = var_2624, x = aw_187_cast_fp16)[name = tensor("op_3101_cast_fp16")]; + tensor var_3102_cast_fp16 = softmax(axis = var_2624, x = aw_189_cast_fp16)[name = tensor("op_3102_cast_fp16")]; + tensor var_3103_cast_fp16 = softmax(axis = var_2624, x = aw_191_cast_fp16)[name = tensor("op_3103_cast_fp16")]; + tensor var_3104_cast_fp16 = softmax(axis = var_2624, x = aw_193_cast_fp16)[name = tensor("op_3104_cast_fp16")]; + tensor var_3105_cast_fp16 = softmax(axis = var_2624, x = aw_195_cast_fp16)[name = tensor("op_3105_cast_fp16")]; + tensor var_3106_cast_fp16 = softmax(axis = var_2624, x = aw_197_cast_fp16)[name = tensor("op_3106_cast_fp16")]; + tensor var_3107_cast_fp16 = softmax(axis = var_2624, x = aw_199_cast_fp16)[name = tensor("op_3107_cast_fp16")]; + tensor var_3109_equation_0 = const()[name = tensor("op_3109_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3109_cast_fp16 = einsum(equation = var_3109_equation_0, values = (var_2929_cast_fp16, var_3088_cast_fp16))[name = tensor("op_3109_cast_fp16")]; + tensor var_3111_equation_0 = const()[name = tensor("op_3111_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3111_cast_fp16 = einsum(equation = var_3111_equation_0, values = (var_2933_cast_fp16, var_3089_cast_fp16))[name = tensor("op_3111_cast_fp16")]; + tensor var_3113_equation_0 = const()[name = tensor("op_3113_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3113_cast_fp16 = einsum(equation = var_3113_equation_0, values = (var_2937_cast_fp16, var_3090_cast_fp16))[name = tensor("op_3113_cast_fp16")]; + tensor var_3115_equation_0 = const()[name = tensor("op_3115_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3115_cast_fp16 = einsum(equation = var_3115_equation_0, values = (var_2941_cast_fp16, var_3091_cast_fp16))[name = tensor("op_3115_cast_fp16")]; + tensor var_3117_equation_0 = const()[name = tensor("op_3117_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3117_cast_fp16 = einsum(equation = var_3117_equation_0, values = (var_2945_cast_fp16, var_3092_cast_fp16))[name = tensor("op_3117_cast_fp16")]; + tensor var_3119_equation_0 = const()[name = tensor("op_3119_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3119_cast_fp16 = einsum(equation = var_3119_equation_0, values = (var_2949_cast_fp16, var_3093_cast_fp16))[name = tensor("op_3119_cast_fp16")]; + tensor var_3121_equation_0 = const()[name = tensor("op_3121_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3121_cast_fp16 = einsum(equation = var_3121_equation_0, values = (var_2953_cast_fp16, var_3094_cast_fp16))[name = tensor("op_3121_cast_fp16")]; + tensor var_3123_equation_0 = const()[name = tensor("op_3123_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3123_cast_fp16 = einsum(equation = var_3123_equation_0, values = (var_2957_cast_fp16, var_3095_cast_fp16))[name = tensor("op_3123_cast_fp16")]; + tensor var_3125_equation_0 = const()[name = tensor("op_3125_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3125_cast_fp16 = einsum(equation = var_3125_equation_0, values = (var_2961_cast_fp16, var_3096_cast_fp16))[name = tensor("op_3125_cast_fp16")]; + tensor var_3127_equation_0 = const()[name = tensor("op_3127_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3127_cast_fp16 = einsum(equation = var_3127_equation_0, values = (var_2965_cast_fp16, var_3097_cast_fp16))[name = tensor("op_3127_cast_fp16")]; + tensor var_3129_equation_0 = const()[name = tensor("op_3129_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3129_cast_fp16 = einsum(equation = var_3129_equation_0, values = (var_2969_cast_fp16, var_3098_cast_fp16))[name = tensor("op_3129_cast_fp16")]; + tensor var_3131_equation_0 = const()[name = tensor("op_3131_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3131_cast_fp16 = einsum(equation = var_3131_equation_0, values = (var_2973_cast_fp16, var_3099_cast_fp16))[name = tensor("op_3131_cast_fp16")]; + tensor var_3133_equation_0 = const()[name = tensor("op_3133_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3133_cast_fp16 = einsum(equation = var_3133_equation_0, values = (var_2977_cast_fp16, var_3100_cast_fp16))[name = tensor("op_3133_cast_fp16")]; + tensor var_3135_equation_0 = const()[name = tensor("op_3135_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3135_cast_fp16 = einsum(equation = var_3135_equation_0, values = (var_2981_cast_fp16, var_3101_cast_fp16))[name = tensor("op_3135_cast_fp16")]; + tensor var_3137_equation_0 = const()[name = tensor("op_3137_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3137_cast_fp16 = einsum(equation = var_3137_equation_0, values = (var_2985_cast_fp16, var_3102_cast_fp16))[name = tensor("op_3137_cast_fp16")]; + tensor var_3139_equation_0 = const()[name = tensor("op_3139_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3139_cast_fp16 = einsum(equation = var_3139_equation_0, values = (var_2989_cast_fp16, var_3103_cast_fp16))[name = tensor("op_3139_cast_fp16")]; + tensor var_3141_equation_0 = const()[name = tensor("op_3141_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3141_cast_fp16 = einsum(equation = var_3141_equation_0, values = (var_2993_cast_fp16, var_3104_cast_fp16))[name = tensor("op_3141_cast_fp16")]; + tensor var_3143_equation_0 = const()[name = tensor("op_3143_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3143_cast_fp16 = einsum(equation = var_3143_equation_0, values = (var_2997_cast_fp16, var_3105_cast_fp16))[name = tensor("op_3143_cast_fp16")]; + tensor var_3145_equation_0 = const()[name = tensor("op_3145_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3145_cast_fp16 = einsum(equation = var_3145_equation_0, values = (var_3001_cast_fp16, var_3106_cast_fp16))[name = tensor("op_3145_cast_fp16")]; + tensor var_3147_equation_0 = const()[name = tensor("op_3147_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3147_cast_fp16 = einsum(equation = var_3147_equation_0, values = (var_3005_cast_fp16, var_3107_cast_fp16))[name = tensor("op_3147_cast_fp16")]; + tensor input_131_interleave_0 = const()[name = tensor("input_131_interleave_0"), val = tensor(false)]; + tensor input_131_cast_fp16 = concat(axis = var_2624, interleave = input_131_interleave_0, values = (var_3109_cast_fp16, var_3111_cast_fp16, var_3113_cast_fp16, var_3115_cast_fp16, var_3117_cast_fp16, var_3119_cast_fp16, var_3121_cast_fp16, var_3123_cast_fp16, var_3125_cast_fp16, var_3127_cast_fp16, var_3129_cast_fp16, var_3131_cast_fp16, var_3133_cast_fp16, var_3135_cast_fp16, var_3137_cast_fp16, var_3139_cast_fp16, var_3141_cast_fp16, var_3143_cast_fp16, var_3145_cast_fp16, var_3147_cast_fp16))[name = tensor("input_131_cast_fp16")]; + tensor var_3153 = const()[name = tensor("op_3153"), val = tensor([1, 1])]; + tensor var_3155 = const()[name = tensor("op_3155"), val = tensor([1, 1])]; + tensor var_3157_pad_type_0 = const()[name = tensor("op_3157_pad_type_0"), val = tensor("custom")]; + tensor var_3157_pad_0 = const()[name = tensor("op_3157_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(208001728)))]; + tensor down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(211278592)))]; + tensor var_3157_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_3155, groups = var_2624, pad = var_3157_pad_0, pad_type = var_3157_pad_type_0, strides = var_3153, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16, x = input_131_cast_fp16)[name = tensor("op_3157_cast_fp16")]; + tensor inputs_27_cast_fp16 = add(x = var_3157_cast_fp16, y = inputs_25_cast_fp16)[name = tensor("inputs_27_cast_fp16")]; + tensor hidden_states_67_axes_0 = const()[name = tensor("hidden_states_67_axes_0"), val = tensor([1])]; + tensor hidden_states_67_gamma_0_to_fp16 = const()[name = tensor("hidden_states_67_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(211281216)))]; + tensor hidden_states_67_beta_0_to_fp16 = const()[name = tensor("hidden_states_67_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(211283840)))]; + tensor var_3167_to_fp16 = const()[name = tensor("op_3167_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_67_cast_fp16 = layer_norm(axes = hidden_states_67_axes_0, beta = hidden_states_67_beta_0_to_fp16, epsilon = var_3167_to_fp16, gamma = hidden_states_67_gamma_0_to_fp16, x = inputs_27_cast_fp16)[name = tensor("hidden_states_67_cast_fp16")]; + tensor var_3182 = const()[name = tensor("op_3182"), val = tensor([1, 1])]; + tensor var_3184 = const()[name = tensor("op_3184"), val = tensor([1, 1])]; + tensor q_19_pad_type_0 = const()[name = tensor("q_19_pad_type_0"), val = tensor("custom")]; + tensor q_19_pad_0 = const()[name = tensor("q_19_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(211286464)))]; + tensor q_19_cast_fp16 = conv(dilations = var_3184, groups = var_2624, pad = q_19_pad_0, pad_type = q_19_pad_type_0, strides = var_3182, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16, x = hidden_states_67_cast_fp16)[name = tensor("q_19_cast_fp16")]; + tensor var_3188 = const()[name = tensor("op_3188"), val = tensor([1, 1])]; + tensor var_3190 = const()[name = tensor("op_3190"), val = tensor([1, 1])]; + tensor k_37_pad_type_0 = const()[name = tensor("k_37_pad_type_0"), val = tensor("custom")]; + tensor k_37_pad_0 = const()[name = tensor("k_37_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(214563328)))]; + tensor k_37_cast_fp16 = conv(dilations = var_3190, groups = var_2624, pad = k_37_pad_0, pad_type = k_37_pad_type_0, strides = var_3188, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_37_cast_fp16")]; + tensor var_3194 = const()[name = tensor("op_3194"), val = tensor([1, 1])]; + tensor var_3196 = const()[name = tensor("op_3196"), val = tensor([1, 1])]; + tensor v_19_pad_type_0 = const()[name = tensor("v_19_pad_type_0"), val = tensor("custom")]; + tensor v_19_pad_0 = const()[name = tensor("v_19_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(219806272)))]; + tensor v_19_cast_fp16 = conv(dilations = var_3196, groups = var_2624, pad = v_19_pad_0, pad_type = v_19_pad_type_0, strides = var_3194, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_19_cast_fp16")]; + tensor var_3200_begin_0 = const()[name = tensor("op_3200_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3200_end_0 = const()[name = tensor("op_3200_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_3200_end_mask_0 = const()[name = tensor("op_3200_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3200_cast_fp16 = slice_by_index(begin = var_3200_begin_0, end = var_3200_end_0, end_mask = var_3200_end_mask_0, x = q_19_cast_fp16)[name = tensor("op_3200_cast_fp16")]; + tensor var_3204_begin_0 = const()[name = tensor("op_3204_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_3204_end_0 = const()[name = tensor("op_3204_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_3204_end_mask_0 = const()[name = tensor("op_3204_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3204_cast_fp16 = slice_by_index(begin = var_3204_begin_0, end = var_3204_end_0, end_mask = var_3204_end_mask_0, x = q_19_cast_fp16)[name = tensor("op_3204_cast_fp16")]; + tensor var_3208_begin_0 = const()[name = tensor("op_3208_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_3208_end_0 = const()[name = tensor("op_3208_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_3208_end_mask_0 = const()[name = tensor("op_3208_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3208_cast_fp16 = slice_by_index(begin = var_3208_begin_0, end = var_3208_end_0, end_mask = var_3208_end_mask_0, x = q_19_cast_fp16)[name = tensor("op_3208_cast_fp16")]; + tensor var_3212_begin_0 = const()[name = tensor("op_3212_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_3212_end_0 = const()[name = tensor("op_3212_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_3212_end_mask_0 = const()[name = tensor("op_3212_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3212_cast_fp16 = slice_by_index(begin = var_3212_begin_0, end = var_3212_end_0, end_mask = var_3212_end_mask_0, x = q_19_cast_fp16)[name = tensor("op_3212_cast_fp16")]; + tensor var_3216_begin_0 = const()[name = tensor("op_3216_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_3216_end_0 = const()[name = tensor("op_3216_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_3216_end_mask_0 = const()[name = tensor("op_3216_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3216_cast_fp16 = slice_by_index(begin = var_3216_begin_0, end = var_3216_end_0, end_mask = var_3216_end_mask_0, x = q_19_cast_fp16)[name = tensor("op_3216_cast_fp16")]; + tensor var_3220_begin_0 = const()[name = tensor("op_3220_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_3220_end_0 = const()[name = tensor("op_3220_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_3220_end_mask_0 = const()[name = tensor("op_3220_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3220_cast_fp16 = slice_by_index(begin = var_3220_begin_0, end = var_3220_end_0, end_mask = var_3220_end_mask_0, x = q_19_cast_fp16)[name = tensor("op_3220_cast_fp16")]; + tensor var_3224_begin_0 = const()[name = tensor("op_3224_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_3224_end_0 = const()[name = tensor("op_3224_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_3224_end_mask_0 = const()[name = tensor("op_3224_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3224_cast_fp16 = slice_by_index(begin = var_3224_begin_0, end = var_3224_end_0, end_mask = var_3224_end_mask_0, x = q_19_cast_fp16)[name = tensor("op_3224_cast_fp16")]; + tensor var_3228_begin_0 = const()[name = tensor("op_3228_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_3228_end_0 = const()[name = tensor("op_3228_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_3228_end_mask_0 = const()[name = tensor("op_3228_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3228_cast_fp16 = slice_by_index(begin = var_3228_begin_0, end = var_3228_end_0, end_mask = var_3228_end_mask_0, x = q_19_cast_fp16)[name = tensor("op_3228_cast_fp16")]; + tensor var_3232_begin_0 = const()[name = tensor("op_3232_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_3232_end_0 = const()[name = tensor("op_3232_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_3232_end_mask_0 = const()[name = tensor("op_3232_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3232_cast_fp16 = slice_by_index(begin = var_3232_begin_0, end = var_3232_end_0, end_mask = var_3232_end_mask_0, x = q_19_cast_fp16)[name = tensor("op_3232_cast_fp16")]; + tensor var_3236_begin_0 = const()[name = tensor("op_3236_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_3236_end_0 = const()[name = tensor("op_3236_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_3236_end_mask_0 = const()[name = tensor("op_3236_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3236_cast_fp16 = slice_by_index(begin = var_3236_begin_0, end = var_3236_end_0, end_mask = var_3236_end_mask_0, x = q_19_cast_fp16)[name = tensor("op_3236_cast_fp16")]; + tensor var_3240_begin_0 = const()[name = tensor("op_3240_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_3240_end_0 = const()[name = tensor("op_3240_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_3240_end_mask_0 = const()[name = tensor("op_3240_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3240_cast_fp16 = slice_by_index(begin = var_3240_begin_0, end = var_3240_end_0, end_mask = var_3240_end_mask_0, x = q_19_cast_fp16)[name = tensor("op_3240_cast_fp16")]; + tensor var_3244_begin_0 = const()[name = tensor("op_3244_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_3244_end_0 = const()[name = tensor("op_3244_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_3244_end_mask_0 = const()[name = tensor("op_3244_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3244_cast_fp16 = slice_by_index(begin = var_3244_begin_0, end = var_3244_end_0, end_mask = var_3244_end_mask_0, x = q_19_cast_fp16)[name = tensor("op_3244_cast_fp16")]; + tensor var_3248_begin_0 = const()[name = tensor("op_3248_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_3248_end_0 = const()[name = tensor("op_3248_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_3248_end_mask_0 = const()[name = tensor("op_3248_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3248_cast_fp16 = slice_by_index(begin = var_3248_begin_0, end = var_3248_end_0, end_mask = var_3248_end_mask_0, x = q_19_cast_fp16)[name = tensor("op_3248_cast_fp16")]; + tensor var_3252_begin_0 = const()[name = tensor("op_3252_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_3252_end_0 = const()[name = tensor("op_3252_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_3252_end_mask_0 = const()[name = tensor("op_3252_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3252_cast_fp16 = slice_by_index(begin = var_3252_begin_0, end = var_3252_end_0, end_mask = var_3252_end_mask_0, x = q_19_cast_fp16)[name = tensor("op_3252_cast_fp16")]; + tensor var_3256_begin_0 = const()[name = tensor("op_3256_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_3256_end_0 = const()[name = tensor("op_3256_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_3256_end_mask_0 = const()[name = tensor("op_3256_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3256_cast_fp16 = slice_by_index(begin = var_3256_begin_0, end = var_3256_end_0, end_mask = var_3256_end_mask_0, x = q_19_cast_fp16)[name = tensor("op_3256_cast_fp16")]; + tensor var_3260_begin_0 = const()[name = tensor("op_3260_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_3260_end_0 = const()[name = tensor("op_3260_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_3260_end_mask_0 = const()[name = tensor("op_3260_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3260_cast_fp16 = slice_by_index(begin = var_3260_begin_0, end = var_3260_end_0, end_mask = var_3260_end_mask_0, x = q_19_cast_fp16)[name = tensor("op_3260_cast_fp16")]; + tensor var_3264_begin_0 = const()[name = tensor("op_3264_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_3264_end_0 = const()[name = tensor("op_3264_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_3264_end_mask_0 = const()[name = tensor("op_3264_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3264_cast_fp16 = slice_by_index(begin = var_3264_begin_0, end = var_3264_end_0, end_mask = var_3264_end_mask_0, x = q_19_cast_fp16)[name = tensor("op_3264_cast_fp16")]; + tensor var_3268_begin_0 = const()[name = tensor("op_3268_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_3268_end_0 = const()[name = tensor("op_3268_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_3268_end_mask_0 = const()[name = tensor("op_3268_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3268_cast_fp16 = slice_by_index(begin = var_3268_begin_0, end = var_3268_end_0, end_mask = var_3268_end_mask_0, x = q_19_cast_fp16)[name = tensor("op_3268_cast_fp16")]; + tensor var_3272_begin_0 = const()[name = tensor("op_3272_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_3272_end_0 = const()[name = tensor("op_3272_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_3272_end_mask_0 = const()[name = tensor("op_3272_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3272_cast_fp16 = slice_by_index(begin = var_3272_begin_0, end = var_3272_end_0, end_mask = var_3272_end_mask_0, x = q_19_cast_fp16)[name = tensor("op_3272_cast_fp16")]; + tensor var_3276_begin_0 = const()[name = tensor("op_3276_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_3276_end_0 = const()[name = tensor("op_3276_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_3276_end_mask_0 = const()[name = tensor("op_3276_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3276_cast_fp16 = slice_by_index(begin = var_3276_begin_0, end = var_3276_end_0, end_mask = var_3276_end_mask_0, x = q_19_cast_fp16)[name = tensor("op_3276_cast_fp16")]; + tensor k_39_perm_0 = const()[name = tensor("k_39_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_3283_begin_0 = const()[name = tensor("op_3283_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3283_end_0 = const()[name = tensor("op_3283_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_3283_end_mask_0 = const()[name = tensor("op_3283_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_39_cast_fp16 = transpose(perm = k_39_perm_0, x = k_37_cast_fp16)[name = tensor("transpose_130")]; + tensor var_3283_cast_fp16 = slice_by_index(begin = var_3283_begin_0, end = var_3283_end_0, end_mask = var_3283_end_mask_0, x = k_39_cast_fp16)[name = tensor("op_3283_cast_fp16")]; + tensor var_3287_begin_0 = const()[name = tensor("op_3287_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_3287_end_0 = const()[name = tensor("op_3287_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_3287_end_mask_0 = const()[name = tensor("op_3287_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3287_cast_fp16 = slice_by_index(begin = var_3287_begin_0, end = var_3287_end_0, end_mask = var_3287_end_mask_0, x = k_39_cast_fp16)[name = tensor("op_3287_cast_fp16")]; + tensor var_3291_begin_0 = const()[name = tensor("op_3291_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_3291_end_0 = const()[name = tensor("op_3291_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_3291_end_mask_0 = const()[name = tensor("op_3291_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3291_cast_fp16 = slice_by_index(begin = var_3291_begin_0, end = var_3291_end_0, end_mask = var_3291_end_mask_0, x = k_39_cast_fp16)[name = tensor("op_3291_cast_fp16")]; + tensor var_3295_begin_0 = const()[name = tensor("op_3295_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_3295_end_0 = const()[name = tensor("op_3295_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_3295_end_mask_0 = const()[name = tensor("op_3295_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3295_cast_fp16 = slice_by_index(begin = var_3295_begin_0, end = var_3295_end_0, end_mask = var_3295_end_mask_0, x = k_39_cast_fp16)[name = tensor("op_3295_cast_fp16")]; + tensor var_3299_begin_0 = const()[name = tensor("op_3299_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_3299_end_0 = const()[name = tensor("op_3299_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_3299_end_mask_0 = const()[name = tensor("op_3299_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3299_cast_fp16 = slice_by_index(begin = var_3299_begin_0, end = var_3299_end_0, end_mask = var_3299_end_mask_0, x = k_39_cast_fp16)[name = tensor("op_3299_cast_fp16")]; + tensor var_3303_begin_0 = const()[name = tensor("op_3303_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_3303_end_0 = const()[name = tensor("op_3303_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_3303_end_mask_0 = const()[name = tensor("op_3303_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3303_cast_fp16 = slice_by_index(begin = var_3303_begin_0, end = var_3303_end_0, end_mask = var_3303_end_mask_0, x = k_39_cast_fp16)[name = tensor("op_3303_cast_fp16")]; + tensor var_3307_begin_0 = const()[name = tensor("op_3307_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_3307_end_0 = const()[name = tensor("op_3307_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_3307_end_mask_0 = const()[name = tensor("op_3307_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3307_cast_fp16 = slice_by_index(begin = var_3307_begin_0, end = var_3307_end_0, end_mask = var_3307_end_mask_0, x = k_39_cast_fp16)[name = tensor("op_3307_cast_fp16")]; + tensor var_3311_begin_0 = const()[name = tensor("op_3311_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_3311_end_0 = const()[name = tensor("op_3311_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_3311_end_mask_0 = const()[name = tensor("op_3311_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3311_cast_fp16 = slice_by_index(begin = var_3311_begin_0, end = var_3311_end_0, end_mask = var_3311_end_mask_0, x = k_39_cast_fp16)[name = tensor("op_3311_cast_fp16")]; + tensor var_3315_begin_0 = const()[name = tensor("op_3315_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_3315_end_0 = const()[name = tensor("op_3315_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_3315_end_mask_0 = const()[name = tensor("op_3315_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3315_cast_fp16 = slice_by_index(begin = var_3315_begin_0, end = var_3315_end_0, end_mask = var_3315_end_mask_0, x = k_39_cast_fp16)[name = tensor("op_3315_cast_fp16")]; + tensor var_3319_begin_0 = const()[name = tensor("op_3319_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_3319_end_0 = const()[name = tensor("op_3319_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_3319_end_mask_0 = const()[name = tensor("op_3319_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3319_cast_fp16 = slice_by_index(begin = var_3319_begin_0, end = var_3319_end_0, end_mask = var_3319_end_mask_0, x = k_39_cast_fp16)[name = tensor("op_3319_cast_fp16")]; + tensor var_3323_begin_0 = const()[name = tensor("op_3323_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_3323_end_0 = const()[name = tensor("op_3323_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_3323_end_mask_0 = const()[name = tensor("op_3323_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3323_cast_fp16 = slice_by_index(begin = var_3323_begin_0, end = var_3323_end_0, end_mask = var_3323_end_mask_0, x = k_39_cast_fp16)[name = tensor("op_3323_cast_fp16")]; + tensor var_3327_begin_0 = const()[name = tensor("op_3327_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_3327_end_0 = const()[name = tensor("op_3327_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_3327_end_mask_0 = const()[name = tensor("op_3327_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3327_cast_fp16 = slice_by_index(begin = var_3327_begin_0, end = var_3327_end_0, end_mask = var_3327_end_mask_0, x = k_39_cast_fp16)[name = tensor("op_3327_cast_fp16")]; + tensor var_3331_begin_0 = const()[name = tensor("op_3331_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_3331_end_0 = const()[name = tensor("op_3331_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_3331_end_mask_0 = const()[name = tensor("op_3331_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3331_cast_fp16 = slice_by_index(begin = var_3331_begin_0, end = var_3331_end_0, end_mask = var_3331_end_mask_0, x = k_39_cast_fp16)[name = tensor("op_3331_cast_fp16")]; + tensor var_3335_begin_0 = const()[name = tensor("op_3335_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_3335_end_0 = const()[name = tensor("op_3335_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_3335_end_mask_0 = const()[name = tensor("op_3335_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3335_cast_fp16 = slice_by_index(begin = var_3335_begin_0, end = var_3335_end_0, end_mask = var_3335_end_mask_0, x = k_39_cast_fp16)[name = tensor("op_3335_cast_fp16")]; + tensor var_3339_begin_0 = const()[name = tensor("op_3339_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_3339_end_0 = const()[name = tensor("op_3339_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_3339_end_mask_0 = const()[name = tensor("op_3339_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3339_cast_fp16 = slice_by_index(begin = var_3339_begin_0, end = var_3339_end_0, end_mask = var_3339_end_mask_0, x = k_39_cast_fp16)[name = tensor("op_3339_cast_fp16")]; + tensor var_3343_begin_0 = const()[name = tensor("op_3343_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_3343_end_0 = const()[name = tensor("op_3343_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_3343_end_mask_0 = const()[name = tensor("op_3343_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3343_cast_fp16 = slice_by_index(begin = var_3343_begin_0, end = var_3343_end_0, end_mask = var_3343_end_mask_0, x = k_39_cast_fp16)[name = tensor("op_3343_cast_fp16")]; + tensor var_3347_begin_0 = const()[name = tensor("op_3347_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_3347_end_0 = const()[name = tensor("op_3347_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_3347_end_mask_0 = const()[name = tensor("op_3347_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3347_cast_fp16 = slice_by_index(begin = var_3347_begin_0, end = var_3347_end_0, end_mask = var_3347_end_mask_0, x = k_39_cast_fp16)[name = tensor("op_3347_cast_fp16")]; + tensor var_3351_begin_0 = const()[name = tensor("op_3351_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_3351_end_0 = const()[name = tensor("op_3351_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_3351_end_mask_0 = const()[name = tensor("op_3351_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3351_cast_fp16 = slice_by_index(begin = var_3351_begin_0, end = var_3351_end_0, end_mask = var_3351_end_mask_0, x = k_39_cast_fp16)[name = tensor("op_3351_cast_fp16")]; + tensor var_3355_begin_0 = const()[name = tensor("op_3355_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_3355_end_0 = const()[name = tensor("op_3355_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_3355_end_mask_0 = const()[name = tensor("op_3355_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3355_cast_fp16 = slice_by_index(begin = var_3355_begin_0, end = var_3355_end_0, end_mask = var_3355_end_mask_0, x = k_39_cast_fp16)[name = tensor("op_3355_cast_fp16")]; + tensor var_3359_begin_0 = const()[name = tensor("op_3359_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_3359_end_0 = const()[name = tensor("op_3359_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_3359_end_mask_0 = const()[name = tensor("op_3359_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3359_cast_fp16 = slice_by_index(begin = var_3359_begin_0, end = var_3359_end_0, end_mask = var_3359_end_mask_0, x = k_39_cast_fp16)[name = tensor("op_3359_cast_fp16")]; + tensor var_3361_begin_0 = const()[name = tensor("op_3361_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3361_end_0 = const()[name = tensor("op_3361_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_3361_end_mask_0 = const()[name = tensor("op_3361_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3361_cast_fp16 = slice_by_index(begin = var_3361_begin_0, end = var_3361_end_0, end_mask = var_3361_end_mask_0, x = v_19_cast_fp16)[name = tensor("op_3361_cast_fp16")]; + tensor var_3365_begin_0 = const()[name = tensor("op_3365_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_3365_end_0 = const()[name = tensor("op_3365_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_3365_end_mask_0 = const()[name = tensor("op_3365_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3365_cast_fp16 = slice_by_index(begin = var_3365_begin_0, end = var_3365_end_0, end_mask = var_3365_end_mask_0, x = v_19_cast_fp16)[name = tensor("op_3365_cast_fp16")]; + tensor var_3369_begin_0 = const()[name = tensor("op_3369_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_3369_end_0 = const()[name = tensor("op_3369_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_3369_end_mask_0 = const()[name = tensor("op_3369_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3369_cast_fp16 = slice_by_index(begin = var_3369_begin_0, end = var_3369_end_0, end_mask = var_3369_end_mask_0, x = v_19_cast_fp16)[name = tensor("op_3369_cast_fp16")]; + tensor var_3373_begin_0 = const()[name = tensor("op_3373_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_3373_end_0 = const()[name = tensor("op_3373_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_3373_end_mask_0 = const()[name = tensor("op_3373_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3373_cast_fp16 = slice_by_index(begin = var_3373_begin_0, end = var_3373_end_0, end_mask = var_3373_end_mask_0, x = v_19_cast_fp16)[name = tensor("op_3373_cast_fp16")]; + tensor var_3377_begin_0 = const()[name = tensor("op_3377_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_3377_end_0 = const()[name = tensor("op_3377_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_3377_end_mask_0 = const()[name = tensor("op_3377_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3377_cast_fp16 = slice_by_index(begin = var_3377_begin_0, end = var_3377_end_0, end_mask = var_3377_end_mask_0, x = v_19_cast_fp16)[name = tensor("op_3377_cast_fp16")]; + tensor var_3381_begin_0 = const()[name = tensor("op_3381_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_3381_end_0 = const()[name = tensor("op_3381_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_3381_end_mask_0 = const()[name = tensor("op_3381_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3381_cast_fp16 = slice_by_index(begin = var_3381_begin_0, end = var_3381_end_0, end_mask = var_3381_end_mask_0, x = v_19_cast_fp16)[name = tensor("op_3381_cast_fp16")]; + tensor var_3385_begin_0 = const()[name = tensor("op_3385_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_3385_end_0 = const()[name = tensor("op_3385_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_3385_end_mask_0 = const()[name = tensor("op_3385_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3385_cast_fp16 = slice_by_index(begin = var_3385_begin_0, end = var_3385_end_0, end_mask = var_3385_end_mask_0, x = v_19_cast_fp16)[name = tensor("op_3385_cast_fp16")]; + tensor var_3389_begin_0 = const()[name = tensor("op_3389_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_3389_end_0 = const()[name = tensor("op_3389_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_3389_end_mask_0 = const()[name = tensor("op_3389_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3389_cast_fp16 = slice_by_index(begin = var_3389_begin_0, end = var_3389_end_0, end_mask = var_3389_end_mask_0, x = v_19_cast_fp16)[name = tensor("op_3389_cast_fp16")]; + tensor var_3393_begin_0 = const()[name = tensor("op_3393_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_3393_end_0 = const()[name = tensor("op_3393_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_3393_end_mask_0 = const()[name = tensor("op_3393_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3393_cast_fp16 = slice_by_index(begin = var_3393_begin_0, end = var_3393_end_0, end_mask = var_3393_end_mask_0, x = v_19_cast_fp16)[name = tensor("op_3393_cast_fp16")]; + tensor var_3397_begin_0 = const()[name = tensor("op_3397_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_3397_end_0 = const()[name = tensor("op_3397_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_3397_end_mask_0 = const()[name = tensor("op_3397_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3397_cast_fp16 = slice_by_index(begin = var_3397_begin_0, end = var_3397_end_0, end_mask = var_3397_end_mask_0, x = v_19_cast_fp16)[name = tensor("op_3397_cast_fp16")]; + tensor var_3401_begin_0 = const()[name = tensor("op_3401_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_3401_end_0 = const()[name = tensor("op_3401_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_3401_end_mask_0 = const()[name = tensor("op_3401_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3401_cast_fp16 = slice_by_index(begin = var_3401_begin_0, end = var_3401_end_0, end_mask = var_3401_end_mask_0, x = v_19_cast_fp16)[name = tensor("op_3401_cast_fp16")]; + tensor var_3405_begin_0 = const()[name = tensor("op_3405_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_3405_end_0 = const()[name = tensor("op_3405_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_3405_end_mask_0 = const()[name = tensor("op_3405_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3405_cast_fp16 = slice_by_index(begin = var_3405_begin_0, end = var_3405_end_0, end_mask = var_3405_end_mask_0, x = v_19_cast_fp16)[name = tensor("op_3405_cast_fp16")]; + tensor var_3409_begin_0 = const()[name = tensor("op_3409_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_3409_end_0 = const()[name = tensor("op_3409_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_3409_end_mask_0 = const()[name = tensor("op_3409_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3409_cast_fp16 = slice_by_index(begin = var_3409_begin_0, end = var_3409_end_0, end_mask = var_3409_end_mask_0, x = v_19_cast_fp16)[name = tensor("op_3409_cast_fp16")]; + tensor var_3413_begin_0 = const()[name = tensor("op_3413_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_3413_end_0 = const()[name = tensor("op_3413_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_3413_end_mask_0 = const()[name = tensor("op_3413_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3413_cast_fp16 = slice_by_index(begin = var_3413_begin_0, end = var_3413_end_0, end_mask = var_3413_end_mask_0, x = v_19_cast_fp16)[name = tensor("op_3413_cast_fp16")]; + tensor var_3417_begin_0 = const()[name = tensor("op_3417_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_3417_end_0 = const()[name = tensor("op_3417_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_3417_end_mask_0 = const()[name = tensor("op_3417_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3417_cast_fp16 = slice_by_index(begin = var_3417_begin_0, end = var_3417_end_0, end_mask = var_3417_end_mask_0, x = v_19_cast_fp16)[name = tensor("op_3417_cast_fp16")]; + tensor var_3421_begin_0 = const()[name = tensor("op_3421_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_3421_end_0 = const()[name = tensor("op_3421_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_3421_end_mask_0 = const()[name = tensor("op_3421_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3421_cast_fp16 = slice_by_index(begin = var_3421_begin_0, end = var_3421_end_0, end_mask = var_3421_end_mask_0, x = v_19_cast_fp16)[name = tensor("op_3421_cast_fp16")]; + tensor var_3425_begin_0 = const()[name = tensor("op_3425_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_3425_end_0 = const()[name = tensor("op_3425_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_3425_end_mask_0 = const()[name = tensor("op_3425_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3425_cast_fp16 = slice_by_index(begin = var_3425_begin_0, end = var_3425_end_0, end_mask = var_3425_end_mask_0, x = v_19_cast_fp16)[name = tensor("op_3425_cast_fp16")]; + tensor var_3429_begin_0 = const()[name = tensor("op_3429_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_3429_end_0 = const()[name = tensor("op_3429_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_3429_end_mask_0 = const()[name = tensor("op_3429_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3429_cast_fp16 = slice_by_index(begin = var_3429_begin_0, end = var_3429_end_0, end_mask = var_3429_end_mask_0, x = v_19_cast_fp16)[name = tensor("op_3429_cast_fp16")]; + tensor var_3433_begin_0 = const()[name = tensor("op_3433_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_3433_end_0 = const()[name = tensor("op_3433_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_3433_end_mask_0 = const()[name = tensor("op_3433_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3433_cast_fp16 = slice_by_index(begin = var_3433_begin_0, end = var_3433_end_0, end_mask = var_3433_end_mask_0, x = v_19_cast_fp16)[name = tensor("op_3433_cast_fp16")]; + tensor var_3437_begin_0 = const()[name = tensor("op_3437_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_3437_end_0 = const()[name = tensor("op_3437_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_3437_end_mask_0 = const()[name = tensor("op_3437_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3437_cast_fp16 = slice_by_index(begin = var_3437_begin_0, end = var_3437_end_0, end_mask = var_3437_end_mask_0, x = v_19_cast_fp16)[name = tensor("op_3437_cast_fp16")]; + tensor var_3441_equation_0 = const()[name = tensor("op_3441_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3441_cast_fp16 = einsum(equation = var_3441_equation_0, values = (var_3283_cast_fp16, var_3200_cast_fp16))[name = tensor("op_3441_cast_fp16")]; + tensor var_3442_to_fp16 = const()[name = tensor("op_3442_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_201_cast_fp16 = mul(x = var_3441_cast_fp16, y = var_3442_to_fp16)[name = tensor("aw_201_cast_fp16")]; + tensor var_3445_equation_0 = const()[name = tensor("op_3445_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3445_cast_fp16 = einsum(equation = var_3445_equation_0, values = (var_3287_cast_fp16, var_3204_cast_fp16))[name = tensor("op_3445_cast_fp16")]; + tensor var_3446_to_fp16 = const()[name = tensor("op_3446_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_203_cast_fp16 = mul(x = var_3445_cast_fp16, y = var_3446_to_fp16)[name = tensor("aw_203_cast_fp16")]; + tensor var_3449_equation_0 = const()[name = tensor("op_3449_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3449_cast_fp16 = einsum(equation = var_3449_equation_0, values = (var_3291_cast_fp16, var_3208_cast_fp16))[name = tensor("op_3449_cast_fp16")]; + tensor var_3450_to_fp16 = const()[name = tensor("op_3450_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_205_cast_fp16 = mul(x = var_3449_cast_fp16, y = var_3450_to_fp16)[name = tensor("aw_205_cast_fp16")]; + tensor var_3453_equation_0 = const()[name = tensor("op_3453_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3453_cast_fp16 = einsum(equation = var_3453_equation_0, values = (var_3295_cast_fp16, var_3212_cast_fp16))[name = tensor("op_3453_cast_fp16")]; + tensor var_3454_to_fp16 = const()[name = tensor("op_3454_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_207_cast_fp16 = mul(x = var_3453_cast_fp16, y = var_3454_to_fp16)[name = tensor("aw_207_cast_fp16")]; + tensor var_3457_equation_0 = const()[name = tensor("op_3457_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3457_cast_fp16 = einsum(equation = var_3457_equation_0, values = (var_3299_cast_fp16, var_3216_cast_fp16))[name = tensor("op_3457_cast_fp16")]; + tensor var_3458_to_fp16 = const()[name = tensor("op_3458_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_209_cast_fp16 = mul(x = var_3457_cast_fp16, y = var_3458_to_fp16)[name = tensor("aw_209_cast_fp16")]; + tensor var_3461_equation_0 = const()[name = tensor("op_3461_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3461_cast_fp16 = einsum(equation = var_3461_equation_0, values = (var_3303_cast_fp16, var_3220_cast_fp16))[name = tensor("op_3461_cast_fp16")]; + tensor var_3462_to_fp16 = const()[name = tensor("op_3462_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_211_cast_fp16 = mul(x = var_3461_cast_fp16, y = var_3462_to_fp16)[name = tensor("aw_211_cast_fp16")]; + tensor var_3465_equation_0 = const()[name = tensor("op_3465_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3465_cast_fp16 = einsum(equation = var_3465_equation_0, values = (var_3307_cast_fp16, var_3224_cast_fp16))[name = tensor("op_3465_cast_fp16")]; + tensor var_3466_to_fp16 = const()[name = tensor("op_3466_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_213_cast_fp16 = mul(x = var_3465_cast_fp16, y = var_3466_to_fp16)[name = tensor("aw_213_cast_fp16")]; + tensor var_3469_equation_0 = const()[name = tensor("op_3469_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3469_cast_fp16 = einsum(equation = var_3469_equation_0, values = (var_3311_cast_fp16, var_3228_cast_fp16))[name = tensor("op_3469_cast_fp16")]; + tensor var_3470_to_fp16 = const()[name = tensor("op_3470_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_215_cast_fp16 = mul(x = var_3469_cast_fp16, y = var_3470_to_fp16)[name = tensor("aw_215_cast_fp16")]; + tensor var_3473_equation_0 = const()[name = tensor("op_3473_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3473_cast_fp16 = einsum(equation = var_3473_equation_0, values = (var_3315_cast_fp16, var_3232_cast_fp16))[name = tensor("op_3473_cast_fp16")]; + tensor var_3474_to_fp16 = const()[name = tensor("op_3474_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_217_cast_fp16 = mul(x = var_3473_cast_fp16, y = var_3474_to_fp16)[name = tensor("aw_217_cast_fp16")]; + tensor var_3477_equation_0 = const()[name = tensor("op_3477_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3477_cast_fp16 = einsum(equation = var_3477_equation_0, values = (var_3319_cast_fp16, var_3236_cast_fp16))[name = tensor("op_3477_cast_fp16")]; + tensor var_3478_to_fp16 = const()[name = tensor("op_3478_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_219_cast_fp16 = mul(x = var_3477_cast_fp16, y = var_3478_to_fp16)[name = tensor("aw_219_cast_fp16")]; + tensor var_3481_equation_0 = const()[name = tensor("op_3481_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3481_cast_fp16 = einsum(equation = var_3481_equation_0, values = (var_3323_cast_fp16, var_3240_cast_fp16))[name = tensor("op_3481_cast_fp16")]; + tensor var_3482_to_fp16 = const()[name = tensor("op_3482_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_221_cast_fp16 = mul(x = var_3481_cast_fp16, y = var_3482_to_fp16)[name = tensor("aw_221_cast_fp16")]; + tensor var_3485_equation_0 = const()[name = tensor("op_3485_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3485_cast_fp16 = einsum(equation = var_3485_equation_0, values = (var_3327_cast_fp16, var_3244_cast_fp16))[name = tensor("op_3485_cast_fp16")]; + tensor var_3486_to_fp16 = const()[name = tensor("op_3486_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_223_cast_fp16 = mul(x = var_3485_cast_fp16, y = var_3486_to_fp16)[name = tensor("aw_223_cast_fp16")]; + tensor var_3489_equation_0 = const()[name = tensor("op_3489_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3489_cast_fp16 = einsum(equation = var_3489_equation_0, values = (var_3331_cast_fp16, var_3248_cast_fp16))[name = tensor("op_3489_cast_fp16")]; + tensor var_3490_to_fp16 = const()[name = tensor("op_3490_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_225_cast_fp16 = mul(x = var_3489_cast_fp16, y = var_3490_to_fp16)[name = tensor("aw_225_cast_fp16")]; + tensor var_3493_equation_0 = const()[name = tensor("op_3493_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3493_cast_fp16 = einsum(equation = var_3493_equation_0, values = (var_3335_cast_fp16, var_3252_cast_fp16))[name = tensor("op_3493_cast_fp16")]; + tensor var_3494_to_fp16 = const()[name = tensor("op_3494_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_227_cast_fp16 = mul(x = var_3493_cast_fp16, y = var_3494_to_fp16)[name = tensor("aw_227_cast_fp16")]; + tensor var_3497_equation_0 = const()[name = tensor("op_3497_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3497_cast_fp16 = einsum(equation = var_3497_equation_0, values = (var_3339_cast_fp16, var_3256_cast_fp16))[name = tensor("op_3497_cast_fp16")]; + tensor var_3498_to_fp16 = const()[name = tensor("op_3498_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_229_cast_fp16 = mul(x = var_3497_cast_fp16, y = var_3498_to_fp16)[name = tensor("aw_229_cast_fp16")]; + tensor var_3501_equation_0 = const()[name = tensor("op_3501_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3501_cast_fp16 = einsum(equation = var_3501_equation_0, values = (var_3343_cast_fp16, var_3260_cast_fp16))[name = tensor("op_3501_cast_fp16")]; + tensor var_3502_to_fp16 = const()[name = tensor("op_3502_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_231_cast_fp16 = mul(x = var_3501_cast_fp16, y = var_3502_to_fp16)[name = tensor("aw_231_cast_fp16")]; + tensor var_3505_equation_0 = const()[name = tensor("op_3505_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3505_cast_fp16 = einsum(equation = var_3505_equation_0, values = (var_3347_cast_fp16, var_3264_cast_fp16))[name = tensor("op_3505_cast_fp16")]; + tensor var_3506_to_fp16 = const()[name = tensor("op_3506_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_233_cast_fp16 = mul(x = var_3505_cast_fp16, y = var_3506_to_fp16)[name = tensor("aw_233_cast_fp16")]; + tensor var_3509_equation_0 = const()[name = tensor("op_3509_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3509_cast_fp16 = einsum(equation = var_3509_equation_0, values = (var_3351_cast_fp16, var_3268_cast_fp16))[name = tensor("op_3509_cast_fp16")]; + tensor var_3510_to_fp16 = const()[name = tensor("op_3510_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_235_cast_fp16 = mul(x = var_3509_cast_fp16, y = var_3510_to_fp16)[name = tensor("aw_235_cast_fp16")]; + tensor var_3513_equation_0 = const()[name = tensor("op_3513_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3513_cast_fp16 = einsum(equation = var_3513_equation_0, values = (var_3355_cast_fp16, var_3272_cast_fp16))[name = tensor("op_3513_cast_fp16")]; + tensor var_3514_to_fp16 = const()[name = tensor("op_3514_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_237_cast_fp16 = mul(x = var_3513_cast_fp16, y = var_3514_to_fp16)[name = tensor("aw_237_cast_fp16")]; + tensor var_3517_equation_0 = const()[name = tensor("op_3517_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3517_cast_fp16 = einsum(equation = var_3517_equation_0, values = (var_3359_cast_fp16, var_3276_cast_fp16))[name = tensor("op_3517_cast_fp16")]; + tensor var_3518_to_fp16 = const()[name = tensor("op_3518_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_239_cast_fp16 = mul(x = var_3517_cast_fp16, y = var_3518_to_fp16)[name = tensor("aw_239_cast_fp16")]; + tensor var_3520_cast_fp16 = softmax(axis = var_2624, x = aw_201_cast_fp16)[name = tensor("op_3520_cast_fp16")]; + tensor var_3521_cast_fp16 = softmax(axis = var_2624, x = aw_203_cast_fp16)[name = tensor("op_3521_cast_fp16")]; + tensor var_3522_cast_fp16 = softmax(axis = var_2624, x = aw_205_cast_fp16)[name = tensor("op_3522_cast_fp16")]; + tensor var_3523_cast_fp16 = softmax(axis = var_2624, x = aw_207_cast_fp16)[name = tensor("op_3523_cast_fp16")]; + tensor var_3524_cast_fp16 = softmax(axis = var_2624, x = aw_209_cast_fp16)[name = tensor("op_3524_cast_fp16")]; + tensor var_3525_cast_fp16 = softmax(axis = var_2624, x = aw_211_cast_fp16)[name = tensor("op_3525_cast_fp16")]; + tensor var_3526_cast_fp16 = softmax(axis = var_2624, x = aw_213_cast_fp16)[name = tensor("op_3526_cast_fp16")]; + tensor var_3527_cast_fp16 = softmax(axis = var_2624, x = aw_215_cast_fp16)[name = tensor("op_3527_cast_fp16")]; + tensor var_3528_cast_fp16 = softmax(axis = var_2624, x = aw_217_cast_fp16)[name = tensor("op_3528_cast_fp16")]; + tensor var_3529_cast_fp16 = softmax(axis = var_2624, x = aw_219_cast_fp16)[name = tensor("op_3529_cast_fp16")]; + tensor var_3530_cast_fp16 = softmax(axis = var_2624, x = aw_221_cast_fp16)[name = tensor("op_3530_cast_fp16")]; + tensor var_3531_cast_fp16 = softmax(axis = var_2624, x = aw_223_cast_fp16)[name = tensor("op_3531_cast_fp16")]; + tensor var_3532_cast_fp16 = softmax(axis = var_2624, x = aw_225_cast_fp16)[name = tensor("op_3532_cast_fp16")]; + tensor var_3533_cast_fp16 = softmax(axis = var_2624, x = aw_227_cast_fp16)[name = tensor("op_3533_cast_fp16")]; + tensor var_3534_cast_fp16 = softmax(axis = var_2624, x = aw_229_cast_fp16)[name = tensor("op_3534_cast_fp16")]; + tensor var_3535_cast_fp16 = softmax(axis = var_2624, x = aw_231_cast_fp16)[name = tensor("op_3535_cast_fp16")]; + tensor var_3536_cast_fp16 = softmax(axis = var_2624, x = aw_233_cast_fp16)[name = tensor("op_3536_cast_fp16")]; + tensor var_3537_cast_fp16 = softmax(axis = var_2624, x = aw_235_cast_fp16)[name = tensor("op_3537_cast_fp16")]; + tensor var_3538_cast_fp16 = softmax(axis = var_2624, x = aw_237_cast_fp16)[name = tensor("op_3538_cast_fp16")]; + tensor var_3539_cast_fp16 = softmax(axis = var_2624, x = aw_239_cast_fp16)[name = tensor("op_3539_cast_fp16")]; + tensor var_3541_equation_0 = const()[name = tensor("op_3541_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3541_cast_fp16 = einsum(equation = var_3541_equation_0, values = (var_3361_cast_fp16, var_3520_cast_fp16))[name = tensor("op_3541_cast_fp16")]; + tensor var_3543_equation_0 = const()[name = tensor("op_3543_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3543_cast_fp16 = einsum(equation = var_3543_equation_0, values = (var_3365_cast_fp16, var_3521_cast_fp16))[name = tensor("op_3543_cast_fp16")]; + tensor var_3545_equation_0 = const()[name = tensor("op_3545_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3545_cast_fp16 = einsum(equation = var_3545_equation_0, values = (var_3369_cast_fp16, var_3522_cast_fp16))[name = tensor("op_3545_cast_fp16")]; + tensor var_3547_equation_0 = const()[name = tensor("op_3547_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3547_cast_fp16 = einsum(equation = var_3547_equation_0, values = (var_3373_cast_fp16, var_3523_cast_fp16))[name = tensor("op_3547_cast_fp16")]; + tensor var_3549_equation_0 = const()[name = tensor("op_3549_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3549_cast_fp16 = einsum(equation = var_3549_equation_0, values = (var_3377_cast_fp16, var_3524_cast_fp16))[name = tensor("op_3549_cast_fp16")]; + tensor var_3551_equation_0 = const()[name = tensor("op_3551_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3551_cast_fp16 = einsum(equation = var_3551_equation_0, values = (var_3381_cast_fp16, var_3525_cast_fp16))[name = tensor("op_3551_cast_fp16")]; + tensor var_3553_equation_0 = const()[name = tensor("op_3553_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3553_cast_fp16 = einsum(equation = var_3553_equation_0, values = (var_3385_cast_fp16, var_3526_cast_fp16))[name = tensor("op_3553_cast_fp16")]; + tensor var_3555_equation_0 = const()[name = tensor("op_3555_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3555_cast_fp16 = einsum(equation = var_3555_equation_0, values = (var_3389_cast_fp16, var_3527_cast_fp16))[name = tensor("op_3555_cast_fp16")]; + tensor var_3557_equation_0 = const()[name = tensor("op_3557_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3557_cast_fp16 = einsum(equation = var_3557_equation_0, values = (var_3393_cast_fp16, var_3528_cast_fp16))[name = tensor("op_3557_cast_fp16")]; + tensor var_3559_equation_0 = const()[name = tensor("op_3559_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3559_cast_fp16 = einsum(equation = var_3559_equation_0, values = (var_3397_cast_fp16, var_3529_cast_fp16))[name = tensor("op_3559_cast_fp16")]; + tensor var_3561_equation_0 = const()[name = tensor("op_3561_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3561_cast_fp16 = einsum(equation = var_3561_equation_0, values = (var_3401_cast_fp16, var_3530_cast_fp16))[name = tensor("op_3561_cast_fp16")]; + tensor var_3563_equation_0 = const()[name = tensor("op_3563_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3563_cast_fp16 = einsum(equation = var_3563_equation_0, values = (var_3405_cast_fp16, var_3531_cast_fp16))[name = tensor("op_3563_cast_fp16")]; + tensor var_3565_equation_0 = const()[name = tensor("op_3565_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3565_cast_fp16 = einsum(equation = var_3565_equation_0, values = (var_3409_cast_fp16, var_3532_cast_fp16))[name = tensor("op_3565_cast_fp16")]; + tensor var_3567_equation_0 = const()[name = tensor("op_3567_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3567_cast_fp16 = einsum(equation = var_3567_equation_0, values = (var_3413_cast_fp16, var_3533_cast_fp16))[name = tensor("op_3567_cast_fp16")]; + tensor var_3569_equation_0 = const()[name = tensor("op_3569_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3569_cast_fp16 = einsum(equation = var_3569_equation_0, values = (var_3417_cast_fp16, var_3534_cast_fp16))[name = tensor("op_3569_cast_fp16")]; + tensor var_3571_equation_0 = const()[name = tensor("op_3571_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3571_cast_fp16 = einsum(equation = var_3571_equation_0, values = (var_3421_cast_fp16, var_3535_cast_fp16))[name = tensor("op_3571_cast_fp16")]; + tensor var_3573_equation_0 = const()[name = tensor("op_3573_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3573_cast_fp16 = einsum(equation = var_3573_equation_0, values = (var_3425_cast_fp16, var_3536_cast_fp16))[name = tensor("op_3573_cast_fp16")]; + tensor var_3575_equation_0 = const()[name = tensor("op_3575_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3575_cast_fp16 = einsum(equation = var_3575_equation_0, values = (var_3429_cast_fp16, var_3537_cast_fp16))[name = tensor("op_3575_cast_fp16")]; + tensor var_3577_equation_0 = const()[name = tensor("op_3577_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3577_cast_fp16 = einsum(equation = var_3577_equation_0, values = (var_3433_cast_fp16, var_3538_cast_fp16))[name = tensor("op_3577_cast_fp16")]; + tensor var_3579_equation_0 = const()[name = tensor("op_3579_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3579_cast_fp16 = einsum(equation = var_3579_equation_0, values = (var_3437_cast_fp16, var_3539_cast_fp16))[name = tensor("op_3579_cast_fp16")]; + tensor input_133_interleave_0 = const()[name = tensor("input_133_interleave_0"), val = tensor(false)]; + tensor input_133_cast_fp16 = concat(axis = var_2624, interleave = input_133_interleave_0, values = (var_3541_cast_fp16, var_3543_cast_fp16, var_3545_cast_fp16, var_3547_cast_fp16, var_3549_cast_fp16, var_3551_cast_fp16, var_3553_cast_fp16, var_3555_cast_fp16, var_3557_cast_fp16, var_3559_cast_fp16, var_3561_cast_fp16, var_3563_cast_fp16, var_3565_cast_fp16, var_3567_cast_fp16, var_3569_cast_fp16, var_3571_cast_fp16, var_3573_cast_fp16, var_3575_cast_fp16, var_3577_cast_fp16, var_3579_cast_fp16))[name = tensor("input_133_cast_fp16")]; + tensor var_3585 = const()[name = tensor("op_3585"), val = tensor([1, 1])]; + tensor var_3587 = const()[name = tensor("op_3587"), val = tensor([1, 1])]; + tensor var_3589_pad_type_0 = const()[name = tensor("op_3589_pad_type_0"), val = tensor("custom")]; + tensor var_3589_pad_0 = const()[name = tensor("op_3589_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(225049216)))]; + tensor down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(228326080)))]; + tensor var_3589_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_3587, groups = var_2624, pad = var_3589_pad_0, pad_type = var_3589_pad_type_0, strides = var_3585, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16, x = input_133_cast_fp16)[name = tensor("op_3589_cast_fp16")]; + tensor inputs_29_cast_fp16 = add(x = var_3589_cast_fp16, y = inputs_27_cast_fp16)[name = tensor("inputs_29_cast_fp16")]; + tensor input_135_axes_0 = const()[name = tensor("input_135_axes_0"), val = tensor([1])]; + tensor input_135_gamma_0_to_fp16 = const()[name = tensor("input_135_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(228328704)))]; + tensor input_135_beta_0_to_fp16 = const()[name = tensor("input_135_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(228331328)))]; + tensor var_3599_to_fp16 = const()[name = tensor("op_3599_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_135_cast_fp16 = layer_norm(axes = input_135_axes_0, beta = input_135_beta_0_to_fp16, epsilon = var_3599_to_fp16, gamma = input_135_gamma_0_to_fp16, x = inputs_29_cast_fp16)[name = tensor("input_135_cast_fp16")]; + tensor var_3615 = const()[name = tensor("op_3615"), val = tensor([1, 1])]; + tensor var_3617 = const()[name = tensor("op_3617"), val = tensor([1, 1])]; + tensor var_3619_pad_type_0 = const()[name = tensor("op_3619_pad_type_0"), val = tensor("custom")]; + tensor var_3619_pad_0 = const()[name = tensor("op_3619_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(228333952)))]; + tensor down_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(254548416)))]; + tensor var_3619_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_3617, groups = var_2624, pad = var_3619_pad_0, pad_type = var_3619_pad_type_0, strides = var_3615, weight = down_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16, x = input_135_cast_fp16)[name = tensor("op_3619_cast_fp16")]; + tensor var_3620_split_sizes_0 = const()[name = tensor("op_3620_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_3620_axis_0 = const()[name = tensor("op_3620_axis_0"), val = tensor(1)]; + tensor var_3620_cast_fp16_0, tensor var_3620_cast_fp16_1 = split(axis = var_3620_axis_0, split_sizes = var_3620_split_sizes_0, x = var_3619_cast_fp16)[name = tensor("op_3620_cast_fp16")]; + tensor var_3622_mode_0 = const()[name = tensor("op_3622_mode_0"), val = tensor("EXACT")]; + tensor var_3622_cast_fp16 = gelu(mode = var_3622_mode_0, x = var_3620_cast_fp16_1)[name = tensor("op_3622_cast_fp16")]; + tensor input_137_cast_fp16 = mul(x = var_3620_cast_fp16_0, y = var_3622_cast_fp16)[name = tensor("input_137_cast_fp16")]; + tensor var_3626 = const()[name = tensor("op_3626"), val = tensor([1, 1])]; + tensor var_3628 = const()[name = tensor("op_3628"), val = tensor([1, 1])]; + tensor var_3630_pad_type_0 = const()[name = tensor("op_3630_pad_type_0"), val = tensor("custom")]; + tensor var_3630_pad_0 = const()[name = tensor("op_3630_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(254568960)))]; + tensor down_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(267676224)))]; + tensor var_3630_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_3628, groups = var_2624, pad = var_3630_pad_0, pad_type = var_3630_pad_type_0, strides = var_3626, weight = down_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16, x = input_137_cast_fp16)[name = tensor("op_3630_cast_fp16")]; + tensor inputs_31_cast_fp16 = add(x = var_3630_cast_fp16, y = inputs_29_cast_fp16)[name = tensor("inputs_31_cast_fp16")]; + tensor hidden_states_71_axes_0 = const()[name = tensor("hidden_states_71_axes_0"), val = tensor([1])]; + tensor hidden_states_71_gamma_0_to_fp16 = const()[name = tensor("hidden_states_71_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(267678848)))]; + tensor hidden_states_71_beta_0_to_fp16 = const()[name = tensor("hidden_states_71_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(267681472)))]; + tensor var_3646_to_fp16 = const()[name = tensor("op_3646_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_71_cast_fp16 = layer_norm(axes = hidden_states_71_axes_0, beta = hidden_states_71_beta_0_to_fp16, epsilon = var_3646_to_fp16, gamma = hidden_states_71_gamma_0_to_fp16, x = inputs_31_cast_fp16)[name = tensor("hidden_states_71_cast_fp16")]; + tensor var_3661 = const()[name = tensor("op_3661"), val = tensor([1, 1])]; + tensor var_3663 = const()[name = tensor("op_3663"), val = tensor([1, 1])]; + tensor q_21_pad_type_0 = const()[name = tensor("q_21_pad_type_0"), val = tensor("custom")]; + tensor q_21_pad_0 = const()[name = tensor("q_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_1_attn1_to_q_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_1_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(267684096)))]; + tensor q_21_cast_fp16 = conv(dilations = var_3663, groups = var_2624, pad = q_21_pad_0, pad_type = q_21_pad_type_0, strides = var_3661, weight = down_blocks_2_attentions_0_transformer_blocks_1_attn1_to_q_weight_to_fp16, x = hidden_states_71_cast_fp16)[name = tensor("q_21_cast_fp16")]; + tensor var_3667 = const()[name = tensor("op_3667"), val = tensor([1, 1])]; + tensor var_3669 = const()[name = tensor("op_3669"), val = tensor([1, 1])]; + tensor k_41_pad_type_0 = const()[name = tensor("k_41_pad_type_0"), val = tensor("custom")]; + tensor k_41_pad_0 = const()[name = tensor("k_41_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_1_attn1_to_k_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_1_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(270960960)))]; + tensor k_41_cast_fp16 = conv(dilations = var_3669, groups = var_2624, pad = k_41_pad_0, pad_type = k_41_pad_type_0, strides = var_3667, weight = down_blocks_2_attentions_0_transformer_blocks_1_attn1_to_k_weight_to_fp16, x = hidden_states_71_cast_fp16)[name = tensor("k_41_cast_fp16")]; + tensor var_3673 = const()[name = tensor("op_3673"), val = tensor([1, 1])]; + tensor var_3675 = const()[name = tensor("op_3675"), val = tensor([1, 1])]; + tensor v_21_pad_type_0 = const()[name = tensor("v_21_pad_type_0"), val = tensor("custom")]; + tensor v_21_pad_0 = const()[name = tensor("v_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_1_attn1_to_v_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_1_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(274237824)))]; + tensor v_21_cast_fp16 = conv(dilations = var_3675, groups = var_2624, pad = v_21_pad_0, pad_type = v_21_pad_type_0, strides = var_3673, weight = down_blocks_2_attentions_0_transformer_blocks_1_attn1_to_v_weight_to_fp16, x = hidden_states_71_cast_fp16)[name = tensor("v_21_cast_fp16")]; + tensor var_3679_begin_0 = const()[name = tensor("op_3679_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3679_end_0 = const()[name = tensor("op_3679_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_3679_end_mask_0 = const()[name = tensor("op_3679_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3679_cast_fp16 = slice_by_index(begin = var_3679_begin_0, end = var_3679_end_0, end_mask = var_3679_end_mask_0, x = q_21_cast_fp16)[name = tensor("op_3679_cast_fp16")]; + tensor var_3683_begin_0 = const()[name = tensor("op_3683_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_3683_end_0 = const()[name = tensor("op_3683_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_3683_end_mask_0 = const()[name = tensor("op_3683_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3683_cast_fp16 = slice_by_index(begin = var_3683_begin_0, end = var_3683_end_0, end_mask = var_3683_end_mask_0, x = q_21_cast_fp16)[name = tensor("op_3683_cast_fp16")]; + tensor var_3687_begin_0 = const()[name = tensor("op_3687_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_3687_end_0 = const()[name = tensor("op_3687_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_3687_end_mask_0 = const()[name = tensor("op_3687_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3687_cast_fp16 = slice_by_index(begin = var_3687_begin_0, end = var_3687_end_0, end_mask = var_3687_end_mask_0, x = q_21_cast_fp16)[name = tensor("op_3687_cast_fp16")]; + tensor var_3691_begin_0 = const()[name = tensor("op_3691_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_3691_end_0 = const()[name = tensor("op_3691_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_3691_end_mask_0 = const()[name = tensor("op_3691_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3691_cast_fp16 = slice_by_index(begin = var_3691_begin_0, end = var_3691_end_0, end_mask = var_3691_end_mask_0, x = q_21_cast_fp16)[name = tensor("op_3691_cast_fp16")]; + tensor var_3695_begin_0 = const()[name = tensor("op_3695_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_3695_end_0 = const()[name = tensor("op_3695_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_3695_end_mask_0 = const()[name = tensor("op_3695_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3695_cast_fp16 = slice_by_index(begin = var_3695_begin_0, end = var_3695_end_0, end_mask = var_3695_end_mask_0, x = q_21_cast_fp16)[name = tensor("op_3695_cast_fp16")]; + tensor var_3699_begin_0 = const()[name = tensor("op_3699_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_3699_end_0 = const()[name = tensor("op_3699_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_3699_end_mask_0 = const()[name = tensor("op_3699_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3699_cast_fp16 = slice_by_index(begin = var_3699_begin_0, end = var_3699_end_0, end_mask = var_3699_end_mask_0, x = q_21_cast_fp16)[name = tensor("op_3699_cast_fp16")]; + tensor var_3703_begin_0 = const()[name = tensor("op_3703_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_3703_end_0 = const()[name = tensor("op_3703_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_3703_end_mask_0 = const()[name = tensor("op_3703_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3703_cast_fp16 = slice_by_index(begin = var_3703_begin_0, end = var_3703_end_0, end_mask = var_3703_end_mask_0, x = q_21_cast_fp16)[name = tensor("op_3703_cast_fp16")]; + tensor var_3707_begin_0 = const()[name = tensor("op_3707_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_3707_end_0 = const()[name = tensor("op_3707_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_3707_end_mask_0 = const()[name = tensor("op_3707_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3707_cast_fp16 = slice_by_index(begin = var_3707_begin_0, end = var_3707_end_0, end_mask = var_3707_end_mask_0, x = q_21_cast_fp16)[name = tensor("op_3707_cast_fp16")]; + tensor var_3711_begin_0 = const()[name = tensor("op_3711_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_3711_end_0 = const()[name = tensor("op_3711_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_3711_end_mask_0 = const()[name = tensor("op_3711_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3711_cast_fp16 = slice_by_index(begin = var_3711_begin_0, end = var_3711_end_0, end_mask = var_3711_end_mask_0, x = q_21_cast_fp16)[name = tensor("op_3711_cast_fp16")]; + tensor var_3715_begin_0 = const()[name = tensor("op_3715_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_3715_end_0 = const()[name = tensor("op_3715_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_3715_end_mask_0 = const()[name = tensor("op_3715_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3715_cast_fp16 = slice_by_index(begin = var_3715_begin_0, end = var_3715_end_0, end_mask = var_3715_end_mask_0, x = q_21_cast_fp16)[name = tensor("op_3715_cast_fp16")]; + tensor var_3719_begin_0 = const()[name = tensor("op_3719_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_3719_end_0 = const()[name = tensor("op_3719_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_3719_end_mask_0 = const()[name = tensor("op_3719_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3719_cast_fp16 = slice_by_index(begin = var_3719_begin_0, end = var_3719_end_0, end_mask = var_3719_end_mask_0, x = q_21_cast_fp16)[name = tensor("op_3719_cast_fp16")]; + tensor var_3723_begin_0 = const()[name = tensor("op_3723_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_3723_end_0 = const()[name = tensor("op_3723_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_3723_end_mask_0 = const()[name = tensor("op_3723_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3723_cast_fp16 = slice_by_index(begin = var_3723_begin_0, end = var_3723_end_0, end_mask = var_3723_end_mask_0, x = q_21_cast_fp16)[name = tensor("op_3723_cast_fp16")]; + tensor var_3727_begin_0 = const()[name = tensor("op_3727_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_3727_end_0 = const()[name = tensor("op_3727_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_3727_end_mask_0 = const()[name = tensor("op_3727_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3727_cast_fp16 = slice_by_index(begin = var_3727_begin_0, end = var_3727_end_0, end_mask = var_3727_end_mask_0, x = q_21_cast_fp16)[name = tensor("op_3727_cast_fp16")]; + tensor var_3731_begin_0 = const()[name = tensor("op_3731_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_3731_end_0 = const()[name = tensor("op_3731_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_3731_end_mask_0 = const()[name = tensor("op_3731_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3731_cast_fp16 = slice_by_index(begin = var_3731_begin_0, end = var_3731_end_0, end_mask = var_3731_end_mask_0, x = q_21_cast_fp16)[name = tensor("op_3731_cast_fp16")]; + tensor var_3735_begin_0 = const()[name = tensor("op_3735_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_3735_end_0 = const()[name = tensor("op_3735_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_3735_end_mask_0 = const()[name = tensor("op_3735_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3735_cast_fp16 = slice_by_index(begin = var_3735_begin_0, end = var_3735_end_0, end_mask = var_3735_end_mask_0, x = q_21_cast_fp16)[name = tensor("op_3735_cast_fp16")]; + tensor var_3739_begin_0 = const()[name = tensor("op_3739_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_3739_end_0 = const()[name = tensor("op_3739_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_3739_end_mask_0 = const()[name = tensor("op_3739_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3739_cast_fp16 = slice_by_index(begin = var_3739_begin_0, end = var_3739_end_0, end_mask = var_3739_end_mask_0, x = q_21_cast_fp16)[name = tensor("op_3739_cast_fp16")]; + tensor var_3743_begin_0 = const()[name = tensor("op_3743_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_3743_end_0 = const()[name = tensor("op_3743_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_3743_end_mask_0 = const()[name = tensor("op_3743_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3743_cast_fp16 = slice_by_index(begin = var_3743_begin_0, end = var_3743_end_0, end_mask = var_3743_end_mask_0, x = q_21_cast_fp16)[name = tensor("op_3743_cast_fp16")]; + tensor var_3747_begin_0 = const()[name = tensor("op_3747_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_3747_end_0 = const()[name = tensor("op_3747_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_3747_end_mask_0 = const()[name = tensor("op_3747_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3747_cast_fp16 = slice_by_index(begin = var_3747_begin_0, end = var_3747_end_0, end_mask = var_3747_end_mask_0, x = q_21_cast_fp16)[name = tensor("op_3747_cast_fp16")]; + tensor var_3751_begin_0 = const()[name = tensor("op_3751_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_3751_end_0 = const()[name = tensor("op_3751_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_3751_end_mask_0 = const()[name = tensor("op_3751_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3751_cast_fp16 = slice_by_index(begin = var_3751_begin_0, end = var_3751_end_0, end_mask = var_3751_end_mask_0, x = q_21_cast_fp16)[name = tensor("op_3751_cast_fp16")]; + tensor var_3755_begin_0 = const()[name = tensor("op_3755_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_3755_end_0 = const()[name = tensor("op_3755_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_3755_end_mask_0 = const()[name = tensor("op_3755_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3755_cast_fp16 = slice_by_index(begin = var_3755_begin_0, end = var_3755_end_0, end_mask = var_3755_end_mask_0, x = q_21_cast_fp16)[name = tensor("op_3755_cast_fp16")]; + tensor k_43_perm_0 = const()[name = tensor("k_43_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_3762_begin_0 = const()[name = tensor("op_3762_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3762_end_0 = const()[name = tensor("op_3762_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_3762_end_mask_0 = const()[name = tensor("op_3762_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_43_cast_fp16 = transpose(perm = k_43_perm_0, x = k_41_cast_fp16)[name = tensor("transpose_129")]; + tensor var_3762_cast_fp16 = slice_by_index(begin = var_3762_begin_0, end = var_3762_end_0, end_mask = var_3762_end_mask_0, x = k_43_cast_fp16)[name = tensor("op_3762_cast_fp16")]; + tensor var_3766_begin_0 = const()[name = tensor("op_3766_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_3766_end_0 = const()[name = tensor("op_3766_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_3766_end_mask_0 = const()[name = tensor("op_3766_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3766_cast_fp16 = slice_by_index(begin = var_3766_begin_0, end = var_3766_end_0, end_mask = var_3766_end_mask_0, x = k_43_cast_fp16)[name = tensor("op_3766_cast_fp16")]; + tensor var_3770_begin_0 = const()[name = tensor("op_3770_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_3770_end_0 = const()[name = tensor("op_3770_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_3770_end_mask_0 = const()[name = tensor("op_3770_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3770_cast_fp16 = slice_by_index(begin = var_3770_begin_0, end = var_3770_end_0, end_mask = var_3770_end_mask_0, x = k_43_cast_fp16)[name = tensor("op_3770_cast_fp16")]; + tensor var_3774_begin_0 = const()[name = tensor("op_3774_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_3774_end_0 = const()[name = tensor("op_3774_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_3774_end_mask_0 = const()[name = tensor("op_3774_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3774_cast_fp16 = slice_by_index(begin = var_3774_begin_0, end = var_3774_end_0, end_mask = var_3774_end_mask_0, x = k_43_cast_fp16)[name = tensor("op_3774_cast_fp16")]; + tensor var_3778_begin_0 = const()[name = tensor("op_3778_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_3778_end_0 = const()[name = tensor("op_3778_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_3778_end_mask_0 = const()[name = tensor("op_3778_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3778_cast_fp16 = slice_by_index(begin = var_3778_begin_0, end = var_3778_end_0, end_mask = var_3778_end_mask_0, x = k_43_cast_fp16)[name = tensor("op_3778_cast_fp16")]; + tensor var_3782_begin_0 = const()[name = tensor("op_3782_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_3782_end_0 = const()[name = tensor("op_3782_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_3782_end_mask_0 = const()[name = tensor("op_3782_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3782_cast_fp16 = slice_by_index(begin = var_3782_begin_0, end = var_3782_end_0, end_mask = var_3782_end_mask_0, x = k_43_cast_fp16)[name = tensor("op_3782_cast_fp16")]; + tensor var_3786_begin_0 = const()[name = tensor("op_3786_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_3786_end_0 = const()[name = tensor("op_3786_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_3786_end_mask_0 = const()[name = tensor("op_3786_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3786_cast_fp16 = slice_by_index(begin = var_3786_begin_0, end = var_3786_end_0, end_mask = var_3786_end_mask_0, x = k_43_cast_fp16)[name = tensor("op_3786_cast_fp16")]; + tensor var_3790_begin_0 = const()[name = tensor("op_3790_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_3790_end_0 = const()[name = tensor("op_3790_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_3790_end_mask_0 = const()[name = tensor("op_3790_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3790_cast_fp16 = slice_by_index(begin = var_3790_begin_0, end = var_3790_end_0, end_mask = var_3790_end_mask_0, x = k_43_cast_fp16)[name = tensor("op_3790_cast_fp16")]; + tensor var_3794_begin_0 = const()[name = tensor("op_3794_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_3794_end_0 = const()[name = tensor("op_3794_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_3794_end_mask_0 = const()[name = tensor("op_3794_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3794_cast_fp16 = slice_by_index(begin = var_3794_begin_0, end = var_3794_end_0, end_mask = var_3794_end_mask_0, x = k_43_cast_fp16)[name = tensor("op_3794_cast_fp16")]; + tensor var_3798_begin_0 = const()[name = tensor("op_3798_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_3798_end_0 = const()[name = tensor("op_3798_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_3798_end_mask_0 = const()[name = tensor("op_3798_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3798_cast_fp16 = slice_by_index(begin = var_3798_begin_0, end = var_3798_end_0, end_mask = var_3798_end_mask_0, x = k_43_cast_fp16)[name = tensor("op_3798_cast_fp16")]; + tensor var_3802_begin_0 = const()[name = tensor("op_3802_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_3802_end_0 = const()[name = tensor("op_3802_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_3802_end_mask_0 = const()[name = tensor("op_3802_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3802_cast_fp16 = slice_by_index(begin = var_3802_begin_0, end = var_3802_end_0, end_mask = var_3802_end_mask_0, x = k_43_cast_fp16)[name = tensor("op_3802_cast_fp16")]; + tensor var_3806_begin_0 = const()[name = tensor("op_3806_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_3806_end_0 = const()[name = tensor("op_3806_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_3806_end_mask_0 = const()[name = tensor("op_3806_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3806_cast_fp16 = slice_by_index(begin = var_3806_begin_0, end = var_3806_end_0, end_mask = var_3806_end_mask_0, x = k_43_cast_fp16)[name = tensor("op_3806_cast_fp16")]; + tensor var_3810_begin_0 = const()[name = tensor("op_3810_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_3810_end_0 = const()[name = tensor("op_3810_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_3810_end_mask_0 = const()[name = tensor("op_3810_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3810_cast_fp16 = slice_by_index(begin = var_3810_begin_0, end = var_3810_end_0, end_mask = var_3810_end_mask_0, x = k_43_cast_fp16)[name = tensor("op_3810_cast_fp16")]; + tensor var_3814_begin_0 = const()[name = tensor("op_3814_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_3814_end_0 = const()[name = tensor("op_3814_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_3814_end_mask_0 = const()[name = tensor("op_3814_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3814_cast_fp16 = slice_by_index(begin = var_3814_begin_0, end = var_3814_end_0, end_mask = var_3814_end_mask_0, x = k_43_cast_fp16)[name = tensor("op_3814_cast_fp16")]; + tensor var_3818_begin_0 = const()[name = tensor("op_3818_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_3818_end_0 = const()[name = tensor("op_3818_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_3818_end_mask_0 = const()[name = tensor("op_3818_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3818_cast_fp16 = slice_by_index(begin = var_3818_begin_0, end = var_3818_end_0, end_mask = var_3818_end_mask_0, x = k_43_cast_fp16)[name = tensor("op_3818_cast_fp16")]; + tensor var_3822_begin_0 = const()[name = tensor("op_3822_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_3822_end_0 = const()[name = tensor("op_3822_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_3822_end_mask_0 = const()[name = tensor("op_3822_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3822_cast_fp16 = slice_by_index(begin = var_3822_begin_0, end = var_3822_end_0, end_mask = var_3822_end_mask_0, x = k_43_cast_fp16)[name = tensor("op_3822_cast_fp16")]; + tensor var_3826_begin_0 = const()[name = tensor("op_3826_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_3826_end_0 = const()[name = tensor("op_3826_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_3826_end_mask_0 = const()[name = tensor("op_3826_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3826_cast_fp16 = slice_by_index(begin = var_3826_begin_0, end = var_3826_end_0, end_mask = var_3826_end_mask_0, x = k_43_cast_fp16)[name = tensor("op_3826_cast_fp16")]; + tensor var_3830_begin_0 = const()[name = tensor("op_3830_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_3830_end_0 = const()[name = tensor("op_3830_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_3830_end_mask_0 = const()[name = tensor("op_3830_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3830_cast_fp16 = slice_by_index(begin = var_3830_begin_0, end = var_3830_end_0, end_mask = var_3830_end_mask_0, x = k_43_cast_fp16)[name = tensor("op_3830_cast_fp16")]; + tensor var_3834_begin_0 = const()[name = tensor("op_3834_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_3834_end_0 = const()[name = tensor("op_3834_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_3834_end_mask_0 = const()[name = tensor("op_3834_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3834_cast_fp16 = slice_by_index(begin = var_3834_begin_0, end = var_3834_end_0, end_mask = var_3834_end_mask_0, x = k_43_cast_fp16)[name = tensor("op_3834_cast_fp16")]; + tensor var_3838_begin_0 = const()[name = tensor("op_3838_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_3838_end_0 = const()[name = tensor("op_3838_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_3838_end_mask_0 = const()[name = tensor("op_3838_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3838_cast_fp16 = slice_by_index(begin = var_3838_begin_0, end = var_3838_end_0, end_mask = var_3838_end_mask_0, x = k_43_cast_fp16)[name = tensor("op_3838_cast_fp16")]; + tensor var_3840_begin_0 = const()[name = tensor("op_3840_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3840_end_0 = const()[name = tensor("op_3840_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_3840_end_mask_0 = const()[name = tensor("op_3840_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3840_cast_fp16 = slice_by_index(begin = var_3840_begin_0, end = var_3840_end_0, end_mask = var_3840_end_mask_0, x = v_21_cast_fp16)[name = tensor("op_3840_cast_fp16")]; + tensor var_3844_begin_0 = const()[name = tensor("op_3844_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_3844_end_0 = const()[name = tensor("op_3844_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_3844_end_mask_0 = const()[name = tensor("op_3844_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3844_cast_fp16 = slice_by_index(begin = var_3844_begin_0, end = var_3844_end_0, end_mask = var_3844_end_mask_0, x = v_21_cast_fp16)[name = tensor("op_3844_cast_fp16")]; + tensor var_3848_begin_0 = const()[name = tensor("op_3848_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_3848_end_0 = const()[name = tensor("op_3848_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_3848_end_mask_0 = const()[name = tensor("op_3848_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3848_cast_fp16 = slice_by_index(begin = var_3848_begin_0, end = var_3848_end_0, end_mask = var_3848_end_mask_0, x = v_21_cast_fp16)[name = tensor("op_3848_cast_fp16")]; + tensor var_3852_begin_0 = const()[name = tensor("op_3852_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_3852_end_0 = const()[name = tensor("op_3852_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_3852_end_mask_0 = const()[name = tensor("op_3852_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3852_cast_fp16 = slice_by_index(begin = var_3852_begin_0, end = var_3852_end_0, end_mask = var_3852_end_mask_0, x = v_21_cast_fp16)[name = tensor("op_3852_cast_fp16")]; + tensor var_3856_begin_0 = const()[name = tensor("op_3856_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_3856_end_0 = const()[name = tensor("op_3856_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_3856_end_mask_0 = const()[name = tensor("op_3856_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3856_cast_fp16 = slice_by_index(begin = var_3856_begin_0, end = var_3856_end_0, end_mask = var_3856_end_mask_0, x = v_21_cast_fp16)[name = tensor("op_3856_cast_fp16")]; + tensor var_3860_begin_0 = const()[name = tensor("op_3860_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_3860_end_0 = const()[name = tensor("op_3860_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_3860_end_mask_0 = const()[name = tensor("op_3860_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3860_cast_fp16 = slice_by_index(begin = var_3860_begin_0, end = var_3860_end_0, end_mask = var_3860_end_mask_0, x = v_21_cast_fp16)[name = tensor("op_3860_cast_fp16")]; + tensor var_3864_begin_0 = const()[name = tensor("op_3864_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_3864_end_0 = const()[name = tensor("op_3864_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_3864_end_mask_0 = const()[name = tensor("op_3864_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3864_cast_fp16 = slice_by_index(begin = var_3864_begin_0, end = var_3864_end_0, end_mask = var_3864_end_mask_0, x = v_21_cast_fp16)[name = tensor("op_3864_cast_fp16")]; + tensor var_3868_begin_0 = const()[name = tensor("op_3868_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_3868_end_0 = const()[name = tensor("op_3868_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_3868_end_mask_0 = const()[name = tensor("op_3868_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3868_cast_fp16 = slice_by_index(begin = var_3868_begin_0, end = var_3868_end_0, end_mask = var_3868_end_mask_0, x = v_21_cast_fp16)[name = tensor("op_3868_cast_fp16")]; + tensor var_3872_begin_0 = const()[name = tensor("op_3872_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_3872_end_0 = const()[name = tensor("op_3872_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_3872_end_mask_0 = const()[name = tensor("op_3872_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3872_cast_fp16 = slice_by_index(begin = var_3872_begin_0, end = var_3872_end_0, end_mask = var_3872_end_mask_0, x = v_21_cast_fp16)[name = tensor("op_3872_cast_fp16")]; + tensor var_3876_begin_0 = const()[name = tensor("op_3876_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_3876_end_0 = const()[name = tensor("op_3876_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_3876_end_mask_0 = const()[name = tensor("op_3876_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3876_cast_fp16 = slice_by_index(begin = var_3876_begin_0, end = var_3876_end_0, end_mask = var_3876_end_mask_0, x = v_21_cast_fp16)[name = tensor("op_3876_cast_fp16")]; + tensor var_3880_begin_0 = const()[name = tensor("op_3880_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_3880_end_0 = const()[name = tensor("op_3880_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_3880_end_mask_0 = const()[name = tensor("op_3880_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3880_cast_fp16 = slice_by_index(begin = var_3880_begin_0, end = var_3880_end_0, end_mask = var_3880_end_mask_0, x = v_21_cast_fp16)[name = tensor("op_3880_cast_fp16")]; + tensor var_3884_begin_0 = const()[name = tensor("op_3884_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_3884_end_0 = const()[name = tensor("op_3884_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_3884_end_mask_0 = const()[name = tensor("op_3884_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3884_cast_fp16 = slice_by_index(begin = var_3884_begin_0, end = var_3884_end_0, end_mask = var_3884_end_mask_0, x = v_21_cast_fp16)[name = tensor("op_3884_cast_fp16")]; + tensor var_3888_begin_0 = const()[name = tensor("op_3888_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_3888_end_0 = const()[name = tensor("op_3888_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_3888_end_mask_0 = const()[name = tensor("op_3888_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3888_cast_fp16 = slice_by_index(begin = var_3888_begin_0, end = var_3888_end_0, end_mask = var_3888_end_mask_0, x = v_21_cast_fp16)[name = tensor("op_3888_cast_fp16")]; + tensor var_3892_begin_0 = const()[name = tensor("op_3892_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_3892_end_0 = const()[name = tensor("op_3892_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_3892_end_mask_0 = const()[name = tensor("op_3892_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3892_cast_fp16 = slice_by_index(begin = var_3892_begin_0, end = var_3892_end_0, end_mask = var_3892_end_mask_0, x = v_21_cast_fp16)[name = tensor("op_3892_cast_fp16")]; + tensor var_3896_begin_0 = const()[name = tensor("op_3896_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_3896_end_0 = const()[name = tensor("op_3896_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_3896_end_mask_0 = const()[name = tensor("op_3896_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3896_cast_fp16 = slice_by_index(begin = var_3896_begin_0, end = var_3896_end_0, end_mask = var_3896_end_mask_0, x = v_21_cast_fp16)[name = tensor("op_3896_cast_fp16")]; + tensor var_3900_begin_0 = const()[name = tensor("op_3900_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_3900_end_0 = const()[name = tensor("op_3900_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_3900_end_mask_0 = const()[name = tensor("op_3900_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3900_cast_fp16 = slice_by_index(begin = var_3900_begin_0, end = var_3900_end_0, end_mask = var_3900_end_mask_0, x = v_21_cast_fp16)[name = tensor("op_3900_cast_fp16")]; + tensor var_3904_begin_0 = const()[name = tensor("op_3904_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_3904_end_0 = const()[name = tensor("op_3904_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_3904_end_mask_0 = const()[name = tensor("op_3904_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3904_cast_fp16 = slice_by_index(begin = var_3904_begin_0, end = var_3904_end_0, end_mask = var_3904_end_mask_0, x = v_21_cast_fp16)[name = tensor("op_3904_cast_fp16")]; + tensor var_3908_begin_0 = const()[name = tensor("op_3908_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_3908_end_0 = const()[name = tensor("op_3908_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_3908_end_mask_0 = const()[name = tensor("op_3908_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3908_cast_fp16 = slice_by_index(begin = var_3908_begin_0, end = var_3908_end_0, end_mask = var_3908_end_mask_0, x = v_21_cast_fp16)[name = tensor("op_3908_cast_fp16")]; + tensor var_3912_begin_0 = const()[name = tensor("op_3912_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_3912_end_0 = const()[name = tensor("op_3912_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_3912_end_mask_0 = const()[name = tensor("op_3912_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3912_cast_fp16 = slice_by_index(begin = var_3912_begin_0, end = var_3912_end_0, end_mask = var_3912_end_mask_0, x = v_21_cast_fp16)[name = tensor("op_3912_cast_fp16")]; + tensor var_3916_begin_0 = const()[name = tensor("op_3916_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_3916_end_0 = const()[name = tensor("op_3916_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_3916_end_mask_0 = const()[name = tensor("op_3916_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3916_cast_fp16 = slice_by_index(begin = var_3916_begin_0, end = var_3916_end_0, end_mask = var_3916_end_mask_0, x = v_21_cast_fp16)[name = tensor("op_3916_cast_fp16")]; + tensor var_3920_equation_0 = const()[name = tensor("op_3920_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3920_cast_fp16 = einsum(equation = var_3920_equation_0, values = (var_3762_cast_fp16, var_3679_cast_fp16))[name = tensor("op_3920_cast_fp16")]; + tensor var_3921_to_fp16 = const()[name = tensor("op_3921_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_241_cast_fp16 = mul(x = var_3920_cast_fp16, y = var_3921_to_fp16)[name = tensor("aw_241_cast_fp16")]; + tensor var_3924_equation_0 = const()[name = tensor("op_3924_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3924_cast_fp16 = einsum(equation = var_3924_equation_0, values = (var_3766_cast_fp16, var_3683_cast_fp16))[name = tensor("op_3924_cast_fp16")]; + tensor var_3925_to_fp16 = const()[name = tensor("op_3925_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_243_cast_fp16 = mul(x = var_3924_cast_fp16, y = var_3925_to_fp16)[name = tensor("aw_243_cast_fp16")]; + tensor var_3928_equation_0 = const()[name = tensor("op_3928_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3928_cast_fp16 = einsum(equation = var_3928_equation_0, values = (var_3770_cast_fp16, var_3687_cast_fp16))[name = tensor("op_3928_cast_fp16")]; + tensor var_3929_to_fp16 = const()[name = tensor("op_3929_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_245_cast_fp16 = mul(x = var_3928_cast_fp16, y = var_3929_to_fp16)[name = tensor("aw_245_cast_fp16")]; + tensor var_3932_equation_0 = const()[name = tensor("op_3932_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3932_cast_fp16 = einsum(equation = var_3932_equation_0, values = (var_3774_cast_fp16, var_3691_cast_fp16))[name = tensor("op_3932_cast_fp16")]; + tensor var_3933_to_fp16 = const()[name = tensor("op_3933_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_247_cast_fp16 = mul(x = var_3932_cast_fp16, y = var_3933_to_fp16)[name = tensor("aw_247_cast_fp16")]; + tensor var_3936_equation_0 = const()[name = tensor("op_3936_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3936_cast_fp16 = einsum(equation = var_3936_equation_0, values = (var_3778_cast_fp16, var_3695_cast_fp16))[name = tensor("op_3936_cast_fp16")]; + tensor var_3937_to_fp16 = const()[name = tensor("op_3937_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_249_cast_fp16 = mul(x = var_3936_cast_fp16, y = var_3937_to_fp16)[name = tensor("aw_249_cast_fp16")]; + tensor var_3940_equation_0 = const()[name = tensor("op_3940_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3940_cast_fp16 = einsum(equation = var_3940_equation_0, values = (var_3782_cast_fp16, var_3699_cast_fp16))[name = tensor("op_3940_cast_fp16")]; + tensor var_3941_to_fp16 = const()[name = tensor("op_3941_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_251_cast_fp16 = mul(x = var_3940_cast_fp16, y = var_3941_to_fp16)[name = tensor("aw_251_cast_fp16")]; + tensor var_3944_equation_0 = const()[name = tensor("op_3944_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3944_cast_fp16 = einsum(equation = var_3944_equation_0, values = (var_3786_cast_fp16, var_3703_cast_fp16))[name = tensor("op_3944_cast_fp16")]; + tensor var_3945_to_fp16 = const()[name = tensor("op_3945_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_253_cast_fp16 = mul(x = var_3944_cast_fp16, y = var_3945_to_fp16)[name = tensor("aw_253_cast_fp16")]; + tensor var_3948_equation_0 = const()[name = tensor("op_3948_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3948_cast_fp16 = einsum(equation = var_3948_equation_0, values = (var_3790_cast_fp16, var_3707_cast_fp16))[name = tensor("op_3948_cast_fp16")]; + tensor var_3949_to_fp16 = const()[name = tensor("op_3949_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_255_cast_fp16 = mul(x = var_3948_cast_fp16, y = var_3949_to_fp16)[name = tensor("aw_255_cast_fp16")]; + tensor var_3952_equation_0 = const()[name = tensor("op_3952_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3952_cast_fp16 = einsum(equation = var_3952_equation_0, values = (var_3794_cast_fp16, var_3711_cast_fp16))[name = tensor("op_3952_cast_fp16")]; + tensor var_3953_to_fp16 = const()[name = tensor("op_3953_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_257_cast_fp16 = mul(x = var_3952_cast_fp16, y = var_3953_to_fp16)[name = tensor("aw_257_cast_fp16")]; + tensor var_3956_equation_0 = const()[name = tensor("op_3956_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3956_cast_fp16 = einsum(equation = var_3956_equation_0, values = (var_3798_cast_fp16, var_3715_cast_fp16))[name = tensor("op_3956_cast_fp16")]; + tensor var_3957_to_fp16 = const()[name = tensor("op_3957_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_259_cast_fp16 = mul(x = var_3956_cast_fp16, y = var_3957_to_fp16)[name = tensor("aw_259_cast_fp16")]; + tensor var_3960_equation_0 = const()[name = tensor("op_3960_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3960_cast_fp16 = einsum(equation = var_3960_equation_0, values = (var_3802_cast_fp16, var_3719_cast_fp16))[name = tensor("op_3960_cast_fp16")]; + tensor var_3961_to_fp16 = const()[name = tensor("op_3961_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_261_cast_fp16 = mul(x = var_3960_cast_fp16, y = var_3961_to_fp16)[name = tensor("aw_261_cast_fp16")]; + tensor var_3964_equation_0 = const()[name = tensor("op_3964_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3964_cast_fp16 = einsum(equation = var_3964_equation_0, values = (var_3806_cast_fp16, var_3723_cast_fp16))[name = tensor("op_3964_cast_fp16")]; + tensor var_3965_to_fp16 = const()[name = tensor("op_3965_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_263_cast_fp16 = mul(x = var_3964_cast_fp16, y = var_3965_to_fp16)[name = tensor("aw_263_cast_fp16")]; + tensor var_3968_equation_0 = const()[name = tensor("op_3968_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3968_cast_fp16 = einsum(equation = var_3968_equation_0, values = (var_3810_cast_fp16, var_3727_cast_fp16))[name = tensor("op_3968_cast_fp16")]; + tensor var_3969_to_fp16 = const()[name = tensor("op_3969_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_265_cast_fp16 = mul(x = var_3968_cast_fp16, y = var_3969_to_fp16)[name = tensor("aw_265_cast_fp16")]; + tensor var_3972_equation_0 = const()[name = tensor("op_3972_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3972_cast_fp16 = einsum(equation = var_3972_equation_0, values = (var_3814_cast_fp16, var_3731_cast_fp16))[name = tensor("op_3972_cast_fp16")]; + tensor var_3973_to_fp16 = const()[name = tensor("op_3973_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_267_cast_fp16 = mul(x = var_3972_cast_fp16, y = var_3973_to_fp16)[name = tensor("aw_267_cast_fp16")]; + tensor var_3976_equation_0 = const()[name = tensor("op_3976_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3976_cast_fp16 = einsum(equation = var_3976_equation_0, values = (var_3818_cast_fp16, var_3735_cast_fp16))[name = tensor("op_3976_cast_fp16")]; + tensor var_3977_to_fp16 = const()[name = tensor("op_3977_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_269_cast_fp16 = mul(x = var_3976_cast_fp16, y = var_3977_to_fp16)[name = tensor("aw_269_cast_fp16")]; + tensor var_3980_equation_0 = const()[name = tensor("op_3980_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3980_cast_fp16 = einsum(equation = var_3980_equation_0, values = (var_3822_cast_fp16, var_3739_cast_fp16))[name = tensor("op_3980_cast_fp16")]; + tensor var_3981_to_fp16 = const()[name = tensor("op_3981_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_271_cast_fp16 = mul(x = var_3980_cast_fp16, y = var_3981_to_fp16)[name = tensor("aw_271_cast_fp16")]; + tensor var_3984_equation_0 = const()[name = tensor("op_3984_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3984_cast_fp16 = einsum(equation = var_3984_equation_0, values = (var_3826_cast_fp16, var_3743_cast_fp16))[name = tensor("op_3984_cast_fp16")]; + tensor var_3985_to_fp16 = const()[name = tensor("op_3985_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_273_cast_fp16 = mul(x = var_3984_cast_fp16, y = var_3985_to_fp16)[name = tensor("aw_273_cast_fp16")]; + tensor var_3988_equation_0 = const()[name = tensor("op_3988_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3988_cast_fp16 = einsum(equation = var_3988_equation_0, values = (var_3830_cast_fp16, var_3747_cast_fp16))[name = tensor("op_3988_cast_fp16")]; + tensor var_3989_to_fp16 = const()[name = tensor("op_3989_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_275_cast_fp16 = mul(x = var_3988_cast_fp16, y = var_3989_to_fp16)[name = tensor("aw_275_cast_fp16")]; + tensor var_3992_equation_0 = const()[name = tensor("op_3992_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3992_cast_fp16 = einsum(equation = var_3992_equation_0, values = (var_3834_cast_fp16, var_3751_cast_fp16))[name = tensor("op_3992_cast_fp16")]; + tensor var_3993_to_fp16 = const()[name = tensor("op_3993_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_277_cast_fp16 = mul(x = var_3992_cast_fp16, y = var_3993_to_fp16)[name = tensor("aw_277_cast_fp16")]; + tensor var_3996_equation_0 = const()[name = tensor("op_3996_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3996_cast_fp16 = einsum(equation = var_3996_equation_0, values = (var_3838_cast_fp16, var_3755_cast_fp16))[name = tensor("op_3996_cast_fp16")]; + tensor var_3997_to_fp16 = const()[name = tensor("op_3997_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_279_cast_fp16 = mul(x = var_3996_cast_fp16, y = var_3997_to_fp16)[name = tensor("aw_279_cast_fp16")]; + tensor var_3999_cast_fp16 = softmax(axis = var_2624, x = aw_241_cast_fp16)[name = tensor("op_3999_cast_fp16")]; + tensor var_4000_cast_fp16 = softmax(axis = var_2624, x = aw_243_cast_fp16)[name = tensor("op_4000_cast_fp16")]; + tensor var_4001_cast_fp16 = softmax(axis = var_2624, x = aw_245_cast_fp16)[name = tensor("op_4001_cast_fp16")]; + tensor var_4002_cast_fp16 = softmax(axis = var_2624, x = aw_247_cast_fp16)[name = tensor("op_4002_cast_fp16")]; + tensor var_4003_cast_fp16 = softmax(axis = var_2624, x = aw_249_cast_fp16)[name = tensor("op_4003_cast_fp16")]; + tensor var_4004_cast_fp16 = softmax(axis = var_2624, x = aw_251_cast_fp16)[name = tensor("op_4004_cast_fp16")]; + tensor var_4005_cast_fp16 = softmax(axis = var_2624, x = aw_253_cast_fp16)[name = tensor("op_4005_cast_fp16")]; + tensor var_4006_cast_fp16 = softmax(axis = var_2624, x = aw_255_cast_fp16)[name = tensor("op_4006_cast_fp16")]; + tensor var_4007_cast_fp16 = softmax(axis = var_2624, x = aw_257_cast_fp16)[name = tensor("op_4007_cast_fp16")]; + tensor var_4008_cast_fp16 = softmax(axis = var_2624, x = aw_259_cast_fp16)[name = tensor("op_4008_cast_fp16")]; + tensor var_4009_cast_fp16 = softmax(axis = var_2624, x = aw_261_cast_fp16)[name = tensor("op_4009_cast_fp16")]; + tensor var_4010_cast_fp16 = softmax(axis = var_2624, x = aw_263_cast_fp16)[name = tensor("op_4010_cast_fp16")]; + tensor var_4011_cast_fp16 = softmax(axis = var_2624, x = aw_265_cast_fp16)[name = tensor("op_4011_cast_fp16")]; + tensor var_4012_cast_fp16 = softmax(axis = var_2624, x = aw_267_cast_fp16)[name = tensor("op_4012_cast_fp16")]; + tensor var_4013_cast_fp16 = softmax(axis = var_2624, x = aw_269_cast_fp16)[name = tensor("op_4013_cast_fp16")]; + tensor var_4014_cast_fp16 = softmax(axis = var_2624, x = aw_271_cast_fp16)[name = tensor("op_4014_cast_fp16")]; + tensor var_4015_cast_fp16 = softmax(axis = var_2624, x = aw_273_cast_fp16)[name = tensor("op_4015_cast_fp16")]; + tensor var_4016_cast_fp16 = softmax(axis = var_2624, x = aw_275_cast_fp16)[name = tensor("op_4016_cast_fp16")]; + tensor var_4017_cast_fp16 = softmax(axis = var_2624, x = aw_277_cast_fp16)[name = tensor("op_4017_cast_fp16")]; + tensor var_4018_cast_fp16 = softmax(axis = var_2624, x = aw_279_cast_fp16)[name = tensor("op_4018_cast_fp16")]; + tensor var_4020_equation_0 = const()[name = tensor("op_4020_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4020_cast_fp16 = einsum(equation = var_4020_equation_0, values = (var_3840_cast_fp16, var_3999_cast_fp16))[name = tensor("op_4020_cast_fp16")]; + tensor var_4022_equation_0 = const()[name = tensor("op_4022_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4022_cast_fp16 = einsum(equation = var_4022_equation_0, values = (var_3844_cast_fp16, var_4000_cast_fp16))[name = tensor("op_4022_cast_fp16")]; + tensor var_4024_equation_0 = const()[name = tensor("op_4024_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4024_cast_fp16 = einsum(equation = var_4024_equation_0, values = (var_3848_cast_fp16, var_4001_cast_fp16))[name = tensor("op_4024_cast_fp16")]; + tensor var_4026_equation_0 = const()[name = tensor("op_4026_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4026_cast_fp16 = einsum(equation = var_4026_equation_0, values = (var_3852_cast_fp16, var_4002_cast_fp16))[name = tensor("op_4026_cast_fp16")]; + tensor var_4028_equation_0 = const()[name = tensor("op_4028_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4028_cast_fp16 = einsum(equation = var_4028_equation_0, values = (var_3856_cast_fp16, var_4003_cast_fp16))[name = tensor("op_4028_cast_fp16")]; + tensor var_4030_equation_0 = const()[name = tensor("op_4030_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4030_cast_fp16 = einsum(equation = var_4030_equation_0, values = (var_3860_cast_fp16, var_4004_cast_fp16))[name = tensor("op_4030_cast_fp16")]; + tensor var_4032_equation_0 = const()[name = tensor("op_4032_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4032_cast_fp16 = einsum(equation = var_4032_equation_0, values = (var_3864_cast_fp16, var_4005_cast_fp16))[name = tensor("op_4032_cast_fp16")]; + tensor var_4034_equation_0 = const()[name = tensor("op_4034_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4034_cast_fp16 = einsum(equation = var_4034_equation_0, values = (var_3868_cast_fp16, var_4006_cast_fp16))[name = tensor("op_4034_cast_fp16")]; + tensor var_4036_equation_0 = const()[name = tensor("op_4036_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4036_cast_fp16 = einsum(equation = var_4036_equation_0, values = (var_3872_cast_fp16, var_4007_cast_fp16))[name = tensor("op_4036_cast_fp16")]; + tensor var_4038_equation_0 = const()[name = tensor("op_4038_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4038_cast_fp16 = einsum(equation = var_4038_equation_0, values = (var_3876_cast_fp16, var_4008_cast_fp16))[name = tensor("op_4038_cast_fp16")]; + tensor var_4040_equation_0 = const()[name = tensor("op_4040_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4040_cast_fp16 = einsum(equation = var_4040_equation_0, values = (var_3880_cast_fp16, var_4009_cast_fp16))[name = tensor("op_4040_cast_fp16")]; + tensor var_4042_equation_0 = const()[name = tensor("op_4042_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4042_cast_fp16 = einsum(equation = var_4042_equation_0, values = (var_3884_cast_fp16, var_4010_cast_fp16))[name = tensor("op_4042_cast_fp16")]; + tensor var_4044_equation_0 = const()[name = tensor("op_4044_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4044_cast_fp16 = einsum(equation = var_4044_equation_0, values = (var_3888_cast_fp16, var_4011_cast_fp16))[name = tensor("op_4044_cast_fp16")]; + tensor var_4046_equation_0 = const()[name = tensor("op_4046_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4046_cast_fp16 = einsum(equation = var_4046_equation_0, values = (var_3892_cast_fp16, var_4012_cast_fp16))[name = tensor("op_4046_cast_fp16")]; + tensor var_4048_equation_0 = const()[name = tensor("op_4048_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4048_cast_fp16 = einsum(equation = var_4048_equation_0, values = (var_3896_cast_fp16, var_4013_cast_fp16))[name = tensor("op_4048_cast_fp16")]; + tensor var_4050_equation_0 = const()[name = tensor("op_4050_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4050_cast_fp16 = einsum(equation = var_4050_equation_0, values = (var_3900_cast_fp16, var_4014_cast_fp16))[name = tensor("op_4050_cast_fp16")]; + tensor var_4052_equation_0 = const()[name = tensor("op_4052_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4052_cast_fp16 = einsum(equation = var_4052_equation_0, values = (var_3904_cast_fp16, var_4015_cast_fp16))[name = tensor("op_4052_cast_fp16")]; + tensor var_4054_equation_0 = const()[name = tensor("op_4054_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4054_cast_fp16 = einsum(equation = var_4054_equation_0, values = (var_3908_cast_fp16, var_4016_cast_fp16))[name = tensor("op_4054_cast_fp16")]; + tensor var_4056_equation_0 = const()[name = tensor("op_4056_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4056_cast_fp16 = einsum(equation = var_4056_equation_0, values = (var_3912_cast_fp16, var_4017_cast_fp16))[name = tensor("op_4056_cast_fp16")]; + tensor var_4058_equation_0 = const()[name = tensor("op_4058_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4058_cast_fp16 = einsum(equation = var_4058_equation_0, values = (var_3916_cast_fp16, var_4018_cast_fp16))[name = tensor("op_4058_cast_fp16")]; + tensor input_139_interleave_0 = const()[name = tensor("input_139_interleave_0"), val = tensor(false)]; + tensor input_139_cast_fp16 = concat(axis = var_2624, interleave = input_139_interleave_0, values = (var_4020_cast_fp16, var_4022_cast_fp16, var_4024_cast_fp16, var_4026_cast_fp16, var_4028_cast_fp16, var_4030_cast_fp16, var_4032_cast_fp16, var_4034_cast_fp16, var_4036_cast_fp16, var_4038_cast_fp16, var_4040_cast_fp16, var_4042_cast_fp16, var_4044_cast_fp16, var_4046_cast_fp16, var_4048_cast_fp16, var_4050_cast_fp16, var_4052_cast_fp16, var_4054_cast_fp16, var_4056_cast_fp16, var_4058_cast_fp16))[name = tensor("input_139_cast_fp16")]; + tensor var_4064 = const()[name = tensor("op_4064"), val = tensor([1, 1])]; + tensor var_4066 = const()[name = tensor("op_4066"), val = tensor([1, 1])]; + tensor var_4068_pad_type_0 = const()[name = tensor("op_4068_pad_type_0"), val = tensor("custom")]; + tensor var_4068_pad_0 = const()[name = tensor("op_4068_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_1_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_1_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(277514688)))]; + tensor down_blocks_2_attentions_0_transformer_blocks_1_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_1_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(280791552)))]; + tensor var_4068_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_1_attn1_to_out_0_bias_to_fp16, dilations = var_4066, groups = var_2624, pad = var_4068_pad_0, pad_type = var_4068_pad_type_0, strides = var_4064, weight = down_blocks_2_attentions_0_transformer_blocks_1_attn1_to_out_0_weight_to_fp16, x = input_139_cast_fp16)[name = tensor("op_4068_cast_fp16")]; + tensor inputs_33_cast_fp16 = add(x = var_4068_cast_fp16, y = inputs_31_cast_fp16)[name = tensor("inputs_33_cast_fp16")]; + tensor hidden_states_73_axes_0 = const()[name = tensor("hidden_states_73_axes_0"), val = tensor([1])]; + tensor hidden_states_73_gamma_0_to_fp16 = const()[name = tensor("hidden_states_73_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(280794176)))]; + tensor hidden_states_73_beta_0_to_fp16 = const()[name = tensor("hidden_states_73_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(280796800)))]; + tensor var_4078_to_fp16 = const()[name = tensor("op_4078_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_73_cast_fp16 = layer_norm(axes = hidden_states_73_axes_0, beta = hidden_states_73_beta_0_to_fp16, epsilon = var_4078_to_fp16, gamma = hidden_states_73_gamma_0_to_fp16, x = inputs_33_cast_fp16)[name = tensor("hidden_states_73_cast_fp16")]; + tensor var_4093 = const()[name = tensor("op_4093"), val = tensor([1, 1])]; + tensor var_4095 = const()[name = tensor("op_4095"), val = tensor([1, 1])]; + tensor q_23_pad_type_0 = const()[name = tensor("q_23_pad_type_0"), val = tensor("custom")]; + tensor q_23_pad_0 = const()[name = tensor("q_23_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_1_attn2_to_q_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_1_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(280799424)))]; + tensor q_23_cast_fp16 = conv(dilations = var_4095, groups = var_2624, pad = q_23_pad_0, pad_type = q_23_pad_type_0, strides = var_4093, weight = down_blocks_2_attentions_0_transformer_blocks_1_attn2_to_q_weight_to_fp16, x = hidden_states_73_cast_fp16)[name = tensor("q_23_cast_fp16")]; + tensor var_4099 = const()[name = tensor("op_4099"), val = tensor([1, 1])]; + tensor var_4101 = const()[name = tensor("op_4101"), val = tensor([1, 1])]; + tensor k_45_pad_type_0 = const()[name = tensor("k_45_pad_type_0"), val = tensor("custom")]; + tensor k_45_pad_0 = const()[name = tensor("k_45_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_1_attn2_to_k_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_1_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(284076288)))]; + tensor k_45_cast_fp16 = conv(dilations = var_4101, groups = var_2624, pad = k_45_pad_0, pad_type = k_45_pad_type_0, strides = var_4099, weight = down_blocks_2_attentions_0_transformer_blocks_1_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_45_cast_fp16")]; + tensor var_4105 = const()[name = tensor("op_4105"), val = tensor([1, 1])]; + tensor var_4107 = const()[name = tensor("op_4107"), val = tensor([1, 1])]; + tensor v_23_pad_type_0 = const()[name = tensor("v_23_pad_type_0"), val = tensor("custom")]; + tensor v_23_pad_0 = const()[name = tensor("v_23_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_1_attn2_to_v_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_1_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(289319232)))]; + tensor v_23_cast_fp16 = conv(dilations = var_4107, groups = var_2624, pad = v_23_pad_0, pad_type = v_23_pad_type_0, strides = var_4105, weight = down_blocks_2_attentions_0_transformer_blocks_1_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_23_cast_fp16")]; + tensor var_4111_begin_0 = const()[name = tensor("op_4111_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4111_end_0 = const()[name = tensor("op_4111_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_4111_end_mask_0 = const()[name = tensor("op_4111_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4111_cast_fp16 = slice_by_index(begin = var_4111_begin_0, end = var_4111_end_0, end_mask = var_4111_end_mask_0, x = q_23_cast_fp16)[name = tensor("op_4111_cast_fp16")]; + tensor var_4115_begin_0 = const()[name = tensor("op_4115_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_4115_end_0 = const()[name = tensor("op_4115_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_4115_end_mask_0 = const()[name = tensor("op_4115_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4115_cast_fp16 = slice_by_index(begin = var_4115_begin_0, end = var_4115_end_0, end_mask = var_4115_end_mask_0, x = q_23_cast_fp16)[name = tensor("op_4115_cast_fp16")]; + tensor var_4119_begin_0 = const()[name = tensor("op_4119_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_4119_end_0 = const()[name = tensor("op_4119_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_4119_end_mask_0 = const()[name = tensor("op_4119_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4119_cast_fp16 = slice_by_index(begin = var_4119_begin_0, end = var_4119_end_0, end_mask = var_4119_end_mask_0, x = q_23_cast_fp16)[name = tensor("op_4119_cast_fp16")]; + tensor var_4123_begin_0 = const()[name = tensor("op_4123_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_4123_end_0 = const()[name = tensor("op_4123_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_4123_end_mask_0 = const()[name = tensor("op_4123_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4123_cast_fp16 = slice_by_index(begin = var_4123_begin_0, end = var_4123_end_0, end_mask = var_4123_end_mask_0, x = q_23_cast_fp16)[name = tensor("op_4123_cast_fp16")]; + tensor var_4127_begin_0 = const()[name = tensor("op_4127_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_4127_end_0 = const()[name = tensor("op_4127_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_4127_end_mask_0 = const()[name = tensor("op_4127_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4127_cast_fp16 = slice_by_index(begin = var_4127_begin_0, end = var_4127_end_0, end_mask = var_4127_end_mask_0, x = q_23_cast_fp16)[name = tensor("op_4127_cast_fp16")]; + tensor var_4131_begin_0 = const()[name = tensor("op_4131_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_4131_end_0 = const()[name = tensor("op_4131_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_4131_end_mask_0 = const()[name = tensor("op_4131_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4131_cast_fp16 = slice_by_index(begin = var_4131_begin_0, end = var_4131_end_0, end_mask = var_4131_end_mask_0, x = q_23_cast_fp16)[name = tensor("op_4131_cast_fp16")]; + tensor var_4135_begin_0 = const()[name = tensor("op_4135_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_4135_end_0 = const()[name = tensor("op_4135_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_4135_end_mask_0 = const()[name = tensor("op_4135_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4135_cast_fp16 = slice_by_index(begin = var_4135_begin_0, end = var_4135_end_0, end_mask = var_4135_end_mask_0, x = q_23_cast_fp16)[name = tensor("op_4135_cast_fp16")]; + tensor var_4139_begin_0 = const()[name = tensor("op_4139_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_4139_end_0 = const()[name = tensor("op_4139_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_4139_end_mask_0 = const()[name = tensor("op_4139_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4139_cast_fp16 = slice_by_index(begin = var_4139_begin_0, end = var_4139_end_0, end_mask = var_4139_end_mask_0, x = q_23_cast_fp16)[name = tensor("op_4139_cast_fp16")]; + tensor var_4143_begin_0 = const()[name = tensor("op_4143_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_4143_end_0 = const()[name = tensor("op_4143_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_4143_end_mask_0 = const()[name = tensor("op_4143_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4143_cast_fp16 = slice_by_index(begin = var_4143_begin_0, end = var_4143_end_0, end_mask = var_4143_end_mask_0, x = q_23_cast_fp16)[name = tensor("op_4143_cast_fp16")]; + tensor var_4147_begin_0 = const()[name = tensor("op_4147_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_4147_end_0 = const()[name = tensor("op_4147_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_4147_end_mask_0 = const()[name = tensor("op_4147_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4147_cast_fp16 = slice_by_index(begin = var_4147_begin_0, end = var_4147_end_0, end_mask = var_4147_end_mask_0, x = q_23_cast_fp16)[name = tensor("op_4147_cast_fp16")]; + tensor var_4151_begin_0 = const()[name = tensor("op_4151_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_4151_end_0 = const()[name = tensor("op_4151_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_4151_end_mask_0 = const()[name = tensor("op_4151_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4151_cast_fp16 = slice_by_index(begin = var_4151_begin_0, end = var_4151_end_0, end_mask = var_4151_end_mask_0, x = q_23_cast_fp16)[name = tensor("op_4151_cast_fp16")]; + tensor var_4155_begin_0 = const()[name = tensor("op_4155_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_4155_end_0 = const()[name = tensor("op_4155_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_4155_end_mask_0 = const()[name = tensor("op_4155_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4155_cast_fp16 = slice_by_index(begin = var_4155_begin_0, end = var_4155_end_0, end_mask = var_4155_end_mask_0, x = q_23_cast_fp16)[name = tensor("op_4155_cast_fp16")]; + tensor var_4159_begin_0 = const()[name = tensor("op_4159_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_4159_end_0 = const()[name = tensor("op_4159_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_4159_end_mask_0 = const()[name = tensor("op_4159_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4159_cast_fp16 = slice_by_index(begin = var_4159_begin_0, end = var_4159_end_0, end_mask = var_4159_end_mask_0, x = q_23_cast_fp16)[name = tensor("op_4159_cast_fp16")]; + tensor var_4163_begin_0 = const()[name = tensor("op_4163_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_4163_end_0 = const()[name = tensor("op_4163_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_4163_end_mask_0 = const()[name = tensor("op_4163_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4163_cast_fp16 = slice_by_index(begin = var_4163_begin_0, end = var_4163_end_0, end_mask = var_4163_end_mask_0, x = q_23_cast_fp16)[name = tensor("op_4163_cast_fp16")]; + tensor var_4167_begin_0 = const()[name = tensor("op_4167_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_4167_end_0 = const()[name = tensor("op_4167_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_4167_end_mask_0 = const()[name = tensor("op_4167_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4167_cast_fp16 = slice_by_index(begin = var_4167_begin_0, end = var_4167_end_0, end_mask = var_4167_end_mask_0, x = q_23_cast_fp16)[name = tensor("op_4167_cast_fp16")]; + tensor var_4171_begin_0 = const()[name = tensor("op_4171_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_4171_end_0 = const()[name = tensor("op_4171_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_4171_end_mask_0 = const()[name = tensor("op_4171_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4171_cast_fp16 = slice_by_index(begin = var_4171_begin_0, end = var_4171_end_0, end_mask = var_4171_end_mask_0, x = q_23_cast_fp16)[name = tensor("op_4171_cast_fp16")]; + tensor var_4175_begin_0 = const()[name = tensor("op_4175_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_4175_end_0 = const()[name = tensor("op_4175_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_4175_end_mask_0 = const()[name = tensor("op_4175_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4175_cast_fp16 = slice_by_index(begin = var_4175_begin_0, end = var_4175_end_0, end_mask = var_4175_end_mask_0, x = q_23_cast_fp16)[name = tensor("op_4175_cast_fp16")]; + tensor var_4179_begin_0 = const()[name = tensor("op_4179_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_4179_end_0 = const()[name = tensor("op_4179_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_4179_end_mask_0 = const()[name = tensor("op_4179_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4179_cast_fp16 = slice_by_index(begin = var_4179_begin_0, end = var_4179_end_0, end_mask = var_4179_end_mask_0, x = q_23_cast_fp16)[name = tensor("op_4179_cast_fp16")]; + tensor var_4183_begin_0 = const()[name = tensor("op_4183_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_4183_end_0 = const()[name = tensor("op_4183_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_4183_end_mask_0 = const()[name = tensor("op_4183_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4183_cast_fp16 = slice_by_index(begin = var_4183_begin_0, end = var_4183_end_0, end_mask = var_4183_end_mask_0, x = q_23_cast_fp16)[name = tensor("op_4183_cast_fp16")]; + tensor var_4187_begin_0 = const()[name = tensor("op_4187_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_4187_end_0 = const()[name = tensor("op_4187_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_4187_end_mask_0 = const()[name = tensor("op_4187_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4187_cast_fp16 = slice_by_index(begin = var_4187_begin_0, end = var_4187_end_0, end_mask = var_4187_end_mask_0, x = q_23_cast_fp16)[name = tensor("op_4187_cast_fp16")]; + tensor k_47_perm_0 = const()[name = tensor("k_47_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_4194_begin_0 = const()[name = tensor("op_4194_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4194_end_0 = const()[name = tensor("op_4194_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_4194_end_mask_0 = const()[name = tensor("op_4194_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_47_cast_fp16 = transpose(perm = k_47_perm_0, x = k_45_cast_fp16)[name = tensor("transpose_128")]; + tensor var_4194_cast_fp16 = slice_by_index(begin = var_4194_begin_0, end = var_4194_end_0, end_mask = var_4194_end_mask_0, x = k_47_cast_fp16)[name = tensor("op_4194_cast_fp16")]; + tensor var_4198_begin_0 = const()[name = tensor("op_4198_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_4198_end_0 = const()[name = tensor("op_4198_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_4198_end_mask_0 = const()[name = tensor("op_4198_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4198_cast_fp16 = slice_by_index(begin = var_4198_begin_0, end = var_4198_end_0, end_mask = var_4198_end_mask_0, x = k_47_cast_fp16)[name = tensor("op_4198_cast_fp16")]; + tensor var_4202_begin_0 = const()[name = tensor("op_4202_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_4202_end_0 = const()[name = tensor("op_4202_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_4202_end_mask_0 = const()[name = tensor("op_4202_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4202_cast_fp16 = slice_by_index(begin = var_4202_begin_0, end = var_4202_end_0, end_mask = var_4202_end_mask_0, x = k_47_cast_fp16)[name = tensor("op_4202_cast_fp16")]; + tensor var_4206_begin_0 = const()[name = tensor("op_4206_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_4206_end_0 = const()[name = tensor("op_4206_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_4206_end_mask_0 = const()[name = tensor("op_4206_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4206_cast_fp16 = slice_by_index(begin = var_4206_begin_0, end = var_4206_end_0, end_mask = var_4206_end_mask_0, x = k_47_cast_fp16)[name = tensor("op_4206_cast_fp16")]; + tensor var_4210_begin_0 = const()[name = tensor("op_4210_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_4210_end_0 = const()[name = tensor("op_4210_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_4210_end_mask_0 = const()[name = tensor("op_4210_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4210_cast_fp16 = slice_by_index(begin = var_4210_begin_0, end = var_4210_end_0, end_mask = var_4210_end_mask_0, x = k_47_cast_fp16)[name = tensor("op_4210_cast_fp16")]; + tensor var_4214_begin_0 = const()[name = tensor("op_4214_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_4214_end_0 = const()[name = tensor("op_4214_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_4214_end_mask_0 = const()[name = tensor("op_4214_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4214_cast_fp16 = slice_by_index(begin = var_4214_begin_0, end = var_4214_end_0, end_mask = var_4214_end_mask_0, x = k_47_cast_fp16)[name = tensor("op_4214_cast_fp16")]; + tensor var_4218_begin_0 = const()[name = tensor("op_4218_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_4218_end_0 = const()[name = tensor("op_4218_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_4218_end_mask_0 = const()[name = tensor("op_4218_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4218_cast_fp16 = slice_by_index(begin = var_4218_begin_0, end = var_4218_end_0, end_mask = var_4218_end_mask_0, x = k_47_cast_fp16)[name = tensor("op_4218_cast_fp16")]; + tensor var_4222_begin_0 = const()[name = tensor("op_4222_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_4222_end_0 = const()[name = tensor("op_4222_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_4222_end_mask_0 = const()[name = tensor("op_4222_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4222_cast_fp16 = slice_by_index(begin = var_4222_begin_0, end = var_4222_end_0, end_mask = var_4222_end_mask_0, x = k_47_cast_fp16)[name = tensor("op_4222_cast_fp16")]; + tensor var_4226_begin_0 = const()[name = tensor("op_4226_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_4226_end_0 = const()[name = tensor("op_4226_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_4226_end_mask_0 = const()[name = tensor("op_4226_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4226_cast_fp16 = slice_by_index(begin = var_4226_begin_0, end = var_4226_end_0, end_mask = var_4226_end_mask_0, x = k_47_cast_fp16)[name = tensor("op_4226_cast_fp16")]; + tensor var_4230_begin_0 = const()[name = tensor("op_4230_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_4230_end_0 = const()[name = tensor("op_4230_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_4230_end_mask_0 = const()[name = tensor("op_4230_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4230_cast_fp16 = slice_by_index(begin = var_4230_begin_0, end = var_4230_end_0, end_mask = var_4230_end_mask_0, x = k_47_cast_fp16)[name = tensor("op_4230_cast_fp16")]; + tensor var_4234_begin_0 = const()[name = tensor("op_4234_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_4234_end_0 = const()[name = tensor("op_4234_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_4234_end_mask_0 = const()[name = tensor("op_4234_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4234_cast_fp16 = slice_by_index(begin = var_4234_begin_0, end = var_4234_end_0, end_mask = var_4234_end_mask_0, x = k_47_cast_fp16)[name = tensor("op_4234_cast_fp16")]; + tensor var_4238_begin_0 = const()[name = tensor("op_4238_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_4238_end_0 = const()[name = tensor("op_4238_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_4238_end_mask_0 = const()[name = tensor("op_4238_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4238_cast_fp16 = slice_by_index(begin = var_4238_begin_0, end = var_4238_end_0, end_mask = var_4238_end_mask_0, x = k_47_cast_fp16)[name = tensor("op_4238_cast_fp16")]; + tensor var_4242_begin_0 = const()[name = tensor("op_4242_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_4242_end_0 = const()[name = tensor("op_4242_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_4242_end_mask_0 = const()[name = tensor("op_4242_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4242_cast_fp16 = slice_by_index(begin = var_4242_begin_0, end = var_4242_end_0, end_mask = var_4242_end_mask_0, x = k_47_cast_fp16)[name = tensor("op_4242_cast_fp16")]; + tensor var_4246_begin_0 = const()[name = tensor("op_4246_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_4246_end_0 = const()[name = tensor("op_4246_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_4246_end_mask_0 = const()[name = tensor("op_4246_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4246_cast_fp16 = slice_by_index(begin = var_4246_begin_0, end = var_4246_end_0, end_mask = var_4246_end_mask_0, x = k_47_cast_fp16)[name = tensor("op_4246_cast_fp16")]; + tensor var_4250_begin_0 = const()[name = tensor("op_4250_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_4250_end_0 = const()[name = tensor("op_4250_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_4250_end_mask_0 = const()[name = tensor("op_4250_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4250_cast_fp16 = slice_by_index(begin = var_4250_begin_0, end = var_4250_end_0, end_mask = var_4250_end_mask_0, x = k_47_cast_fp16)[name = tensor("op_4250_cast_fp16")]; + tensor var_4254_begin_0 = const()[name = tensor("op_4254_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_4254_end_0 = const()[name = tensor("op_4254_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_4254_end_mask_0 = const()[name = tensor("op_4254_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4254_cast_fp16 = slice_by_index(begin = var_4254_begin_0, end = var_4254_end_0, end_mask = var_4254_end_mask_0, x = k_47_cast_fp16)[name = tensor("op_4254_cast_fp16")]; + tensor var_4258_begin_0 = const()[name = tensor("op_4258_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_4258_end_0 = const()[name = tensor("op_4258_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_4258_end_mask_0 = const()[name = tensor("op_4258_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4258_cast_fp16 = slice_by_index(begin = var_4258_begin_0, end = var_4258_end_0, end_mask = var_4258_end_mask_0, x = k_47_cast_fp16)[name = tensor("op_4258_cast_fp16")]; + tensor var_4262_begin_0 = const()[name = tensor("op_4262_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_4262_end_0 = const()[name = tensor("op_4262_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_4262_end_mask_0 = const()[name = tensor("op_4262_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4262_cast_fp16 = slice_by_index(begin = var_4262_begin_0, end = var_4262_end_0, end_mask = var_4262_end_mask_0, x = k_47_cast_fp16)[name = tensor("op_4262_cast_fp16")]; + tensor var_4266_begin_0 = const()[name = tensor("op_4266_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_4266_end_0 = const()[name = tensor("op_4266_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_4266_end_mask_0 = const()[name = tensor("op_4266_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4266_cast_fp16 = slice_by_index(begin = var_4266_begin_0, end = var_4266_end_0, end_mask = var_4266_end_mask_0, x = k_47_cast_fp16)[name = tensor("op_4266_cast_fp16")]; + tensor var_4270_begin_0 = const()[name = tensor("op_4270_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_4270_end_0 = const()[name = tensor("op_4270_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_4270_end_mask_0 = const()[name = tensor("op_4270_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4270_cast_fp16 = slice_by_index(begin = var_4270_begin_0, end = var_4270_end_0, end_mask = var_4270_end_mask_0, x = k_47_cast_fp16)[name = tensor("op_4270_cast_fp16")]; + tensor var_4272_begin_0 = const()[name = tensor("op_4272_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4272_end_0 = const()[name = tensor("op_4272_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_4272_end_mask_0 = const()[name = tensor("op_4272_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4272_cast_fp16 = slice_by_index(begin = var_4272_begin_0, end = var_4272_end_0, end_mask = var_4272_end_mask_0, x = v_23_cast_fp16)[name = tensor("op_4272_cast_fp16")]; + tensor var_4276_begin_0 = const()[name = tensor("op_4276_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_4276_end_0 = const()[name = tensor("op_4276_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_4276_end_mask_0 = const()[name = tensor("op_4276_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4276_cast_fp16 = slice_by_index(begin = var_4276_begin_0, end = var_4276_end_0, end_mask = var_4276_end_mask_0, x = v_23_cast_fp16)[name = tensor("op_4276_cast_fp16")]; + tensor var_4280_begin_0 = const()[name = tensor("op_4280_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_4280_end_0 = const()[name = tensor("op_4280_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_4280_end_mask_0 = const()[name = tensor("op_4280_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4280_cast_fp16 = slice_by_index(begin = var_4280_begin_0, end = var_4280_end_0, end_mask = var_4280_end_mask_0, x = v_23_cast_fp16)[name = tensor("op_4280_cast_fp16")]; + tensor var_4284_begin_0 = const()[name = tensor("op_4284_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_4284_end_0 = const()[name = tensor("op_4284_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_4284_end_mask_0 = const()[name = tensor("op_4284_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4284_cast_fp16 = slice_by_index(begin = var_4284_begin_0, end = var_4284_end_0, end_mask = var_4284_end_mask_0, x = v_23_cast_fp16)[name = tensor("op_4284_cast_fp16")]; + tensor var_4288_begin_0 = const()[name = tensor("op_4288_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_4288_end_0 = const()[name = tensor("op_4288_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_4288_end_mask_0 = const()[name = tensor("op_4288_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4288_cast_fp16 = slice_by_index(begin = var_4288_begin_0, end = var_4288_end_0, end_mask = var_4288_end_mask_0, x = v_23_cast_fp16)[name = tensor("op_4288_cast_fp16")]; + tensor var_4292_begin_0 = const()[name = tensor("op_4292_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_4292_end_0 = const()[name = tensor("op_4292_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_4292_end_mask_0 = const()[name = tensor("op_4292_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4292_cast_fp16 = slice_by_index(begin = var_4292_begin_0, end = var_4292_end_0, end_mask = var_4292_end_mask_0, x = v_23_cast_fp16)[name = tensor("op_4292_cast_fp16")]; + tensor var_4296_begin_0 = const()[name = tensor("op_4296_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_4296_end_0 = const()[name = tensor("op_4296_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_4296_end_mask_0 = const()[name = tensor("op_4296_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4296_cast_fp16 = slice_by_index(begin = var_4296_begin_0, end = var_4296_end_0, end_mask = var_4296_end_mask_0, x = v_23_cast_fp16)[name = tensor("op_4296_cast_fp16")]; + tensor var_4300_begin_0 = const()[name = tensor("op_4300_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_4300_end_0 = const()[name = tensor("op_4300_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_4300_end_mask_0 = const()[name = tensor("op_4300_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4300_cast_fp16 = slice_by_index(begin = var_4300_begin_0, end = var_4300_end_0, end_mask = var_4300_end_mask_0, x = v_23_cast_fp16)[name = tensor("op_4300_cast_fp16")]; + tensor var_4304_begin_0 = const()[name = tensor("op_4304_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_4304_end_0 = const()[name = tensor("op_4304_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_4304_end_mask_0 = const()[name = tensor("op_4304_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4304_cast_fp16 = slice_by_index(begin = var_4304_begin_0, end = var_4304_end_0, end_mask = var_4304_end_mask_0, x = v_23_cast_fp16)[name = tensor("op_4304_cast_fp16")]; + tensor var_4308_begin_0 = const()[name = tensor("op_4308_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_4308_end_0 = const()[name = tensor("op_4308_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_4308_end_mask_0 = const()[name = tensor("op_4308_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4308_cast_fp16 = slice_by_index(begin = var_4308_begin_0, end = var_4308_end_0, end_mask = var_4308_end_mask_0, x = v_23_cast_fp16)[name = tensor("op_4308_cast_fp16")]; + tensor var_4312_begin_0 = const()[name = tensor("op_4312_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_4312_end_0 = const()[name = tensor("op_4312_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_4312_end_mask_0 = const()[name = tensor("op_4312_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4312_cast_fp16 = slice_by_index(begin = var_4312_begin_0, end = var_4312_end_0, end_mask = var_4312_end_mask_0, x = v_23_cast_fp16)[name = tensor("op_4312_cast_fp16")]; + tensor var_4316_begin_0 = const()[name = tensor("op_4316_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_4316_end_0 = const()[name = tensor("op_4316_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_4316_end_mask_0 = const()[name = tensor("op_4316_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4316_cast_fp16 = slice_by_index(begin = var_4316_begin_0, end = var_4316_end_0, end_mask = var_4316_end_mask_0, x = v_23_cast_fp16)[name = tensor("op_4316_cast_fp16")]; + tensor var_4320_begin_0 = const()[name = tensor("op_4320_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_4320_end_0 = const()[name = tensor("op_4320_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_4320_end_mask_0 = const()[name = tensor("op_4320_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4320_cast_fp16 = slice_by_index(begin = var_4320_begin_0, end = var_4320_end_0, end_mask = var_4320_end_mask_0, x = v_23_cast_fp16)[name = tensor("op_4320_cast_fp16")]; + tensor var_4324_begin_0 = const()[name = tensor("op_4324_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_4324_end_0 = const()[name = tensor("op_4324_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_4324_end_mask_0 = const()[name = tensor("op_4324_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4324_cast_fp16 = slice_by_index(begin = var_4324_begin_0, end = var_4324_end_0, end_mask = var_4324_end_mask_0, x = v_23_cast_fp16)[name = tensor("op_4324_cast_fp16")]; + tensor var_4328_begin_0 = const()[name = tensor("op_4328_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_4328_end_0 = const()[name = tensor("op_4328_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_4328_end_mask_0 = const()[name = tensor("op_4328_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4328_cast_fp16 = slice_by_index(begin = var_4328_begin_0, end = var_4328_end_0, end_mask = var_4328_end_mask_0, x = v_23_cast_fp16)[name = tensor("op_4328_cast_fp16")]; + tensor var_4332_begin_0 = const()[name = tensor("op_4332_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_4332_end_0 = const()[name = tensor("op_4332_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_4332_end_mask_0 = const()[name = tensor("op_4332_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4332_cast_fp16 = slice_by_index(begin = var_4332_begin_0, end = var_4332_end_0, end_mask = var_4332_end_mask_0, x = v_23_cast_fp16)[name = tensor("op_4332_cast_fp16")]; + tensor var_4336_begin_0 = const()[name = tensor("op_4336_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_4336_end_0 = const()[name = tensor("op_4336_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_4336_end_mask_0 = const()[name = tensor("op_4336_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4336_cast_fp16 = slice_by_index(begin = var_4336_begin_0, end = var_4336_end_0, end_mask = var_4336_end_mask_0, x = v_23_cast_fp16)[name = tensor("op_4336_cast_fp16")]; + tensor var_4340_begin_0 = const()[name = tensor("op_4340_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_4340_end_0 = const()[name = tensor("op_4340_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_4340_end_mask_0 = const()[name = tensor("op_4340_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4340_cast_fp16 = slice_by_index(begin = var_4340_begin_0, end = var_4340_end_0, end_mask = var_4340_end_mask_0, x = v_23_cast_fp16)[name = tensor("op_4340_cast_fp16")]; + tensor var_4344_begin_0 = const()[name = tensor("op_4344_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_4344_end_0 = const()[name = tensor("op_4344_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_4344_end_mask_0 = const()[name = tensor("op_4344_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4344_cast_fp16 = slice_by_index(begin = var_4344_begin_0, end = var_4344_end_0, end_mask = var_4344_end_mask_0, x = v_23_cast_fp16)[name = tensor("op_4344_cast_fp16")]; + tensor var_4348_begin_0 = const()[name = tensor("op_4348_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_4348_end_0 = const()[name = tensor("op_4348_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_4348_end_mask_0 = const()[name = tensor("op_4348_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4348_cast_fp16 = slice_by_index(begin = var_4348_begin_0, end = var_4348_end_0, end_mask = var_4348_end_mask_0, x = v_23_cast_fp16)[name = tensor("op_4348_cast_fp16")]; + tensor var_4352_equation_0 = const()[name = tensor("op_4352_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4352_cast_fp16 = einsum(equation = var_4352_equation_0, values = (var_4194_cast_fp16, var_4111_cast_fp16))[name = tensor("op_4352_cast_fp16")]; + tensor var_4353_to_fp16 = const()[name = tensor("op_4353_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_281_cast_fp16 = mul(x = var_4352_cast_fp16, y = var_4353_to_fp16)[name = tensor("aw_281_cast_fp16")]; + tensor var_4356_equation_0 = const()[name = tensor("op_4356_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4356_cast_fp16 = einsum(equation = var_4356_equation_0, values = (var_4198_cast_fp16, var_4115_cast_fp16))[name = tensor("op_4356_cast_fp16")]; + tensor var_4357_to_fp16 = const()[name = tensor("op_4357_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_283_cast_fp16 = mul(x = var_4356_cast_fp16, y = var_4357_to_fp16)[name = tensor("aw_283_cast_fp16")]; + tensor var_4360_equation_0 = const()[name = tensor("op_4360_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4360_cast_fp16 = einsum(equation = var_4360_equation_0, values = (var_4202_cast_fp16, var_4119_cast_fp16))[name = tensor("op_4360_cast_fp16")]; + tensor var_4361_to_fp16 = const()[name = tensor("op_4361_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_285_cast_fp16 = mul(x = var_4360_cast_fp16, y = var_4361_to_fp16)[name = tensor("aw_285_cast_fp16")]; + tensor var_4364_equation_0 = const()[name = tensor("op_4364_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4364_cast_fp16 = einsum(equation = var_4364_equation_0, values = (var_4206_cast_fp16, var_4123_cast_fp16))[name = tensor("op_4364_cast_fp16")]; + tensor var_4365_to_fp16 = const()[name = tensor("op_4365_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_287_cast_fp16 = mul(x = var_4364_cast_fp16, y = var_4365_to_fp16)[name = tensor("aw_287_cast_fp16")]; + tensor var_4368_equation_0 = const()[name = tensor("op_4368_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4368_cast_fp16 = einsum(equation = var_4368_equation_0, values = (var_4210_cast_fp16, var_4127_cast_fp16))[name = tensor("op_4368_cast_fp16")]; + tensor var_4369_to_fp16 = const()[name = tensor("op_4369_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_289_cast_fp16 = mul(x = var_4368_cast_fp16, y = var_4369_to_fp16)[name = tensor("aw_289_cast_fp16")]; + tensor var_4372_equation_0 = const()[name = tensor("op_4372_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4372_cast_fp16 = einsum(equation = var_4372_equation_0, values = (var_4214_cast_fp16, var_4131_cast_fp16))[name = tensor("op_4372_cast_fp16")]; + tensor var_4373_to_fp16 = const()[name = tensor("op_4373_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_291_cast_fp16 = mul(x = var_4372_cast_fp16, y = var_4373_to_fp16)[name = tensor("aw_291_cast_fp16")]; + tensor var_4376_equation_0 = const()[name = tensor("op_4376_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4376_cast_fp16 = einsum(equation = var_4376_equation_0, values = (var_4218_cast_fp16, var_4135_cast_fp16))[name = tensor("op_4376_cast_fp16")]; + tensor var_4377_to_fp16 = const()[name = tensor("op_4377_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_293_cast_fp16 = mul(x = var_4376_cast_fp16, y = var_4377_to_fp16)[name = tensor("aw_293_cast_fp16")]; + tensor var_4380_equation_0 = const()[name = tensor("op_4380_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4380_cast_fp16 = einsum(equation = var_4380_equation_0, values = (var_4222_cast_fp16, var_4139_cast_fp16))[name = tensor("op_4380_cast_fp16")]; + tensor var_4381_to_fp16 = const()[name = tensor("op_4381_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_295_cast_fp16 = mul(x = var_4380_cast_fp16, y = var_4381_to_fp16)[name = tensor("aw_295_cast_fp16")]; + tensor var_4384_equation_0 = const()[name = tensor("op_4384_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4384_cast_fp16 = einsum(equation = var_4384_equation_0, values = (var_4226_cast_fp16, var_4143_cast_fp16))[name = tensor("op_4384_cast_fp16")]; + tensor var_4385_to_fp16 = const()[name = tensor("op_4385_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_297_cast_fp16 = mul(x = var_4384_cast_fp16, y = var_4385_to_fp16)[name = tensor("aw_297_cast_fp16")]; + tensor var_4388_equation_0 = const()[name = tensor("op_4388_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4388_cast_fp16 = einsum(equation = var_4388_equation_0, values = (var_4230_cast_fp16, var_4147_cast_fp16))[name = tensor("op_4388_cast_fp16")]; + tensor var_4389_to_fp16 = const()[name = tensor("op_4389_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_299_cast_fp16 = mul(x = var_4388_cast_fp16, y = var_4389_to_fp16)[name = tensor("aw_299_cast_fp16")]; + tensor var_4392_equation_0 = const()[name = tensor("op_4392_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4392_cast_fp16 = einsum(equation = var_4392_equation_0, values = (var_4234_cast_fp16, var_4151_cast_fp16))[name = tensor("op_4392_cast_fp16")]; + tensor var_4393_to_fp16 = const()[name = tensor("op_4393_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_301_cast_fp16 = mul(x = var_4392_cast_fp16, y = var_4393_to_fp16)[name = tensor("aw_301_cast_fp16")]; + tensor var_4396_equation_0 = const()[name = tensor("op_4396_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4396_cast_fp16 = einsum(equation = var_4396_equation_0, values = (var_4238_cast_fp16, var_4155_cast_fp16))[name = tensor("op_4396_cast_fp16")]; + tensor var_4397_to_fp16 = const()[name = tensor("op_4397_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_303_cast_fp16 = mul(x = var_4396_cast_fp16, y = var_4397_to_fp16)[name = tensor("aw_303_cast_fp16")]; + tensor var_4400_equation_0 = const()[name = tensor("op_4400_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4400_cast_fp16 = einsum(equation = var_4400_equation_0, values = (var_4242_cast_fp16, var_4159_cast_fp16))[name = tensor("op_4400_cast_fp16")]; + tensor var_4401_to_fp16 = const()[name = tensor("op_4401_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_305_cast_fp16 = mul(x = var_4400_cast_fp16, y = var_4401_to_fp16)[name = tensor("aw_305_cast_fp16")]; + tensor var_4404_equation_0 = const()[name = tensor("op_4404_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4404_cast_fp16 = einsum(equation = var_4404_equation_0, values = (var_4246_cast_fp16, var_4163_cast_fp16))[name = tensor("op_4404_cast_fp16")]; + tensor var_4405_to_fp16 = const()[name = tensor("op_4405_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_307_cast_fp16 = mul(x = var_4404_cast_fp16, y = var_4405_to_fp16)[name = tensor("aw_307_cast_fp16")]; + tensor var_4408_equation_0 = const()[name = tensor("op_4408_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4408_cast_fp16 = einsum(equation = var_4408_equation_0, values = (var_4250_cast_fp16, var_4167_cast_fp16))[name = tensor("op_4408_cast_fp16")]; + tensor var_4409_to_fp16 = const()[name = tensor("op_4409_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_309_cast_fp16 = mul(x = var_4408_cast_fp16, y = var_4409_to_fp16)[name = tensor("aw_309_cast_fp16")]; + tensor var_4412_equation_0 = const()[name = tensor("op_4412_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4412_cast_fp16 = einsum(equation = var_4412_equation_0, values = (var_4254_cast_fp16, var_4171_cast_fp16))[name = tensor("op_4412_cast_fp16")]; + tensor var_4413_to_fp16 = const()[name = tensor("op_4413_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_311_cast_fp16 = mul(x = var_4412_cast_fp16, y = var_4413_to_fp16)[name = tensor("aw_311_cast_fp16")]; + tensor var_4416_equation_0 = const()[name = tensor("op_4416_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4416_cast_fp16 = einsum(equation = var_4416_equation_0, values = (var_4258_cast_fp16, var_4175_cast_fp16))[name = tensor("op_4416_cast_fp16")]; + tensor var_4417_to_fp16 = const()[name = tensor("op_4417_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_313_cast_fp16 = mul(x = var_4416_cast_fp16, y = var_4417_to_fp16)[name = tensor("aw_313_cast_fp16")]; + tensor var_4420_equation_0 = const()[name = tensor("op_4420_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4420_cast_fp16 = einsum(equation = var_4420_equation_0, values = (var_4262_cast_fp16, var_4179_cast_fp16))[name = tensor("op_4420_cast_fp16")]; + tensor var_4421_to_fp16 = const()[name = tensor("op_4421_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_315_cast_fp16 = mul(x = var_4420_cast_fp16, y = var_4421_to_fp16)[name = tensor("aw_315_cast_fp16")]; + tensor var_4424_equation_0 = const()[name = tensor("op_4424_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4424_cast_fp16 = einsum(equation = var_4424_equation_0, values = (var_4266_cast_fp16, var_4183_cast_fp16))[name = tensor("op_4424_cast_fp16")]; + tensor var_4425_to_fp16 = const()[name = tensor("op_4425_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_317_cast_fp16 = mul(x = var_4424_cast_fp16, y = var_4425_to_fp16)[name = tensor("aw_317_cast_fp16")]; + tensor var_4428_equation_0 = const()[name = tensor("op_4428_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4428_cast_fp16 = einsum(equation = var_4428_equation_0, values = (var_4270_cast_fp16, var_4187_cast_fp16))[name = tensor("op_4428_cast_fp16")]; + tensor var_4429_to_fp16 = const()[name = tensor("op_4429_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_319_cast_fp16 = mul(x = var_4428_cast_fp16, y = var_4429_to_fp16)[name = tensor("aw_319_cast_fp16")]; + tensor var_4431_cast_fp16 = softmax(axis = var_2624, x = aw_281_cast_fp16)[name = tensor("op_4431_cast_fp16")]; + tensor var_4432_cast_fp16 = softmax(axis = var_2624, x = aw_283_cast_fp16)[name = tensor("op_4432_cast_fp16")]; + tensor var_4433_cast_fp16 = softmax(axis = var_2624, x = aw_285_cast_fp16)[name = tensor("op_4433_cast_fp16")]; + tensor var_4434_cast_fp16 = softmax(axis = var_2624, x = aw_287_cast_fp16)[name = tensor("op_4434_cast_fp16")]; + tensor var_4435_cast_fp16 = softmax(axis = var_2624, x = aw_289_cast_fp16)[name = tensor("op_4435_cast_fp16")]; + tensor var_4436_cast_fp16 = softmax(axis = var_2624, x = aw_291_cast_fp16)[name = tensor("op_4436_cast_fp16")]; + tensor var_4437_cast_fp16 = softmax(axis = var_2624, x = aw_293_cast_fp16)[name = tensor("op_4437_cast_fp16")]; + tensor var_4438_cast_fp16 = softmax(axis = var_2624, x = aw_295_cast_fp16)[name = tensor("op_4438_cast_fp16")]; + tensor var_4439_cast_fp16 = softmax(axis = var_2624, x = aw_297_cast_fp16)[name = tensor("op_4439_cast_fp16")]; + tensor var_4440_cast_fp16 = softmax(axis = var_2624, x = aw_299_cast_fp16)[name = tensor("op_4440_cast_fp16")]; + tensor var_4441_cast_fp16 = softmax(axis = var_2624, x = aw_301_cast_fp16)[name = tensor("op_4441_cast_fp16")]; + tensor var_4442_cast_fp16 = softmax(axis = var_2624, x = aw_303_cast_fp16)[name = tensor("op_4442_cast_fp16")]; + tensor var_4443_cast_fp16 = softmax(axis = var_2624, x = aw_305_cast_fp16)[name = tensor("op_4443_cast_fp16")]; + tensor var_4444_cast_fp16 = softmax(axis = var_2624, x = aw_307_cast_fp16)[name = tensor("op_4444_cast_fp16")]; + tensor var_4445_cast_fp16 = softmax(axis = var_2624, x = aw_309_cast_fp16)[name = tensor("op_4445_cast_fp16")]; + tensor var_4446_cast_fp16 = softmax(axis = var_2624, x = aw_311_cast_fp16)[name = tensor("op_4446_cast_fp16")]; + tensor var_4447_cast_fp16 = softmax(axis = var_2624, x = aw_313_cast_fp16)[name = tensor("op_4447_cast_fp16")]; + tensor var_4448_cast_fp16 = softmax(axis = var_2624, x = aw_315_cast_fp16)[name = tensor("op_4448_cast_fp16")]; + tensor var_4449_cast_fp16 = softmax(axis = var_2624, x = aw_317_cast_fp16)[name = tensor("op_4449_cast_fp16")]; + tensor var_4450_cast_fp16 = softmax(axis = var_2624, x = aw_319_cast_fp16)[name = tensor("op_4450_cast_fp16")]; + tensor var_4452_equation_0 = const()[name = tensor("op_4452_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4452_cast_fp16 = einsum(equation = var_4452_equation_0, values = (var_4272_cast_fp16, var_4431_cast_fp16))[name = tensor("op_4452_cast_fp16")]; + tensor var_4454_equation_0 = const()[name = tensor("op_4454_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4454_cast_fp16 = einsum(equation = var_4454_equation_0, values = (var_4276_cast_fp16, var_4432_cast_fp16))[name = tensor("op_4454_cast_fp16")]; + tensor var_4456_equation_0 = const()[name = tensor("op_4456_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4456_cast_fp16 = einsum(equation = var_4456_equation_0, values = (var_4280_cast_fp16, var_4433_cast_fp16))[name = tensor("op_4456_cast_fp16")]; + tensor var_4458_equation_0 = const()[name = tensor("op_4458_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4458_cast_fp16 = einsum(equation = var_4458_equation_0, values = (var_4284_cast_fp16, var_4434_cast_fp16))[name = tensor("op_4458_cast_fp16")]; + tensor var_4460_equation_0 = const()[name = tensor("op_4460_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4460_cast_fp16 = einsum(equation = var_4460_equation_0, values = (var_4288_cast_fp16, var_4435_cast_fp16))[name = tensor("op_4460_cast_fp16")]; + tensor var_4462_equation_0 = const()[name = tensor("op_4462_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4462_cast_fp16 = einsum(equation = var_4462_equation_0, values = (var_4292_cast_fp16, var_4436_cast_fp16))[name = tensor("op_4462_cast_fp16")]; + tensor var_4464_equation_0 = const()[name = tensor("op_4464_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4464_cast_fp16 = einsum(equation = var_4464_equation_0, values = (var_4296_cast_fp16, var_4437_cast_fp16))[name = tensor("op_4464_cast_fp16")]; + tensor var_4466_equation_0 = const()[name = tensor("op_4466_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4466_cast_fp16 = einsum(equation = var_4466_equation_0, values = (var_4300_cast_fp16, var_4438_cast_fp16))[name = tensor("op_4466_cast_fp16")]; + tensor var_4468_equation_0 = const()[name = tensor("op_4468_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4468_cast_fp16 = einsum(equation = var_4468_equation_0, values = (var_4304_cast_fp16, var_4439_cast_fp16))[name = tensor("op_4468_cast_fp16")]; + tensor var_4470_equation_0 = const()[name = tensor("op_4470_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4470_cast_fp16 = einsum(equation = var_4470_equation_0, values = (var_4308_cast_fp16, var_4440_cast_fp16))[name = tensor("op_4470_cast_fp16")]; + tensor var_4472_equation_0 = const()[name = tensor("op_4472_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4472_cast_fp16 = einsum(equation = var_4472_equation_0, values = (var_4312_cast_fp16, var_4441_cast_fp16))[name = tensor("op_4472_cast_fp16")]; + tensor var_4474_equation_0 = const()[name = tensor("op_4474_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4474_cast_fp16 = einsum(equation = var_4474_equation_0, values = (var_4316_cast_fp16, var_4442_cast_fp16))[name = tensor("op_4474_cast_fp16")]; + tensor var_4476_equation_0 = const()[name = tensor("op_4476_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4476_cast_fp16 = einsum(equation = var_4476_equation_0, values = (var_4320_cast_fp16, var_4443_cast_fp16))[name = tensor("op_4476_cast_fp16")]; + tensor var_4478_equation_0 = const()[name = tensor("op_4478_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4478_cast_fp16 = einsum(equation = var_4478_equation_0, values = (var_4324_cast_fp16, var_4444_cast_fp16))[name = tensor("op_4478_cast_fp16")]; + tensor var_4480_equation_0 = const()[name = tensor("op_4480_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4480_cast_fp16 = einsum(equation = var_4480_equation_0, values = (var_4328_cast_fp16, var_4445_cast_fp16))[name = tensor("op_4480_cast_fp16")]; + tensor var_4482_equation_0 = const()[name = tensor("op_4482_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4482_cast_fp16 = einsum(equation = var_4482_equation_0, values = (var_4332_cast_fp16, var_4446_cast_fp16))[name = tensor("op_4482_cast_fp16")]; + tensor var_4484_equation_0 = const()[name = tensor("op_4484_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4484_cast_fp16 = einsum(equation = var_4484_equation_0, values = (var_4336_cast_fp16, var_4447_cast_fp16))[name = tensor("op_4484_cast_fp16")]; + tensor var_4486_equation_0 = const()[name = tensor("op_4486_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4486_cast_fp16 = einsum(equation = var_4486_equation_0, values = (var_4340_cast_fp16, var_4448_cast_fp16))[name = tensor("op_4486_cast_fp16")]; + tensor var_4488_equation_0 = const()[name = tensor("op_4488_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4488_cast_fp16 = einsum(equation = var_4488_equation_0, values = (var_4344_cast_fp16, var_4449_cast_fp16))[name = tensor("op_4488_cast_fp16")]; + tensor var_4490_equation_0 = const()[name = tensor("op_4490_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4490_cast_fp16 = einsum(equation = var_4490_equation_0, values = (var_4348_cast_fp16, var_4450_cast_fp16))[name = tensor("op_4490_cast_fp16")]; + tensor input_141_interleave_0 = const()[name = tensor("input_141_interleave_0"), val = tensor(false)]; + tensor input_141_cast_fp16 = concat(axis = var_2624, interleave = input_141_interleave_0, values = (var_4452_cast_fp16, var_4454_cast_fp16, var_4456_cast_fp16, var_4458_cast_fp16, var_4460_cast_fp16, var_4462_cast_fp16, var_4464_cast_fp16, var_4466_cast_fp16, var_4468_cast_fp16, var_4470_cast_fp16, var_4472_cast_fp16, var_4474_cast_fp16, var_4476_cast_fp16, var_4478_cast_fp16, var_4480_cast_fp16, var_4482_cast_fp16, var_4484_cast_fp16, var_4486_cast_fp16, var_4488_cast_fp16, var_4490_cast_fp16))[name = tensor("input_141_cast_fp16")]; + tensor var_4496 = const()[name = tensor("op_4496"), val = tensor([1, 1])]; + tensor var_4498 = const()[name = tensor("op_4498"), val = tensor([1, 1])]; + tensor var_4500_pad_type_0 = const()[name = tensor("op_4500_pad_type_0"), val = tensor("custom")]; + tensor var_4500_pad_0 = const()[name = tensor("op_4500_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_1_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_1_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(294562176)))]; + tensor down_blocks_2_attentions_0_transformer_blocks_1_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_1_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(297839040)))]; + tensor var_4500_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_1_attn2_to_out_0_bias_to_fp16, dilations = var_4498, groups = var_2624, pad = var_4500_pad_0, pad_type = var_4500_pad_type_0, strides = var_4496, weight = down_blocks_2_attentions_0_transformer_blocks_1_attn2_to_out_0_weight_to_fp16, x = input_141_cast_fp16)[name = tensor("op_4500_cast_fp16")]; + tensor inputs_35_cast_fp16 = add(x = var_4500_cast_fp16, y = inputs_33_cast_fp16)[name = tensor("inputs_35_cast_fp16")]; + tensor input_143_axes_0 = const()[name = tensor("input_143_axes_0"), val = tensor([1])]; + tensor input_143_gamma_0_to_fp16 = const()[name = tensor("input_143_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(297841664)))]; + tensor input_143_beta_0_to_fp16 = const()[name = tensor("input_143_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(297844288)))]; + tensor var_4510_to_fp16 = const()[name = tensor("op_4510_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_143_cast_fp16 = layer_norm(axes = input_143_axes_0, beta = input_143_beta_0_to_fp16, epsilon = var_4510_to_fp16, gamma = input_143_gamma_0_to_fp16, x = inputs_35_cast_fp16)[name = tensor("input_143_cast_fp16")]; + tensor var_4526 = const()[name = tensor("op_4526"), val = tensor([1, 1])]; + tensor var_4528 = const()[name = tensor("op_4528"), val = tensor([1, 1])]; + tensor var_4530_pad_type_0 = const()[name = tensor("op_4530_pad_type_0"), val = tensor("custom")]; + tensor var_4530_pad_0 = const()[name = tensor("op_4530_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_1_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_1_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(297846912)))]; + tensor down_blocks_2_attentions_0_transformer_blocks_1_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_1_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(324061376)))]; + tensor var_4530_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_1_ff_net_0_proj_bias_to_fp16, dilations = var_4528, groups = var_2624, pad = var_4530_pad_0, pad_type = var_4530_pad_type_0, strides = var_4526, weight = down_blocks_2_attentions_0_transformer_blocks_1_ff_net_0_proj_weight_to_fp16, x = input_143_cast_fp16)[name = tensor("op_4530_cast_fp16")]; + tensor var_4531_split_sizes_0 = const()[name = tensor("op_4531_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_4531_axis_0 = const()[name = tensor("op_4531_axis_0"), val = tensor(1)]; + tensor var_4531_cast_fp16_0, tensor var_4531_cast_fp16_1 = split(axis = var_4531_axis_0, split_sizes = var_4531_split_sizes_0, x = var_4530_cast_fp16)[name = tensor("op_4531_cast_fp16")]; + tensor var_4533_mode_0 = const()[name = tensor("op_4533_mode_0"), val = tensor("EXACT")]; + tensor var_4533_cast_fp16 = gelu(mode = var_4533_mode_0, x = var_4531_cast_fp16_1)[name = tensor("op_4533_cast_fp16")]; + tensor input_145_cast_fp16 = mul(x = var_4531_cast_fp16_0, y = var_4533_cast_fp16)[name = tensor("input_145_cast_fp16")]; + tensor var_4537 = const()[name = tensor("op_4537"), val = tensor([1, 1])]; + tensor var_4539 = const()[name = tensor("op_4539"), val = tensor([1, 1])]; + tensor var_4541_pad_type_0 = const()[name = tensor("op_4541_pad_type_0"), val = tensor("custom")]; + tensor var_4541_pad_0 = const()[name = tensor("op_4541_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_1_ff_net_2_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_1_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(324081920)))]; + tensor down_blocks_2_attentions_0_transformer_blocks_1_ff_net_2_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_1_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(337189184)))]; + tensor var_4541_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_1_ff_net_2_bias_to_fp16, dilations = var_4539, groups = var_2624, pad = var_4541_pad_0, pad_type = var_4541_pad_type_0, strides = var_4537, weight = down_blocks_2_attentions_0_transformer_blocks_1_ff_net_2_weight_to_fp16, x = input_145_cast_fp16)[name = tensor("op_4541_cast_fp16")]; + tensor inputs_37_cast_fp16 = add(x = var_4541_cast_fp16, y = inputs_35_cast_fp16)[name = tensor("inputs_37_cast_fp16")]; + tensor hidden_states_77_axes_0 = const()[name = tensor("hidden_states_77_axes_0"), val = tensor([1])]; + tensor hidden_states_77_gamma_0_to_fp16 = const()[name = tensor("hidden_states_77_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(337191808)))]; + tensor hidden_states_77_beta_0_to_fp16 = const()[name = tensor("hidden_states_77_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(337194432)))]; + tensor var_4557_to_fp16 = const()[name = tensor("op_4557_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_77_cast_fp16 = layer_norm(axes = hidden_states_77_axes_0, beta = hidden_states_77_beta_0_to_fp16, epsilon = var_4557_to_fp16, gamma = hidden_states_77_gamma_0_to_fp16, x = inputs_37_cast_fp16)[name = tensor("hidden_states_77_cast_fp16")]; + tensor var_4572 = const()[name = tensor("op_4572"), val = tensor([1, 1])]; + tensor var_4574 = const()[name = tensor("op_4574"), val = tensor([1, 1])]; + tensor q_25_pad_type_0 = const()[name = tensor("q_25_pad_type_0"), val = tensor("custom")]; + tensor q_25_pad_0 = const()[name = tensor("q_25_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_2_attn1_to_q_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_2_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(337197056)))]; + tensor q_25_cast_fp16 = conv(dilations = var_4574, groups = var_2624, pad = q_25_pad_0, pad_type = q_25_pad_type_0, strides = var_4572, weight = down_blocks_2_attentions_0_transformer_blocks_2_attn1_to_q_weight_to_fp16, x = hidden_states_77_cast_fp16)[name = tensor("q_25_cast_fp16")]; + tensor var_4578 = const()[name = tensor("op_4578"), val = tensor([1, 1])]; + tensor var_4580 = const()[name = tensor("op_4580"), val = tensor([1, 1])]; + tensor k_49_pad_type_0 = const()[name = tensor("k_49_pad_type_0"), val = tensor("custom")]; + tensor k_49_pad_0 = const()[name = tensor("k_49_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_2_attn1_to_k_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_2_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(340473920)))]; + tensor k_49_cast_fp16 = conv(dilations = var_4580, groups = var_2624, pad = k_49_pad_0, pad_type = k_49_pad_type_0, strides = var_4578, weight = down_blocks_2_attentions_0_transformer_blocks_2_attn1_to_k_weight_to_fp16, x = hidden_states_77_cast_fp16)[name = tensor("k_49_cast_fp16")]; + tensor var_4584 = const()[name = tensor("op_4584"), val = tensor([1, 1])]; + tensor var_4586 = const()[name = tensor("op_4586"), val = tensor([1, 1])]; + tensor v_25_pad_type_0 = const()[name = tensor("v_25_pad_type_0"), val = tensor("custom")]; + tensor v_25_pad_0 = const()[name = tensor("v_25_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_2_attn1_to_v_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_2_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(343750784)))]; + tensor v_25_cast_fp16 = conv(dilations = var_4586, groups = var_2624, pad = v_25_pad_0, pad_type = v_25_pad_type_0, strides = var_4584, weight = down_blocks_2_attentions_0_transformer_blocks_2_attn1_to_v_weight_to_fp16, x = hidden_states_77_cast_fp16)[name = tensor("v_25_cast_fp16")]; + tensor var_4590_begin_0 = const()[name = tensor("op_4590_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4590_end_0 = const()[name = tensor("op_4590_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_4590_end_mask_0 = const()[name = tensor("op_4590_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4590_cast_fp16 = slice_by_index(begin = var_4590_begin_0, end = var_4590_end_0, end_mask = var_4590_end_mask_0, x = q_25_cast_fp16)[name = tensor("op_4590_cast_fp16")]; + tensor var_4594_begin_0 = const()[name = tensor("op_4594_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_4594_end_0 = const()[name = tensor("op_4594_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_4594_end_mask_0 = const()[name = tensor("op_4594_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4594_cast_fp16 = slice_by_index(begin = var_4594_begin_0, end = var_4594_end_0, end_mask = var_4594_end_mask_0, x = q_25_cast_fp16)[name = tensor("op_4594_cast_fp16")]; + tensor var_4598_begin_0 = const()[name = tensor("op_4598_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_4598_end_0 = const()[name = tensor("op_4598_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_4598_end_mask_0 = const()[name = tensor("op_4598_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4598_cast_fp16 = slice_by_index(begin = var_4598_begin_0, end = var_4598_end_0, end_mask = var_4598_end_mask_0, x = q_25_cast_fp16)[name = tensor("op_4598_cast_fp16")]; + tensor var_4602_begin_0 = const()[name = tensor("op_4602_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_4602_end_0 = const()[name = tensor("op_4602_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_4602_end_mask_0 = const()[name = tensor("op_4602_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4602_cast_fp16 = slice_by_index(begin = var_4602_begin_0, end = var_4602_end_0, end_mask = var_4602_end_mask_0, x = q_25_cast_fp16)[name = tensor("op_4602_cast_fp16")]; + tensor var_4606_begin_0 = const()[name = tensor("op_4606_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_4606_end_0 = const()[name = tensor("op_4606_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_4606_end_mask_0 = const()[name = tensor("op_4606_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4606_cast_fp16 = slice_by_index(begin = var_4606_begin_0, end = var_4606_end_0, end_mask = var_4606_end_mask_0, x = q_25_cast_fp16)[name = tensor("op_4606_cast_fp16")]; + tensor var_4610_begin_0 = const()[name = tensor("op_4610_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_4610_end_0 = const()[name = tensor("op_4610_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_4610_end_mask_0 = const()[name = tensor("op_4610_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4610_cast_fp16 = slice_by_index(begin = var_4610_begin_0, end = var_4610_end_0, end_mask = var_4610_end_mask_0, x = q_25_cast_fp16)[name = tensor("op_4610_cast_fp16")]; + tensor var_4614_begin_0 = const()[name = tensor("op_4614_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_4614_end_0 = const()[name = tensor("op_4614_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_4614_end_mask_0 = const()[name = tensor("op_4614_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4614_cast_fp16 = slice_by_index(begin = var_4614_begin_0, end = var_4614_end_0, end_mask = var_4614_end_mask_0, x = q_25_cast_fp16)[name = tensor("op_4614_cast_fp16")]; + tensor var_4618_begin_0 = const()[name = tensor("op_4618_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_4618_end_0 = const()[name = tensor("op_4618_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_4618_end_mask_0 = const()[name = tensor("op_4618_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4618_cast_fp16 = slice_by_index(begin = var_4618_begin_0, end = var_4618_end_0, end_mask = var_4618_end_mask_0, x = q_25_cast_fp16)[name = tensor("op_4618_cast_fp16")]; + tensor var_4622_begin_0 = const()[name = tensor("op_4622_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_4622_end_0 = const()[name = tensor("op_4622_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_4622_end_mask_0 = const()[name = tensor("op_4622_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4622_cast_fp16 = slice_by_index(begin = var_4622_begin_0, end = var_4622_end_0, end_mask = var_4622_end_mask_0, x = q_25_cast_fp16)[name = tensor("op_4622_cast_fp16")]; + tensor var_4626_begin_0 = const()[name = tensor("op_4626_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_4626_end_0 = const()[name = tensor("op_4626_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_4626_end_mask_0 = const()[name = tensor("op_4626_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4626_cast_fp16 = slice_by_index(begin = var_4626_begin_0, end = var_4626_end_0, end_mask = var_4626_end_mask_0, x = q_25_cast_fp16)[name = tensor("op_4626_cast_fp16")]; + tensor var_4630_begin_0 = const()[name = tensor("op_4630_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_4630_end_0 = const()[name = tensor("op_4630_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_4630_end_mask_0 = const()[name = tensor("op_4630_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4630_cast_fp16 = slice_by_index(begin = var_4630_begin_0, end = var_4630_end_0, end_mask = var_4630_end_mask_0, x = q_25_cast_fp16)[name = tensor("op_4630_cast_fp16")]; + tensor var_4634_begin_0 = const()[name = tensor("op_4634_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_4634_end_0 = const()[name = tensor("op_4634_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_4634_end_mask_0 = const()[name = tensor("op_4634_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4634_cast_fp16 = slice_by_index(begin = var_4634_begin_0, end = var_4634_end_0, end_mask = var_4634_end_mask_0, x = q_25_cast_fp16)[name = tensor("op_4634_cast_fp16")]; + tensor var_4638_begin_0 = const()[name = tensor("op_4638_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_4638_end_0 = const()[name = tensor("op_4638_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_4638_end_mask_0 = const()[name = tensor("op_4638_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4638_cast_fp16 = slice_by_index(begin = var_4638_begin_0, end = var_4638_end_0, end_mask = var_4638_end_mask_0, x = q_25_cast_fp16)[name = tensor("op_4638_cast_fp16")]; + tensor var_4642_begin_0 = const()[name = tensor("op_4642_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_4642_end_0 = const()[name = tensor("op_4642_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_4642_end_mask_0 = const()[name = tensor("op_4642_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4642_cast_fp16 = slice_by_index(begin = var_4642_begin_0, end = var_4642_end_0, end_mask = var_4642_end_mask_0, x = q_25_cast_fp16)[name = tensor("op_4642_cast_fp16")]; + tensor var_4646_begin_0 = const()[name = tensor("op_4646_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_4646_end_0 = const()[name = tensor("op_4646_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_4646_end_mask_0 = const()[name = tensor("op_4646_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4646_cast_fp16 = slice_by_index(begin = var_4646_begin_0, end = var_4646_end_0, end_mask = var_4646_end_mask_0, x = q_25_cast_fp16)[name = tensor("op_4646_cast_fp16")]; + tensor var_4650_begin_0 = const()[name = tensor("op_4650_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_4650_end_0 = const()[name = tensor("op_4650_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_4650_end_mask_0 = const()[name = tensor("op_4650_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4650_cast_fp16 = slice_by_index(begin = var_4650_begin_0, end = var_4650_end_0, end_mask = var_4650_end_mask_0, x = q_25_cast_fp16)[name = tensor("op_4650_cast_fp16")]; + tensor var_4654_begin_0 = const()[name = tensor("op_4654_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_4654_end_0 = const()[name = tensor("op_4654_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_4654_end_mask_0 = const()[name = tensor("op_4654_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4654_cast_fp16 = slice_by_index(begin = var_4654_begin_0, end = var_4654_end_0, end_mask = var_4654_end_mask_0, x = q_25_cast_fp16)[name = tensor("op_4654_cast_fp16")]; + tensor var_4658_begin_0 = const()[name = tensor("op_4658_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_4658_end_0 = const()[name = tensor("op_4658_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_4658_end_mask_0 = const()[name = tensor("op_4658_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4658_cast_fp16 = slice_by_index(begin = var_4658_begin_0, end = var_4658_end_0, end_mask = var_4658_end_mask_0, x = q_25_cast_fp16)[name = tensor("op_4658_cast_fp16")]; + tensor var_4662_begin_0 = const()[name = tensor("op_4662_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_4662_end_0 = const()[name = tensor("op_4662_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_4662_end_mask_0 = const()[name = tensor("op_4662_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4662_cast_fp16 = slice_by_index(begin = var_4662_begin_0, end = var_4662_end_0, end_mask = var_4662_end_mask_0, x = q_25_cast_fp16)[name = tensor("op_4662_cast_fp16")]; + tensor var_4666_begin_0 = const()[name = tensor("op_4666_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_4666_end_0 = const()[name = tensor("op_4666_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_4666_end_mask_0 = const()[name = tensor("op_4666_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4666_cast_fp16 = slice_by_index(begin = var_4666_begin_0, end = var_4666_end_0, end_mask = var_4666_end_mask_0, x = q_25_cast_fp16)[name = tensor("op_4666_cast_fp16")]; + tensor k_51_perm_0 = const()[name = tensor("k_51_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_4673_begin_0 = const()[name = tensor("op_4673_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4673_end_0 = const()[name = tensor("op_4673_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_4673_end_mask_0 = const()[name = tensor("op_4673_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_51_cast_fp16 = transpose(perm = k_51_perm_0, x = k_49_cast_fp16)[name = tensor("transpose_127")]; + tensor var_4673_cast_fp16 = slice_by_index(begin = var_4673_begin_0, end = var_4673_end_0, end_mask = var_4673_end_mask_0, x = k_51_cast_fp16)[name = tensor("op_4673_cast_fp16")]; + tensor var_4677_begin_0 = const()[name = tensor("op_4677_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_4677_end_0 = const()[name = tensor("op_4677_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_4677_end_mask_0 = const()[name = tensor("op_4677_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4677_cast_fp16 = slice_by_index(begin = var_4677_begin_0, end = var_4677_end_0, end_mask = var_4677_end_mask_0, x = k_51_cast_fp16)[name = tensor("op_4677_cast_fp16")]; + tensor var_4681_begin_0 = const()[name = tensor("op_4681_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_4681_end_0 = const()[name = tensor("op_4681_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_4681_end_mask_0 = const()[name = tensor("op_4681_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4681_cast_fp16 = slice_by_index(begin = var_4681_begin_0, end = var_4681_end_0, end_mask = var_4681_end_mask_0, x = k_51_cast_fp16)[name = tensor("op_4681_cast_fp16")]; + tensor var_4685_begin_0 = const()[name = tensor("op_4685_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_4685_end_0 = const()[name = tensor("op_4685_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_4685_end_mask_0 = const()[name = tensor("op_4685_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4685_cast_fp16 = slice_by_index(begin = var_4685_begin_0, end = var_4685_end_0, end_mask = var_4685_end_mask_0, x = k_51_cast_fp16)[name = tensor("op_4685_cast_fp16")]; + tensor var_4689_begin_0 = const()[name = tensor("op_4689_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_4689_end_0 = const()[name = tensor("op_4689_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_4689_end_mask_0 = const()[name = tensor("op_4689_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4689_cast_fp16 = slice_by_index(begin = var_4689_begin_0, end = var_4689_end_0, end_mask = var_4689_end_mask_0, x = k_51_cast_fp16)[name = tensor("op_4689_cast_fp16")]; + tensor var_4693_begin_0 = const()[name = tensor("op_4693_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_4693_end_0 = const()[name = tensor("op_4693_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_4693_end_mask_0 = const()[name = tensor("op_4693_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4693_cast_fp16 = slice_by_index(begin = var_4693_begin_0, end = var_4693_end_0, end_mask = var_4693_end_mask_0, x = k_51_cast_fp16)[name = tensor("op_4693_cast_fp16")]; + tensor var_4697_begin_0 = const()[name = tensor("op_4697_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_4697_end_0 = const()[name = tensor("op_4697_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_4697_end_mask_0 = const()[name = tensor("op_4697_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4697_cast_fp16 = slice_by_index(begin = var_4697_begin_0, end = var_4697_end_0, end_mask = var_4697_end_mask_0, x = k_51_cast_fp16)[name = tensor("op_4697_cast_fp16")]; + tensor var_4701_begin_0 = const()[name = tensor("op_4701_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_4701_end_0 = const()[name = tensor("op_4701_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_4701_end_mask_0 = const()[name = tensor("op_4701_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4701_cast_fp16 = slice_by_index(begin = var_4701_begin_0, end = var_4701_end_0, end_mask = var_4701_end_mask_0, x = k_51_cast_fp16)[name = tensor("op_4701_cast_fp16")]; + tensor var_4705_begin_0 = const()[name = tensor("op_4705_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_4705_end_0 = const()[name = tensor("op_4705_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_4705_end_mask_0 = const()[name = tensor("op_4705_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4705_cast_fp16 = slice_by_index(begin = var_4705_begin_0, end = var_4705_end_0, end_mask = var_4705_end_mask_0, x = k_51_cast_fp16)[name = tensor("op_4705_cast_fp16")]; + tensor var_4709_begin_0 = const()[name = tensor("op_4709_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_4709_end_0 = const()[name = tensor("op_4709_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_4709_end_mask_0 = const()[name = tensor("op_4709_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4709_cast_fp16 = slice_by_index(begin = var_4709_begin_0, end = var_4709_end_0, end_mask = var_4709_end_mask_0, x = k_51_cast_fp16)[name = tensor("op_4709_cast_fp16")]; + tensor var_4713_begin_0 = const()[name = tensor("op_4713_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_4713_end_0 = const()[name = tensor("op_4713_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_4713_end_mask_0 = const()[name = tensor("op_4713_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4713_cast_fp16 = slice_by_index(begin = var_4713_begin_0, end = var_4713_end_0, end_mask = var_4713_end_mask_0, x = k_51_cast_fp16)[name = tensor("op_4713_cast_fp16")]; + tensor var_4717_begin_0 = const()[name = tensor("op_4717_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_4717_end_0 = const()[name = tensor("op_4717_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_4717_end_mask_0 = const()[name = tensor("op_4717_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4717_cast_fp16 = slice_by_index(begin = var_4717_begin_0, end = var_4717_end_0, end_mask = var_4717_end_mask_0, x = k_51_cast_fp16)[name = tensor("op_4717_cast_fp16")]; + tensor var_4721_begin_0 = const()[name = tensor("op_4721_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_4721_end_0 = const()[name = tensor("op_4721_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_4721_end_mask_0 = const()[name = tensor("op_4721_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4721_cast_fp16 = slice_by_index(begin = var_4721_begin_0, end = var_4721_end_0, end_mask = var_4721_end_mask_0, x = k_51_cast_fp16)[name = tensor("op_4721_cast_fp16")]; + tensor var_4725_begin_0 = const()[name = tensor("op_4725_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_4725_end_0 = const()[name = tensor("op_4725_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_4725_end_mask_0 = const()[name = tensor("op_4725_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4725_cast_fp16 = slice_by_index(begin = var_4725_begin_0, end = var_4725_end_0, end_mask = var_4725_end_mask_0, x = k_51_cast_fp16)[name = tensor("op_4725_cast_fp16")]; + tensor var_4729_begin_0 = const()[name = tensor("op_4729_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_4729_end_0 = const()[name = tensor("op_4729_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_4729_end_mask_0 = const()[name = tensor("op_4729_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4729_cast_fp16 = slice_by_index(begin = var_4729_begin_0, end = var_4729_end_0, end_mask = var_4729_end_mask_0, x = k_51_cast_fp16)[name = tensor("op_4729_cast_fp16")]; + tensor var_4733_begin_0 = const()[name = tensor("op_4733_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_4733_end_0 = const()[name = tensor("op_4733_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_4733_end_mask_0 = const()[name = tensor("op_4733_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4733_cast_fp16 = slice_by_index(begin = var_4733_begin_0, end = var_4733_end_0, end_mask = var_4733_end_mask_0, x = k_51_cast_fp16)[name = tensor("op_4733_cast_fp16")]; + tensor var_4737_begin_0 = const()[name = tensor("op_4737_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_4737_end_0 = const()[name = tensor("op_4737_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_4737_end_mask_0 = const()[name = tensor("op_4737_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4737_cast_fp16 = slice_by_index(begin = var_4737_begin_0, end = var_4737_end_0, end_mask = var_4737_end_mask_0, x = k_51_cast_fp16)[name = tensor("op_4737_cast_fp16")]; + tensor var_4741_begin_0 = const()[name = tensor("op_4741_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_4741_end_0 = const()[name = tensor("op_4741_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_4741_end_mask_0 = const()[name = tensor("op_4741_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4741_cast_fp16 = slice_by_index(begin = var_4741_begin_0, end = var_4741_end_0, end_mask = var_4741_end_mask_0, x = k_51_cast_fp16)[name = tensor("op_4741_cast_fp16")]; + tensor var_4745_begin_0 = const()[name = tensor("op_4745_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_4745_end_0 = const()[name = tensor("op_4745_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_4745_end_mask_0 = const()[name = tensor("op_4745_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4745_cast_fp16 = slice_by_index(begin = var_4745_begin_0, end = var_4745_end_0, end_mask = var_4745_end_mask_0, x = k_51_cast_fp16)[name = tensor("op_4745_cast_fp16")]; + tensor var_4749_begin_0 = const()[name = tensor("op_4749_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_4749_end_0 = const()[name = tensor("op_4749_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_4749_end_mask_0 = const()[name = tensor("op_4749_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4749_cast_fp16 = slice_by_index(begin = var_4749_begin_0, end = var_4749_end_0, end_mask = var_4749_end_mask_0, x = k_51_cast_fp16)[name = tensor("op_4749_cast_fp16")]; + tensor var_4751_begin_0 = const()[name = tensor("op_4751_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4751_end_0 = const()[name = tensor("op_4751_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_4751_end_mask_0 = const()[name = tensor("op_4751_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4751_cast_fp16 = slice_by_index(begin = var_4751_begin_0, end = var_4751_end_0, end_mask = var_4751_end_mask_0, x = v_25_cast_fp16)[name = tensor("op_4751_cast_fp16")]; + tensor var_4755_begin_0 = const()[name = tensor("op_4755_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_4755_end_0 = const()[name = tensor("op_4755_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_4755_end_mask_0 = const()[name = tensor("op_4755_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4755_cast_fp16 = slice_by_index(begin = var_4755_begin_0, end = var_4755_end_0, end_mask = var_4755_end_mask_0, x = v_25_cast_fp16)[name = tensor("op_4755_cast_fp16")]; + tensor var_4759_begin_0 = const()[name = tensor("op_4759_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_4759_end_0 = const()[name = tensor("op_4759_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_4759_end_mask_0 = const()[name = tensor("op_4759_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4759_cast_fp16 = slice_by_index(begin = var_4759_begin_0, end = var_4759_end_0, end_mask = var_4759_end_mask_0, x = v_25_cast_fp16)[name = tensor("op_4759_cast_fp16")]; + tensor var_4763_begin_0 = const()[name = tensor("op_4763_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_4763_end_0 = const()[name = tensor("op_4763_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_4763_end_mask_0 = const()[name = tensor("op_4763_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4763_cast_fp16 = slice_by_index(begin = var_4763_begin_0, end = var_4763_end_0, end_mask = var_4763_end_mask_0, x = v_25_cast_fp16)[name = tensor("op_4763_cast_fp16")]; + tensor var_4767_begin_0 = const()[name = tensor("op_4767_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_4767_end_0 = const()[name = tensor("op_4767_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_4767_end_mask_0 = const()[name = tensor("op_4767_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4767_cast_fp16 = slice_by_index(begin = var_4767_begin_0, end = var_4767_end_0, end_mask = var_4767_end_mask_0, x = v_25_cast_fp16)[name = tensor("op_4767_cast_fp16")]; + tensor var_4771_begin_0 = const()[name = tensor("op_4771_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_4771_end_0 = const()[name = tensor("op_4771_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_4771_end_mask_0 = const()[name = tensor("op_4771_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4771_cast_fp16 = slice_by_index(begin = var_4771_begin_0, end = var_4771_end_0, end_mask = var_4771_end_mask_0, x = v_25_cast_fp16)[name = tensor("op_4771_cast_fp16")]; + tensor var_4775_begin_0 = const()[name = tensor("op_4775_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_4775_end_0 = const()[name = tensor("op_4775_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_4775_end_mask_0 = const()[name = tensor("op_4775_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4775_cast_fp16 = slice_by_index(begin = var_4775_begin_0, end = var_4775_end_0, end_mask = var_4775_end_mask_0, x = v_25_cast_fp16)[name = tensor("op_4775_cast_fp16")]; + tensor var_4779_begin_0 = const()[name = tensor("op_4779_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_4779_end_0 = const()[name = tensor("op_4779_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_4779_end_mask_0 = const()[name = tensor("op_4779_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4779_cast_fp16 = slice_by_index(begin = var_4779_begin_0, end = var_4779_end_0, end_mask = var_4779_end_mask_0, x = v_25_cast_fp16)[name = tensor("op_4779_cast_fp16")]; + tensor var_4783_begin_0 = const()[name = tensor("op_4783_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_4783_end_0 = const()[name = tensor("op_4783_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_4783_end_mask_0 = const()[name = tensor("op_4783_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4783_cast_fp16 = slice_by_index(begin = var_4783_begin_0, end = var_4783_end_0, end_mask = var_4783_end_mask_0, x = v_25_cast_fp16)[name = tensor("op_4783_cast_fp16")]; + tensor var_4787_begin_0 = const()[name = tensor("op_4787_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_4787_end_0 = const()[name = tensor("op_4787_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_4787_end_mask_0 = const()[name = tensor("op_4787_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4787_cast_fp16 = slice_by_index(begin = var_4787_begin_0, end = var_4787_end_0, end_mask = var_4787_end_mask_0, x = v_25_cast_fp16)[name = tensor("op_4787_cast_fp16")]; + tensor var_4791_begin_0 = const()[name = tensor("op_4791_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_4791_end_0 = const()[name = tensor("op_4791_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_4791_end_mask_0 = const()[name = tensor("op_4791_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4791_cast_fp16 = slice_by_index(begin = var_4791_begin_0, end = var_4791_end_0, end_mask = var_4791_end_mask_0, x = v_25_cast_fp16)[name = tensor("op_4791_cast_fp16")]; + tensor var_4795_begin_0 = const()[name = tensor("op_4795_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_4795_end_0 = const()[name = tensor("op_4795_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_4795_end_mask_0 = const()[name = tensor("op_4795_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4795_cast_fp16 = slice_by_index(begin = var_4795_begin_0, end = var_4795_end_0, end_mask = var_4795_end_mask_0, x = v_25_cast_fp16)[name = tensor("op_4795_cast_fp16")]; + tensor var_4799_begin_0 = const()[name = tensor("op_4799_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_4799_end_0 = const()[name = tensor("op_4799_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_4799_end_mask_0 = const()[name = tensor("op_4799_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4799_cast_fp16 = slice_by_index(begin = var_4799_begin_0, end = var_4799_end_0, end_mask = var_4799_end_mask_0, x = v_25_cast_fp16)[name = tensor("op_4799_cast_fp16")]; + tensor var_4803_begin_0 = const()[name = tensor("op_4803_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_4803_end_0 = const()[name = tensor("op_4803_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_4803_end_mask_0 = const()[name = tensor("op_4803_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4803_cast_fp16 = slice_by_index(begin = var_4803_begin_0, end = var_4803_end_0, end_mask = var_4803_end_mask_0, x = v_25_cast_fp16)[name = tensor("op_4803_cast_fp16")]; + tensor var_4807_begin_0 = const()[name = tensor("op_4807_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_4807_end_0 = const()[name = tensor("op_4807_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_4807_end_mask_0 = const()[name = tensor("op_4807_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4807_cast_fp16 = slice_by_index(begin = var_4807_begin_0, end = var_4807_end_0, end_mask = var_4807_end_mask_0, x = v_25_cast_fp16)[name = tensor("op_4807_cast_fp16")]; + tensor var_4811_begin_0 = const()[name = tensor("op_4811_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_4811_end_0 = const()[name = tensor("op_4811_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_4811_end_mask_0 = const()[name = tensor("op_4811_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4811_cast_fp16 = slice_by_index(begin = var_4811_begin_0, end = var_4811_end_0, end_mask = var_4811_end_mask_0, x = v_25_cast_fp16)[name = tensor("op_4811_cast_fp16")]; + tensor var_4815_begin_0 = const()[name = tensor("op_4815_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_4815_end_0 = const()[name = tensor("op_4815_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_4815_end_mask_0 = const()[name = tensor("op_4815_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4815_cast_fp16 = slice_by_index(begin = var_4815_begin_0, end = var_4815_end_0, end_mask = var_4815_end_mask_0, x = v_25_cast_fp16)[name = tensor("op_4815_cast_fp16")]; + tensor var_4819_begin_0 = const()[name = tensor("op_4819_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_4819_end_0 = const()[name = tensor("op_4819_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_4819_end_mask_0 = const()[name = tensor("op_4819_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4819_cast_fp16 = slice_by_index(begin = var_4819_begin_0, end = var_4819_end_0, end_mask = var_4819_end_mask_0, x = v_25_cast_fp16)[name = tensor("op_4819_cast_fp16")]; + tensor var_4823_begin_0 = const()[name = tensor("op_4823_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_4823_end_0 = const()[name = tensor("op_4823_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_4823_end_mask_0 = const()[name = tensor("op_4823_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4823_cast_fp16 = slice_by_index(begin = var_4823_begin_0, end = var_4823_end_0, end_mask = var_4823_end_mask_0, x = v_25_cast_fp16)[name = tensor("op_4823_cast_fp16")]; + tensor var_4827_begin_0 = const()[name = tensor("op_4827_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_4827_end_0 = const()[name = tensor("op_4827_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_4827_end_mask_0 = const()[name = tensor("op_4827_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4827_cast_fp16 = slice_by_index(begin = var_4827_begin_0, end = var_4827_end_0, end_mask = var_4827_end_mask_0, x = v_25_cast_fp16)[name = tensor("op_4827_cast_fp16")]; + tensor var_4831_equation_0 = const()[name = tensor("op_4831_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4831_cast_fp16 = einsum(equation = var_4831_equation_0, values = (var_4673_cast_fp16, var_4590_cast_fp16))[name = tensor("op_4831_cast_fp16")]; + tensor var_4832_to_fp16 = const()[name = tensor("op_4832_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_321_cast_fp16 = mul(x = var_4831_cast_fp16, y = var_4832_to_fp16)[name = tensor("aw_321_cast_fp16")]; + tensor var_4835_equation_0 = const()[name = tensor("op_4835_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4835_cast_fp16 = einsum(equation = var_4835_equation_0, values = (var_4677_cast_fp16, var_4594_cast_fp16))[name = tensor("op_4835_cast_fp16")]; + tensor var_4836_to_fp16 = const()[name = tensor("op_4836_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_323_cast_fp16 = mul(x = var_4835_cast_fp16, y = var_4836_to_fp16)[name = tensor("aw_323_cast_fp16")]; + tensor var_4839_equation_0 = const()[name = tensor("op_4839_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4839_cast_fp16 = einsum(equation = var_4839_equation_0, values = (var_4681_cast_fp16, var_4598_cast_fp16))[name = tensor("op_4839_cast_fp16")]; + tensor var_4840_to_fp16 = const()[name = tensor("op_4840_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_325_cast_fp16 = mul(x = var_4839_cast_fp16, y = var_4840_to_fp16)[name = tensor("aw_325_cast_fp16")]; + tensor var_4843_equation_0 = const()[name = tensor("op_4843_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4843_cast_fp16 = einsum(equation = var_4843_equation_0, values = (var_4685_cast_fp16, var_4602_cast_fp16))[name = tensor("op_4843_cast_fp16")]; + tensor var_4844_to_fp16 = const()[name = tensor("op_4844_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_327_cast_fp16 = mul(x = var_4843_cast_fp16, y = var_4844_to_fp16)[name = tensor("aw_327_cast_fp16")]; + tensor var_4847_equation_0 = const()[name = tensor("op_4847_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4847_cast_fp16 = einsum(equation = var_4847_equation_0, values = (var_4689_cast_fp16, var_4606_cast_fp16))[name = tensor("op_4847_cast_fp16")]; + tensor var_4848_to_fp16 = const()[name = tensor("op_4848_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_329_cast_fp16 = mul(x = var_4847_cast_fp16, y = var_4848_to_fp16)[name = tensor("aw_329_cast_fp16")]; + tensor var_4851_equation_0 = const()[name = tensor("op_4851_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4851_cast_fp16 = einsum(equation = var_4851_equation_0, values = (var_4693_cast_fp16, var_4610_cast_fp16))[name = tensor("op_4851_cast_fp16")]; + tensor var_4852_to_fp16 = const()[name = tensor("op_4852_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_331_cast_fp16 = mul(x = var_4851_cast_fp16, y = var_4852_to_fp16)[name = tensor("aw_331_cast_fp16")]; + tensor var_4855_equation_0 = const()[name = tensor("op_4855_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4855_cast_fp16 = einsum(equation = var_4855_equation_0, values = (var_4697_cast_fp16, var_4614_cast_fp16))[name = tensor("op_4855_cast_fp16")]; + tensor var_4856_to_fp16 = const()[name = tensor("op_4856_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_333_cast_fp16 = mul(x = var_4855_cast_fp16, y = var_4856_to_fp16)[name = tensor("aw_333_cast_fp16")]; + tensor var_4859_equation_0 = const()[name = tensor("op_4859_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4859_cast_fp16 = einsum(equation = var_4859_equation_0, values = (var_4701_cast_fp16, var_4618_cast_fp16))[name = tensor("op_4859_cast_fp16")]; + tensor var_4860_to_fp16 = const()[name = tensor("op_4860_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_335_cast_fp16 = mul(x = var_4859_cast_fp16, y = var_4860_to_fp16)[name = tensor("aw_335_cast_fp16")]; + tensor var_4863_equation_0 = const()[name = tensor("op_4863_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4863_cast_fp16 = einsum(equation = var_4863_equation_0, values = (var_4705_cast_fp16, var_4622_cast_fp16))[name = tensor("op_4863_cast_fp16")]; + tensor var_4864_to_fp16 = const()[name = tensor("op_4864_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_337_cast_fp16 = mul(x = var_4863_cast_fp16, y = var_4864_to_fp16)[name = tensor("aw_337_cast_fp16")]; + tensor var_4867_equation_0 = const()[name = tensor("op_4867_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4867_cast_fp16 = einsum(equation = var_4867_equation_0, values = (var_4709_cast_fp16, var_4626_cast_fp16))[name = tensor("op_4867_cast_fp16")]; + tensor var_4868_to_fp16 = const()[name = tensor("op_4868_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_339_cast_fp16 = mul(x = var_4867_cast_fp16, y = var_4868_to_fp16)[name = tensor("aw_339_cast_fp16")]; + tensor var_4871_equation_0 = const()[name = tensor("op_4871_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4871_cast_fp16 = einsum(equation = var_4871_equation_0, values = (var_4713_cast_fp16, var_4630_cast_fp16))[name = tensor("op_4871_cast_fp16")]; + tensor var_4872_to_fp16 = const()[name = tensor("op_4872_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_341_cast_fp16 = mul(x = var_4871_cast_fp16, y = var_4872_to_fp16)[name = tensor("aw_341_cast_fp16")]; + tensor var_4875_equation_0 = const()[name = tensor("op_4875_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4875_cast_fp16 = einsum(equation = var_4875_equation_0, values = (var_4717_cast_fp16, var_4634_cast_fp16))[name = tensor("op_4875_cast_fp16")]; + tensor var_4876_to_fp16 = const()[name = tensor("op_4876_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_343_cast_fp16 = mul(x = var_4875_cast_fp16, y = var_4876_to_fp16)[name = tensor("aw_343_cast_fp16")]; + tensor var_4879_equation_0 = const()[name = tensor("op_4879_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4879_cast_fp16 = einsum(equation = var_4879_equation_0, values = (var_4721_cast_fp16, var_4638_cast_fp16))[name = tensor("op_4879_cast_fp16")]; + tensor var_4880_to_fp16 = const()[name = tensor("op_4880_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_345_cast_fp16 = mul(x = var_4879_cast_fp16, y = var_4880_to_fp16)[name = tensor("aw_345_cast_fp16")]; + tensor var_4883_equation_0 = const()[name = tensor("op_4883_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4883_cast_fp16 = einsum(equation = var_4883_equation_0, values = (var_4725_cast_fp16, var_4642_cast_fp16))[name = tensor("op_4883_cast_fp16")]; + tensor var_4884_to_fp16 = const()[name = tensor("op_4884_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_347_cast_fp16 = mul(x = var_4883_cast_fp16, y = var_4884_to_fp16)[name = tensor("aw_347_cast_fp16")]; + tensor var_4887_equation_0 = const()[name = tensor("op_4887_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4887_cast_fp16 = einsum(equation = var_4887_equation_0, values = (var_4729_cast_fp16, var_4646_cast_fp16))[name = tensor("op_4887_cast_fp16")]; + tensor var_4888_to_fp16 = const()[name = tensor("op_4888_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_349_cast_fp16 = mul(x = var_4887_cast_fp16, y = var_4888_to_fp16)[name = tensor("aw_349_cast_fp16")]; + tensor var_4891_equation_0 = const()[name = tensor("op_4891_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4891_cast_fp16 = einsum(equation = var_4891_equation_0, values = (var_4733_cast_fp16, var_4650_cast_fp16))[name = tensor("op_4891_cast_fp16")]; + tensor var_4892_to_fp16 = const()[name = tensor("op_4892_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_351_cast_fp16 = mul(x = var_4891_cast_fp16, y = var_4892_to_fp16)[name = tensor("aw_351_cast_fp16")]; + tensor var_4895_equation_0 = const()[name = tensor("op_4895_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4895_cast_fp16 = einsum(equation = var_4895_equation_0, values = (var_4737_cast_fp16, var_4654_cast_fp16))[name = tensor("op_4895_cast_fp16")]; + tensor var_4896_to_fp16 = const()[name = tensor("op_4896_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_353_cast_fp16 = mul(x = var_4895_cast_fp16, y = var_4896_to_fp16)[name = tensor("aw_353_cast_fp16")]; + tensor var_4899_equation_0 = const()[name = tensor("op_4899_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4899_cast_fp16 = einsum(equation = var_4899_equation_0, values = (var_4741_cast_fp16, var_4658_cast_fp16))[name = tensor("op_4899_cast_fp16")]; + tensor var_4900_to_fp16 = const()[name = tensor("op_4900_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_355_cast_fp16 = mul(x = var_4899_cast_fp16, y = var_4900_to_fp16)[name = tensor("aw_355_cast_fp16")]; + tensor var_4903_equation_0 = const()[name = tensor("op_4903_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4903_cast_fp16 = einsum(equation = var_4903_equation_0, values = (var_4745_cast_fp16, var_4662_cast_fp16))[name = tensor("op_4903_cast_fp16")]; + tensor var_4904_to_fp16 = const()[name = tensor("op_4904_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_357_cast_fp16 = mul(x = var_4903_cast_fp16, y = var_4904_to_fp16)[name = tensor("aw_357_cast_fp16")]; + tensor var_4907_equation_0 = const()[name = tensor("op_4907_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4907_cast_fp16 = einsum(equation = var_4907_equation_0, values = (var_4749_cast_fp16, var_4666_cast_fp16))[name = tensor("op_4907_cast_fp16")]; + tensor var_4908_to_fp16 = const()[name = tensor("op_4908_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_359_cast_fp16 = mul(x = var_4907_cast_fp16, y = var_4908_to_fp16)[name = tensor("aw_359_cast_fp16")]; + tensor var_4910_cast_fp16 = softmax(axis = var_2624, x = aw_321_cast_fp16)[name = tensor("op_4910_cast_fp16")]; + tensor var_4911_cast_fp16 = softmax(axis = var_2624, x = aw_323_cast_fp16)[name = tensor("op_4911_cast_fp16")]; + tensor var_4912_cast_fp16 = softmax(axis = var_2624, x = aw_325_cast_fp16)[name = tensor("op_4912_cast_fp16")]; + tensor var_4913_cast_fp16 = softmax(axis = var_2624, x = aw_327_cast_fp16)[name = tensor("op_4913_cast_fp16")]; + tensor var_4914_cast_fp16 = softmax(axis = var_2624, x = aw_329_cast_fp16)[name = tensor("op_4914_cast_fp16")]; + tensor var_4915_cast_fp16 = softmax(axis = var_2624, x = aw_331_cast_fp16)[name = tensor("op_4915_cast_fp16")]; + tensor var_4916_cast_fp16 = softmax(axis = var_2624, x = aw_333_cast_fp16)[name = tensor("op_4916_cast_fp16")]; + tensor var_4917_cast_fp16 = softmax(axis = var_2624, x = aw_335_cast_fp16)[name = tensor("op_4917_cast_fp16")]; + tensor var_4918_cast_fp16 = softmax(axis = var_2624, x = aw_337_cast_fp16)[name = tensor("op_4918_cast_fp16")]; + tensor var_4919_cast_fp16 = softmax(axis = var_2624, x = aw_339_cast_fp16)[name = tensor("op_4919_cast_fp16")]; + tensor var_4920_cast_fp16 = softmax(axis = var_2624, x = aw_341_cast_fp16)[name = tensor("op_4920_cast_fp16")]; + tensor var_4921_cast_fp16 = softmax(axis = var_2624, x = aw_343_cast_fp16)[name = tensor("op_4921_cast_fp16")]; + tensor var_4922_cast_fp16 = softmax(axis = var_2624, x = aw_345_cast_fp16)[name = tensor("op_4922_cast_fp16")]; + tensor var_4923_cast_fp16 = softmax(axis = var_2624, x = aw_347_cast_fp16)[name = tensor("op_4923_cast_fp16")]; + tensor var_4924_cast_fp16 = softmax(axis = var_2624, x = aw_349_cast_fp16)[name = tensor("op_4924_cast_fp16")]; + tensor var_4925_cast_fp16 = softmax(axis = var_2624, x = aw_351_cast_fp16)[name = tensor("op_4925_cast_fp16")]; + tensor var_4926_cast_fp16 = softmax(axis = var_2624, x = aw_353_cast_fp16)[name = tensor("op_4926_cast_fp16")]; + tensor var_4927_cast_fp16 = softmax(axis = var_2624, x = aw_355_cast_fp16)[name = tensor("op_4927_cast_fp16")]; + tensor var_4928_cast_fp16 = softmax(axis = var_2624, x = aw_357_cast_fp16)[name = tensor("op_4928_cast_fp16")]; + tensor var_4929_cast_fp16 = softmax(axis = var_2624, x = aw_359_cast_fp16)[name = tensor("op_4929_cast_fp16")]; + tensor var_4931_equation_0 = const()[name = tensor("op_4931_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4931_cast_fp16 = einsum(equation = var_4931_equation_0, values = (var_4751_cast_fp16, var_4910_cast_fp16))[name = tensor("op_4931_cast_fp16")]; + tensor var_4933_equation_0 = const()[name = tensor("op_4933_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4933_cast_fp16 = einsum(equation = var_4933_equation_0, values = (var_4755_cast_fp16, var_4911_cast_fp16))[name = tensor("op_4933_cast_fp16")]; + tensor var_4935_equation_0 = const()[name = tensor("op_4935_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4935_cast_fp16 = einsum(equation = var_4935_equation_0, values = (var_4759_cast_fp16, var_4912_cast_fp16))[name = tensor("op_4935_cast_fp16")]; + tensor var_4937_equation_0 = const()[name = tensor("op_4937_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4937_cast_fp16 = einsum(equation = var_4937_equation_0, values = (var_4763_cast_fp16, var_4913_cast_fp16))[name = tensor("op_4937_cast_fp16")]; + tensor var_4939_equation_0 = const()[name = tensor("op_4939_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4939_cast_fp16 = einsum(equation = var_4939_equation_0, values = (var_4767_cast_fp16, var_4914_cast_fp16))[name = tensor("op_4939_cast_fp16")]; + tensor var_4941_equation_0 = const()[name = tensor("op_4941_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4941_cast_fp16 = einsum(equation = var_4941_equation_0, values = (var_4771_cast_fp16, var_4915_cast_fp16))[name = tensor("op_4941_cast_fp16")]; + tensor var_4943_equation_0 = const()[name = tensor("op_4943_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4943_cast_fp16 = einsum(equation = var_4943_equation_0, values = (var_4775_cast_fp16, var_4916_cast_fp16))[name = tensor("op_4943_cast_fp16")]; + tensor var_4945_equation_0 = const()[name = tensor("op_4945_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4945_cast_fp16 = einsum(equation = var_4945_equation_0, values = (var_4779_cast_fp16, var_4917_cast_fp16))[name = tensor("op_4945_cast_fp16")]; + tensor var_4947_equation_0 = const()[name = tensor("op_4947_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4947_cast_fp16 = einsum(equation = var_4947_equation_0, values = (var_4783_cast_fp16, var_4918_cast_fp16))[name = tensor("op_4947_cast_fp16")]; + tensor var_4949_equation_0 = const()[name = tensor("op_4949_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4949_cast_fp16 = einsum(equation = var_4949_equation_0, values = (var_4787_cast_fp16, var_4919_cast_fp16))[name = tensor("op_4949_cast_fp16")]; + tensor var_4951_equation_0 = const()[name = tensor("op_4951_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4951_cast_fp16 = einsum(equation = var_4951_equation_0, values = (var_4791_cast_fp16, var_4920_cast_fp16))[name = tensor("op_4951_cast_fp16")]; + tensor var_4953_equation_0 = const()[name = tensor("op_4953_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4953_cast_fp16 = einsum(equation = var_4953_equation_0, values = (var_4795_cast_fp16, var_4921_cast_fp16))[name = tensor("op_4953_cast_fp16")]; + tensor var_4955_equation_0 = const()[name = tensor("op_4955_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4955_cast_fp16 = einsum(equation = var_4955_equation_0, values = (var_4799_cast_fp16, var_4922_cast_fp16))[name = tensor("op_4955_cast_fp16")]; + tensor var_4957_equation_0 = const()[name = tensor("op_4957_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4957_cast_fp16 = einsum(equation = var_4957_equation_0, values = (var_4803_cast_fp16, var_4923_cast_fp16))[name = tensor("op_4957_cast_fp16")]; + tensor var_4959_equation_0 = const()[name = tensor("op_4959_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4959_cast_fp16 = einsum(equation = var_4959_equation_0, values = (var_4807_cast_fp16, var_4924_cast_fp16))[name = tensor("op_4959_cast_fp16")]; + tensor var_4961_equation_0 = const()[name = tensor("op_4961_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4961_cast_fp16 = einsum(equation = var_4961_equation_0, values = (var_4811_cast_fp16, var_4925_cast_fp16))[name = tensor("op_4961_cast_fp16")]; + tensor var_4963_equation_0 = const()[name = tensor("op_4963_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4963_cast_fp16 = einsum(equation = var_4963_equation_0, values = (var_4815_cast_fp16, var_4926_cast_fp16))[name = tensor("op_4963_cast_fp16")]; + tensor var_4965_equation_0 = const()[name = tensor("op_4965_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4965_cast_fp16 = einsum(equation = var_4965_equation_0, values = (var_4819_cast_fp16, var_4927_cast_fp16))[name = tensor("op_4965_cast_fp16")]; + tensor var_4967_equation_0 = const()[name = tensor("op_4967_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4967_cast_fp16 = einsum(equation = var_4967_equation_0, values = (var_4823_cast_fp16, var_4928_cast_fp16))[name = tensor("op_4967_cast_fp16")]; + tensor var_4969_equation_0 = const()[name = tensor("op_4969_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4969_cast_fp16 = einsum(equation = var_4969_equation_0, values = (var_4827_cast_fp16, var_4929_cast_fp16))[name = tensor("op_4969_cast_fp16")]; + tensor input_147_interleave_0 = const()[name = tensor("input_147_interleave_0"), val = tensor(false)]; + tensor input_147_cast_fp16 = concat(axis = var_2624, interleave = input_147_interleave_0, values = (var_4931_cast_fp16, var_4933_cast_fp16, var_4935_cast_fp16, var_4937_cast_fp16, var_4939_cast_fp16, var_4941_cast_fp16, var_4943_cast_fp16, var_4945_cast_fp16, var_4947_cast_fp16, var_4949_cast_fp16, var_4951_cast_fp16, var_4953_cast_fp16, var_4955_cast_fp16, var_4957_cast_fp16, var_4959_cast_fp16, var_4961_cast_fp16, var_4963_cast_fp16, var_4965_cast_fp16, var_4967_cast_fp16, var_4969_cast_fp16))[name = tensor("input_147_cast_fp16")]; + tensor var_4975 = const()[name = tensor("op_4975"), val = tensor([1, 1])]; + tensor var_4977 = const()[name = tensor("op_4977"), val = tensor([1, 1])]; + tensor var_4979_pad_type_0 = const()[name = tensor("op_4979_pad_type_0"), val = tensor("custom")]; + tensor var_4979_pad_0 = const()[name = tensor("op_4979_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_2_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_2_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(347027648)))]; + tensor down_blocks_2_attentions_0_transformer_blocks_2_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_2_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(350304512)))]; + tensor var_4979_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_2_attn1_to_out_0_bias_to_fp16, dilations = var_4977, groups = var_2624, pad = var_4979_pad_0, pad_type = var_4979_pad_type_0, strides = var_4975, weight = down_blocks_2_attentions_0_transformer_blocks_2_attn1_to_out_0_weight_to_fp16, x = input_147_cast_fp16)[name = tensor("op_4979_cast_fp16")]; + tensor inputs_39_cast_fp16 = add(x = var_4979_cast_fp16, y = inputs_37_cast_fp16)[name = tensor("inputs_39_cast_fp16")]; + tensor hidden_states_79_axes_0 = const()[name = tensor("hidden_states_79_axes_0"), val = tensor([1])]; + tensor hidden_states_79_gamma_0_to_fp16 = const()[name = tensor("hidden_states_79_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(350307136)))]; + tensor hidden_states_79_beta_0_to_fp16 = const()[name = tensor("hidden_states_79_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(350309760)))]; + tensor var_4989_to_fp16 = const()[name = tensor("op_4989_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_79_cast_fp16 = layer_norm(axes = hidden_states_79_axes_0, beta = hidden_states_79_beta_0_to_fp16, epsilon = var_4989_to_fp16, gamma = hidden_states_79_gamma_0_to_fp16, x = inputs_39_cast_fp16)[name = tensor("hidden_states_79_cast_fp16")]; + tensor var_5004 = const()[name = tensor("op_5004"), val = tensor([1, 1])]; + tensor var_5006 = const()[name = tensor("op_5006"), val = tensor([1, 1])]; + tensor q_27_pad_type_0 = const()[name = tensor("q_27_pad_type_0"), val = tensor("custom")]; + tensor q_27_pad_0 = const()[name = tensor("q_27_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_2_attn2_to_q_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_2_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(350312384)))]; + tensor q_27_cast_fp16 = conv(dilations = var_5006, groups = var_2624, pad = q_27_pad_0, pad_type = q_27_pad_type_0, strides = var_5004, weight = down_blocks_2_attentions_0_transformer_blocks_2_attn2_to_q_weight_to_fp16, x = hidden_states_79_cast_fp16)[name = tensor("q_27_cast_fp16")]; + tensor var_5010 = const()[name = tensor("op_5010"), val = tensor([1, 1])]; + tensor var_5012 = const()[name = tensor("op_5012"), val = tensor([1, 1])]; + tensor k_53_pad_type_0 = const()[name = tensor("k_53_pad_type_0"), val = tensor("custom")]; + tensor k_53_pad_0 = const()[name = tensor("k_53_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_2_attn2_to_k_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_2_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(353589248)))]; + tensor k_53_cast_fp16 = conv(dilations = var_5012, groups = var_2624, pad = k_53_pad_0, pad_type = k_53_pad_type_0, strides = var_5010, weight = down_blocks_2_attentions_0_transformer_blocks_2_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_53_cast_fp16")]; + tensor var_5016 = const()[name = tensor("op_5016"), val = tensor([1, 1])]; + tensor var_5018 = const()[name = tensor("op_5018"), val = tensor([1, 1])]; + tensor v_27_pad_type_0 = const()[name = tensor("v_27_pad_type_0"), val = tensor("custom")]; + tensor v_27_pad_0 = const()[name = tensor("v_27_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_2_attn2_to_v_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_2_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(358832192)))]; + tensor v_27_cast_fp16 = conv(dilations = var_5018, groups = var_2624, pad = v_27_pad_0, pad_type = v_27_pad_type_0, strides = var_5016, weight = down_blocks_2_attentions_0_transformer_blocks_2_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_27_cast_fp16")]; + tensor var_5022_begin_0 = const()[name = tensor("op_5022_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5022_end_0 = const()[name = tensor("op_5022_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_5022_end_mask_0 = const()[name = tensor("op_5022_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5022_cast_fp16 = slice_by_index(begin = var_5022_begin_0, end = var_5022_end_0, end_mask = var_5022_end_mask_0, x = q_27_cast_fp16)[name = tensor("op_5022_cast_fp16")]; + tensor var_5026_begin_0 = const()[name = tensor("op_5026_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_5026_end_0 = const()[name = tensor("op_5026_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_5026_end_mask_0 = const()[name = tensor("op_5026_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5026_cast_fp16 = slice_by_index(begin = var_5026_begin_0, end = var_5026_end_0, end_mask = var_5026_end_mask_0, x = q_27_cast_fp16)[name = tensor("op_5026_cast_fp16")]; + tensor var_5030_begin_0 = const()[name = tensor("op_5030_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_5030_end_0 = const()[name = tensor("op_5030_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_5030_end_mask_0 = const()[name = tensor("op_5030_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5030_cast_fp16 = slice_by_index(begin = var_5030_begin_0, end = var_5030_end_0, end_mask = var_5030_end_mask_0, x = q_27_cast_fp16)[name = tensor("op_5030_cast_fp16")]; + tensor var_5034_begin_0 = const()[name = tensor("op_5034_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_5034_end_0 = const()[name = tensor("op_5034_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_5034_end_mask_0 = const()[name = tensor("op_5034_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5034_cast_fp16 = slice_by_index(begin = var_5034_begin_0, end = var_5034_end_0, end_mask = var_5034_end_mask_0, x = q_27_cast_fp16)[name = tensor("op_5034_cast_fp16")]; + tensor var_5038_begin_0 = const()[name = tensor("op_5038_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_5038_end_0 = const()[name = tensor("op_5038_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_5038_end_mask_0 = const()[name = tensor("op_5038_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5038_cast_fp16 = slice_by_index(begin = var_5038_begin_0, end = var_5038_end_0, end_mask = var_5038_end_mask_0, x = q_27_cast_fp16)[name = tensor("op_5038_cast_fp16")]; + tensor var_5042_begin_0 = const()[name = tensor("op_5042_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_5042_end_0 = const()[name = tensor("op_5042_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_5042_end_mask_0 = const()[name = tensor("op_5042_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5042_cast_fp16 = slice_by_index(begin = var_5042_begin_0, end = var_5042_end_0, end_mask = var_5042_end_mask_0, x = q_27_cast_fp16)[name = tensor("op_5042_cast_fp16")]; + tensor var_5046_begin_0 = const()[name = tensor("op_5046_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_5046_end_0 = const()[name = tensor("op_5046_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_5046_end_mask_0 = const()[name = tensor("op_5046_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5046_cast_fp16 = slice_by_index(begin = var_5046_begin_0, end = var_5046_end_0, end_mask = var_5046_end_mask_0, x = q_27_cast_fp16)[name = tensor("op_5046_cast_fp16")]; + tensor var_5050_begin_0 = const()[name = tensor("op_5050_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_5050_end_0 = const()[name = tensor("op_5050_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_5050_end_mask_0 = const()[name = tensor("op_5050_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5050_cast_fp16 = slice_by_index(begin = var_5050_begin_0, end = var_5050_end_0, end_mask = var_5050_end_mask_0, x = q_27_cast_fp16)[name = tensor("op_5050_cast_fp16")]; + tensor var_5054_begin_0 = const()[name = tensor("op_5054_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_5054_end_0 = const()[name = tensor("op_5054_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_5054_end_mask_0 = const()[name = tensor("op_5054_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5054_cast_fp16 = slice_by_index(begin = var_5054_begin_0, end = var_5054_end_0, end_mask = var_5054_end_mask_0, x = q_27_cast_fp16)[name = tensor("op_5054_cast_fp16")]; + tensor var_5058_begin_0 = const()[name = tensor("op_5058_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_5058_end_0 = const()[name = tensor("op_5058_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_5058_end_mask_0 = const()[name = tensor("op_5058_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5058_cast_fp16 = slice_by_index(begin = var_5058_begin_0, end = var_5058_end_0, end_mask = var_5058_end_mask_0, x = q_27_cast_fp16)[name = tensor("op_5058_cast_fp16")]; + tensor var_5062_begin_0 = const()[name = tensor("op_5062_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_5062_end_0 = const()[name = tensor("op_5062_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_5062_end_mask_0 = const()[name = tensor("op_5062_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5062_cast_fp16 = slice_by_index(begin = var_5062_begin_0, end = var_5062_end_0, end_mask = var_5062_end_mask_0, x = q_27_cast_fp16)[name = tensor("op_5062_cast_fp16")]; + tensor var_5066_begin_0 = const()[name = tensor("op_5066_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_5066_end_0 = const()[name = tensor("op_5066_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_5066_end_mask_0 = const()[name = tensor("op_5066_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5066_cast_fp16 = slice_by_index(begin = var_5066_begin_0, end = var_5066_end_0, end_mask = var_5066_end_mask_0, x = q_27_cast_fp16)[name = tensor("op_5066_cast_fp16")]; + tensor var_5070_begin_0 = const()[name = tensor("op_5070_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_5070_end_0 = const()[name = tensor("op_5070_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_5070_end_mask_0 = const()[name = tensor("op_5070_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5070_cast_fp16 = slice_by_index(begin = var_5070_begin_0, end = var_5070_end_0, end_mask = var_5070_end_mask_0, x = q_27_cast_fp16)[name = tensor("op_5070_cast_fp16")]; + tensor var_5074_begin_0 = const()[name = tensor("op_5074_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_5074_end_0 = const()[name = tensor("op_5074_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_5074_end_mask_0 = const()[name = tensor("op_5074_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5074_cast_fp16 = slice_by_index(begin = var_5074_begin_0, end = var_5074_end_0, end_mask = var_5074_end_mask_0, x = q_27_cast_fp16)[name = tensor("op_5074_cast_fp16")]; + tensor var_5078_begin_0 = const()[name = tensor("op_5078_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_5078_end_0 = const()[name = tensor("op_5078_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_5078_end_mask_0 = const()[name = tensor("op_5078_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5078_cast_fp16 = slice_by_index(begin = var_5078_begin_0, end = var_5078_end_0, end_mask = var_5078_end_mask_0, x = q_27_cast_fp16)[name = tensor("op_5078_cast_fp16")]; + tensor var_5082_begin_0 = const()[name = tensor("op_5082_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_5082_end_0 = const()[name = tensor("op_5082_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_5082_end_mask_0 = const()[name = tensor("op_5082_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5082_cast_fp16 = slice_by_index(begin = var_5082_begin_0, end = var_5082_end_0, end_mask = var_5082_end_mask_0, x = q_27_cast_fp16)[name = tensor("op_5082_cast_fp16")]; + tensor var_5086_begin_0 = const()[name = tensor("op_5086_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_5086_end_0 = const()[name = tensor("op_5086_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_5086_end_mask_0 = const()[name = tensor("op_5086_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5086_cast_fp16 = slice_by_index(begin = var_5086_begin_0, end = var_5086_end_0, end_mask = var_5086_end_mask_0, x = q_27_cast_fp16)[name = tensor("op_5086_cast_fp16")]; + tensor var_5090_begin_0 = const()[name = tensor("op_5090_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_5090_end_0 = const()[name = tensor("op_5090_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_5090_end_mask_0 = const()[name = tensor("op_5090_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5090_cast_fp16 = slice_by_index(begin = var_5090_begin_0, end = var_5090_end_0, end_mask = var_5090_end_mask_0, x = q_27_cast_fp16)[name = tensor("op_5090_cast_fp16")]; + tensor var_5094_begin_0 = const()[name = tensor("op_5094_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_5094_end_0 = const()[name = tensor("op_5094_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_5094_end_mask_0 = const()[name = tensor("op_5094_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5094_cast_fp16 = slice_by_index(begin = var_5094_begin_0, end = var_5094_end_0, end_mask = var_5094_end_mask_0, x = q_27_cast_fp16)[name = tensor("op_5094_cast_fp16")]; + tensor var_5098_begin_0 = const()[name = tensor("op_5098_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_5098_end_0 = const()[name = tensor("op_5098_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_5098_end_mask_0 = const()[name = tensor("op_5098_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5098_cast_fp16 = slice_by_index(begin = var_5098_begin_0, end = var_5098_end_0, end_mask = var_5098_end_mask_0, x = q_27_cast_fp16)[name = tensor("op_5098_cast_fp16")]; + tensor k_55_perm_0 = const()[name = tensor("k_55_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_5105_begin_0 = const()[name = tensor("op_5105_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5105_end_0 = const()[name = tensor("op_5105_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_5105_end_mask_0 = const()[name = tensor("op_5105_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_55_cast_fp16 = transpose(perm = k_55_perm_0, x = k_53_cast_fp16)[name = tensor("transpose_126")]; + tensor var_5105_cast_fp16 = slice_by_index(begin = var_5105_begin_0, end = var_5105_end_0, end_mask = var_5105_end_mask_0, x = k_55_cast_fp16)[name = tensor("op_5105_cast_fp16")]; + tensor var_5109_begin_0 = const()[name = tensor("op_5109_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_5109_end_0 = const()[name = tensor("op_5109_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_5109_end_mask_0 = const()[name = tensor("op_5109_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5109_cast_fp16 = slice_by_index(begin = var_5109_begin_0, end = var_5109_end_0, end_mask = var_5109_end_mask_0, x = k_55_cast_fp16)[name = tensor("op_5109_cast_fp16")]; + tensor var_5113_begin_0 = const()[name = tensor("op_5113_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_5113_end_0 = const()[name = tensor("op_5113_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_5113_end_mask_0 = const()[name = tensor("op_5113_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5113_cast_fp16 = slice_by_index(begin = var_5113_begin_0, end = var_5113_end_0, end_mask = var_5113_end_mask_0, x = k_55_cast_fp16)[name = tensor("op_5113_cast_fp16")]; + tensor var_5117_begin_0 = const()[name = tensor("op_5117_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_5117_end_0 = const()[name = tensor("op_5117_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_5117_end_mask_0 = const()[name = tensor("op_5117_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5117_cast_fp16 = slice_by_index(begin = var_5117_begin_0, end = var_5117_end_0, end_mask = var_5117_end_mask_0, x = k_55_cast_fp16)[name = tensor("op_5117_cast_fp16")]; + tensor var_5121_begin_0 = const()[name = tensor("op_5121_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_5121_end_0 = const()[name = tensor("op_5121_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_5121_end_mask_0 = const()[name = tensor("op_5121_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5121_cast_fp16 = slice_by_index(begin = var_5121_begin_0, end = var_5121_end_0, end_mask = var_5121_end_mask_0, x = k_55_cast_fp16)[name = tensor("op_5121_cast_fp16")]; + tensor var_5125_begin_0 = const()[name = tensor("op_5125_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_5125_end_0 = const()[name = tensor("op_5125_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_5125_end_mask_0 = const()[name = tensor("op_5125_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5125_cast_fp16 = slice_by_index(begin = var_5125_begin_0, end = var_5125_end_0, end_mask = var_5125_end_mask_0, x = k_55_cast_fp16)[name = tensor("op_5125_cast_fp16")]; + tensor var_5129_begin_0 = const()[name = tensor("op_5129_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_5129_end_0 = const()[name = tensor("op_5129_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_5129_end_mask_0 = const()[name = tensor("op_5129_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5129_cast_fp16 = slice_by_index(begin = var_5129_begin_0, end = var_5129_end_0, end_mask = var_5129_end_mask_0, x = k_55_cast_fp16)[name = tensor("op_5129_cast_fp16")]; + tensor var_5133_begin_0 = const()[name = tensor("op_5133_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_5133_end_0 = const()[name = tensor("op_5133_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_5133_end_mask_0 = const()[name = tensor("op_5133_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5133_cast_fp16 = slice_by_index(begin = var_5133_begin_0, end = var_5133_end_0, end_mask = var_5133_end_mask_0, x = k_55_cast_fp16)[name = tensor("op_5133_cast_fp16")]; + tensor var_5137_begin_0 = const()[name = tensor("op_5137_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_5137_end_0 = const()[name = tensor("op_5137_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_5137_end_mask_0 = const()[name = tensor("op_5137_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5137_cast_fp16 = slice_by_index(begin = var_5137_begin_0, end = var_5137_end_0, end_mask = var_5137_end_mask_0, x = k_55_cast_fp16)[name = tensor("op_5137_cast_fp16")]; + tensor var_5141_begin_0 = const()[name = tensor("op_5141_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_5141_end_0 = const()[name = tensor("op_5141_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_5141_end_mask_0 = const()[name = tensor("op_5141_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5141_cast_fp16 = slice_by_index(begin = var_5141_begin_0, end = var_5141_end_0, end_mask = var_5141_end_mask_0, x = k_55_cast_fp16)[name = tensor("op_5141_cast_fp16")]; + tensor var_5145_begin_0 = const()[name = tensor("op_5145_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_5145_end_0 = const()[name = tensor("op_5145_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_5145_end_mask_0 = const()[name = tensor("op_5145_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5145_cast_fp16 = slice_by_index(begin = var_5145_begin_0, end = var_5145_end_0, end_mask = var_5145_end_mask_0, x = k_55_cast_fp16)[name = tensor("op_5145_cast_fp16")]; + tensor var_5149_begin_0 = const()[name = tensor("op_5149_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_5149_end_0 = const()[name = tensor("op_5149_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_5149_end_mask_0 = const()[name = tensor("op_5149_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5149_cast_fp16 = slice_by_index(begin = var_5149_begin_0, end = var_5149_end_0, end_mask = var_5149_end_mask_0, x = k_55_cast_fp16)[name = tensor("op_5149_cast_fp16")]; + tensor var_5153_begin_0 = const()[name = tensor("op_5153_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_5153_end_0 = const()[name = tensor("op_5153_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_5153_end_mask_0 = const()[name = tensor("op_5153_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5153_cast_fp16 = slice_by_index(begin = var_5153_begin_0, end = var_5153_end_0, end_mask = var_5153_end_mask_0, x = k_55_cast_fp16)[name = tensor("op_5153_cast_fp16")]; + tensor var_5157_begin_0 = const()[name = tensor("op_5157_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_5157_end_0 = const()[name = tensor("op_5157_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_5157_end_mask_0 = const()[name = tensor("op_5157_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5157_cast_fp16 = slice_by_index(begin = var_5157_begin_0, end = var_5157_end_0, end_mask = var_5157_end_mask_0, x = k_55_cast_fp16)[name = tensor("op_5157_cast_fp16")]; + tensor var_5161_begin_0 = const()[name = tensor("op_5161_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_5161_end_0 = const()[name = tensor("op_5161_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_5161_end_mask_0 = const()[name = tensor("op_5161_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5161_cast_fp16 = slice_by_index(begin = var_5161_begin_0, end = var_5161_end_0, end_mask = var_5161_end_mask_0, x = k_55_cast_fp16)[name = tensor("op_5161_cast_fp16")]; + tensor var_5165_begin_0 = const()[name = tensor("op_5165_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_5165_end_0 = const()[name = tensor("op_5165_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_5165_end_mask_0 = const()[name = tensor("op_5165_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5165_cast_fp16 = slice_by_index(begin = var_5165_begin_0, end = var_5165_end_0, end_mask = var_5165_end_mask_0, x = k_55_cast_fp16)[name = tensor("op_5165_cast_fp16")]; + tensor var_5169_begin_0 = const()[name = tensor("op_5169_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_5169_end_0 = const()[name = tensor("op_5169_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_5169_end_mask_0 = const()[name = tensor("op_5169_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5169_cast_fp16 = slice_by_index(begin = var_5169_begin_0, end = var_5169_end_0, end_mask = var_5169_end_mask_0, x = k_55_cast_fp16)[name = tensor("op_5169_cast_fp16")]; + tensor var_5173_begin_0 = const()[name = tensor("op_5173_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_5173_end_0 = const()[name = tensor("op_5173_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_5173_end_mask_0 = const()[name = tensor("op_5173_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5173_cast_fp16 = slice_by_index(begin = var_5173_begin_0, end = var_5173_end_0, end_mask = var_5173_end_mask_0, x = k_55_cast_fp16)[name = tensor("op_5173_cast_fp16")]; + tensor var_5177_begin_0 = const()[name = tensor("op_5177_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_5177_end_0 = const()[name = tensor("op_5177_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_5177_end_mask_0 = const()[name = tensor("op_5177_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5177_cast_fp16 = slice_by_index(begin = var_5177_begin_0, end = var_5177_end_0, end_mask = var_5177_end_mask_0, x = k_55_cast_fp16)[name = tensor("op_5177_cast_fp16")]; + tensor var_5181_begin_0 = const()[name = tensor("op_5181_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_5181_end_0 = const()[name = tensor("op_5181_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_5181_end_mask_0 = const()[name = tensor("op_5181_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5181_cast_fp16 = slice_by_index(begin = var_5181_begin_0, end = var_5181_end_0, end_mask = var_5181_end_mask_0, x = k_55_cast_fp16)[name = tensor("op_5181_cast_fp16")]; + tensor var_5183_begin_0 = const()[name = tensor("op_5183_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5183_end_0 = const()[name = tensor("op_5183_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_5183_end_mask_0 = const()[name = tensor("op_5183_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5183_cast_fp16 = slice_by_index(begin = var_5183_begin_0, end = var_5183_end_0, end_mask = var_5183_end_mask_0, x = v_27_cast_fp16)[name = tensor("op_5183_cast_fp16")]; + tensor var_5187_begin_0 = const()[name = tensor("op_5187_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_5187_end_0 = const()[name = tensor("op_5187_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_5187_end_mask_0 = const()[name = tensor("op_5187_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5187_cast_fp16 = slice_by_index(begin = var_5187_begin_0, end = var_5187_end_0, end_mask = var_5187_end_mask_0, x = v_27_cast_fp16)[name = tensor("op_5187_cast_fp16")]; + tensor var_5191_begin_0 = const()[name = tensor("op_5191_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_5191_end_0 = const()[name = tensor("op_5191_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_5191_end_mask_0 = const()[name = tensor("op_5191_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5191_cast_fp16 = slice_by_index(begin = var_5191_begin_0, end = var_5191_end_0, end_mask = var_5191_end_mask_0, x = v_27_cast_fp16)[name = tensor("op_5191_cast_fp16")]; + tensor var_5195_begin_0 = const()[name = tensor("op_5195_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_5195_end_0 = const()[name = tensor("op_5195_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_5195_end_mask_0 = const()[name = tensor("op_5195_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5195_cast_fp16 = slice_by_index(begin = var_5195_begin_0, end = var_5195_end_0, end_mask = var_5195_end_mask_0, x = v_27_cast_fp16)[name = tensor("op_5195_cast_fp16")]; + tensor var_5199_begin_0 = const()[name = tensor("op_5199_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_5199_end_0 = const()[name = tensor("op_5199_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_5199_end_mask_0 = const()[name = tensor("op_5199_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5199_cast_fp16 = slice_by_index(begin = var_5199_begin_0, end = var_5199_end_0, end_mask = var_5199_end_mask_0, x = v_27_cast_fp16)[name = tensor("op_5199_cast_fp16")]; + tensor var_5203_begin_0 = const()[name = tensor("op_5203_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_5203_end_0 = const()[name = tensor("op_5203_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_5203_end_mask_0 = const()[name = tensor("op_5203_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5203_cast_fp16 = slice_by_index(begin = var_5203_begin_0, end = var_5203_end_0, end_mask = var_5203_end_mask_0, x = v_27_cast_fp16)[name = tensor("op_5203_cast_fp16")]; + tensor var_5207_begin_0 = const()[name = tensor("op_5207_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_5207_end_0 = const()[name = tensor("op_5207_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_5207_end_mask_0 = const()[name = tensor("op_5207_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5207_cast_fp16 = slice_by_index(begin = var_5207_begin_0, end = var_5207_end_0, end_mask = var_5207_end_mask_0, x = v_27_cast_fp16)[name = tensor("op_5207_cast_fp16")]; + tensor var_5211_begin_0 = const()[name = tensor("op_5211_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_5211_end_0 = const()[name = tensor("op_5211_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_5211_end_mask_0 = const()[name = tensor("op_5211_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5211_cast_fp16 = slice_by_index(begin = var_5211_begin_0, end = var_5211_end_0, end_mask = var_5211_end_mask_0, x = v_27_cast_fp16)[name = tensor("op_5211_cast_fp16")]; + tensor var_5215_begin_0 = const()[name = tensor("op_5215_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_5215_end_0 = const()[name = tensor("op_5215_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_5215_end_mask_0 = const()[name = tensor("op_5215_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5215_cast_fp16 = slice_by_index(begin = var_5215_begin_0, end = var_5215_end_0, end_mask = var_5215_end_mask_0, x = v_27_cast_fp16)[name = tensor("op_5215_cast_fp16")]; + tensor var_5219_begin_0 = const()[name = tensor("op_5219_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_5219_end_0 = const()[name = tensor("op_5219_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_5219_end_mask_0 = const()[name = tensor("op_5219_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5219_cast_fp16 = slice_by_index(begin = var_5219_begin_0, end = var_5219_end_0, end_mask = var_5219_end_mask_0, x = v_27_cast_fp16)[name = tensor("op_5219_cast_fp16")]; + tensor var_5223_begin_0 = const()[name = tensor("op_5223_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_5223_end_0 = const()[name = tensor("op_5223_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_5223_end_mask_0 = const()[name = tensor("op_5223_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5223_cast_fp16 = slice_by_index(begin = var_5223_begin_0, end = var_5223_end_0, end_mask = var_5223_end_mask_0, x = v_27_cast_fp16)[name = tensor("op_5223_cast_fp16")]; + tensor var_5227_begin_0 = const()[name = tensor("op_5227_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_5227_end_0 = const()[name = tensor("op_5227_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_5227_end_mask_0 = const()[name = tensor("op_5227_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5227_cast_fp16 = slice_by_index(begin = var_5227_begin_0, end = var_5227_end_0, end_mask = var_5227_end_mask_0, x = v_27_cast_fp16)[name = tensor("op_5227_cast_fp16")]; + tensor var_5231_begin_0 = const()[name = tensor("op_5231_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_5231_end_0 = const()[name = tensor("op_5231_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_5231_end_mask_0 = const()[name = tensor("op_5231_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5231_cast_fp16 = slice_by_index(begin = var_5231_begin_0, end = var_5231_end_0, end_mask = var_5231_end_mask_0, x = v_27_cast_fp16)[name = tensor("op_5231_cast_fp16")]; + tensor var_5235_begin_0 = const()[name = tensor("op_5235_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_5235_end_0 = const()[name = tensor("op_5235_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_5235_end_mask_0 = const()[name = tensor("op_5235_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5235_cast_fp16 = slice_by_index(begin = var_5235_begin_0, end = var_5235_end_0, end_mask = var_5235_end_mask_0, x = v_27_cast_fp16)[name = tensor("op_5235_cast_fp16")]; + tensor var_5239_begin_0 = const()[name = tensor("op_5239_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_5239_end_0 = const()[name = tensor("op_5239_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_5239_end_mask_0 = const()[name = tensor("op_5239_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5239_cast_fp16 = slice_by_index(begin = var_5239_begin_0, end = var_5239_end_0, end_mask = var_5239_end_mask_0, x = v_27_cast_fp16)[name = tensor("op_5239_cast_fp16")]; + tensor var_5243_begin_0 = const()[name = tensor("op_5243_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_5243_end_0 = const()[name = tensor("op_5243_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_5243_end_mask_0 = const()[name = tensor("op_5243_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5243_cast_fp16 = slice_by_index(begin = var_5243_begin_0, end = var_5243_end_0, end_mask = var_5243_end_mask_0, x = v_27_cast_fp16)[name = tensor("op_5243_cast_fp16")]; + tensor var_5247_begin_0 = const()[name = tensor("op_5247_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_5247_end_0 = const()[name = tensor("op_5247_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_5247_end_mask_0 = const()[name = tensor("op_5247_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5247_cast_fp16 = slice_by_index(begin = var_5247_begin_0, end = var_5247_end_0, end_mask = var_5247_end_mask_0, x = v_27_cast_fp16)[name = tensor("op_5247_cast_fp16")]; + tensor var_5251_begin_0 = const()[name = tensor("op_5251_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_5251_end_0 = const()[name = tensor("op_5251_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_5251_end_mask_0 = const()[name = tensor("op_5251_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5251_cast_fp16 = slice_by_index(begin = var_5251_begin_0, end = var_5251_end_0, end_mask = var_5251_end_mask_0, x = v_27_cast_fp16)[name = tensor("op_5251_cast_fp16")]; + tensor var_5255_begin_0 = const()[name = tensor("op_5255_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_5255_end_0 = const()[name = tensor("op_5255_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_5255_end_mask_0 = const()[name = tensor("op_5255_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5255_cast_fp16 = slice_by_index(begin = var_5255_begin_0, end = var_5255_end_0, end_mask = var_5255_end_mask_0, x = v_27_cast_fp16)[name = tensor("op_5255_cast_fp16")]; + tensor var_5259_begin_0 = const()[name = tensor("op_5259_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_5259_end_0 = const()[name = tensor("op_5259_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_5259_end_mask_0 = const()[name = tensor("op_5259_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5259_cast_fp16 = slice_by_index(begin = var_5259_begin_0, end = var_5259_end_0, end_mask = var_5259_end_mask_0, x = v_27_cast_fp16)[name = tensor("op_5259_cast_fp16")]; + tensor var_5263_equation_0 = const()[name = tensor("op_5263_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5263_cast_fp16 = einsum(equation = var_5263_equation_0, values = (var_5105_cast_fp16, var_5022_cast_fp16))[name = tensor("op_5263_cast_fp16")]; + tensor var_5264_to_fp16 = const()[name = tensor("op_5264_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_361_cast_fp16 = mul(x = var_5263_cast_fp16, y = var_5264_to_fp16)[name = tensor("aw_361_cast_fp16")]; + tensor var_5267_equation_0 = const()[name = tensor("op_5267_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5267_cast_fp16 = einsum(equation = var_5267_equation_0, values = (var_5109_cast_fp16, var_5026_cast_fp16))[name = tensor("op_5267_cast_fp16")]; + tensor var_5268_to_fp16 = const()[name = tensor("op_5268_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_363_cast_fp16 = mul(x = var_5267_cast_fp16, y = var_5268_to_fp16)[name = tensor("aw_363_cast_fp16")]; + tensor var_5271_equation_0 = const()[name = tensor("op_5271_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5271_cast_fp16 = einsum(equation = var_5271_equation_0, values = (var_5113_cast_fp16, var_5030_cast_fp16))[name = tensor("op_5271_cast_fp16")]; + tensor var_5272_to_fp16 = const()[name = tensor("op_5272_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_365_cast_fp16 = mul(x = var_5271_cast_fp16, y = var_5272_to_fp16)[name = tensor("aw_365_cast_fp16")]; + tensor var_5275_equation_0 = const()[name = tensor("op_5275_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5275_cast_fp16 = einsum(equation = var_5275_equation_0, values = (var_5117_cast_fp16, var_5034_cast_fp16))[name = tensor("op_5275_cast_fp16")]; + tensor var_5276_to_fp16 = const()[name = tensor("op_5276_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_367_cast_fp16 = mul(x = var_5275_cast_fp16, y = var_5276_to_fp16)[name = tensor("aw_367_cast_fp16")]; + tensor var_5279_equation_0 = const()[name = tensor("op_5279_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5279_cast_fp16 = einsum(equation = var_5279_equation_0, values = (var_5121_cast_fp16, var_5038_cast_fp16))[name = tensor("op_5279_cast_fp16")]; + tensor var_5280_to_fp16 = const()[name = tensor("op_5280_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_369_cast_fp16 = mul(x = var_5279_cast_fp16, y = var_5280_to_fp16)[name = tensor("aw_369_cast_fp16")]; + tensor var_5283_equation_0 = const()[name = tensor("op_5283_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5283_cast_fp16 = einsum(equation = var_5283_equation_0, values = (var_5125_cast_fp16, var_5042_cast_fp16))[name = tensor("op_5283_cast_fp16")]; + tensor var_5284_to_fp16 = const()[name = tensor("op_5284_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_371_cast_fp16 = mul(x = var_5283_cast_fp16, y = var_5284_to_fp16)[name = tensor("aw_371_cast_fp16")]; + tensor var_5287_equation_0 = const()[name = tensor("op_5287_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5287_cast_fp16 = einsum(equation = var_5287_equation_0, values = (var_5129_cast_fp16, var_5046_cast_fp16))[name = tensor("op_5287_cast_fp16")]; + tensor var_5288_to_fp16 = const()[name = tensor("op_5288_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_373_cast_fp16 = mul(x = var_5287_cast_fp16, y = var_5288_to_fp16)[name = tensor("aw_373_cast_fp16")]; + tensor var_5291_equation_0 = const()[name = tensor("op_5291_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5291_cast_fp16 = einsum(equation = var_5291_equation_0, values = (var_5133_cast_fp16, var_5050_cast_fp16))[name = tensor("op_5291_cast_fp16")]; + tensor var_5292_to_fp16 = const()[name = tensor("op_5292_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_375_cast_fp16 = mul(x = var_5291_cast_fp16, y = var_5292_to_fp16)[name = tensor("aw_375_cast_fp16")]; + tensor var_5295_equation_0 = const()[name = tensor("op_5295_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5295_cast_fp16 = einsum(equation = var_5295_equation_0, values = (var_5137_cast_fp16, var_5054_cast_fp16))[name = tensor("op_5295_cast_fp16")]; + tensor var_5296_to_fp16 = const()[name = tensor("op_5296_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_377_cast_fp16 = mul(x = var_5295_cast_fp16, y = var_5296_to_fp16)[name = tensor("aw_377_cast_fp16")]; + tensor var_5299_equation_0 = const()[name = tensor("op_5299_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5299_cast_fp16 = einsum(equation = var_5299_equation_0, values = (var_5141_cast_fp16, var_5058_cast_fp16))[name = tensor("op_5299_cast_fp16")]; + tensor var_5300_to_fp16 = const()[name = tensor("op_5300_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_379_cast_fp16 = mul(x = var_5299_cast_fp16, y = var_5300_to_fp16)[name = tensor("aw_379_cast_fp16")]; + tensor var_5303_equation_0 = const()[name = tensor("op_5303_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5303_cast_fp16 = einsum(equation = var_5303_equation_0, values = (var_5145_cast_fp16, var_5062_cast_fp16))[name = tensor("op_5303_cast_fp16")]; + tensor var_5304_to_fp16 = const()[name = tensor("op_5304_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_381_cast_fp16 = mul(x = var_5303_cast_fp16, y = var_5304_to_fp16)[name = tensor("aw_381_cast_fp16")]; + tensor var_5307_equation_0 = const()[name = tensor("op_5307_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5307_cast_fp16 = einsum(equation = var_5307_equation_0, values = (var_5149_cast_fp16, var_5066_cast_fp16))[name = tensor("op_5307_cast_fp16")]; + tensor var_5308_to_fp16 = const()[name = tensor("op_5308_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_383_cast_fp16 = mul(x = var_5307_cast_fp16, y = var_5308_to_fp16)[name = tensor("aw_383_cast_fp16")]; + tensor var_5311_equation_0 = const()[name = tensor("op_5311_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5311_cast_fp16 = einsum(equation = var_5311_equation_0, values = (var_5153_cast_fp16, var_5070_cast_fp16))[name = tensor("op_5311_cast_fp16")]; + tensor var_5312_to_fp16 = const()[name = tensor("op_5312_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_385_cast_fp16 = mul(x = var_5311_cast_fp16, y = var_5312_to_fp16)[name = tensor("aw_385_cast_fp16")]; + tensor var_5315_equation_0 = const()[name = tensor("op_5315_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5315_cast_fp16 = einsum(equation = var_5315_equation_0, values = (var_5157_cast_fp16, var_5074_cast_fp16))[name = tensor("op_5315_cast_fp16")]; + tensor var_5316_to_fp16 = const()[name = tensor("op_5316_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_387_cast_fp16 = mul(x = var_5315_cast_fp16, y = var_5316_to_fp16)[name = tensor("aw_387_cast_fp16")]; + tensor var_5319_equation_0 = const()[name = tensor("op_5319_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5319_cast_fp16 = einsum(equation = var_5319_equation_0, values = (var_5161_cast_fp16, var_5078_cast_fp16))[name = tensor("op_5319_cast_fp16")]; + tensor var_5320_to_fp16 = const()[name = tensor("op_5320_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_389_cast_fp16 = mul(x = var_5319_cast_fp16, y = var_5320_to_fp16)[name = tensor("aw_389_cast_fp16")]; + tensor var_5323_equation_0 = const()[name = tensor("op_5323_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5323_cast_fp16 = einsum(equation = var_5323_equation_0, values = (var_5165_cast_fp16, var_5082_cast_fp16))[name = tensor("op_5323_cast_fp16")]; + tensor var_5324_to_fp16 = const()[name = tensor("op_5324_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_391_cast_fp16 = mul(x = var_5323_cast_fp16, y = var_5324_to_fp16)[name = tensor("aw_391_cast_fp16")]; + tensor var_5327_equation_0 = const()[name = tensor("op_5327_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5327_cast_fp16 = einsum(equation = var_5327_equation_0, values = (var_5169_cast_fp16, var_5086_cast_fp16))[name = tensor("op_5327_cast_fp16")]; + tensor var_5328_to_fp16 = const()[name = tensor("op_5328_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_393_cast_fp16 = mul(x = var_5327_cast_fp16, y = var_5328_to_fp16)[name = tensor("aw_393_cast_fp16")]; + tensor var_5331_equation_0 = const()[name = tensor("op_5331_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5331_cast_fp16 = einsum(equation = var_5331_equation_0, values = (var_5173_cast_fp16, var_5090_cast_fp16))[name = tensor("op_5331_cast_fp16")]; + tensor var_5332_to_fp16 = const()[name = tensor("op_5332_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_395_cast_fp16 = mul(x = var_5331_cast_fp16, y = var_5332_to_fp16)[name = tensor("aw_395_cast_fp16")]; + tensor var_5335_equation_0 = const()[name = tensor("op_5335_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5335_cast_fp16 = einsum(equation = var_5335_equation_0, values = (var_5177_cast_fp16, var_5094_cast_fp16))[name = tensor("op_5335_cast_fp16")]; + tensor var_5336_to_fp16 = const()[name = tensor("op_5336_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_397_cast_fp16 = mul(x = var_5335_cast_fp16, y = var_5336_to_fp16)[name = tensor("aw_397_cast_fp16")]; + tensor var_5339_equation_0 = const()[name = tensor("op_5339_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5339_cast_fp16 = einsum(equation = var_5339_equation_0, values = (var_5181_cast_fp16, var_5098_cast_fp16))[name = tensor("op_5339_cast_fp16")]; + tensor var_5340_to_fp16 = const()[name = tensor("op_5340_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_399_cast_fp16 = mul(x = var_5339_cast_fp16, y = var_5340_to_fp16)[name = tensor("aw_399_cast_fp16")]; + tensor var_5342_cast_fp16 = softmax(axis = var_2624, x = aw_361_cast_fp16)[name = tensor("op_5342_cast_fp16")]; + tensor var_5343_cast_fp16 = softmax(axis = var_2624, x = aw_363_cast_fp16)[name = tensor("op_5343_cast_fp16")]; + tensor var_5344_cast_fp16 = softmax(axis = var_2624, x = aw_365_cast_fp16)[name = tensor("op_5344_cast_fp16")]; + tensor var_5345_cast_fp16 = softmax(axis = var_2624, x = aw_367_cast_fp16)[name = tensor("op_5345_cast_fp16")]; + tensor var_5346_cast_fp16 = softmax(axis = var_2624, x = aw_369_cast_fp16)[name = tensor("op_5346_cast_fp16")]; + tensor var_5347_cast_fp16 = softmax(axis = var_2624, x = aw_371_cast_fp16)[name = tensor("op_5347_cast_fp16")]; + tensor var_5348_cast_fp16 = softmax(axis = var_2624, x = aw_373_cast_fp16)[name = tensor("op_5348_cast_fp16")]; + tensor var_5349_cast_fp16 = softmax(axis = var_2624, x = aw_375_cast_fp16)[name = tensor("op_5349_cast_fp16")]; + tensor var_5350_cast_fp16 = softmax(axis = var_2624, x = aw_377_cast_fp16)[name = tensor("op_5350_cast_fp16")]; + tensor var_5351_cast_fp16 = softmax(axis = var_2624, x = aw_379_cast_fp16)[name = tensor("op_5351_cast_fp16")]; + tensor var_5352_cast_fp16 = softmax(axis = var_2624, x = aw_381_cast_fp16)[name = tensor("op_5352_cast_fp16")]; + tensor var_5353_cast_fp16 = softmax(axis = var_2624, x = aw_383_cast_fp16)[name = tensor("op_5353_cast_fp16")]; + tensor var_5354_cast_fp16 = softmax(axis = var_2624, x = aw_385_cast_fp16)[name = tensor("op_5354_cast_fp16")]; + tensor var_5355_cast_fp16 = softmax(axis = var_2624, x = aw_387_cast_fp16)[name = tensor("op_5355_cast_fp16")]; + tensor var_5356_cast_fp16 = softmax(axis = var_2624, x = aw_389_cast_fp16)[name = tensor("op_5356_cast_fp16")]; + tensor var_5357_cast_fp16 = softmax(axis = var_2624, x = aw_391_cast_fp16)[name = tensor("op_5357_cast_fp16")]; + tensor var_5358_cast_fp16 = softmax(axis = var_2624, x = aw_393_cast_fp16)[name = tensor("op_5358_cast_fp16")]; + tensor var_5359_cast_fp16 = softmax(axis = var_2624, x = aw_395_cast_fp16)[name = tensor("op_5359_cast_fp16")]; + tensor var_5360_cast_fp16 = softmax(axis = var_2624, x = aw_397_cast_fp16)[name = tensor("op_5360_cast_fp16")]; + tensor var_5361_cast_fp16 = softmax(axis = var_2624, x = aw_399_cast_fp16)[name = tensor("op_5361_cast_fp16")]; + tensor var_5363_equation_0 = const()[name = tensor("op_5363_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5363_cast_fp16 = einsum(equation = var_5363_equation_0, values = (var_5183_cast_fp16, var_5342_cast_fp16))[name = tensor("op_5363_cast_fp16")]; + tensor var_5365_equation_0 = const()[name = tensor("op_5365_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5365_cast_fp16 = einsum(equation = var_5365_equation_0, values = (var_5187_cast_fp16, var_5343_cast_fp16))[name = tensor("op_5365_cast_fp16")]; + tensor var_5367_equation_0 = const()[name = tensor("op_5367_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5367_cast_fp16 = einsum(equation = var_5367_equation_0, values = (var_5191_cast_fp16, var_5344_cast_fp16))[name = tensor("op_5367_cast_fp16")]; + tensor var_5369_equation_0 = const()[name = tensor("op_5369_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5369_cast_fp16 = einsum(equation = var_5369_equation_0, values = (var_5195_cast_fp16, var_5345_cast_fp16))[name = tensor("op_5369_cast_fp16")]; + tensor var_5371_equation_0 = const()[name = tensor("op_5371_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5371_cast_fp16 = einsum(equation = var_5371_equation_0, values = (var_5199_cast_fp16, var_5346_cast_fp16))[name = tensor("op_5371_cast_fp16")]; + tensor var_5373_equation_0 = const()[name = tensor("op_5373_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5373_cast_fp16 = einsum(equation = var_5373_equation_0, values = (var_5203_cast_fp16, var_5347_cast_fp16))[name = tensor("op_5373_cast_fp16")]; + tensor var_5375_equation_0 = const()[name = tensor("op_5375_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5375_cast_fp16 = einsum(equation = var_5375_equation_0, values = (var_5207_cast_fp16, var_5348_cast_fp16))[name = tensor("op_5375_cast_fp16")]; + tensor var_5377_equation_0 = const()[name = tensor("op_5377_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5377_cast_fp16 = einsum(equation = var_5377_equation_0, values = (var_5211_cast_fp16, var_5349_cast_fp16))[name = tensor("op_5377_cast_fp16")]; + tensor var_5379_equation_0 = const()[name = tensor("op_5379_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5379_cast_fp16 = einsum(equation = var_5379_equation_0, values = (var_5215_cast_fp16, var_5350_cast_fp16))[name = tensor("op_5379_cast_fp16")]; + tensor var_5381_equation_0 = const()[name = tensor("op_5381_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5381_cast_fp16 = einsum(equation = var_5381_equation_0, values = (var_5219_cast_fp16, var_5351_cast_fp16))[name = tensor("op_5381_cast_fp16")]; + tensor var_5383_equation_0 = const()[name = tensor("op_5383_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5383_cast_fp16 = einsum(equation = var_5383_equation_0, values = (var_5223_cast_fp16, var_5352_cast_fp16))[name = tensor("op_5383_cast_fp16")]; + tensor var_5385_equation_0 = const()[name = tensor("op_5385_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5385_cast_fp16 = einsum(equation = var_5385_equation_0, values = (var_5227_cast_fp16, var_5353_cast_fp16))[name = tensor("op_5385_cast_fp16")]; + tensor var_5387_equation_0 = const()[name = tensor("op_5387_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5387_cast_fp16 = einsum(equation = var_5387_equation_0, values = (var_5231_cast_fp16, var_5354_cast_fp16))[name = tensor("op_5387_cast_fp16")]; + tensor var_5389_equation_0 = const()[name = tensor("op_5389_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5389_cast_fp16 = einsum(equation = var_5389_equation_0, values = (var_5235_cast_fp16, var_5355_cast_fp16))[name = tensor("op_5389_cast_fp16")]; + tensor var_5391_equation_0 = const()[name = tensor("op_5391_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5391_cast_fp16 = einsum(equation = var_5391_equation_0, values = (var_5239_cast_fp16, var_5356_cast_fp16))[name = tensor("op_5391_cast_fp16")]; + tensor var_5393_equation_0 = const()[name = tensor("op_5393_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5393_cast_fp16 = einsum(equation = var_5393_equation_0, values = (var_5243_cast_fp16, var_5357_cast_fp16))[name = tensor("op_5393_cast_fp16")]; + tensor var_5395_equation_0 = const()[name = tensor("op_5395_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5395_cast_fp16 = einsum(equation = var_5395_equation_0, values = (var_5247_cast_fp16, var_5358_cast_fp16))[name = tensor("op_5395_cast_fp16")]; + tensor var_5397_equation_0 = const()[name = tensor("op_5397_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5397_cast_fp16 = einsum(equation = var_5397_equation_0, values = (var_5251_cast_fp16, var_5359_cast_fp16))[name = tensor("op_5397_cast_fp16")]; + tensor var_5399_equation_0 = const()[name = tensor("op_5399_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5399_cast_fp16 = einsum(equation = var_5399_equation_0, values = (var_5255_cast_fp16, var_5360_cast_fp16))[name = tensor("op_5399_cast_fp16")]; + tensor var_5401_equation_0 = const()[name = tensor("op_5401_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5401_cast_fp16 = einsum(equation = var_5401_equation_0, values = (var_5259_cast_fp16, var_5361_cast_fp16))[name = tensor("op_5401_cast_fp16")]; + tensor input_149_interleave_0 = const()[name = tensor("input_149_interleave_0"), val = tensor(false)]; + tensor input_149_cast_fp16 = concat(axis = var_2624, interleave = input_149_interleave_0, values = (var_5363_cast_fp16, var_5365_cast_fp16, var_5367_cast_fp16, var_5369_cast_fp16, var_5371_cast_fp16, var_5373_cast_fp16, var_5375_cast_fp16, var_5377_cast_fp16, var_5379_cast_fp16, var_5381_cast_fp16, var_5383_cast_fp16, var_5385_cast_fp16, var_5387_cast_fp16, var_5389_cast_fp16, var_5391_cast_fp16, var_5393_cast_fp16, var_5395_cast_fp16, var_5397_cast_fp16, var_5399_cast_fp16, var_5401_cast_fp16))[name = tensor("input_149_cast_fp16")]; + tensor var_5407 = const()[name = tensor("op_5407"), val = tensor([1, 1])]; + tensor var_5409 = const()[name = tensor("op_5409"), val = tensor([1, 1])]; + tensor var_5411_pad_type_0 = const()[name = tensor("op_5411_pad_type_0"), val = tensor("custom")]; + tensor var_5411_pad_0 = const()[name = tensor("op_5411_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_2_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_2_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(364075136)))]; + tensor down_blocks_2_attentions_0_transformer_blocks_2_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_2_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(367352000)))]; + tensor var_5411_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_2_attn2_to_out_0_bias_to_fp16, dilations = var_5409, groups = var_2624, pad = var_5411_pad_0, pad_type = var_5411_pad_type_0, strides = var_5407, weight = down_blocks_2_attentions_0_transformer_blocks_2_attn2_to_out_0_weight_to_fp16, x = input_149_cast_fp16)[name = tensor("op_5411_cast_fp16")]; + tensor inputs_41_cast_fp16 = add(x = var_5411_cast_fp16, y = inputs_39_cast_fp16)[name = tensor("inputs_41_cast_fp16")]; + tensor input_151_axes_0 = const()[name = tensor("input_151_axes_0"), val = tensor([1])]; + tensor input_151_gamma_0_to_fp16 = const()[name = tensor("input_151_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(367354624)))]; + tensor input_151_beta_0_to_fp16 = const()[name = tensor("input_151_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(367357248)))]; + tensor var_5421_to_fp16 = const()[name = tensor("op_5421_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_151_cast_fp16 = layer_norm(axes = input_151_axes_0, beta = input_151_beta_0_to_fp16, epsilon = var_5421_to_fp16, gamma = input_151_gamma_0_to_fp16, x = inputs_41_cast_fp16)[name = tensor("input_151_cast_fp16")]; + tensor var_5437 = const()[name = tensor("op_5437"), val = tensor([1, 1])]; + tensor var_5439 = const()[name = tensor("op_5439"), val = tensor([1, 1])]; + tensor var_5441_pad_type_0 = const()[name = tensor("op_5441_pad_type_0"), val = tensor("custom")]; + tensor var_5441_pad_0 = const()[name = tensor("op_5441_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_2_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_2_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(367359872)))]; + tensor down_blocks_2_attentions_0_transformer_blocks_2_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_2_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(393574336)))]; + tensor var_5441_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_2_ff_net_0_proj_bias_to_fp16, dilations = var_5439, groups = var_2624, pad = var_5441_pad_0, pad_type = var_5441_pad_type_0, strides = var_5437, weight = down_blocks_2_attentions_0_transformer_blocks_2_ff_net_0_proj_weight_to_fp16, x = input_151_cast_fp16)[name = tensor("op_5441_cast_fp16")]; + tensor var_5442_split_sizes_0 = const()[name = tensor("op_5442_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_5442_axis_0 = const()[name = tensor("op_5442_axis_0"), val = tensor(1)]; + tensor var_5442_cast_fp16_0, tensor var_5442_cast_fp16_1 = split(axis = var_5442_axis_0, split_sizes = var_5442_split_sizes_0, x = var_5441_cast_fp16)[name = tensor("op_5442_cast_fp16")]; + tensor var_5444_mode_0 = const()[name = tensor("op_5444_mode_0"), val = tensor("EXACT")]; + tensor var_5444_cast_fp16 = gelu(mode = var_5444_mode_0, x = var_5442_cast_fp16_1)[name = tensor("op_5444_cast_fp16")]; + tensor input_153_cast_fp16 = mul(x = var_5442_cast_fp16_0, y = var_5444_cast_fp16)[name = tensor("input_153_cast_fp16")]; + tensor var_5448 = const()[name = tensor("op_5448"), val = tensor([1, 1])]; + tensor var_5450 = const()[name = tensor("op_5450"), val = tensor([1, 1])]; + tensor var_5452_pad_type_0 = const()[name = tensor("op_5452_pad_type_0"), val = tensor("custom")]; + tensor var_5452_pad_0 = const()[name = tensor("op_5452_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_2_ff_net_2_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_2_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(393594880)))]; + tensor down_blocks_2_attentions_0_transformer_blocks_2_ff_net_2_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_2_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(406702144)))]; + tensor var_5452_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_2_ff_net_2_bias_to_fp16, dilations = var_5450, groups = var_2624, pad = var_5452_pad_0, pad_type = var_5452_pad_type_0, strides = var_5448, weight = down_blocks_2_attentions_0_transformer_blocks_2_ff_net_2_weight_to_fp16, x = input_153_cast_fp16)[name = tensor("op_5452_cast_fp16")]; + tensor inputs_43_cast_fp16 = add(x = var_5452_cast_fp16, y = inputs_41_cast_fp16)[name = tensor("inputs_43_cast_fp16")]; + tensor hidden_states_83_axes_0 = const()[name = tensor("hidden_states_83_axes_0"), val = tensor([1])]; + tensor hidden_states_83_gamma_0_to_fp16 = const()[name = tensor("hidden_states_83_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(406704768)))]; + tensor hidden_states_83_beta_0_to_fp16 = const()[name = tensor("hidden_states_83_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(406707392)))]; + tensor var_5468_to_fp16 = const()[name = tensor("op_5468_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_83_cast_fp16 = layer_norm(axes = hidden_states_83_axes_0, beta = hidden_states_83_beta_0_to_fp16, epsilon = var_5468_to_fp16, gamma = hidden_states_83_gamma_0_to_fp16, x = inputs_43_cast_fp16)[name = tensor("hidden_states_83_cast_fp16")]; + tensor var_5483 = const()[name = tensor("op_5483"), val = tensor([1, 1])]; + tensor var_5485 = const()[name = tensor("op_5485"), val = tensor([1, 1])]; + tensor q_29_pad_type_0 = const()[name = tensor("q_29_pad_type_0"), val = tensor("custom")]; + tensor q_29_pad_0 = const()[name = tensor("q_29_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_3_attn1_to_q_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_3_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(406710016)))]; + tensor q_29_cast_fp16 = conv(dilations = var_5485, groups = var_2624, pad = q_29_pad_0, pad_type = q_29_pad_type_0, strides = var_5483, weight = down_blocks_2_attentions_0_transformer_blocks_3_attn1_to_q_weight_to_fp16, x = hidden_states_83_cast_fp16)[name = tensor("q_29_cast_fp16")]; + tensor var_5489 = const()[name = tensor("op_5489"), val = tensor([1, 1])]; + tensor var_5491 = const()[name = tensor("op_5491"), val = tensor([1, 1])]; + tensor k_57_pad_type_0 = const()[name = tensor("k_57_pad_type_0"), val = tensor("custom")]; + tensor k_57_pad_0 = const()[name = tensor("k_57_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_3_attn1_to_k_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_3_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(409986880)))]; + tensor k_57_cast_fp16 = conv(dilations = var_5491, groups = var_2624, pad = k_57_pad_0, pad_type = k_57_pad_type_0, strides = var_5489, weight = down_blocks_2_attentions_0_transformer_blocks_3_attn1_to_k_weight_to_fp16, x = hidden_states_83_cast_fp16)[name = tensor("k_57_cast_fp16")]; + tensor var_5495 = const()[name = tensor("op_5495"), val = tensor([1, 1])]; + tensor var_5497 = const()[name = tensor("op_5497"), val = tensor([1, 1])]; + tensor v_29_pad_type_0 = const()[name = tensor("v_29_pad_type_0"), val = tensor("custom")]; + tensor v_29_pad_0 = const()[name = tensor("v_29_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_3_attn1_to_v_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_3_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(413263744)))]; + tensor v_29_cast_fp16 = conv(dilations = var_5497, groups = var_2624, pad = v_29_pad_0, pad_type = v_29_pad_type_0, strides = var_5495, weight = down_blocks_2_attentions_0_transformer_blocks_3_attn1_to_v_weight_to_fp16, x = hidden_states_83_cast_fp16)[name = tensor("v_29_cast_fp16")]; + tensor var_5501_begin_0 = const()[name = tensor("op_5501_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5501_end_0 = const()[name = tensor("op_5501_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_5501_end_mask_0 = const()[name = tensor("op_5501_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5501_cast_fp16 = slice_by_index(begin = var_5501_begin_0, end = var_5501_end_0, end_mask = var_5501_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5501_cast_fp16")]; + tensor var_5505_begin_0 = const()[name = tensor("op_5505_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_5505_end_0 = const()[name = tensor("op_5505_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_5505_end_mask_0 = const()[name = tensor("op_5505_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5505_cast_fp16 = slice_by_index(begin = var_5505_begin_0, end = var_5505_end_0, end_mask = var_5505_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5505_cast_fp16")]; + tensor var_5509_begin_0 = const()[name = tensor("op_5509_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_5509_end_0 = const()[name = tensor("op_5509_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_5509_end_mask_0 = const()[name = tensor("op_5509_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5509_cast_fp16 = slice_by_index(begin = var_5509_begin_0, end = var_5509_end_0, end_mask = var_5509_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5509_cast_fp16")]; + tensor var_5513_begin_0 = const()[name = tensor("op_5513_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_5513_end_0 = const()[name = tensor("op_5513_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_5513_end_mask_0 = const()[name = tensor("op_5513_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5513_cast_fp16 = slice_by_index(begin = var_5513_begin_0, end = var_5513_end_0, end_mask = var_5513_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5513_cast_fp16")]; + tensor var_5517_begin_0 = const()[name = tensor("op_5517_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_5517_end_0 = const()[name = tensor("op_5517_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_5517_end_mask_0 = const()[name = tensor("op_5517_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5517_cast_fp16 = slice_by_index(begin = var_5517_begin_0, end = var_5517_end_0, end_mask = var_5517_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5517_cast_fp16")]; + tensor var_5521_begin_0 = const()[name = tensor("op_5521_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_5521_end_0 = const()[name = tensor("op_5521_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_5521_end_mask_0 = const()[name = tensor("op_5521_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5521_cast_fp16 = slice_by_index(begin = var_5521_begin_0, end = var_5521_end_0, end_mask = var_5521_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5521_cast_fp16")]; + tensor var_5525_begin_0 = const()[name = tensor("op_5525_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_5525_end_0 = const()[name = tensor("op_5525_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_5525_end_mask_0 = const()[name = tensor("op_5525_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5525_cast_fp16 = slice_by_index(begin = var_5525_begin_0, end = var_5525_end_0, end_mask = var_5525_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5525_cast_fp16")]; + tensor var_5529_begin_0 = const()[name = tensor("op_5529_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_5529_end_0 = const()[name = tensor("op_5529_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_5529_end_mask_0 = const()[name = tensor("op_5529_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5529_cast_fp16 = slice_by_index(begin = var_5529_begin_0, end = var_5529_end_0, end_mask = var_5529_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5529_cast_fp16")]; + tensor var_5533_begin_0 = const()[name = tensor("op_5533_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_5533_end_0 = const()[name = tensor("op_5533_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_5533_end_mask_0 = const()[name = tensor("op_5533_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5533_cast_fp16 = slice_by_index(begin = var_5533_begin_0, end = var_5533_end_0, end_mask = var_5533_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5533_cast_fp16")]; + tensor var_5537_begin_0 = const()[name = tensor("op_5537_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_5537_end_0 = const()[name = tensor("op_5537_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_5537_end_mask_0 = const()[name = tensor("op_5537_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5537_cast_fp16 = slice_by_index(begin = var_5537_begin_0, end = var_5537_end_0, end_mask = var_5537_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5537_cast_fp16")]; + tensor var_5541_begin_0 = const()[name = tensor("op_5541_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_5541_end_0 = const()[name = tensor("op_5541_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_5541_end_mask_0 = const()[name = tensor("op_5541_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5541_cast_fp16 = slice_by_index(begin = var_5541_begin_0, end = var_5541_end_0, end_mask = var_5541_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5541_cast_fp16")]; + tensor var_5545_begin_0 = const()[name = tensor("op_5545_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_5545_end_0 = const()[name = tensor("op_5545_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_5545_end_mask_0 = const()[name = tensor("op_5545_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5545_cast_fp16 = slice_by_index(begin = var_5545_begin_0, end = var_5545_end_0, end_mask = var_5545_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5545_cast_fp16")]; + tensor var_5549_begin_0 = const()[name = tensor("op_5549_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_5549_end_0 = const()[name = tensor("op_5549_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_5549_end_mask_0 = const()[name = tensor("op_5549_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5549_cast_fp16 = slice_by_index(begin = var_5549_begin_0, end = var_5549_end_0, end_mask = var_5549_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5549_cast_fp16")]; + tensor var_5553_begin_0 = const()[name = tensor("op_5553_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_5553_end_0 = const()[name = tensor("op_5553_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_5553_end_mask_0 = const()[name = tensor("op_5553_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5553_cast_fp16 = slice_by_index(begin = var_5553_begin_0, end = var_5553_end_0, end_mask = var_5553_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5553_cast_fp16")]; + tensor var_5557_begin_0 = const()[name = tensor("op_5557_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_5557_end_0 = const()[name = tensor("op_5557_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_5557_end_mask_0 = const()[name = tensor("op_5557_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5557_cast_fp16 = slice_by_index(begin = var_5557_begin_0, end = var_5557_end_0, end_mask = var_5557_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5557_cast_fp16")]; + tensor var_5561_begin_0 = const()[name = tensor("op_5561_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_5561_end_0 = const()[name = tensor("op_5561_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_5561_end_mask_0 = const()[name = tensor("op_5561_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5561_cast_fp16 = slice_by_index(begin = var_5561_begin_0, end = var_5561_end_0, end_mask = var_5561_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5561_cast_fp16")]; + tensor var_5565_begin_0 = const()[name = tensor("op_5565_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_5565_end_0 = const()[name = tensor("op_5565_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_5565_end_mask_0 = const()[name = tensor("op_5565_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5565_cast_fp16 = slice_by_index(begin = var_5565_begin_0, end = var_5565_end_0, end_mask = var_5565_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5565_cast_fp16")]; + tensor var_5569_begin_0 = const()[name = tensor("op_5569_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_5569_end_0 = const()[name = tensor("op_5569_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_5569_end_mask_0 = const()[name = tensor("op_5569_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5569_cast_fp16 = slice_by_index(begin = var_5569_begin_0, end = var_5569_end_0, end_mask = var_5569_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5569_cast_fp16")]; + tensor var_5573_begin_0 = const()[name = tensor("op_5573_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_5573_end_0 = const()[name = tensor("op_5573_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_5573_end_mask_0 = const()[name = tensor("op_5573_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5573_cast_fp16 = slice_by_index(begin = var_5573_begin_0, end = var_5573_end_0, end_mask = var_5573_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5573_cast_fp16")]; + tensor var_5577_begin_0 = const()[name = tensor("op_5577_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_5577_end_0 = const()[name = tensor("op_5577_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_5577_end_mask_0 = const()[name = tensor("op_5577_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5577_cast_fp16 = slice_by_index(begin = var_5577_begin_0, end = var_5577_end_0, end_mask = var_5577_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5577_cast_fp16")]; + tensor k_59_perm_0 = const()[name = tensor("k_59_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_5584_begin_0 = const()[name = tensor("op_5584_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5584_end_0 = const()[name = tensor("op_5584_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_5584_end_mask_0 = const()[name = tensor("op_5584_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_59_cast_fp16 = transpose(perm = k_59_perm_0, x = k_57_cast_fp16)[name = tensor("transpose_125")]; + tensor var_5584_cast_fp16 = slice_by_index(begin = var_5584_begin_0, end = var_5584_end_0, end_mask = var_5584_end_mask_0, x = k_59_cast_fp16)[name = tensor("op_5584_cast_fp16")]; + tensor var_5588_begin_0 = const()[name = tensor("op_5588_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_5588_end_0 = const()[name = tensor("op_5588_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_5588_end_mask_0 = const()[name = tensor("op_5588_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5588_cast_fp16 = slice_by_index(begin = var_5588_begin_0, end = var_5588_end_0, end_mask = var_5588_end_mask_0, x = k_59_cast_fp16)[name = tensor("op_5588_cast_fp16")]; + tensor var_5592_begin_0 = const()[name = tensor("op_5592_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_5592_end_0 = const()[name = tensor("op_5592_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_5592_end_mask_0 = const()[name = tensor("op_5592_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5592_cast_fp16 = slice_by_index(begin = var_5592_begin_0, end = var_5592_end_0, end_mask = var_5592_end_mask_0, x = k_59_cast_fp16)[name = tensor("op_5592_cast_fp16")]; + tensor var_5596_begin_0 = const()[name = tensor("op_5596_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_5596_end_0 = const()[name = tensor("op_5596_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_5596_end_mask_0 = const()[name = tensor("op_5596_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5596_cast_fp16 = slice_by_index(begin = var_5596_begin_0, end = var_5596_end_0, end_mask = var_5596_end_mask_0, x = k_59_cast_fp16)[name = tensor("op_5596_cast_fp16")]; + tensor var_5600_begin_0 = const()[name = tensor("op_5600_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_5600_end_0 = const()[name = tensor("op_5600_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_5600_end_mask_0 = const()[name = tensor("op_5600_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5600_cast_fp16 = slice_by_index(begin = var_5600_begin_0, end = var_5600_end_0, end_mask = var_5600_end_mask_0, x = k_59_cast_fp16)[name = tensor("op_5600_cast_fp16")]; + tensor var_5604_begin_0 = const()[name = tensor("op_5604_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_5604_end_0 = const()[name = tensor("op_5604_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_5604_end_mask_0 = const()[name = tensor("op_5604_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5604_cast_fp16 = slice_by_index(begin = var_5604_begin_0, end = var_5604_end_0, end_mask = var_5604_end_mask_0, x = k_59_cast_fp16)[name = tensor("op_5604_cast_fp16")]; + tensor var_5608_begin_0 = const()[name = tensor("op_5608_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_5608_end_0 = const()[name = tensor("op_5608_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_5608_end_mask_0 = const()[name = tensor("op_5608_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5608_cast_fp16 = slice_by_index(begin = var_5608_begin_0, end = var_5608_end_0, end_mask = var_5608_end_mask_0, x = k_59_cast_fp16)[name = tensor("op_5608_cast_fp16")]; + tensor var_5612_begin_0 = const()[name = tensor("op_5612_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_5612_end_0 = const()[name = tensor("op_5612_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_5612_end_mask_0 = const()[name = tensor("op_5612_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5612_cast_fp16 = slice_by_index(begin = var_5612_begin_0, end = var_5612_end_0, end_mask = var_5612_end_mask_0, x = k_59_cast_fp16)[name = tensor("op_5612_cast_fp16")]; + tensor var_5616_begin_0 = const()[name = tensor("op_5616_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_5616_end_0 = const()[name = tensor("op_5616_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_5616_end_mask_0 = const()[name = tensor("op_5616_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5616_cast_fp16 = slice_by_index(begin = var_5616_begin_0, end = var_5616_end_0, end_mask = var_5616_end_mask_0, x = k_59_cast_fp16)[name = tensor("op_5616_cast_fp16")]; + tensor var_5620_begin_0 = const()[name = tensor("op_5620_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_5620_end_0 = const()[name = tensor("op_5620_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_5620_end_mask_0 = const()[name = tensor("op_5620_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5620_cast_fp16 = slice_by_index(begin = var_5620_begin_0, end = var_5620_end_0, end_mask = var_5620_end_mask_0, x = k_59_cast_fp16)[name = tensor("op_5620_cast_fp16")]; + tensor var_5624_begin_0 = const()[name = tensor("op_5624_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_5624_end_0 = const()[name = tensor("op_5624_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_5624_end_mask_0 = const()[name = tensor("op_5624_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5624_cast_fp16 = slice_by_index(begin = var_5624_begin_0, end = var_5624_end_0, end_mask = var_5624_end_mask_0, x = k_59_cast_fp16)[name = tensor("op_5624_cast_fp16")]; + tensor var_5628_begin_0 = const()[name = tensor("op_5628_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_5628_end_0 = const()[name = tensor("op_5628_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_5628_end_mask_0 = const()[name = tensor("op_5628_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5628_cast_fp16 = slice_by_index(begin = var_5628_begin_0, end = var_5628_end_0, end_mask = var_5628_end_mask_0, x = k_59_cast_fp16)[name = tensor("op_5628_cast_fp16")]; + tensor var_5632_begin_0 = const()[name = tensor("op_5632_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_5632_end_0 = const()[name = tensor("op_5632_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_5632_end_mask_0 = const()[name = tensor("op_5632_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5632_cast_fp16 = slice_by_index(begin = var_5632_begin_0, end = var_5632_end_0, end_mask = var_5632_end_mask_0, x = k_59_cast_fp16)[name = tensor("op_5632_cast_fp16")]; + tensor var_5636_begin_0 = const()[name = tensor("op_5636_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_5636_end_0 = const()[name = tensor("op_5636_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_5636_end_mask_0 = const()[name = tensor("op_5636_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5636_cast_fp16 = slice_by_index(begin = var_5636_begin_0, end = var_5636_end_0, end_mask = var_5636_end_mask_0, x = k_59_cast_fp16)[name = tensor("op_5636_cast_fp16")]; + tensor var_5640_begin_0 = const()[name = tensor("op_5640_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_5640_end_0 = const()[name = tensor("op_5640_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_5640_end_mask_0 = const()[name = tensor("op_5640_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5640_cast_fp16 = slice_by_index(begin = var_5640_begin_0, end = var_5640_end_0, end_mask = var_5640_end_mask_0, x = k_59_cast_fp16)[name = tensor("op_5640_cast_fp16")]; + tensor var_5644_begin_0 = const()[name = tensor("op_5644_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_5644_end_0 = const()[name = tensor("op_5644_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_5644_end_mask_0 = const()[name = tensor("op_5644_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5644_cast_fp16 = slice_by_index(begin = var_5644_begin_0, end = var_5644_end_0, end_mask = var_5644_end_mask_0, x = k_59_cast_fp16)[name = tensor("op_5644_cast_fp16")]; + tensor var_5648_begin_0 = const()[name = tensor("op_5648_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_5648_end_0 = const()[name = tensor("op_5648_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_5648_end_mask_0 = const()[name = tensor("op_5648_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5648_cast_fp16 = slice_by_index(begin = var_5648_begin_0, end = var_5648_end_0, end_mask = var_5648_end_mask_0, x = k_59_cast_fp16)[name = tensor("op_5648_cast_fp16")]; + tensor var_5652_begin_0 = const()[name = tensor("op_5652_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_5652_end_0 = const()[name = tensor("op_5652_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_5652_end_mask_0 = const()[name = tensor("op_5652_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5652_cast_fp16 = slice_by_index(begin = var_5652_begin_0, end = var_5652_end_0, end_mask = var_5652_end_mask_0, x = k_59_cast_fp16)[name = tensor("op_5652_cast_fp16")]; + tensor var_5656_begin_0 = const()[name = tensor("op_5656_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_5656_end_0 = const()[name = tensor("op_5656_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_5656_end_mask_0 = const()[name = tensor("op_5656_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5656_cast_fp16 = slice_by_index(begin = var_5656_begin_0, end = var_5656_end_0, end_mask = var_5656_end_mask_0, x = k_59_cast_fp16)[name = tensor("op_5656_cast_fp16")]; + tensor var_5660_begin_0 = const()[name = tensor("op_5660_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_5660_end_0 = const()[name = tensor("op_5660_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_5660_end_mask_0 = const()[name = tensor("op_5660_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5660_cast_fp16 = slice_by_index(begin = var_5660_begin_0, end = var_5660_end_0, end_mask = var_5660_end_mask_0, x = k_59_cast_fp16)[name = tensor("op_5660_cast_fp16")]; + tensor var_5662_begin_0 = const()[name = tensor("op_5662_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5662_end_0 = const()[name = tensor("op_5662_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_5662_end_mask_0 = const()[name = tensor("op_5662_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5662_cast_fp16 = slice_by_index(begin = var_5662_begin_0, end = var_5662_end_0, end_mask = var_5662_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5662_cast_fp16")]; + tensor var_5666_begin_0 = const()[name = tensor("op_5666_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_5666_end_0 = const()[name = tensor("op_5666_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_5666_end_mask_0 = const()[name = tensor("op_5666_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5666_cast_fp16 = slice_by_index(begin = var_5666_begin_0, end = var_5666_end_0, end_mask = var_5666_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5666_cast_fp16")]; + tensor var_5670_begin_0 = const()[name = tensor("op_5670_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_5670_end_0 = const()[name = tensor("op_5670_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_5670_end_mask_0 = const()[name = tensor("op_5670_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5670_cast_fp16 = slice_by_index(begin = var_5670_begin_0, end = var_5670_end_0, end_mask = var_5670_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5670_cast_fp16")]; + tensor var_5674_begin_0 = const()[name = tensor("op_5674_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_5674_end_0 = const()[name = tensor("op_5674_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_5674_end_mask_0 = const()[name = tensor("op_5674_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5674_cast_fp16 = slice_by_index(begin = var_5674_begin_0, end = var_5674_end_0, end_mask = var_5674_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5674_cast_fp16")]; + tensor var_5678_begin_0 = const()[name = tensor("op_5678_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_5678_end_0 = const()[name = tensor("op_5678_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_5678_end_mask_0 = const()[name = tensor("op_5678_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5678_cast_fp16 = slice_by_index(begin = var_5678_begin_0, end = var_5678_end_0, end_mask = var_5678_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5678_cast_fp16")]; + tensor var_5682_begin_0 = const()[name = tensor("op_5682_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_5682_end_0 = const()[name = tensor("op_5682_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_5682_end_mask_0 = const()[name = tensor("op_5682_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5682_cast_fp16 = slice_by_index(begin = var_5682_begin_0, end = var_5682_end_0, end_mask = var_5682_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5682_cast_fp16")]; + tensor var_5686_begin_0 = const()[name = tensor("op_5686_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_5686_end_0 = const()[name = tensor("op_5686_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_5686_end_mask_0 = const()[name = tensor("op_5686_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5686_cast_fp16 = slice_by_index(begin = var_5686_begin_0, end = var_5686_end_0, end_mask = var_5686_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5686_cast_fp16")]; + tensor var_5690_begin_0 = const()[name = tensor("op_5690_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_5690_end_0 = const()[name = tensor("op_5690_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_5690_end_mask_0 = const()[name = tensor("op_5690_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5690_cast_fp16 = slice_by_index(begin = var_5690_begin_0, end = var_5690_end_0, end_mask = var_5690_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5690_cast_fp16")]; + tensor var_5694_begin_0 = const()[name = tensor("op_5694_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_5694_end_0 = const()[name = tensor("op_5694_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_5694_end_mask_0 = const()[name = tensor("op_5694_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5694_cast_fp16 = slice_by_index(begin = var_5694_begin_0, end = var_5694_end_0, end_mask = var_5694_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5694_cast_fp16")]; + tensor var_5698_begin_0 = const()[name = tensor("op_5698_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_5698_end_0 = const()[name = tensor("op_5698_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_5698_end_mask_0 = const()[name = tensor("op_5698_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5698_cast_fp16 = slice_by_index(begin = var_5698_begin_0, end = var_5698_end_0, end_mask = var_5698_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5698_cast_fp16")]; + tensor var_5702_begin_0 = const()[name = tensor("op_5702_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_5702_end_0 = const()[name = tensor("op_5702_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_5702_end_mask_0 = const()[name = tensor("op_5702_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5702_cast_fp16 = slice_by_index(begin = var_5702_begin_0, end = var_5702_end_0, end_mask = var_5702_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5702_cast_fp16")]; + tensor var_5706_begin_0 = const()[name = tensor("op_5706_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_5706_end_0 = const()[name = tensor("op_5706_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_5706_end_mask_0 = const()[name = tensor("op_5706_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5706_cast_fp16 = slice_by_index(begin = var_5706_begin_0, end = var_5706_end_0, end_mask = var_5706_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5706_cast_fp16")]; + tensor var_5710_begin_0 = const()[name = tensor("op_5710_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_5710_end_0 = const()[name = tensor("op_5710_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_5710_end_mask_0 = const()[name = tensor("op_5710_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5710_cast_fp16 = slice_by_index(begin = var_5710_begin_0, end = var_5710_end_0, end_mask = var_5710_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5710_cast_fp16")]; + tensor var_5714_begin_0 = const()[name = tensor("op_5714_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_5714_end_0 = const()[name = tensor("op_5714_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_5714_end_mask_0 = const()[name = tensor("op_5714_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5714_cast_fp16 = slice_by_index(begin = var_5714_begin_0, end = var_5714_end_0, end_mask = var_5714_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5714_cast_fp16")]; + tensor var_5718_begin_0 = const()[name = tensor("op_5718_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_5718_end_0 = const()[name = tensor("op_5718_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_5718_end_mask_0 = const()[name = tensor("op_5718_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5718_cast_fp16 = slice_by_index(begin = var_5718_begin_0, end = var_5718_end_0, end_mask = var_5718_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5718_cast_fp16")]; + tensor var_5722_begin_0 = const()[name = tensor("op_5722_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_5722_end_0 = const()[name = tensor("op_5722_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_5722_end_mask_0 = const()[name = tensor("op_5722_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5722_cast_fp16 = slice_by_index(begin = var_5722_begin_0, end = var_5722_end_0, end_mask = var_5722_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5722_cast_fp16")]; + tensor var_5726_begin_0 = const()[name = tensor("op_5726_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_5726_end_0 = const()[name = tensor("op_5726_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_5726_end_mask_0 = const()[name = tensor("op_5726_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5726_cast_fp16 = slice_by_index(begin = var_5726_begin_0, end = var_5726_end_0, end_mask = var_5726_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5726_cast_fp16")]; + tensor var_5730_begin_0 = const()[name = tensor("op_5730_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_5730_end_0 = const()[name = tensor("op_5730_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_5730_end_mask_0 = const()[name = tensor("op_5730_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5730_cast_fp16 = slice_by_index(begin = var_5730_begin_0, end = var_5730_end_0, end_mask = var_5730_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5730_cast_fp16")]; + tensor var_5734_begin_0 = const()[name = tensor("op_5734_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_5734_end_0 = const()[name = tensor("op_5734_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_5734_end_mask_0 = const()[name = tensor("op_5734_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5734_cast_fp16 = slice_by_index(begin = var_5734_begin_0, end = var_5734_end_0, end_mask = var_5734_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5734_cast_fp16")]; + tensor var_5738_begin_0 = const()[name = tensor("op_5738_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_5738_end_0 = const()[name = tensor("op_5738_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_5738_end_mask_0 = const()[name = tensor("op_5738_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5738_cast_fp16 = slice_by_index(begin = var_5738_begin_0, end = var_5738_end_0, end_mask = var_5738_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5738_cast_fp16")]; + tensor var_5742_equation_0 = const()[name = tensor("op_5742_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5742_cast_fp16 = einsum(equation = var_5742_equation_0, values = (var_5584_cast_fp16, var_5501_cast_fp16))[name = tensor("op_5742_cast_fp16")]; + tensor var_5743_to_fp16 = const()[name = tensor("op_5743_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_401_cast_fp16 = mul(x = var_5742_cast_fp16, y = var_5743_to_fp16)[name = tensor("aw_401_cast_fp16")]; + tensor var_5746_equation_0 = const()[name = tensor("op_5746_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5746_cast_fp16 = einsum(equation = var_5746_equation_0, values = (var_5588_cast_fp16, var_5505_cast_fp16))[name = tensor("op_5746_cast_fp16")]; + tensor var_5747_to_fp16 = const()[name = tensor("op_5747_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_403_cast_fp16 = mul(x = var_5746_cast_fp16, y = var_5747_to_fp16)[name = tensor("aw_403_cast_fp16")]; + tensor var_5750_equation_0 = const()[name = tensor("op_5750_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5750_cast_fp16 = einsum(equation = var_5750_equation_0, values = (var_5592_cast_fp16, var_5509_cast_fp16))[name = tensor("op_5750_cast_fp16")]; + tensor var_5751_to_fp16 = const()[name = tensor("op_5751_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_405_cast_fp16 = mul(x = var_5750_cast_fp16, y = var_5751_to_fp16)[name = tensor("aw_405_cast_fp16")]; + tensor var_5754_equation_0 = const()[name = tensor("op_5754_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5754_cast_fp16 = einsum(equation = var_5754_equation_0, values = (var_5596_cast_fp16, var_5513_cast_fp16))[name = tensor("op_5754_cast_fp16")]; + tensor var_5755_to_fp16 = const()[name = tensor("op_5755_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_407_cast_fp16 = mul(x = var_5754_cast_fp16, y = var_5755_to_fp16)[name = tensor("aw_407_cast_fp16")]; + tensor var_5758_equation_0 = const()[name = tensor("op_5758_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5758_cast_fp16 = einsum(equation = var_5758_equation_0, values = (var_5600_cast_fp16, var_5517_cast_fp16))[name = tensor("op_5758_cast_fp16")]; + tensor var_5759_to_fp16 = const()[name = tensor("op_5759_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_409_cast_fp16 = mul(x = var_5758_cast_fp16, y = var_5759_to_fp16)[name = tensor("aw_409_cast_fp16")]; + tensor var_5762_equation_0 = const()[name = tensor("op_5762_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5762_cast_fp16 = einsum(equation = var_5762_equation_0, values = (var_5604_cast_fp16, var_5521_cast_fp16))[name = tensor("op_5762_cast_fp16")]; + tensor var_5763_to_fp16 = const()[name = tensor("op_5763_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_411_cast_fp16 = mul(x = var_5762_cast_fp16, y = var_5763_to_fp16)[name = tensor("aw_411_cast_fp16")]; + tensor var_5766_equation_0 = const()[name = tensor("op_5766_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5766_cast_fp16 = einsum(equation = var_5766_equation_0, values = (var_5608_cast_fp16, var_5525_cast_fp16))[name = tensor("op_5766_cast_fp16")]; + tensor var_5767_to_fp16 = const()[name = tensor("op_5767_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_413_cast_fp16 = mul(x = var_5766_cast_fp16, y = var_5767_to_fp16)[name = tensor("aw_413_cast_fp16")]; + tensor var_5770_equation_0 = const()[name = tensor("op_5770_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5770_cast_fp16 = einsum(equation = var_5770_equation_0, values = (var_5612_cast_fp16, var_5529_cast_fp16))[name = tensor("op_5770_cast_fp16")]; + tensor var_5771_to_fp16 = const()[name = tensor("op_5771_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_415_cast_fp16 = mul(x = var_5770_cast_fp16, y = var_5771_to_fp16)[name = tensor("aw_415_cast_fp16")]; + tensor var_5774_equation_0 = const()[name = tensor("op_5774_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5774_cast_fp16 = einsum(equation = var_5774_equation_0, values = (var_5616_cast_fp16, var_5533_cast_fp16))[name = tensor("op_5774_cast_fp16")]; + tensor var_5775_to_fp16 = const()[name = tensor("op_5775_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_417_cast_fp16 = mul(x = var_5774_cast_fp16, y = var_5775_to_fp16)[name = tensor("aw_417_cast_fp16")]; + tensor var_5778_equation_0 = const()[name = tensor("op_5778_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5778_cast_fp16 = einsum(equation = var_5778_equation_0, values = (var_5620_cast_fp16, var_5537_cast_fp16))[name = tensor("op_5778_cast_fp16")]; + tensor var_5779_to_fp16 = const()[name = tensor("op_5779_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_419_cast_fp16 = mul(x = var_5778_cast_fp16, y = var_5779_to_fp16)[name = tensor("aw_419_cast_fp16")]; + tensor var_5782_equation_0 = const()[name = tensor("op_5782_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5782_cast_fp16 = einsum(equation = var_5782_equation_0, values = (var_5624_cast_fp16, var_5541_cast_fp16))[name = tensor("op_5782_cast_fp16")]; + tensor var_5783_to_fp16 = const()[name = tensor("op_5783_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_421_cast_fp16 = mul(x = var_5782_cast_fp16, y = var_5783_to_fp16)[name = tensor("aw_421_cast_fp16")]; + tensor var_5786_equation_0 = const()[name = tensor("op_5786_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5786_cast_fp16 = einsum(equation = var_5786_equation_0, values = (var_5628_cast_fp16, var_5545_cast_fp16))[name = tensor("op_5786_cast_fp16")]; + tensor var_5787_to_fp16 = const()[name = tensor("op_5787_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_423_cast_fp16 = mul(x = var_5786_cast_fp16, y = var_5787_to_fp16)[name = tensor("aw_423_cast_fp16")]; + tensor var_5790_equation_0 = const()[name = tensor("op_5790_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5790_cast_fp16 = einsum(equation = var_5790_equation_0, values = (var_5632_cast_fp16, var_5549_cast_fp16))[name = tensor("op_5790_cast_fp16")]; + tensor var_5791_to_fp16 = const()[name = tensor("op_5791_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_425_cast_fp16 = mul(x = var_5790_cast_fp16, y = var_5791_to_fp16)[name = tensor("aw_425_cast_fp16")]; + tensor var_5794_equation_0 = const()[name = tensor("op_5794_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5794_cast_fp16 = einsum(equation = var_5794_equation_0, values = (var_5636_cast_fp16, var_5553_cast_fp16))[name = tensor("op_5794_cast_fp16")]; + tensor var_5795_to_fp16 = const()[name = tensor("op_5795_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_427_cast_fp16 = mul(x = var_5794_cast_fp16, y = var_5795_to_fp16)[name = tensor("aw_427_cast_fp16")]; + tensor var_5798_equation_0 = const()[name = tensor("op_5798_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5798_cast_fp16 = einsum(equation = var_5798_equation_0, values = (var_5640_cast_fp16, var_5557_cast_fp16))[name = tensor("op_5798_cast_fp16")]; + tensor var_5799_to_fp16 = const()[name = tensor("op_5799_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_429_cast_fp16 = mul(x = var_5798_cast_fp16, y = var_5799_to_fp16)[name = tensor("aw_429_cast_fp16")]; + tensor var_5802_equation_0 = const()[name = tensor("op_5802_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5802_cast_fp16 = einsum(equation = var_5802_equation_0, values = (var_5644_cast_fp16, var_5561_cast_fp16))[name = tensor("op_5802_cast_fp16")]; + tensor var_5803_to_fp16 = const()[name = tensor("op_5803_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_431_cast_fp16 = mul(x = var_5802_cast_fp16, y = var_5803_to_fp16)[name = tensor("aw_431_cast_fp16")]; + tensor var_5806_equation_0 = const()[name = tensor("op_5806_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5806_cast_fp16 = einsum(equation = var_5806_equation_0, values = (var_5648_cast_fp16, var_5565_cast_fp16))[name = tensor("op_5806_cast_fp16")]; + tensor var_5807_to_fp16 = const()[name = tensor("op_5807_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_433_cast_fp16 = mul(x = var_5806_cast_fp16, y = var_5807_to_fp16)[name = tensor("aw_433_cast_fp16")]; + tensor var_5810_equation_0 = const()[name = tensor("op_5810_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5810_cast_fp16 = einsum(equation = var_5810_equation_0, values = (var_5652_cast_fp16, var_5569_cast_fp16))[name = tensor("op_5810_cast_fp16")]; + tensor var_5811_to_fp16 = const()[name = tensor("op_5811_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_435_cast_fp16 = mul(x = var_5810_cast_fp16, y = var_5811_to_fp16)[name = tensor("aw_435_cast_fp16")]; + tensor var_5814_equation_0 = const()[name = tensor("op_5814_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5814_cast_fp16 = einsum(equation = var_5814_equation_0, values = (var_5656_cast_fp16, var_5573_cast_fp16))[name = tensor("op_5814_cast_fp16")]; + tensor var_5815_to_fp16 = const()[name = tensor("op_5815_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_437_cast_fp16 = mul(x = var_5814_cast_fp16, y = var_5815_to_fp16)[name = tensor("aw_437_cast_fp16")]; + tensor var_5818_equation_0 = const()[name = tensor("op_5818_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5818_cast_fp16 = einsum(equation = var_5818_equation_0, values = (var_5660_cast_fp16, var_5577_cast_fp16))[name = tensor("op_5818_cast_fp16")]; + tensor var_5819_to_fp16 = const()[name = tensor("op_5819_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_439_cast_fp16 = mul(x = var_5818_cast_fp16, y = var_5819_to_fp16)[name = tensor("aw_439_cast_fp16")]; + tensor var_5821_cast_fp16 = softmax(axis = var_2624, x = aw_401_cast_fp16)[name = tensor("op_5821_cast_fp16")]; + tensor var_5822_cast_fp16 = softmax(axis = var_2624, x = aw_403_cast_fp16)[name = tensor("op_5822_cast_fp16")]; + tensor var_5823_cast_fp16 = softmax(axis = var_2624, x = aw_405_cast_fp16)[name = tensor("op_5823_cast_fp16")]; + tensor var_5824_cast_fp16 = softmax(axis = var_2624, x = aw_407_cast_fp16)[name = tensor("op_5824_cast_fp16")]; + tensor var_5825_cast_fp16 = softmax(axis = var_2624, x = aw_409_cast_fp16)[name = tensor("op_5825_cast_fp16")]; + tensor var_5826_cast_fp16 = softmax(axis = var_2624, x = aw_411_cast_fp16)[name = tensor("op_5826_cast_fp16")]; + tensor var_5827_cast_fp16 = softmax(axis = var_2624, x = aw_413_cast_fp16)[name = tensor("op_5827_cast_fp16")]; + tensor var_5828_cast_fp16 = softmax(axis = var_2624, x = aw_415_cast_fp16)[name = tensor("op_5828_cast_fp16")]; + tensor var_5829_cast_fp16 = softmax(axis = var_2624, x = aw_417_cast_fp16)[name = tensor("op_5829_cast_fp16")]; + tensor var_5830_cast_fp16 = softmax(axis = var_2624, x = aw_419_cast_fp16)[name = tensor("op_5830_cast_fp16")]; + tensor var_5831_cast_fp16 = softmax(axis = var_2624, x = aw_421_cast_fp16)[name = tensor("op_5831_cast_fp16")]; + tensor var_5832_cast_fp16 = softmax(axis = var_2624, x = aw_423_cast_fp16)[name = tensor("op_5832_cast_fp16")]; + tensor var_5833_cast_fp16 = softmax(axis = var_2624, x = aw_425_cast_fp16)[name = tensor("op_5833_cast_fp16")]; + tensor var_5834_cast_fp16 = softmax(axis = var_2624, x = aw_427_cast_fp16)[name = tensor("op_5834_cast_fp16")]; + tensor var_5835_cast_fp16 = softmax(axis = var_2624, x = aw_429_cast_fp16)[name = tensor("op_5835_cast_fp16")]; + tensor var_5836_cast_fp16 = softmax(axis = var_2624, x = aw_431_cast_fp16)[name = tensor("op_5836_cast_fp16")]; + tensor var_5837_cast_fp16 = softmax(axis = var_2624, x = aw_433_cast_fp16)[name = tensor("op_5837_cast_fp16")]; + tensor var_5838_cast_fp16 = softmax(axis = var_2624, x = aw_435_cast_fp16)[name = tensor("op_5838_cast_fp16")]; + tensor var_5839_cast_fp16 = softmax(axis = var_2624, x = aw_437_cast_fp16)[name = tensor("op_5839_cast_fp16")]; + tensor var_5840_cast_fp16 = softmax(axis = var_2624, x = aw_439_cast_fp16)[name = tensor("op_5840_cast_fp16")]; + tensor var_5842_equation_0 = const()[name = tensor("op_5842_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5842_cast_fp16 = einsum(equation = var_5842_equation_0, values = (var_5662_cast_fp16, var_5821_cast_fp16))[name = tensor("op_5842_cast_fp16")]; + tensor var_5844_equation_0 = const()[name = tensor("op_5844_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5844_cast_fp16 = einsum(equation = var_5844_equation_0, values = (var_5666_cast_fp16, var_5822_cast_fp16))[name = tensor("op_5844_cast_fp16")]; + tensor var_5846_equation_0 = const()[name = tensor("op_5846_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5846_cast_fp16 = einsum(equation = var_5846_equation_0, values = (var_5670_cast_fp16, var_5823_cast_fp16))[name = tensor("op_5846_cast_fp16")]; + tensor var_5848_equation_0 = const()[name = tensor("op_5848_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5848_cast_fp16 = einsum(equation = var_5848_equation_0, values = (var_5674_cast_fp16, var_5824_cast_fp16))[name = tensor("op_5848_cast_fp16")]; + tensor var_5850_equation_0 = const()[name = tensor("op_5850_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5850_cast_fp16 = einsum(equation = var_5850_equation_0, values = (var_5678_cast_fp16, var_5825_cast_fp16))[name = tensor("op_5850_cast_fp16")]; + tensor var_5852_equation_0 = const()[name = tensor("op_5852_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5852_cast_fp16 = einsum(equation = var_5852_equation_0, values = (var_5682_cast_fp16, var_5826_cast_fp16))[name = tensor("op_5852_cast_fp16")]; + tensor var_5854_equation_0 = const()[name = tensor("op_5854_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5854_cast_fp16 = einsum(equation = var_5854_equation_0, values = (var_5686_cast_fp16, var_5827_cast_fp16))[name = tensor("op_5854_cast_fp16")]; + tensor var_5856_equation_0 = const()[name = tensor("op_5856_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5856_cast_fp16 = einsum(equation = var_5856_equation_0, values = (var_5690_cast_fp16, var_5828_cast_fp16))[name = tensor("op_5856_cast_fp16")]; + tensor var_5858_equation_0 = const()[name = tensor("op_5858_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5858_cast_fp16 = einsum(equation = var_5858_equation_0, values = (var_5694_cast_fp16, var_5829_cast_fp16))[name = tensor("op_5858_cast_fp16")]; + tensor var_5860_equation_0 = const()[name = tensor("op_5860_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5860_cast_fp16 = einsum(equation = var_5860_equation_0, values = (var_5698_cast_fp16, var_5830_cast_fp16))[name = tensor("op_5860_cast_fp16")]; + tensor var_5862_equation_0 = const()[name = tensor("op_5862_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5862_cast_fp16 = einsum(equation = var_5862_equation_0, values = (var_5702_cast_fp16, var_5831_cast_fp16))[name = tensor("op_5862_cast_fp16")]; + tensor var_5864_equation_0 = const()[name = tensor("op_5864_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5864_cast_fp16 = einsum(equation = var_5864_equation_0, values = (var_5706_cast_fp16, var_5832_cast_fp16))[name = tensor("op_5864_cast_fp16")]; + tensor var_5866_equation_0 = const()[name = tensor("op_5866_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5866_cast_fp16 = einsum(equation = var_5866_equation_0, values = (var_5710_cast_fp16, var_5833_cast_fp16))[name = tensor("op_5866_cast_fp16")]; + tensor var_5868_equation_0 = const()[name = tensor("op_5868_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5868_cast_fp16 = einsum(equation = var_5868_equation_0, values = (var_5714_cast_fp16, var_5834_cast_fp16))[name = tensor("op_5868_cast_fp16")]; + tensor var_5870_equation_0 = const()[name = tensor("op_5870_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5870_cast_fp16 = einsum(equation = var_5870_equation_0, values = (var_5718_cast_fp16, var_5835_cast_fp16))[name = tensor("op_5870_cast_fp16")]; + tensor var_5872_equation_0 = const()[name = tensor("op_5872_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5872_cast_fp16 = einsum(equation = var_5872_equation_0, values = (var_5722_cast_fp16, var_5836_cast_fp16))[name = tensor("op_5872_cast_fp16")]; + tensor var_5874_equation_0 = const()[name = tensor("op_5874_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5874_cast_fp16 = einsum(equation = var_5874_equation_0, values = (var_5726_cast_fp16, var_5837_cast_fp16))[name = tensor("op_5874_cast_fp16")]; + tensor var_5876_equation_0 = const()[name = tensor("op_5876_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5876_cast_fp16 = einsum(equation = var_5876_equation_0, values = (var_5730_cast_fp16, var_5838_cast_fp16))[name = tensor("op_5876_cast_fp16")]; + tensor var_5878_equation_0 = const()[name = tensor("op_5878_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5878_cast_fp16 = einsum(equation = var_5878_equation_0, values = (var_5734_cast_fp16, var_5839_cast_fp16))[name = tensor("op_5878_cast_fp16")]; + tensor var_5880_equation_0 = const()[name = tensor("op_5880_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5880_cast_fp16 = einsum(equation = var_5880_equation_0, values = (var_5738_cast_fp16, var_5840_cast_fp16))[name = tensor("op_5880_cast_fp16")]; + tensor input_155_interleave_0 = const()[name = tensor("input_155_interleave_0"), val = tensor(false)]; + tensor input_155_cast_fp16 = concat(axis = var_2624, interleave = input_155_interleave_0, values = (var_5842_cast_fp16, var_5844_cast_fp16, var_5846_cast_fp16, var_5848_cast_fp16, var_5850_cast_fp16, var_5852_cast_fp16, var_5854_cast_fp16, var_5856_cast_fp16, var_5858_cast_fp16, var_5860_cast_fp16, var_5862_cast_fp16, var_5864_cast_fp16, var_5866_cast_fp16, var_5868_cast_fp16, var_5870_cast_fp16, var_5872_cast_fp16, var_5874_cast_fp16, var_5876_cast_fp16, var_5878_cast_fp16, var_5880_cast_fp16))[name = tensor("input_155_cast_fp16")]; + tensor var_5886 = const()[name = tensor("op_5886"), val = tensor([1, 1])]; + tensor var_5888 = const()[name = tensor("op_5888"), val = tensor([1, 1])]; + tensor var_5890_pad_type_0 = const()[name = tensor("op_5890_pad_type_0"), val = tensor("custom")]; + tensor var_5890_pad_0 = const()[name = tensor("op_5890_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_3_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_3_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(416540608)))]; + tensor down_blocks_2_attentions_0_transformer_blocks_3_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_3_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(419817472)))]; + tensor var_5890_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_3_attn1_to_out_0_bias_to_fp16, dilations = var_5888, groups = var_2624, pad = var_5890_pad_0, pad_type = var_5890_pad_type_0, strides = var_5886, weight = down_blocks_2_attentions_0_transformer_blocks_3_attn1_to_out_0_weight_to_fp16, x = input_155_cast_fp16)[name = tensor("op_5890_cast_fp16")]; + tensor inputs_45_cast_fp16 = add(x = var_5890_cast_fp16, y = inputs_43_cast_fp16)[name = tensor("inputs_45_cast_fp16")]; + tensor hidden_states_85_axes_0 = const()[name = tensor("hidden_states_85_axes_0"), val = tensor([1])]; + tensor hidden_states_85_gamma_0_to_fp16 = const()[name = tensor("hidden_states_85_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(419820096)))]; + tensor hidden_states_85_beta_0_to_fp16 = const()[name = tensor("hidden_states_85_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(419822720)))]; + tensor var_5900_to_fp16 = const()[name = tensor("op_5900_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_85_cast_fp16 = layer_norm(axes = hidden_states_85_axes_0, beta = hidden_states_85_beta_0_to_fp16, epsilon = var_5900_to_fp16, gamma = hidden_states_85_gamma_0_to_fp16, x = inputs_45_cast_fp16)[name = tensor("hidden_states_85_cast_fp16")]; + tensor var_5915 = const()[name = tensor("op_5915"), val = tensor([1, 1])]; + tensor var_5917 = const()[name = tensor("op_5917"), val = tensor([1, 1])]; + tensor q_31_pad_type_0 = const()[name = tensor("q_31_pad_type_0"), val = tensor("custom")]; + tensor q_31_pad_0 = const()[name = tensor("q_31_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_3_attn2_to_q_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_3_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(419825344)))]; + tensor q_31_cast_fp16 = conv(dilations = var_5917, groups = var_2624, pad = q_31_pad_0, pad_type = q_31_pad_type_0, strides = var_5915, weight = down_blocks_2_attentions_0_transformer_blocks_3_attn2_to_q_weight_to_fp16, x = hidden_states_85_cast_fp16)[name = tensor("q_31_cast_fp16")]; + tensor var_5921 = const()[name = tensor("op_5921"), val = tensor([1, 1])]; + tensor var_5923 = const()[name = tensor("op_5923"), val = tensor([1, 1])]; + tensor k_61_pad_type_0 = const()[name = tensor("k_61_pad_type_0"), val = tensor("custom")]; + tensor k_61_pad_0 = const()[name = tensor("k_61_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_3_attn2_to_k_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_3_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(423102208)))]; + tensor k_61_cast_fp16 = conv(dilations = var_5923, groups = var_2624, pad = k_61_pad_0, pad_type = k_61_pad_type_0, strides = var_5921, weight = down_blocks_2_attentions_0_transformer_blocks_3_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_61_cast_fp16")]; + tensor var_5927 = const()[name = tensor("op_5927"), val = tensor([1, 1])]; + tensor var_5929 = const()[name = tensor("op_5929"), val = tensor([1, 1])]; + tensor v_31_pad_type_0 = const()[name = tensor("v_31_pad_type_0"), val = tensor("custom")]; + tensor v_31_pad_0 = const()[name = tensor("v_31_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_3_attn2_to_v_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_3_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(428345152)))]; + tensor v_31_cast_fp16 = conv(dilations = var_5929, groups = var_2624, pad = v_31_pad_0, pad_type = v_31_pad_type_0, strides = var_5927, weight = down_blocks_2_attentions_0_transformer_blocks_3_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_31_cast_fp16")]; + tensor var_5933_begin_0 = const()[name = tensor("op_5933_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5933_end_0 = const()[name = tensor("op_5933_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_5933_end_mask_0 = const()[name = tensor("op_5933_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5933_cast_fp16 = slice_by_index(begin = var_5933_begin_0, end = var_5933_end_0, end_mask = var_5933_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_5933_cast_fp16")]; + tensor var_5937_begin_0 = const()[name = tensor("op_5937_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_5937_end_0 = const()[name = tensor("op_5937_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_5937_end_mask_0 = const()[name = tensor("op_5937_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5937_cast_fp16 = slice_by_index(begin = var_5937_begin_0, end = var_5937_end_0, end_mask = var_5937_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_5937_cast_fp16")]; + tensor var_5941_begin_0 = const()[name = tensor("op_5941_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_5941_end_0 = const()[name = tensor("op_5941_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_5941_end_mask_0 = const()[name = tensor("op_5941_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5941_cast_fp16 = slice_by_index(begin = var_5941_begin_0, end = var_5941_end_0, end_mask = var_5941_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_5941_cast_fp16")]; + tensor var_5945_begin_0 = const()[name = tensor("op_5945_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_5945_end_0 = const()[name = tensor("op_5945_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_5945_end_mask_0 = const()[name = tensor("op_5945_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5945_cast_fp16 = slice_by_index(begin = var_5945_begin_0, end = var_5945_end_0, end_mask = var_5945_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_5945_cast_fp16")]; + tensor var_5949_begin_0 = const()[name = tensor("op_5949_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_5949_end_0 = const()[name = tensor("op_5949_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_5949_end_mask_0 = const()[name = tensor("op_5949_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5949_cast_fp16 = slice_by_index(begin = var_5949_begin_0, end = var_5949_end_0, end_mask = var_5949_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_5949_cast_fp16")]; + tensor var_5953_begin_0 = const()[name = tensor("op_5953_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_5953_end_0 = const()[name = tensor("op_5953_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_5953_end_mask_0 = const()[name = tensor("op_5953_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5953_cast_fp16 = slice_by_index(begin = var_5953_begin_0, end = var_5953_end_0, end_mask = var_5953_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_5953_cast_fp16")]; + tensor var_5957_begin_0 = const()[name = tensor("op_5957_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_5957_end_0 = const()[name = tensor("op_5957_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_5957_end_mask_0 = const()[name = tensor("op_5957_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5957_cast_fp16 = slice_by_index(begin = var_5957_begin_0, end = var_5957_end_0, end_mask = var_5957_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_5957_cast_fp16")]; + tensor var_5961_begin_0 = const()[name = tensor("op_5961_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_5961_end_0 = const()[name = tensor("op_5961_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_5961_end_mask_0 = const()[name = tensor("op_5961_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5961_cast_fp16 = slice_by_index(begin = var_5961_begin_0, end = var_5961_end_0, end_mask = var_5961_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_5961_cast_fp16")]; + tensor var_5965_begin_0 = const()[name = tensor("op_5965_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_5965_end_0 = const()[name = tensor("op_5965_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_5965_end_mask_0 = const()[name = tensor("op_5965_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5965_cast_fp16 = slice_by_index(begin = var_5965_begin_0, end = var_5965_end_0, end_mask = var_5965_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_5965_cast_fp16")]; + tensor var_5969_begin_0 = const()[name = tensor("op_5969_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_5969_end_0 = const()[name = tensor("op_5969_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_5969_end_mask_0 = const()[name = tensor("op_5969_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5969_cast_fp16 = slice_by_index(begin = var_5969_begin_0, end = var_5969_end_0, end_mask = var_5969_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_5969_cast_fp16")]; + tensor var_5973_begin_0 = const()[name = tensor("op_5973_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_5973_end_0 = const()[name = tensor("op_5973_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_5973_end_mask_0 = const()[name = tensor("op_5973_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5973_cast_fp16 = slice_by_index(begin = var_5973_begin_0, end = var_5973_end_0, end_mask = var_5973_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_5973_cast_fp16")]; + tensor var_5977_begin_0 = const()[name = tensor("op_5977_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_5977_end_0 = const()[name = tensor("op_5977_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_5977_end_mask_0 = const()[name = tensor("op_5977_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5977_cast_fp16 = slice_by_index(begin = var_5977_begin_0, end = var_5977_end_0, end_mask = var_5977_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_5977_cast_fp16")]; + tensor var_5981_begin_0 = const()[name = tensor("op_5981_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_5981_end_0 = const()[name = tensor("op_5981_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_5981_end_mask_0 = const()[name = tensor("op_5981_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5981_cast_fp16 = slice_by_index(begin = var_5981_begin_0, end = var_5981_end_0, end_mask = var_5981_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_5981_cast_fp16")]; + tensor var_5985_begin_0 = const()[name = tensor("op_5985_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_5985_end_0 = const()[name = tensor("op_5985_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_5985_end_mask_0 = const()[name = tensor("op_5985_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5985_cast_fp16 = slice_by_index(begin = var_5985_begin_0, end = var_5985_end_0, end_mask = var_5985_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_5985_cast_fp16")]; + tensor var_5989_begin_0 = const()[name = tensor("op_5989_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_5989_end_0 = const()[name = tensor("op_5989_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_5989_end_mask_0 = const()[name = tensor("op_5989_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5989_cast_fp16 = slice_by_index(begin = var_5989_begin_0, end = var_5989_end_0, end_mask = var_5989_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_5989_cast_fp16")]; + tensor var_5993_begin_0 = const()[name = tensor("op_5993_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_5993_end_0 = const()[name = tensor("op_5993_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_5993_end_mask_0 = const()[name = tensor("op_5993_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5993_cast_fp16 = slice_by_index(begin = var_5993_begin_0, end = var_5993_end_0, end_mask = var_5993_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_5993_cast_fp16")]; + tensor var_5997_begin_0 = const()[name = tensor("op_5997_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_5997_end_0 = const()[name = tensor("op_5997_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_5997_end_mask_0 = const()[name = tensor("op_5997_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5997_cast_fp16 = slice_by_index(begin = var_5997_begin_0, end = var_5997_end_0, end_mask = var_5997_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_5997_cast_fp16")]; + tensor var_6001_begin_0 = const()[name = tensor("op_6001_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_6001_end_0 = const()[name = tensor("op_6001_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_6001_end_mask_0 = const()[name = tensor("op_6001_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6001_cast_fp16 = slice_by_index(begin = var_6001_begin_0, end = var_6001_end_0, end_mask = var_6001_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_6001_cast_fp16")]; + tensor var_6005_begin_0 = const()[name = tensor("op_6005_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_6005_end_0 = const()[name = tensor("op_6005_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_6005_end_mask_0 = const()[name = tensor("op_6005_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6005_cast_fp16 = slice_by_index(begin = var_6005_begin_0, end = var_6005_end_0, end_mask = var_6005_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_6005_cast_fp16")]; + tensor var_6009_begin_0 = const()[name = tensor("op_6009_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_6009_end_0 = const()[name = tensor("op_6009_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_6009_end_mask_0 = const()[name = tensor("op_6009_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6009_cast_fp16 = slice_by_index(begin = var_6009_begin_0, end = var_6009_end_0, end_mask = var_6009_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_6009_cast_fp16")]; + tensor k_63_perm_0 = const()[name = tensor("k_63_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_6016_begin_0 = const()[name = tensor("op_6016_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6016_end_0 = const()[name = tensor("op_6016_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_6016_end_mask_0 = const()[name = tensor("op_6016_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_63_cast_fp16 = transpose(perm = k_63_perm_0, x = k_61_cast_fp16)[name = tensor("transpose_124")]; + tensor var_6016_cast_fp16 = slice_by_index(begin = var_6016_begin_0, end = var_6016_end_0, end_mask = var_6016_end_mask_0, x = k_63_cast_fp16)[name = tensor("op_6016_cast_fp16")]; + tensor var_6020_begin_0 = const()[name = tensor("op_6020_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_6020_end_0 = const()[name = tensor("op_6020_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_6020_end_mask_0 = const()[name = tensor("op_6020_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6020_cast_fp16 = slice_by_index(begin = var_6020_begin_0, end = var_6020_end_0, end_mask = var_6020_end_mask_0, x = k_63_cast_fp16)[name = tensor("op_6020_cast_fp16")]; + tensor var_6024_begin_0 = const()[name = tensor("op_6024_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_6024_end_0 = const()[name = tensor("op_6024_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_6024_end_mask_0 = const()[name = tensor("op_6024_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6024_cast_fp16 = slice_by_index(begin = var_6024_begin_0, end = var_6024_end_0, end_mask = var_6024_end_mask_0, x = k_63_cast_fp16)[name = tensor("op_6024_cast_fp16")]; + tensor var_6028_begin_0 = const()[name = tensor("op_6028_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_6028_end_0 = const()[name = tensor("op_6028_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_6028_end_mask_0 = const()[name = tensor("op_6028_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6028_cast_fp16 = slice_by_index(begin = var_6028_begin_0, end = var_6028_end_0, end_mask = var_6028_end_mask_0, x = k_63_cast_fp16)[name = tensor("op_6028_cast_fp16")]; + tensor var_6032_begin_0 = const()[name = tensor("op_6032_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_6032_end_0 = const()[name = tensor("op_6032_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_6032_end_mask_0 = const()[name = tensor("op_6032_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6032_cast_fp16 = slice_by_index(begin = var_6032_begin_0, end = var_6032_end_0, end_mask = var_6032_end_mask_0, x = k_63_cast_fp16)[name = tensor("op_6032_cast_fp16")]; + tensor var_6036_begin_0 = const()[name = tensor("op_6036_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_6036_end_0 = const()[name = tensor("op_6036_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_6036_end_mask_0 = const()[name = tensor("op_6036_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6036_cast_fp16 = slice_by_index(begin = var_6036_begin_0, end = var_6036_end_0, end_mask = var_6036_end_mask_0, x = k_63_cast_fp16)[name = tensor("op_6036_cast_fp16")]; + tensor var_6040_begin_0 = const()[name = tensor("op_6040_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_6040_end_0 = const()[name = tensor("op_6040_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_6040_end_mask_0 = const()[name = tensor("op_6040_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6040_cast_fp16 = slice_by_index(begin = var_6040_begin_0, end = var_6040_end_0, end_mask = var_6040_end_mask_0, x = k_63_cast_fp16)[name = tensor("op_6040_cast_fp16")]; + tensor var_6044_begin_0 = const()[name = tensor("op_6044_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_6044_end_0 = const()[name = tensor("op_6044_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_6044_end_mask_0 = const()[name = tensor("op_6044_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6044_cast_fp16 = slice_by_index(begin = var_6044_begin_0, end = var_6044_end_0, end_mask = var_6044_end_mask_0, x = k_63_cast_fp16)[name = tensor("op_6044_cast_fp16")]; + tensor var_6048_begin_0 = const()[name = tensor("op_6048_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_6048_end_0 = const()[name = tensor("op_6048_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_6048_end_mask_0 = const()[name = tensor("op_6048_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6048_cast_fp16 = slice_by_index(begin = var_6048_begin_0, end = var_6048_end_0, end_mask = var_6048_end_mask_0, x = k_63_cast_fp16)[name = tensor("op_6048_cast_fp16")]; + tensor var_6052_begin_0 = const()[name = tensor("op_6052_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_6052_end_0 = const()[name = tensor("op_6052_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_6052_end_mask_0 = const()[name = tensor("op_6052_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6052_cast_fp16 = slice_by_index(begin = var_6052_begin_0, end = var_6052_end_0, end_mask = var_6052_end_mask_0, x = k_63_cast_fp16)[name = tensor("op_6052_cast_fp16")]; + tensor var_6056_begin_0 = const()[name = tensor("op_6056_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_6056_end_0 = const()[name = tensor("op_6056_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_6056_end_mask_0 = const()[name = tensor("op_6056_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6056_cast_fp16 = slice_by_index(begin = var_6056_begin_0, end = var_6056_end_0, end_mask = var_6056_end_mask_0, x = k_63_cast_fp16)[name = tensor("op_6056_cast_fp16")]; + tensor var_6060_begin_0 = const()[name = tensor("op_6060_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_6060_end_0 = const()[name = tensor("op_6060_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_6060_end_mask_0 = const()[name = tensor("op_6060_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6060_cast_fp16 = slice_by_index(begin = var_6060_begin_0, end = var_6060_end_0, end_mask = var_6060_end_mask_0, x = k_63_cast_fp16)[name = tensor("op_6060_cast_fp16")]; + tensor var_6064_begin_0 = const()[name = tensor("op_6064_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_6064_end_0 = const()[name = tensor("op_6064_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_6064_end_mask_0 = const()[name = tensor("op_6064_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6064_cast_fp16 = slice_by_index(begin = var_6064_begin_0, end = var_6064_end_0, end_mask = var_6064_end_mask_0, x = k_63_cast_fp16)[name = tensor("op_6064_cast_fp16")]; + tensor var_6068_begin_0 = const()[name = tensor("op_6068_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_6068_end_0 = const()[name = tensor("op_6068_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_6068_end_mask_0 = const()[name = tensor("op_6068_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6068_cast_fp16 = slice_by_index(begin = var_6068_begin_0, end = var_6068_end_0, end_mask = var_6068_end_mask_0, x = k_63_cast_fp16)[name = tensor("op_6068_cast_fp16")]; + tensor var_6072_begin_0 = const()[name = tensor("op_6072_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_6072_end_0 = const()[name = tensor("op_6072_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_6072_end_mask_0 = const()[name = tensor("op_6072_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6072_cast_fp16 = slice_by_index(begin = var_6072_begin_0, end = var_6072_end_0, end_mask = var_6072_end_mask_0, x = k_63_cast_fp16)[name = tensor("op_6072_cast_fp16")]; + tensor var_6076_begin_0 = const()[name = tensor("op_6076_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_6076_end_0 = const()[name = tensor("op_6076_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_6076_end_mask_0 = const()[name = tensor("op_6076_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6076_cast_fp16 = slice_by_index(begin = var_6076_begin_0, end = var_6076_end_0, end_mask = var_6076_end_mask_0, x = k_63_cast_fp16)[name = tensor("op_6076_cast_fp16")]; + tensor var_6080_begin_0 = const()[name = tensor("op_6080_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_6080_end_0 = const()[name = tensor("op_6080_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_6080_end_mask_0 = const()[name = tensor("op_6080_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6080_cast_fp16 = slice_by_index(begin = var_6080_begin_0, end = var_6080_end_0, end_mask = var_6080_end_mask_0, x = k_63_cast_fp16)[name = tensor("op_6080_cast_fp16")]; + tensor var_6084_begin_0 = const()[name = tensor("op_6084_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_6084_end_0 = const()[name = tensor("op_6084_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_6084_end_mask_0 = const()[name = tensor("op_6084_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6084_cast_fp16 = slice_by_index(begin = var_6084_begin_0, end = var_6084_end_0, end_mask = var_6084_end_mask_0, x = k_63_cast_fp16)[name = tensor("op_6084_cast_fp16")]; + tensor var_6088_begin_0 = const()[name = tensor("op_6088_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_6088_end_0 = const()[name = tensor("op_6088_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_6088_end_mask_0 = const()[name = tensor("op_6088_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6088_cast_fp16 = slice_by_index(begin = var_6088_begin_0, end = var_6088_end_0, end_mask = var_6088_end_mask_0, x = k_63_cast_fp16)[name = tensor("op_6088_cast_fp16")]; + tensor var_6092_begin_0 = const()[name = tensor("op_6092_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_6092_end_0 = const()[name = tensor("op_6092_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_6092_end_mask_0 = const()[name = tensor("op_6092_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6092_cast_fp16 = slice_by_index(begin = var_6092_begin_0, end = var_6092_end_0, end_mask = var_6092_end_mask_0, x = k_63_cast_fp16)[name = tensor("op_6092_cast_fp16")]; + tensor var_6094_begin_0 = const()[name = tensor("op_6094_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6094_end_0 = const()[name = tensor("op_6094_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_6094_end_mask_0 = const()[name = tensor("op_6094_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6094_cast_fp16 = slice_by_index(begin = var_6094_begin_0, end = var_6094_end_0, end_mask = var_6094_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6094_cast_fp16")]; + tensor var_6098_begin_0 = const()[name = tensor("op_6098_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_6098_end_0 = const()[name = tensor("op_6098_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_6098_end_mask_0 = const()[name = tensor("op_6098_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6098_cast_fp16 = slice_by_index(begin = var_6098_begin_0, end = var_6098_end_0, end_mask = var_6098_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6098_cast_fp16")]; + tensor var_6102_begin_0 = const()[name = tensor("op_6102_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_6102_end_0 = const()[name = tensor("op_6102_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_6102_end_mask_0 = const()[name = tensor("op_6102_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6102_cast_fp16 = slice_by_index(begin = var_6102_begin_0, end = var_6102_end_0, end_mask = var_6102_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6102_cast_fp16")]; + tensor var_6106_begin_0 = const()[name = tensor("op_6106_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_6106_end_0 = const()[name = tensor("op_6106_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_6106_end_mask_0 = const()[name = tensor("op_6106_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6106_cast_fp16 = slice_by_index(begin = var_6106_begin_0, end = var_6106_end_0, end_mask = var_6106_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6106_cast_fp16")]; + tensor var_6110_begin_0 = const()[name = tensor("op_6110_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_6110_end_0 = const()[name = tensor("op_6110_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_6110_end_mask_0 = const()[name = tensor("op_6110_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6110_cast_fp16 = slice_by_index(begin = var_6110_begin_0, end = var_6110_end_0, end_mask = var_6110_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6110_cast_fp16")]; + tensor var_6114_begin_0 = const()[name = tensor("op_6114_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_6114_end_0 = const()[name = tensor("op_6114_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_6114_end_mask_0 = const()[name = tensor("op_6114_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6114_cast_fp16 = slice_by_index(begin = var_6114_begin_0, end = var_6114_end_0, end_mask = var_6114_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6114_cast_fp16")]; + tensor var_6118_begin_0 = const()[name = tensor("op_6118_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_6118_end_0 = const()[name = tensor("op_6118_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_6118_end_mask_0 = const()[name = tensor("op_6118_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6118_cast_fp16 = slice_by_index(begin = var_6118_begin_0, end = var_6118_end_0, end_mask = var_6118_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6118_cast_fp16")]; + tensor var_6122_begin_0 = const()[name = tensor("op_6122_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_6122_end_0 = const()[name = tensor("op_6122_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_6122_end_mask_0 = const()[name = tensor("op_6122_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6122_cast_fp16 = slice_by_index(begin = var_6122_begin_0, end = var_6122_end_0, end_mask = var_6122_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6122_cast_fp16")]; + tensor var_6126_begin_0 = const()[name = tensor("op_6126_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_6126_end_0 = const()[name = tensor("op_6126_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_6126_end_mask_0 = const()[name = tensor("op_6126_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6126_cast_fp16 = slice_by_index(begin = var_6126_begin_0, end = var_6126_end_0, end_mask = var_6126_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6126_cast_fp16")]; + tensor var_6130_begin_0 = const()[name = tensor("op_6130_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_6130_end_0 = const()[name = tensor("op_6130_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_6130_end_mask_0 = const()[name = tensor("op_6130_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6130_cast_fp16 = slice_by_index(begin = var_6130_begin_0, end = var_6130_end_0, end_mask = var_6130_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6130_cast_fp16")]; + tensor var_6134_begin_0 = const()[name = tensor("op_6134_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_6134_end_0 = const()[name = tensor("op_6134_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_6134_end_mask_0 = const()[name = tensor("op_6134_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6134_cast_fp16 = slice_by_index(begin = var_6134_begin_0, end = var_6134_end_0, end_mask = var_6134_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6134_cast_fp16")]; + tensor var_6138_begin_0 = const()[name = tensor("op_6138_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_6138_end_0 = const()[name = tensor("op_6138_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_6138_end_mask_0 = const()[name = tensor("op_6138_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6138_cast_fp16 = slice_by_index(begin = var_6138_begin_0, end = var_6138_end_0, end_mask = var_6138_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6138_cast_fp16")]; + tensor var_6142_begin_0 = const()[name = tensor("op_6142_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_6142_end_0 = const()[name = tensor("op_6142_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_6142_end_mask_0 = const()[name = tensor("op_6142_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6142_cast_fp16 = slice_by_index(begin = var_6142_begin_0, end = var_6142_end_0, end_mask = var_6142_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6142_cast_fp16")]; + tensor var_6146_begin_0 = const()[name = tensor("op_6146_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_6146_end_0 = const()[name = tensor("op_6146_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_6146_end_mask_0 = const()[name = tensor("op_6146_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6146_cast_fp16 = slice_by_index(begin = var_6146_begin_0, end = var_6146_end_0, end_mask = var_6146_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6146_cast_fp16")]; + tensor var_6150_begin_0 = const()[name = tensor("op_6150_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_6150_end_0 = const()[name = tensor("op_6150_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_6150_end_mask_0 = const()[name = tensor("op_6150_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6150_cast_fp16 = slice_by_index(begin = var_6150_begin_0, end = var_6150_end_0, end_mask = var_6150_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6150_cast_fp16")]; + tensor var_6154_begin_0 = const()[name = tensor("op_6154_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_6154_end_0 = const()[name = tensor("op_6154_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_6154_end_mask_0 = const()[name = tensor("op_6154_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6154_cast_fp16 = slice_by_index(begin = var_6154_begin_0, end = var_6154_end_0, end_mask = var_6154_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6154_cast_fp16")]; + tensor var_6158_begin_0 = const()[name = tensor("op_6158_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_6158_end_0 = const()[name = tensor("op_6158_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_6158_end_mask_0 = const()[name = tensor("op_6158_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6158_cast_fp16 = slice_by_index(begin = var_6158_begin_0, end = var_6158_end_0, end_mask = var_6158_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6158_cast_fp16")]; + tensor var_6162_begin_0 = const()[name = tensor("op_6162_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_6162_end_0 = const()[name = tensor("op_6162_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_6162_end_mask_0 = const()[name = tensor("op_6162_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6162_cast_fp16 = slice_by_index(begin = var_6162_begin_0, end = var_6162_end_0, end_mask = var_6162_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6162_cast_fp16")]; + tensor var_6166_begin_0 = const()[name = tensor("op_6166_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_6166_end_0 = const()[name = tensor("op_6166_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_6166_end_mask_0 = const()[name = tensor("op_6166_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6166_cast_fp16 = slice_by_index(begin = var_6166_begin_0, end = var_6166_end_0, end_mask = var_6166_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6166_cast_fp16")]; + tensor var_6170_begin_0 = const()[name = tensor("op_6170_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_6170_end_0 = const()[name = tensor("op_6170_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_6170_end_mask_0 = const()[name = tensor("op_6170_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6170_cast_fp16 = slice_by_index(begin = var_6170_begin_0, end = var_6170_end_0, end_mask = var_6170_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6170_cast_fp16")]; + tensor var_6174_equation_0 = const()[name = tensor("op_6174_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6174_cast_fp16 = einsum(equation = var_6174_equation_0, values = (var_6016_cast_fp16, var_5933_cast_fp16))[name = tensor("op_6174_cast_fp16")]; + tensor var_6175_to_fp16 = const()[name = tensor("op_6175_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_441_cast_fp16 = mul(x = var_6174_cast_fp16, y = var_6175_to_fp16)[name = tensor("aw_441_cast_fp16")]; + tensor var_6178_equation_0 = const()[name = tensor("op_6178_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6178_cast_fp16 = einsum(equation = var_6178_equation_0, values = (var_6020_cast_fp16, var_5937_cast_fp16))[name = tensor("op_6178_cast_fp16")]; + tensor var_6179_to_fp16 = const()[name = tensor("op_6179_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_443_cast_fp16 = mul(x = var_6178_cast_fp16, y = var_6179_to_fp16)[name = tensor("aw_443_cast_fp16")]; + tensor var_6182_equation_0 = const()[name = tensor("op_6182_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6182_cast_fp16 = einsum(equation = var_6182_equation_0, values = (var_6024_cast_fp16, var_5941_cast_fp16))[name = tensor("op_6182_cast_fp16")]; + tensor var_6183_to_fp16 = const()[name = tensor("op_6183_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_445_cast_fp16 = mul(x = var_6182_cast_fp16, y = var_6183_to_fp16)[name = tensor("aw_445_cast_fp16")]; + tensor var_6186_equation_0 = const()[name = tensor("op_6186_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6186_cast_fp16 = einsum(equation = var_6186_equation_0, values = (var_6028_cast_fp16, var_5945_cast_fp16))[name = tensor("op_6186_cast_fp16")]; + tensor var_6187_to_fp16 = const()[name = tensor("op_6187_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_447_cast_fp16 = mul(x = var_6186_cast_fp16, y = var_6187_to_fp16)[name = tensor("aw_447_cast_fp16")]; + tensor var_6190_equation_0 = const()[name = tensor("op_6190_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6190_cast_fp16 = einsum(equation = var_6190_equation_0, values = (var_6032_cast_fp16, var_5949_cast_fp16))[name = tensor("op_6190_cast_fp16")]; + tensor var_6191_to_fp16 = const()[name = tensor("op_6191_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_449_cast_fp16 = mul(x = var_6190_cast_fp16, y = var_6191_to_fp16)[name = tensor("aw_449_cast_fp16")]; + tensor var_6194_equation_0 = const()[name = tensor("op_6194_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6194_cast_fp16 = einsum(equation = var_6194_equation_0, values = (var_6036_cast_fp16, var_5953_cast_fp16))[name = tensor("op_6194_cast_fp16")]; + tensor var_6195_to_fp16 = const()[name = tensor("op_6195_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_451_cast_fp16 = mul(x = var_6194_cast_fp16, y = var_6195_to_fp16)[name = tensor("aw_451_cast_fp16")]; + tensor var_6198_equation_0 = const()[name = tensor("op_6198_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6198_cast_fp16 = einsum(equation = var_6198_equation_0, values = (var_6040_cast_fp16, var_5957_cast_fp16))[name = tensor("op_6198_cast_fp16")]; + tensor var_6199_to_fp16 = const()[name = tensor("op_6199_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_453_cast_fp16 = mul(x = var_6198_cast_fp16, y = var_6199_to_fp16)[name = tensor("aw_453_cast_fp16")]; + tensor var_6202_equation_0 = const()[name = tensor("op_6202_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6202_cast_fp16 = einsum(equation = var_6202_equation_0, values = (var_6044_cast_fp16, var_5961_cast_fp16))[name = tensor("op_6202_cast_fp16")]; + tensor var_6203_to_fp16 = const()[name = tensor("op_6203_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_455_cast_fp16 = mul(x = var_6202_cast_fp16, y = var_6203_to_fp16)[name = tensor("aw_455_cast_fp16")]; + tensor var_6206_equation_0 = const()[name = tensor("op_6206_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6206_cast_fp16 = einsum(equation = var_6206_equation_0, values = (var_6048_cast_fp16, var_5965_cast_fp16))[name = tensor("op_6206_cast_fp16")]; + tensor var_6207_to_fp16 = const()[name = tensor("op_6207_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_457_cast_fp16 = mul(x = var_6206_cast_fp16, y = var_6207_to_fp16)[name = tensor("aw_457_cast_fp16")]; + tensor var_6210_equation_0 = const()[name = tensor("op_6210_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6210_cast_fp16 = einsum(equation = var_6210_equation_0, values = (var_6052_cast_fp16, var_5969_cast_fp16))[name = tensor("op_6210_cast_fp16")]; + tensor var_6211_to_fp16 = const()[name = tensor("op_6211_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_459_cast_fp16 = mul(x = var_6210_cast_fp16, y = var_6211_to_fp16)[name = tensor("aw_459_cast_fp16")]; + tensor var_6214_equation_0 = const()[name = tensor("op_6214_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6214_cast_fp16 = einsum(equation = var_6214_equation_0, values = (var_6056_cast_fp16, var_5973_cast_fp16))[name = tensor("op_6214_cast_fp16")]; + tensor var_6215_to_fp16 = const()[name = tensor("op_6215_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_461_cast_fp16 = mul(x = var_6214_cast_fp16, y = var_6215_to_fp16)[name = tensor("aw_461_cast_fp16")]; + tensor var_6218_equation_0 = const()[name = tensor("op_6218_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6218_cast_fp16 = einsum(equation = var_6218_equation_0, values = (var_6060_cast_fp16, var_5977_cast_fp16))[name = tensor("op_6218_cast_fp16")]; + tensor var_6219_to_fp16 = const()[name = tensor("op_6219_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_463_cast_fp16 = mul(x = var_6218_cast_fp16, y = var_6219_to_fp16)[name = tensor("aw_463_cast_fp16")]; + tensor var_6222_equation_0 = const()[name = tensor("op_6222_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6222_cast_fp16 = einsum(equation = var_6222_equation_0, values = (var_6064_cast_fp16, var_5981_cast_fp16))[name = tensor("op_6222_cast_fp16")]; + tensor var_6223_to_fp16 = const()[name = tensor("op_6223_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_465_cast_fp16 = mul(x = var_6222_cast_fp16, y = var_6223_to_fp16)[name = tensor("aw_465_cast_fp16")]; + tensor var_6226_equation_0 = const()[name = tensor("op_6226_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6226_cast_fp16 = einsum(equation = var_6226_equation_0, values = (var_6068_cast_fp16, var_5985_cast_fp16))[name = tensor("op_6226_cast_fp16")]; + tensor var_6227_to_fp16 = const()[name = tensor("op_6227_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_467_cast_fp16 = mul(x = var_6226_cast_fp16, y = var_6227_to_fp16)[name = tensor("aw_467_cast_fp16")]; + tensor var_6230_equation_0 = const()[name = tensor("op_6230_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6230_cast_fp16 = einsum(equation = var_6230_equation_0, values = (var_6072_cast_fp16, var_5989_cast_fp16))[name = tensor("op_6230_cast_fp16")]; + tensor var_6231_to_fp16 = const()[name = tensor("op_6231_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_469_cast_fp16 = mul(x = var_6230_cast_fp16, y = var_6231_to_fp16)[name = tensor("aw_469_cast_fp16")]; + tensor var_6234_equation_0 = const()[name = tensor("op_6234_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6234_cast_fp16 = einsum(equation = var_6234_equation_0, values = (var_6076_cast_fp16, var_5993_cast_fp16))[name = tensor("op_6234_cast_fp16")]; + tensor var_6235_to_fp16 = const()[name = tensor("op_6235_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_471_cast_fp16 = mul(x = var_6234_cast_fp16, y = var_6235_to_fp16)[name = tensor("aw_471_cast_fp16")]; + tensor var_6238_equation_0 = const()[name = tensor("op_6238_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6238_cast_fp16 = einsum(equation = var_6238_equation_0, values = (var_6080_cast_fp16, var_5997_cast_fp16))[name = tensor("op_6238_cast_fp16")]; + tensor var_6239_to_fp16 = const()[name = tensor("op_6239_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_473_cast_fp16 = mul(x = var_6238_cast_fp16, y = var_6239_to_fp16)[name = tensor("aw_473_cast_fp16")]; + tensor var_6242_equation_0 = const()[name = tensor("op_6242_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6242_cast_fp16 = einsum(equation = var_6242_equation_0, values = (var_6084_cast_fp16, var_6001_cast_fp16))[name = tensor("op_6242_cast_fp16")]; + tensor var_6243_to_fp16 = const()[name = tensor("op_6243_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_475_cast_fp16 = mul(x = var_6242_cast_fp16, y = var_6243_to_fp16)[name = tensor("aw_475_cast_fp16")]; + tensor var_6246_equation_0 = const()[name = tensor("op_6246_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6246_cast_fp16 = einsum(equation = var_6246_equation_0, values = (var_6088_cast_fp16, var_6005_cast_fp16))[name = tensor("op_6246_cast_fp16")]; + tensor var_6247_to_fp16 = const()[name = tensor("op_6247_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_477_cast_fp16 = mul(x = var_6246_cast_fp16, y = var_6247_to_fp16)[name = tensor("aw_477_cast_fp16")]; + tensor var_6250_equation_0 = const()[name = tensor("op_6250_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6250_cast_fp16 = einsum(equation = var_6250_equation_0, values = (var_6092_cast_fp16, var_6009_cast_fp16))[name = tensor("op_6250_cast_fp16")]; + tensor var_6251_to_fp16 = const()[name = tensor("op_6251_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_479_cast_fp16 = mul(x = var_6250_cast_fp16, y = var_6251_to_fp16)[name = tensor("aw_479_cast_fp16")]; + tensor var_6253_cast_fp16 = softmax(axis = var_2624, x = aw_441_cast_fp16)[name = tensor("op_6253_cast_fp16")]; + tensor var_6254_cast_fp16 = softmax(axis = var_2624, x = aw_443_cast_fp16)[name = tensor("op_6254_cast_fp16")]; + tensor var_6255_cast_fp16 = softmax(axis = var_2624, x = aw_445_cast_fp16)[name = tensor("op_6255_cast_fp16")]; + tensor var_6256_cast_fp16 = softmax(axis = var_2624, x = aw_447_cast_fp16)[name = tensor("op_6256_cast_fp16")]; + tensor var_6257_cast_fp16 = softmax(axis = var_2624, x = aw_449_cast_fp16)[name = tensor("op_6257_cast_fp16")]; + tensor var_6258_cast_fp16 = softmax(axis = var_2624, x = aw_451_cast_fp16)[name = tensor("op_6258_cast_fp16")]; + tensor var_6259_cast_fp16 = softmax(axis = var_2624, x = aw_453_cast_fp16)[name = tensor("op_6259_cast_fp16")]; + tensor var_6260_cast_fp16 = softmax(axis = var_2624, x = aw_455_cast_fp16)[name = tensor("op_6260_cast_fp16")]; + tensor var_6261_cast_fp16 = softmax(axis = var_2624, x = aw_457_cast_fp16)[name = tensor("op_6261_cast_fp16")]; + tensor var_6262_cast_fp16 = softmax(axis = var_2624, x = aw_459_cast_fp16)[name = tensor("op_6262_cast_fp16")]; + tensor var_6263_cast_fp16 = softmax(axis = var_2624, x = aw_461_cast_fp16)[name = tensor("op_6263_cast_fp16")]; + tensor var_6264_cast_fp16 = softmax(axis = var_2624, x = aw_463_cast_fp16)[name = tensor("op_6264_cast_fp16")]; + tensor var_6265_cast_fp16 = softmax(axis = var_2624, x = aw_465_cast_fp16)[name = tensor("op_6265_cast_fp16")]; + tensor var_6266_cast_fp16 = softmax(axis = var_2624, x = aw_467_cast_fp16)[name = tensor("op_6266_cast_fp16")]; + tensor var_6267_cast_fp16 = softmax(axis = var_2624, x = aw_469_cast_fp16)[name = tensor("op_6267_cast_fp16")]; + tensor var_6268_cast_fp16 = softmax(axis = var_2624, x = aw_471_cast_fp16)[name = tensor("op_6268_cast_fp16")]; + tensor var_6269_cast_fp16 = softmax(axis = var_2624, x = aw_473_cast_fp16)[name = tensor("op_6269_cast_fp16")]; + tensor var_6270_cast_fp16 = softmax(axis = var_2624, x = aw_475_cast_fp16)[name = tensor("op_6270_cast_fp16")]; + tensor var_6271_cast_fp16 = softmax(axis = var_2624, x = aw_477_cast_fp16)[name = tensor("op_6271_cast_fp16")]; + tensor var_6272_cast_fp16 = softmax(axis = var_2624, x = aw_479_cast_fp16)[name = tensor("op_6272_cast_fp16")]; + tensor var_6274_equation_0 = const()[name = tensor("op_6274_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6274_cast_fp16 = einsum(equation = var_6274_equation_0, values = (var_6094_cast_fp16, var_6253_cast_fp16))[name = tensor("op_6274_cast_fp16")]; + tensor var_6276_equation_0 = const()[name = tensor("op_6276_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6276_cast_fp16 = einsum(equation = var_6276_equation_0, values = (var_6098_cast_fp16, var_6254_cast_fp16))[name = tensor("op_6276_cast_fp16")]; + tensor var_6278_equation_0 = const()[name = tensor("op_6278_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6278_cast_fp16 = einsum(equation = var_6278_equation_0, values = (var_6102_cast_fp16, var_6255_cast_fp16))[name = tensor("op_6278_cast_fp16")]; + tensor var_6280_equation_0 = const()[name = tensor("op_6280_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6280_cast_fp16 = einsum(equation = var_6280_equation_0, values = (var_6106_cast_fp16, var_6256_cast_fp16))[name = tensor("op_6280_cast_fp16")]; + tensor var_6282_equation_0 = const()[name = tensor("op_6282_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6282_cast_fp16 = einsum(equation = var_6282_equation_0, values = (var_6110_cast_fp16, var_6257_cast_fp16))[name = tensor("op_6282_cast_fp16")]; + tensor var_6284_equation_0 = const()[name = tensor("op_6284_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6284_cast_fp16 = einsum(equation = var_6284_equation_0, values = (var_6114_cast_fp16, var_6258_cast_fp16))[name = tensor("op_6284_cast_fp16")]; + tensor var_6286_equation_0 = const()[name = tensor("op_6286_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6286_cast_fp16 = einsum(equation = var_6286_equation_0, values = (var_6118_cast_fp16, var_6259_cast_fp16))[name = tensor("op_6286_cast_fp16")]; + tensor var_6288_equation_0 = const()[name = tensor("op_6288_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6288_cast_fp16 = einsum(equation = var_6288_equation_0, values = (var_6122_cast_fp16, var_6260_cast_fp16))[name = tensor("op_6288_cast_fp16")]; + tensor var_6290_equation_0 = const()[name = tensor("op_6290_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6290_cast_fp16 = einsum(equation = var_6290_equation_0, values = (var_6126_cast_fp16, var_6261_cast_fp16))[name = tensor("op_6290_cast_fp16")]; + tensor var_6292_equation_0 = const()[name = tensor("op_6292_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6292_cast_fp16 = einsum(equation = var_6292_equation_0, values = (var_6130_cast_fp16, var_6262_cast_fp16))[name = tensor("op_6292_cast_fp16")]; + tensor var_6294_equation_0 = const()[name = tensor("op_6294_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6294_cast_fp16 = einsum(equation = var_6294_equation_0, values = (var_6134_cast_fp16, var_6263_cast_fp16))[name = tensor("op_6294_cast_fp16")]; + tensor var_6296_equation_0 = const()[name = tensor("op_6296_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6296_cast_fp16 = einsum(equation = var_6296_equation_0, values = (var_6138_cast_fp16, var_6264_cast_fp16))[name = tensor("op_6296_cast_fp16")]; + tensor var_6298_equation_0 = const()[name = tensor("op_6298_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6298_cast_fp16 = einsum(equation = var_6298_equation_0, values = (var_6142_cast_fp16, var_6265_cast_fp16))[name = tensor("op_6298_cast_fp16")]; + tensor var_6300_equation_0 = const()[name = tensor("op_6300_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6300_cast_fp16 = einsum(equation = var_6300_equation_0, values = (var_6146_cast_fp16, var_6266_cast_fp16))[name = tensor("op_6300_cast_fp16")]; + tensor var_6302_equation_0 = const()[name = tensor("op_6302_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6302_cast_fp16 = einsum(equation = var_6302_equation_0, values = (var_6150_cast_fp16, var_6267_cast_fp16))[name = tensor("op_6302_cast_fp16")]; + tensor var_6304_equation_0 = const()[name = tensor("op_6304_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6304_cast_fp16 = einsum(equation = var_6304_equation_0, values = (var_6154_cast_fp16, var_6268_cast_fp16))[name = tensor("op_6304_cast_fp16")]; + tensor var_6306_equation_0 = const()[name = tensor("op_6306_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6306_cast_fp16 = einsum(equation = var_6306_equation_0, values = (var_6158_cast_fp16, var_6269_cast_fp16))[name = tensor("op_6306_cast_fp16")]; + tensor var_6308_equation_0 = const()[name = tensor("op_6308_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6308_cast_fp16 = einsum(equation = var_6308_equation_0, values = (var_6162_cast_fp16, var_6270_cast_fp16))[name = tensor("op_6308_cast_fp16")]; + tensor var_6310_equation_0 = const()[name = tensor("op_6310_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6310_cast_fp16 = einsum(equation = var_6310_equation_0, values = (var_6166_cast_fp16, var_6271_cast_fp16))[name = tensor("op_6310_cast_fp16")]; + tensor var_6312_equation_0 = const()[name = tensor("op_6312_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6312_cast_fp16 = einsum(equation = var_6312_equation_0, values = (var_6170_cast_fp16, var_6272_cast_fp16))[name = tensor("op_6312_cast_fp16")]; + tensor input_157_interleave_0 = const()[name = tensor("input_157_interleave_0"), val = tensor(false)]; + tensor input_157_cast_fp16 = concat(axis = var_2624, interleave = input_157_interleave_0, values = (var_6274_cast_fp16, var_6276_cast_fp16, var_6278_cast_fp16, var_6280_cast_fp16, var_6282_cast_fp16, var_6284_cast_fp16, var_6286_cast_fp16, var_6288_cast_fp16, var_6290_cast_fp16, var_6292_cast_fp16, var_6294_cast_fp16, var_6296_cast_fp16, var_6298_cast_fp16, var_6300_cast_fp16, var_6302_cast_fp16, var_6304_cast_fp16, var_6306_cast_fp16, var_6308_cast_fp16, var_6310_cast_fp16, var_6312_cast_fp16))[name = tensor("input_157_cast_fp16")]; + tensor var_6318 = const()[name = tensor("op_6318"), val = tensor([1, 1])]; + tensor var_6320 = const()[name = tensor("op_6320"), val = tensor([1, 1])]; + tensor var_6322_pad_type_0 = const()[name = tensor("op_6322_pad_type_0"), val = tensor("custom")]; + tensor var_6322_pad_0 = const()[name = tensor("op_6322_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_3_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_3_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(433588096)))]; + tensor down_blocks_2_attentions_0_transformer_blocks_3_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_3_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(436864960)))]; + tensor var_6322_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_3_attn2_to_out_0_bias_to_fp16, dilations = var_6320, groups = var_2624, pad = var_6322_pad_0, pad_type = var_6322_pad_type_0, strides = var_6318, weight = down_blocks_2_attentions_0_transformer_blocks_3_attn2_to_out_0_weight_to_fp16, x = input_157_cast_fp16)[name = tensor("op_6322_cast_fp16")]; + tensor inputs_47_cast_fp16 = add(x = var_6322_cast_fp16, y = inputs_45_cast_fp16)[name = tensor("inputs_47_cast_fp16")]; + tensor input_159_axes_0 = const()[name = tensor("input_159_axes_0"), val = tensor([1])]; + tensor input_159_gamma_0_to_fp16 = const()[name = tensor("input_159_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(436867584)))]; + tensor input_159_beta_0_to_fp16 = const()[name = tensor("input_159_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(436870208)))]; + tensor var_6332_to_fp16 = const()[name = tensor("op_6332_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_159_cast_fp16 = layer_norm(axes = input_159_axes_0, beta = input_159_beta_0_to_fp16, epsilon = var_6332_to_fp16, gamma = input_159_gamma_0_to_fp16, x = inputs_47_cast_fp16)[name = tensor("input_159_cast_fp16")]; + tensor var_6348 = const()[name = tensor("op_6348"), val = tensor([1, 1])]; + tensor var_6350 = const()[name = tensor("op_6350"), val = tensor([1, 1])]; + tensor var_6352_pad_type_0 = const()[name = tensor("op_6352_pad_type_0"), val = tensor("custom")]; + tensor var_6352_pad_0 = const()[name = tensor("op_6352_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_3_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_3_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(436872832)))]; + tensor down_blocks_2_attentions_0_transformer_blocks_3_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_3_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(463087296)))]; + tensor var_6352_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_3_ff_net_0_proj_bias_to_fp16, dilations = var_6350, groups = var_2624, pad = var_6352_pad_0, pad_type = var_6352_pad_type_0, strides = var_6348, weight = down_blocks_2_attentions_0_transformer_blocks_3_ff_net_0_proj_weight_to_fp16, x = input_159_cast_fp16)[name = tensor("op_6352_cast_fp16")]; + tensor var_6353_split_sizes_0 = const()[name = tensor("op_6353_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_6353_axis_0 = const()[name = tensor("op_6353_axis_0"), val = tensor(1)]; + tensor var_6353_cast_fp16_0, tensor var_6353_cast_fp16_1 = split(axis = var_6353_axis_0, split_sizes = var_6353_split_sizes_0, x = var_6352_cast_fp16)[name = tensor("op_6353_cast_fp16")]; + tensor var_6355_mode_0 = const()[name = tensor("op_6355_mode_0"), val = tensor("EXACT")]; + tensor var_6355_cast_fp16 = gelu(mode = var_6355_mode_0, x = var_6353_cast_fp16_1)[name = tensor("op_6355_cast_fp16")]; + tensor input_161_cast_fp16 = mul(x = var_6353_cast_fp16_0, y = var_6355_cast_fp16)[name = tensor("input_161_cast_fp16")]; + tensor var_6359 = const()[name = tensor("op_6359"), val = tensor([1, 1])]; + tensor var_6361 = const()[name = tensor("op_6361"), val = tensor([1, 1])]; + tensor var_6363_pad_type_0 = const()[name = tensor("op_6363_pad_type_0"), val = tensor("custom")]; + tensor var_6363_pad_0 = const()[name = tensor("op_6363_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_3_ff_net_2_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_3_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(463107840)))]; + tensor down_blocks_2_attentions_0_transformer_blocks_3_ff_net_2_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_3_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(476215104)))]; + tensor var_6363_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_3_ff_net_2_bias_to_fp16, dilations = var_6361, groups = var_2624, pad = var_6363_pad_0, pad_type = var_6363_pad_type_0, strides = var_6359, weight = down_blocks_2_attentions_0_transformer_blocks_3_ff_net_2_weight_to_fp16, x = input_161_cast_fp16)[name = tensor("op_6363_cast_fp16")]; + tensor inputs_49_cast_fp16 = add(x = var_6363_cast_fp16, y = inputs_47_cast_fp16)[name = tensor("inputs_49_cast_fp16")]; + tensor hidden_states_89_axes_0 = const()[name = tensor("hidden_states_89_axes_0"), val = tensor([1])]; + tensor hidden_states_89_gamma_0_to_fp16 = const()[name = tensor("hidden_states_89_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(476217728)))]; + tensor hidden_states_89_beta_0_to_fp16 = const()[name = tensor("hidden_states_89_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(476220352)))]; + tensor var_6379_to_fp16 = const()[name = tensor("op_6379_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_89_cast_fp16 = layer_norm(axes = hidden_states_89_axes_0, beta = hidden_states_89_beta_0_to_fp16, epsilon = var_6379_to_fp16, gamma = hidden_states_89_gamma_0_to_fp16, x = inputs_49_cast_fp16)[name = tensor("hidden_states_89_cast_fp16")]; + tensor var_6394 = const()[name = tensor("op_6394"), val = tensor([1, 1])]; + tensor var_6396 = const()[name = tensor("op_6396"), val = tensor([1, 1])]; + tensor q_33_pad_type_0 = const()[name = tensor("q_33_pad_type_0"), val = tensor("custom")]; + tensor q_33_pad_0 = const()[name = tensor("q_33_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_4_attn1_to_q_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_4_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(476222976)))]; + tensor q_33_cast_fp16 = conv(dilations = var_6396, groups = var_2624, pad = q_33_pad_0, pad_type = q_33_pad_type_0, strides = var_6394, weight = down_blocks_2_attentions_0_transformer_blocks_4_attn1_to_q_weight_to_fp16, x = hidden_states_89_cast_fp16)[name = tensor("q_33_cast_fp16")]; + tensor var_6400 = const()[name = tensor("op_6400"), val = tensor([1, 1])]; + tensor var_6402 = const()[name = tensor("op_6402"), val = tensor([1, 1])]; + tensor k_65_pad_type_0 = const()[name = tensor("k_65_pad_type_0"), val = tensor("custom")]; + tensor k_65_pad_0 = const()[name = tensor("k_65_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_4_attn1_to_k_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_4_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(479499840)))]; + tensor k_65_cast_fp16 = conv(dilations = var_6402, groups = var_2624, pad = k_65_pad_0, pad_type = k_65_pad_type_0, strides = var_6400, weight = down_blocks_2_attentions_0_transformer_blocks_4_attn1_to_k_weight_to_fp16, x = hidden_states_89_cast_fp16)[name = tensor("k_65_cast_fp16")]; + tensor var_6406 = const()[name = tensor("op_6406"), val = tensor([1, 1])]; + tensor var_6408 = const()[name = tensor("op_6408"), val = tensor([1, 1])]; + tensor v_33_pad_type_0 = const()[name = tensor("v_33_pad_type_0"), val = tensor("custom")]; + tensor v_33_pad_0 = const()[name = tensor("v_33_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_4_attn1_to_v_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_4_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(482776704)))]; + tensor v_33_cast_fp16 = conv(dilations = var_6408, groups = var_2624, pad = v_33_pad_0, pad_type = v_33_pad_type_0, strides = var_6406, weight = down_blocks_2_attentions_0_transformer_blocks_4_attn1_to_v_weight_to_fp16, x = hidden_states_89_cast_fp16)[name = tensor("v_33_cast_fp16")]; + tensor var_6412_begin_0 = const()[name = tensor("op_6412_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6412_end_0 = const()[name = tensor("op_6412_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_6412_end_mask_0 = const()[name = tensor("op_6412_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6412_cast_fp16 = slice_by_index(begin = var_6412_begin_0, end = var_6412_end_0, end_mask = var_6412_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_6412_cast_fp16")]; + tensor var_6416_begin_0 = const()[name = tensor("op_6416_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_6416_end_0 = const()[name = tensor("op_6416_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_6416_end_mask_0 = const()[name = tensor("op_6416_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6416_cast_fp16 = slice_by_index(begin = var_6416_begin_0, end = var_6416_end_0, end_mask = var_6416_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_6416_cast_fp16")]; + tensor var_6420_begin_0 = const()[name = tensor("op_6420_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_6420_end_0 = const()[name = tensor("op_6420_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_6420_end_mask_0 = const()[name = tensor("op_6420_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6420_cast_fp16 = slice_by_index(begin = var_6420_begin_0, end = var_6420_end_0, end_mask = var_6420_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_6420_cast_fp16")]; + tensor var_6424_begin_0 = const()[name = tensor("op_6424_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_6424_end_0 = const()[name = tensor("op_6424_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_6424_end_mask_0 = const()[name = tensor("op_6424_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6424_cast_fp16 = slice_by_index(begin = var_6424_begin_0, end = var_6424_end_0, end_mask = var_6424_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_6424_cast_fp16")]; + tensor var_6428_begin_0 = const()[name = tensor("op_6428_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_6428_end_0 = const()[name = tensor("op_6428_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_6428_end_mask_0 = const()[name = tensor("op_6428_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6428_cast_fp16 = slice_by_index(begin = var_6428_begin_0, end = var_6428_end_0, end_mask = var_6428_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_6428_cast_fp16")]; + tensor var_6432_begin_0 = const()[name = tensor("op_6432_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_6432_end_0 = const()[name = tensor("op_6432_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_6432_end_mask_0 = const()[name = tensor("op_6432_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6432_cast_fp16 = slice_by_index(begin = var_6432_begin_0, end = var_6432_end_0, end_mask = var_6432_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_6432_cast_fp16")]; + tensor var_6436_begin_0 = const()[name = tensor("op_6436_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_6436_end_0 = const()[name = tensor("op_6436_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_6436_end_mask_0 = const()[name = tensor("op_6436_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6436_cast_fp16 = slice_by_index(begin = var_6436_begin_0, end = var_6436_end_0, end_mask = var_6436_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_6436_cast_fp16")]; + tensor var_6440_begin_0 = const()[name = tensor("op_6440_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_6440_end_0 = const()[name = tensor("op_6440_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_6440_end_mask_0 = const()[name = tensor("op_6440_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6440_cast_fp16 = slice_by_index(begin = var_6440_begin_0, end = var_6440_end_0, end_mask = var_6440_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_6440_cast_fp16")]; + tensor var_6444_begin_0 = const()[name = tensor("op_6444_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_6444_end_0 = const()[name = tensor("op_6444_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_6444_end_mask_0 = const()[name = tensor("op_6444_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6444_cast_fp16 = slice_by_index(begin = var_6444_begin_0, end = var_6444_end_0, end_mask = var_6444_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_6444_cast_fp16")]; + tensor var_6448_begin_0 = const()[name = tensor("op_6448_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_6448_end_0 = const()[name = tensor("op_6448_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_6448_end_mask_0 = const()[name = tensor("op_6448_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6448_cast_fp16 = slice_by_index(begin = var_6448_begin_0, end = var_6448_end_0, end_mask = var_6448_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_6448_cast_fp16")]; + tensor var_6452_begin_0 = const()[name = tensor("op_6452_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_6452_end_0 = const()[name = tensor("op_6452_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_6452_end_mask_0 = const()[name = tensor("op_6452_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6452_cast_fp16 = slice_by_index(begin = var_6452_begin_0, end = var_6452_end_0, end_mask = var_6452_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_6452_cast_fp16")]; + tensor var_6456_begin_0 = const()[name = tensor("op_6456_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_6456_end_0 = const()[name = tensor("op_6456_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_6456_end_mask_0 = const()[name = tensor("op_6456_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6456_cast_fp16 = slice_by_index(begin = var_6456_begin_0, end = var_6456_end_0, end_mask = var_6456_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_6456_cast_fp16")]; + tensor var_6460_begin_0 = const()[name = tensor("op_6460_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_6460_end_0 = const()[name = tensor("op_6460_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_6460_end_mask_0 = const()[name = tensor("op_6460_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6460_cast_fp16 = slice_by_index(begin = var_6460_begin_0, end = var_6460_end_0, end_mask = var_6460_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_6460_cast_fp16")]; + tensor var_6464_begin_0 = const()[name = tensor("op_6464_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_6464_end_0 = const()[name = tensor("op_6464_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_6464_end_mask_0 = const()[name = tensor("op_6464_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6464_cast_fp16 = slice_by_index(begin = var_6464_begin_0, end = var_6464_end_0, end_mask = var_6464_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_6464_cast_fp16")]; + tensor var_6468_begin_0 = const()[name = tensor("op_6468_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_6468_end_0 = const()[name = tensor("op_6468_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_6468_end_mask_0 = const()[name = tensor("op_6468_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6468_cast_fp16 = slice_by_index(begin = var_6468_begin_0, end = var_6468_end_0, end_mask = var_6468_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_6468_cast_fp16")]; + tensor var_6472_begin_0 = const()[name = tensor("op_6472_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_6472_end_0 = const()[name = tensor("op_6472_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_6472_end_mask_0 = const()[name = tensor("op_6472_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6472_cast_fp16 = slice_by_index(begin = var_6472_begin_0, end = var_6472_end_0, end_mask = var_6472_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_6472_cast_fp16")]; + tensor var_6476_begin_0 = const()[name = tensor("op_6476_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_6476_end_0 = const()[name = tensor("op_6476_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_6476_end_mask_0 = const()[name = tensor("op_6476_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6476_cast_fp16 = slice_by_index(begin = var_6476_begin_0, end = var_6476_end_0, end_mask = var_6476_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_6476_cast_fp16")]; + tensor var_6480_begin_0 = const()[name = tensor("op_6480_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_6480_end_0 = const()[name = tensor("op_6480_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_6480_end_mask_0 = const()[name = tensor("op_6480_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6480_cast_fp16 = slice_by_index(begin = var_6480_begin_0, end = var_6480_end_0, end_mask = var_6480_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_6480_cast_fp16")]; + tensor var_6484_begin_0 = const()[name = tensor("op_6484_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_6484_end_0 = const()[name = tensor("op_6484_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_6484_end_mask_0 = const()[name = tensor("op_6484_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6484_cast_fp16 = slice_by_index(begin = var_6484_begin_0, end = var_6484_end_0, end_mask = var_6484_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_6484_cast_fp16")]; + tensor var_6488_begin_0 = const()[name = tensor("op_6488_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_6488_end_0 = const()[name = tensor("op_6488_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_6488_end_mask_0 = const()[name = tensor("op_6488_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6488_cast_fp16 = slice_by_index(begin = var_6488_begin_0, end = var_6488_end_0, end_mask = var_6488_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_6488_cast_fp16")]; + tensor k_67_perm_0 = const()[name = tensor("k_67_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_6495_begin_0 = const()[name = tensor("op_6495_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6495_end_0 = const()[name = tensor("op_6495_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_6495_end_mask_0 = const()[name = tensor("op_6495_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_67_cast_fp16 = transpose(perm = k_67_perm_0, x = k_65_cast_fp16)[name = tensor("transpose_123")]; + tensor var_6495_cast_fp16 = slice_by_index(begin = var_6495_begin_0, end = var_6495_end_0, end_mask = var_6495_end_mask_0, x = k_67_cast_fp16)[name = tensor("op_6495_cast_fp16")]; + tensor var_6499_begin_0 = const()[name = tensor("op_6499_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_6499_end_0 = const()[name = tensor("op_6499_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_6499_end_mask_0 = const()[name = tensor("op_6499_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6499_cast_fp16 = slice_by_index(begin = var_6499_begin_0, end = var_6499_end_0, end_mask = var_6499_end_mask_0, x = k_67_cast_fp16)[name = tensor("op_6499_cast_fp16")]; + tensor var_6503_begin_0 = const()[name = tensor("op_6503_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_6503_end_0 = const()[name = tensor("op_6503_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_6503_end_mask_0 = const()[name = tensor("op_6503_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6503_cast_fp16 = slice_by_index(begin = var_6503_begin_0, end = var_6503_end_0, end_mask = var_6503_end_mask_0, x = k_67_cast_fp16)[name = tensor("op_6503_cast_fp16")]; + tensor var_6507_begin_0 = const()[name = tensor("op_6507_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_6507_end_0 = const()[name = tensor("op_6507_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_6507_end_mask_0 = const()[name = tensor("op_6507_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6507_cast_fp16 = slice_by_index(begin = var_6507_begin_0, end = var_6507_end_0, end_mask = var_6507_end_mask_0, x = k_67_cast_fp16)[name = tensor("op_6507_cast_fp16")]; + tensor var_6511_begin_0 = const()[name = tensor("op_6511_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_6511_end_0 = const()[name = tensor("op_6511_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_6511_end_mask_0 = const()[name = tensor("op_6511_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6511_cast_fp16 = slice_by_index(begin = var_6511_begin_0, end = var_6511_end_0, end_mask = var_6511_end_mask_0, x = k_67_cast_fp16)[name = tensor("op_6511_cast_fp16")]; + tensor var_6515_begin_0 = const()[name = tensor("op_6515_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_6515_end_0 = const()[name = tensor("op_6515_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_6515_end_mask_0 = const()[name = tensor("op_6515_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6515_cast_fp16 = slice_by_index(begin = var_6515_begin_0, end = var_6515_end_0, end_mask = var_6515_end_mask_0, x = k_67_cast_fp16)[name = tensor("op_6515_cast_fp16")]; + tensor var_6519_begin_0 = const()[name = tensor("op_6519_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_6519_end_0 = const()[name = tensor("op_6519_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_6519_end_mask_0 = const()[name = tensor("op_6519_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6519_cast_fp16 = slice_by_index(begin = var_6519_begin_0, end = var_6519_end_0, end_mask = var_6519_end_mask_0, x = k_67_cast_fp16)[name = tensor("op_6519_cast_fp16")]; + tensor var_6523_begin_0 = const()[name = tensor("op_6523_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_6523_end_0 = const()[name = tensor("op_6523_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_6523_end_mask_0 = const()[name = tensor("op_6523_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6523_cast_fp16 = slice_by_index(begin = var_6523_begin_0, end = var_6523_end_0, end_mask = var_6523_end_mask_0, x = k_67_cast_fp16)[name = tensor("op_6523_cast_fp16")]; + tensor var_6527_begin_0 = const()[name = tensor("op_6527_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_6527_end_0 = const()[name = tensor("op_6527_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_6527_end_mask_0 = const()[name = tensor("op_6527_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6527_cast_fp16 = slice_by_index(begin = var_6527_begin_0, end = var_6527_end_0, end_mask = var_6527_end_mask_0, x = k_67_cast_fp16)[name = tensor("op_6527_cast_fp16")]; + tensor var_6531_begin_0 = const()[name = tensor("op_6531_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_6531_end_0 = const()[name = tensor("op_6531_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_6531_end_mask_0 = const()[name = tensor("op_6531_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6531_cast_fp16 = slice_by_index(begin = var_6531_begin_0, end = var_6531_end_0, end_mask = var_6531_end_mask_0, x = k_67_cast_fp16)[name = tensor("op_6531_cast_fp16")]; + tensor var_6535_begin_0 = const()[name = tensor("op_6535_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_6535_end_0 = const()[name = tensor("op_6535_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_6535_end_mask_0 = const()[name = tensor("op_6535_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6535_cast_fp16 = slice_by_index(begin = var_6535_begin_0, end = var_6535_end_0, end_mask = var_6535_end_mask_0, x = k_67_cast_fp16)[name = tensor("op_6535_cast_fp16")]; + tensor var_6539_begin_0 = const()[name = tensor("op_6539_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_6539_end_0 = const()[name = tensor("op_6539_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_6539_end_mask_0 = const()[name = tensor("op_6539_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6539_cast_fp16 = slice_by_index(begin = var_6539_begin_0, end = var_6539_end_0, end_mask = var_6539_end_mask_0, x = k_67_cast_fp16)[name = tensor("op_6539_cast_fp16")]; + tensor var_6543_begin_0 = const()[name = tensor("op_6543_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_6543_end_0 = const()[name = tensor("op_6543_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_6543_end_mask_0 = const()[name = tensor("op_6543_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6543_cast_fp16 = slice_by_index(begin = var_6543_begin_0, end = var_6543_end_0, end_mask = var_6543_end_mask_0, x = k_67_cast_fp16)[name = tensor("op_6543_cast_fp16")]; + tensor var_6547_begin_0 = const()[name = tensor("op_6547_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_6547_end_0 = const()[name = tensor("op_6547_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_6547_end_mask_0 = const()[name = tensor("op_6547_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6547_cast_fp16 = slice_by_index(begin = var_6547_begin_0, end = var_6547_end_0, end_mask = var_6547_end_mask_0, x = k_67_cast_fp16)[name = tensor("op_6547_cast_fp16")]; + tensor var_6551_begin_0 = const()[name = tensor("op_6551_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_6551_end_0 = const()[name = tensor("op_6551_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_6551_end_mask_0 = const()[name = tensor("op_6551_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6551_cast_fp16 = slice_by_index(begin = var_6551_begin_0, end = var_6551_end_0, end_mask = var_6551_end_mask_0, x = k_67_cast_fp16)[name = tensor("op_6551_cast_fp16")]; + tensor var_6555_begin_0 = const()[name = tensor("op_6555_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_6555_end_0 = const()[name = tensor("op_6555_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_6555_end_mask_0 = const()[name = tensor("op_6555_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6555_cast_fp16 = slice_by_index(begin = var_6555_begin_0, end = var_6555_end_0, end_mask = var_6555_end_mask_0, x = k_67_cast_fp16)[name = tensor("op_6555_cast_fp16")]; + tensor var_6559_begin_0 = const()[name = tensor("op_6559_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_6559_end_0 = const()[name = tensor("op_6559_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_6559_end_mask_0 = const()[name = tensor("op_6559_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6559_cast_fp16 = slice_by_index(begin = var_6559_begin_0, end = var_6559_end_0, end_mask = var_6559_end_mask_0, x = k_67_cast_fp16)[name = tensor("op_6559_cast_fp16")]; + tensor var_6563_begin_0 = const()[name = tensor("op_6563_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_6563_end_0 = const()[name = tensor("op_6563_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_6563_end_mask_0 = const()[name = tensor("op_6563_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6563_cast_fp16 = slice_by_index(begin = var_6563_begin_0, end = var_6563_end_0, end_mask = var_6563_end_mask_0, x = k_67_cast_fp16)[name = tensor("op_6563_cast_fp16")]; + tensor var_6567_begin_0 = const()[name = tensor("op_6567_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_6567_end_0 = const()[name = tensor("op_6567_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_6567_end_mask_0 = const()[name = tensor("op_6567_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6567_cast_fp16 = slice_by_index(begin = var_6567_begin_0, end = var_6567_end_0, end_mask = var_6567_end_mask_0, x = k_67_cast_fp16)[name = tensor("op_6567_cast_fp16")]; + tensor var_6571_begin_0 = const()[name = tensor("op_6571_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_6571_end_0 = const()[name = tensor("op_6571_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_6571_end_mask_0 = const()[name = tensor("op_6571_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6571_cast_fp16 = slice_by_index(begin = var_6571_begin_0, end = var_6571_end_0, end_mask = var_6571_end_mask_0, x = k_67_cast_fp16)[name = tensor("op_6571_cast_fp16")]; + tensor var_6573_begin_0 = const()[name = tensor("op_6573_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6573_end_0 = const()[name = tensor("op_6573_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_6573_end_mask_0 = const()[name = tensor("op_6573_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6573_cast_fp16 = slice_by_index(begin = var_6573_begin_0, end = var_6573_end_0, end_mask = var_6573_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_6573_cast_fp16")]; + tensor var_6577_begin_0 = const()[name = tensor("op_6577_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_6577_end_0 = const()[name = tensor("op_6577_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_6577_end_mask_0 = const()[name = tensor("op_6577_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6577_cast_fp16 = slice_by_index(begin = var_6577_begin_0, end = var_6577_end_0, end_mask = var_6577_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_6577_cast_fp16")]; + tensor var_6581_begin_0 = const()[name = tensor("op_6581_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_6581_end_0 = const()[name = tensor("op_6581_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_6581_end_mask_0 = const()[name = tensor("op_6581_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6581_cast_fp16 = slice_by_index(begin = var_6581_begin_0, end = var_6581_end_0, end_mask = var_6581_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_6581_cast_fp16")]; + tensor var_6585_begin_0 = const()[name = tensor("op_6585_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_6585_end_0 = const()[name = tensor("op_6585_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_6585_end_mask_0 = const()[name = tensor("op_6585_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6585_cast_fp16 = slice_by_index(begin = var_6585_begin_0, end = var_6585_end_0, end_mask = var_6585_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_6585_cast_fp16")]; + tensor var_6589_begin_0 = const()[name = tensor("op_6589_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_6589_end_0 = const()[name = tensor("op_6589_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_6589_end_mask_0 = const()[name = tensor("op_6589_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6589_cast_fp16 = slice_by_index(begin = var_6589_begin_0, end = var_6589_end_0, end_mask = var_6589_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_6589_cast_fp16")]; + tensor var_6593_begin_0 = const()[name = tensor("op_6593_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_6593_end_0 = const()[name = tensor("op_6593_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_6593_end_mask_0 = const()[name = tensor("op_6593_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6593_cast_fp16 = slice_by_index(begin = var_6593_begin_0, end = var_6593_end_0, end_mask = var_6593_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_6593_cast_fp16")]; + tensor var_6597_begin_0 = const()[name = tensor("op_6597_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_6597_end_0 = const()[name = tensor("op_6597_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_6597_end_mask_0 = const()[name = tensor("op_6597_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6597_cast_fp16 = slice_by_index(begin = var_6597_begin_0, end = var_6597_end_0, end_mask = var_6597_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_6597_cast_fp16")]; + tensor var_6601_begin_0 = const()[name = tensor("op_6601_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_6601_end_0 = const()[name = tensor("op_6601_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_6601_end_mask_0 = const()[name = tensor("op_6601_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6601_cast_fp16 = slice_by_index(begin = var_6601_begin_0, end = var_6601_end_0, end_mask = var_6601_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_6601_cast_fp16")]; + tensor var_6605_begin_0 = const()[name = tensor("op_6605_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_6605_end_0 = const()[name = tensor("op_6605_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_6605_end_mask_0 = const()[name = tensor("op_6605_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6605_cast_fp16 = slice_by_index(begin = var_6605_begin_0, end = var_6605_end_0, end_mask = var_6605_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_6605_cast_fp16")]; + tensor var_6609_begin_0 = const()[name = tensor("op_6609_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_6609_end_0 = const()[name = tensor("op_6609_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_6609_end_mask_0 = const()[name = tensor("op_6609_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6609_cast_fp16 = slice_by_index(begin = var_6609_begin_0, end = var_6609_end_0, end_mask = var_6609_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_6609_cast_fp16")]; + tensor var_6613_begin_0 = const()[name = tensor("op_6613_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_6613_end_0 = const()[name = tensor("op_6613_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_6613_end_mask_0 = const()[name = tensor("op_6613_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6613_cast_fp16 = slice_by_index(begin = var_6613_begin_0, end = var_6613_end_0, end_mask = var_6613_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_6613_cast_fp16")]; + tensor var_6617_begin_0 = const()[name = tensor("op_6617_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_6617_end_0 = const()[name = tensor("op_6617_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_6617_end_mask_0 = const()[name = tensor("op_6617_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6617_cast_fp16 = slice_by_index(begin = var_6617_begin_0, end = var_6617_end_0, end_mask = var_6617_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_6617_cast_fp16")]; + tensor var_6621_begin_0 = const()[name = tensor("op_6621_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_6621_end_0 = const()[name = tensor("op_6621_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_6621_end_mask_0 = const()[name = tensor("op_6621_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6621_cast_fp16 = slice_by_index(begin = var_6621_begin_0, end = var_6621_end_0, end_mask = var_6621_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_6621_cast_fp16")]; + tensor var_6625_begin_0 = const()[name = tensor("op_6625_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_6625_end_0 = const()[name = tensor("op_6625_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_6625_end_mask_0 = const()[name = tensor("op_6625_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6625_cast_fp16 = slice_by_index(begin = var_6625_begin_0, end = var_6625_end_0, end_mask = var_6625_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_6625_cast_fp16")]; + tensor var_6629_begin_0 = const()[name = tensor("op_6629_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_6629_end_0 = const()[name = tensor("op_6629_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_6629_end_mask_0 = const()[name = tensor("op_6629_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6629_cast_fp16 = slice_by_index(begin = var_6629_begin_0, end = var_6629_end_0, end_mask = var_6629_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_6629_cast_fp16")]; + tensor var_6633_begin_0 = const()[name = tensor("op_6633_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_6633_end_0 = const()[name = tensor("op_6633_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_6633_end_mask_0 = const()[name = tensor("op_6633_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6633_cast_fp16 = slice_by_index(begin = var_6633_begin_0, end = var_6633_end_0, end_mask = var_6633_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_6633_cast_fp16")]; + tensor var_6637_begin_0 = const()[name = tensor("op_6637_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_6637_end_0 = const()[name = tensor("op_6637_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_6637_end_mask_0 = const()[name = tensor("op_6637_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6637_cast_fp16 = slice_by_index(begin = var_6637_begin_0, end = var_6637_end_0, end_mask = var_6637_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_6637_cast_fp16")]; + tensor var_6641_begin_0 = const()[name = tensor("op_6641_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_6641_end_0 = const()[name = tensor("op_6641_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_6641_end_mask_0 = const()[name = tensor("op_6641_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6641_cast_fp16 = slice_by_index(begin = var_6641_begin_0, end = var_6641_end_0, end_mask = var_6641_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_6641_cast_fp16")]; + tensor var_6645_begin_0 = const()[name = tensor("op_6645_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_6645_end_0 = const()[name = tensor("op_6645_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_6645_end_mask_0 = const()[name = tensor("op_6645_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6645_cast_fp16 = slice_by_index(begin = var_6645_begin_0, end = var_6645_end_0, end_mask = var_6645_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_6645_cast_fp16")]; + tensor var_6649_begin_0 = const()[name = tensor("op_6649_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_6649_end_0 = const()[name = tensor("op_6649_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_6649_end_mask_0 = const()[name = tensor("op_6649_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6649_cast_fp16 = slice_by_index(begin = var_6649_begin_0, end = var_6649_end_0, end_mask = var_6649_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_6649_cast_fp16")]; + tensor var_6653_equation_0 = const()[name = tensor("op_6653_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6653_cast_fp16 = einsum(equation = var_6653_equation_0, values = (var_6495_cast_fp16, var_6412_cast_fp16))[name = tensor("op_6653_cast_fp16")]; + tensor var_6654_to_fp16 = const()[name = tensor("op_6654_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_481_cast_fp16 = mul(x = var_6653_cast_fp16, y = var_6654_to_fp16)[name = tensor("aw_481_cast_fp16")]; + tensor var_6657_equation_0 = const()[name = tensor("op_6657_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6657_cast_fp16 = einsum(equation = var_6657_equation_0, values = (var_6499_cast_fp16, var_6416_cast_fp16))[name = tensor("op_6657_cast_fp16")]; + tensor var_6658_to_fp16 = const()[name = tensor("op_6658_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_483_cast_fp16 = mul(x = var_6657_cast_fp16, y = var_6658_to_fp16)[name = tensor("aw_483_cast_fp16")]; + tensor var_6661_equation_0 = const()[name = tensor("op_6661_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6661_cast_fp16 = einsum(equation = var_6661_equation_0, values = (var_6503_cast_fp16, var_6420_cast_fp16))[name = tensor("op_6661_cast_fp16")]; + tensor var_6662_to_fp16 = const()[name = tensor("op_6662_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_485_cast_fp16 = mul(x = var_6661_cast_fp16, y = var_6662_to_fp16)[name = tensor("aw_485_cast_fp16")]; + tensor var_6665_equation_0 = const()[name = tensor("op_6665_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6665_cast_fp16 = einsum(equation = var_6665_equation_0, values = (var_6507_cast_fp16, var_6424_cast_fp16))[name = tensor("op_6665_cast_fp16")]; + tensor var_6666_to_fp16 = const()[name = tensor("op_6666_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_487_cast_fp16 = mul(x = var_6665_cast_fp16, y = var_6666_to_fp16)[name = tensor("aw_487_cast_fp16")]; + tensor var_6669_equation_0 = const()[name = tensor("op_6669_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6669_cast_fp16 = einsum(equation = var_6669_equation_0, values = (var_6511_cast_fp16, var_6428_cast_fp16))[name = tensor("op_6669_cast_fp16")]; + tensor var_6670_to_fp16 = const()[name = tensor("op_6670_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_489_cast_fp16 = mul(x = var_6669_cast_fp16, y = var_6670_to_fp16)[name = tensor("aw_489_cast_fp16")]; + tensor var_6673_equation_0 = const()[name = tensor("op_6673_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6673_cast_fp16 = einsum(equation = var_6673_equation_0, values = (var_6515_cast_fp16, var_6432_cast_fp16))[name = tensor("op_6673_cast_fp16")]; + tensor var_6674_to_fp16 = const()[name = tensor("op_6674_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_491_cast_fp16 = mul(x = var_6673_cast_fp16, y = var_6674_to_fp16)[name = tensor("aw_491_cast_fp16")]; + tensor var_6677_equation_0 = const()[name = tensor("op_6677_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6677_cast_fp16 = einsum(equation = var_6677_equation_0, values = (var_6519_cast_fp16, var_6436_cast_fp16))[name = tensor("op_6677_cast_fp16")]; + tensor var_6678_to_fp16 = const()[name = tensor("op_6678_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_493_cast_fp16 = mul(x = var_6677_cast_fp16, y = var_6678_to_fp16)[name = tensor("aw_493_cast_fp16")]; + tensor var_6681_equation_0 = const()[name = tensor("op_6681_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6681_cast_fp16 = einsum(equation = var_6681_equation_0, values = (var_6523_cast_fp16, var_6440_cast_fp16))[name = tensor("op_6681_cast_fp16")]; + tensor var_6682_to_fp16 = const()[name = tensor("op_6682_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_495_cast_fp16 = mul(x = var_6681_cast_fp16, y = var_6682_to_fp16)[name = tensor("aw_495_cast_fp16")]; + tensor var_6685_equation_0 = const()[name = tensor("op_6685_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6685_cast_fp16 = einsum(equation = var_6685_equation_0, values = (var_6527_cast_fp16, var_6444_cast_fp16))[name = tensor("op_6685_cast_fp16")]; + tensor var_6686_to_fp16 = const()[name = tensor("op_6686_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_497_cast_fp16 = mul(x = var_6685_cast_fp16, y = var_6686_to_fp16)[name = tensor("aw_497_cast_fp16")]; + tensor var_6689_equation_0 = const()[name = tensor("op_6689_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6689_cast_fp16 = einsum(equation = var_6689_equation_0, values = (var_6531_cast_fp16, var_6448_cast_fp16))[name = tensor("op_6689_cast_fp16")]; + tensor var_6690_to_fp16 = const()[name = tensor("op_6690_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_499_cast_fp16 = mul(x = var_6689_cast_fp16, y = var_6690_to_fp16)[name = tensor("aw_499_cast_fp16")]; + tensor var_6693_equation_0 = const()[name = tensor("op_6693_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6693_cast_fp16 = einsum(equation = var_6693_equation_0, values = (var_6535_cast_fp16, var_6452_cast_fp16))[name = tensor("op_6693_cast_fp16")]; + tensor var_6694_to_fp16 = const()[name = tensor("op_6694_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_501_cast_fp16 = mul(x = var_6693_cast_fp16, y = var_6694_to_fp16)[name = tensor("aw_501_cast_fp16")]; + tensor var_6697_equation_0 = const()[name = tensor("op_6697_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6697_cast_fp16 = einsum(equation = var_6697_equation_0, values = (var_6539_cast_fp16, var_6456_cast_fp16))[name = tensor("op_6697_cast_fp16")]; + tensor var_6698_to_fp16 = const()[name = tensor("op_6698_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_503_cast_fp16 = mul(x = var_6697_cast_fp16, y = var_6698_to_fp16)[name = tensor("aw_503_cast_fp16")]; + tensor var_6701_equation_0 = const()[name = tensor("op_6701_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6701_cast_fp16 = einsum(equation = var_6701_equation_0, values = (var_6543_cast_fp16, var_6460_cast_fp16))[name = tensor("op_6701_cast_fp16")]; + tensor var_6702_to_fp16 = const()[name = tensor("op_6702_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_505_cast_fp16 = mul(x = var_6701_cast_fp16, y = var_6702_to_fp16)[name = tensor("aw_505_cast_fp16")]; + tensor var_6705_equation_0 = const()[name = tensor("op_6705_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6705_cast_fp16 = einsum(equation = var_6705_equation_0, values = (var_6547_cast_fp16, var_6464_cast_fp16))[name = tensor("op_6705_cast_fp16")]; + tensor var_6706_to_fp16 = const()[name = tensor("op_6706_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_507_cast_fp16 = mul(x = var_6705_cast_fp16, y = var_6706_to_fp16)[name = tensor("aw_507_cast_fp16")]; + tensor var_6709_equation_0 = const()[name = tensor("op_6709_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6709_cast_fp16 = einsum(equation = var_6709_equation_0, values = (var_6551_cast_fp16, var_6468_cast_fp16))[name = tensor("op_6709_cast_fp16")]; + tensor var_6710_to_fp16 = const()[name = tensor("op_6710_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_509_cast_fp16 = mul(x = var_6709_cast_fp16, y = var_6710_to_fp16)[name = tensor("aw_509_cast_fp16")]; + tensor var_6713_equation_0 = const()[name = tensor("op_6713_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6713_cast_fp16 = einsum(equation = var_6713_equation_0, values = (var_6555_cast_fp16, var_6472_cast_fp16))[name = tensor("op_6713_cast_fp16")]; + tensor var_6714_to_fp16 = const()[name = tensor("op_6714_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_511_cast_fp16 = mul(x = var_6713_cast_fp16, y = var_6714_to_fp16)[name = tensor("aw_511_cast_fp16")]; + tensor var_6717_equation_0 = const()[name = tensor("op_6717_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6717_cast_fp16 = einsum(equation = var_6717_equation_0, values = (var_6559_cast_fp16, var_6476_cast_fp16))[name = tensor("op_6717_cast_fp16")]; + tensor var_6718_to_fp16 = const()[name = tensor("op_6718_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_513_cast_fp16 = mul(x = var_6717_cast_fp16, y = var_6718_to_fp16)[name = tensor("aw_513_cast_fp16")]; + tensor var_6721_equation_0 = const()[name = tensor("op_6721_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6721_cast_fp16 = einsum(equation = var_6721_equation_0, values = (var_6563_cast_fp16, var_6480_cast_fp16))[name = tensor("op_6721_cast_fp16")]; + tensor var_6722_to_fp16 = const()[name = tensor("op_6722_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_515_cast_fp16 = mul(x = var_6721_cast_fp16, y = var_6722_to_fp16)[name = tensor("aw_515_cast_fp16")]; + tensor var_6725_equation_0 = const()[name = tensor("op_6725_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6725_cast_fp16 = einsum(equation = var_6725_equation_0, values = (var_6567_cast_fp16, var_6484_cast_fp16))[name = tensor("op_6725_cast_fp16")]; + tensor var_6726_to_fp16 = const()[name = tensor("op_6726_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_517_cast_fp16 = mul(x = var_6725_cast_fp16, y = var_6726_to_fp16)[name = tensor("aw_517_cast_fp16")]; + tensor var_6729_equation_0 = const()[name = tensor("op_6729_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6729_cast_fp16 = einsum(equation = var_6729_equation_0, values = (var_6571_cast_fp16, var_6488_cast_fp16))[name = tensor("op_6729_cast_fp16")]; + tensor var_6730_to_fp16 = const()[name = tensor("op_6730_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_519_cast_fp16 = mul(x = var_6729_cast_fp16, y = var_6730_to_fp16)[name = tensor("aw_519_cast_fp16")]; + tensor var_6732_cast_fp16 = softmax(axis = var_2624, x = aw_481_cast_fp16)[name = tensor("op_6732_cast_fp16")]; + tensor var_6733_cast_fp16 = softmax(axis = var_2624, x = aw_483_cast_fp16)[name = tensor("op_6733_cast_fp16")]; + tensor var_6734_cast_fp16 = softmax(axis = var_2624, x = aw_485_cast_fp16)[name = tensor("op_6734_cast_fp16")]; + tensor var_6735_cast_fp16 = softmax(axis = var_2624, x = aw_487_cast_fp16)[name = tensor("op_6735_cast_fp16")]; + tensor var_6736_cast_fp16 = softmax(axis = var_2624, x = aw_489_cast_fp16)[name = tensor("op_6736_cast_fp16")]; + tensor var_6737_cast_fp16 = softmax(axis = var_2624, x = aw_491_cast_fp16)[name = tensor("op_6737_cast_fp16")]; + tensor var_6738_cast_fp16 = softmax(axis = var_2624, x = aw_493_cast_fp16)[name = tensor("op_6738_cast_fp16")]; + tensor var_6739_cast_fp16 = softmax(axis = var_2624, x = aw_495_cast_fp16)[name = tensor("op_6739_cast_fp16")]; + tensor var_6740_cast_fp16 = softmax(axis = var_2624, x = aw_497_cast_fp16)[name = tensor("op_6740_cast_fp16")]; + tensor var_6741_cast_fp16 = softmax(axis = var_2624, x = aw_499_cast_fp16)[name = tensor("op_6741_cast_fp16")]; + tensor var_6742_cast_fp16 = softmax(axis = var_2624, x = aw_501_cast_fp16)[name = tensor("op_6742_cast_fp16")]; + tensor var_6743_cast_fp16 = softmax(axis = var_2624, x = aw_503_cast_fp16)[name = tensor("op_6743_cast_fp16")]; + tensor var_6744_cast_fp16 = softmax(axis = var_2624, x = aw_505_cast_fp16)[name = tensor("op_6744_cast_fp16")]; + tensor var_6745_cast_fp16 = softmax(axis = var_2624, x = aw_507_cast_fp16)[name = tensor("op_6745_cast_fp16")]; + tensor var_6746_cast_fp16 = softmax(axis = var_2624, x = aw_509_cast_fp16)[name = tensor("op_6746_cast_fp16")]; + tensor var_6747_cast_fp16 = softmax(axis = var_2624, x = aw_511_cast_fp16)[name = tensor("op_6747_cast_fp16")]; + tensor var_6748_cast_fp16 = softmax(axis = var_2624, x = aw_513_cast_fp16)[name = tensor("op_6748_cast_fp16")]; + tensor var_6749_cast_fp16 = softmax(axis = var_2624, x = aw_515_cast_fp16)[name = tensor("op_6749_cast_fp16")]; + tensor var_6750_cast_fp16 = softmax(axis = var_2624, x = aw_517_cast_fp16)[name = tensor("op_6750_cast_fp16")]; + tensor var_6751_cast_fp16 = softmax(axis = var_2624, x = aw_519_cast_fp16)[name = tensor("op_6751_cast_fp16")]; + tensor var_6753_equation_0 = const()[name = tensor("op_6753_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6753_cast_fp16 = einsum(equation = var_6753_equation_0, values = (var_6573_cast_fp16, var_6732_cast_fp16))[name = tensor("op_6753_cast_fp16")]; + tensor var_6755_equation_0 = const()[name = tensor("op_6755_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6755_cast_fp16 = einsum(equation = var_6755_equation_0, values = (var_6577_cast_fp16, var_6733_cast_fp16))[name = tensor("op_6755_cast_fp16")]; + tensor var_6757_equation_0 = const()[name = tensor("op_6757_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6757_cast_fp16 = einsum(equation = var_6757_equation_0, values = (var_6581_cast_fp16, var_6734_cast_fp16))[name = tensor("op_6757_cast_fp16")]; + tensor var_6759_equation_0 = const()[name = tensor("op_6759_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6759_cast_fp16 = einsum(equation = var_6759_equation_0, values = (var_6585_cast_fp16, var_6735_cast_fp16))[name = tensor("op_6759_cast_fp16")]; + tensor var_6761_equation_0 = const()[name = tensor("op_6761_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6761_cast_fp16 = einsum(equation = var_6761_equation_0, values = (var_6589_cast_fp16, var_6736_cast_fp16))[name = tensor("op_6761_cast_fp16")]; + tensor var_6763_equation_0 = const()[name = tensor("op_6763_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6763_cast_fp16 = einsum(equation = var_6763_equation_0, values = (var_6593_cast_fp16, var_6737_cast_fp16))[name = tensor("op_6763_cast_fp16")]; + tensor var_6765_equation_0 = const()[name = tensor("op_6765_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6765_cast_fp16 = einsum(equation = var_6765_equation_0, values = (var_6597_cast_fp16, var_6738_cast_fp16))[name = tensor("op_6765_cast_fp16")]; + tensor var_6767_equation_0 = const()[name = tensor("op_6767_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6767_cast_fp16 = einsum(equation = var_6767_equation_0, values = (var_6601_cast_fp16, var_6739_cast_fp16))[name = tensor("op_6767_cast_fp16")]; + tensor var_6769_equation_0 = const()[name = tensor("op_6769_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6769_cast_fp16 = einsum(equation = var_6769_equation_0, values = (var_6605_cast_fp16, var_6740_cast_fp16))[name = tensor("op_6769_cast_fp16")]; + tensor var_6771_equation_0 = const()[name = tensor("op_6771_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6771_cast_fp16 = einsum(equation = var_6771_equation_0, values = (var_6609_cast_fp16, var_6741_cast_fp16))[name = tensor("op_6771_cast_fp16")]; + tensor var_6773_equation_0 = const()[name = tensor("op_6773_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6773_cast_fp16 = einsum(equation = var_6773_equation_0, values = (var_6613_cast_fp16, var_6742_cast_fp16))[name = tensor("op_6773_cast_fp16")]; + tensor var_6775_equation_0 = const()[name = tensor("op_6775_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6775_cast_fp16 = einsum(equation = var_6775_equation_0, values = (var_6617_cast_fp16, var_6743_cast_fp16))[name = tensor("op_6775_cast_fp16")]; + tensor var_6777_equation_0 = const()[name = tensor("op_6777_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6777_cast_fp16 = einsum(equation = var_6777_equation_0, values = (var_6621_cast_fp16, var_6744_cast_fp16))[name = tensor("op_6777_cast_fp16")]; + tensor var_6779_equation_0 = const()[name = tensor("op_6779_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6779_cast_fp16 = einsum(equation = var_6779_equation_0, values = (var_6625_cast_fp16, var_6745_cast_fp16))[name = tensor("op_6779_cast_fp16")]; + tensor var_6781_equation_0 = const()[name = tensor("op_6781_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6781_cast_fp16 = einsum(equation = var_6781_equation_0, values = (var_6629_cast_fp16, var_6746_cast_fp16))[name = tensor("op_6781_cast_fp16")]; + tensor var_6783_equation_0 = const()[name = tensor("op_6783_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6783_cast_fp16 = einsum(equation = var_6783_equation_0, values = (var_6633_cast_fp16, var_6747_cast_fp16))[name = tensor("op_6783_cast_fp16")]; + tensor var_6785_equation_0 = const()[name = tensor("op_6785_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6785_cast_fp16 = einsum(equation = var_6785_equation_0, values = (var_6637_cast_fp16, var_6748_cast_fp16))[name = tensor("op_6785_cast_fp16")]; + tensor var_6787_equation_0 = const()[name = tensor("op_6787_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6787_cast_fp16 = einsum(equation = var_6787_equation_0, values = (var_6641_cast_fp16, var_6749_cast_fp16))[name = tensor("op_6787_cast_fp16")]; + tensor var_6789_equation_0 = const()[name = tensor("op_6789_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6789_cast_fp16 = einsum(equation = var_6789_equation_0, values = (var_6645_cast_fp16, var_6750_cast_fp16))[name = tensor("op_6789_cast_fp16")]; + tensor var_6791_equation_0 = const()[name = tensor("op_6791_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6791_cast_fp16 = einsum(equation = var_6791_equation_0, values = (var_6649_cast_fp16, var_6751_cast_fp16))[name = tensor("op_6791_cast_fp16")]; + tensor input_163_interleave_0 = const()[name = tensor("input_163_interleave_0"), val = tensor(false)]; + tensor input_163_cast_fp16 = concat(axis = var_2624, interleave = input_163_interleave_0, values = (var_6753_cast_fp16, var_6755_cast_fp16, var_6757_cast_fp16, var_6759_cast_fp16, var_6761_cast_fp16, var_6763_cast_fp16, var_6765_cast_fp16, var_6767_cast_fp16, var_6769_cast_fp16, var_6771_cast_fp16, var_6773_cast_fp16, var_6775_cast_fp16, var_6777_cast_fp16, var_6779_cast_fp16, var_6781_cast_fp16, var_6783_cast_fp16, var_6785_cast_fp16, var_6787_cast_fp16, var_6789_cast_fp16, var_6791_cast_fp16))[name = tensor("input_163_cast_fp16")]; + tensor var_6797 = const()[name = tensor("op_6797"), val = tensor([1, 1])]; + tensor var_6799 = const()[name = tensor("op_6799"), val = tensor([1, 1])]; + tensor var_6801_pad_type_0 = const()[name = tensor("op_6801_pad_type_0"), val = tensor("custom")]; + tensor var_6801_pad_0 = const()[name = tensor("op_6801_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_4_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_4_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(486053568)))]; + tensor down_blocks_2_attentions_0_transformer_blocks_4_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_4_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(489330432)))]; + tensor var_6801_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_4_attn1_to_out_0_bias_to_fp16, dilations = var_6799, groups = var_2624, pad = var_6801_pad_0, pad_type = var_6801_pad_type_0, strides = var_6797, weight = down_blocks_2_attentions_0_transformer_blocks_4_attn1_to_out_0_weight_to_fp16, x = input_163_cast_fp16)[name = tensor("op_6801_cast_fp16")]; + tensor inputs_51_cast_fp16 = add(x = var_6801_cast_fp16, y = inputs_49_cast_fp16)[name = tensor("inputs_51_cast_fp16")]; + tensor hidden_states_91_axes_0 = const()[name = tensor("hidden_states_91_axes_0"), val = tensor([1])]; + tensor hidden_states_91_gamma_0_to_fp16 = const()[name = tensor("hidden_states_91_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(489333056)))]; + tensor hidden_states_91_beta_0_to_fp16 = const()[name = tensor("hidden_states_91_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(489335680)))]; + tensor var_6811_to_fp16 = const()[name = tensor("op_6811_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_91_cast_fp16 = layer_norm(axes = hidden_states_91_axes_0, beta = hidden_states_91_beta_0_to_fp16, epsilon = var_6811_to_fp16, gamma = hidden_states_91_gamma_0_to_fp16, x = inputs_51_cast_fp16)[name = tensor("hidden_states_91_cast_fp16")]; + tensor var_6826 = const()[name = tensor("op_6826"), val = tensor([1, 1])]; + tensor var_6828 = const()[name = tensor("op_6828"), val = tensor([1, 1])]; + tensor q_35_pad_type_0 = const()[name = tensor("q_35_pad_type_0"), val = tensor("custom")]; + tensor q_35_pad_0 = const()[name = tensor("q_35_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_4_attn2_to_q_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_4_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(489338304)))]; + tensor q_35_cast_fp16 = conv(dilations = var_6828, groups = var_2624, pad = q_35_pad_0, pad_type = q_35_pad_type_0, strides = var_6826, weight = down_blocks_2_attentions_0_transformer_blocks_4_attn2_to_q_weight_to_fp16, x = hidden_states_91_cast_fp16)[name = tensor("q_35_cast_fp16")]; + tensor var_6832 = const()[name = tensor("op_6832"), val = tensor([1, 1])]; + tensor var_6834 = const()[name = tensor("op_6834"), val = tensor([1, 1])]; + tensor k_69_pad_type_0 = const()[name = tensor("k_69_pad_type_0"), val = tensor("custom")]; + tensor k_69_pad_0 = const()[name = tensor("k_69_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_4_attn2_to_k_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_4_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(492615168)))]; + tensor k_69_cast_fp16 = conv(dilations = var_6834, groups = var_2624, pad = k_69_pad_0, pad_type = k_69_pad_type_0, strides = var_6832, weight = down_blocks_2_attentions_0_transformer_blocks_4_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_69_cast_fp16")]; + tensor var_6838 = const()[name = tensor("op_6838"), val = tensor([1, 1])]; + tensor var_6840 = const()[name = tensor("op_6840"), val = tensor([1, 1])]; + tensor v_35_pad_type_0 = const()[name = tensor("v_35_pad_type_0"), val = tensor("custom")]; + tensor v_35_pad_0 = const()[name = tensor("v_35_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_4_attn2_to_v_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_4_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(497858112)))]; + tensor v_35_cast_fp16 = conv(dilations = var_6840, groups = var_2624, pad = v_35_pad_0, pad_type = v_35_pad_type_0, strides = var_6838, weight = down_blocks_2_attentions_0_transformer_blocks_4_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_35_cast_fp16")]; + tensor var_6844_begin_0 = const()[name = tensor("op_6844_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6844_end_0 = const()[name = tensor("op_6844_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_6844_end_mask_0 = const()[name = tensor("op_6844_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6844_cast_fp16 = slice_by_index(begin = var_6844_begin_0, end = var_6844_end_0, end_mask = var_6844_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_6844_cast_fp16")]; + tensor var_6848_begin_0 = const()[name = tensor("op_6848_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_6848_end_0 = const()[name = tensor("op_6848_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_6848_end_mask_0 = const()[name = tensor("op_6848_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6848_cast_fp16 = slice_by_index(begin = var_6848_begin_0, end = var_6848_end_0, end_mask = var_6848_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_6848_cast_fp16")]; + tensor var_6852_begin_0 = const()[name = tensor("op_6852_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_6852_end_0 = const()[name = tensor("op_6852_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_6852_end_mask_0 = const()[name = tensor("op_6852_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6852_cast_fp16 = slice_by_index(begin = var_6852_begin_0, end = var_6852_end_0, end_mask = var_6852_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_6852_cast_fp16")]; + tensor var_6856_begin_0 = const()[name = tensor("op_6856_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_6856_end_0 = const()[name = tensor("op_6856_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_6856_end_mask_0 = const()[name = tensor("op_6856_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6856_cast_fp16 = slice_by_index(begin = var_6856_begin_0, end = var_6856_end_0, end_mask = var_6856_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_6856_cast_fp16")]; + tensor var_6860_begin_0 = const()[name = tensor("op_6860_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_6860_end_0 = const()[name = tensor("op_6860_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_6860_end_mask_0 = const()[name = tensor("op_6860_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6860_cast_fp16 = slice_by_index(begin = var_6860_begin_0, end = var_6860_end_0, end_mask = var_6860_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_6860_cast_fp16")]; + tensor var_6864_begin_0 = const()[name = tensor("op_6864_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_6864_end_0 = const()[name = tensor("op_6864_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_6864_end_mask_0 = const()[name = tensor("op_6864_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6864_cast_fp16 = slice_by_index(begin = var_6864_begin_0, end = var_6864_end_0, end_mask = var_6864_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_6864_cast_fp16")]; + tensor var_6868_begin_0 = const()[name = tensor("op_6868_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_6868_end_0 = const()[name = tensor("op_6868_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_6868_end_mask_0 = const()[name = tensor("op_6868_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6868_cast_fp16 = slice_by_index(begin = var_6868_begin_0, end = var_6868_end_0, end_mask = var_6868_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_6868_cast_fp16")]; + tensor var_6872_begin_0 = const()[name = tensor("op_6872_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_6872_end_0 = const()[name = tensor("op_6872_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_6872_end_mask_0 = const()[name = tensor("op_6872_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6872_cast_fp16 = slice_by_index(begin = var_6872_begin_0, end = var_6872_end_0, end_mask = var_6872_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_6872_cast_fp16")]; + tensor var_6876_begin_0 = const()[name = tensor("op_6876_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_6876_end_0 = const()[name = tensor("op_6876_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_6876_end_mask_0 = const()[name = tensor("op_6876_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6876_cast_fp16 = slice_by_index(begin = var_6876_begin_0, end = var_6876_end_0, end_mask = var_6876_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_6876_cast_fp16")]; + tensor var_6880_begin_0 = const()[name = tensor("op_6880_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_6880_end_0 = const()[name = tensor("op_6880_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_6880_end_mask_0 = const()[name = tensor("op_6880_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6880_cast_fp16 = slice_by_index(begin = var_6880_begin_0, end = var_6880_end_0, end_mask = var_6880_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_6880_cast_fp16")]; + tensor var_6884_begin_0 = const()[name = tensor("op_6884_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_6884_end_0 = const()[name = tensor("op_6884_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_6884_end_mask_0 = const()[name = tensor("op_6884_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6884_cast_fp16 = slice_by_index(begin = var_6884_begin_0, end = var_6884_end_0, end_mask = var_6884_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_6884_cast_fp16")]; + tensor var_6888_begin_0 = const()[name = tensor("op_6888_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_6888_end_0 = const()[name = tensor("op_6888_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_6888_end_mask_0 = const()[name = tensor("op_6888_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6888_cast_fp16 = slice_by_index(begin = var_6888_begin_0, end = var_6888_end_0, end_mask = var_6888_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_6888_cast_fp16")]; + tensor var_6892_begin_0 = const()[name = tensor("op_6892_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_6892_end_0 = const()[name = tensor("op_6892_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_6892_end_mask_0 = const()[name = tensor("op_6892_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6892_cast_fp16 = slice_by_index(begin = var_6892_begin_0, end = var_6892_end_0, end_mask = var_6892_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_6892_cast_fp16")]; + tensor var_6896_begin_0 = const()[name = tensor("op_6896_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_6896_end_0 = const()[name = tensor("op_6896_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_6896_end_mask_0 = const()[name = tensor("op_6896_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6896_cast_fp16 = slice_by_index(begin = var_6896_begin_0, end = var_6896_end_0, end_mask = var_6896_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_6896_cast_fp16")]; + tensor var_6900_begin_0 = const()[name = tensor("op_6900_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_6900_end_0 = const()[name = tensor("op_6900_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_6900_end_mask_0 = const()[name = tensor("op_6900_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6900_cast_fp16 = slice_by_index(begin = var_6900_begin_0, end = var_6900_end_0, end_mask = var_6900_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_6900_cast_fp16")]; + tensor var_6904_begin_0 = const()[name = tensor("op_6904_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_6904_end_0 = const()[name = tensor("op_6904_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_6904_end_mask_0 = const()[name = tensor("op_6904_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6904_cast_fp16 = slice_by_index(begin = var_6904_begin_0, end = var_6904_end_0, end_mask = var_6904_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_6904_cast_fp16")]; + tensor var_6908_begin_0 = const()[name = tensor("op_6908_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_6908_end_0 = const()[name = tensor("op_6908_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_6908_end_mask_0 = const()[name = tensor("op_6908_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6908_cast_fp16 = slice_by_index(begin = var_6908_begin_0, end = var_6908_end_0, end_mask = var_6908_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_6908_cast_fp16")]; + tensor var_6912_begin_0 = const()[name = tensor("op_6912_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_6912_end_0 = const()[name = tensor("op_6912_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_6912_end_mask_0 = const()[name = tensor("op_6912_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6912_cast_fp16 = slice_by_index(begin = var_6912_begin_0, end = var_6912_end_0, end_mask = var_6912_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_6912_cast_fp16")]; + tensor var_6916_begin_0 = const()[name = tensor("op_6916_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_6916_end_0 = const()[name = tensor("op_6916_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_6916_end_mask_0 = const()[name = tensor("op_6916_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6916_cast_fp16 = slice_by_index(begin = var_6916_begin_0, end = var_6916_end_0, end_mask = var_6916_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_6916_cast_fp16")]; + tensor var_6920_begin_0 = const()[name = tensor("op_6920_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_6920_end_0 = const()[name = tensor("op_6920_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_6920_end_mask_0 = const()[name = tensor("op_6920_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6920_cast_fp16 = slice_by_index(begin = var_6920_begin_0, end = var_6920_end_0, end_mask = var_6920_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_6920_cast_fp16")]; + tensor k_71_perm_0 = const()[name = tensor("k_71_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_6927_begin_0 = const()[name = tensor("op_6927_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6927_end_0 = const()[name = tensor("op_6927_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_6927_end_mask_0 = const()[name = tensor("op_6927_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_71_cast_fp16 = transpose(perm = k_71_perm_0, x = k_69_cast_fp16)[name = tensor("transpose_122")]; + tensor var_6927_cast_fp16 = slice_by_index(begin = var_6927_begin_0, end = var_6927_end_0, end_mask = var_6927_end_mask_0, x = k_71_cast_fp16)[name = tensor("op_6927_cast_fp16")]; + tensor var_6931_begin_0 = const()[name = tensor("op_6931_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_6931_end_0 = const()[name = tensor("op_6931_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_6931_end_mask_0 = const()[name = tensor("op_6931_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6931_cast_fp16 = slice_by_index(begin = var_6931_begin_0, end = var_6931_end_0, end_mask = var_6931_end_mask_0, x = k_71_cast_fp16)[name = tensor("op_6931_cast_fp16")]; + tensor var_6935_begin_0 = const()[name = tensor("op_6935_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_6935_end_0 = const()[name = tensor("op_6935_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_6935_end_mask_0 = const()[name = tensor("op_6935_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6935_cast_fp16 = slice_by_index(begin = var_6935_begin_0, end = var_6935_end_0, end_mask = var_6935_end_mask_0, x = k_71_cast_fp16)[name = tensor("op_6935_cast_fp16")]; + tensor var_6939_begin_0 = const()[name = tensor("op_6939_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_6939_end_0 = const()[name = tensor("op_6939_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_6939_end_mask_0 = const()[name = tensor("op_6939_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6939_cast_fp16 = slice_by_index(begin = var_6939_begin_0, end = var_6939_end_0, end_mask = var_6939_end_mask_0, x = k_71_cast_fp16)[name = tensor("op_6939_cast_fp16")]; + tensor var_6943_begin_0 = const()[name = tensor("op_6943_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_6943_end_0 = const()[name = tensor("op_6943_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_6943_end_mask_0 = const()[name = tensor("op_6943_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6943_cast_fp16 = slice_by_index(begin = var_6943_begin_0, end = var_6943_end_0, end_mask = var_6943_end_mask_0, x = k_71_cast_fp16)[name = tensor("op_6943_cast_fp16")]; + tensor var_6947_begin_0 = const()[name = tensor("op_6947_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_6947_end_0 = const()[name = tensor("op_6947_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_6947_end_mask_0 = const()[name = tensor("op_6947_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6947_cast_fp16 = slice_by_index(begin = var_6947_begin_0, end = var_6947_end_0, end_mask = var_6947_end_mask_0, x = k_71_cast_fp16)[name = tensor("op_6947_cast_fp16")]; + tensor var_6951_begin_0 = const()[name = tensor("op_6951_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_6951_end_0 = const()[name = tensor("op_6951_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_6951_end_mask_0 = const()[name = tensor("op_6951_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6951_cast_fp16 = slice_by_index(begin = var_6951_begin_0, end = var_6951_end_0, end_mask = var_6951_end_mask_0, x = k_71_cast_fp16)[name = tensor("op_6951_cast_fp16")]; + tensor var_6955_begin_0 = const()[name = tensor("op_6955_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_6955_end_0 = const()[name = tensor("op_6955_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_6955_end_mask_0 = const()[name = tensor("op_6955_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6955_cast_fp16 = slice_by_index(begin = var_6955_begin_0, end = var_6955_end_0, end_mask = var_6955_end_mask_0, x = k_71_cast_fp16)[name = tensor("op_6955_cast_fp16")]; + tensor var_6959_begin_0 = const()[name = tensor("op_6959_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_6959_end_0 = const()[name = tensor("op_6959_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_6959_end_mask_0 = const()[name = tensor("op_6959_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6959_cast_fp16 = slice_by_index(begin = var_6959_begin_0, end = var_6959_end_0, end_mask = var_6959_end_mask_0, x = k_71_cast_fp16)[name = tensor("op_6959_cast_fp16")]; + tensor var_6963_begin_0 = const()[name = tensor("op_6963_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_6963_end_0 = const()[name = tensor("op_6963_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_6963_end_mask_0 = const()[name = tensor("op_6963_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6963_cast_fp16 = slice_by_index(begin = var_6963_begin_0, end = var_6963_end_0, end_mask = var_6963_end_mask_0, x = k_71_cast_fp16)[name = tensor("op_6963_cast_fp16")]; + tensor var_6967_begin_0 = const()[name = tensor("op_6967_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_6967_end_0 = const()[name = tensor("op_6967_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_6967_end_mask_0 = const()[name = tensor("op_6967_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6967_cast_fp16 = slice_by_index(begin = var_6967_begin_0, end = var_6967_end_0, end_mask = var_6967_end_mask_0, x = k_71_cast_fp16)[name = tensor("op_6967_cast_fp16")]; + tensor var_6971_begin_0 = const()[name = tensor("op_6971_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_6971_end_0 = const()[name = tensor("op_6971_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_6971_end_mask_0 = const()[name = tensor("op_6971_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6971_cast_fp16 = slice_by_index(begin = var_6971_begin_0, end = var_6971_end_0, end_mask = var_6971_end_mask_0, x = k_71_cast_fp16)[name = tensor("op_6971_cast_fp16")]; + tensor var_6975_begin_0 = const()[name = tensor("op_6975_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_6975_end_0 = const()[name = tensor("op_6975_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_6975_end_mask_0 = const()[name = tensor("op_6975_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6975_cast_fp16 = slice_by_index(begin = var_6975_begin_0, end = var_6975_end_0, end_mask = var_6975_end_mask_0, x = k_71_cast_fp16)[name = tensor("op_6975_cast_fp16")]; + tensor var_6979_begin_0 = const()[name = tensor("op_6979_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_6979_end_0 = const()[name = tensor("op_6979_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_6979_end_mask_0 = const()[name = tensor("op_6979_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6979_cast_fp16 = slice_by_index(begin = var_6979_begin_0, end = var_6979_end_0, end_mask = var_6979_end_mask_0, x = k_71_cast_fp16)[name = tensor("op_6979_cast_fp16")]; + tensor var_6983_begin_0 = const()[name = tensor("op_6983_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_6983_end_0 = const()[name = tensor("op_6983_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_6983_end_mask_0 = const()[name = tensor("op_6983_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6983_cast_fp16 = slice_by_index(begin = var_6983_begin_0, end = var_6983_end_0, end_mask = var_6983_end_mask_0, x = k_71_cast_fp16)[name = tensor("op_6983_cast_fp16")]; + tensor var_6987_begin_0 = const()[name = tensor("op_6987_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_6987_end_0 = const()[name = tensor("op_6987_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_6987_end_mask_0 = const()[name = tensor("op_6987_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6987_cast_fp16 = slice_by_index(begin = var_6987_begin_0, end = var_6987_end_0, end_mask = var_6987_end_mask_0, x = k_71_cast_fp16)[name = tensor("op_6987_cast_fp16")]; + tensor var_6991_begin_0 = const()[name = tensor("op_6991_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_6991_end_0 = const()[name = tensor("op_6991_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_6991_end_mask_0 = const()[name = tensor("op_6991_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6991_cast_fp16 = slice_by_index(begin = var_6991_begin_0, end = var_6991_end_0, end_mask = var_6991_end_mask_0, x = k_71_cast_fp16)[name = tensor("op_6991_cast_fp16")]; + tensor var_6995_begin_0 = const()[name = tensor("op_6995_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_6995_end_0 = const()[name = tensor("op_6995_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_6995_end_mask_0 = const()[name = tensor("op_6995_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6995_cast_fp16 = slice_by_index(begin = var_6995_begin_0, end = var_6995_end_0, end_mask = var_6995_end_mask_0, x = k_71_cast_fp16)[name = tensor("op_6995_cast_fp16")]; + tensor var_6999_begin_0 = const()[name = tensor("op_6999_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_6999_end_0 = const()[name = tensor("op_6999_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_6999_end_mask_0 = const()[name = tensor("op_6999_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6999_cast_fp16 = slice_by_index(begin = var_6999_begin_0, end = var_6999_end_0, end_mask = var_6999_end_mask_0, x = k_71_cast_fp16)[name = tensor("op_6999_cast_fp16")]; + tensor var_7003_begin_0 = const()[name = tensor("op_7003_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_7003_end_0 = const()[name = tensor("op_7003_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_7003_end_mask_0 = const()[name = tensor("op_7003_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7003_cast_fp16 = slice_by_index(begin = var_7003_begin_0, end = var_7003_end_0, end_mask = var_7003_end_mask_0, x = k_71_cast_fp16)[name = tensor("op_7003_cast_fp16")]; + tensor var_7005_begin_0 = const()[name = tensor("op_7005_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7005_end_0 = const()[name = tensor("op_7005_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_7005_end_mask_0 = const()[name = tensor("op_7005_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7005_cast_fp16 = slice_by_index(begin = var_7005_begin_0, end = var_7005_end_0, end_mask = var_7005_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7005_cast_fp16")]; + tensor var_7009_begin_0 = const()[name = tensor("op_7009_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_7009_end_0 = const()[name = tensor("op_7009_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_7009_end_mask_0 = const()[name = tensor("op_7009_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7009_cast_fp16 = slice_by_index(begin = var_7009_begin_0, end = var_7009_end_0, end_mask = var_7009_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7009_cast_fp16")]; + tensor var_7013_begin_0 = const()[name = tensor("op_7013_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_7013_end_0 = const()[name = tensor("op_7013_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_7013_end_mask_0 = const()[name = tensor("op_7013_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7013_cast_fp16 = slice_by_index(begin = var_7013_begin_0, end = var_7013_end_0, end_mask = var_7013_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7013_cast_fp16")]; + tensor var_7017_begin_0 = const()[name = tensor("op_7017_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_7017_end_0 = const()[name = tensor("op_7017_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_7017_end_mask_0 = const()[name = tensor("op_7017_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7017_cast_fp16 = slice_by_index(begin = var_7017_begin_0, end = var_7017_end_0, end_mask = var_7017_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7017_cast_fp16")]; + tensor var_7021_begin_0 = const()[name = tensor("op_7021_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_7021_end_0 = const()[name = tensor("op_7021_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_7021_end_mask_0 = const()[name = tensor("op_7021_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7021_cast_fp16 = slice_by_index(begin = var_7021_begin_0, end = var_7021_end_0, end_mask = var_7021_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7021_cast_fp16")]; + tensor var_7025_begin_0 = const()[name = tensor("op_7025_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_7025_end_0 = const()[name = tensor("op_7025_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_7025_end_mask_0 = const()[name = tensor("op_7025_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7025_cast_fp16 = slice_by_index(begin = var_7025_begin_0, end = var_7025_end_0, end_mask = var_7025_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7025_cast_fp16")]; + tensor var_7029_begin_0 = const()[name = tensor("op_7029_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_7029_end_0 = const()[name = tensor("op_7029_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_7029_end_mask_0 = const()[name = tensor("op_7029_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7029_cast_fp16 = slice_by_index(begin = var_7029_begin_0, end = var_7029_end_0, end_mask = var_7029_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7029_cast_fp16")]; + tensor var_7033_begin_0 = const()[name = tensor("op_7033_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_7033_end_0 = const()[name = tensor("op_7033_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_7033_end_mask_0 = const()[name = tensor("op_7033_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7033_cast_fp16 = slice_by_index(begin = var_7033_begin_0, end = var_7033_end_0, end_mask = var_7033_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7033_cast_fp16")]; + tensor var_7037_begin_0 = const()[name = tensor("op_7037_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_7037_end_0 = const()[name = tensor("op_7037_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_7037_end_mask_0 = const()[name = tensor("op_7037_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7037_cast_fp16 = slice_by_index(begin = var_7037_begin_0, end = var_7037_end_0, end_mask = var_7037_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7037_cast_fp16")]; + tensor var_7041_begin_0 = const()[name = tensor("op_7041_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_7041_end_0 = const()[name = tensor("op_7041_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_7041_end_mask_0 = const()[name = tensor("op_7041_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7041_cast_fp16 = slice_by_index(begin = var_7041_begin_0, end = var_7041_end_0, end_mask = var_7041_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7041_cast_fp16")]; + tensor var_7045_begin_0 = const()[name = tensor("op_7045_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_7045_end_0 = const()[name = tensor("op_7045_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_7045_end_mask_0 = const()[name = tensor("op_7045_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7045_cast_fp16 = slice_by_index(begin = var_7045_begin_0, end = var_7045_end_0, end_mask = var_7045_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7045_cast_fp16")]; + tensor var_7049_begin_0 = const()[name = tensor("op_7049_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_7049_end_0 = const()[name = tensor("op_7049_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_7049_end_mask_0 = const()[name = tensor("op_7049_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7049_cast_fp16 = slice_by_index(begin = var_7049_begin_0, end = var_7049_end_0, end_mask = var_7049_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7049_cast_fp16")]; + tensor var_7053_begin_0 = const()[name = tensor("op_7053_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_7053_end_0 = const()[name = tensor("op_7053_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_7053_end_mask_0 = const()[name = tensor("op_7053_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7053_cast_fp16 = slice_by_index(begin = var_7053_begin_0, end = var_7053_end_0, end_mask = var_7053_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7053_cast_fp16")]; + tensor var_7057_begin_0 = const()[name = tensor("op_7057_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_7057_end_0 = const()[name = tensor("op_7057_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_7057_end_mask_0 = const()[name = tensor("op_7057_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7057_cast_fp16 = slice_by_index(begin = var_7057_begin_0, end = var_7057_end_0, end_mask = var_7057_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7057_cast_fp16")]; + tensor var_7061_begin_0 = const()[name = tensor("op_7061_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_7061_end_0 = const()[name = tensor("op_7061_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_7061_end_mask_0 = const()[name = tensor("op_7061_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7061_cast_fp16 = slice_by_index(begin = var_7061_begin_0, end = var_7061_end_0, end_mask = var_7061_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7061_cast_fp16")]; + tensor var_7065_begin_0 = const()[name = tensor("op_7065_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_7065_end_0 = const()[name = tensor("op_7065_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_7065_end_mask_0 = const()[name = tensor("op_7065_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7065_cast_fp16 = slice_by_index(begin = var_7065_begin_0, end = var_7065_end_0, end_mask = var_7065_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7065_cast_fp16")]; + tensor var_7069_begin_0 = const()[name = tensor("op_7069_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_7069_end_0 = const()[name = tensor("op_7069_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_7069_end_mask_0 = const()[name = tensor("op_7069_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7069_cast_fp16 = slice_by_index(begin = var_7069_begin_0, end = var_7069_end_0, end_mask = var_7069_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7069_cast_fp16")]; + tensor var_7073_begin_0 = const()[name = tensor("op_7073_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_7073_end_0 = const()[name = tensor("op_7073_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_7073_end_mask_0 = const()[name = tensor("op_7073_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7073_cast_fp16 = slice_by_index(begin = var_7073_begin_0, end = var_7073_end_0, end_mask = var_7073_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7073_cast_fp16")]; + tensor var_7077_begin_0 = const()[name = tensor("op_7077_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_7077_end_0 = const()[name = tensor("op_7077_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_7077_end_mask_0 = const()[name = tensor("op_7077_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7077_cast_fp16 = slice_by_index(begin = var_7077_begin_0, end = var_7077_end_0, end_mask = var_7077_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7077_cast_fp16")]; + tensor var_7081_begin_0 = const()[name = tensor("op_7081_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_7081_end_0 = const()[name = tensor("op_7081_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_7081_end_mask_0 = const()[name = tensor("op_7081_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7081_cast_fp16 = slice_by_index(begin = var_7081_begin_0, end = var_7081_end_0, end_mask = var_7081_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7081_cast_fp16")]; + tensor var_7085_equation_0 = const()[name = tensor("op_7085_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7085_cast_fp16 = einsum(equation = var_7085_equation_0, values = (var_6927_cast_fp16, var_6844_cast_fp16))[name = tensor("op_7085_cast_fp16")]; + tensor var_7086_to_fp16 = const()[name = tensor("op_7086_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_521_cast_fp16 = mul(x = var_7085_cast_fp16, y = var_7086_to_fp16)[name = tensor("aw_521_cast_fp16")]; + tensor var_7089_equation_0 = const()[name = tensor("op_7089_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7089_cast_fp16 = einsum(equation = var_7089_equation_0, values = (var_6931_cast_fp16, var_6848_cast_fp16))[name = tensor("op_7089_cast_fp16")]; + tensor var_7090_to_fp16 = const()[name = tensor("op_7090_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_523_cast_fp16 = mul(x = var_7089_cast_fp16, y = var_7090_to_fp16)[name = tensor("aw_523_cast_fp16")]; + tensor var_7093_equation_0 = const()[name = tensor("op_7093_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7093_cast_fp16 = einsum(equation = var_7093_equation_0, values = (var_6935_cast_fp16, var_6852_cast_fp16))[name = tensor("op_7093_cast_fp16")]; + tensor var_7094_to_fp16 = const()[name = tensor("op_7094_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_525_cast_fp16 = mul(x = var_7093_cast_fp16, y = var_7094_to_fp16)[name = tensor("aw_525_cast_fp16")]; + tensor var_7097_equation_0 = const()[name = tensor("op_7097_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7097_cast_fp16 = einsum(equation = var_7097_equation_0, values = (var_6939_cast_fp16, var_6856_cast_fp16))[name = tensor("op_7097_cast_fp16")]; + tensor var_7098_to_fp16 = const()[name = tensor("op_7098_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_527_cast_fp16 = mul(x = var_7097_cast_fp16, y = var_7098_to_fp16)[name = tensor("aw_527_cast_fp16")]; + tensor var_7101_equation_0 = const()[name = tensor("op_7101_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7101_cast_fp16 = einsum(equation = var_7101_equation_0, values = (var_6943_cast_fp16, var_6860_cast_fp16))[name = tensor("op_7101_cast_fp16")]; + tensor var_7102_to_fp16 = const()[name = tensor("op_7102_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_529_cast_fp16 = mul(x = var_7101_cast_fp16, y = var_7102_to_fp16)[name = tensor("aw_529_cast_fp16")]; + tensor var_7105_equation_0 = const()[name = tensor("op_7105_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7105_cast_fp16 = einsum(equation = var_7105_equation_0, values = (var_6947_cast_fp16, var_6864_cast_fp16))[name = tensor("op_7105_cast_fp16")]; + tensor var_7106_to_fp16 = const()[name = tensor("op_7106_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_531_cast_fp16 = mul(x = var_7105_cast_fp16, y = var_7106_to_fp16)[name = tensor("aw_531_cast_fp16")]; + tensor var_7109_equation_0 = const()[name = tensor("op_7109_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7109_cast_fp16 = einsum(equation = var_7109_equation_0, values = (var_6951_cast_fp16, var_6868_cast_fp16))[name = tensor("op_7109_cast_fp16")]; + tensor var_7110_to_fp16 = const()[name = tensor("op_7110_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_533_cast_fp16 = mul(x = var_7109_cast_fp16, y = var_7110_to_fp16)[name = tensor("aw_533_cast_fp16")]; + tensor var_7113_equation_0 = const()[name = tensor("op_7113_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7113_cast_fp16 = einsum(equation = var_7113_equation_0, values = (var_6955_cast_fp16, var_6872_cast_fp16))[name = tensor("op_7113_cast_fp16")]; + tensor var_7114_to_fp16 = const()[name = tensor("op_7114_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_535_cast_fp16 = mul(x = var_7113_cast_fp16, y = var_7114_to_fp16)[name = tensor("aw_535_cast_fp16")]; + tensor var_7117_equation_0 = const()[name = tensor("op_7117_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7117_cast_fp16 = einsum(equation = var_7117_equation_0, values = (var_6959_cast_fp16, var_6876_cast_fp16))[name = tensor("op_7117_cast_fp16")]; + tensor var_7118_to_fp16 = const()[name = tensor("op_7118_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_537_cast_fp16 = mul(x = var_7117_cast_fp16, y = var_7118_to_fp16)[name = tensor("aw_537_cast_fp16")]; + tensor var_7121_equation_0 = const()[name = tensor("op_7121_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7121_cast_fp16 = einsum(equation = var_7121_equation_0, values = (var_6963_cast_fp16, var_6880_cast_fp16))[name = tensor("op_7121_cast_fp16")]; + tensor var_7122_to_fp16 = const()[name = tensor("op_7122_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_539_cast_fp16 = mul(x = var_7121_cast_fp16, y = var_7122_to_fp16)[name = tensor("aw_539_cast_fp16")]; + tensor var_7125_equation_0 = const()[name = tensor("op_7125_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7125_cast_fp16 = einsum(equation = var_7125_equation_0, values = (var_6967_cast_fp16, var_6884_cast_fp16))[name = tensor("op_7125_cast_fp16")]; + tensor var_7126_to_fp16 = const()[name = tensor("op_7126_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_541_cast_fp16 = mul(x = var_7125_cast_fp16, y = var_7126_to_fp16)[name = tensor("aw_541_cast_fp16")]; + tensor var_7129_equation_0 = const()[name = tensor("op_7129_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7129_cast_fp16 = einsum(equation = var_7129_equation_0, values = (var_6971_cast_fp16, var_6888_cast_fp16))[name = tensor("op_7129_cast_fp16")]; + tensor var_7130_to_fp16 = const()[name = tensor("op_7130_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_543_cast_fp16 = mul(x = var_7129_cast_fp16, y = var_7130_to_fp16)[name = tensor("aw_543_cast_fp16")]; + tensor var_7133_equation_0 = const()[name = tensor("op_7133_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7133_cast_fp16 = einsum(equation = var_7133_equation_0, values = (var_6975_cast_fp16, var_6892_cast_fp16))[name = tensor("op_7133_cast_fp16")]; + tensor var_7134_to_fp16 = const()[name = tensor("op_7134_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_545_cast_fp16 = mul(x = var_7133_cast_fp16, y = var_7134_to_fp16)[name = tensor("aw_545_cast_fp16")]; + tensor var_7137_equation_0 = const()[name = tensor("op_7137_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7137_cast_fp16 = einsum(equation = var_7137_equation_0, values = (var_6979_cast_fp16, var_6896_cast_fp16))[name = tensor("op_7137_cast_fp16")]; + tensor var_7138_to_fp16 = const()[name = tensor("op_7138_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_547_cast_fp16 = mul(x = var_7137_cast_fp16, y = var_7138_to_fp16)[name = tensor("aw_547_cast_fp16")]; + tensor var_7141_equation_0 = const()[name = tensor("op_7141_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7141_cast_fp16 = einsum(equation = var_7141_equation_0, values = (var_6983_cast_fp16, var_6900_cast_fp16))[name = tensor("op_7141_cast_fp16")]; + tensor var_7142_to_fp16 = const()[name = tensor("op_7142_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_549_cast_fp16 = mul(x = var_7141_cast_fp16, y = var_7142_to_fp16)[name = tensor("aw_549_cast_fp16")]; + tensor var_7145_equation_0 = const()[name = tensor("op_7145_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7145_cast_fp16 = einsum(equation = var_7145_equation_0, values = (var_6987_cast_fp16, var_6904_cast_fp16))[name = tensor("op_7145_cast_fp16")]; + tensor var_7146_to_fp16 = const()[name = tensor("op_7146_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_551_cast_fp16 = mul(x = var_7145_cast_fp16, y = var_7146_to_fp16)[name = tensor("aw_551_cast_fp16")]; + tensor var_7149_equation_0 = const()[name = tensor("op_7149_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7149_cast_fp16 = einsum(equation = var_7149_equation_0, values = (var_6991_cast_fp16, var_6908_cast_fp16))[name = tensor("op_7149_cast_fp16")]; + tensor var_7150_to_fp16 = const()[name = tensor("op_7150_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_553_cast_fp16 = mul(x = var_7149_cast_fp16, y = var_7150_to_fp16)[name = tensor("aw_553_cast_fp16")]; + tensor var_7153_equation_0 = const()[name = tensor("op_7153_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7153_cast_fp16 = einsum(equation = var_7153_equation_0, values = (var_6995_cast_fp16, var_6912_cast_fp16))[name = tensor("op_7153_cast_fp16")]; + tensor var_7154_to_fp16 = const()[name = tensor("op_7154_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_555_cast_fp16 = mul(x = var_7153_cast_fp16, y = var_7154_to_fp16)[name = tensor("aw_555_cast_fp16")]; + tensor var_7157_equation_0 = const()[name = tensor("op_7157_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7157_cast_fp16 = einsum(equation = var_7157_equation_0, values = (var_6999_cast_fp16, var_6916_cast_fp16))[name = tensor("op_7157_cast_fp16")]; + tensor var_7158_to_fp16 = const()[name = tensor("op_7158_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_557_cast_fp16 = mul(x = var_7157_cast_fp16, y = var_7158_to_fp16)[name = tensor("aw_557_cast_fp16")]; + tensor var_7161_equation_0 = const()[name = tensor("op_7161_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7161_cast_fp16 = einsum(equation = var_7161_equation_0, values = (var_7003_cast_fp16, var_6920_cast_fp16))[name = tensor("op_7161_cast_fp16")]; + tensor var_7162_to_fp16 = const()[name = tensor("op_7162_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_559_cast_fp16 = mul(x = var_7161_cast_fp16, y = var_7162_to_fp16)[name = tensor("aw_559_cast_fp16")]; + tensor var_7164_cast_fp16 = softmax(axis = var_2624, x = aw_521_cast_fp16)[name = tensor("op_7164_cast_fp16")]; + tensor var_7165_cast_fp16 = softmax(axis = var_2624, x = aw_523_cast_fp16)[name = tensor("op_7165_cast_fp16")]; + tensor var_7166_cast_fp16 = softmax(axis = var_2624, x = aw_525_cast_fp16)[name = tensor("op_7166_cast_fp16")]; + tensor var_7167_cast_fp16 = softmax(axis = var_2624, x = aw_527_cast_fp16)[name = tensor("op_7167_cast_fp16")]; + tensor var_7168_cast_fp16 = softmax(axis = var_2624, x = aw_529_cast_fp16)[name = tensor("op_7168_cast_fp16")]; + tensor var_7169_cast_fp16 = softmax(axis = var_2624, x = aw_531_cast_fp16)[name = tensor("op_7169_cast_fp16")]; + tensor var_7170_cast_fp16 = softmax(axis = var_2624, x = aw_533_cast_fp16)[name = tensor("op_7170_cast_fp16")]; + tensor var_7171_cast_fp16 = softmax(axis = var_2624, x = aw_535_cast_fp16)[name = tensor("op_7171_cast_fp16")]; + tensor var_7172_cast_fp16 = softmax(axis = var_2624, x = aw_537_cast_fp16)[name = tensor("op_7172_cast_fp16")]; + tensor var_7173_cast_fp16 = softmax(axis = var_2624, x = aw_539_cast_fp16)[name = tensor("op_7173_cast_fp16")]; + tensor var_7174_cast_fp16 = softmax(axis = var_2624, x = aw_541_cast_fp16)[name = tensor("op_7174_cast_fp16")]; + tensor var_7175_cast_fp16 = softmax(axis = var_2624, x = aw_543_cast_fp16)[name = tensor("op_7175_cast_fp16")]; + tensor var_7176_cast_fp16 = softmax(axis = var_2624, x = aw_545_cast_fp16)[name = tensor("op_7176_cast_fp16")]; + tensor var_7177_cast_fp16 = softmax(axis = var_2624, x = aw_547_cast_fp16)[name = tensor("op_7177_cast_fp16")]; + tensor var_7178_cast_fp16 = softmax(axis = var_2624, x = aw_549_cast_fp16)[name = tensor("op_7178_cast_fp16")]; + tensor var_7179_cast_fp16 = softmax(axis = var_2624, x = aw_551_cast_fp16)[name = tensor("op_7179_cast_fp16")]; + tensor var_7180_cast_fp16 = softmax(axis = var_2624, x = aw_553_cast_fp16)[name = tensor("op_7180_cast_fp16")]; + tensor var_7181_cast_fp16 = softmax(axis = var_2624, x = aw_555_cast_fp16)[name = tensor("op_7181_cast_fp16")]; + tensor var_7182_cast_fp16 = softmax(axis = var_2624, x = aw_557_cast_fp16)[name = tensor("op_7182_cast_fp16")]; + tensor var_7183_cast_fp16 = softmax(axis = var_2624, x = aw_559_cast_fp16)[name = tensor("op_7183_cast_fp16")]; + tensor var_7185_equation_0 = const()[name = tensor("op_7185_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7185_cast_fp16 = einsum(equation = var_7185_equation_0, values = (var_7005_cast_fp16, var_7164_cast_fp16))[name = tensor("op_7185_cast_fp16")]; + tensor var_7187_equation_0 = const()[name = tensor("op_7187_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7187_cast_fp16 = einsum(equation = var_7187_equation_0, values = (var_7009_cast_fp16, var_7165_cast_fp16))[name = tensor("op_7187_cast_fp16")]; + tensor var_7189_equation_0 = const()[name = tensor("op_7189_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7189_cast_fp16 = einsum(equation = var_7189_equation_0, values = (var_7013_cast_fp16, var_7166_cast_fp16))[name = tensor("op_7189_cast_fp16")]; + tensor var_7191_equation_0 = const()[name = tensor("op_7191_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7191_cast_fp16 = einsum(equation = var_7191_equation_0, values = (var_7017_cast_fp16, var_7167_cast_fp16))[name = tensor("op_7191_cast_fp16")]; + tensor var_7193_equation_0 = const()[name = tensor("op_7193_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7193_cast_fp16 = einsum(equation = var_7193_equation_0, values = (var_7021_cast_fp16, var_7168_cast_fp16))[name = tensor("op_7193_cast_fp16")]; + tensor var_7195_equation_0 = const()[name = tensor("op_7195_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7195_cast_fp16 = einsum(equation = var_7195_equation_0, values = (var_7025_cast_fp16, var_7169_cast_fp16))[name = tensor("op_7195_cast_fp16")]; + tensor var_7197_equation_0 = const()[name = tensor("op_7197_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7197_cast_fp16 = einsum(equation = var_7197_equation_0, values = (var_7029_cast_fp16, var_7170_cast_fp16))[name = tensor("op_7197_cast_fp16")]; + tensor var_7199_equation_0 = const()[name = tensor("op_7199_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7199_cast_fp16 = einsum(equation = var_7199_equation_0, values = (var_7033_cast_fp16, var_7171_cast_fp16))[name = tensor("op_7199_cast_fp16")]; + tensor var_7201_equation_0 = const()[name = tensor("op_7201_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7201_cast_fp16 = einsum(equation = var_7201_equation_0, values = (var_7037_cast_fp16, var_7172_cast_fp16))[name = tensor("op_7201_cast_fp16")]; + tensor var_7203_equation_0 = const()[name = tensor("op_7203_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7203_cast_fp16 = einsum(equation = var_7203_equation_0, values = (var_7041_cast_fp16, var_7173_cast_fp16))[name = tensor("op_7203_cast_fp16")]; + tensor var_7205_equation_0 = const()[name = tensor("op_7205_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7205_cast_fp16 = einsum(equation = var_7205_equation_0, values = (var_7045_cast_fp16, var_7174_cast_fp16))[name = tensor("op_7205_cast_fp16")]; + tensor var_7207_equation_0 = const()[name = tensor("op_7207_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7207_cast_fp16 = einsum(equation = var_7207_equation_0, values = (var_7049_cast_fp16, var_7175_cast_fp16))[name = tensor("op_7207_cast_fp16")]; + tensor var_7209_equation_0 = const()[name = tensor("op_7209_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7209_cast_fp16 = einsum(equation = var_7209_equation_0, values = (var_7053_cast_fp16, var_7176_cast_fp16))[name = tensor("op_7209_cast_fp16")]; + tensor var_7211_equation_0 = const()[name = tensor("op_7211_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7211_cast_fp16 = einsum(equation = var_7211_equation_0, values = (var_7057_cast_fp16, var_7177_cast_fp16))[name = tensor("op_7211_cast_fp16")]; + tensor var_7213_equation_0 = const()[name = tensor("op_7213_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7213_cast_fp16 = einsum(equation = var_7213_equation_0, values = (var_7061_cast_fp16, var_7178_cast_fp16))[name = tensor("op_7213_cast_fp16")]; + tensor var_7215_equation_0 = const()[name = tensor("op_7215_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7215_cast_fp16 = einsum(equation = var_7215_equation_0, values = (var_7065_cast_fp16, var_7179_cast_fp16))[name = tensor("op_7215_cast_fp16")]; + tensor var_7217_equation_0 = const()[name = tensor("op_7217_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7217_cast_fp16 = einsum(equation = var_7217_equation_0, values = (var_7069_cast_fp16, var_7180_cast_fp16))[name = tensor("op_7217_cast_fp16")]; + tensor var_7219_equation_0 = const()[name = tensor("op_7219_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7219_cast_fp16 = einsum(equation = var_7219_equation_0, values = (var_7073_cast_fp16, var_7181_cast_fp16))[name = tensor("op_7219_cast_fp16")]; + tensor var_7221_equation_0 = const()[name = tensor("op_7221_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7221_cast_fp16 = einsum(equation = var_7221_equation_0, values = (var_7077_cast_fp16, var_7182_cast_fp16))[name = tensor("op_7221_cast_fp16")]; + tensor var_7223_equation_0 = const()[name = tensor("op_7223_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7223_cast_fp16 = einsum(equation = var_7223_equation_0, values = (var_7081_cast_fp16, var_7183_cast_fp16))[name = tensor("op_7223_cast_fp16")]; + tensor input_165_interleave_0 = const()[name = tensor("input_165_interleave_0"), val = tensor(false)]; + tensor input_165_cast_fp16 = concat(axis = var_2624, interleave = input_165_interleave_0, values = (var_7185_cast_fp16, var_7187_cast_fp16, var_7189_cast_fp16, var_7191_cast_fp16, var_7193_cast_fp16, var_7195_cast_fp16, var_7197_cast_fp16, var_7199_cast_fp16, var_7201_cast_fp16, var_7203_cast_fp16, var_7205_cast_fp16, var_7207_cast_fp16, var_7209_cast_fp16, var_7211_cast_fp16, var_7213_cast_fp16, var_7215_cast_fp16, var_7217_cast_fp16, var_7219_cast_fp16, var_7221_cast_fp16, var_7223_cast_fp16))[name = tensor("input_165_cast_fp16")]; + tensor var_7229 = const()[name = tensor("op_7229"), val = tensor([1, 1])]; + tensor var_7231 = const()[name = tensor("op_7231"), val = tensor([1, 1])]; + tensor var_7233_pad_type_0 = const()[name = tensor("op_7233_pad_type_0"), val = tensor("custom")]; + tensor var_7233_pad_0 = const()[name = tensor("op_7233_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_4_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_4_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(503101056)))]; + tensor down_blocks_2_attentions_0_transformer_blocks_4_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_4_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(506377920)))]; + tensor var_7233_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_4_attn2_to_out_0_bias_to_fp16, dilations = var_7231, groups = var_2624, pad = var_7233_pad_0, pad_type = var_7233_pad_type_0, strides = var_7229, weight = down_blocks_2_attentions_0_transformer_blocks_4_attn2_to_out_0_weight_to_fp16, x = input_165_cast_fp16)[name = tensor("op_7233_cast_fp16")]; + tensor inputs_53_cast_fp16 = add(x = var_7233_cast_fp16, y = inputs_51_cast_fp16)[name = tensor("inputs_53_cast_fp16")]; + tensor input_167_axes_0 = const()[name = tensor("input_167_axes_0"), val = tensor([1])]; + tensor input_167_gamma_0_to_fp16 = const()[name = tensor("input_167_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(506380544)))]; + tensor input_167_beta_0_to_fp16 = const()[name = tensor("input_167_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(506383168)))]; + tensor var_7243_to_fp16 = const()[name = tensor("op_7243_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_167_cast_fp16 = layer_norm(axes = input_167_axes_0, beta = input_167_beta_0_to_fp16, epsilon = var_7243_to_fp16, gamma = input_167_gamma_0_to_fp16, x = inputs_53_cast_fp16)[name = tensor("input_167_cast_fp16")]; + tensor var_7259 = const()[name = tensor("op_7259"), val = tensor([1, 1])]; + tensor var_7261 = const()[name = tensor("op_7261"), val = tensor([1, 1])]; + tensor var_7263_pad_type_0 = const()[name = tensor("op_7263_pad_type_0"), val = tensor("custom")]; + tensor var_7263_pad_0 = const()[name = tensor("op_7263_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_4_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_4_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(506385792)))]; + tensor down_blocks_2_attentions_0_transformer_blocks_4_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_4_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(532600256)))]; + tensor var_7263_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_4_ff_net_0_proj_bias_to_fp16, dilations = var_7261, groups = var_2624, pad = var_7263_pad_0, pad_type = var_7263_pad_type_0, strides = var_7259, weight = down_blocks_2_attentions_0_transformer_blocks_4_ff_net_0_proj_weight_to_fp16, x = input_167_cast_fp16)[name = tensor("op_7263_cast_fp16")]; + tensor var_7264_split_sizes_0 = const()[name = tensor("op_7264_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_7264_axis_0 = const()[name = tensor("op_7264_axis_0"), val = tensor(1)]; + tensor var_7264_cast_fp16_0, tensor var_7264_cast_fp16_1 = split(axis = var_7264_axis_0, split_sizes = var_7264_split_sizes_0, x = var_7263_cast_fp16)[name = tensor("op_7264_cast_fp16")]; + tensor var_7266_mode_0 = const()[name = tensor("op_7266_mode_0"), val = tensor("EXACT")]; + tensor var_7266_cast_fp16 = gelu(mode = var_7266_mode_0, x = var_7264_cast_fp16_1)[name = tensor("op_7266_cast_fp16")]; + tensor input_169_cast_fp16 = mul(x = var_7264_cast_fp16_0, y = var_7266_cast_fp16)[name = tensor("input_169_cast_fp16")]; + tensor var_7270 = const()[name = tensor("op_7270"), val = tensor([1, 1])]; + tensor var_7272 = const()[name = tensor("op_7272"), val = tensor([1, 1])]; + tensor var_7274_pad_type_0 = const()[name = tensor("op_7274_pad_type_0"), val = tensor("custom")]; + tensor var_7274_pad_0 = const()[name = tensor("op_7274_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_4_ff_net_2_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_4_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(532620800)))]; + tensor down_blocks_2_attentions_0_transformer_blocks_4_ff_net_2_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_4_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(545728064)))]; + tensor var_7274_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_4_ff_net_2_bias_to_fp16, dilations = var_7272, groups = var_2624, pad = var_7274_pad_0, pad_type = var_7274_pad_type_0, strides = var_7270, weight = down_blocks_2_attentions_0_transformer_blocks_4_ff_net_2_weight_to_fp16, x = input_169_cast_fp16)[name = tensor("op_7274_cast_fp16")]; + tensor inputs_55_cast_fp16 = add(x = var_7274_cast_fp16, y = inputs_53_cast_fp16)[name = tensor("inputs_55_cast_fp16")]; + tensor hidden_states_95_axes_0 = const()[name = tensor("hidden_states_95_axes_0"), val = tensor([1])]; + tensor hidden_states_95_gamma_0_to_fp16 = const()[name = tensor("hidden_states_95_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(545730688)))]; + tensor hidden_states_95_beta_0_to_fp16 = const()[name = tensor("hidden_states_95_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(545733312)))]; + tensor var_7290_to_fp16 = const()[name = tensor("op_7290_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_95_cast_fp16 = layer_norm(axes = hidden_states_95_axes_0, beta = hidden_states_95_beta_0_to_fp16, epsilon = var_7290_to_fp16, gamma = hidden_states_95_gamma_0_to_fp16, x = inputs_55_cast_fp16)[name = tensor("hidden_states_95_cast_fp16")]; + tensor var_7305 = const()[name = tensor("op_7305"), val = tensor([1, 1])]; + tensor var_7307 = const()[name = tensor("op_7307"), val = tensor([1, 1])]; + tensor q_37_pad_type_0 = const()[name = tensor("q_37_pad_type_0"), val = tensor("custom")]; + tensor q_37_pad_0 = const()[name = tensor("q_37_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_5_attn1_to_q_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_5_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(545735936)))]; + tensor q_37_cast_fp16 = conv(dilations = var_7307, groups = var_2624, pad = q_37_pad_0, pad_type = q_37_pad_type_0, strides = var_7305, weight = down_blocks_2_attentions_0_transformer_blocks_5_attn1_to_q_weight_to_fp16, x = hidden_states_95_cast_fp16)[name = tensor("q_37_cast_fp16")]; + tensor var_7311 = const()[name = tensor("op_7311"), val = tensor([1, 1])]; + tensor var_7313 = const()[name = tensor("op_7313"), val = tensor([1, 1])]; + tensor k_73_pad_type_0 = const()[name = tensor("k_73_pad_type_0"), val = tensor("custom")]; + tensor k_73_pad_0 = const()[name = tensor("k_73_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_5_attn1_to_k_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_5_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(549012800)))]; + tensor k_73_cast_fp16 = conv(dilations = var_7313, groups = var_2624, pad = k_73_pad_0, pad_type = k_73_pad_type_0, strides = var_7311, weight = down_blocks_2_attentions_0_transformer_blocks_5_attn1_to_k_weight_to_fp16, x = hidden_states_95_cast_fp16)[name = tensor("k_73_cast_fp16")]; + tensor var_7317 = const()[name = tensor("op_7317"), val = tensor([1, 1])]; + tensor var_7319 = const()[name = tensor("op_7319"), val = tensor([1, 1])]; + tensor v_37_pad_type_0 = const()[name = tensor("v_37_pad_type_0"), val = tensor("custom")]; + tensor v_37_pad_0 = const()[name = tensor("v_37_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_5_attn1_to_v_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_5_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(552289664)))]; + tensor v_37_cast_fp16 = conv(dilations = var_7319, groups = var_2624, pad = v_37_pad_0, pad_type = v_37_pad_type_0, strides = var_7317, weight = down_blocks_2_attentions_0_transformer_blocks_5_attn1_to_v_weight_to_fp16, x = hidden_states_95_cast_fp16)[name = tensor("v_37_cast_fp16")]; + tensor var_7323_begin_0 = const()[name = tensor("op_7323_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7323_end_0 = const()[name = tensor("op_7323_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_7323_end_mask_0 = const()[name = tensor("op_7323_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7323_cast_fp16 = slice_by_index(begin = var_7323_begin_0, end = var_7323_end_0, end_mask = var_7323_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7323_cast_fp16")]; + tensor var_7327_begin_0 = const()[name = tensor("op_7327_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_7327_end_0 = const()[name = tensor("op_7327_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_7327_end_mask_0 = const()[name = tensor("op_7327_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7327_cast_fp16 = slice_by_index(begin = var_7327_begin_0, end = var_7327_end_0, end_mask = var_7327_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7327_cast_fp16")]; + tensor var_7331_begin_0 = const()[name = tensor("op_7331_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_7331_end_0 = const()[name = tensor("op_7331_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_7331_end_mask_0 = const()[name = tensor("op_7331_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7331_cast_fp16 = slice_by_index(begin = var_7331_begin_0, end = var_7331_end_0, end_mask = var_7331_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7331_cast_fp16")]; + tensor var_7335_begin_0 = const()[name = tensor("op_7335_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_7335_end_0 = const()[name = tensor("op_7335_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_7335_end_mask_0 = const()[name = tensor("op_7335_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7335_cast_fp16 = slice_by_index(begin = var_7335_begin_0, end = var_7335_end_0, end_mask = var_7335_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7335_cast_fp16")]; + tensor var_7339_begin_0 = const()[name = tensor("op_7339_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_7339_end_0 = const()[name = tensor("op_7339_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_7339_end_mask_0 = const()[name = tensor("op_7339_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7339_cast_fp16 = slice_by_index(begin = var_7339_begin_0, end = var_7339_end_0, end_mask = var_7339_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7339_cast_fp16")]; + tensor var_7343_begin_0 = const()[name = tensor("op_7343_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_7343_end_0 = const()[name = tensor("op_7343_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_7343_end_mask_0 = const()[name = tensor("op_7343_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7343_cast_fp16 = slice_by_index(begin = var_7343_begin_0, end = var_7343_end_0, end_mask = var_7343_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7343_cast_fp16")]; + tensor var_7347_begin_0 = const()[name = tensor("op_7347_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_7347_end_0 = const()[name = tensor("op_7347_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_7347_end_mask_0 = const()[name = tensor("op_7347_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7347_cast_fp16 = slice_by_index(begin = var_7347_begin_0, end = var_7347_end_0, end_mask = var_7347_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7347_cast_fp16")]; + tensor var_7351_begin_0 = const()[name = tensor("op_7351_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_7351_end_0 = const()[name = tensor("op_7351_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_7351_end_mask_0 = const()[name = tensor("op_7351_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7351_cast_fp16 = slice_by_index(begin = var_7351_begin_0, end = var_7351_end_0, end_mask = var_7351_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7351_cast_fp16")]; + tensor var_7355_begin_0 = const()[name = tensor("op_7355_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_7355_end_0 = const()[name = tensor("op_7355_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_7355_end_mask_0 = const()[name = tensor("op_7355_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7355_cast_fp16 = slice_by_index(begin = var_7355_begin_0, end = var_7355_end_0, end_mask = var_7355_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7355_cast_fp16")]; + tensor var_7359_begin_0 = const()[name = tensor("op_7359_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_7359_end_0 = const()[name = tensor("op_7359_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_7359_end_mask_0 = const()[name = tensor("op_7359_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7359_cast_fp16 = slice_by_index(begin = var_7359_begin_0, end = var_7359_end_0, end_mask = var_7359_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7359_cast_fp16")]; + tensor var_7363_begin_0 = const()[name = tensor("op_7363_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_7363_end_0 = const()[name = tensor("op_7363_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_7363_end_mask_0 = const()[name = tensor("op_7363_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7363_cast_fp16 = slice_by_index(begin = var_7363_begin_0, end = var_7363_end_0, end_mask = var_7363_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7363_cast_fp16")]; + tensor var_7367_begin_0 = const()[name = tensor("op_7367_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_7367_end_0 = const()[name = tensor("op_7367_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_7367_end_mask_0 = const()[name = tensor("op_7367_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7367_cast_fp16 = slice_by_index(begin = var_7367_begin_0, end = var_7367_end_0, end_mask = var_7367_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7367_cast_fp16")]; + tensor var_7371_begin_0 = const()[name = tensor("op_7371_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_7371_end_0 = const()[name = tensor("op_7371_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_7371_end_mask_0 = const()[name = tensor("op_7371_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7371_cast_fp16 = slice_by_index(begin = var_7371_begin_0, end = var_7371_end_0, end_mask = var_7371_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7371_cast_fp16")]; + tensor var_7375_begin_0 = const()[name = tensor("op_7375_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_7375_end_0 = const()[name = tensor("op_7375_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_7375_end_mask_0 = const()[name = tensor("op_7375_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7375_cast_fp16 = slice_by_index(begin = var_7375_begin_0, end = var_7375_end_0, end_mask = var_7375_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7375_cast_fp16")]; + tensor var_7379_begin_0 = const()[name = tensor("op_7379_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_7379_end_0 = const()[name = tensor("op_7379_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_7379_end_mask_0 = const()[name = tensor("op_7379_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7379_cast_fp16 = slice_by_index(begin = var_7379_begin_0, end = var_7379_end_0, end_mask = var_7379_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7379_cast_fp16")]; + tensor var_7383_begin_0 = const()[name = tensor("op_7383_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_7383_end_0 = const()[name = tensor("op_7383_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_7383_end_mask_0 = const()[name = tensor("op_7383_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7383_cast_fp16 = slice_by_index(begin = var_7383_begin_0, end = var_7383_end_0, end_mask = var_7383_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7383_cast_fp16")]; + tensor var_7387_begin_0 = const()[name = tensor("op_7387_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_7387_end_0 = const()[name = tensor("op_7387_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_7387_end_mask_0 = const()[name = tensor("op_7387_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7387_cast_fp16 = slice_by_index(begin = var_7387_begin_0, end = var_7387_end_0, end_mask = var_7387_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7387_cast_fp16")]; + tensor var_7391_begin_0 = const()[name = tensor("op_7391_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_7391_end_0 = const()[name = tensor("op_7391_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_7391_end_mask_0 = const()[name = tensor("op_7391_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7391_cast_fp16 = slice_by_index(begin = var_7391_begin_0, end = var_7391_end_0, end_mask = var_7391_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7391_cast_fp16")]; + tensor var_7395_begin_0 = const()[name = tensor("op_7395_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_7395_end_0 = const()[name = tensor("op_7395_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_7395_end_mask_0 = const()[name = tensor("op_7395_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7395_cast_fp16 = slice_by_index(begin = var_7395_begin_0, end = var_7395_end_0, end_mask = var_7395_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7395_cast_fp16")]; + tensor var_7399_begin_0 = const()[name = tensor("op_7399_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_7399_end_0 = const()[name = tensor("op_7399_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_7399_end_mask_0 = const()[name = tensor("op_7399_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7399_cast_fp16 = slice_by_index(begin = var_7399_begin_0, end = var_7399_end_0, end_mask = var_7399_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7399_cast_fp16")]; + tensor k_75_perm_0 = const()[name = tensor("k_75_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_7406_begin_0 = const()[name = tensor("op_7406_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7406_end_0 = const()[name = tensor("op_7406_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_7406_end_mask_0 = const()[name = tensor("op_7406_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_75_cast_fp16 = transpose(perm = k_75_perm_0, x = k_73_cast_fp16)[name = tensor("transpose_121")]; + tensor var_7406_cast_fp16 = slice_by_index(begin = var_7406_begin_0, end = var_7406_end_0, end_mask = var_7406_end_mask_0, x = k_75_cast_fp16)[name = tensor("op_7406_cast_fp16")]; + tensor var_7410_begin_0 = const()[name = tensor("op_7410_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_7410_end_0 = const()[name = tensor("op_7410_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_7410_end_mask_0 = const()[name = tensor("op_7410_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7410_cast_fp16 = slice_by_index(begin = var_7410_begin_0, end = var_7410_end_0, end_mask = var_7410_end_mask_0, x = k_75_cast_fp16)[name = tensor("op_7410_cast_fp16")]; + tensor var_7414_begin_0 = const()[name = tensor("op_7414_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_7414_end_0 = const()[name = tensor("op_7414_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_7414_end_mask_0 = const()[name = tensor("op_7414_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7414_cast_fp16 = slice_by_index(begin = var_7414_begin_0, end = var_7414_end_0, end_mask = var_7414_end_mask_0, x = k_75_cast_fp16)[name = tensor("op_7414_cast_fp16")]; + tensor var_7418_begin_0 = const()[name = tensor("op_7418_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_7418_end_0 = const()[name = tensor("op_7418_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_7418_end_mask_0 = const()[name = tensor("op_7418_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7418_cast_fp16 = slice_by_index(begin = var_7418_begin_0, end = var_7418_end_0, end_mask = var_7418_end_mask_0, x = k_75_cast_fp16)[name = tensor("op_7418_cast_fp16")]; + tensor var_7422_begin_0 = const()[name = tensor("op_7422_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_7422_end_0 = const()[name = tensor("op_7422_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_7422_end_mask_0 = const()[name = tensor("op_7422_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7422_cast_fp16 = slice_by_index(begin = var_7422_begin_0, end = var_7422_end_0, end_mask = var_7422_end_mask_0, x = k_75_cast_fp16)[name = tensor("op_7422_cast_fp16")]; + tensor var_7426_begin_0 = const()[name = tensor("op_7426_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_7426_end_0 = const()[name = tensor("op_7426_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_7426_end_mask_0 = const()[name = tensor("op_7426_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7426_cast_fp16 = slice_by_index(begin = var_7426_begin_0, end = var_7426_end_0, end_mask = var_7426_end_mask_0, x = k_75_cast_fp16)[name = tensor("op_7426_cast_fp16")]; + tensor var_7430_begin_0 = const()[name = tensor("op_7430_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_7430_end_0 = const()[name = tensor("op_7430_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_7430_end_mask_0 = const()[name = tensor("op_7430_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7430_cast_fp16 = slice_by_index(begin = var_7430_begin_0, end = var_7430_end_0, end_mask = var_7430_end_mask_0, x = k_75_cast_fp16)[name = tensor("op_7430_cast_fp16")]; + tensor var_7434_begin_0 = const()[name = tensor("op_7434_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_7434_end_0 = const()[name = tensor("op_7434_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_7434_end_mask_0 = const()[name = tensor("op_7434_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7434_cast_fp16 = slice_by_index(begin = var_7434_begin_0, end = var_7434_end_0, end_mask = var_7434_end_mask_0, x = k_75_cast_fp16)[name = tensor("op_7434_cast_fp16")]; + tensor var_7438_begin_0 = const()[name = tensor("op_7438_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_7438_end_0 = const()[name = tensor("op_7438_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_7438_end_mask_0 = const()[name = tensor("op_7438_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7438_cast_fp16 = slice_by_index(begin = var_7438_begin_0, end = var_7438_end_0, end_mask = var_7438_end_mask_0, x = k_75_cast_fp16)[name = tensor("op_7438_cast_fp16")]; + tensor var_7442_begin_0 = const()[name = tensor("op_7442_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_7442_end_0 = const()[name = tensor("op_7442_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_7442_end_mask_0 = const()[name = tensor("op_7442_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7442_cast_fp16 = slice_by_index(begin = var_7442_begin_0, end = var_7442_end_0, end_mask = var_7442_end_mask_0, x = k_75_cast_fp16)[name = tensor("op_7442_cast_fp16")]; + tensor var_7446_begin_0 = const()[name = tensor("op_7446_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_7446_end_0 = const()[name = tensor("op_7446_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_7446_end_mask_0 = const()[name = tensor("op_7446_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7446_cast_fp16 = slice_by_index(begin = var_7446_begin_0, end = var_7446_end_0, end_mask = var_7446_end_mask_0, x = k_75_cast_fp16)[name = tensor("op_7446_cast_fp16")]; + tensor var_7450_begin_0 = const()[name = tensor("op_7450_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_7450_end_0 = const()[name = tensor("op_7450_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_7450_end_mask_0 = const()[name = tensor("op_7450_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7450_cast_fp16 = slice_by_index(begin = var_7450_begin_0, end = var_7450_end_0, end_mask = var_7450_end_mask_0, x = k_75_cast_fp16)[name = tensor("op_7450_cast_fp16")]; + tensor var_7454_begin_0 = const()[name = tensor("op_7454_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_7454_end_0 = const()[name = tensor("op_7454_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_7454_end_mask_0 = const()[name = tensor("op_7454_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7454_cast_fp16 = slice_by_index(begin = var_7454_begin_0, end = var_7454_end_0, end_mask = var_7454_end_mask_0, x = k_75_cast_fp16)[name = tensor("op_7454_cast_fp16")]; + tensor var_7458_begin_0 = const()[name = tensor("op_7458_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_7458_end_0 = const()[name = tensor("op_7458_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_7458_end_mask_0 = const()[name = tensor("op_7458_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7458_cast_fp16 = slice_by_index(begin = var_7458_begin_0, end = var_7458_end_0, end_mask = var_7458_end_mask_0, x = k_75_cast_fp16)[name = tensor("op_7458_cast_fp16")]; + tensor var_7462_begin_0 = const()[name = tensor("op_7462_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_7462_end_0 = const()[name = tensor("op_7462_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_7462_end_mask_0 = const()[name = tensor("op_7462_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7462_cast_fp16 = slice_by_index(begin = var_7462_begin_0, end = var_7462_end_0, end_mask = var_7462_end_mask_0, x = k_75_cast_fp16)[name = tensor("op_7462_cast_fp16")]; + tensor var_7466_begin_0 = const()[name = tensor("op_7466_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_7466_end_0 = const()[name = tensor("op_7466_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_7466_end_mask_0 = const()[name = tensor("op_7466_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7466_cast_fp16 = slice_by_index(begin = var_7466_begin_0, end = var_7466_end_0, end_mask = var_7466_end_mask_0, x = k_75_cast_fp16)[name = tensor("op_7466_cast_fp16")]; + tensor var_7470_begin_0 = const()[name = tensor("op_7470_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_7470_end_0 = const()[name = tensor("op_7470_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_7470_end_mask_0 = const()[name = tensor("op_7470_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7470_cast_fp16 = slice_by_index(begin = var_7470_begin_0, end = var_7470_end_0, end_mask = var_7470_end_mask_0, x = k_75_cast_fp16)[name = tensor("op_7470_cast_fp16")]; + tensor var_7474_begin_0 = const()[name = tensor("op_7474_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_7474_end_0 = const()[name = tensor("op_7474_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_7474_end_mask_0 = const()[name = tensor("op_7474_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7474_cast_fp16 = slice_by_index(begin = var_7474_begin_0, end = var_7474_end_0, end_mask = var_7474_end_mask_0, x = k_75_cast_fp16)[name = tensor("op_7474_cast_fp16")]; + tensor var_7478_begin_0 = const()[name = tensor("op_7478_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_7478_end_0 = const()[name = tensor("op_7478_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_7478_end_mask_0 = const()[name = tensor("op_7478_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7478_cast_fp16 = slice_by_index(begin = var_7478_begin_0, end = var_7478_end_0, end_mask = var_7478_end_mask_0, x = k_75_cast_fp16)[name = tensor("op_7478_cast_fp16")]; + tensor var_7482_begin_0 = const()[name = tensor("op_7482_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_7482_end_0 = const()[name = tensor("op_7482_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_7482_end_mask_0 = const()[name = tensor("op_7482_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7482_cast_fp16 = slice_by_index(begin = var_7482_begin_0, end = var_7482_end_0, end_mask = var_7482_end_mask_0, x = k_75_cast_fp16)[name = tensor("op_7482_cast_fp16")]; + tensor var_7484_begin_0 = const()[name = tensor("op_7484_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7484_end_0 = const()[name = tensor("op_7484_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_7484_end_mask_0 = const()[name = tensor("op_7484_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7484_cast_fp16 = slice_by_index(begin = var_7484_begin_0, end = var_7484_end_0, end_mask = var_7484_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7484_cast_fp16")]; + tensor var_7488_begin_0 = const()[name = tensor("op_7488_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_7488_end_0 = const()[name = tensor("op_7488_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_7488_end_mask_0 = const()[name = tensor("op_7488_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7488_cast_fp16 = slice_by_index(begin = var_7488_begin_0, end = var_7488_end_0, end_mask = var_7488_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7488_cast_fp16")]; + tensor var_7492_begin_0 = const()[name = tensor("op_7492_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_7492_end_0 = const()[name = tensor("op_7492_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_7492_end_mask_0 = const()[name = tensor("op_7492_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7492_cast_fp16 = slice_by_index(begin = var_7492_begin_0, end = var_7492_end_0, end_mask = var_7492_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7492_cast_fp16")]; + tensor var_7496_begin_0 = const()[name = tensor("op_7496_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_7496_end_0 = const()[name = tensor("op_7496_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_7496_end_mask_0 = const()[name = tensor("op_7496_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7496_cast_fp16 = slice_by_index(begin = var_7496_begin_0, end = var_7496_end_0, end_mask = var_7496_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7496_cast_fp16")]; + tensor var_7500_begin_0 = const()[name = tensor("op_7500_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_7500_end_0 = const()[name = tensor("op_7500_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_7500_end_mask_0 = const()[name = tensor("op_7500_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7500_cast_fp16 = slice_by_index(begin = var_7500_begin_0, end = var_7500_end_0, end_mask = var_7500_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7500_cast_fp16")]; + tensor var_7504_begin_0 = const()[name = tensor("op_7504_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_7504_end_0 = const()[name = tensor("op_7504_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_7504_end_mask_0 = const()[name = tensor("op_7504_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7504_cast_fp16 = slice_by_index(begin = var_7504_begin_0, end = var_7504_end_0, end_mask = var_7504_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7504_cast_fp16")]; + tensor var_7508_begin_0 = const()[name = tensor("op_7508_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_7508_end_0 = const()[name = tensor("op_7508_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_7508_end_mask_0 = const()[name = tensor("op_7508_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7508_cast_fp16 = slice_by_index(begin = var_7508_begin_0, end = var_7508_end_0, end_mask = var_7508_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7508_cast_fp16")]; + tensor var_7512_begin_0 = const()[name = tensor("op_7512_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_7512_end_0 = const()[name = tensor("op_7512_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_7512_end_mask_0 = const()[name = tensor("op_7512_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7512_cast_fp16 = slice_by_index(begin = var_7512_begin_0, end = var_7512_end_0, end_mask = var_7512_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7512_cast_fp16")]; + tensor var_7516_begin_0 = const()[name = tensor("op_7516_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_7516_end_0 = const()[name = tensor("op_7516_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_7516_end_mask_0 = const()[name = tensor("op_7516_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7516_cast_fp16 = slice_by_index(begin = var_7516_begin_0, end = var_7516_end_0, end_mask = var_7516_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7516_cast_fp16")]; + tensor var_7520_begin_0 = const()[name = tensor("op_7520_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_7520_end_0 = const()[name = tensor("op_7520_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_7520_end_mask_0 = const()[name = tensor("op_7520_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7520_cast_fp16 = slice_by_index(begin = var_7520_begin_0, end = var_7520_end_0, end_mask = var_7520_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7520_cast_fp16")]; + tensor var_7524_begin_0 = const()[name = tensor("op_7524_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_7524_end_0 = const()[name = tensor("op_7524_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_7524_end_mask_0 = const()[name = tensor("op_7524_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7524_cast_fp16 = slice_by_index(begin = var_7524_begin_0, end = var_7524_end_0, end_mask = var_7524_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7524_cast_fp16")]; + tensor var_7528_begin_0 = const()[name = tensor("op_7528_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_7528_end_0 = const()[name = tensor("op_7528_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_7528_end_mask_0 = const()[name = tensor("op_7528_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7528_cast_fp16 = slice_by_index(begin = var_7528_begin_0, end = var_7528_end_0, end_mask = var_7528_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7528_cast_fp16")]; + tensor var_7532_begin_0 = const()[name = tensor("op_7532_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_7532_end_0 = const()[name = tensor("op_7532_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_7532_end_mask_0 = const()[name = tensor("op_7532_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7532_cast_fp16 = slice_by_index(begin = var_7532_begin_0, end = var_7532_end_0, end_mask = var_7532_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7532_cast_fp16")]; + tensor var_7536_begin_0 = const()[name = tensor("op_7536_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_7536_end_0 = const()[name = tensor("op_7536_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_7536_end_mask_0 = const()[name = tensor("op_7536_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7536_cast_fp16 = slice_by_index(begin = var_7536_begin_0, end = var_7536_end_0, end_mask = var_7536_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7536_cast_fp16")]; + tensor var_7540_begin_0 = const()[name = tensor("op_7540_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_7540_end_0 = const()[name = tensor("op_7540_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_7540_end_mask_0 = const()[name = tensor("op_7540_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7540_cast_fp16 = slice_by_index(begin = var_7540_begin_0, end = var_7540_end_0, end_mask = var_7540_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7540_cast_fp16")]; + tensor var_7544_begin_0 = const()[name = tensor("op_7544_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_7544_end_0 = const()[name = tensor("op_7544_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_7544_end_mask_0 = const()[name = tensor("op_7544_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7544_cast_fp16 = slice_by_index(begin = var_7544_begin_0, end = var_7544_end_0, end_mask = var_7544_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7544_cast_fp16")]; + tensor var_7548_begin_0 = const()[name = tensor("op_7548_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_7548_end_0 = const()[name = tensor("op_7548_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_7548_end_mask_0 = const()[name = tensor("op_7548_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7548_cast_fp16 = slice_by_index(begin = var_7548_begin_0, end = var_7548_end_0, end_mask = var_7548_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7548_cast_fp16")]; + tensor var_7552_begin_0 = const()[name = tensor("op_7552_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_7552_end_0 = const()[name = tensor("op_7552_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_7552_end_mask_0 = const()[name = tensor("op_7552_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7552_cast_fp16 = slice_by_index(begin = var_7552_begin_0, end = var_7552_end_0, end_mask = var_7552_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7552_cast_fp16")]; + tensor var_7556_begin_0 = const()[name = tensor("op_7556_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_7556_end_0 = const()[name = tensor("op_7556_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_7556_end_mask_0 = const()[name = tensor("op_7556_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7556_cast_fp16 = slice_by_index(begin = var_7556_begin_0, end = var_7556_end_0, end_mask = var_7556_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7556_cast_fp16")]; + tensor var_7560_begin_0 = const()[name = tensor("op_7560_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_7560_end_0 = const()[name = tensor("op_7560_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_7560_end_mask_0 = const()[name = tensor("op_7560_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7560_cast_fp16 = slice_by_index(begin = var_7560_begin_0, end = var_7560_end_0, end_mask = var_7560_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7560_cast_fp16")]; + tensor var_7564_equation_0 = const()[name = tensor("op_7564_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7564_cast_fp16 = einsum(equation = var_7564_equation_0, values = (var_7406_cast_fp16, var_7323_cast_fp16))[name = tensor("op_7564_cast_fp16")]; + tensor var_7565_to_fp16 = const()[name = tensor("op_7565_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_561_cast_fp16 = mul(x = var_7564_cast_fp16, y = var_7565_to_fp16)[name = tensor("aw_561_cast_fp16")]; + tensor var_7568_equation_0 = const()[name = tensor("op_7568_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7568_cast_fp16 = einsum(equation = var_7568_equation_0, values = (var_7410_cast_fp16, var_7327_cast_fp16))[name = tensor("op_7568_cast_fp16")]; + tensor var_7569_to_fp16 = const()[name = tensor("op_7569_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_563_cast_fp16 = mul(x = var_7568_cast_fp16, y = var_7569_to_fp16)[name = tensor("aw_563_cast_fp16")]; + tensor var_7572_equation_0 = const()[name = tensor("op_7572_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7572_cast_fp16 = einsum(equation = var_7572_equation_0, values = (var_7414_cast_fp16, var_7331_cast_fp16))[name = tensor("op_7572_cast_fp16")]; + tensor var_7573_to_fp16 = const()[name = tensor("op_7573_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_565_cast_fp16 = mul(x = var_7572_cast_fp16, y = var_7573_to_fp16)[name = tensor("aw_565_cast_fp16")]; + tensor var_7576_equation_0 = const()[name = tensor("op_7576_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7576_cast_fp16 = einsum(equation = var_7576_equation_0, values = (var_7418_cast_fp16, var_7335_cast_fp16))[name = tensor("op_7576_cast_fp16")]; + tensor var_7577_to_fp16 = const()[name = tensor("op_7577_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_567_cast_fp16 = mul(x = var_7576_cast_fp16, y = var_7577_to_fp16)[name = tensor("aw_567_cast_fp16")]; + tensor var_7580_equation_0 = const()[name = tensor("op_7580_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7580_cast_fp16 = einsum(equation = var_7580_equation_0, values = (var_7422_cast_fp16, var_7339_cast_fp16))[name = tensor("op_7580_cast_fp16")]; + tensor var_7581_to_fp16 = const()[name = tensor("op_7581_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_569_cast_fp16 = mul(x = var_7580_cast_fp16, y = var_7581_to_fp16)[name = tensor("aw_569_cast_fp16")]; + tensor var_7584_equation_0 = const()[name = tensor("op_7584_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7584_cast_fp16 = einsum(equation = var_7584_equation_0, values = (var_7426_cast_fp16, var_7343_cast_fp16))[name = tensor("op_7584_cast_fp16")]; + tensor var_7585_to_fp16 = const()[name = tensor("op_7585_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_571_cast_fp16 = mul(x = var_7584_cast_fp16, y = var_7585_to_fp16)[name = tensor("aw_571_cast_fp16")]; + tensor var_7588_equation_0 = const()[name = tensor("op_7588_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7588_cast_fp16 = einsum(equation = var_7588_equation_0, values = (var_7430_cast_fp16, var_7347_cast_fp16))[name = tensor("op_7588_cast_fp16")]; + tensor var_7589_to_fp16 = const()[name = tensor("op_7589_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_573_cast_fp16 = mul(x = var_7588_cast_fp16, y = var_7589_to_fp16)[name = tensor("aw_573_cast_fp16")]; + tensor var_7592_equation_0 = const()[name = tensor("op_7592_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7592_cast_fp16 = einsum(equation = var_7592_equation_0, values = (var_7434_cast_fp16, var_7351_cast_fp16))[name = tensor("op_7592_cast_fp16")]; + tensor var_7593_to_fp16 = const()[name = tensor("op_7593_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_575_cast_fp16 = mul(x = var_7592_cast_fp16, y = var_7593_to_fp16)[name = tensor("aw_575_cast_fp16")]; + tensor var_7596_equation_0 = const()[name = tensor("op_7596_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7596_cast_fp16 = einsum(equation = var_7596_equation_0, values = (var_7438_cast_fp16, var_7355_cast_fp16))[name = tensor("op_7596_cast_fp16")]; + tensor var_7597_to_fp16 = const()[name = tensor("op_7597_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_577_cast_fp16 = mul(x = var_7596_cast_fp16, y = var_7597_to_fp16)[name = tensor("aw_577_cast_fp16")]; + tensor var_7600_equation_0 = const()[name = tensor("op_7600_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7600_cast_fp16 = einsum(equation = var_7600_equation_0, values = (var_7442_cast_fp16, var_7359_cast_fp16))[name = tensor("op_7600_cast_fp16")]; + tensor var_7601_to_fp16 = const()[name = tensor("op_7601_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_579_cast_fp16 = mul(x = var_7600_cast_fp16, y = var_7601_to_fp16)[name = tensor("aw_579_cast_fp16")]; + tensor var_7604_equation_0 = const()[name = tensor("op_7604_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7604_cast_fp16 = einsum(equation = var_7604_equation_0, values = (var_7446_cast_fp16, var_7363_cast_fp16))[name = tensor("op_7604_cast_fp16")]; + tensor var_7605_to_fp16 = const()[name = tensor("op_7605_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_581_cast_fp16 = mul(x = var_7604_cast_fp16, y = var_7605_to_fp16)[name = tensor("aw_581_cast_fp16")]; + tensor var_7608_equation_0 = const()[name = tensor("op_7608_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7608_cast_fp16 = einsum(equation = var_7608_equation_0, values = (var_7450_cast_fp16, var_7367_cast_fp16))[name = tensor("op_7608_cast_fp16")]; + tensor var_7609_to_fp16 = const()[name = tensor("op_7609_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_583_cast_fp16 = mul(x = var_7608_cast_fp16, y = var_7609_to_fp16)[name = tensor("aw_583_cast_fp16")]; + tensor var_7612_equation_0 = const()[name = tensor("op_7612_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7612_cast_fp16 = einsum(equation = var_7612_equation_0, values = (var_7454_cast_fp16, var_7371_cast_fp16))[name = tensor("op_7612_cast_fp16")]; + tensor var_7613_to_fp16 = const()[name = tensor("op_7613_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_585_cast_fp16 = mul(x = var_7612_cast_fp16, y = var_7613_to_fp16)[name = tensor("aw_585_cast_fp16")]; + tensor var_7616_equation_0 = const()[name = tensor("op_7616_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7616_cast_fp16 = einsum(equation = var_7616_equation_0, values = (var_7458_cast_fp16, var_7375_cast_fp16))[name = tensor("op_7616_cast_fp16")]; + tensor var_7617_to_fp16 = const()[name = tensor("op_7617_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_587_cast_fp16 = mul(x = var_7616_cast_fp16, y = var_7617_to_fp16)[name = tensor("aw_587_cast_fp16")]; + tensor var_7620_equation_0 = const()[name = tensor("op_7620_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7620_cast_fp16 = einsum(equation = var_7620_equation_0, values = (var_7462_cast_fp16, var_7379_cast_fp16))[name = tensor("op_7620_cast_fp16")]; + tensor var_7621_to_fp16 = const()[name = tensor("op_7621_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_589_cast_fp16 = mul(x = var_7620_cast_fp16, y = var_7621_to_fp16)[name = tensor("aw_589_cast_fp16")]; + tensor var_7624_equation_0 = const()[name = tensor("op_7624_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7624_cast_fp16 = einsum(equation = var_7624_equation_0, values = (var_7466_cast_fp16, var_7383_cast_fp16))[name = tensor("op_7624_cast_fp16")]; + tensor var_7625_to_fp16 = const()[name = tensor("op_7625_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_591_cast_fp16 = mul(x = var_7624_cast_fp16, y = var_7625_to_fp16)[name = tensor("aw_591_cast_fp16")]; + tensor var_7628_equation_0 = const()[name = tensor("op_7628_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7628_cast_fp16 = einsum(equation = var_7628_equation_0, values = (var_7470_cast_fp16, var_7387_cast_fp16))[name = tensor("op_7628_cast_fp16")]; + tensor var_7629_to_fp16 = const()[name = tensor("op_7629_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_593_cast_fp16 = mul(x = var_7628_cast_fp16, y = var_7629_to_fp16)[name = tensor("aw_593_cast_fp16")]; + tensor var_7632_equation_0 = const()[name = tensor("op_7632_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7632_cast_fp16 = einsum(equation = var_7632_equation_0, values = (var_7474_cast_fp16, var_7391_cast_fp16))[name = tensor("op_7632_cast_fp16")]; + tensor var_7633_to_fp16 = const()[name = tensor("op_7633_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_595_cast_fp16 = mul(x = var_7632_cast_fp16, y = var_7633_to_fp16)[name = tensor("aw_595_cast_fp16")]; + tensor var_7636_equation_0 = const()[name = tensor("op_7636_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7636_cast_fp16 = einsum(equation = var_7636_equation_0, values = (var_7478_cast_fp16, var_7395_cast_fp16))[name = tensor("op_7636_cast_fp16")]; + tensor var_7637_to_fp16 = const()[name = tensor("op_7637_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_597_cast_fp16 = mul(x = var_7636_cast_fp16, y = var_7637_to_fp16)[name = tensor("aw_597_cast_fp16")]; + tensor var_7640_equation_0 = const()[name = tensor("op_7640_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7640_cast_fp16 = einsum(equation = var_7640_equation_0, values = (var_7482_cast_fp16, var_7399_cast_fp16))[name = tensor("op_7640_cast_fp16")]; + tensor var_7641_to_fp16 = const()[name = tensor("op_7641_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_599_cast_fp16 = mul(x = var_7640_cast_fp16, y = var_7641_to_fp16)[name = tensor("aw_599_cast_fp16")]; + tensor var_7643_cast_fp16 = softmax(axis = var_2624, x = aw_561_cast_fp16)[name = tensor("op_7643_cast_fp16")]; + tensor var_7644_cast_fp16 = softmax(axis = var_2624, x = aw_563_cast_fp16)[name = tensor("op_7644_cast_fp16")]; + tensor var_7645_cast_fp16 = softmax(axis = var_2624, x = aw_565_cast_fp16)[name = tensor("op_7645_cast_fp16")]; + tensor var_7646_cast_fp16 = softmax(axis = var_2624, x = aw_567_cast_fp16)[name = tensor("op_7646_cast_fp16")]; + tensor var_7647_cast_fp16 = softmax(axis = var_2624, x = aw_569_cast_fp16)[name = tensor("op_7647_cast_fp16")]; + tensor var_7648_cast_fp16 = softmax(axis = var_2624, x = aw_571_cast_fp16)[name = tensor("op_7648_cast_fp16")]; + tensor var_7649_cast_fp16 = softmax(axis = var_2624, x = aw_573_cast_fp16)[name = tensor("op_7649_cast_fp16")]; + tensor var_7650_cast_fp16 = softmax(axis = var_2624, x = aw_575_cast_fp16)[name = tensor("op_7650_cast_fp16")]; + tensor var_7651_cast_fp16 = softmax(axis = var_2624, x = aw_577_cast_fp16)[name = tensor("op_7651_cast_fp16")]; + tensor var_7652_cast_fp16 = softmax(axis = var_2624, x = aw_579_cast_fp16)[name = tensor("op_7652_cast_fp16")]; + tensor var_7653_cast_fp16 = softmax(axis = var_2624, x = aw_581_cast_fp16)[name = tensor("op_7653_cast_fp16")]; + tensor var_7654_cast_fp16 = softmax(axis = var_2624, x = aw_583_cast_fp16)[name = tensor("op_7654_cast_fp16")]; + tensor var_7655_cast_fp16 = softmax(axis = var_2624, x = aw_585_cast_fp16)[name = tensor("op_7655_cast_fp16")]; + tensor var_7656_cast_fp16 = softmax(axis = var_2624, x = aw_587_cast_fp16)[name = tensor("op_7656_cast_fp16")]; + tensor var_7657_cast_fp16 = softmax(axis = var_2624, x = aw_589_cast_fp16)[name = tensor("op_7657_cast_fp16")]; + tensor var_7658_cast_fp16 = softmax(axis = var_2624, x = aw_591_cast_fp16)[name = tensor("op_7658_cast_fp16")]; + tensor var_7659_cast_fp16 = softmax(axis = var_2624, x = aw_593_cast_fp16)[name = tensor("op_7659_cast_fp16")]; + tensor var_7660_cast_fp16 = softmax(axis = var_2624, x = aw_595_cast_fp16)[name = tensor("op_7660_cast_fp16")]; + tensor var_7661_cast_fp16 = softmax(axis = var_2624, x = aw_597_cast_fp16)[name = tensor("op_7661_cast_fp16")]; + tensor var_7662_cast_fp16 = softmax(axis = var_2624, x = aw_599_cast_fp16)[name = tensor("op_7662_cast_fp16")]; + tensor var_7664_equation_0 = const()[name = tensor("op_7664_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7664_cast_fp16 = einsum(equation = var_7664_equation_0, values = (var_7484_cast_fp16, var_7643_cast_fp16))[name = tensor("op_7664_cast_fp16")]; + tensor var_7666_equation_0 = const()[name = tensor("op_7666_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7666_cast_fp16 = einsum(equation = var_7666_equation_0, values = (var_7488_cast_fp16, var_7644_cast_fp16))[name = tensor("op_7666_cast_fp16")]; + tensor var_7668_equation_0 = const()[name = tensor("op_7668_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7668_cast_fp16 = einsum(equation = var_7668_equation_0, values = (var_7492_cast_fp16, var_7645_cast_fp16))[name = tensor("op_7668_cast_fp16")]; + tensor var_7670_equation_0 = const()[name = tensor("op_7670_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7670_cast_fp16 = einsum(equation = var_7670_equation_0, values = (var_7496_cast_fp16, var_7646_cast_fp16))[name = tensor("op_7670_cast_fp16")]; + tensor var_7672_equation_0 = const()[name = tensor("op_7672_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7672_cast_fp16 = einsum(equation = var_7672_equation_0, values = (var_7500_cast_fp16, var_7647_cast_fp16))[name = tensor("op_7672_cast_fp16")]; + tensor var_7674_equation_0 = const()[name = tensor("op_7674_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7674_cast_fp16 = einsum(equation = var_7674_equation_0, values = (var_7504_cast_fp16, var_7648_cast_fp16))[name = tensor("op_7674_cast_fp16")]; + tensor var_7676_equation_0 = const()[name = tensor("op_7676_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7676_cast_fp16 = einsum(equation = var_7676_equation_0, values = (var_7508_cast_fp16, var_7649_cast_fp16))[name = tensor("op_7676_cast_fp16")]; + tensor var_7678_equation_0 = const()[name = tensor("op_7678_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7678_cast_fp16 = einsum(equation = var_7678_equation_0, values = (var_7512_cast_fp16, var_7650_cast_fp16))[name = tensor("op_7678_cast_fp16")]; + tensor var_7680_equation_0 = const()[name = tensor("op_7680_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7680_cast_fp16 = einsum(equation = var_7680_equation_0, values = (var_7516_cast_fp16, var_7651_cast_fp16))[name = tensor("op_7680_cast_fp16")]; + tensor var_7682_equation_0 = const()[name = tensor("op_7682_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7682_cast_fp16 = einsum(equation = var_7682_equation_0, values = (var_7520_cast_fp16, var_7652_cast_fp16))[name = tensor("op_7682_cast_fp16")]; + tensor var_7684_equation_0 = const()[name = tensor("op_7684_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7684_cast_fp16 = einsum(equation = var_7684_equation_0, values = (var_7524_cast_fp16, var_7653_cast_fp16))[name = tensor("op_7684_cast_fp16")]; + tensor var_7686_equation_0 = const()[name = tensor("op_7686_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7686_cast_fp16 = einsum(equation = var_7686_equation_0, values = (var_7528_cast_fp16, var_7654_cast_fp16))[name = tensor("op_7686_cast_fp16")]; + tensor var_7688_equation_0 = const()[name = tensor("op_7688_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7688_cast_fp16 = einsum(equation = var_7688_equation_0, values = (var_7532_cast_fp16, var_7655_cast_fp16))[name = tensor("op_7688_cast_fp16")]; + tensor var_7690_equation_0 = const()[name = tensor("op_7690_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7690_cast_fp16 = einsum(equation = var_7690_equation_0, values = (var_7536_cast_fp16, var_7656_cast_fp16))[name = tensor("op_7690_cast_fp16")]; + tensor var_7692_equation_0 = const()[name = tensor("op_7692_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7692_cast_fp16 = einsum(equation = var_7692_equation_0, values = (var_7540_cast_fp16, var_7657_cast_fp16))[name = tensor("op_7692_cast_fp16")]; + tensor var_7694_equation_0 = const()[name = tensor("op_7694_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7694_cast_fp16 = einsum(equation = var_7694_equation_0, values = (var_7544_cast_fp16, var_7658_cast_fp16))[name = tensor("op_7694_cast_fp16")]; + tensor var_7696_equation_0 = const()[name = tensor("op_7696_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7696_cast_fp16 = einsum(equation = var_7696_equation_0, values = (var_7548_cast_fp16, var_7659_cast_fp16))[name = tensor("op_7696_cast_fp16")]; + tensor var_7698_equation_0 = const()[name = tensor("op_7698_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7698_cast_fp16 = einsum(equation = var_7698_equation_0, values = (var_7552_cast_fp16, var_7660_cast_fp16))[name = tensor("op_7698_cast_fp16")]; + tensor var_7700_equation_0 = const()[name = tensor("op_7700_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7700_cast_fp16 = einsum(equation = var_7700_equation_0, values = (var_7556_cast_fp16, var_7661_cast_fp16))[name = tensor("op_7700_cast_fp16")]; + tensor var_7702_equation_0 = const()[name = tensor("op_7702_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7702_cast_fp16 = einsum(equation = var_7702_equation_0, values = (var_7560_cast_fp16, var_7662_cast_fp16))[name = tensor("op_7702_cast_fp16")]; + tensor input_171_interleave_0 = const()[name = tensor("input_171_interleave_0"), val = tensor(false)]; + tensor input_171_cast_fp16 = concat(axis = var_2624, interleave = input_171_interleave_0, values = (var_7664_cast_fp16, var_7666_cast_fp16, var_7668_cast_fp16, var_7670_cast_fp16, var_7672_cast_fp16, var_7674_cast_fp16, var_7676_cast_fp16, var_7678_cast_fp16, var_7680_cast_fp16, var_7682_cast_fp16, var_7684_cast_fp16, var_7686_cast_fp16, var_7688_cast_fp16, var_7690_cast_fp16, var_7692_cast_fp16, var_7694_cast_fp16, var_7696_cast_fp16, var_7698_cast_fp16, var_7700_cast_fp16, var_7702_cast_fp16))[name = tensor("input_171_cast_fp16")]; + tensor var_7708 = const()[name = tensor("op_7708"), val = tensor([1, 1])]; + tensor var_7710 = const()[name = tensor("op_7710"), val = tensor([1, 1])]; + tensor var_7712_pad_type_0 = const()[name = tensor("op_7712_pad_type_0"), val = tensor("custom")]; + tensor var_7712_pad_0 = const()[name = tensor("op_7712_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_5_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_5_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(555566528)))]; + tensor down_blocks_2_attentions_0_transformer_blocks_5_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_5_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(558843392)))]; + tensor var_7712_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_5_attn1_to_out_0_bias_to_fp16, dilations = var_7710, groups = var_2624, pad = var_7712_pad_0, pad_type = var_7712_pad_type_0, strides = var_7708, weight = down_blocks_2_attentions_0_transformer_blocks_5_attn1_to_out_0_weight_to_fp16, x = input_171_cast_fp16)[name = tensor("op_7712_cast_fp16")]; + tensor inputs_57_cast_fp16 = add(x = var_7712_cast_fp16, y = inputs_55_cast_fp16)[name = tensor("inputs_57_cast_fp16")]; + tensor hidden_states_97_axes_0 = const()[name = tensor("hidden_states_97_axes_0"), val = tensor([1])]; + tensor hidden_states_97_gamma_0_to_fp16 = const()[name = tensor("hidden_states_97_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(558846016)))]; + tensor hidden_states_97_beta_0_to_fp16 = const()[name = tensor("hidden_states_97_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(558848640)))]; + tensor var_7722_to_fp16 = const()[name = tensor("op_7722_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_97_cast_fp16 = layer_norm(axes = hidden_states_97_axes_0, beta = hidden_states_97_beta_0_to_fp16, epsilon = var_7722_to_fp16, gamma = hidden_states_97_gamma_0_to_fp16, x = inputs_57_cast_fp16)[name = tensor("hidden_states_97_cast_fp16")]; + tensor var_7737 = const()[name = tensor("op_7737"), val = tensor([1, 1])]; + tensor var_7739 = const()[name = tensor("op_7739"), val = tensor([1, 1])]; + tensor q_39_pad_type_0 = const()[name = tensor("q_39_pad_type_0"), val = tensor("custom")]; + tensor q_39_pad_0 = const()[name = tensor("q_39_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_5_attn2_to_q_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_5_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(558851264)))]; + tensor q_39_cast_fp16 = conv(dilations = var_7739, groups = var_2624, pad = q_39_pad_0, pad_type = q_39_pad_type_0, strides = var_7737, weight = down_blocks_2_attentions_0_transformer_blocks_5_attn2_to_q_weight_to_fp16, x = hidden_states_97_cast_fp16)[name = tensor("q_39_cast_fp16")]; + tensor var_7743 = const()[name = tensor("op_7743"), val = tensor([1, 1])]; + tensor var_7745 = const()[name = tensor("op_7745"), val = tensor([1, 1])]; + tensor k_77_pad_type_0 = const()[name = tensor("k_77_pad_type_0"), val = tensor("custom")]; + tensor k_77_pad_0 = const()[name = tensor("k_77_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_5_attn2_to_k_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_5_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(562128128)))]; + tensor k_77_cast_fp16 = conv(dilations = var_7745, groups = var_2624, pad = k_77_pad_0, pad_type = k_77_pad_type_0, strides = var_7743, weight = down_blocks_2_attentions_0_transformer_blocks_5_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_77_cast_fp16")]; + tensor var_7749 = const()[name = tensor("op_7749"), val = tensor([1, 1])]; + tensor var_7751 = const()[name = tensor("op_7751"), val = tensor([1, 1])]; + tensor v_39_pad_type_0 = const()[name = tensor("v_39_pad_type_0"), val = tensor("custom")]; + tensor v_39_pad_0 = const()[name = tensor("v_39_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_5_attn2_to_v_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_5_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(567371072)))]; + tensor v_39_cast_fp16 = conv(dilations = var_7751, groups = var_2624, pad = v_39_pad_0, pad_type = v_39_pad_type_0, strides = var_7749, weight = down_blocks_2_attentions_0_transformer_blocks_5_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_39_cast_fp16")]; + tensor var_7755_begin_0 = const()[name = tensor("op_7755_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7755_end_0 = const()[name = tensor("op_7755_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_7755_end_mask_0 = const()[name = tensor("op_7755_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7755_cast_fp16 = slice_by_index(begin = var_7755_begin_0, end = var_7755_end_0, end_mask = var_7755_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_7755_cast_fp16")]; + tensor var_7759_begin_0 = const()[name = tensor("op_7759_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_7759_end_0 = const()[name = tensor("op_7759_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_7759_end_mask_0 = const()[name = tensor("op_7759_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7759_cast_fp16 = slice_by_index(begin = var_7759_begin_0, end = var_7759_end_0, end_mask = var_7759_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_7759_cast_fp16")]; + tensor var_7763_begin_0 = const()[name = tensor("op_7763_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_7763_end_0 = const()[name = tensor("op_7763_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_7763_end_mask_0 = const()[name = tensor("op_7763_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7763_cast_fp16 = slice_by_index(begin = var_7763_begin_0, end = var_7763_end_0, end_mask = var_7763_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_7763_cast_fp16")]; + tensor var_7767_begin_0 = const()[name = tensor("op_7767_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_7767_end_0 = const()[name = tensor("op_7767_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_7767_end_mask_0 = const()[name = tensor("op_7767_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7767_cast_fp16 = slice_by_index(begin = var_7767_begin_0, end = var_7767_end_0, end_mask = var_7767_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_7767_cast_fp16")]; + tensor var_7771_begin_0 = const()[name = tensor("op_7771_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_7771_end_0 = const()[name = tensor("op_7771_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_7771_end_mask_0 = const()[name = tensor("op_7771_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7771_cast_fp16 = slice_by_index(begin = var_7771_begin_0, end = var_7771_end_0, end_mask = var_7771_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_7771_cast_fp16")]; + tensor var_7775_begin_0 = const()[name = tensor("op_7775_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_7775_end_0 = const()[name = tensor("op_7775_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_7775_end_mask_0 = const()[name = tensor("op_7775_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7775_cast_fp16 = slice_by_index(begin = var_7775_begin_0, end = var_7775_end_0, end_mask = var_7775_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_7775_cast_fp16")]; + tensor var_7779_begin_0 = const()[name = tensor("op_7779_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_7779_end_0 = const()[name = tensor("op_7779_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_7779_end_mask_0 = const()[name = tensor("op_7779_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7779_cast_fp16 = slice_by_index(begin = var_7779_begin_0, end = var_7779_end_0, end_mask = var_7779_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_7779_cast_fp16")]; + tensor var_7783_begin_0 = const()[name = tensor("op_7783_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_7783_end_0 = const()[name = tensor("op_7783_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_7783_end_mask_0 = const()[name = tensor("op_7783_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7783_cast_fp16 = slice_by_index(begin = var_7783_begin_0, end = var_7783_end_0, end_mask = var_7783_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_7783_cast_fp16")]; + tensor var_7787_begin_0 = const()[name = tensor("op_7787_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_7787_end_0 = const()[name = tensor("op_7787_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_7787_end_mask_0 = const()[name = tensor("op_7787_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7787_cast_fp16 = slice_by_index(begin = var_7787_begin_0, end = var_7787_end_0, end_mask = var_7787_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_7787_cast_fp16")]; + tensor var_7791_begin_0 = const()[name = tensor("op_7791_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_7791_end_0 = const()[name = tensor("op_7791_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_7791_end_mask_0 = const()[name = tensor("op_7791_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7791_cast_fp16 = slice_by_index(begin = var_7791_begin_0, end = var_7791_end_0, end_mask = var_7791_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_7791_cast_fp16")]; + tensor var_7795_begin_0 = const()[name = tensor("op_7795_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_7795_end_0 = const()[name = tensor("op_7795_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_7795_end_mask_0 = const()[name = tensor("op_7795_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7795_cast_fp16 = slice_by_index(begin = var_7795_begin_0, end = var_7795_end_0, end_mask = var_7795_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_7795_cast_fp16")]; + tensor var_7799_begin_0 = const()[name = tensor("op_7799_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_7799_end_0 = const()[name = tensor("op_7799_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_7799_end_mask_0 = const()[name = tensor("op_7799_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7799_cast_fp16 = slice_by_index(begin = var_7799_begin_0, end = var_7799_end_0, end_mask = var_7799_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_7799_cast_fp16")]; + tensor var_7803_begin_0 = const()[name = tensor("op_7803_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_7803_end_0 = const()[name = tensor("op_7803_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_7803_end_mask_0 = const()[name = tensor("op_7803_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7803_cast_fp16 = slice_by_index(begin = var_7803_begin_0, end = var_7803_end_0, end_mask = var_7803_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_7803_cast_fp16")]; + tensor var_7807_begin_0 = const()[name = tensor("op_7807_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_7807_end_0 = const()[name = tensor("op_7807_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_7807_end_mask_0 = const()[name = tensor("op_7807_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7807_cast_fp16 = slice_by_index(begin = var_7807_begin_0, end = var_7807_end_0, end_mask = var_7807_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_7807_cast_fp16")]; + tensor var_7811_begin_0 = const()[name = tensor("op_7811_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_7811_end_0 = const()[name = tensor("op_7811_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_7811_end_mask_0 = const()[name = tensor("op_7811_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7811_cast_fp16 = slice_by_index(begin = var_7811_begin_0, end = var_7811_end_0, end_mask = var_7811_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_7811_cast_fp16")]; + tensor var_7815_begin_0 = const()[name = tensor("op_7815_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_7815_end_0 = const()[name = tensor("op_7815_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_7815_end_mask_0 = const()[name = tensor("op_7815_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7815_cast_fp16 = slice_by_index(begin = var_7815_begin_0, end = var_7815_end_0, end_mask = var_7815_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_7815_cast_fp16")]; + tensor var_7819_begin_0 = const()[name = tensor("op_7819_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_7819_end_0 = const()[name = tensor("op_7819_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_7819_end_mask_0 = const()[name = tensor("op_7819_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7819_cast_fp16 = slice_by_index(begin = var_7819_begin_0, end = var_7819_end_0, end_mask = var_7819_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_7819_cast_fp16")]; + tensor var_7823_begin_0 = const()[name = tensor("op_7823_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_7823_end_0 = const()[name = tensor("op_7823_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_7823_end_mask_0 = const()[name = tensor("op_7823_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7823_cast_fp16 = slice_by_index(begin = var_7823_begin_0, end = var_7823_end_0, end_mask = var_7823_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_7823_cast_fp16")]; + tensor var_7827_begin_0 = const()[name = tensor("op_7827_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_7827_end_0 = const()[name = tensor("op_7827_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_7827_end_mask_0 = const()[name = tensor("op_7827_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7827_cast_fp16 = slice_by_index(begin = var_7827_begin_0, end = var_7827_end_0, end_mask = var_7827_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_7827_cast_fp16")]; + tensor var_7831_begin_0 = const()[name = tensor("op_7831_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_7831_end_0 = const()[name = tensor("op_7831_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_7831_end_mask_0 = const()[name = tensor("op_7831_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7831_cast_fp16 = slice_by_index(begin = var_7831_begin_0, end = var_7831_end_0, end_mask = var_7831_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_7831_cast_fp16")]; + tensor k_79_perm_0 = const()[name = tensor("k_79_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_7838_begin_0 = const()[name = tensor("op_7838_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7838_end_0 = const()[name = tensor("op_7838_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_7838_end_mask_0 = const()[name = tensor("op_7838_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_79_cast_fp16 = transpose(perm = k_79_perm_0, x = k_77_cast_fp16)[name = tensor("transpose_120")]; + tensor var_7838_cast_fp16 = slice_by_index(begin = var_7838_begin_0, end = var_7838_end_0, end_mask = var_7838_end_mask_0, x = k_79_cast_fp16)[name = tensor("op_7838_cast_fp16")]; + tensor var_7842_begin_0 = const()[name = tensor("op_7842_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_7842_end_0 = const()[name = tensor("op_7842_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_7842_end_mask_0 = const()[name = tensor("op_7842_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7842_cast_fp16 = slice_by_index(begin = var_7842_begin_0, end = var_7842_end_0, end_mask = var_7842_end_mask_0, x = k_79_cast_fp16)[name = tensor("op_7842_cast_fp16")]; + tensor var_7846_begin_0 = const()[name = tensor("op_7846_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_7846_end_0 = const()[name = tensor("op_7846_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_7846_end_mask_0 = const()[name = tensor("op_7846_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7846_cast_fp16 = slice_by_index(begin = var_7846_begin_0, end = var_7846_end_0, end_mask = var_7846_end_mask_0, x = k_79_cast_fp16)[name = tensor("op_7846_cast_fp16")]; + tensor var_7850_begin_0 = const()[name = tensor("op_7850_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_7850_end_0 = const()[name = tensor("op_7850_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_7850_end_mask_0 = const()[name = tensor("op_7850_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7850_cast_fp16 = slice_by_index(begin = var_7850_begin_0, end = var_7850_end_0, end_mask = var_7850_end_mask_0, x = k_79_cast_fp16)[name = tensor("op_7850_cast_fp16")]; + tensor var_7854_begin_0 = const()[name = tensor("op_7854_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_7854_end_0 = const()[name = tensor("op_7854_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_7854_end_mask_0 = const()[name = tensor("op_7854_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7854_cast_fp16 = slice_by_index(begin = var_7854_begin_0, end = var_7854_end_0, end_mask = var_7854_end_mask_0, x = k_79_cast_fp16)[name = tensor("op_7854_cast_fp16")]; + tensor var_7858_begin_0 = const()[name = tensor("op_7858_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_7858_end_0 = const()[name = tensor("op_7858_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_7858_end_mask_0 = const()[name = tensor("op_7858_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7858_cast_fp16 = slice_by_index(begin = var_7858_begin_0, end = var_7858_end_0, end_mask = var_7858_end_mask_0, x = k_79_cast_fp16)[name = tensor("op_7858_cast_fp16")]; + tensor var_7862_begin_0 = const()[name = tensor("op_7862_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_7862_end_0 = const()[name = tensor("op_7862_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_7862_end_mask_0 = const()[name = tensor("op_7862_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7862_cast_fp16 = slice_by_index(begin = var_7862_begin_0, end = var_7862_end_0, end_mask = var_7862_end_mask_0, x = k_79_cast_fp16)[name = tensor("op_7862_cast_fp16")]; + tensor var_7866_begin_0 = const()[name = tensor("op_7866_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_7866_end_0 = const()[name = tensor("op_7866_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_7866_end_mask_0 = const()[name = tensor("op_7866_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7866_cast_fp16 = slice_by_index(begin = var_7866_begin_0, end = var_7866_end_0, end_mask = var_7866_end_mask_0, x = k_79_cast_fp16)[name = tensor("op_7866_cast_fp16")]; + tensor var_7870_begin_0 = const()[name = tensor("op_7870_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_7870_end_0 = const()[name = tensor("op_7870_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_7870_end_mask_0 = const()[name = tensor("op_7870_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7870_cast_fp16 = slice_by_index(begin = var_7870_begin_0, end = var_7870_end_0, end_mask = var_7870_end_mask_0, x = k_79_cast_fp16)[name = tensor("op_7870_cast_fp16")]; + tensor var_7874_begin_0 = const()[name = tensor("op_7874_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_7874_end_0 = const()[name = tensor("op_7874_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_7874_end_mask_0 = const()[name = tensor("op_7874_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7874_cast_fp16 = slice_by_index(begin = var_7874_begin_0, end = var_7874_end_0, end_mask = var_7874_end_mask_0, x = k_79_cast_fp16)[name = tensor("op_7874_cast_fp16")]; + tensor var_7878_begin_0 = const()[name = tensor("op_7878_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_7878_end_0 = const()[name = tensor("op_7878_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_7878_end_mask_0 = const()[name = tensor("op_7878_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7878_cast_fp16 = slice_by_index(begin = var_7878_begin_0, end = var_7878_end_0, end_mask = var_7878_end_mask_0, x = k_79_cast_fp16)[name = tensor("op_7878_cast_fp16")]; + tensor var_7882_begin_0 = const()[name = tensor("op_7882_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_7882_end_0 = const()[name = tensor("op_7882_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_7882_end_mask_0 = const()[name = tensor("op_7882_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7882_cast_fp16 = slice_by_index(begin = var_7882_begin_0, end = var_7882_end_0, end_mask = var_7882_end_mask_0, x = k_79_cast_fp16)[name = tensor("op_7882_cast_fp16")]; + tensor var_7886_begin_0 = const()[name = tensor("op_7886_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_7886_end_0 = const()[name = tensor("op_7886_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_7886_end_mask_0 = const()[name = tensor("op_7886_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7886_cast_fp16 = slice_by_index(begin = var_7886_begin_0, end = var_7886_end_0, end_mask = var_7886_end_mask_0, x = k_79_cast_fp16)[name = tensor("op_7886_cast_fp16")]; + tensor var_7890_begin_0 = const()[name = tensor("op_7890_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_7890_end_0 = const()[name = tensor("op_7890_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_7890_end_mask_0 = const()[name = tensor("op_7890_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7890_cast_fp16 = slice_by_index(begin = var_7890_begin_0, end = var_7890_end_0, end_mask = var_7890_end_mask_0, x = k_79_cast_fp16)[name = tensor("op_7890_cast_fp16")]; + tensor var_7894_begin_0 = const()[name = tensor("op_7894_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_7894_end_0 = const()[name = tensor("op_7894_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_7894_end_mask_0 = const()[name = tensor("op_7894_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7894_cast_fp16 = slice_by_index(begin = var_7894_begin_0, end = var_7894_end_0, end_mask = var_7894_end_mask_0, x = k_79_cast_fp16)[name = tensor("op_7894_cast_fp16")]; + tensor var_7898_begin_0 = const()[name = tensor("op_7898_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_7898_end_0 = const()[name = tensor("op_7898_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_7898_end_mask_0 = const()[name = tensor("op_7898_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7898_cast_fp16 = slice_by_index(begin = var_7898_begin_0, end = var_7898_end_0, end_mask = var_7898_end_mask_0, x = k_79_cast_fp16)[name = tensor("op_7898_cast_fp16")]; + tensor var_7902_begin_0 = const()[name = tensor("op_7902_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_7902_end_0 = const()[name = tensor("op_7902_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_7902_end_mask_0 = const()[name = tensor("op_7902_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7902_cast_fp16 = slice_by_index(begin = var_7902_begin_0, end = var_7902_end_0, end_mask = var_7902_end_mask_0, x = k_79_cast_fp16)[name = tensor("op_7902_cast_fp16")]; + tensor var_7906_begin_0 = const()[name = tensor("op_7906_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_7906_end_0 = const()[name = tensor("op_7906_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_7906_end_mask_0 = const()[name = tensor("op_7906_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7906_cast_fp16 = slice_by_index(begin = var_7906_begin_0, end = var_7906_end_0, end_mask = var_7906_end_mask_0, x = k_79_cast_fp16)[name = tensor("op_7906_cast_fp16")]; + tensor var_7910_begin_0 = const()[name = tensor("op_7910_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_7910_end_0 = const()[name = tensor("op_7910_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_7910_end_mask_0 = const()[name = tensor("op_7910_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7910_cast_fp16 = slice_by_index(begin = var_7910_begin_0, end = var_7910_end_0, end_mask = var_7910_end_mask_0, x = k_79_cast_fp16)[name = tensor("op_7910_cast_fp16")]; + tensor var_7914_begin_0 = const()[name = tensor("op_7914_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_7914_end_0 = const()[name = tensor("op_7914_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_7914_end_mask_0 = const()[name = tensor("op_7914_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7914_cast_fp16 = slice_by_index(begin = var_7914_begin_0, end = var_7914_end_0, end_mask = var_7914_end_mask_0, x = k_79_cast_fp16)[name = tensor("op_7914_cast_fp16")]; + tensor var_7916_begin_0 = const()[name = tensor("op_7916_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7916_end_0 = const()[name = tensor("op_7916_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_7916_end_mask_0 = const()[name = tensor("op_7916_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7916_cast_fp16 = slice_by_index(begin = var_7916_begin_0, end = var_7916_end_0, end_mask = var_7916_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_7916_cast_fp16")]; + tensor var_7920_begin_0 = const()[name = tensor("op_7920_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_7920_end_0 = const()[name = tensor("op_7920_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_7920_end_mask_0 = const()[name = tensor("op_7920_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7920_cast_fp16 = slice_by_index(begin = var_7920_begin_0, end = var_7920_end_0, end_mask = var_7920_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_7920_cast_fp16")]; + tensor var_7924_begin_0 = const()[name = tensor("op_7924_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_7924_end_0 = const()[name = tensor("op_7924_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_7924_end_mask_0 = const()[name = tensor("op_7924_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7924_cast_fp16 = slice_by_index(begin = var_7924_begin_0, end = var_7924_end_0, end_mask = var_7924_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_7924_cast_fp16")]; + tensor var_7928_begin_0 = const()[name = tensor("op_7928_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_7928_end_0 = const()[name = tensor("op_7928_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_7928_end_mask_0 = const()[name = tensor("op_7928_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7928_cast_fp16 = slice_by_index(begin = var_7928_begin_0, end = var_7928_end_0, end_mask = var_7928_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_7928_cast_fp16")]; + tensor var_7932_begin_0 = const()[name = tensor("op_7932_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_7932_end_0 = const()[name = tensor("op_7932_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_7932_end_mask_0 = const()[name = tensor("op_7932_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7932_cast_fp16 = slice_by_index(begin = var_7932_begin_0, end = var_7932_end_0, end_mask = var_7932_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_7932_cast_fp16")]; + tensor var_7936_begin_0 = const()[name = tensor("op_7936_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_7936_end_0 = const()[name = tensor("op_7936_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_7936_end_mask_0 = const()[name = tensor("op_7936_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7936_cast_fp16 = slice_by_index(begin = var_7936_begin_0, end = var_7936_end_0, end_mask = var_7936_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_7936_cast_fp16")]; + tensor var_7940_begin_0 = const()[name = tensor("op_7940_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_7940_end_0 = const()[name = tensor("op_7940_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_7940_end_mask_0 = const()[name = tensor("op_7940_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7940_cast_fp16 = slice_by_index(begin = var_7940_begin_0, end = var_7940_end_0, end_mask = var_7940_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_7940_cast_fp16")]; + tensor var_7944_begin_0 = const()[name = tensor("op_7944_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_7944_end_0 = const()[name = tensor("op_7944_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_7944_end_mask_0 = const()[name = tensor("op_7944_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7944_cast_fp16 = slice_by_index(begin = var_7944_begin_0, end = var_7944_end_0, end_mask = var_7944_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_7944_cast_fp16")]; + tensor var_7948_begin_0 = const()[name = tensor("op_7948_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_7948_end_0 = const()[name = tensor("op_7948_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_7948_end_mask_0 = const()[name = tensor("op_7948_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7948_cast_fp16 = slice_by_index(begin = var_7948_begin_0, end = var_7948_end_0, end_mask = var_7948_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_7948_cast_fp16")]; + tensor var_7952_begin_0 = const()[name = tensor("op_7952_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_7952_end_0 = const()[name = tensor("op_7952_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_7952_end_mask_0 = const()[name = tensor("op_7952_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7952_cast_fp16 = slice_by_index(begin = var_7952_begin_0, end = var_7952_end_0, end_mask = var_7952_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_7952_cast_fp16")]; + tensor var_7956_begin_0 = const()[name = tensor("op_7956_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_7956_end_0 = const()[name = tensor("op_7956_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_7956_end_mask_0 = const()[name = tensor("op_7956_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7956_cast_fp16 = slice_by_index(begin = var_7956_begin_0, end = var_7956_end_0, end_mask = var_7956_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_7956_cast_fp16")]; + tensor var_7960_begin_0 = const()[name = tensor("op_7960_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_7960_end_0 = const()[name = tensor("op_7960_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_7960_end_mask_0 = const()[name = tensor("op_7960_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7960_cast_fp16 = slice_by_index(begin = var_7960_begin_0, end = var_7960_end_0, end_mask = var_7960_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_7960_cast_fp16")]; + tensor var_7964_begin_0 = const()[name = tensor("op_7964_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_7964_end_0 = const()[name = tensor("op_7964_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_7964_end_mask_0 = const()[name = tensor("op_7964_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7964_cast_fp16 = slice_by_index(begin = var_7964_begin_0, end = var_7964_end_0, end_mask = var_7964_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_7964_cast_fp16")]; + tensor var_7968_begin_0 = const()[name = tensor("op_7968_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_7968_end_0 = const()[name = tensor("op_7968_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_7968_end_mask_0 = const()[name = tensor("op_7968_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7968_cast_fp16 = slice_by_index(begin = var_7968_begin_0, end = var_7968_end_0, end_mask = var_7968_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_7968_cast_fp16")]; + tensor var_7972_begin_0 = const()[name = tensor("op_7972_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_7972_end_0 = const()[name = tensor("op_7972_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_7972_end_mask_0 = const()[name = tensor("op_7972_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7972_cast_fp16 = slice_by_index(begin = var_7972_begin_0, end = var_7972_end_0, end_mask = var_7972_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_7972_cast_fp16")]; + tensor var_7976_begin_0 = const()[name = tensor("op_7976_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_7976_end_0 = const()[name = tensor("op_7976_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_7976_end_mask_0 = const()[name = tensor("op_7976_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7976_cast_fp16 = slice_by_index(begin = var_7976_begin_0, end = var_7976_end_0, end_mask = var_7976_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_7976_cast_fp16")]; + tensor var_7980_begin_0 = const()[name = tensor("op_7980_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_7980_end_0 = const()[name = tensor("op_7980_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_7980_end_mask_0 = const()[name = tensor("op_7980_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7980_cast_fp16 = slice_by_index(begin = var_7980_begin_0, end = var_7980_end_0, end_mask = var_7980_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_7980_cast_fp16")]; + tensor var_7984_begin_0 = const()[name = tensor("op_7984_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_7984_end_0 = const()[name = tensor("op_7984_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_7984_end_mask_0 = const()[name = tensor("op_7984_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7984_cast_fp16 = slice_by_index(begin = var_7984_begin_0, end = var_7984_end_0, end_mask = var_7984_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_7984_cast_fp16")]; + tensor var_7988_begin_0 = const()[name = tensor("op_7988_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_7988_end_0 = const()[name = tensor("op_7988_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_7988_end_mask_0 = const()[name = tensor("op_7988_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7988_cast_fp16 = slice_by_index(begin = var_7988_begin_0, end = var_7988_end_0, end_mask = var_7988_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_7988_cast_fp16")]; + tensor var_7992_begin_0 = const()[name = tensor("op_7992_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_7992_end_0 = const()[name = tensor("op_7992_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_7992_end_mask_0 = const()[name = tensor("op_7992_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7992_cast_fp16 = slice_by_index(begin = var_7992_begin_0, end = var_7992_end_0, end_mask = var_7992_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_7992_cast_fp16")]; + tensor var_7996_equation_0 = const()[name = tensor("op_7996_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7996_cast_fp16 = einsum(equation = var_7996_equation_0, values = (var_7838_cast_fp16, var_7755_cast_fp16))[name = tensor("op_7996_cast_fp16")]; + tensor var_7997_to_fp16 = const()[name = tensor("op_7997_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_601_cast_fp16 = mul(x = var_7996_cast_fp16, y = var_7997_to_fp16)[name = tensor("aw_601_cast_fp16")]; + tensor var_8000_equation_0 = const()[name = tensor("op_8000_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8000_cast_fp16 = einsum(equation = var_8000_equation_0, values = (var_7842_cast_fp16, var_7759_cast_fp16))[name = tensor("op_8000_cast_fp16")]; + tensor var_8001_to_fp16 = const()[name = tensor("op_8001_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_603_cast_fp16 = mul(x = var_8000_cast_fp16, y = var_8001_to_fp16)[name = tensor("aw_603_cast_fp16")]; + tensor var_8004_equation_0 = const()[name = tensor("op_8004_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8004_cast_fp16 = einsum(equation = var_8004_equation_0, values = (var_7846_cast_fp16, var_7763_cast_fp16))[name = tensor("op_8004_cast_fp16")]; + tensor var_8005_to_fp16 = const()[name = tensor("op_8005_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_605_cast_fp16 = mul(x = var_8004_cast_fp16, y = var_8005_to_fp16)[name = tensor("aw_605_cast_fp16")]; + tensor var_8008_equation_0 = const()[name = tensor("op_8008_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8008_cast_fp16 = einsum(equation = var_8008_equation_0, values = (var_7850_cast_fp16, var_7767_cast_fp16))[name = tensor("op_8008_cast_fp16")]; + tensor var_8009_to_fp16 = const()[name = tensor("op_8009_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_607_cast_fp16 = mul(x = var_8008_cast_fp16, y = var_8009_to_fp16)[name = tensor("aw_607_cast_fp16")]; + tensor var_8012_equation_0 = const()[name = tensor("op_8012_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8012_cast_fp16 = einsum(equation = var_8012_equation_0, values = (var_7854_cast_fp16, var_7771_cast_fp16))[name = tensor("op_8012_cast_fp16")]; + tensor var_8013_to_fp16 = const()[name = tensor("op_8013_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_609_cast_fp16 = mul(x = var_8012_cast_fp16, y = var_8013_to_fp16)[name = tensor("aw_609_cast_fp16")]; + tensor var_8016_equation_0 = const()[name = tensor("op_8016_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8016_cast_fp16 = einsum(equation = var_8016_equation_0, values = (var_7858_cast_fp16, var_7775_cast_fp16))[name = tensor("op_8016_cast_fp16")]; + tensor var_8017_to_fp16 = const()[name = tensor("op_8017_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_611_cast_fp16 = mul(x = var_8016_cast_fp16, y = var_8017_to_fp16)[name = tensor("aw_611_cast_fp16")]; + tensor var_8020_equation_0 = const()[name = tensor("op_8020_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8020_cast_fp16 = einsum(equation = var_8020_equation_0, values = (var_7862_cast_fp16, var_7779_cast_fp16))[name = tensor("op_8020_cast_fp16")]; + tensor var_8021_to_fp16 = const()[name = tensor("op_8021_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_613_cast_fp16 = mul(x = var_8020_cast_fp16, y = var_8021_to_fp16)[name = tensor("aw_613_cast_fp16")]; + tensor var_8024_equation_0 = const()[name = tensor("op_8024_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8024_cast_fp16 = einsum(equation = var_8024_equation_0, values = (var_7866_cast_fp16, var_7783_cast_fp16))[name = tensor("op_8024_cast_fp16")]; + tensor var_8025_to_fp16 = const()[name = tensor("op_8025_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_615_cast_fp16 = mul(x = var_8024_cast_fp16, y = var_8025_to_fp16)[name = tensor("aw_615_cast_fp16")]; + tensor var_8028_equation_0 = const()[name = tensor("op_8028_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8028_cast_fp16 = einsum(equation = var_8028_equation_0, values = (var_7870_cast_fp16, var_7787_cast_fp16))[name = tensor("op_8028_cast_fp16")]; + tensor var_8029_to_fp16 = const()[name = tensor("op_8029_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_617_cast_fp16 = mul(x = var_8028_cast_fp16, y = var_8029_to_fp16)[name = tensor("aw_617_cast_fp16")]; + tensor var_8032_equation_0 = const()[name = tensor("op_8032_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8032_cast_fp16 = einsum(equation = var_8032_equation_0, values = (var_7874_cast_fp16, var_7791_cast_fp16))[name = tensor("op_8032_cast_fp16")]; + tensor var_8033_to_fp16 = const()[name = tensor("op_8033_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_619_cast_fp16 = mul(x = var_8032_cast_fp16, y = var_8033_to_fp16)[name = tensor("aw_619_cast_fp16")]; + tensor var_8036_equation_0 = const()[name = tensor("op_8036_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8036_cast_fp16 = einsum(equation = var_8036_equation_0, values = (var_7878_cast_fp16, var_7795_cast_fp16))[name = tensor("op_8036_cast_fp16")]; + tensor var_8037_to_fp16 = const()[name = tensor("op_8037_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_621_cast_fp16 = mul(x = var_8036_cast_fp16, y = var_8037_to_fp16)[name = tensor("aw_621_cast_fp16")]; + tensor var_8040_equation_0 = const()[name = tensor("op_8040_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8040_cast_fp16 = einsum(equation = var_8040_equation_0, values = (var_7882_cast_fp16, var_7799_cast_fp16))[name = tensor("op_8040_cast_fp16")]; + tensor var_8041_to_fp16 = const()[name = tensor("op_8041_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_623_cast_fp16 = mul(x = var_8040_cast_fp16, y = var_8041_to_fp16)[name = tensor("aw_623_cast_fp16")]; + tensor var_8044_equation_0 = const()[name = tensor("op_8044_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8044_cast_fp16 = einsum(equation = var_8044_equation_0, values = (var_7886_cast_fp16, var_7803_cast_fp16))[name = tensor("op_8044_cast_fp16")]; + tensor var_8045_to_fp16 = const()[name = tensor("op_8045_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_625_cast_fp16 = mul(x = var_8044_cast_fp16, y = var_8045_to_fp16)[name = tensor("aw_625_cast_fp16")]; + tensor var_8048_equation_0 = const()[name = tensor("op_8048_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8048_cast_fp16 = einsum(equation = var_8048_equation_0, values = (var_7890_cast_fp16, var_7807_cast_fp16))[name = tensor("op_8048_cast_fp16")]; + tensor var_8049_to_fp16 = const()[name = tensor("op_8049_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_627_cast_fp16 = mul(x = var_8048_cast_fp16, y = var_8049_to_fp16)[name = tensor("aw_627_cast_fp16")]; + tensor var_8052_equation_0 = const()[name = tensor("op_8052_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8052_cast_fp16 = einsum(equation = var_8052_equation_0, values = (var_7894_cast_fp16, var_7811_cast_fp16))[name = tensor("op_8052_cast_fp16")]; + tensor var_8053_to_fp16 = const()[name = tensor("op_8053_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_629_cast_fp16 = mul(x = var_8052_cast_fp16, y = var_8053_to_fp16)[name = tensor("aw_629_cast_fp16")]; + tensor var_8056_equation_0 = const()[name = tensor("op_8056_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8056_cast_fp16 = einsum(equation = var_8056_equation_0, values = (var_7898_cast_fp16, var_7815_cast_fp16))[name = tensor("op_8056_cast_fp16")]; + tensor var_8057_to_fp16 = const()[name = tensor("op_8057_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_631_cast_fp16 = mul(x = var_8056_cast_fp16, y = var_8057_to_fp16)[name = tensor("aw_631_cast_fp16")]; + tensor var_8060_equation_0 = const()[name = tensor("op_8060_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8060_cast_fp16 = einsum(equation = var_8060_equation_0, values = (var_7902_cast_fp16, var_7819_cast_fp16))[name = tensor("op_8060_cast_fp16")]; + tensor var_8061_to_fp16 = const()[name = tensor("op_8061_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_633_cast_fp16 = mul(x = var_8060_cast_fp16, y = var_8061_to_fp16)[name = tensor("aw_633_cast_fp16")]; + tensor var_8064_equation_0 = const()[name = tensor("op_8064_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8064_cast_fp16 = einsum(equation = var_8064_equation_0, values = (var_7906_cast_fp16, var_7823_cast_fp16))[name = tensor("op_8064_cast_fp16")]; + tensor var_8065_to_fp16 = const()[name = tensor("op_8065_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_635_cast_fp16 = mul(x = var_8064_cast_fp16, y = var_8065_to_fp16)[name = tensor("aw_635_cast_fp16")]; + tensor var_8068_equation_0 = const()[name = tensor("op_8068_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8068_cast_fp16 = einsum(equation = var_8068_equation_0, values = (var_7910_cast_fp16, var_7827_cast_fp16))[name = tensor("op_8068_cast_fp16")]; + tensor var_8069_to_fp16 = const()[name = tensor("op_8069_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_637_cast_fp16 = mul(x = var_8068_cast_fp16, y = var_8069_to_fp16)[name = tensor("aw_637_cast_fp16")]; + tensor var_8072_equation_0 = const()[name = tensor("op_8072_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8072_cast_fp16 = einsum(equation = var_8072_equation_0, values = (var_7914_cast_fp16, var_7831_cast_fp16))[name = tensor("op_8072_cast_fp16")]; + tensor var_8073_to_fp16 = const()[name = tensor("op_8073_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_639_cast_fp16 = mul(x = var_8072_cast_fp16, y = var_8073_to_fp16)[name = tensor("aw_639_cast_fp16")]; + tensor var_8075_cast_fp16 = softmax(axis = var_2624, x = aw_601_cast_fp16)[name = tensor("op_8075_cast_fp16")]; + tensor var_8076_cast_fp16 = softmax(axis = var_2624, x = aw_603_cast_fp16)[name = tensor("op_8076_cast_fp16")]; + tensor var_8077_cast_fp16 = softmax(axis = var_2624, x = aw_605_cast_fp16)[name = tensor("op_8077_cast_fp16")]; + tensor var_8078_cast_fp16 = softmax(axis = var_2624, x = aw_607_cast_fp16)[name = tensor("op_8078_cast_fp16")]; + tensor var_8079_cast_fp16 = softmax(axis = var_2624, x = aw_609_cast_fp16)[name = tensor("op_8079_cast_fp16")]; + tensor var_8080_cast_fp16 = softmax(axis = var_2624, x = aw_611_cast_fp16)[name = tensor("op_8080_cast_fp16")]; + tensor var_8081_cast_fp16 = softmax(axis = var_2624, x = aw_613_cast_fp16)[name = tensor("op_8081_cast_fp16")]; + tensor var_8082_cast_fp16 = softmax(axis = var_2624, x = aw_615_cast_fp16)[name = tensor("op_8082_cast_fp16")]; + tensor var_8083_cast_fp16 = softmax(axis = var_2624, x = aw_617_cast_fp16)[name = tensor("op_8083_cast_fp16")]; + tensor var_8084_cast_fp16 = softmax(axis = var_2624, x = aw_619_cast_fp16)[name = tensor("op_8084_cast_fp16")]; + tensor var_8085_cast_fp16 = softmax(axis = var_2624, x = aw_621_cast_fp16)[name = tensor("op_8085_cast_fp16")]; + tensor var_8086_cast_fp16 = softmax(axis = var_2624, x = aw_623_cast_fp16)[name = tensor("op_8086_cast_fp16")]; + tensor var_8087_cast_fp16 = softmax(axis = var_2624, x = aw_625_cast_fp16)[name = tensor("op_8087_cast_fp16")]; + tensor var_8088_cast_fp16 = softmax(axis = var_2624, x = aw_627_cast_fp16)[name = tensor("op_8088_cast_fp16")]; + tensor var_8089_cast_fp16 = softmax(axis = var_2624, x = aw_629_cast_fp16)[name = tensor("op_8089_cast_fp16")]; + tensor var_8090_cast_fp16 = softmax(axis = var_2624, x = aw_631_cast_fp16)[name = tensor("op_8090_cast_fp16")]; + tensor var_8091_cast_fp16 = softmax(axis = var_2624, x = aw_633_cast_fp16)[name = tensor("op_8091_cast_fp16")]; + tensor var_8092_cast_fp16 = softmax(axis = var_2624, x = aw_635_cast_fp16)[name = tensor("op_8092_cast_fp16")]; + tensor var_8093_cast_fp16 = softmax(axis = var_2624, x = aw_637_cast_fp16)[name = tensor("op_8093_cast_fp16")]; + tensor var_8094_cast_fp16 = softmax(axis = var_2624, x = aw_639_cast_fp16)[name = tensor("op_8094_cast_fp16")]; + tensor var_8096_equation_0 = const()[name = tensor("op_8096_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8096_cast_fp16 = einsum(equation = var_8096_equation_0, values = (var_7916_cast_fp16, var_8075_cast_fp16))[name = tensor("op_8096_cast_fp16")]; + tensor var_8098_equation_0 = const()[name = tensor("op_8098_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8098_cast_fp16 = einsum(equation = var_8098_equation_0, values = (var_7920_cast_fp16, var_8076_cast_fp16))[name = tensor("op_8098_cast_fp16")]; + tensor var_8100_equation_0 = const()[name = tensor("op_8100_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8100_cast_fp16 = einsum(equation = var_8100_equation_0, values = (var_7924_cast_fp16, var_8077_cast_fp16))[name = tensor("op_8100_cast_fp16")]; + tensor var_8102_equation_0 = const()[name = tensor("op_8102_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8102_cast_fp16 = einsum(equation = var_8102_equation_0, values = (var_7928_cast_fp16, var_8078_cast_fp16))[name = tensor("op_8102_cast_fp16")]; + tensor var_8104_equation_0 = const()[name = tensor("op_8104_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8104_cast_fp16 = einsum(equation = var_8104_equation_0, values = (var_7932_cast_fp16, var_8079_cast_fp16))[name = tensor("op_8104_cast_fp16")]; + tensor var_8106_equation_0 = const()[name = tensor("op_8106_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8106_cast_fp16 = einsum(equation = var_8106_equation_0, values = (var_7936_cast_fp16, var_8080_cast_fp16))[name = tensor("op_8106_cast_fp16")]; + tensor var_8108_equation_0 = const()[name = tensor("op_8108_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8108_cast_fp16 = einsum(equation = var_8108_equation_0, values = (var_7940_cast_fp16, var_8081_cast_fp16))[name = tensor("op_8108_cast_fp16")]; + tensor var_8110_equation_0 = const()[name = tensor("op_8110_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8110_cast_fp16 = einsum(equation = var_8110_equation_0, values = (var_7944_cast_fp16, var_8082_cast_fp16))[name = tensor("op_8110_cast_fp16")]; + tensor var_8112_equation_0 = const()[name = tensor("op_8112_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8112_cast_fp16 = einsum(equation = var_8112_equation_0, values = (var_7948_cast_fp16, var_8083_cast_fp16))[name = tensor("op_8112_cast_fp16")]; + tensor var_8114_equation_0 = const()[name = tensor("op_8114_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8114_cast_fp16 = einsum(equation = var_8114_equation_0, values = (var_7952_cast_fp16, var_8084_cast_fp16))[name = tensor("op_8114_cast_fp16")]; + tensor var_8116_equation_0 = const()[name = tensor("op_8116_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8116_cast_fp16 = einsum(equation = var_8116_equation_0, values = (var_7956_cast_fp16, var_8085_cast_fp16))[name = tensor("op_8116_cast_fp16")]; + tensor var_8118_equation_0 = const()[name = tensor("op_8118_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8118_cast_fp16 = einsum(equation = var_8118_equation_0, values = (var_7960_cast_fp16, var_8086_cast_fp16))[name = tensor("op_8118_cast_fp16")]; + tensor var_8120_equation_0 = const()[name = tensor("op_8120_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8120_cast_fp16 = einsum(equation = var_8120_equation_0, values = (var_7964_cast_fp16, var_8087_cast_fp16))[name = tensor("op_8120_cast_fp16")]; + tensor var_8122_equation_0 = const()[name = tensor("op_8122_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8122_cast_fp16 = einsum(equation = var_8122_equation_0, values = (var_7968_cast_fp16, var_8088_cast_fp16))[name = tensor("op_8122_cast_fp16")]; + tensor var_8124_equation_0 = const()[name = tensor("op_8124_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8124_cast_fp16 = einsum(equation = var_8124_equation_0, values = (var_7972_cast_fp16, var_8089_cast_fp16))[name = tensor("op_8124_cast_fp16")]; + tensor var_8126_equation_0 = const()[name = tensor("op_8126_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8126_cast_fp16 = einsum(equation = var_8126_equation_0, values = (var_7976_cast_fp16, var_8090_cast_fp16))[name = tensor("op_8126_cast_fp16")]; + tensor var_8128_equation_0 = const()[name = tensor("op_8128_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8128_cast_fp16 = einsum(equation = var_8128_equation_0, values = (var_7980_cast_fp16, var_8091_cast_fp16))[name = tensor("op_8128_cast_fp16")]; + tensor var_8130_equation_0 = const()[name = tensor("op_8130_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8130_cast_fp16 = einsum(equation = var_8130_equation_0, values = (var_7984_cast_fp16, var_8092_cast_fp16))[name = tensor("op_8130_cast_fp16")]; + tensor var_8132_equation_0 = const()[name = tensor("op_8132_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8132_cast_fp16 = einsum(equation = var_8132_equation_0, values = (var_7988_cast_fp16, var_8093_cast_fp16))[name = tensor("op_8132_cast_fp16")]; + tensor var_8134_equation_0 = const()[name = tensor("op_8134_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8134_cast_fp16 = einsum(equation = var_8134_equation_0, values = (var_7992_cast_fp16, var_8094_cast_fp16))[name = tensor("op_8134_cast_fp16")]; + tensor input_173_interleave_0 = const()[name = tensor("input_173_interleave_0"), val = tensor(false)]; + tensor input_173_cast_fp16 = concat(axis = var_2624, interleave = input_173_interleave_0, values = (var_8096_cast_fp16, var_8098_cast_fp16, var_8100_cast_fp16, var_8102_cast_fp16, var_8104_cast_fp16, var_8106_cast_fp16, var_8108_cast_fp16, var_8110_cast_fp16, var_8112_cast_fp16, var_8114_cast_fp16, var_8116_cast_fp16, var_8118_cast_fp16, var_8120_cast_fp16, var_8122_cast_fp16, var_8124_cast_fp16, var_8126_cast_fp16, var_8128_cast_fp16, var_8130_cast_fp16, var_8132_cast_fp16, var_8134_cast_fp16))[name = tensor("input_173_cast_fp16")]; + tensor var_8140 = const()[name = tensor("op_8140"), val = tensor([1, 1])]; + tensor var_8142 = const()[name = tensor("op_8142"), val = tensor([1, 1])]; + tensor var_8144_pad_type_0 = const()[name = tensor("op_8144_pad_type_0"), val = tensor("custom")]; + tensor var_8144_pad_0 = const()[name = tensor("op_8144_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_5_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_5_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(572614016)))]; + tensor down_blocks_2_attentions_0_transformer_blocks_5_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_5_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(575890880)))]; + tensor var_8144_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_5_attn2_to_out_0_bias_to_fp16, dilations = var_8142, groups = var_2624, pad = var_8144_pad_0, pad_type = var_8144_pad_type_0, strides = var_8140, weight = down_blocks_2_attentions_0_transformer_blocks_5_attn2_to_out_0_weight_to_fp16, x = input_173_cast_fp16)[name = tensor("op_8144_cast_fp16")]; + tensor inputs_59_cast_fp16 = add(x = var_8144_cast_fp16, y = inputs_57_cast_fp16)[name = tensor("inputs_59_cast_fp16")]; + tensor input_175_axes_0 = const()[name = tensor("input_175_axes_0"), val = tensor([1])]; + tensor input_175_gamma_0_to_fp16 = const()[name = tensor("input_175_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(575893504)))]; + tensor input_175_beta_0_to_fp16 = const()[name = tensor("input_175_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(575896128)))]; + tensor var_8154_to_fp16 = const()[name = tensor("op_8154_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_175_cast_fp16 = layer_norm(axes = input_175_axes_0, beta = input_175_beta_0_to_fp16, epsilon = var_8154_to_fp16, gamma = input_175_gamma_0_to_fp16, x = inputs_59_cast_fp16)[name = tensor("input_175_cast_fp16")]; + tensor var_8170 = const()[name = tensor("op_8170"), val = tensor([1, 1])]; + tensor var_8172 = const()[name = tensor("op_8172"), val = tensor([1, 1])]; + tensor var_8174_pad_type_0 = const()[name = tensor("op_8174_pad_type_0"), val = tensor("custom")]; + tensor var_8174_pad_0 = const()[name = tensor("op_8174_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_5_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_5_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(575898752)))]; + tensor down_blocks_2_attentions_0_transformer_blocks_5_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_5_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(602113216)))]; + tensor var_8174_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_5_ff_net_0_proj_bias_to_fp16, dilations = var_8172, groups = var_2624, pad = var_8174_pad_0, pad_type = var_8174_pad_type_0, strides = var_8170, weight = down_blocks_2_attentions_0_transformer_blocks_5_ff_net_0_proj_weight_to_fp16, x = input_175_cast_fp16)[name = tensor("op_8174_cast_fp16")]; + tensor var_8175_split_sizes_0 = const()[name = tensor("op_8175_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_8175_axis_0 = const()[name = tensor("op_8175_axis_0"), val = tensor(1)]; + tensor var_8175_cast_fp16_0, tensor var_8175_cast_fp16_1 = split(axis = var_8175_axis_0, split_sizes = var_8175_split_sizes_0, x = var_8174_cast_fp16)[name = tensor("op_8175_cast_fp16")]; + tensor var_8177_mode_0 = const()[name = tensor("op_8177_mode_0"), val = tensor("EXACT")]; + tensor var_8177_cast_fp16 = gelu(mode = var_8177_mode_0, x = var_8175_cast_fp16_1)[name = tensor("op_8177_cast_fp16")]; + tensor input_177_cast_fp16 = mul(x = var_8175_cast_fp16_0, y = var_8177_cast_fp16)[name = tensor("input_177_cast_fp16")]; + tensor var_8181 = const()[name = tensor("op_8181"), val = tensor([1, 1])]; + tensor var_8183 = const()[name = tensor("op_8183"), val = tensor([1, 1])]; + tensor var_8185_pad_type_0 = const()[name = tensor("op_8185_pad_type_0"), val = tensor("custom")]; + tensor var_8185_pad_0 = const()[name = tensor("op_8185_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_5_ff_net_2_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_5_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(602133760)))]; + tensor down_blocks_2_attentions_0_transformer_blocks_5_ff_net_2_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_5_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(615241024)))]; + tensor var_8185_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_5_ff_net_2_bias_to_fp16, dilations = var_8183, groups = var_2624, pad = var_8185_pad_0, pad_type = var_8185_pad_type_0, strides = var_8181, weight = down_blocks_2_attentions_0_transformer_blocks_5_ff_net_2_weight_to_fp16, x = input_177_cast_fp16)[name = tensor("op_8185_cast_fp16")]; + tensor inputs_61_cast_fp16 = add(x = var_8185_cast_fp16, y = inputs_59_cast_fp16)[name = tensor("inputs_61_cast_fp16")]; + tensor hidden_states_101_axes_0 = const()[name = tensor("hidden_states_101_axes_0"), val = tensor([1])]; + tensor hidden_states_101_gamma_0_to_fp16 = const()[name = tensor("hidden_states_101_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(615243648)))]; + tensor hidden_states_101_beta_0_to_fp16 = const()[name = tensor("hidden_states_101_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(615246272)))]; + tensor var_8201_to_fp16 = const()[name = tensor("op_8201_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_101_cast_fp16 = layer_norm(axes = hidden_states_101_axes_0, beta = hidden_states_101_beta_0_to_fp16, epsilon = var_8201_to_fp16, gamma = hidden_states_101_gamma_0_to_fp16, x = inputs_61_cast_fp16)[name = tensor("hidden_states_101_cast_fp16")]; + tensor var_8216 = const()[name = tensor("op_8216"), val = tensor([1, 1])]; + tensor var_8218 = const()[name = tensor("op_8218"), val = tensor([1, 1])]; + tensor q_41_pad_type_0 = const()[name = tensor("q_41_pad_type_0"), val = tensor("custom")]; + tensor q_41_pad_0 = const()[name = tensor("q_41_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_6_attn1_to_q_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_6_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(615248896)))]; + tensor q_41_cast_fp16 = conv(dilations = var_8218, groups = var_2624, pad = q_41_pad_0, pad_type = q_41_pad_type_0, strides = var_8216, weight = down_blocks_2_attentions_0_transformer_blocks_6_attn1_to_q_weight_to_fp16, x = hidden_states_101_cast_fp16)[name = tensor("q_41_cast_fp16")]; + tensor var_8222 = const()[name = tensor("op_8222"), val = tensor([1, 1])]; + tensor var_8224 = const()[name = tensor("op_8224"), val = tensor([1, 1])]; + tensor k_81_pad_type_0 = const()[name = tensor("k_81_pad_type_0"), val = tensor("custom")]; + tensor k_81_pad_0 = const()[name = tensor("k_81_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_6_attn1_to_k_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_6_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(618525760)))]; + tensor k_81_cast_fp16 = conv(dilations = var_8224, groups = var_2624, pad = k_81_pad_0, pad_type = k_81_pad_type_0, strides = var_8222, weight = down_blocks_2_attentions_0_transformer_blocks_6_attn1_to_k_weight_to_fp16, x = hidden_states_101_cast_fp16)[name = tensor("k_81_cast_fp16")]; + tensor var_8228 = const()[name = tensor("op_8228"), val = tensor([1, 1])]; + tensor var_8230 = const()[name = tensor("op_8230"), val = tensor([1, 1])]; + tensor v_41_pad_type_0 = const()[name = tensor("v_41_pad_type_0"), val = tensor("custom")]; + tensor v_41_pad_0 = const()[name = tensor("v_41_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_6_attn1_to_v_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_6_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(621802624)))]; + tensor v_41_cast_fp16 = conv(dilations = var_8230, groups = var_2624, pad = v_41_pad_0, pad_type = v_41_pad_type_0, strides = var_8228, weight = down_blocks_2_attentions_0_transformer_blocks_6_attn1_to_v_weight_to_fp16, x = hidden_states_101_cast_fp16)[name = tensor("v_41_cast_fp16")]; + tensor var_8234_begin_0 = const()[name = tensor("op_8234_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8234_end_0 = const()[name = tensor("op_8234_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_8234_end_mask_0 = const()[name = tensor("op_8234_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8234_cast_fp16 = slice_by_index(begin = var_8234_begin_0, end = var_8234_end_0, end_mask = var_8234_end_mask_0, x = q_41_cast_fp16)[name = tensor("op_8234_cast_fp16")]; + tensor var_8238_begin_0 = const()[name = tensor("op_8238_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_8238_end_0 = const()[name = tensor("op_8238_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_8238_end_mask_0 = const()[name = tensor("op_8238_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8238_cast_fp16 = slice_by_index(begin = var_8238_begin_0, end = var_8238_end_0, end_mask = var_8238_end_mask_0, x = q_41_cast_fp16)[name = tensor("op_8238_cast_fp16")]; + tensor var_8242_begin_0 = const()[name = tensor("op_8242_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_8242_end_0 = const()[name = tensor("op_8242_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_8242_end_mask_0 = const()[name = tensor("op_8242_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8242_cast_fp16 = slice_by_index(begin = var_8242_begin_0, end = var_8242_end_0, end_mask = var_8242_end_mask_0, x = q_41_cast_fp16)[name = tensor("op_8242_cast_fp16")]; + tensor var_8246_begin_0 = const()[name = tensor("op_8246_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_8246_end_0 = const()[name = tensor("op_8246_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_8246_end_mask_0 = const()[name = tensor("op_8246_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8246_cast_fp16 = slice_by_index(begin = var_8246_begin_0, end = var_8246_end_0, end_mask = var_8246_end_mask_0, x = q_41_cast_fp16)[name = tensor("op_8246_cast_fp16")]; + tensor var_8250_begin_0 = const()[name = tensor("op_8250_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_8250_end_0 = const()[name = tensor("op_8250_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_8250_end_mask_0 = const()[name = tensor("op_8250_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8250_cast_fp16 = slice_by_index(begin = var_8250_begin_0, end = var_8250_end_0, end_mask = var_8250_end_mask_0, x = q_41_cast_fp16)[name = tensor("op_8250_cast_fp16")]; + tensor var_8254_begin_0 = const()[name = tensor("op_8254_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_8254_end_0 = const()[name = tensor("op_8254_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_8254_end_mask_0 = const()[name = tensor("op_8254_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8254_cast_fp16 = slice_by_index(begin = var_8254_begin_0, end = var_8254_end_0, end_mask = var_8254_end_mask_0, x = q_41_cast_fp16)[name = tensor("op_8254_cast_fp16")]; + tensor var_8258_begin_0 = const()[name = tensor("op_8258_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_8258_end_0 = const()[name = tensor("op_8258_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_8258_end_mask_0 = const()[name = tensor("op_8258_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8258_cast_fp16 = slice_by_index(begin = var_8258_begin_0, end = var_8258_end_0, end_mask = var_8258_end_mask_0, x = q_41_cast_fp16)[name = tensor("op_8258_cast_fp16")]; + tensor var_8262_begin_0 = const()[name = tensor("op_8262_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_8262_end_0 = const()[name = tensor("op_8262_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_8262_end_mask_0 = const()[name = tensor("op_8262_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8262_cast_fp16 = slice_by_index(begin = var_8262_begin_0, end = var_8262_end_0, end_mask = var_8262_end_mask_0, x = q_41_cast_fp16)[name = tensor("op_8262_cast_fp16")]; + tensor var_8266_begin_0 = const()[name = tensor("op_8266_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_8266_end_0 = const()[name = tensor("op_8266_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_8266_end_mask_0 = const()[name = tensor("op_8266_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8266_cast_fp16 = slice_by_index(begin = var_8266_begin_0, end = var_8266_end_0, end_mask = var_8266_end_mask_0, x = q_41_cast_fp16)[name = tensor("op_8266_cast_fp16")]; + tensor var_8270_begin_0 = const()[name = tensor("op_8270_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_8270_end_0 = const()[name = tensor("op_8270_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_8270_end_mask_0 = const()[name = tensor("op_8270_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8270_cast_fp16 = slice_by_index(begin = var_8270_begin_0, end = var_8270_end_0, end_mask = var_8270_end_mask_0, x = q_41_cast_fp16)[name = tensor("op_8270_cast_fp16")]; + tensor var_8274_begin_0 = const()[name = tensor("op_8274_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_8274_end_0 = const()[name = tensor("op_8274_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_8274_end_mask_0 = const()[name = tensor("op_8274_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8274_cast_fp16 = slice_by_index(begin = var_8274_begin_0, end = var_8274_end_0, end_mask = var_8274_end_mask_0, x = q_41_cast_fp16)[name = tensor("op_8274_cast_fp16")]; + tensor var_8278_begin_0 = const()[name = tensor("op_8278_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_8278_end_0 = const()[name = tensor("op_8278_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_8278_end_mask_0 = const()[name = tensor("op_8278_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8278_cast_fp16 = slice_by_index(begin = var_8278_begin_0, end = var_8278_end_0, end_mask = var_8278_end_mask_0, x = q_41_cast_fp16)[name = tensor("op_8278_cast_fp16")]; + tensor var_8282_begin_0 = const()[name = tensor("op_8282_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_8282_end_0 = const()[name = tensor("op_8282_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_8282_end_mask_0 = const()[name = tensor("op_8282_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8282_cast_fp16 = slice_by_index(begin = var_8282_begin_0, end = var_8282_end_0, end_mask = var_8282_end_mask_0, x = q_41_cast_fp16)[name = tensor("op_8282_cast_fp16")]; + tensor var_8286_begin_0 = const()[name = tensor("op_8286_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_8286_end_0 = const()[name = tensor("op_8286_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_8286_end_mask_0 = const()[name = tensor("op_8286_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8286_cast_fp16 = slice_by_index(begin = var_8286_begin_0, end = var_8286_end_0, end_mask = var_8286_end_mask_0, x = q_41_cast_fp16)[name = tensor("op_8286_cast_fp16")]; + tensor var_8290_begin_0 = const()[name = tensor("op_8290_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_8290_end_0 = const()[name = tensor("op_8290_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_8290_end_mask_0 = const()[name = tensor("op_8290_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8290_cast_fp16 = slice_by_index(begin = var_8290_begin_0, end = var_8290_end_0, end_mask = var_8290_end_mask_0, x = q_41_cast_fp16)[name = tensor("op_8290_cast_fp16")]; + tensor var_8294_begin_0 = const()[name = tensor("op_8294_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_8294_end_0 = const()[name = tensor("op_8294_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_8294_end_mask_0 = const()[name = tensor("op_8294_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8294_cast_fp16 = slice_by_index(begin = var_8294_begin_0, end = var_8294_end_0, end_mask = var_8294_end_mask_0, x = q_41_cast_fp16)[name = tensor("op_8294_cast_fp16")]; + tensor var_8298_begin_0 = const()[name = tensor("op_8298_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_8298_end_0 = const()[name = tensor("op_8298_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_8298_end_mask_0 = const()[name = tensor("op_8298_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8298_cast_fp16 = slice_by_index(begin = var_8298_begin_0, end = var_8298_end_0, end_mask = var_8298_end_mask_0, x = q_41_cast_fp16)[name = tensor("op_8298_cast_fp16")]; + tensor var_8302_begin_0 = const()[name = tensor("op_8302_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_8302_end_0 = const()[name = tensor("op_8302_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_8302_end_mask_0 = const()[name = tensor("op_8302_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8302_cast_fp16 = slice_by_index(begin = var_8302_begin_0, end = var_8302_end_0, end_mask = var_8302_end_mask_0, x = q_41_cast_fp16)[name = tensor("op_8302_cast_fp16")]; + tensor var_8306_begin_0 = const()[name = tensor("op_8306_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_8306_end_0 = const()[name = tensor("op_8306_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_8306_end_mask_0 = const()[name = tensor("op_8306_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8306_cast_fp16 = slice_by_index(begin = var_8306_begin_0, end = var_8306_end_0, end_mask = var_8306_end_mask_0, x = q_41_cast_fp16)[name = tensor("op_8306_cast_fp16")]; + tensor var_8310_begin_0 = const()[name = tensor("op_8310_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_8310_end_0 = const()[name = tensor("op_8310_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_8310_end_mask_0 = const()[name = tensor("op_8310_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8310_cast_fp16 = slice_by_index(begin = var_8310_begin_0, end = var_8310_end_0, end_mask = var_8310_end_mask_0, x = q_41_cast_fp16)[name = tensor("op_8310_cast_fp16")]; + tensor k_83_perm_0 = const()[name = tensor("k_83_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_8317_begin_0 = const()[name = tensor("op_8317_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8317_end_0 = const()[name = tensor("op_8317_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_8317_end_mask_0 = const()[name = tensor("op_8317_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_83_cast_fp16 = transpose(perm = k_83_perm_0, x = k_81_cast_fp16)[name = tensor("transpose_119")]; + tensor var_8317_cast_fp16 = slice_by_index(begin = var_8317_begin_0, end = var_8317_end_0, end_mask = var_8317_end_mask_0, x = k_83_cast_fp16)[name = tensor("op_8317_cast_fp16")]; + tensor var_8321_begin_0 = const()[name = tensor("op_8321_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_8321_end_0 = const()[name = tensor("op_8321_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_8321_end_mask_0 = const()[name = tensor("op_8321_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8321_cast_fp16 = slice_by_index(begin = var_8321_begin_0, end = var_8321_end_0, end_mask = var_8321_end_mask_0, x = k_83_cast_fp16)[name = tensor("op_8321_cast_fp16")]; + tensor var_8325_begin_0 = const()[name = tensor("op_8325_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_8325_end_0 = const()[name = tensor("op_8325_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_8325_end_mask_0 = const()[name = tensor("op_8325_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8325_cast_fp16 = slice_by_index(begin = var_8325_begin_0, end = var_8325_end_0, end_mask = var_8325_end_mask_0, x = k_83_cast_fp16)[name = tensor("op_8325_cast_fp16")]; + tensor var_8329_begin_0 = const()[name = tensor("op_8329_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_8329_end_0 = const()[name = tensor("op_8329_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_8329_end_mask_0 = const()[name = tensor("op_8329_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8329_cast_fp16 = slice_by_index(begin = var_8329_begin_0, end = var_8329_end_0, end_mask = var_8329_end_mask_0, x = k_83_cast_fp16)[name = tensor("op_8329_cast_fp16")]; + tensor var_8333_begin_0 = const()[name = tensor("op_8333_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_8333_end_0 = const()[name = tensor("op_8333_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_8333_end_mask_0 = const()[name = tensor("op_8333_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8333_cast_fp16 = slice_by_index(begin = var_8333_begin_0, end = var_8333_end_0, end_mask = var_8333_end_mask_0, x = k_83_cast_fp16)[name = tensor("op_8333_cast_fp16")]; + tensor var_8337_begin_0 = const()[name = tensor("op_8337_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_8337_end_0 = const()[name = tensor("op_8337_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_8337_end_mask_0 = const()[name = tensor("op_8337_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8337_cast_fp16 = slice_by_index(begin = var_8337_begin_0, end = var_8337_end_0, end_mask = var_8337_end_mask_0, x = k_83_cast_fp16)[name = tensor("op_8337_cast_fp16")]; + tensor var_8341_begin_0 = const()[name = tensor("op_8341_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_8341_end_0 = const()[name = tensor("op_8341_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_8341_end_mask_0 = const()[name = tensor("op_8341_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8341_cast_fp16 = slice_by_index(begin = var_8341_begin_0, end = var_8341_end_0, end_mask = var_8341_end_mask_0, x = k_83_cast_fp16)[name = tensor("op_8341_cast_fp16")]; + tensor var_8345_begin_0 = const()[name = tensor("op_8345_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_8345_end_0 = const()[name = tensor("op_8345_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_8345_end_mask_0 = const()[name = tensor("op_8345_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8345_cast_fp16 = slice_by_index(begin = var_8345_begin_0, end = var_8345_end_0, end_mask = var_8345_end_mask_0, x = k_83_cast_fp16)[name = tensor("op_8345_cast_fp16")]; + tensor var_8349_begin_0 = const()[name = tensor("op_8349_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_8349_end_0 = const()[name = tensor("op_8349_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_8349_end_mask_0 = const()[name = tensor("op_8349_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8349_cast_fp16 = slice_by_index(begin = var_8349_begin_0, end = var_8349_end_0, end_mask = var_8349_end_mask_0, x = k_83_cast_fp16)[name = tensor("op_8349_cast_fp16")]; + tensor var_8353_begin_0 = const()[name = tensor("op_8353_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_8353_end_0 = const()[name = tensor("op_8353_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_8353_end_mask_0 = const()[name = tensor("op_8353_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8353_cast_fp16 = slice_by_index(begin = var_8353_begin_0, end = var_8353_end_0, end_mask = var_8353_end_mask_0, x = k_83_cast_fp16)[name = tensor("op_8353_cast_fp16")]; + tensor var_8357_begin_0 = const()[name = tensor("op_8357_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_8357_end_0 = const()[name = tensor("op_8357_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_8357_end_mask_0 = const()[name = tensor("op_8357_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8357_cast_fp16 = slice_by_index(begin = var_8357_begin_0, end = var_8357_end_0, end_mask = var_8357_end_mask_0, x = k_83_cast_fp16)[name = tensor("op_8357_cast_fp16")]; + tensor var_8361_begin_0 = const()[name = tensor("op_8361_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_8361_end_0 = const()[name = tensor("op_8361_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_8361_end_mask_0 = const()[name = tensor("op_8361_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8361_cast_fp16 = slice_by_index(begin = var_8361_begin_0, end = var_8361_end_0, end_mask = var_8361_end_mask_0, x = k_83_cast_fp16)[name = tensor("op_8361_cast_fp16")]; + tensor var_8365_begin_0 = const()[name = tensor("op_8365_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_8365_end_0 = const()[name = tensor("op_8365_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_8365_end_mask_0 = const()[name = tensor("op_8365_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8365_cast_fp16 = slice_by_index(begin = var_8365_begin_0, end = var_8365_end_0, end_mask = var_8365_end_mask_0, x = k_83_cast_fp16)[name = tensor("op_8365_cast_fp16")]; + tensor var_8369_begin_0 = const()[name = tensor("op_8369_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_8369_end_0 = const()[name = tensor("op_8369_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_8369_end_mask_0 = const()[name = tensor("op_8369_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8369_cast_fp16 = slice_by_index(begin = var_8369_begin_0, end = var_8369_end_0, end_mask = var_8369_end_mask_0, x = k_83_cast_fp16)[name = tensor("op_8369_cast_fp16")]; + tensor var_8373_begin_0 = const()[name = tensor("op_8373_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_8373_end_0 = const()[name = tensor("op_8373_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_8373_end_mask_0 = const()[name = tensor("op_8373_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8373_cast_fp16 = slice_by_index(begin = var_8373_begin_0, end = var_8373_end_0, end_mask = var_8373_end_mask_0, x = k_83_cast_fp16)[name = tensor("op_8373_cast_fp16")]; + tensor var_8377_begin_0 = const()[name = tensor("op_8377_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_8377_end_0 = const()[name = tensor("op_8377_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_8377_end_mask_0 = const()[name = tensor("op_8377_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8377_cast_fp16 = slice_by_index(begin = var_8377_begin_0, end = var_8377_end_0, end_mask = var_8377_end_mask_0, x = k_83_cast_fp16)[name = tensor("op_8377_cast_fp16")]; + tensor var_8381_begin_0 = const()[name = tensor("op_8381_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_8381_end_0 = const()[name = tensor("op_8381_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_8381_end_mask_0 = const()[name = tensor("op_8381_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8381_cast_fp16 = slice_by_index(begin = var_8381_begin_0, end = var_8381_end_0, end_mask = var_8381_end_mask_0, x = k_83_cast_fp16)[name = tensor("op_8381_cast_fp16")]; + tensor var_8385_begin_0 = const()[name = tensor("op_8385_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_8385_end_0 = const()[name = tensor("op_8385_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_8385_end_mask_0 = const()[name = tensor("op_8385_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8385_cast_fp16 = slice_by_index(begin = var_8385_begin_0, end = var_8385_end_0, end_mask = var_8385_end_mask_0, x = k_83_cast_fp16)[name = tensor("op_8385_cast_fp16")]; + tensor var_8389_begin_0 = const()[name = tensor("op_8389_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_8389_end_0 = const()[name = tensor("op_8389_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_8389_end_mask_0 = const()[name = tensor("op_8389_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8389_cast_fp16 = slice_by_index(begin = var_8389_begin_0, end = var_8389_end_0, end_mask = var_8389_end_mask_0, x = k_83_cast_fp16)[name = tensor("op_8389_cast_fp16")]; + tensor var_8393_begin_0 = const()[name = tensor("op_8393_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_8393_end_0 = const()[name = tensor("op_8393_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_8393_end_mask_0 = const()[name = tensor("op_8393_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8393_cast_fp16 = slice_by_index(begin = var_8393_begin_0, end = var_8393_end_0, end_mask = var_8393_end_mask_0, x = k_83_cast_fp16)[name = tensor("op_8393_cast_fp16")]; + tensor var_8395_begin_0 = const()[name = tensor("op_8395_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8395_end_0 = const()[name = tensor("op_8395_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_8395_end_mask_0 = const()[name = tensor("op_8395_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8395_cast_fp16 = slice_by_index(begin = var_8395_begin_0, end = var_8395_end_0, end_mask = var_8395_end_mask_0, x = v_41_cast_fp16)[name = tensor("op_8395_cast_fp16")]; + tensor var_8399_begin_0 = const()[name = tensor("op_8399_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_8399_end_0 = const()[name = tensor("op_8399_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_8399_end_mask_0 = const()[name = tensor("op_8399_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8399_cast_fp16 = slice_by_index(begin = var_8399_begin_0, end = var_8399_end_0, end_mask = var_8399_end_mask_0, x = v_41_cast_fp16)[name = tensor("op_8399_cast_fp16")]; + tensor var_8403_begin_0 = const()[name = tensor("op_8403_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_8403_end_0 = const()[name = tensor("op_8403_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_8403_end_mask_0 = const()[name = tensor("op_8403_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8403_cast_fp16 = slice_by_index(begin = var_8403_begin_0, end = var_8403_end_0, end_mask = var_8403_end_mask_0, x = v_41_cast_fp16)[name = tensor("op_8403_cast_fp16")]; + tensor var_8407_begin_0 = const()[name = tensor("op_8407_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_8407_end_0 = const()[name = tensor("op_8407_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_8407_end_mask_0 = const()[name = tensor("op_8407_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8407_cast_fp16 = slice_by_index(begin = var_8407_begin_0, end = var_8407_end_0, end_mask = var_8407_end_mask_0, x = v_41_cast_fp16)[name = tensor("op_8407_cast_fp16")]; + tensor var_8411_begin_0 = const()[name = tensor("op_8411_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_8411_end_0 = const()[name = tensor("op_8411_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_8411_end_mask_0 = const()[name = tensor("op_8411_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8411_cast_fp16 = slice_by_index(begin = var_8411_begin_0, end = var_8411_end_0, end_mask = var_8411_end_mask_0, x = v_41_cast_fp16)[name = tensor("op_8411_cast_fp16")]; + tensor var_8415_begin_0 = const()[name = tensor("op_8415_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_8415_end_0 = const()[name = tensor("op_8415_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_8415_end_mask_0 = const()[name = tensor("op_8415_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8415_cast_fp16 = slice_by_index(begin = var_8415_begin_0, end = var_8415_end_0, end_mask = var_8415_end_mask_0, x = v_41_cast_fp16)[name = tensor("op_8415_cast_fp16")]; + tensor var_8419_begin_0 = const()[name = tensor("op_8419_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_8419_end_0 = const()[name = tensor("op_8419_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_8419_end_mask_0 = const()[name = tensor("op_8419_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8419_cast_fp16 = slice_by_index(begin = var_8419_begin_0, end = var_8419_end_0, end_mask = var_8419_end_mask_0, x = v_41_cast_fp16)[name = tensor("op_8419_cast_fp16")]; + tensor var_8423_begin_0 = const()[name = tensor("op_8423_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_8423_end_0 = const()[name = tensor("op_8423_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_8423_end_mask_0 = const()[name = tensor("op_8423_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8423_cast_fp16 = slice_by_index(begin = var_8423_begin_0, end = var_8423_end_0, end_mask = var_8423_end_mask_0, x = v_41_cast_fp16)[name = tensor("op_8423_cast_fp16")]; + tensor var_8427_begin_0 = const()[name = tensor("op_8427_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_8427_end_0 = const()[name = tensor("op_8427_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_8427_end_mask_0 = const()[name = tensor("op_8427_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8427_cast_fp16 = slice_by_index(begin = var_8427_begin_0, end = var_8427_end_0, end_mask = var_8427_end_mask_0, x = v_41_cast_fp16)[name = tensor("op_8427_cast_fp16")]; + tensor var_8431_begin_0 = const()[name = tensor("op_8431_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_8431_end_0 = const()[name = tensor("op_8431_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_8431_end_mask_0 = const()[name = tensor("op_8431_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8431_cast_fp16 = slice_by_index(begin = var_8431_begin_0, end = var_8431_end_0, end_mask = var_8431_end_mask_0, x = v_41_cast_fp16)[name = tensor("op_8431_cast_fp16")]; + tensor var_8435_begin_0 = const()[name = tensor("op_8435_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_8435_end_0 = const()[name = tensor("op_8435_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_8435_end_mask_0 = const()[name = tensor("op_8435_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8435_cast_fp16 = slice_by_index(begin = var_8435_begin_0, end = var_8435_end_0, end_mask = var_8435_end_mask_0, x = v_41_cast_fp16)[name = tensor("op_8435_cast_fp16")]; + tensor var_8439_begin_0 = const()[name = tensor("op_8439_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_8439_end_0 = const()[name = tensor("op_8439_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_8439_end_mask_0 = const()[name = tensor("op_8439_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8439_cast_fp16 = slice_by_index(begin = var_8439_begin_0, end = var_8439_end_0, end_mask = var_8439_end_mask_0, x = v_41_cast_fp16)[name = tensor("op_8439_cast_fp16")]; + tensor var_8443_begin_0 = const()[name = tensor("op_8443_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_8443_end_0 = const()[name = tensor("op_8443_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_8443_end_mask_0 = const()[name = tensor("op_8443_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8443_cast_fp16 = slice_by_index(begin = var_8443_begin_0, end = var_8443_end_0, end_mask = var_8443_end_mask_0, x = v_41_cast_fp16)[name = tensor("op_8443_cast_fp16")]; + tensor var_8447_begin_0 = const()[name = tensor("op_8447_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_8447_end_0 = const()[name = tensor("op_8447_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_8447_end_mask_0 = const()[name = tensor("op_8447_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8447_cast_fp16 = slice_by_index(begin = var_8447_begin_0, end = var_8447_end_0, end_mask = var_8447_end_mask_0, x = v_41_cast_fp16)[name = tensor("op_8447_cast_fp16")]; + tensor var_8451_begin_0 = const()[name = tensor("op_8451_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_8451_end_0 = const()[name = tensor("op_8451_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_8451_end_mask_0 = const()[name = tensor("op_8451_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8451_cast_fp16 = slice_by_index(begin = var_8451_begin_0, end = var_8451_end_0, end_mask = var_8451_end_mask_0, x = v_41_cast_fp16)[name = tensor("op_8451_cast_fp16")]; + tensor var_8455_begin_0 = const()[name = tensor("op_8455_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_8455_end_0 = const()[name = tensor("op_8455_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_8455_end_mask_0 = const()[name = tensor("op_8455_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8455_cast_fp16 = slice_by_index(begin = var_8455_begin_0, end = var_8455_end_0, end_mask = var_8455_end_mask_0, x = v_41_cast_fp16)[name = tensor("op_8455_cast_fp16")]; + tensor var_8459_begin_0 = const()[name = tensor("op_8459_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_8459_end_0 = const()[name = tensor("op_8459_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_8459_end_mask_0 = const()[name = tensor("op_8459_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8459_cast_fp16 = slice_by_index(begin = var_8459_begin_0, end = var_8459_end_0, end_mask = var_8459_end_mask_0, x = v_41_cast_fp16)[name = tensor("op_8459_cast_fp16")]; + tensor var_8463_begin_0 = const()[name = tensor("op_8463_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_8463_end_0 = const()[name = tensor("op_8463_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_8463_end_mask_0 = const()[name = tensor("op_8463_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8463_cast_fp16 = slice_by_index(begin = var_8463_begin_0, end = var_8463_end_0, end_mask = var_8463_end_mask_0, x = v_41_cast_fp16)[name = tensor("op_8463_cast_fp16")]; + tensor var_8467_begin_0 = const()[name = tensor("op_8467_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_8467_end_0 = const()[name = tensor("op_8467_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_8467_end_mask_0 = const()[name = tensor("op_8467_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8467_cast_fp16 = slice_by_index(begin = var_8467_begin_0, end = var_8467_end_0, end_mask = var_8467_end_mask_0, x = v_41_cast_fp16)[name = tensor("op_8467_cast_fp16")]; + tensor var_8471_begin_0 = const()[name = tensor("op_8471_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_8471_end_0 = const()[name = tensor("op_8471_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_8471_end_mask_0 = const()[name = tensor("op_8471_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8471_cast_fp16 = slice_by_index(begin = var_8471_begin_0, end = var_8471_end_0, end_mask = var_8471_end_mask_0, x = v_41_cast_fp16)[name = tensor("op_8471_cast_fp16")]; + tensor var_8475_equation_0 = const()[name = tensor("op_8475_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8475_cast_fp16 = einsum(equation = var_8475_equation_0, values = (var_8317_cast_fp16, var_8234_cast_fp16))[name = tensor("op_8475_cast_fp16")]; + tensor var_8476_to_fp16 = const()[name = tensor("op_8476_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_641_cast_fp16 = mul(x = var_8475_cast_fp16, y = var_8476_to_fp16)[name = tensor("aw_641_cast_fp16")]; + tensor var_8479_equation_0 = const()[name = tensor("op_8479_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8479_cast_fp16 = einsum(equation = var_8479_equation_0, values = (var_8321_cast_fp16, var_8238_cast_fp16))[name = tensor("op_8479_cast_fp16")]; + tensor var_8480_to_fp16 = const()[name = tensor("op_8480_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_643_cast_fp16 = mul(x = var_8479_cast_fp16, y = var_8480_to_fp16)[name = tensor("aw_643_cast_fp16")]; + tensor var_8483_equation_0 = const()[name = tensor("op_8483_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8483_cast_fp16 = einsum(equation = var_8483_equation_0, values = (var_8325_cast_fp16, var_8242_cast_fp16))[name = tensor("op_8483_cast_fp16")]; + tensor var_8484_to_fp16 = const()[name = tensor("op_8484_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_645_cast_fp16 = mul(x = var_8483_cast_fp16, y = var_8484_to_fp16)[name = tensor("aw_645_cast_fp16")]; + tensor var_8487_equation_0 = const()[name = tensor("op_8487_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8487_cast_fp16 = einsum(equation = var_8487_equation_0, values = (var_8329_cast_fp16, var_8246_cast_fp16))[name = tensor("op_8487_cast_fp16")]; + tensor var_8488_to_fp16 = const()[name = tensor("op_8488_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_647_cast_fp16 = mul(x = var_8487_cast_fp16, y = var_8488_to_fp16)[name = tensor("aw_647_cast_fp16")]; + tensor var_8491_equation_0 = const()[name = tensor("op_8491_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8491_cast_fp16 = einsum(equation = var_8491_equation_0, values = (var_8333_cast_fp16, var_8250_cast_fp16))[name = tensor("op_8491_cast_fp16")]; + tensor var_8492_to_fp16 = const()[name = tensor("op_8492_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_649_cast_fp16 = mul(x = var_8491_cast_fp16, y = var_8492_to_fp16)[name = tensor("aw_649_cast_fp16")]; + tensor var_8495_equation_0 = const()[name = tensor("op_8495_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8495_cast_fp16 = einsum(equation = var_8495_equation_0, values = (var_8337_cast_fp16, var_8254_cast_fp16))[name = tensor("op_8495_cast_fp16")]; + tensor var_8496_to_fp16 = const()[name = tensor("op_8496_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_651_cast_fp16 = mul(x = var_8495_cast_fp16, y = var_8496_to_fp16)[name = tensor("aw_651_cast_fp16")]; + tensor var_8499_equation_0 = const()[name = tensor("op_8499_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8499_cast_fp16 = einsum(equation = var_8499_equation_0, values = (var_8341_cast_fp16, var_8258_cast_fp16))[name = tensor("op_8499_cast_fp16")]; + tensor var_8500_to_fp16 = const()[name = tensor("op_8500_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_653_cast_fp16 = mul(x = var_8499_cast_fp16, y = var_8500_to_fp16)[name = tensor("aw_653_cast_fp16")]; + tensor var_8503_equation_0 = const()[name = tensor("op_8503_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8503_cast_fp16 = einsum(equation = var_8503_equation_0, values = (var_8345_cast_fp16, var_8262_cast_fp16))[name = tensor("op_8503_cast_fp16")]; + tensor var_8504_to_fp16 = const()[name = tensor("op_8504_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_655_cast_fp16 = mul(x = var_8503_cast_fp16, y = var_8504_to_fp16)[name = tensor("aw_655_cast_fp16")]; + tensor var_8507_equation_0 = const()[name = tensor("op_8507_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8507_cast_fp16 = einsum(equation = var_8507_equation_0, values = (var_8349_cast_fp16, var_8266_cast_fp16))[name = tensor("op_8507_cast_fp16")]; + tensor var_8508_to_fp16 = const()[name = tensor("op_8508_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_657_cast_fp16 = mul(x = var_8507_cast_fp16, y = var_8508_to_fp16)[name = tensor("aw_657_cast_fp16")]; + tensor var_8511_equation_0 = const()[name = tensor("op_8511_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8511_cast_fp16 = einsum(equation = var_8511_equation_0, values = (var_8353_cast_fp16, var_8270_cast_fp16))[name = tensor("op_8511_cast_fp16")]; + tensor var_8512_to_fp16 = const()[name = tensor("op_8512_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_659_cast_fp16 = mul(x = var_8511_cast_fp16, y = var_8512_to_fp16)[name = tensor("aw_659_cast_fp16")]; + tensor var_8515_equation_0 = const()[name = tensor("op_8515_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8515_cast_fp16 = einsum(equation = var_8515_equation_0, values = (var_8357_cast_fp16, var_8274_cast_fp16))[name = tensor("op_8515_cast_fp16")]; + tensor var_8516_to_fp16 = const()[name = tensor("op_8516_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_661_cast_fp16 = mul(x = var_8515_cast_fp16, y = var_8516_to_fp16)[name = tensor("aw_661_cast_fp16")]; + tensor var_8519_equation_0 = const()[name = tensor("op_8519_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8519_cast_fp16 = einsum(equation = var_8519_equation_0, values = (var_8361_cast_fp16, var_8278_cast_fp16))[name = tensor("op_8519_cast_fp16")]; + tensor var_8520_to_fp16 = const()[name = tensor("op_8520_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_663_cast_fp16 = mul(x = var_8519_cast_fp16, y = var_8520_to_fp16)[name = tensor("aw_663_cast_fp16")]; + tensor var_8523_equation_0 = const()[name = tensor("op_8523_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8523_cast_fp16 = einsum(equation = var_8523_equation_0, values = (var_8365_cast_fp16, var_8282_cast_fp16))[name = tensor("op_8523_cast_fp16")]; + tensor var_8524_to_fp16 = const()[name = tensor("op_8524_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_665_cast_fp16 = mul(x = var_8523_cast_fp16, y = var_8524_to_fp16)[name = tensor("aw_665_cast_fp16")]; + tensor var_8527_equation_0 = const()[name = tensor("op_8527_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8527_cast_fp16 = einsum(equation = var_8527_equation_0, values = (var_8369_cast_fp16, var_8286_cast_fp16))[name = tensor("op_8527_cast_fp16")]; + tensor var_8528_to_fp16 = const()[name = tensor("op_8528_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_667_cast_fp16 = mul(x = var_8527_cast_fp16, y = var_8528_to_fp16)[name = tensor("aw_667_cast_fp16")]; + tensor var_8531_equation_0 = const()[name = tensor("op_8531_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8531_cast_fp16 = einsum(equation = var_8531_equation_0, values = (var_8373_cast_fp16, var_8290_cast_fp16))[name = tensor("op_8531_cast_fp16")]; + tensor var_8532_to_fp16 = const()[name = tensor("op_8532_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_669_cast_fp16 = mul(x = var_8531_cast_fp16, y = var_8532_to_fp16)[name = tensor("aw_669_cast_fp16")]; + tensor var_8535_equation_0 = const()[name = tensor("op_8535_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8535_cast_fp16 = einsum(equation = var_8535_equation_0, values = (var_8377_cast_fp16, var_8294_cast_fp16))[name = tensor("op_8535_cast_fp16")]; + tensor var_8536_to_fp16 = const()[name = tensor("op_8536_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_671_cast_fp16 = mul(x = var_8535_cast_fp16, y = var_8536_to_fp16)[name = tensor("aw_671_cast_fp16")]; + tensor var_8539_equation_0 = const()[name = tensor("op_8539_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8539_cast_fp16 = einsum(equation = var_8539_equation_0, values = (var_8381_cast_fp16, var_8298_cast_fp16))[name = tensor("op_8539_cast_fp16")]; + tensor var_8540_to_fp16 = const()[name = tensor("op_8540_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_673_cast_fp16 = mul(x = var_8539_cast_fp16, y = var_8540_to_fp16)[name = tensor("aw_673_cast_fp16")]; + tensor var_8543_equation_0 = const()[name = tensor("op_8543_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8543_cast_fp16 = einsum(equation = var_8543_equation_0, values = (var_8385_cast_fp16, var_8302_cast_fp16))[name = tensor("op_8543_cast_fp16")]; + tensor var_8544_to_fp16 = const()[name = tensor("op_8544_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_675_cast_fp16 = mul(x = var_8543_cast_fp16, y = var_8544_to_fp16)[name = tensor("aw_675_cast_fp16")]; + tensor var_8547_equation_0 = const()[name = tensor("op_8547_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8547_cast_fp16 = einsum(equation = var_8547_equation_0, values = (var_8389_cast_fp16, var_8306_cast_fp16))[name = tensor("op_8547_cast_fp16")]; + tensor var_8548_to_fp16 = const()[name = tensor("op_8548_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_677_cast_fp16 = mul(x = var_8547_cast_fp16, y = var_8548_to_fp16)[name = tensor("aw_677_cast_fp16")]; + tensor var_8551_equation_0 = const()[name = tensor("op_8551_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8551_cast_fp16 = einsum(equation = var_8551_equation_0, values = (var_8393_cast_fp16, var_8310_cast_fp16))[name = tensor("op_8551_cast_fp16")]; + tensor var_8552_to_fp16 = const()[name = tensor("op_8552_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_679_cast_fp16 = mul(x = var_8551_cast_fp16, y = var_8552_to_fp16)[name = tensor("aw_679_cast_fp16")]; + tensor var_8554_cast_fp16 = softmax(axis = var_2624, x = aw_641_cast_fp16)[name = tensor("op_8554_cast_fp16")]; + tensor var_8555_cast_fp16 = softmax(axis = var_2624, x = aw_643_cast_fp16)[name = tensor("op_8555_cast_fp16")]; + tensor var_8556_cast_fp16 = softmax(axis = var_2624, x = aw_645_cast_fp16)[name = tensor("op_8556_cast_fp16")]; + tensor var_8557_cast_fp16 = softmax(axis = var_2624, x = aw_647_cast_fp16)[name = tensor("op_8557_cast_fp16")]; + tensor var_8558_cast_fp16 = softmax(axis = var_2624, x = aw_649_cast_fp16)[name = tensor("op_8558_cast_fp16")]; + tensor var_8559_cast_fp16 = softmax(axis = var_2624, x = aw_651_cast_fp16)[name = tensor("op_8559_cast_fp16")]; + tensor var_8560_cast_fp16 = softmax(axis = var_2624, x = aw_653_cast_fp16)[name = tensor("op_8560_cast_fp16")]; + tensor var_8561_cast_fp16 = softmax(axis = var_2624, x = aw_655_cast_fp16)[name = tensor("op_8561_cast_fp16")]; + tensor var_8562_cast_fp16 = softmax(axis = var_2624, x = aw_657_cast_fp16)[name = tensor("op_8562_cast_fp16")]; + tensor var_8563_cast_fp16 = softmax(axis = var_2624, x = aw_659_cast_fp16)[name = tensor("op_8563_cast_fp16")]; + tensor var_8564_cast_fp16 = softmax(axis = var_2624, x = aw_661_cast_fp16)[name = tensor("op_8564_cast_fp16")]; + tensor var_8565_cast_fp16 = softmax(axis = var_2624, x = aw_663_cast_fp16)[name = tensor("op_8565_cast_fp16")]; + tensor var_8566_cast_fp16 = softmax(axis = var_2624, x = aw_665_cast_fp16)[name = tensor("op_8566_cast_fp16")]; + tensor var_8567_cast_fp16 = softmax(axis = var_2624, x = aw_667_cast_fp16)[name = tensor("op_8567_cast_fp16")]; + tensor var_8568_cast_fp16 = softmax(axis = var_2624, x = aw_669_cast_fp16)[name = tensor("op_8568_cast_fp16")]; + tensor var_8569_cast_fp16 = softmax(axis = var_2624, x = aw_671_cast_fp16)[name = tensor("op_8569_cast_fp16")]; + tensor var_8570_cast_fp16 = softmax(axis = var_2624, x = aw_673_cast_fp16)[name = tensor("op_8570_cast_fp16")]; + tensor var_8571_cast_fp16 = softmax(axis = var_2624, x = aw_675_cast_fp16)[name = tensor("op_8571_cast_fp16")]; + tensor var_8572_cast_fp16 = softmax(axis = var_2624, x = aw_677_cast_fp16)[name = tensor("op_8572_cast_fp16")]; + tensor var_8573_cast_fp16 = softmax(axis = var_2624, x = aw_679_cast_fp16)[name = tensor("op_8573_cast_fp16")]; + tensor var_8575_equation_0 = const()[name = tensor("op_8575_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8575_cast_fp16 = einsum(equation = var_8575_equation_0, values = (var_8395_cast_fp16, var_8554_cast_fp16))[name = tensor("op_8575_cast_fp16")]; + tensor var_8577_equation_0 = const()[name = tensor("op_8577_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8577_cast_fp16 = einsum(equation = var_8577_equation_0, values = (var_8399_cast_fp16, var_8555_cast_fp16))[name = tensor("op_8577_cast_fp16")]; + tensor var_8579_equation_0 = const()[name = tensor("op_8579_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8579_cast_fp16 = einsum(equation = var_8579_equation_0, values = (var_8403_cast_fp16, var_8556_cast_fp16))[name = tensor("op_8579_cast_fp16")]; + tensor var_8581_equation_0 = const()[name = tensor("op_8581_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8581_cast_fp16 = einsum(equation = var_8581_equation_0, values = (var_8407_cast_fp16, var_8557_cast_fp16))[name = tensor("op_8581_cast_fp16")]; + tensor var_8583_equation_0 = const()[name = tensor("op_8583_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8583_cast_fp16 = einsum(equation = var_8583_equation_0, values = (var_8411_cast_fp16, var_8558_cast_fp16))[name = tensor("op_8583_cast_fp16")]; + tensor var_8585_equation_0 = const()[name = tensor("op_8585_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8585_cast_fp16 = einsum(equation = var_8585_equation_0, values = (var_8415_cast_fp16, var_8559_cast_fp16))[name = tensor("op_8585_cast_fp16")]; + tensor var_8587_equation_0 = const()[name = tensor("op_8587_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8587_cast_fp16 = einsum(equation = var_8587_equation_0, values = (var_8419_cast_fp16, var_8560_cast_fp16))[name = tensor("op_8587_cast_fp16")]; + tensor var_8589_equation_0 = const()[name = tensor("op_8589_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8589_cast_fp16 = einsum(equation = var_8589_equation_0, values = (var_8423_cast_fp16, var_8561_cast_fp16))[name = tensor("op_8589_cast_fp16")]; + tensor var_8591_equation_0 = const()[name = tensor("op_8591_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8591_cast_fp16 = einsum(equation = var_8591_equation_0, values = (var_8427_cast_fp16, var_8562_cast_fp16))[name = tensor("op_8591_cast_fp16")]; + tensor var_8593_equation_0 = const()[name = tensor("op_8593_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8593_cast_fp16 = einsum(equation = var_8593_equation_0, values = (var_8431_cast_fp16, var_8563_cast_fp16))[name = tensor("op_8593_cast_fp16")]; + tensor var_8595_equation_0 = const()[name = tensor("op_8595_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8595_cast_fp16 = einsum(equation = var_8595_equation_0, values = (var_8435_cast_fp16, var_8564_cast_fp16))[name = tensor("op_8595_cast_fp16")]; + tensor var_8597_equation_0 = const()[name = tensor("op_8597_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8597_cast_fp16 = einsum(equation = var_8597_equation_0, values = (var_8439_cast_fp16, var_8565_cast_fp16))[name = tensor("op_8597_cast_fp16")]; + tensor var_8599_equation_0 = const()[name = tensor("op_8599_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8599_cast_fp16 = einsum(equation = var_8599_equation_0, values = (var_8443_cast_fp16, var_8566_cast_fp16))[name = tensor("op_8599_cast_fp16")]; + tensor var_8601_equation_0 = const()[name = tensor("op_8601_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8601_cast_fp16 = einsum(equation = var_8601_equation_0, values = (var_8447_cast_fp16, var_8567_cast_fp16))[name = tensor("op_8601_cast_fp16")]; + tensor var_8603_equation_0 = const()[name = tensor("op_8603_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8603_cast_fp16 = einsum(equation = var_8603_equation_0, values = (var_8451_cast_fp16, var_8568_cast_fp16))[name = tensor("op_8603_cast_fp16")]; + tensor var_8605_equation_0 = const()[name = tensor("op_8605_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8605_cast_fp16 = einsum(equation = var_8605_equation_0, values = (var_8455_cast_fp16, var_8569_cast_fp16))[name = tensor("op_8605_cast_fp16")]; + tensor var_8607_equation_0 = const()[name = tensor("op_8607_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8607_cast_fp16 = einsum(equation = var_8607_equation_0, values = (var_8459_cast_fp16, var_8570_cast_fp16))[name = tensor("op_8607_cast_fp16")]; + tensor var_8609_equation_0 = const()[name = tensor("op_8609_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8609_cast_fp16 = einsum(equation = var_8609_equation_0, values = (var_8463_cast_fp16, var_8571_cast_fp16))[name = tensor("op_8609_cast_fp16")]; + tensor var_8611_equation_0 = const()[name = tensor("op_8611_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8611_cast_fp16 = einsum(equation = var_8611_equation_0, values = (var_8467_cast_fp16, var_8572_cast_fp16))[name = tensor("op_8611_cast_fp16")]; + tensor var_8613_equation_0 = const()[name = tensor("op_8613_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8613_cast_fp16 = einsum(equation = var_8613_equation_0, values = (var_8471_cast_fp16, var_8573_cast_fp16))[name = tensor("op_8613_cast_fp16")]; + tensor input_179_interleave_0 = const()[name = tensor("input_179_interleave_0"), val = tensor(false)]; + tensor input_179_cast_fp16 = concat(axis = var_2624, interleave = input_179_interleave_0, values = (var_8575_cast_fp16, var_8577_cast_fp16, var_8579_cast_fp16, var_8581_cast_fp16, var_8583_cast_fp16, var_8585_cast_fp16, var_8587_cast_fp16, var_8589_cast_fp16, var_8591_cast_fp16, var_8593_cast_fp16, var_8595_cast_fp16, var_8597_cast_fp16, var_8599_cast_fp16, var_8601_cast_fp16, var_8603_cast_fp16, var_8605_cast_fp16, var_8607_cast_fp16, var_8609_cast_fp16, var_8611_cast_fp16, var_8613_cast_fp16))[name = tensor("input_179_cast_fp16")]; + tensor var_8619 = const()[name = tensor("op_8619"), val = tensor([1, 1])]; + tensor var_8621 = const()[name = tensor("op_8621"), val = tensor([1, 1])]; + tensor var_8623_pad_type_0 = const()[name = tensor("op_8623_pad_type_0"), val = tensor("custom")]; + tensor var_8623_pad_0 = const()[name = tensor("op_8623_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_6_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_6_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(625079488)))]; + tensor down_blocks_2_attentions_0_transformer_blocks_6_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_6_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(628356352)))]; + tensor var_8623_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_6_attn1_to_out_0_bias_to_fp16, dilations = var_8621, groups = var_2624, pad = var_8623_pad_0, pad_type = var_8623_pad_type_0, strides = var_8619, weight = down_blocks_2_attentions_0_transformer_blocks_6_attn1_to_out_0_weight_to_fp16, x = input_179_cast_fp16)[name = tensor("op_8623_cast_fp16")]; + tensor inputs_63_cast_fp16 = add(x = var_8623_cast_fp16, y = inputs_61_cast_fp16)[name = tensor("inputs_63_cast_fp16")]; + tensor hidden_states_103_axes_0 = const()[name = tensor("hidden_states_103_axes_0"), val = tensor([1])]; + tensor hidden_states_103_gamma_0_to_fp16 = const()[name = tensor("hidden_states_103_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(628358976)))]; + tensor hidden_states_103_beta_0_to_fp16 = const()[name = tensor("hidden_states_103_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(628361600)))]; + tensor var_8633_to_fp16 = const()[name = tensor("op_8633_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_103_cast_fp16 = layer_norm(axes = hidden_states_103_axes_0, beta = hidden_states_103_beta_0_to_fp16, epsilon = var_8633_to_fp16, gamma = hidden_states_103_gamma_0_to_fp16, x = inputs_63_cast_fp16)[name = tensor("hidden_states_103_cast_fp16")]; + tensor var_8648 = const()[name = tensor("op_8648"), val = tensor([1, 1])]; + tensor var_8650 = const()[name = tensor("op_8650"), val = tensor([1, 1])]; + tensor q_43_pad_type_0 = const()[name = tensor("q_43_pad_type_0"), val = tensor("custom")]; + tensor q_43_pad_0 = const()[name = tensor("q_43_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_6_attn2_to_q_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_6_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(628364224)))]; + tensor q_43_cast_fp16 = conv(dilations = var_8650, groups = var_2624, pad = q_43_pad_0, pad_type = q_43_pad_type_0, strides = var_8648, weight = down_blocks_2_attentions_0_transformer_blocks_6_attn2_to_q_weight_to_fp16, x = hidden_states_103_cast_fp16)[name = tensor("q_43_cast_fp16")]; + tensor var_8654 = const()[name = tensor("op_8654"), val = tensor([1, 1])]; + tensor var_8656 = const()[name = tensor("op_8656"), val = tensor([1, 1])]; + tensor k_85_pad_type_0 = const()[name = tensor("k_85_pad_type_0"), val = tensor("custom")]; + tensor k_85_pad_0 = const()[name = tensor("k_85_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_6_attn2_to_k_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_6_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(631641088)))]; + tensor k_85_cast_fp16 = conv(dilations = var_8656, groups = var_2624, pad = k_85_pad_0, pad_type = k_85_pad_type_0, strides = var_8654, weight = down_blocks_2_attentions_0_transformer_blocks_6_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_85_cast_fp16")]; + tensor var_8660 = const()[name = tensor("op_8660"), val = tensor([1, 1])]; + tensor var_8662 = const()[name = tensor("op_8662"), val = tensor([1, 1])]; + tensor v_43_pad_type_0 = const()[name = tensor("v_43_pad_type_0"), val = tensor("custom")]; + tensor v_43_pad_0 = const()[name = tensor("v_43_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_6_attn2_to_v_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_6_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(636884032)))]; + tensor v_43_cast_fp16 = conv(dilations = var_8662, groups = var_2624, pad = v_43_pad_0, pad_type = v_43_pad_type_0, strides = var_8660, weight = down_blocks_2_attentions_0_transformer_blocks_6_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_43_cast_fp16")]; + tensor var_8666_begin_0 = const()[name = tensor("op_8666_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8666_end_0 = const()[name = tensor("op_8666_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_8666_end_mask_0 = const()[name = tensor("op_8666_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8666_cast_fp16 = slice_by_index(begin = var_8666_begin_0, end = var_8666_end_0, end_mask = var_8666_end_mask_0, x = q_43_cast_fp16)[name = tensor("op_8666_cast_fp16")]; + tensor var_8670_begin_0 = const()[name = tensor("op_8670_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_8670_end_0 = const()[name = tensor("op_8670_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_8670_end_mask_0 = const()[name = tensor("op_8670_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8670_cast_fp16 = slice_by_index(begin = var_8670_begin_0, end = var_8670_end_0, end_mask = var_8670_end_mask_0, x = q_43_cast_fp16)[name = tensor("op_8670_cast_fp16")]; + tensor var_8674_begin_0 = const()[name = tensor("op_8674_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_8674_end_0 = const()[name = tensor("op_8674_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_8674_end_mask_0 = const()[name = tensor("op_8674_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8674_cast_fp16 = slice_by_index(begin = var_8674_begin_0, end = var_8674_end_0, end_mask = var_8674_end_mask_0, x = q_43_cast_fp16)[name = tensor("op_8674_cast_fp16")]; + tensor var_8678_begin_0 = const()[name = tensor("op_8678_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_8678_end_0 = const()[name = tensor("op_8678_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_8678_end_mask_0 = const()[name = tensor("op_8678_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8678_cast_fp16 = slice_by_index(begin = var_8678_begin_0, end = var_8678_end_0, end_mask = var_8678_end_mask_0, x = q_43_cast_fp16)[name = tensor("op_8678_cast_fp16")]; + tensor var_8682_begin_0 = const()[name = tensor("op_8682_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_8682_end_0 = const()[name = tensor("op_8682_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_8682_end_mask_0 = const()[name = tensor("op_8682_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8682_cast_fp16 = slice_by_index(begin = var_8682_begin_0, end = var_8682_end_0, end_mask = var_8682_end_mask_0, x = q_43_cast_fp16)[name = tensor("op_8682_cast_fp16")]; + tensor var_8686_begin_0 = const()[name = tensor("op_8686_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_8686_end_0 = const()[name = tensor("op_8686_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_8686_end_mask_0 = const()[name = tensor("op_8686_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8686_cast_fp16 = slice_by_index(begin = var_8686_begin_0, end = var_8686_end_0, end_mask = var_8686_end_mask_0, x = q_43_cast_fp16)[name = tensor("op_8686_cast_fp16")]; + tensor var_8690_begin_0 = const()[name = tensor("op_8690_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_8690_end_0 = const()[name = tensor("op_8690_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_8690_end_mask_0 = const()[name = tensor("op_8690_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8690_cast_fp16 = slice_by_index(begin = var_8690_begin_0, end = var_8690_end_0, end_mask = var_8690_end_mask_0, x = q_43_cast_fp16)[name = tensor("op_8690_cast_fp16")]; + tensor var_8694_begin_0 = const()[name = tensor("op_8694_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_8694_end_0 = const()[name = tensor("op_8694_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_8694_end_mask_0 = const()[name = tensor("op_8694_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8694_cast_fp16 = slice_by_index(begin = var_8694_begin_0, end = var_8694_end_0, end_mask = var_8694_end_mask_0, x = q_43_cast_fp16)[name = tensor("op_8694_cast_fp16")]; + tensor var_8698_begin_0 = const()[name = tensor("op_8698_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_8698_end_0 = const()[name = tensor("op_8698_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_8698_end_mask_0 = const()[name = tensor("op_8698_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8698_cast_fp16 = slice_by_index(begin = var_8698_begin_0, end = var_8698_end_0, end_mask = var_8698_end_mask_0, x = q_43_cast_fp16)[name = tensor("op_8698_cast_fp16")]; + tensor var_8702_begin_0 = const()[name = tensor("op_8702_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_8702_end_0 = const()[name = tensor("op_8702_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_8702_end_mask_0 = const()[name = tensor("op_8702_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8702_cast_fp16 = slice_by_index(begin = var_8702_begin_0, end = var_8702_end_0, end_mask = var_8702_end_mask_0, x = q_43_cast_fp16)[name = tensor("op_8702_cast_fp16")]; + tensor var_8706_begin_0 = const()[name = tensor("op_8706_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_8706_end_0 = const()[name = tensor("op_8706_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_8706_end_mask_0 = const()[name = tensor("op_8706_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8706_cast_fp16 = slice_by_index(begin = var_8706_begin_0, end = var_8706_end_0, end_mask = var_8706_end_mask_0, x = q_43_cast_fp16)[name = tensor("op_8706_cast_fp16")]; + tensor var_8710_begin_0 = const()[name = tensor("op_8710_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_8710_end_0 = const()[name = tensor("op_8710_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_8710_end_mask_0 = const()[name = tensor("op_8710_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8710_cast_fp16 = slice_by_index(begin = var_8710_begin_0, end = var_8710_end_0, end_mask = var_8710_end_mask_0, x = q_43_cast_fp16)[name = tensor("op_8710_cast_fp16")]; + tensor var_8714_begin_0 = const()[name = tensor("op_8714_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_8714_end_0 = const()[name = tensor("op_8714_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_8714_end_mask_0 = const()[name = tensor("op_8714_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8714_cast_fp16 = slice_by_index(begin = var_8714_begin_0, end = var_8714_end_0, end_mask = var_8714_end_mask_0, x = q_43_cast_fp16)[name = tensor("op_8714_cast_fp16")]; + tensor var_8718_begin_0 = const()[name = tensor("op_8718_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_8718_end_0 = const()[name = tensor("op_8718_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_8718_end_mask_0 = const()[name = tensor("op_8718_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8718_cast_fp16 = slice_by_index(begin = var_8718_begin_0, end = var_8718_end_0, end_mask = var_8718_end_mask_0, x = q_43_cast_fp16)[name = tensor("op_8718_cast_fp16")]; + tensor var_8722_begin_0 = const()[name = tensor("op_8722_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_8722_end_0 = const()[name = tensor("op_8722_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_8722_end_mask_0 = const()[name = tensor("op_8722_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8722_cast_fp16 = slice_by_index(begin = var_8722_begin_0, end = var_8722_end_0, end_mask = var_8722_end_mask_0, x = q_43_cast_fp16)[name = tensor("op_8722_cast_fp16")]; + tensor var_8726_begin_0 = const()[name = tensor("op_8726_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_8726_end_0 = const()[name = tensor("op_8726_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_8726_end_mask_0 = const()[name = tensor("op_8726_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8726_cast_fp16 = slice_by_index(begin = var_8726_begin_0, end = var_8726_end_0, end_mask = var_8726_end_mask_0, x = q_43_cast_fp16)[name = tensor("op_8726_cast_fp16")]; + tensor var_8730_begin_0 = const()[name = tensor("op_8730_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_8730_end_0 = const()[name = tensor("op_8730_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_8730_end_mask_0 = const()[name = tensor("op_8730_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8730_cast_fp16 = slice_by_index(begin = var_8730_begin_0, end = var_8730_end_0, end_mask = var_8730_end_mask_0, x = q_43_cast_fp16)[name = tensor("op_8730_cast_fp16")]; + tensor var_8734_begin_0 = const()[name = tensor("op_8734_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_8734_end_0 = const()[name = tensor("op_8734_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_8734_end_mask_0 = const()[name = tensor("op_8734_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8734_cast_fp16 = slice_by_index(begin = var_8734_begin_0, end = var_8734_end_0, end_mask = var_8734_end_mask_0, x = q_43_cast_fp16)[name = tensor("op_8734_cast_fp16")]; + tensor var_8738_begin_0 = const()[name = tensor("op_8738_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_8738_end_0 = const()[name = tensor("op_8738_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_8738_end_mask_0 = const()[name = tensor("op_8738_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8738_cast_fp16 = slice_by_index(begin = var_8738_begin_0, end = var_8738_end_0, end_mask = var_8738_end_mask_0, x = q_43_cast_fp16)[name = tensor("op_8738_cast_fp16")]; + tensor var_8742_begin_0 = const()[name = tensor("op_8742_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_8742_end_0 = const()[name = tensor("op_8742_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_8742_end_mask_0 = const()[name = tensor("op_8742_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8742_cast_fp16 = slice_by_index(begin = var_8742_begin_0, end = var_8742_end_0, end_mask = var_8742_end_mask_0, x = q_43_cast_fp16)[name = tensor("op_8742_cast_fp16")]; + tensor k_87_perm_0 = const()[name = tensor("k_87_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_8749_begin_0 = const()[name = tensor("op_8749_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8749_end_0 = const()[name = tensor("op_8749_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_8749_end_mask_0 = const()[name = tensor("op_8749_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_87_cast_fp16 = transpose(perm = k_87_perm_0, x = k_85_cast_fp16)[name = tensor("transpose_118")]; + tensor var_8749_cast_fp16 = slice_by_index(begin = var_8749_begin_0, end = var_8749_end_0, end_mask = var_8749_end_mask_0, x = k_87_cast_fp16)[name = tensor("op_8749_cast_fp16")]; + tensor var_8753_begin_0 = const()[name = tensor("op_8753_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_8753_end_0 = const()[name = tensor("op_8753_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_8753_end_mask_0 = const()[name = tensor("op_8753_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8753_cast_fp16 = slice_by_index(begin = var_8753_begin_0, end = var_8753_end_0, end_mask = var_8753_end_mask_0, x = k_87_cast_fp16)[name = tensor("op_8753_cast_fp16")]; + tensor var_8757_begin_0 = const()[name = tensor("op_8757_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_8757_end_0 = const()[name = tensor("op_8757_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_8757_end_mask_0 = const()[name = tensor("op_8757_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8757_cast_fp16 = slice_by_index(begin = var_8757_begin_0, end = var_8757_end_0, end_mask = var_8757_end_mask_0, x = k_87_cast_fp16)[name = tensor("op_8757_cast_fp16")]; + tensor var_8761_begin_0 = const()[name = tensor("op_8761_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_8761_end_0 = const()[name = tensor("op_8761_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_8761_end_mask_0 = const()[name = tensor("op_8761_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8761_cast_fp16 = slice_by_index(begin = var_8761_begin_0, end = var_8761_end_0, end_mask = var_8761_end_mask_0, x = k_87_cast_fp16)[name = tensor("op_8761_cast_fp16")]; + tensor var_8765_begin_0 = const()[name = tensor("op_8765_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_8765_end_0 = const()[name = tensor("op_8765_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_8765_end_mask_0 = const()[name = tensor("op_8765_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8765_cast_fp16 = slice_by_index(begin = var_8765_begin_0, end = var_8765_end_0, end_mask = var_8765_end_mask_0, x = k_87_cast_fp16)[name = tensor("op_8765_cast_fp16")]; + tensor var_8769_begin_0 = const()[name = tensor("op_8769_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_8769_end_0 = const()[name = tensor("op_8769_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_8769_end_mask_0 = const()[name = tensor("op_8769_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8769_cast_fp16 = slice_by_index(begin = var_8769_begin_0, end = var_8769_end_0, end_mask = var_8769_end_mask_0, x = k_87_cast_fp16)[name = tensor("op_8769_cast_fp16")]; + tensor var_8773_begin_0 = const()[name = tensor("op_8773_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_8773_end_0 = const()[name = tensor("op_8773_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_8773_end_mask_0 = const()[name = tensor("op_8773_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8773_cast_fp16 = slice_by_index(begin = var_8773_begin_0, end = var_8773_end_0, end_mask = var_8773_end_mask_0, x = k_87_cast_fp16)[name = tensor("op_8773_cast_fp16")]; + tensor var_8777_begin_0 = const()[name = tensor("op_8777_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_8777_end_0 = const()[name = tensor("op_8777_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_8777_end_mask_0 = const()[name = tensor("op_8777_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8777_cast_fp16 = slice_by_index(begin = var_8777_begin_0, end = var_8777_end_0, end_mask = var_8777_end_mask_0, x = k_87_cast_fp16)[name = tensor("op_8777_cast_fp16")]; + tensor var_8781_begin_0 = const()[name = tensor("op_8781_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_8781_end_0 = const()[name = tensor("op_8781_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_8781_end_mask_0 = const()[name = tensor("op_8781_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8781_cast_fp16 = slice_by_index(begin = var_8781_begin_0, end = var_8781_end_0, end_mask = var_8781_end_mask_0, x = k_87_cast_fp16)[name = tensor("op_8781_cast_fp16")]; + tensor var_8785_begin_0 = const()[name = tensor("op_8785_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_8785_end_0 = const()[name = tensor("op_8785_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_8785_end_mask_0 = const()[name = tensor("op_8785_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8785_cast_fp16 = slice_by_index(begin = var_8785_begin_0, end = var_8785_end_0, end_mask = var_8785_end_mask_0, x = k_87_cast_fp16)[name = tensor("op_8785_cast_fp16")]; + tensor var_8789_begin_0 = const()[name = tensor("op_8789_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_8789_end_0 = const()[name = tensor("op_8789_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_8789_end_mask_0 = const()[name = tensor("op_8789_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8789_cast_fp16 = slice_by_index(begin = var_8789_begin_0, end = var_8789_end_0, end_mask = var_8789_end_mask_0, x = k_87_cast_fp16)[name = tensor("op_8789_cast_fp16")]; + tensor var_8793_begin_0 = const()[name = tensor("op_8793_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_8793_end_0 = const()[name = tensor("op_8793_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_8793_end_mask_0 = const()[name = tensor("op_8793_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8793_cast_fp16 = slice_by_index(begin = var_8793_begin_0, end = var_8793_end_0, end_mask = var_8793_end_mask_0, x = k_87_cast_fp16)[name = tensor("op_8793_cast_fp16")]; + tensor var_8797_begin_0 = const()[name = tensor("op_8797_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_8797_end_0 = const()[name = tensor("op_8797_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_8797_end_mask_0 = const()[name = tensor("op_8797_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8797_cast_fp16 = slice_by_index(begin = var_8797_begin_0, end = var_8797_end_0, end_mask = var_8797_end_mask_0, x = k_87_cast_fp16)[name = tensor("op_8797_cast_fp16")]; + tensor var_8801_begin_0 = const()[name = tensor("op_8801_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_8801_end_0 = const()[name = tensor("op_8801_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_8801_end_mask_0 = const()[name = tensor("op_8801_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8801_cast_fp16 = slice_by_index(begin = var_8801_begin_0, end = var_8801_end_0, end_mask = var_8801_end_mask_0, x = k_87_cast_fp16)[name = tensor("op_8801_cast_fp16")]; + tensor var_8805_begin_0 = const()[name = tensor("op_8805_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_8805_end_0 = const()[name = tensor("op_8805_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_8805_end_mask_0 = const()[name = tensor("op_8805_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8805_cast_fp16 = slice_by_index(begin = var_8805_begin_0, end = var_8805_end_0, end_mask = var_8805_end_mask_0, x = k_87_cast_fp16)[name = tensor("op_8805_cast_fp16")]; + tensor var_8809_begin_0 = const()[name = tensor("op_8809_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_8809_end_0 = const()[name = tensor("op_8809_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_8809_end_mask_0 = const()[name = tensor("op_8809_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8809_cast_fp16 = slice_by_index(begin = var_8809_begin_0, end = var_8809_end_0, end_mask = var_8809_end_mask_0, x = k_87_cast_fp16)[name = tensor("op_8809_cast_fp16")]; + tensor var_8813_begin_0 = const()[name = tensor("op_8813_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_8813_end_0 = const()[name = tensor("op_8813_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_8813_end_mask_0 = const()[name = tensor("op_8813_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8813_cast_fp16 = slice_by_index(begin = var_8813_begin_0, end = var_8813_end_0, end_mask = var_8813_end_mask_0, x = k_87_cast_fp16)[name = tensor("op_8813_cast_fp16")]; + tensor var_8817_begin_0 = const()[name = tensor("op_8817_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_8817_end_0 = const()[name = tensor("op_8817_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_8817_end_mask_0 = const()[name = tensor("op_8817_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8817_cast_fp16 = slice_by_index(begin = var_8817_begin_0, end = var_8817_end_0, end_mask = var_8817_end_mask_0, x = k_87_cast_fp16)[name = tensor("op_8817_cast_fp16")]; + tensor var_8821_begin_0 = const()[name = tensor("op_8821_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_8821_end_0 = const()[name = tensor("op_8821_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_8821_end_mask_0 = const()[name = tensor("op_8821_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8821_cast_fp16 = slice_by_index(begin = var_8821_begin_0, end = var_8821_end_0, end_mask = var_8821_end_mask_0, x = k_87_cast_fp16)[name = tensor("op_8821_cast_fp16")]; + tensor var_8825_begin_0 = const()[name = tensor("op_8825_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_8825_end_0 = const()[name = tensor("op_8825_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_8825_end_mask_0 = const()[name = tensor("op_8825_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8825_cast_fp16 = slice_by_index(begin = var_8825_begin_0, end = var_8825_end_0, end_mask = var_8825_end_mask_0, x = k_87_cast_fp16)[name = tensor("op_8825_cast_fp16")]; + tensor var_8827_begin_0 = const()[name = tensor("op_8827_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8827_end_0 = const()[name = tensor("op_8827_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_8827_end_mask_0 = const()[name = tensor("op_8827_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8827_cast_fp16 = slice_by_index(begin = var_8827_begin_0, end = var_8827_end_0, end_mask = var_8827_end_mask_0, x = v_43_cast_fp16)[name = tensor("op_8827_cast_fp16")]; + tensor var_8831_begin_0 = const()[name = tensor("op_8831_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_8831_end_0 = const()[name = tensor("op_8831_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_8831_end_mask_0 = const()[name = tensor("op_8831_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8831_cast_fp16 = slice_by_index(begin = var_8831_begin_0, end = var_8831_end_0, end_mask = var_8831_end_mask_0, x = v_43_cast_fp16)[name = tensor("op_8831_cast_fp16")]; + tensor var_8835_begin_0 = const()[name = tensor("op_8835_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_8835_end_0 = const()[name = tensor("op_8835_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_8835_end_mask_0 = const()[name = tensor("op_8835_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8835_cast_fp16 = slice_by_index(begin = var_8835_begin_0, end = var_8835_end_0, end_mask = var_8835_end_mask_0, x = v_43_cast_fp16)[name = tensor("op_8835_cast_fp16")]; + tensor var_8839_begin_0 = const()[name = tensor("op_8839_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_8839_end_0 = const()[name = tensor("op_8839_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_8839_end_mask_0 = const()[name = tensor("op_8839_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8839_cast_fp16 = slice_by_index(begin = var_8839_begin_0, end = var_8839_end_0, end_mask = var_8839_end_mask_0, x = v_43_cast_fp16)[name = tensor("op_8839_cast_fp16")]; + tensor var_8843_begin_0 = const()[name = tensor("op_8843_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_8843_end_0 = const()[name = tensor("op_8843_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_8843_end_mask_0 = const()[name = tensor("op_8843_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8843_cast_fp16 = slice_by_index(begin = var_8843_begin_0, end = var_8843_end_0, end_mask = var_8843_end_mask_0, x = v_43_cast_fp16)[name = tensor("op_8843_cast_fp16")]; + tensor var_8847_begin_0 = const()[name = tensor("op_8847_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_8847_end_0 = const()[name = tensor("op_8847_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_8847_end_mask_0 = const()[name = tensor("op_8847_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8847_cast_fp16 = slice_by_index(begin = var_8847_begin_0, end = var_8847_end_0, end_mask = var_8847_end_mask_0, x = v_43_cast_fp16)[name = tensor("op_8847_cast_fp16")]; + tensor var_8851_begin_0 = const()[name = tensor("op_8851_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_8851_end_0 = const()[name = tensor("op_8851_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_8851_end_mask_0 = const()[name = tensor("op_8851_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8851_cast_fp16 = slice_by_index(begin = var_8851_begin_0, end = var_8851_end_0, end_mask = var_8851_end_mask_0, x = v_43_cast_fp16)[name = tensor("op_8851_cast_fp16")]; + tensor var_8855_begin_0 = const()[name = tensor("op_8855_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_8855_end_0 = const()[name = tensor("op_8855_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_8855_end_mask_0 = const()[name = tensor("op_8855_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8855_cast_fp16 = slice_by_index(begin = var_8855_begin_0, end = var_8855_end_0, end_mask = var_8855_end_mask_0, x = v_43_cast_fp16)[name = tensor("op_8855_cast_fp16")]; + tensor var_8859_begin_0 = const()[name = tensor("op_8859_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_8859_end_0 = const()[name = tensor("op_8859_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_8859_end_mask_0 = const()[name = tensor("op_8859_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8859_cast_fp16 = slice_by_index(begin = var_8859_begin_0, end = var_8859_end_0, end_mask = var_8859_end_mask_0, x = v_43_cast_fp16)[name = tensor("op_8859_cast_fp16")]; + tensor var_8863_begin_0 = const()[name = tensor("op_8863_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_8863_end_0 = const()[name = tensor("op_8863_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_8863_end_mask_0 = const()[name = tensor("op_8863_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8863_cast_fp16 = slice_by_index(begin = var_8863_begin_0, end = var_8863_end_0, end_mask = var_8863_end_mask_0, x = v_43_cast_fp16)[name = tensor("op_8863_cast_fp16")]; + tensor var_8867_begin_0 = const()[name = tensor("op_8867_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_8867_end_0 = const()[name = tensor("op_8867_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_8867_end_mask_0 = const()[name = tensor("op_8867_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8867_cast_fp16 = slice_by_index(begin = var_8867_begin_0, end = var_8867_end_0, end_mask = var_8867_end_mask_0, x = v_43_cast_fp16)[name = tensor("op_8867_cast_fp16")]; + tensor var_8871_begin_0 = const()[name = tensor("op_8871_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_8871_end_0 = const()[name = tensor("op_8871_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_8871_end_mask_0 = const()[name = tensor("op_8871_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8871_cast_fp16 = slice_by_index(begin = var_8871_begin_0, end = var_8871_end_0, end_mask = var_8871_end_mask_0, x = v_43_cast_fp16)[name = tensor("op_8871_cast_fp16")]; + tensor var_8875_begin_0 = const()[name = tensor("op_8875_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_8875_end_0 = const()[name = tensor("op_8875_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_8875_end_mask_0 = const()[name = tensor("op_8875_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8875_cast_fp16 = slice_by_index(begin = var_8875_begin_0, end = var_8875_end_0, end_mask = var_8875_end_mask_0, x = v_43_cast_fp16)[name = tensor("op_8875_cast_fp16")]; + tensor var_8879_begin_0 = const()[name = tensor("op_8879_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_8879_end_0 = const()[name = tensor("op_8879_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_8879_end_mask_0 = const()[name = tensor("op_8879_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8879_cast_fp16 = slice_by_index(begin = var_8879_begin_0, end = var_8879_end_0, end_mask = var_8879_end_mask_0, x = v_43_cast_fp16)[name = tensor("op_8879_cast_fp16")]; + tensor var_8883_begin_0 = const()[name = tensor("op_8883_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_8883_end_0 = const()[name = tensor("op_8883_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_8883_end_mask_0 = const()[name = tensor("op_8883_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8883_cast_fp16 = slice_by_index(begin = var_8883_begin_0, end = var_8883_end_0, end_mask = var_8883_end_mask_0, x = v_43_cast_fp16)[name = tensor("op_8883_cast_fp16")]; + tensor var_8887_begin_0 = const()[name = tensor("op_8887_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_8887_end_0 = const()[name = tensor("op_8887_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_8887_end_mask_0 = const()[name = tensor("op_8887_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8887_cast_fp16 = slice_by_index(begin = var_8887_begin_0, end = var_8887_end_0, end_mask = var_8887_end_mask_0, x = v_43_cast_fp16)[name = tensor("op_8887_cast_fp16")]; + tensor var_8891_begin_0 = const()[name = tensor("op_8891_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_8891_end_0 = const()[name = tensor("op_8891_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_8891_end_mask_0 = const()[name = tensor("op_8891_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8891_cast_fp16 = slice_by_index(begin = var_8891_begin_0, end = var_8891_end_0, end_mask = var_8891_end_mask_0, x = v_43_cast_fp16)[name = tensor("op_8891_cast_fp16")]; + tensor var_8895_begin_0 = const()[name = tensor("op_8895_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_8895_end_0 = const()[name = tensor("op_8895_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_8895_end_mask_0 = const()[name = tensor("op_8895_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8895_cast_fp16 = slice_by_index(begin = var_8895_begin_0, end = var_8895_end_0, end_mask = var_8895_end_mask_0, x = v_43_cast_fp16)[name = tensor("op_8895_cast_fp16")]; + tensor var_8899_begin_0 = const()[name = tensor("op_8899_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_8899_end_0 = const()[name = tensor("op_8899_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_8899_end_mask_0 = const()[name = tensor("op_8899_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8899_cast_fp16 = slice_by_index(begin = var_8899_begin_0, end = var_8899_end_0, end_mask = var_8899_end_mask_0, x = v_43_cast_fp16)[name = tensor("op_8899_cast_fp16")]; + tensor var_8903_begin_0 = const()[name = tensor("op_8903_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_8903_end_0 = const()[name = tensor("op_8903_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_8903_end_mask_0 = const()[name = tensor("op_8903_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8903_cast_fp16 = slice_by_index(begin = var_8903_begin_0, end = var_8903_end_0, end_mask = var_8903_end_mask_0, x = v_43_cast_fp16)[name = tensor("op_8903_cast_fp16")]; + tensor var_8907_equation_0 = const()[name = tensor("op_8907_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8907_cast_fp16 = einsum(equation = var_8907_equation_0, values = (var_8749_cast_fp16, var_8666_cast_fp16))[name = tensor("op_8907_cast_fp16")]; + tensor var_8908_to_fp16 = const()[name = tensor("op_8908_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_681_cast_fp16 = mul(x = var_8907_cast_fp16, y = var_8908_to_fp16)[name = tensor("aw_681_cast_fp16")]; + tensor var_8911_equation_0 = const()[name = tensor("op_8911_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8911_cast_fp16 = einsum(equation = var_8911_equation_0, values = (var_8753_cast_fp16, var_8670_cast_fp16))[name = tensor("op_8911_cast_fp16")]; + tensor var_8912_to_fp16 = const()[name = tensor("op_8912_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_683_cast_fp16 = mul(x = var_8911_cast_fp16, y = var_8912_to_fp16)[name = tensor("aw_683_cast_fp16")]; + tensor var_8915_equation_0 = const()[name = tensor("op_8915_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8915_cast_fp16 = einsum(equation = var_8915_equation_0, values = (var_8757_cast_fp16, var_8674_cast_fp16))[name = tensor("op_8915_cast_fp16")]; + tensor var_8916_to_fp16 = const()[name = tensor("op_8916_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_685_cast_fp16 = mul(x = var_8915_cast_fp16, y = var_8916_to_fp16)[name = tensor("aw_685_cast_fp16")]; + tensor var_8919_equation_0 = const()[name = tensor("op_8919_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8919_cast_fp16 = einsum(equation = var_8919_equation_0, values = (var_8761_cast_fp16, var_8678_cast_fp16))[name = tensor("op_8919_cast_fp16")]; + tensor var_8920_to_fp16 = const()[name = tensor("op_8920_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_687_cast_fp16 = mul(x = var_8919_cast_fp16, y = var_8920_to_fp16)[name = tensor("aw_687_cast_fp16")]; + tensor var_8923_equation_0 = const()[name = tensor("op_8923_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8923_cast_fp16 = einsum(equation = var_8923_equation_0, values = (var_8765_cast_fp16, var_8682_cast_fp16))[name = tensor("op_8923_cast_fp16")]; + tensor var_8924_to_fp16 = const()[name = tensor("op_8924_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_689_cast_fp16 = mul(x = var_8923_cast_fp16, y = var_8924_to_fp16)[name = tensor("aw_689_cast_fp16")]; + tensor var_8927_equation_0 = const()[name = tensor("op_8927_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8927_cast_fp16 = einsum(equation = var_8927_equation_0, values = (var_8769_cast_fp16, var_8686_cast_fp16))[name = tensor("op_8927_cast_fp16")]; + tensor var_8928_to_fp16 = const()[name = tensor("op_8928_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_691_cast_fp16 = mul(x = var_8927_cast_fp16, y = var_8928_to_fp16)[name = tensor("aw_691_cast_fp16")]; + tensor var_8931_equation_0 = const()[name = tensor("op_8931_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8931_cast_fp16 = einsum(equation = var_8931_equation_0, values = (var_8773_cast_fp16, var_8690_cast_fp16))[name = tensor("op_8931_cast_fp16")]; + tensor var_8932_to_fp16 = const()[name = tensor("op_8932_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_693_cast_fp16 = mul(x = var_8931_cast_fp16, y = var_8932_to_fp16)[name = tensor("aw_693_cast_fp16")]; + tensor var_8935_equation_0 = const()[name = tensor("op_8935_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8935_cast_fp16 = einsum(equation = var_8935_equation_0, values = (var_8777_cast_fp16, var_8694_cast_fp16))[name = tensor("op_8935_cast_fp16")]; + tensor var_8936_to_fp16 = const()[name = tensor("op_8936_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_695_cast_fp16 = mul(x = var_8935_cast_fp16, y = var_8936_to_fp16)[name = tensor("aw_695_cast_fp16")]; + tensor var_8939_equation_0 = const()[name = tensor("op_8939_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8939_cast_fp16 = einsum(equation = var_8939_equation_0, values = (var_8781_cast_fp16, var_8698_cast_fp16))[name = tensor("op_8939_cast_fp16")]; + tensor var_8940_to_fp16 = const()[name = tensor("op_8940_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_697_cast_fp16 = mul(x = var_8939_cast_fp16, y = var_8940_to_fp16)[name = tensor("aw_697_cast_fp16")]; + tensor var_8943_equation_0 = const()[name = tensor("op_8943_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8943_cast_fp16 = einsum(equation = var_8943_equation_0, values = (var_8785_cast_fp16, var_8702_cast_fp16))[name = tensor("op_8943_cast_fp16")]; + tensor var_8944_to_fp16 = const()[name = tensor("op_8944_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_699_cast_fp16 = mul(x = var_8943_cast_fp16, y = var_8944_to_fp16)[name = tensor("aw_699_cast_fp16")]; + tensor var_8947_equation_0 = const()[name = tensor("op_8947_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8947_cast_fp16 = einsum(equation = var_8947_equation_0, values = (var_8789_cast_fp16, var_8706_cast_fp16))[name = tensor("op_8947_cast_fp16")]; + tensor var_8948_to_fp16 = const()[name = tensor("op_8948_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_701_cast_fp16 = mul(x = var_8947_cast_fp16, y = var_8948_to_fp16)[name = tensor("aw_701_cast_fp16")]; + tensor var_8951_equation_0 = const()[name = tensor("op_8951_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8951_cast_fp16 = einsum(equation = var_8951_equation_0, values = (var_8793_cast_fp16, var_8710_cast_fp16))[name = tensor("op_8951_cast_fp16")]; + tensor var_8952_to_fp16 = const()[name = tensor("op_8952_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_703_cast_fp16 = mul(x = var_8951_cast_fp16, y = var_8952_to_fp16)[name = tensor("aw_703_cast_fp16")]; + tensor var_8955_equation_0 = const()[name = tensor("op_8955_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8955_cast_fp16 = einsum(equation = var_8955_equation_0, values = (var_8797_cast_fp16, var_8714_cast_fp16))[name = tensor("op_8955_cast_fp16")]; + tensor var_8956_to_fp16 = const()[name = tensor("op_8956_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_705_cast_fp16 = mul(x = var_8955_cast_fp16, y = var_8956_to_fp16)[name = tensor("aw_705_cast_fp16")]; + tensor var_8959_equation_0 = const()[name = tensor("op_8959_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8959_cast_fp16 = einsum(equation = var_8959_equation_0, values = (var_8801_cast_fp16, var_8718_cast_fp16))[name = tensor("op_8959_cast_fp16")]; + tensor var_8960_to_fp16 = const()[name = tensor("op_8960_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_707_cast_fp16 = mul(x = var_8959_cast_fp16, y = var_8960_to_fp16)[name = tensor("aw_707_cast_fp16")]; + tensor var_8963_equation_0 = const()[name = tensor("op_8963_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8963_cast_fp16 = einsum(equation = var_8963_equation_0, values = (var_8805_cast_fp16, var_8722_cast_fp16))[name = tensor("op_8963_cast_fp16")]; + tensor var_8964_to_fp16 = const()[name = tensor("op_8964_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_709_cast_fp16 = mul(x = var_8963_cast_fp16, y = var_8964_to_fp16)[name = tensor("aw_709_cast_fp16")]; + tensor var_8967_equation_0 = const()[name = tensor("op_8967_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8967_cast_fp16 = einsum(equation = var_8967_equation_0, values = (var_8809_cast_fp16, var_8726_cast_fp16))[name = tensor("op_8967_cast_fp16")]; + tensor var_8968_to_fp16 = const()[name = tensor("op_8968_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_711_cast_fp16 = mul(x = var_8967_cast_fp16, y = var_8968_to_fp16)[name = tensor("aw_711_cast_fp16")]; + tensor var_8971_equation_0 = const()[name = tensor("op_8971_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8971_cast_fp16 = einsum(equation = var_8971_equation_0, values = (var_8813_cast_fp16, var_8730_cast_fp16))[name = tensor("op_8971_cast_fp16")]; + tensor var_8972_to_fp16 = const()[name = tensor("op_8972_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_713_cast_fp16 = mul(x = var_8971_cast_fp16, y = var_8972_to_fp16)[name = tensor("aw_713_cast_fp16")]; + tensor var_8975_equation_0 = const()[name = tensor("op_8975_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8975_cast_fp16 = einsum(equation = var_8975_equation_0, values = (var_8817_cast_fp16, var_8734_cast_fp16))[name = tensor("op_8975_cast_fp16")]; + tensor var_8976_to_fp16 = const()[name = tensor("op_8976_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_715_cast_fp16 = mul(x = var_8975_cast_fp16, y = var_8976_to_fp16)[name = tensor("aw_715_cast_fp16")]; + tensor var_8979_equation_0 = const()[name = tensor("op_8979_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8979_cast_fp16 = einsum(equation = var_8979_equation_0, values = (var_8821_cast_fp16, var_8738_cast_fp16))[name = tensor("op_8979_cast_fp16")]; + tensor var_8980_to_fp16 = const()[name = tensor("op_8980_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_717_cast_fp16 = mul(x = var_8979_cast_fp16, y = var_8980_to_fp16)[name = tensor("aw_717_cast_fp16")]; + tensor var_8983_equation_0 = const()[name = tensor("op_8983_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8983_cast_fp16 = einsum(equation = var_8983_equation_0, values = (var_8825_cast_fp16, var_8742_cast_fp16))[name = tensor("op_8983_cast_fp16")]; + tensor var_8984_to_fp16 = const()[name = tensor("op_8984_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_719_cast_fp16 = mul(x = var_8983_cast_fp16, y = var_8984_to_fp16)[name = tensor("aw_719_cast_fp16")]; + tensor var_8986_cast_fp16 = softmax(axis = var_2624, x = aw_681_cast_fp16)[name = tensor("op_8986_cast_fp16")]; + tensor var_8987_cast_fp16 = softmax(axis = var_2624, x = aw_683_cast_fp16)[name = tensor("op_8987_cast_fp16")]; + tensor var_8988_cast_fp16 = softmax(axis = var_2624, x = aw_685_cast_fp16)[name = tensor("op_8988_cast_fp16")]; + tensor var_8989_cast_fp16 = softmax(axis = var_2624, x = aw_687_cast_fp16)[name = tensor("op_8989_cast_fp16")]; + tensor var_8990_cast_fp16 = softmax(axis = var_2624, x = aw_689_cast_fp16)[name = tensor("op_8990_cast_fp16")]; + tensor var_8991_cast_fp16 = softmax(axis = var_2624, x = aw_691_cast_fp16)[name = tensor("op_8991_cast_fp16")]; + tensor var_8992_cast_fp16 = softmax(axis = var_2624, x = aw_693_cast_fp16)[name = tensor("op_8992_cast_fp16")]; + tensor var_8993_cast_fp16 = softmax(axis = var_2624, x = aw_695_cast_fp16)[name = tensor("op_8993_cast_fp16")]; + tensor var_8994_cast_fp16 = softmax(axis = var_2624, x = aw_697_cast_fp16)[name = tensor("op_8994_cast_fp16")]; + tensor var_8995_cast_fp16 = softmax(axis = var_2624, x = aw_699_cast_fp16)[name = tensor("op_8995_cast_fp16")]; + tensor var_8996_cast_fp16 = softmax(axis = var_2624, x = aw_701_cast_fp16)[name = tensor("op_8996_cast_fp16")]; + tensor var_8997_cast_fp16 = softmax(axis = var_2624, x = aw_703_cast_fp16)[name = tensor("op_8997_cast_fp16")]; + tensor var_8998_cast_fp16 = softmax(axis = var_2624, x = aw_705_cast_fp16)[name = tensor("op_8998_cast_fp16")]; + tensor var_8999_cast_fp16 = softmax(axis = var_2624, x = aw_707_cast_fp16)[name = tensor("op_8999_cast_fp16")]; + tensor var_9000_cast_fp16 = softmax(axis = var_2624, x = aw_709_cast_fp16)[name = tensor("op_9000_cast_fp16")]; + tensor var_9001_cast_fp16 = softmax(axis = var_2624, x = aw_711_cast_fp16)[name = tensor("op_9001_cast_fp16")]; + tensor var_9002_cast_fp16 = softmax(axis = var_2624, x = aw_713_cast_fp16)[name = tensor("op_9002_cast_fp16")]; + tensor var_9003_cast_fp16 = softmax(axis = var_2624, x = aw_715_cast_fp16)[name = tensor("op_9003_cast_fp16")]; + tensor var_9004_cast_fp16 = softmax(axis = var_2624, x = aw_717_cast_fp16)[name = tensor("op_9004_cast_fp16")]; + tensor var_9005_cast_fp16 = softmax(axis = var_2624, x = aw_719_cast_fp16)[name = tensor("op_9005_cast_fp16")]; + tensor var_9007_equation_0 = const()[name = tensor("op_9007_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9007_cast_fp16 = einsum(equation = var_9007_equation_0, values = (var_8827_cast_fp16, var_8986_cast_fp16))[name = tensor("op_9007_cast_fp16")]; + tensor var_9009_equation_0 = const()[name = tensor("op_9009_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9009_cast_fp16 = einsum(equation = var_9009_equation_0, values = (var_8831_cast_fp16, var_8987_cast_fp16))[name = tensor("op_9009_cast_fp16")]; + tensor var_9011_equation_0 = const()[name = tensor("op_9011_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9011_cast_fp16 = einsum(equation = var_9011_equation_0, values = (var_8835_cast_fp16, var_8988_cast_fp16))[name = tensor("op_9011_cast_fp16")]; + tensor var_9013_equation_0 = const()[name = tensor("op_9013_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9013_cast_fp16 = einsum(equation = var_9013_equation_0, values = (var_8839_cast_fp16, var_8989_cast_fp16))[name = tensor("op_9013_cast_fp16")]; + tensor var_9015_equation_0 = const()[name = tensor("op_9015_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9015_cast_fp16 = einsum(equation = var_9015_equation_0, values = (var_8843_cast_fp16, var_8990_cast_fp16))[name = tensor("op_9015_cast_fp16")]; + tensor var_9017_equation_0 = const()[name = tensor("op_9017_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9017_cast_fp16 = einsum(equation = var_9017_equation_0, values = (var_8847_cast_fp16, var_8991_cast_fp16))[name = tensor("op_9017_cast_fp16")]; + tensor var_9019_equation_0 = const()[name = tensor("op_9019_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9019_cast_fp16 = einsum(equation = var_9019_equation_0, values = (var_8851_cast_fp16, var_8992_cast_fp16))[name = tensor("op_9019_cast_fp16")]; + tensor var_9021_equation_0 = const()[name = tensor("op_9021_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9021_cast_fp16 = einsum(equation = var_9021_equation_0, values = (var_8855_cast_fp16, var_8993_cast_fp16))[name = tensor("op_9021_cast_fp16")]; + tensor var_9023_equation_0 = const()[name = tensor("op_9023_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9023_cast_fp16 = einsum(equation = var_9023_equation_0, values = (var_8859_cast_fp16, var_8994_cast_fp16))[name = tensor("op_9023_cast_fp16")]; + tensor var_9025_equation_0 = const()[name = tensor("op_9025_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9025_cast_fp16 = einsum(equation = var_9025_equation_0, values = (var_8863_cast_fp16, var_8995_cast_fp16))[name = tensor("op_9025_cast_fp16")]; + tensor var_9027_equation_0 = const()[name = tensor("op_9027_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9027_cast_fp16 = einsum(equation = var_9027_equation_0, values = (var_8867_cast_fp16, var_8996_cast_fp16))[name = tensor("op_9027_cast_fp16")]; + tensor var_9029_equation_0 = const()[name = tensor("op_9029_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9029_cast_fp16 = einsum(equation = var_9029_equation_0, values = (var_8871_cast_fp16, var_8997_cast_fp16))[name = tensor("op_9029_cast_fp16")]; + tensor var_9031_equation_0 = const()[name = tensor("op_9031_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9031_cast_fp16 = einsum(equation = var_9031_equation_0, values = (var_8875_cast_fp16, var_8998_cast_fp16))[name = tensor("op_9031_cast_fp16")]; + tensor var_9033_equation_0 = const()[name = tensor("op_9033_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9033_cast_fp16 = einsum(equation = var_9033_equation_0, values = (var_8879_cast_fp16, var_8999_cast_fp16))[name = tensor("op_9033_cast_fp16")]; + tensor var_9035_equation_0 = const()[name = tensor("op_9035_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9035_cast_fp16 = einsum(equation = var_9035_equation_0, values = (var_8883_cast_fp16, var_9000_cast_fp16))[name = tensor("op_9035_cast_fp16")]; + tensor var_9037_equation_0 = const()[name = tensor("op_9037_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9037_cast_fp16 = einsum(equation = var_9037_equation_0, values = (var_8887_cast_fp16, var_9001_cast_fp16))[name = tensor("op_9037_cast_fp16")]; + tensor var_9039_equation_0 = const()[name = tensor("op_9039_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9039_cast_fp16 = einsum(equation = var_9039_equation_0, values = (var_8891_cast_fp16, var_9002_cast_fp16))[name = tensor("op_9039_cast_fp16")]; + tensor var_9041_equation_0 = const()[name = tensor("op_9041_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9041_cast_fp16 = einsum(equation = var_9041_equation_0, values = (var_8895_cast_fp16, var_9003_cast_fp16))[name = tensor("op_9041_cast_fp16")]; + tensor var_9043_equation_0 = const()[name = tensor("op_9043_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9043_cast_fp16 = einsum(equation = var_9043_equation_0, values = (var_8899_cast_fp16, var_9004_cast_fp16))[name = tensor("op_9043_cast_fp16")]; + tensor var_9045_equation_0 = const()[name = tensor("op_9045_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9045_cast_fp16 = einsum(equation = var_9045_equation_0, values = (var_8903_cast_fp16, var_9005_cast_fp16))[name = tensor("op_9045_cast_fp16")]; + tensor input_181_interleave_0 = const()[name = tensor("input_181_interleave_0"), val = tensor(false)]; + tensor input_181_cast_fp16 = concat(axis = var_2624, interleave = input_181_interleave_0, values = (var_9007_cast_fp16, var_9009_cast_fp16, var_9011_cast_fp16, var_9013_cast_fp16, var_9015_cast_fp16, var_9017_cast_fp16, var_9019_cast_fp16, var_9021_cast_fp16, var_9023_cast_fp16, var_9025_cast_fp16, var_9027_cast_fp16, var_9029_cast_fp16, var_9031_cast_fp16, var_9033_cast_fp16, var_9035_cast_fp16, var_9037_cast_fp16, var_9039_cast_fp16, var_9041_cast_fp16, var_9043_cast_fp16, var_9045_cast_fp16))[name = tensor("input_181_cast_fp16")]; + tensor var_9051 = const()[name = tensor("op_9051"), val = tensor([1, 1])]; + tensor var_9053 = const()[name = tensor("op_9053"), val = tensor([1, 1])]; + tensor var_9055_pad_type_0 = const()[name = tensor("op_9055_pad_type_0"), val = tensor("custom")]; + tensor var_9055_pad_0 = const()[name = tensor("op_9055_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_6_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_6_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(642126976)))]; + tensor down_blocks_2_attentions_0_transformer_blocks_6_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_6_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(645403840)))]; + tensor var_9055_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_6_attn2_to_out_0_bias_to_fp16, dilations = var_9053, groups = var_2624, pad = var_9055_pad_0, pad_type = var_9055_pad_type_0, strides = var_9051, weight = down_blocks_2_attentions_0_transformer_blocks_6_attn2_to_out_0_weight_to_fp16, x = input_181_cast_fp16)[name = tensor("op_9055_cast_fp16")]; + tensor inputs_65_cast_fp16 = add(x = var_9055_cast_fp16, y = inputs_63_cast_fp16)[name = tensor("inputs_65_cast_fp16")]; + tensor input_183_axes_0 = const()[name = tensor("input_183_axes_0"), val = tensor([1])]; + tensor input_183_gamma_0_to_fp16 = const()[name = tensor("input_183_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(645406464)))]; + tensor input_183_beta_0_to_fp16 = const()[name = tensor("input_183_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(645409088)))]; + tensor var_9065_to_fp16 = const()[name = tensor("op_9065_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_183_cast_fp16 = layer_norm(axes = input_183_axes_0, beta = input_183_beta_0_to_fp16, epsilon = var_9065_to_fp16, gamma = input_183_gamma_0_to_fp16, x = inputs_65_cast_fp16)[name = tensor("input_183_cast_fp16")]; + tensor var_9081 = const()[name = tensor("op_9081"), val = tensor([1, 1])]; + tensor var_9083 = const()[name = tensor("op_9083"), val = tensor([1, 1])]; + tensor var_9085_pad_type_0 = const()[name = tensor("op_9085_pad_type_0"), val = tensor("custom")]; + tensor var_9085_pad_0 = const()[name = tensor("op_9085_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_6_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_6_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(645411712)))]; + tensor down_blocks_2_attentions_0_transformer_blocks_6_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_6_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(671626176)))]; + tensor var_9085_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_6_ff_net_0_proj_bias_to_fp16, dilations = var_9083, groups = var_2624, pad = var_9085_pad_0, pad_type = var_9085_pad_type_0, strides = var_9081, weight = down_blocks_2_attentions_0_transformer_blocks_6_ff_net_0_proj_weight_to_fp16, x = input_183_cast_fp16)[name = tensor("op_9085_cast_fp16")]; + tensor var_9086_split_sizes_0 = const()[name = tensor("op_9086_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_9086_axis_0 = const()[name = tensor("op_9086_axis_0"), val = tensor(1)]; + tensor var_9086_cast_fp16_0, tensor var_9086_cast_fp16_1 = split(axis = var_9086_axis_0, split_sizes = var_9086_split_sizes_0, x = var_9085_cast_fp16)[name = tensor("op_9086_cast_fp16")]; + tensor var_9088_mode_0 = const()[name = tensor("op_9088_mode_0"), val = tensor("EXACT")]; + tensor var_9088_cast_fp16 = gelu(mode = var_9088_mode_0, x = var_9086_cast_fp16_1)[name = tensor("op_9088_cast_fp16")]; + tensor input_185_cast_fp16 = mul(x = var_9086_cast_fp16_0, y = var_9088_cast_fp16)[name = tensor("input_185_cast_fp16")]; + tensor var_9092 = const()[name = tensor("op_9092"), val = tensor([1, 1])]; + tensor var_9094 = const()[name = tensor("op_9094"), val = tensor([1, 1])]; + tensor var_9096_pad_type_0 = const()[name = tensor("op_9096_pad_type_0"), val = tensor("custom")]; + tensor var_9096_pad_0 = const()[name = tensor("op_9096_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_6_ff_net_2_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_6_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(671646720)))]; + tensor down_blocks_2_attentions_0_transformer_blocks_6_ff_net_2_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_6_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(684753984)))]; + tensor var_9096_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_6_ff_net_2_bias_to_fp16, dilations = var_9094, groups = var_2624, pad = var_9096_pad_0, pad_type = var_9096_pad_type_0, strides = var_9092, weight = down_blocks_2_attentions_0_transformer_blocks_6_ff_net_2_weight_to_fp16, x = input_185_cast_fp16)[name = tensor("op_9096_cast_fp16")]; + tensor inputs_67_cast_fp16 = add(x = var_9096_cast_fp16, y = inputs_65_cast_fp16)[name = tensor("inputs_67_cast_fp16")]; + tensor hidden_states_107_axes_0 = const()[name = tensor("hidden_states_107_axes_0"), val = tensor([1])]; + tensor hidden_states_107_gamma_0_to_fp16 = const()[name = tensor("hidden_states_107_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(684756608)))]; + tensor hidden_states_107_beta_0_to_fp16 = const()[name = tensor("hidden_states_107_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(684759232)))]; + tensor var_9112_to_fp16 = const()[name = tensor("op_9112_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_107_cast_fp16 = layer_norm(axes = hidden_states_107_axes_0, beta = hidden_states_107_beta_0_to_fp16, epsilon = var_9112_to_fp16, gamma = hidden_states_107_gamma_0_to_fp16, x = inputs_67_cast_fp16)[name = tensor("hidden_states_107_cast_fp16")]; + tensor var_9127 = const()[name = tensor("op_9127"), val = tensor([1, 1])]; + tensor var_9129 = const()[name = tensor("op_9129"), val = tensor([1, 1])]; + tensor q_45_pad_type_0 = const()[name = tensor("q_45_pad_type_0"), val = tensor("custom")]; + tensor q_45_pad_0 = const()[name = tensor("q_45_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_7_attn1_to_q_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_7_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(684761856)))]; + tensor q_45_cast_fp16 = conv(dilations = var_9129, groups = var_2624, pad = q_45_pad_0, pad_type = q_45_pad_type_0, strides = var_9127, weight = down_blocks_2_attentions_0_transformer_blocks_7_attn1_to_q_weight_to_fp16, x = hidden_states_107_cast_fp16)[name = tensor("q_45_cast_fp16")]; + tensor var_9133 = const()[name = tensor("op_9133"), val = tensor([1, 1])]; + tensor var_9135 = const()[name = tensor("op_9135"), val = tensor([1, 1])]; + tensor k_89_pad_type_0 = const()[name = tensor("k_89_pad_type_0"), val = tensor("custom")]; + tensor k_89_pad_0 = const()[name = tensor("k_89_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_7_attn1_to_k_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_7_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(688038720)))]; + tensor k_89_cast_fp16 = conv(dilations = var_9135, groups = var_2624, pad = k_89_pad_0, pad_type = k_89_pad_type_0, strides = var_9133, weight = down_blocks_2_attentions_0_transformer_blocks_7_attn1_to_k_weight_to_fp16, x = hidden_states_107_cast_fp16)[name = tensor("k_89_cast_fp16")]; + tensor var_9139 = const()[name = tensor("op_9139"), val = tensor([1, 1])]; + tensor var_9141 = const()[name = tensor("op_9141"), val = tensor([1, 1])]; + tensor v_45_pad_type_0 = const()[name = tensor("v_45_pad_type_0"), val = tensor("custom")]; + tensor v_45_pad_0 = const()[name = tensor("v_45_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_7_attn1_to_v_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_7_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(691315584)))]; + tensor v_45_cast_fp16 = conv(dilations = var_9141, groups = var_2624, pad = v_45_pad_0, pad_type = v_45_pad_type_0, strides = var_9139, weight = down_blocks_2_attentions_0_transformer_blocks_7_attn1_to_v_weight_to_fp16, x = hidden_states_107_cast_fp16)[name = tensor("v_45_cast_fp16")]; + tensor var_9145_begin_0 = const()[name = tensor("op_9145_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9145_end_0 = const()[name = tensor("op_9145_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_9145_end_mask_0 = const()[name = tensor("op_9145_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9145_cast_fp16 = slice_by_index(begin = var_9145_begin_0, end = var_9145_end_0, end_mask = var_9145_end_mask_0, x = q_45_cast_fp16)[name = tensor("op_9145_cast_fp16")]; + tensor var_9149_begin_0 = const()[name = tensor("op_9149_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_9149_end_0 = const()[name = tensor("op_9149_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_9149_end_mask_0 = const()[name = tensor("op_9149_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9149_cast_fp16 = slice_by_index(begin = var_9149_begin_0, end = var_9149_end_0, end_mask = var_9149_end_mask_0, x = q_45_cast_fp16)[name = tensor("op_9149_cast_fp16")]; + tensor var_9153_begin_0 = const()[name = tensor("op_9153_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_9153_end_0 = const()[name = tensor("op_9153_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_9153_end_mask_0 = const()[name = tensor("op_9153_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9153_cast_fp16 = slice_by_index(begin = var_9153_begin_0, end = var_9153_end_0, end_mask = var_9153_end_mask_0, x = q_45_cast_fp16)[name = tensor("op_9153_cast_fp16")]; + tensor var_9157_begin_0 = const()[name = tensor("op_9157_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_9157_end_0 = const()[name = tensor("op_9157_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_9157_end_mask_0 = const()[name = tensor("op_9157_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9157_cast_fp16 = slice_by_index(begin = var_9157_begin_0, end = var_9157_end_0, end_mask = var_9157_end_mask_0, x = q_45_cast_fp16)[name = tensor("op_9157_cast_fp16")]; + tensor var_9161_begin_0 = const()[name = tensor("op_9161_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_9161_end_0 = const()[name = tensor("op_9161_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_9161_end_mask_0 = const()[name = tensor("op_9161_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9161_cast_fp16 = slice_by_index(begin = var_9161_begin_0, end = var_9161_end_0, end_mask = var_9161_end_mask_0, x = q_45_cast_fp16)[name = tensor("op_9161_cast_fp16")]; + tensor var_9165_begin_0 = const()[name = tensor("op_9165_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_9165_end_0 = const()[name = tensor("op_9165_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_9165_end_mask_0 = const()[name = tensor("op_9165_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9165_cast_fp16 = slice_by_index(begin = var_9165_begin_0, end = var_9165_end_0, end_mask = var_9165_end_mask_0, x = q_45_cast_fp16)[name = tensor("op_9165_cast_fp16")]; + tensor var_9169_begin_0 = const()[name = tensor("op_9169_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_9169_end_0 = const()[name = tensor("op_9169_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_9169_end_mask_0 = const()[name = tensor("op_9169_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9169_cast_fp16 = slice_by_index(begin = var_9169_begin_0, end = var_9169_end_0, end_mask = var_9169_end_mask_0, x = q_45_cast_fp16)[name = tensor("op_9169_cast_fp16")]; + tensor var_9173_begin_0 = const()[name = tensor("op_9173_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_9173_end_0 = const()[name = tensor("op_9173_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_9173_end_mask_0 = const()[name = tensor("op_9173_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9173_cast_fp16 = slice_by_index(begin = var_9173_begin_0, end = var_9173_end_0, end_mask = var_9173_end_mask_0, x = q_45_cast_fp16)[name = tensor("op_9173_cast_fp16")]; + tensor var_9177_begin_0 = const()[name = tensor("op_9177_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_9177_end_0 = const()[name = tensor("op_9177_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_9177_end_mask_0 = const()[name = tensor("op_9177_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9177_cast_fp16 = slice_by_index(begin = var_9177_begin_0, end = var_9177_end_0, end_mask = var_9177_end_mask_0, x = q_45_cast_fp16)[name = tensor("op_9177_cast_fp16")]; + tensor var_9181_begin_0 = const()[name = tensor("op_9181_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_9181_end_0 = const()[name = tensor("op_9181_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_9181_end_mask_0 = const()[name = tensor("op_9181_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9181_cast_fp16 = slice_by_index(begin = var_9181_begin_0, end = var_9181_end_0, end_mask = var_9181_end_mask_0, x = q_45_cast_fp16)[name = tensor("op_9181_cast_fp16")]; + tensor var_9185_begin_0 = const()[name = tensor("op_9185_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_9185_end_0 = const()[name = tensor("op_9185_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_9185_end_mask_0 = const()[name = tensor("op_9185_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9185_cast_fp16 = slice_by_index(begin = var_9185_begin_0, end = var_9185_end_0, end_mask = var_9185_end_mask_0, x = q_45_cast_fp16)[name = tensor("op_9185_cast_fp16")]; + tensor var_9189_begin_0 = const()[name = tensor("op_9189_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_9189_end_0 = const()[name = tensor("op_9189_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_9189_end_mask_0 = const()[name = tensor("op_9189_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9189_cast_fp16 = slice_by_index(begin = var_9189_begin_0, end = var_9189_end_0, end_mask = var_9189_end_mask_0, x = q_45_cast_fp16)[name = tensor("op_9189_cast_fp16")]; + tensor var_9193_begin_0 = const()[name = tensor("op_9193_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_9193_end_0 = const()[name = tensor("op_9193_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_9193_end_mask_0 = const()[name = tensor("op_9193_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9193_cast_fp16 = slice_by_index(begin = var_9193_begin_0, end = var_9193_end_0, end_mask = var_9193_end_mask_0, x = q_45_cast_fp16)[name = tensor("op_9193_cast_fp16")]; + tensor var_9197_begin_0 = const()[name = tensor("op_9197_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_9197_end_0 = const()[name = tensor("op_9197_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_9197_end_mask_0 = const()[name = tensor("op_9197_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9197_cast_fp16 = slice_by_index(begin = var_9197_begin_0, end = var_9197_end_0, end_mask = var_9197_end_mask_0, x = q_45_cast_fp16)[name = tensor("op_9197_cast_fp16")]; + tensor var_9201_begin_0 = const()[name = tensor("op_9201_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_9201_end_0 = const()[name = tensor("op_9201_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_9201_end_mask_0 = const()[name = tensor("op_9201_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9201_cast_fp16 = slice_by_index(begin = var_9201_begin_0, end = var_9201_end_0, end_mask = var_9201_end_mask_0, x = q_45_cast_fp16)[name = tensor("op_9201_cast_fp16")]; + tensor var_9205_begin_0 = const()[name = tensor("op_9205_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_9205_end_0 = const()[name = tensor("op_9205_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_9205_end_mask_0 = const()[name = tensor("op_9205_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9205_cast_fp16 = slice_by_index(begin = var_9205_begin_0, end = var_9205_end_0, end_mask = var_9205_end_mask_0, x = q_45_cast_fp16)[name = tensor("op_9205_cast_fp16")]; + tensor var_9209_begin_0 = const()[name = tensor("op_9209_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_9209_end_0 = const()[name = tensor("op_9209_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_9209_end_mask_0 = const()[name = tensor("op_9209_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9209_cast_fp16 = slice_by_index(begin = var_9209_begin_0, end = var_9209_end_0, end_mask = var_9209_end_mask_0, x = q_45_cast_fp16)[name = tensor("op_9209_cast_fp16")]; + tensor var_9213_begin_0 = const()[name = tensor("op_9213_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_9213_end_0 = const()[name = tensor("op_9213_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_9213_end_mask_0 = const()[name = tensor("op_9213_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9213_cast_fp16 = slice_by_index(begin = var_9213_begin_0, end = var_9213_end_0, end_mask = var_9213_end_mask_0, x = q_45_cast_fp16)[name = tensor("op_9213_cast_fp16")]; + tensor var_9217_begin_0 = const()[name = tensor("op_9217_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_9217_end_0 = const()[name = tensor("op_9217_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_9217_end_mask_0 = const()[name = tensor("op_9217_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9217_cast_fp16 = slice_by_index(begin = var_9217_begin_0, end = var_9217_end_0, end_mask = var_9217_end_mask_0, x = q_45_cast_fp16)[name = tensor("op_9217_cast_fp16")]; + tensor var_9221_begin_0 = const()[name = tensor("op_9221_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_9221_end_0 = const()[name = tensor("op_9221_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_9221_end_mask_0 = const()[name = tensor("op_9221_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9221_cast_fp16 = slice_by_index(begin = var_9221_begin_0, end = var_9221_end_0, end_mask = var_9221_end_mask_0, x = q_45_cast_fp16)[name = tensor("op_9221_cast_fp16")]; + tensor k_91_perm_0 = const()[name = tensor("k_91_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_9228_begin_0 = const()[name = tensor("op_9228_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9228_end_0 = const()[name = tensor("op_9228_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_9228_end_mask_0 = const()[name = tensor("op_9228_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_91_cast_fp16 = transpose(perm = k_91_perm_0, x = k_89_cast_fp16)[name = tensor("transpose_117")]; + tensor var_9228_cast_fp16 = slice_by_index(begin = var_9228_begin_0, end = var_9228_end_0, end_mask = var_9228_end_mask_0, x = k_91_cast_fp16)[name = tensor("op_9228_cast_fp16")]; + tensor var_9232_begin_0 = const()[name = tensor("op_9232_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_9232_end_0 = const()[name = tensor("op_9232_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_9232_end_mask_0 = const()[name = tensor("op_9232_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9232_cast_fp16 = slice_by_index(begin = var_9232_begin_0, end = var_9232_end_0, end_mask = var_9232_end_mask_0, x = k_91_cast_fp16)[name = tensor("op_9232_cast_fp16")]; + tensor var_9236_begin_0 = const()[name = tensor("op_9236_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_9236_end_0 = const()[name = tensor("op_9236_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_9236_end_mask_0 = const()[name = tensor("op_9236_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9236_cast_fp16 = slice_by_index(begin = var_9236_begin_0, end = var_9236_end_0, end_mask = var_9236_end_mask_0, x = k_91_cast_fp16)[name = tensor("op_9236_cast_fp16")]; + tensor var_9240_begin_0 = const()[name = tensor("op_9240_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_9240_end_0 = const()[name = tensor("op_9240_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_9240_end_mask_0 = const()[name = tensor("op_9240_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9240_cast_fp16 = slice_by_index(begin = var_9240_begin_0, end = var_9240_end_0, end_mask = var_9240_end_mask_0, x = k_91_cast_fp16)[name = tensor("op_9240_cast_fp16")]; + tensor var_9244_begin_0 = const()[name = tensor("op_9244_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_9244_end_0 = const()[name = tensor("op_9244_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_9244_end_mask_0 = const()[name = tensor("op_9244_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9244_cast_fp16 = slice_by_index(begin = var_9244_begin_0, end = var_9244_end_0, end_mask = var_9244_end_mask_0, x = k_91_cast_fp16)[name = tensor("op_9244_cast_fp16")]; + tensor var_9248_begin_0 = const()[name = tensor("op_9248_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_9248_end_0 = const()[name = tensor("op_9248_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_9248_end_mask_0 = const()[name = tensor("op_9248_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9248_cast_fp16 = slice_by_index(begin = var_9248_begin_0, end = var_9248_end_0, end_mask = var_9248_end_mask_0, x = k_91_cast_fp16)[name = tensor("op_9248_cast_fp16")]; + tensor var_9252_begin_0 = const()[name = tensor("op_9252_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_9252_end_0 = const()[name = tensor("op_9252_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_9252_end_mask_0 = const()[name = tensor("op_9252_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9252_cast_fp16 = slice_by_index(begin = var_9252_begin_0, end = var_9252_end_0, end_mask = var_9252_end_mask_0, x = k_91_cast_fp16)[name = tensor("op_9252_cast_fp16")]; + tensor var_9256_begin_0 = const()[name = tensor("op_9256_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_9256_end_0 = const()[name = tensor("op_9256_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_9256_end_mask_0 = const()[name = tensor("op_9256_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9256_cast_fp16 = slice_by_index(begin = var_9256_begin_0, end = var_9256_end_0, end_mask = var_9256_end_mask_0, x = k_91_cast_fp16)[name = tensor("op_9256_cast_fp16")]; + tensor var_9260_begin_0 = const()[name = tensor("op_9260_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_9260_end_0 = const()[name = tensor("op_9260_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_9260_end_mask_0 = const()[name = tensor("op_9260_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9260_cast_fp16 = slice_by_index(begin = var_9260_begin_0, end = var_9260_end_0, end_mask = var_9260_end_mask_0, x = k_91_cast_fp16)[name = tensor("op_9260_cast_fp16")]; + tensor var_9264_begin_0 = const()[name = tensor("op_9264_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_9264_end_0 = const()[name = tensor("op_9264_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_9264_end_mask_0 = const()[name = tensor("op_9264_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9264_cast_fp16 = slice_by_index(begin = var_9264_begin_0, end = var_9264_end_0, end_mask = var_9264_end_mask_0, x = k_91_cast_fp16)[name = tensor("op_9264_cast_fp16")]; + tensor var_9268_begin_0 = const()[name = tensor("op_9268_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_9268_end_0 = const()[name = tensor("op_9268_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_9268_end_mask_0 = const()[name = tensor("op_9268_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9268_cast_fp16 = slice_by_index(begin = var_9268_begin_0, end = var_9268_end_0, end_mask = var_9268_end_mask_0, x = k_91_cast_fp16)[name = tensor("op_9268_cast_fp16")]; + tensor var_9272_begin_0 = const()[name = tensor("op_9272_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_9272_end_0 = const()[name = tensor("op_9272_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_9272_end_mask_0 = const()[name = tensor("op_9272_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9272_cast_fp16 = slice_by_index(begin = var_9272_begin_0, end = var_9272_end_0, end_mask = var_9272_end_mask_0, x = k_91_cast_fp16)[name = tensor("op_9272_cast_fp16")]; + tensor var_9276_begin_0 = const()[name = tensor("op_9276_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_9276_end_0 = const()[name = tensor("op_9276_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_9276_end_mask_0 = const()[name = tensor("op_9276_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9276_cast_fp16 = slice_by_index(begin = var_9276_begin_0, end = var_9276_end_0, end_mask = var_9276_end_mask_0, x = k_91_cast_fp16)[name = tensor("op_9276_cast_fp16")]; + tensor var_9280_begin_0 = const()[name = tensor("op_9280_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_9280_end_0 = const()[name = tensor("op_9280_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_9280_end_mask_0 = const()[name = tensor("op_9280_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9280_cast_fp16 = slice_by_index(begin = var_9280_begin_0, end = var_9280_end_0, end_mask = var_9280_end_mask_0, x = k_91_cast_fp16)[name = tensor("op_9280_cast_fp16")]; + tensor var_9284_begin_0 = const()[name = tensor("op_9284_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_9284_end_0 = const()[name = tensor("op_9284_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_9284_end_mask_0 = const()[name = tensor("op_9284_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9284_cast_fp16 = slice_by_index(begin = var_9284_begin_0, end = var_9284_end_0, end_mask = var_9284_end_mask_0, x = k_91_cast_fp16)[name = tensor("op_9284_cast_fp16")]; + tensor var_9288_begin_0 = const()[name = tensor("op_9288_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_9288_end_0 = const()[name = tensor("op_9288_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_9288_end_mask_0 = const()[name = tensor("op_9288_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9288_cast_fp16 = slice_by_index(begin = var_9288_begin_0, end = var_9288_end_0, end_mask = var_9288_end_mask_0, x = k_91_cast_fp16)[name = tensor("op_9288_cast_fp16")]; + tensor var_9292_begin_0 = const()[name = tensor("op_9292_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_9292_end_0 = const()[name = tensor("op_9292_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_9292_end_mask_0 = const()[name = tensor("op_9292_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9292_cast_fp16 = slice_by_index(begin = var_9292_begin_0, end = var_9292_end_0, end_mask = var_9292_end_mask_0, x = k_91_cast_fp16)[name = tensor("op_9292_cast_fp16")]; + tensor var_9296_begin_0 = const()[name = tensor("op_9296_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_9296_end_0 = const()[name = tensor("op_9296_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_9296_end_mask_0 = const()[name = tensor("op_9296_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9296_cast_fp16 = slice_by_index(begin = var_9296_begin_0, end = var_9296_end_0, end_mask = var_9296_end_mask_0, x = k_91_cast_fp16)[name = tensor("op_9296_cast_fp16")]; + tensor var_9300_begin_0 = const()[name = tensor("op_9300_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_9300_end_0 = const()[name = tensor("op_9300_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_9300_end_mask_0 = const()[name = tensor("op_9300_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9300_cast_fp16 = slice_by_index(begin = var_9300_begin_0, end = var_9300_end_0, end_mask = var_9300_end_mask_0, x = k_91_cast_fp16)[name = tensor("op_9300_cast_fp16")]; + tensor var_9304_begin_0 = const()[name = tensor("op_9304_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_9304_end_0 = const()[name = tensor("op_9304_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_9304_end_mask_0 = const()[name = tensor("op_9304_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9304_cast_fp16 = slice_by_index(begin = var_9304_begin_0, end = var_9304_end_0, end_mask = var_9304_end_mask_0, x = k_91_cast_fp16)[name = tensor("op_9304_cast_fp16")]; + tensor var_9306_begin_0 = const()[name = tensor("op_9306_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9306_end_0 = const()[name = tensor("op_9306_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_9306_end_mask_0 = const()[name = tensor("op_9306_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9306_cast_fp16 = slice_by_index(begin = var_9306_begin_0, end = var_9306_end_0, end_mask = var_9306_end_mask_0, x = v_45_cast_fp16)[name = tensor("op_9306_cast_fp16")]; + tensor var_9310_begin_0 = const()[name = tensor("op_9310_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_9310_end_0 = const()[name = tensor("op_9310_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_9310_end_mask_0 = const()[name = tensor("op_9310_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9310_cast_fp16 = slice_by_index(begin = var_9310_begin_0, end = var_9310_end_0, end_mask = var_9310_end_mask_0, x = v_45_cast_fp16)[name = tensor("op_9310_cast_fp16")]; + tensor var_9314_begin_0 = const()[name = tensor("op_9314_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_9314_end_0 = const()[name = tensor("op_9314_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_9314_end_mask_0 = const()[name = tensor("op_9314_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9314_cast_fp16 = slice_by_index(begin = var_9314_begin_0, end = var_9314_end_0, end_mask = var_9314_end_mask_0, x = v_45_cast_fp16)[name = tensor("op_9314_cast_fp16")]; + tensor var_9318_begin_0 = const()[name = tensor("op_9318_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_9318_end_0 = const()[name = tensor("op_9318_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_9318_end_mask_0 = const()[name = tensor("op_9318_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9318_cast_fp16 = slice_by_index(begin = var_9318_begin_0, end = var_9318_end_0, end_mask = var_9318_end_mask_0, x = v_45_cast_fp16)[name = tensor("op_9318_cast_fp16")]; + tensor var_9322_begin_0 = const()[name = tensor("op_9322_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_9322_end_0 = const()[name = tensor("op_9322_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_9322_end_mask_0 = const()[name = tensor("op_9322_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9322_cast_fp16 = slice_by_index(begin = var_9322_begin_0, end = var_9322_end_0, end_mask = var_9322_end_mask_0, x = v_45_cast_fp16)[name = tensor("op_9322_cast_fp16")]; + tensor var_9326_begin_0 = const()[name = tensor("op_9326_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_9326_end_0 = const()[name = tensor("op_9326_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_9326_end_mask_0 = const()[name = tensor("op_9326_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9326_cast_fp16 = slice_by_index(begin = var_9326_begin_0, end = var_9326_end_0, end_mask = var_9326_end_mask_0, x = v_45_cast_fp16)[name = tensor("op_9326_cast_fp16")]; + tensor var_9330_begin_0 = const()[name = tensor("op_9330_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_9330_end_0 = const()[name = tensor("op_9330_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_9330_end_mask_0 = const()[name = tensor("op_9330_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9330_cast_fp16 = slice_by_index(begin = var_9330_begin_0, end = var_9330_end_0, end_mask = var_9330_end_mask_0, x = v_45_cast_fp16)[name = tensor("op_9330_cast_fp16")]; + tensor var_9334_begin_0 = const()[name = tensor("op_9334_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_9334_end_0 = const()[name = tensor("op_9334_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_9334_end_mask_0 = const()[name = tensor("op_9334_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9334_cast_fp16 = slice_by_index(begin = var_9334_begin_0, end = var_9334_end_0, end_mask = var_9334_end_mask_0, x = v_45_cast_fp16)[name = tensor("op_9334_cast_fp16")]; + tensor var_9338_begin_0 = const()[name = tensor("op_9338_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_9338_end_0 = const()[name = tensor("op_9338_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_9338_end_mask_0 = const()[name = tensor("op_9338_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9338_cast_fp16 = slice_by_index(begin = var_9338_begin_0, end = var_9338_end_0, end_mask = var_9338_end_mask_0, x = v_45_cast_fp16)[name = tensor("op_9338_cast_fp16")]; + tensor var_9342_begin_0 = const()[name = tensor("op_9342_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_9342_end_0 = const()[name = tensor("op_9342_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_9342_end_mask_0 = const()[name = tensor("op_9342_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9342_cast_fp16 = slice_by_index(begin = var_9342_begin_0, end = var_9342_end_0, end_mask = var_9342_end_mask_0, x = v_45_cast_fp16)[name = tensor("op_9342_cast_fp16")]; + tensor var_9346_begin_0 = const()[name = tensor("op_9346_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_9346_end_0 = const()[name = tensor("op_9346_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_9346_end_mask_0 = const()[name = tensor("op_9346_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9346_cast_fp16 = slice_by_index(begin = var_9346_begin_0, end = var_9346_end_0, end_mask = var_9346_end_mask_0, x = v_45_cast_fp16)[name = tensor("op_9346_cast_fp16")]; + tensor var_9350_begin_0 = const()[name = tensor("op_9350_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_9350_end_0 = const()[name = tensor("op_9350_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_9350_end_mask_0 = const()[name = tensor("op_9350_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9350_cast_fp16 = slice_by_index(begin = var_9350_begin_0, end = var_9350_end_0, end_mask = var_9350_end_mask_0, x = v_45_cast_fp16)[name = tensor("op_9350_cast_fp16")]; + tensor var_9354_begin_0 = const()[name = tensor("op_9354_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_9354_end_0 = const()[name = tensor("op_9354_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_9354_end_mask_0 = const()[name = tensor("op_9354_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9354_cast_fp16 = slice_by_index(begin = var_9354_begin_0, end = var_9354_end_0, end_mask = var_9354_end_mask_0, x = v_45_cast_fp16)[name = tensor("op_9354_cast_fp16")]; + tensor var_9358_begin_0 = const()[name = tensor("op_9358_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_9358_end_0 = const()[name = tensor("op_9358_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_9358_end_mask_0 = const()[name = tensor("op_9358_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9358_cast_fp16 = slice_by_index(begin = var_9358_begin_0, end = var_9358_end_0, end_mask = var_9358_end_mask_0, x = v_45_cast_fp16)[name = tensor("op_9358_cast_fp16")]; + tensor var_9362_begin_0 = const()[name = tensor("op_9362_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_9362_end_0 = const()[name = tensor("op_9362_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_9362_end_mask_0 = const()[name = tensor("op_9362_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9362_cast_fp16 = slice_by_index(begin = var_9362_begin_0, end = var_9362_end_0, end_mask = var_9362_end_mask_0, x = v_45_cast_fp16)[name = tensor("op_9362_cast_fp16")]; + tensor var_9366_begin_0 = const()[name = tensor("op_9366_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_9366_end_0 = const()[name = tensor("op_9366_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_9366_end_mask_0 = const()[name = tensor("op_9366_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9366_cast_fp16 = slice_by_index(begin = var_9366_begin_0, end = var_9366_end_0, end_mask = var_9366_end_mask_0, x = v_45_cast_fp16)[name = tensor("op_9366_cast_fp16")]; + tensor var_9370_begin_0 = const()[name = tensor("op_9370_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_9370_end_0 = const()[name = tensor("op_9370_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_9370_end_mask_0 = const()[name = tensor("op_9370_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9370_cast_fp16 = slice_by_index(begin = var_9370_begin_0, end = var_9370_end_0, end_mask = var_9370_end_mask_0, x = v_45_cast_fp16)[name = tensor("op_9370_cast_fp16")]; + tensor var_9374_begin_0 = const()[name = tensor("op_9374_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_9374_end_0 = const()[name = tensor("op_9374_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_9374_end_mask_0 = const()[name = tensor("op_9374_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9374_cast_fp16 = slice_by_index(begin = var_9374_begin_0, end = var_9374_end_0, end_mask = var_9374_end_mask_0, x = v_45_cast_fp16)[name = tensor("op_9374_cast_fp16")]; + tensor var_9378_begin_0 = const()[name = tensor("op_9378_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_9378_end_0 = const()[name = tensor("op_9378_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_9378_end_mask_0 = const()[name = tensor("op_9378_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9378_cast_fp16 = slice_by_index(begin = var_9378_begin_0, end = var_9378_end_0, end_mask = var_9378_end_mask_0, x = v_45_cast_fp16)[name = tensor("op_9378_cast_fp16")]; + tensor var_9382_begin_0 = const()[name = tensor("op_9382_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_9382_end_0 = const()[name = tensor("op_9382_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_9382_end_mask_0 = const()[name = tensor("op_9382_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9382_cast_fp16 = slice_by_index(begin = var_9382_begin_0, end = var_9382_end_0, end_mask = var_9382_end_mask_0, x = v_45_cast_fp16)[name = tensor("op_9382_cast_fp16")]; + tensor var_9386_equation_0 = const()[name = tensor("op_9386_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9386_cast_fp16 = einsum(equation = var_9386_equation_0, values = (var_9228_cast_fp16, var_9145_cast_fp16))[name = tensor("op_9386_cast_fp16")]; + tensor var_9387_to_fp16 = const()[name = tensor("op_9387_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_721_cast_fp16 = mul(x = var_9386_cast_fp16, y = var_9387_to_fp16)[name = tensor("aw_721_cast_fp16")]; + tensor var_9390_equation_0 = const()[name = tensor("op_9390_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9390_cast_fp16 = einsum(equation = var_9390_equation_0, values = (var_9232_cast_fp16, var_9149_cast_fp16))[name = tensor("op_9390_cast_fp16")]; + tensor var_9391_to_fp16 = const()[name = tensor("op_9391_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_723_cast_fp16 = mul(x = var_9390_cast_fp16, y = var_9391_to_fp16)[name = tensor("aw_723_cast_fp16")]; + tensor var_9394_equation_0 = const()[name = tensor("op_9394_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9394_cast_fp16 = einsum(equation = var_9394_equation_0, values = (var_9236_cast_fp16, var_9153_cast_fp16))[name = tensor("op_9394_cast_fp16")]; + tensor var_9395_to_fp16 = const()[name = tensor("op_9395_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_725_cast_fp16 = mul(x = var_9394_cast_fp16, y = var_9395_to_fp16)[name = tensor("aw_725_cast_fp16")]; + tensor var_9398_equation_0 = const()[name = tensor("op_9398_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9398_cast_fp16 = einsum(equation = var_9398_equation_0, values = (var_9240_cast_fp16, var_9157_cast_fp16))[name = tensor("op_9398_cast_fp16")]; + tensor var_9399_to_fp16 = const()[name = tensor("op_9399_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_727_cast_fp16 = mul(x = var_9398_cast_fp16, y = var_9399_to_fp16)[name = tensor("aw_727_cast_fp16")]; + tensor var_9402_equation_0 = const()[name = tensor("op_9402_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9402_cast_fp16 = einsum(equation = var_9402_equation_0, values = (var_9244_cast_fp16, var_9161_cast_fp16))[name = tensor("op_9402_cast_fp16")]; + tensor var_9403_to_fp16 = const()[name = tensor("op_9403_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_729_cast_fp16 = mul(x = var_9402_cast_fp16, y = var_9403_to_fp16)[name = tensor("aw_729_cast_fp16")]; + tensor var_9406_equation_0 = const()[name = tensor("op_9406_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9406_cast_fp16 = einsum(equation = var_9406_equation_0, values = (var_9248_cast_fp16, var_9165_cast_fp16))[name = tensor("op_9406_cast_fp16")]; + tensor var_9407_to_fp16 = const()[name = tensor("op_9407_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_731_cast_fp16 = mul(x = var_9406_cast_fp16, y = var_9407_to_fp16)[name = tensor("aw_731_cast_fp16")]; + tensor var_9410_equation_0 = const()[name = tensor("op_9410_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9410_cast_fp16 = einsum(equation = var_9410_equation_0, values = (var_9252_cast_fp16, var_9169_cast_fp16))[name = tensor("op_9410_cast_fp16")]; + tensor var_9411_to_fp16 = const()[name = tensor("op_9411_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_733_cast_fp16 = mul(x = var_9410_cast_fp16, y = var_9411_to_fp16)[name = tensor("aw_733_cast_fp16")]; + tensor var_9414_equation_0 = const()[name = tensor("op_9414_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9414_cast_fp16 = einsum(equation = var_9414_equation_0, values = (var_9256_cast_fp16, var_9173_cast_fp16))[name = tensor("op_9414_cast_fp16")]; + tensor var_9415_to_fp16 = const()[name = tensor("op_9415_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_735_cast_fp16 = mul(x = var_9414_cast_fp16, y = var_9415_to_fp16)[name = tensor("aw_735_cast_fp16")]; + tensor var_9418_equation_0 = const()[name = tensor("op_9418_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9418_cast_fp16 = einsum(equation = var_9418_equation_0, values = (var_9260_cast_fp16, var_9177_cast_fp16))[name = tensor("op_9418_cast_fp16")]; + tensor var_9419_to_fp16 = const()[name = tensor("op_9419_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_737_cast_fp16 = mul(x = var_9418_cast_fp16, y = var_9419_to_fp16)[name = tensor("aw_737_cast_fp16")]; + tensor var_9422_equation_0 = const()[name = tensor("op_9422_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9422_cast_fp16 = einsum(equation = var_9422_equation_0, values = (var_9264_cast_fp16, var_9181_cast_fp16))[name = tensor("op_9422_cast_fp16")]; + tensor var_9423_to_fp16 = const()[name = tensor("op_9423_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_739_cast_fp16 = mul(x = var_9422_cast_fp16, y = var_9423_to_fp16)[name = tensor("aw_739_cast_fp16")]; + tensor var_9426_equation_0 = const()[name = tensor("op_9426_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9426_cast_fp16 = einsum(equation = var_9426_equation_0, values = (var_9268_cast_fp16, var_9185_cast_fp16))[name = tensor("op_9426_cast_fp16")]; + tensor var_9427_to_fp16 = const()[name = tensor("op_9427_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_741_cast_fp16 = mul(x = var_9426_cast_fp16, y = var_9427_to_fp16)[name = tensor("aw_741_cast_fp16")]; + tensor var_9430_equation_0 = const()[name = tensor("op_9430_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9430_cast_fp16 = einsum(equation = var_9430_equation_0, values = (var_9272_cast_fp16, var_9189_cast_fp16))[name = tensor("op_9430_cast_fp16")]; + tensor var_9431_to_fp16 = const()[name = tensor("op_9431_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_743_cast_fp16 = mul(x = var_9430_cast_fp16, y = var_9431_to_fp16)[name = tensor("aw_743_cast_fp16")]; + tensor var_9434_equation_0 = const()[name = tensor("op_9434_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9434_cast_fp16 = einsum(equation = var_9434_equation_0, values = (var_9276_cast_fp16, var_9193_cast_fp16))[name = tensor("op_9434_cast_fp16")]; + tensor var_9435_to_fp16 = const()[name = tensor("op_9435_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_745_cast_fp16 = mul(x = var_9434_cast_fp16, y = var_9435_to_fp16)[name = tensor("aw_745_cast_fp16")]; + tensor var_9438_equation_0 = const()[name = tensor("op_9438_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9438_cast_fp16 = einsum(equation = var_9438_equation_0, values = (var_9280_cast_fp16, var_9197_cast_fp16))[name = tensor("op_9438_cast_fp16")]; + tensor var_9439_to_fp16 = const()[name = tensor("op_9439_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_747_cast_fp16 = mul(x = var_9438_cast_fp16, y = var_9439_to_fp16)[name = tensor("aw_747_cast_fp16")]; + tensor var_9442_equation_0 = const()[name = tensor("op_9442_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9442_cast_fp16 = einsum(equation = var_9442_equation_0, values = (var_9284_cast_fp16, var_9201_cast_fp16))[name = tensor("op_9442_cast_fp16")]; + tensor var_9443_to_fp16 = const()[name = tensor("op_9443_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_749_cast_fp16 = mul(x = var_9442_cast_fp16, y = var_9443_to_fp16)[name = tensor("aw_749_cast_fp16")]; + tensor var_9446_equation_0 = const()[name = tensor("op_9446_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9446_cast_fp16 = einsum(equation = var_9446_equation_0, values = (var_9288_cast_fp16, var_9205_cast_fp16))[name = tensor("op_9446_cast_fp16")]; + tensor var_9447_to_fp16 = const()[name = tensor("op_9447_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_751_cast_fp16 = mul(x = var_9446_cast_fp16, y = var_9447_to_fp16)[name = tensor("aw_751_cast_fp16")]; + tensor var_9450_equation_0 = const()[name = tensor("op_9450_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9450_cast_fp16 = einsum(equation = var_9450_equation_0, values = (var_9292_cast_fp16, var_9209_cast_fp16))[name = tensor("op_9450_cast_fp16")]; + tensor var_9451_to_fp16 = const()[name = tensor("op_9451_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_753_cast_fp16 = mul(x = var_9450_cast_fp16, y = var_9451_to_fp16)[name = tensor("aw_753_cast_fp16")]; + tensor var_9454_equation_0 = const()[name = tensor("op_9454_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9454_cast_fp16 = einsum(equation = var_9454_equation_0, values = (var_9296_cast_fp16, var_9213_cast_fp16))[name = tensor("op_9454_cast_fp16")]; + tensor var_9455_to_fp16 = const()[name = tensor("op_9455_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_755_cast_fp16 = mul(x = var_9454_cast_fp16, y = var_9455_to_fp16)[name = tensor("aw_755_cast_fp16")]; + tensor var_9458_equation_0 = const()[name = tensor("op_9458_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9458_cast_fp16 = einsum(equation = var_9458_equation_0, values = (var_9300_cast_fp16, var_9217_cast_fp16))[name = tensor("op_9458_cast_fp16")]; + tensor var_9459_to_fp16 = const()[name = tensor("op_9459_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_757_cast_fp16 = mul(x = var_9458_cast_fp16, y = var_9459_to_fp16)[name = tensor("aw_757_cast_fp16")]; + tensor var_9462_equation_0 = const()[name = tensor("op_9462_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9462_cast_fp16 = einsum(equation = var_9462_equation_0, values = (var_9304_cast_fp16, var_9221_cast_fp16))[name = tensor("op_9462_cast_fp16")]; + tensor var_9463_to_fp16 = const()[name = tensor("op_9463_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_759_cast_fp16 = mul(x = var_9462_cast_fp16, y = var_9463_to_fp16)[name = tensor("aw_759_cast_fp16")]; + tensor var_9465_cast_fp16 = softmax(axis = var_2624, x = aw_721_cast_fp16)[name = tensor("op_9465_cast_fp16")]; + tensor var_9466_cast_fp16 = softmax(axis = var_2624, x = aw_723_cast_fp16)[name = tensor("op_9466_cast_fp16")]; + tensor var_9467_cast_fp16 = softmax(axis = var_2624, x = aw_725_cast_fp16)[name = tensor("op_9467_cast_fp16")]; + tensor var_9468_cast_fp16 = softmax(axis = var_2624, x = aw_727_cast_fp16)[name = tensor("op_9468_cast_fp16")]; + tensor var_9469_cast_fp16 = softmax(axis = var_2624, x = aw_729_cast_fp16)[name = tensor("op_9469_cast_fp16")]; + tensor var_9470_cast_fp16 = softmax(axis = var_2624, x = aw_731_cast_fp16)[name = tensor("op_9470_cast_fp16")]; + tensor var_9471_cast_fp16 = softmax(axis = var_2624, x = aw_733_cast_fp16)[name = tensor("op_9471_cast_fp16")]; + tensor var_9472_cast_fp16 = softmax(axis = var_2624, x = aw_735_cast_fp16)[name = tensor("op_9472_cast_fp16")]; + tensor var_9473_cast_fp16 = softmax(axis = var_2624, x = aw_737_cast_fp16)[name = tensor("op_9473_cast_fp16")]; + tensor var_9474_cast_fp16 = softmax(axis = var_2624, x = aw_739_cast_fp16)[name = tensor("op_9474_cast_fp16")]; + tensor var_9475_cast_fp16 = softmax(axis = var_2624, x = aw_741_cast_fp16)[name = tensor("op_9475_cast_fp16")]; + tensor var_9476_cast_fp16 = softmax(axis = var_2624, x = aw_743_cast_fp16)[name = tensor("op_9476_cast_fp16")]; + tensor var_9477_cast_fp16 = softmax(axis = var_2624, x = aw_745_cast_fp16)[name = tensor("op_9477_cast_fp16")]; + tensor var_9478_cast_fp16 = softmax(axis = var_2624, x = aw_747_cast_fp16)[name = tensor("op_9478_cast_fp16")]; + tensor var_9479_cast_fp16 = softmax(axis = var_2624, x = aw_749_cast_fp16)[name = tensor("op_9479_cast_fp16")]; + tensor var_9480_cast_fp16 = softmax(axis = var_2624, x = aw_751_cast_fp16)[name = tensor("op_9480_cast_fp16")]; + tensor var_9481_cast_fp16 = softmax(axis = var_2624, x = aw_753_cast_fp16)[name = tensor("op_9481_cast_fp16")]; + tensor var_9482_cast_fp16 = softmax(axis = var_2624, x = aw_755_cast_fp16)[name = tensor("op_9482_cast_fp16")]; + tensor var_9483_cast_fp16 = softmax(axis = var_2624, x = aw_757_cast_fp16)[name = tensor("op_9483_cast_fp16")]; + tensor var_9484_cast_fp16 = softmax(axis = var_2624, x = aw_759_cast_fp16)[name = tensor("op_9484_cast_fp16")]; + tensor var_9486_equation_0 = const()[name = tensor("op_9486_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9486_cast_fp16 = einsum(equation = var_9486_equation_0, values = (var_9306_cast_fp16, var_9465_cast_fp16))[name = tensor("op_9486_cast_fp16")]; + tensor var_9488_equation_0 = const()[name = tensor("op_9488_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9488_cast_fp16 = einsum(equation = var_9488_equation_0, values = (var_9310_cast_fp16, var_9466_cast_fp16))[name = tensor("op_9488_cast_fp16")]; + tensor var_9490_equation_0 = const()[name = tensor("op_9490_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9490_cast_fp16 = einsum(equation = var_9490_equation_0, values = (var_9314_cast_fp16, var_9467_cast_fp16))[name = tensor("op_9490_cast_fp16")]; + tensor var_9492_equation_0 = const()[name = tensor("op_9492_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9492_cast_fp16 = einsum(equation = var_9492_equation_0, values = (var_9318_cast_fp16, var_9468_cast_fp16))[name = tensor("op_9492_cast_fp16")]; + tensor var_9494_equation_0 = const()[name = tensor("op_9494_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9494_cast_fp16 = einsum(equation = var_9494_equation_0, values = (var_9322_cast_fp16, var_9469_cast_fp16))[name = tensor("op_9494_cast_fp16")]; + tensor var_9496_equation_0 = const()[name = tensor("op_9496_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9496_cast_fp16 = einsum(equation = var_9496_equation_0, values = (var_9326_cast_fp16, var_9470_cast_fp16))[name = tensor("op_9496_cast_fp16")]; + tensor var_9498_equation_0 = const()[name = tensor("op_9498_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9498_cast_fp16 = einsum(equation = var_9498_equation_0, values = (var_9330_cast_fp16, var_9471_cast_fp16))[name = tensor("op_9498_cast_fp16")]; + tensor var_9500_equation_0 = const()[name = tensor("op_9500_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9500_cast_fp16 = einsum(equation = var_9500_equation_0, values = (var_9334_cast_fp16, var_9472_cast_fp16))[name = tensor("op_9500_cast_fp16")]; + tensor var_9502_equation_0 = const()[name = tensor("op_9502_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9502_cast_fp16 = einsum(equation = var_9502_equation_0, values = (var_9338_cast_fp16, var_9473_cast_fp16))[name = tensor("op_9502_cast_fp16")]; + tensor var_9504_equation_0 = const()[name = tensor("op_9504_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9504_cast_fp16 = einsum(equation = var_9504_equation_0, values = (var_9342_cast_fp16, var_9474_cast_fp16))[name = tensor("op_9504_cast_fp16")]; + tensor var_9506_equation_0 = const()[name = tensor("op_9506_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9506_cast_fp16 = einsum(equation = var_9506_equation_0, values = (var_9346_cast_fp16, var_9475_cast_fp16))[name = tensor("op_9506_cast_fp16")]; + tensor var_9508_equation_0 = const()[name = tensor("op_9508_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9508_cast_fp16 = einsum(equation = var_9508_equation_0, values = (var_9350_cast_fp16, var_9476_cast_fp16))[name = tensor("op_9508_cast_fp16")]; + tensor var_9510_equation_0 = const()[name = tensor("op_9510_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9510_cast_fp16 = einsum(equation = var_9510_equation_0, values = (var_9354_cast_fp16, var_9477_cast_fp16))[name = tensor("op_9510_cast_fp16")]; + tensor var_9512_equation_0 = const()[name = tensor("op_9512_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9512_cast_fp16 = einsum(equation = var_9512_equation_0, values = (var_9358_cast_fp16, var_9478_cast_fp16))[name = tensor("op_9512_cast_fp16")]; + tensor var_9514_equation_0 = const()[name = tensor("op_9514_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9514_cast_fp16 = einsum(equation = var_9514_equation_0, values = (var_9362_cast_fp16, var_9479_cast_fp16))[name = tensor("op_9514_cast_fp16")]; + tensor var_9516_equation_0 = const()[name = tensor("op_9516_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9516_cast_fp16 = einsum(equation = var_9516_equation_0, values = (var_9366_cast_fp16, var_9480_cast_fp16))[name = tensor("op_9516_cast_fp16")]; + tensor var_9518_equation_0 = const()[name = tensor("op_9518_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9518_cast_fp16 = einsum(equation = var_9518_equation_0, values = (var_9370_cast_fp16, var_9481_cast_fp16))[name = tensor("op_9518_cast_fp16")]; + tensor var_9520_equation_0 = const()[name = tensor("op_9520_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9520_cast_fp16 = einsum(equation = var_9520_equation_0, values = (var_9374_cast_fp16, var_9482_cast_fp16))[name = tensor("op_9520_cast_fp16")]; + tensor var_9522_equation_0 = const()[name = tensor("op_9522_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9522_cast_fp16 = einsum(equation = var_9522_equation_0, values = (var_9378_cast_fp16, var_9483_cast_fp16))[name = tensor("op_9522_cast_fp16")]; + tensor var_9524_equation_0 = const()[name = tensor("op_9524_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9524_cast_fp16 = einsum(equation = var_9524_equation_0, values = (var_9382_cast_fp16, var_9484_cast_fp16))[name = tensor("op_9524_cast_fp16")]; + tensor input_187_interleave_0 = const()[name = tensor("input_187_interleave_0"), val = tensor(false)]; + tensor input_187_cast_fp16 = concat(axis = var_2624, interleave = input_187_interleave_0, values = (var_9486_cast_fp16, var_9488_cast_fp16, var_9490_cast_fp16, var_9492_cast_fp16, var_9494_cast_fp16, var_9496_cast_fp16, var_9498_cast_fp16, var_9500_cast_fp16, var_9502_cast_fp16, var_9504_cast_fp16, var_9506_cast_fp16, var_9508_cast_fp16, var_9510_cast_fp16, var_9512_cast_fp16, var_9514_cast_fp16, var_9516_cast_fp16, var_9518_cast_fp16, var_9520_cast_fp16, var_9522_cast_fp16, var_9524_cast_fp16))[name = tensor("input_187_cast_fp16")]; + tensor var_9530 = const()[name = tensor("op_9530"), val = tensor([1, 1])]; + tensor var_9532 = const()[name = tensor("op_9532"), val = tensor([1, 1])]; + tensor var_9534_pad_type_0 = const()[name = tensor("op_9534_pad_type_0"), val = tensor("custom")]; + tensor var_9534_pad_0 = const()[name = tensor("op_9534_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_7_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_7_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(694592448)))]; + tensor down_blocks_2_attentions_0_transformer_blocks_7_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_7_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(697869312)))]; + tensor var_9534_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_7_attn1_to_out_0_bias_to_fp16, dilations = var_9532, groups = var_2624, pad = var_9534_pad_0, pad_type = var_9534_pad_type_0, strides = var_9530, weight = down_blocks_2_attentions_0_transformer_blocks_7_attn1_to_out_0_weight_to_fp16, x = input_187_cast_fp16)[name = tensor("op_9534_cast_fp16")]; + tensor inputs_69_cast_fp16 = add(x = var_9534_cast_fp16, y = inputs_67_cast_fp16)[name = tensor("inputs_69_cast_fp16")]; + tensor hidden_states_109_axes_0 = const()[name = tensor("hidden_states_109_axes_0"), val = tensor([1])]; + tensor hidden_states_109_gamma_0_to_fp16 = const()[name = tensor("hidden_states_109_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(697871936)))]; + tensor hidden_states_109_beta_0_to_fp16 = const()[name = tensor("hidden_states_109_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(697874560)))]; + tensor var_9544_to_fp16 = const()[name = tensor("op_9544_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_109_cast_fp16 = layer_norm(axes = hidden_states_109_axes_0, beta = hidden_states_109_beta_0_to_fp16, epsilon = var_9544_to_fp16, gamma = hidden_states_109_gamma_0_to_fp16, x = inputs_69_cast_fp16)[name = tensor("hidden_states_109_cast_fp16")]; + tensor var_9559 = const()[name = tensor("op_9559"), val = tensor([1, 1])]; + tensor var_9561 = const()[name = tensor("op_9561"), val = tensor([1, 1])]; + tensor q_47_pad_type_0 = const()[name = tensor("q_47_pad_type_0"), val = tensor("custom")]; + tensor q_47_pad_0 = const()[name = tensor("q_47_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_7_attn2_to_q_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_7_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(697877184)))]; + tensor q_47_cast_fp16 = conv(dilations = var_9561, groups = var_2624, pad = q_47_pad_0, pad_type = q_47_pad_type_0, strides = var_9559, weight = down_blocks_2_attentions_0_transformer_blocks_7_attn2_to_q_weight_to_fp16, x = hidden_states_109_cast_fp16)[name = tensor("q_47_cast_fp16")]; + tensor var_9565 = const()[name = tensor("op_9565"), val = tensor([1, 1])]; + tensor var_9567 = const()[name = tensor("op_9567"), val = tensor([1, 1])]; + tensor k_93_pad_type_0 = const()[name = tensor("k_93_pad_type_0"), val = tensor("custom")]; + tensor k_93_pad_0 = const()[name = tensor("k_93_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_7_attn2_to_k_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_7_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(701154048)))]; + tensor k_93_cast_fp16 = conv(dilations = var_9567, groups = var_2624, pad = k_93_pad_0, pad_type = k_93_pad_type_0, strides = var_9565, weight = down_blocks_2_attentions_0_transformer_blocks_7_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_93_cast_fp16")]; + tensor var_9571 = const()[name = tensor("op_9571"), val = tensor([1, 1])]; + tensor var_9573 = const()[name = tensor("op_9573"), val = tensor([1, 1])]; + tensor v_47_pad_type_0 = const()[name = tensor("v_47_pad_type_0"), val = tensor("custom")]; + tensor v_47_pad_0 = const()[name = tensor("v_47_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_7_attn2_to_v_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_7_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(706396992)))]; + tensor v_47_cast_fp16 = conv(dilations = var_9573, groups = var_2624, pad = v_47_pad_0, pad_type = v_47_pad_type_0, strides = var_9571, weight = down_blocks_2_attentions_0_transformer_blocks_7_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_47_cast_fp16")]; + tensor var_9577_begin_0 = const()[name = tensor("op_9577_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9577_end_0 = const()[name = tensor("op_9577_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_9577_end_mask_0 = const()[name = tensor("op_9577_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9577_cast_fp16 = slice_by_index(begin = var_9577_begin_0, end = var_9577_end_0, end_mask = var_9577_end_mask_0, x = q_47_cast_fp16)[name = tensor("op_9577_cast_fp16")]; + tensor var_9581_begin_0 = const()[name = tensor("op_9581_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_9581_end_0 = const()[name = tensor("op_9581_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_9581_end_mask_0 = const()[name = tensor("op_9581_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9581_cast_fp16 = slice_by_index(begin = var_9581_begin_0, end = var_9581_end_0, end_mask = var_9581_end_mask_0, x = q_47_cast_fp16)[name = tensor("op_9581_cast_fp16")]; + tensor var_9585_begin_0 = const()[name = tensor("op_9585_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_9585_end_0 = const()[name = tensor("op_9585_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_9585_end_mask_0 = const()[name = tensor("op_9585_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9585_cast_fp16 = slice_by_index(begin = var_9585_begin_0, end = var_9585_end_0, end_mask = var_9585_end_mask_0, x = q_47_cast_fp16)[name = tensor("op_9585_cast_fp16")]; + tensor var_9589_begin_0 = const()[name = tensor("op_9589_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_9589_end_0 = const()[name = tensor("op_9589_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_9589_end_mask_0 = const()[name = tensor("op_9589_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9589_cast_fp16 = slice_by_index(begin = var_9589_begin_0, end = var_9589_end_0, end_mask = var_9589_end_mask_0, x = q_47_cast_fp16)[name = tensor("op_9589_cast_fp16")]; + tensor var_9593_begin_0 = const()[name = tensor("op_9593_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_9593_end_0 = const()[name = tensor("op_9593_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_9593_end_mask_0 = const()[name = tensor("op_9593_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9593_cast_fp16 = slice_by_index(begin = var_9593_begin_0, end = var_9593_end_0, end_mask = var_9593_end_mask_0, x = q_47_cast_fp16)[name = tensor("op_9593_cast_fp16")]; + tensor var_9597_begin_0 = const()[name = tensor("op_9597_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_9597_end_0 = const()[name = tensor("op_9597_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_9597_end_mask_0 = const()[name = tensor("op_9597_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9597_cast_fp16 = slice_by_index(begin = var_9597_begin_0, end = var_9597_end_0, end_mask = var_9597_end_mask_0, x = q_47_cast_fp16)[name = tensor("op_9597_cast_fp16")]; + tensor var_9601_begin_0 = const()[name = tensor("op_9601_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_9601_end_0 = const()[name = tensor("op_9601_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_9601_end_mask_0 = const()[name = tensor("op_9601_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9601_cast_fp16 = slice_by_index(begin = var_9601_begin_0, end = var_9601_end_0, end_mask = var_9601_end_mask_0, x = q_47_cast_fp16)[name = tensor("op_9601_cast_fp16")]; + tensor var_9605_begin_0 = const()[name = tensor("op_9605_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_9605_end_0 = const()[name = tensor("op_9605_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_9605_end_mask_0 = const()[name = tensor("op_9605_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9605_cast_fp16 = slice_by_index(begin = var_9605_begin_0, end = var_9605_end_0, end_mask = var_9605_end_mask_0, x = q_47_cast_fp16)[name = tensor("op_9605_cast_fp16")]; + tensor var_9609_begin_0 = const()[name = tensor("op_9609_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_9609_end_0 = const()[name = tensor("op_9609_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_9609_end_mask_0 = const()[name = tensor("op_9609_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9609_cast_fp16 = slice_by_index(begin = var_9609_begin_0, end = var_9609_end_0, end_mask = var_9609_end_mask_0, x = q_47_cast_fp16)[name = tensor("op_9609_cast_fp16")]; + tensor var_9613_begin_0 = const()[name = tensor("op_9613_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_9613_end_0 = const()[name = tensor("op_9613_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_9613_end_mask_0 = const()[name = tensor("op_9613_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9613_cast_fp16 = slice_by_index(begin = var_9613_begin_0, end = var_9613_end_0, end_mask = var_9613_end_mask_0, x = q_47_cast_fp16)[name = tensor("op_9613_cast_fp16")]; + tensor var_9617_begin_0 = const()[name = tensor("op_9617_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_9617_end_0 = const()[name = tensor("op_9617_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_9617_end_mask_0 = const()[name = tensor("op_9617_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9617_cast_fp16 = slice_by_index(begin = var_9617_begin_0, end = var_9617_end_0, end_mask = var_9617_end_mask_0, x = q_47_cast_fp16)[name = tensor("op_9617_cast_fp16")]; + tensor var_9621_begin_0 = const()[name = tensor("op_9621_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_9621_end_0 = const()[name = tensor("op_9621_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_9621_end_mask_0 = const()[name = tensor("op_9621_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9621_cast_fp16 = slice_by_index(begin = var_9621_begin_0, end = var_9621_end_0, end_mask = var_9621_end_mask_0, x = q_47_cast_fp16)[name = tensor("op_9621_cast_fp16")]; + tensor var_9625_begin_0 = const()[name = tensor("op_9625_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_9625_end_0 = const()[name = tensor("op_9625_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_9625_end_mask_0 = const()[name = tensor("op_9625_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9625_cast_fp16 = slice_by_index(begin = var_9625_begin_0, end = var_9625_end_0, end_mask = var_9625_end_mask_0, x = q_47_cast_fp16)[name = tensor("op_9625_cast_fp16")]; + tensor var_9629_begin_0 = const()[name = tensor("op_9629_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_9629_end_0 = const()[name = tensor("op_9629_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_9629_end_mask_0 = const()[name = tensor("op_9629_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9629_cast_fp16 = slice_by_index(begin = var_9629_begin_0, end = var_9629_end_0, end_mask = var_9629_end_mask_0, x = q_47_cast_fp16)[name = tensor("op_9629_cast_fp16")]; + tensor var_9633_begin_0 = const()[name = tensor("op_9633_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_9633_end_0 = const()[name = tensor("op_9633_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_9633_end_mask_0 = const()[name = tensor("op_9633_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9633_cast_fp16 = slice_by_index(begin = var_9633_begin_0, end = var_9633_end_0, end_mask = var_9633_end_mask_0, x = q_47_cast_fp16)[name = tensor("op_9633_cast_fp16")]; + tensor var_9637_begin_0 = const()[name = tensor("op_9637_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_9637_end_0 = const()[name = tensor("op_9637_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_9637_end_mask_0 = const()[name = tensor("op_9637_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9637_cast_fp16 = slice_by_index(begin = var_9637_begin_0, end = var_9637_end_0, end_mask = var_9637_end_mask_0, x = q_47_cast_fp16)[name = tensor("op_9637_cast_fp16")]; + tensor var_9641_begin_0 = const()[name = tensor("op_9641_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_9641_end_0 = const()[name = tensor("op_9641_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_9641_end_mask_0 = const()[name = tensor("op_9641_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9641_cast_fp16 = slice_by_index(begin = var_9641_begin_0, end = var_9641_end_0, end_mask = var_9641_end_mask_0, x = q_47_cast_fp16)[name = tensor("op_9641_cast_fp16")]; + tensor var_9645_begin_0 = const()[name = tensor("op_9645_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_9645_end_0 = const()[name = tensor("op_9645_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_9645_end_mask_0 = const()[name = tensor("op_9645_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9645_cast_fp16 = slice_by_index(begin = var_9645_begin_0, end = var_9645_end_0, end_mask = var_9645_end_mask_0, x = q_47_cast_fp16)[name = tensor("op_9645_cast_fp16")]; + tensor var_9649_begin_0 = const()[name = tensor("op_9649_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_9649_end_0 = const()[name = tensor("op_9649_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_9649_end_mask_0 = const()[name = tensor("op_9649_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9649_cast_fp16 = slice_by_index(begin = var_9649_begin_0, end = var_9649_end_0, end_mask = var_9649_end_mask_0, x = q_47_cast_fp16)[name = tensor("op_9649_cast_fp16")]; + tensor var_9653_begin_0 = const()[name = tensor("op_9653_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_9653_end_0 = const()[name = tensor("op_9653_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_9653_end_mask_0 = const()[name = tensor("op_9653_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9653_cast_fp16 = slice_by_index(begin = var_9653_begin_0, end = var_9653_end_0, end_mask = var_9653_end_mask_0, x = q_47_cast_fp16)[name = tensor("op_9653_cast_fp16")]; + tensor k_95_perm_0 = const()[name = tensor("k_95_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_9660_begin_0 = const()[name = tensor("op_9660_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9660_end_0 = const()[name = tensor("op_9660_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_9660_end_mask_0 = const()[name = tensor("op_9660_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_95_cast_fp16 = transpose(perm = k_95_perm_0, x = k_93_cast_fp16)[name = tensor("transpose_116")]; + tensor var_9660_cast_fp16 = slice_by_index(begin = var_9660_begin_0, end = var_9660_end_0, end_mask = var_9660_end_mask_0, x = k_95_cast_fp16)[name = tensor("op_9660_cast_fp16")]; + tensor var_9664_begin_0 = const()[name = tensor("op_9664_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_9664_end_0 = const()[name = tensor("op_9664_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_9664_end_mask_0 = const()[name = tensor("op_9664_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9664_cast_fp16 = slice_by_index(begin = var_9664_begin_0, end = var_9664_end_0, end_mask = var_9664_end_mask_0, x = k_95_cast_fp16)[name = tensor("op_9664_cast_fp16")]; + tensor var_9668_begin_0 = const()[name = tensor("op_9668_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_9668_end_0 = const()[name = tensor("op_9668_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_9668_end_mask_0 = const()[name = tensor("op_9668_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9668_cast_fp16 = slice_by_index(begin = var_9668_begin_0, end = var_9668_end_0, end_mask = var_9668_end_mask_0, x = k_95_cast_fp16)[name = tensor("op_9668_cast_fp16")]; + tensor var_9672_begin_0 = const()[name = tensor("op_9672_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_9672_end_0 = const()[name = tensor("op_9672_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_9672_end_mask_0 = const()[name = tensor("op_9672_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9672_cast_fp16 = slice_by_index(begin = var_9672_begin_0, end = var_9672_end_0, end_mask = var_9672_end_mask_0, x = k_95_cast_fp16)[name = tensor("op_9672_cast_fp16")]; + tensor var_9676_begin_0 = const()[name = tensor("op_9676_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_9676_end_0 = const()[name = tensor("op_9676_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_9676_end_mask_0 = const()[name = tensor("op_9676_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9676_cast_fp16 = slice_by_index(begin = var_9676_begin_0, end = var_9676_end_0, end_mask = var_9676_end_mask_0, x = k_95_cast_fp16)[name = tensor("op_9676_cast_fp16")]; + tensor var_9680_begin_0 = const()[name = tensor("op_9680_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_9680_end_0 = const()[name = tensor("op_9680_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_9680_end_mask_0 = const()[name = tensor("op_9680_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9680_cast_fp16 = slice_by_index(begin = var_9680_begin_0, end = var_9680_end_0, end_mask = var_9680_end_mask_0, x = k_95_cast_fp16)[name = tensor("op_9680_cast_fp16")]; + tensor var_9684_begin_0 = const()[name = tensor("op_9684_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_9684_end_0 = const()[name = tensor("op_9684_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_9684_end_mask_0 = const()[name = tensor("op_9684_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9684_cast_fp16 = slice_by_index(begin = var_9684_begin_0, end = var_9684_end_0, end_mask = var_9684_end_mask_0, x = k_95_cast_fp16)[name = tensor("op_9684_cast_fp16")]; + tensor var_9688_begin_0 = const()[name = tensor("op_9688_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_9688_end_0 = const()[name = tensor("op_9688_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_9688_end_mask_0 = const()[name = tensor("op_9688_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9688_cast_fp16 = slice_by_index(begin = var_9688_begin_0, end = var_9688_end_0, end_mask = var_9688_end_mask_0, x = k_95_cast_fp16)[name = tensor("op_9688_cast_fp16")]; + tensor var_9692_begin_0 = const()[name = tensor("op_9692_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_9692_end_0 = const()[name = tensor("op_9692_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_9692_end_mask_0 = const()[name = tensor("op_9692_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9692_cast_fp16 = slice_by_index(begin = var_9692_begin_0, end = var_9692_end_0, end_mask = var_9692_end_mask_0, x = k_95_cast_fp16)[name = tensor("op_9692_cast_fp16")]; + tensor var_9696_begin_0 = const()[name = tensor("op_9696_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_9696_end_0 = const()[name = tensor("op_9696_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_9696_end_mask_0 = const()[name = tensor("op_9696_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9696_cast_fp16 = slice_by_index(begin = var_9696_begin_0, end = var_9696_end_0, end_mask = var_9696_end_mask_0, x = k_95_cast_fp16)[name = tensor("op_9696_cast_fp16")]; + tensor var_9700_begin_0 = const()[name = tensor("op_9700_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_9700_end_0 = const()[name = tensor("op_9700_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_9700_end_mask_0 = const()[name = tensor("op_9700_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9700_cast_fp16 = slice_by_index(begin = var_9700_begin_0, end = var_9700_end_0, end_mask = var_9700_end_mask_0, x = k_95_cast_fp16)[name = tensor("op_9700_cast_fp16")]; + tensor var_9704_begin_0 = const()[name = tensor("op_9704_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_9704_end_0 = const()[name = tensor("op_9704_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_9704_end_mask_0 = const()[name = tensor("op_9704_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9704_cast_fp16 = slice_by_index(begin = var_9704_begin_0, end = var_9704_end_0, end_mask = var_9704_end_mask_0, x = k_95_cast_fp16)[name = tensor("op_9704_cast_fp16")]; + tensor var_9708_begin_0 = const()[name = tensor("op_9708_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_9708_end_0 = const()[name = tensor("op_9708_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_9708_end_mask_0 = const()[name = tensor("op_9708_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9708_cast_fp16 = slice_by_index(begin = var_9708_begin_0, end = var_9708_end_0, end_mask = var_9708_end_mask_0, x = k_95_cast_fp16)[name = tensor("op_9708_cast_fp16")]; + tensor var_9712_begin_0 = const()[name = tensor("op_9712_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_9712_end_0 = const()[name = tensor("op_9712_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_9712_end_mask_0 = const()[name = tensor("op_9712_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9712_cast_fp16 = slice_by_index(begin = var_9712_begin_0, end = var_9712_end_0, end_mask = var_9712_end_mask_0, x = k_95_cast_fp16)[name = tensor("op_9712_cast_fp16")]; + tensor var_9716_begin_0 = const()[name = tensor("op_9716_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_9716_end_0 = const()[name = tensor("op_9716_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_9716_end_mask_0 = const()[name = tensor("op_9716_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9716_cast_fp16 = slice_by_index(begin = var_9716_begin_0, end = var_9716_end_0, end_mask = var_9716_end_mask_0, x = k_95_cast_fp16)[name = tensor("op_9716_cast_fp16")]; + tensor var_9720_begin_0 = const()[name = tensor("op_9720_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_9720_end_0 = const()[name = tensor("op_9720_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_9720_end_mask_0 = const()[name = tensor("op_9720_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9720_cast_fp16 = slice_by_index(begin = var_9720_begin_0, end = var_9720_end_0, end_mask = var_9720_end_mask_0, x = k_95_cast_fp16)[name = tensor("op_9720_cast_fp16")]; + tensor var_9724_begin_0 = const()[name = tensor("op_9724_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_9724_end_0 = const()[name = tensor("op_9724_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_9724_end_mask_0 = const()[name = tensor("op_9724_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9724_cast_fp16 = slice_by_index(begin = var_9724_begin_0, end = var_9724_end_0, end_mask = var_9724_end_mask_0, x = k_95_cast_fp16)[name = tensor("op_9724_cast_fp16")]; + tensor var_9728_begin_0 = const()[name = tensor("op_9728_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_9728_end_0 = const()[name = tensor("op_9728_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_9728_end_mask_0 = const()[name = tensor("op_9728_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9728_cast_fp16 = slice_by_index(begin = var_9728_begin_0, end = var_9728_end_0, end_mask = var_9728_end_mask_0, x = k_95_cast_fp16)[name = tensor("op_9728_cast_fp16")]; + tensor var_9732_begin_0 = const()[name = tensor("op_9732_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_9732_end_0 = const()[name = tensor("op_9732_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_9732_end_mask_0 = const()[name = tensor("op_9732_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9732_cast_fp16 = slice_by_index(begin = var_9732_begin_0, end = var_9732_end_0, end_mask = var_9732_end_mask_0, x = k_95_cast_fp16)[name = tensor("op_9732_cast_fp16")]; + tensor var_9736_begin_0 = const()[name = tensor("op_9736_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_9736_end_0 = const()[name = tensor("op_9736_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_9736_end_mask_0 = const()[name = tensor("op_9736_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9736_cast_fp16 = slice_by_index(begin = var_9736_begin_0, end = var_9736_end_0, end_mask = var_9736_end_mask_0, x = k_95_cast_fp16)[name = tensor("op_9736_cast_fp16")]; + tensor var_9738_begin_0 = const()[name = tensor("op_9738_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9738_end_0 = const()[name = tensor("op_9738_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_9738_end_mask_0 = const()[name = tensor("op_9738_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9738_cast_fp16 = slice_by_index(begin = var_9738_begin_0, end = var_9738_end_0, end_mask = var_9738_end_mask_0, x = v_47_cast_fp16)[name = tensor("op_9738_cast_fp16")]; + tensor var_9742_begin_0 = const()[name = tensor("op_9742_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_9742_end_0 = const()[name = tensor("op_9742_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_9742_end_mask_0 = const()[name = tensor("op_9742_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9742_cast_fp16 = slice_by_index(begin = var_9742_begin_0, end = var_9742_end_0, end_mask = var_9742_end_mask_0, x = v_47_cast_fp16)[name = tensor("op_9742_cast_fp16")]; + tensor var_9746_begin_0 = const()[name = tensor("op_9746_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_9746_end_0 = const()[name = tensor("op_9746_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_9746_end_mask_0 = const()[name = tensor("op_9746_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9746_cast_fp16 = slice_by_index(begin = var_9746_begin_0, end = var_9746_end_0, end_mask = var_9746_end_mask_0, x = v_47_cast_fp16)[name = tensor("op_9746_cast_fp16")]; + tensor var_9750_begin_0 = const()[name = tensor("op_9750_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_9750_end_0 = const()[name = tensor("op_9750_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_9750_end_mask_0 = const()[name = tensor("op_9750_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9750_cast_fp16 = slice_by_index(begin = var_9750_begin_0, end = var_9750_end_0, end_mask = var_9750_end_mask_0, x = v_47_cast_fp16)[name = tensor("op_9750_cast_fp16")]; + tensor var_9754_begin_0 = const()[name = tensor("op_9754_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_9754_end_0 = const()[name = tensor("op_9754_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_9754_end_mask_0 = const()[name = tensor("op_9754_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9754_cast_fp16 = slice_by_index(begin = var_9754_begin_0, end = var_9754_end_0, end_mask = var_9754_end_mask_0, x = v_47_cast_fp16)[name = tensor("op_9754_cast_fp16")]; + tensor var_9758_begin_0 = const()[name = tensor("op_9758_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_9758_end_0 = const()[name = tensor("op_9758_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_9758_end_mask_0 = const()[name = tensor("op_9758_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9758_cast_fp16 = slice_by_index(begin = var_9758_begin_0, end = var_9758_end_0, end_mask = var_9758_end_mask_0, x = v_47_cast_fp16)[name = tensor("op_9758_cast_fp16")]; + tensor var_9762_begin_0 = const()[name = tensor("op_9762_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_9762_end_0 = const()[name = tensor("op_9762_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_9762_end_mask_0 = const()[name = tensor("op_9762_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9762_cast_fp16 = slice_by_index(begin = var_9762_begin_0, end = var_9762_end_0, end_mask = var_9762_end_mask_0, x = v_47_cast_fp16)[name = tensor("op_9762_cast_fp16")]; + tensor var_9766_begin_0 = const()[name = tensor("op_9766_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_9766_end_0 = const()[name = tensor("op_9766_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_9766_end_mask_0 = const()[name = tensor("op_9766_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9766_cast_fp16 = slice_by_index(begin = var_9766_begin_0, end = var_9766_end_0, end_mask = var_9766_end_mask_0, x = v_47_cast_fp16)[name = tensor("op_9766_cast_fp16")]; + tensor var_9770_begin_0 = const()[name = tensor("op_9770_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_9770_end_0 = const()[name = tensor("op_9770_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_9770_end_mask_0 = const()[name = tensor("op_9770_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9770_cast_fp16 = slice_by_index(begin = var_9770_begin_0, end = var_9770_end_0, end_mask = var_9770_end_mask_0, x = v_47_cast_fp16)[name = tensor("op_9770_cast_fp16")]; + tensor var_9774_begin_0 = const()[name = tensor("op_9774_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_9774_end_0 = const()[name = tensor("op_9774_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_9774_end_mask_0 = const()[name = tensor("op_9774_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9774_cast_fp16 = slice_by_index(begin = var_9774_begin_0, end = var_9774_end_0, end_mask = var_9774_end_mask_0, x = v_47_cast_fp16)[name = tensor("op_9774_cast_fp16")]; + tensor var_9778_begin_0 = const()[name = tensor("op_9778_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_9778_end_0 = const()[name = tensor("op_9778_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_9778_end_mask_0 = const()[name = tensor("op_9778_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9778_cast_fp16 = slice_by_index(begin = var_9778_begin_0, end = var_9778_end_0, end_mask = var_9778_end_mask_0, x = v_47_cast_fp16)[name = tensor("op_9778_cast_fp16")]; + tensor var_9782_begin_0 = const()[name = tensor("op_9782_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_9782_end_0 = const()[name = tensor("op_9782_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_9782_end_mask_0 = const()[name = tensor("op_9782_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9782_cast_fp16 = slice_by_index(begin = var_9782_begin_0, end = var_9782_end_0, end_mask = var_9782_end_mask_0, x = v_47_cast_fp16)[name = tensor("op_9782_cast_fp16")]; + tensor var_9786_begin_0 = const()[name = tensor("op_9786_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_9786_end_0 = const()[name = tensor("op_9786_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_9786_end_mask_0 = const()[name = tensor("op_9786_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9786_cast_fp16 = slice_by_index(begin = var_9786_begin_0, end = var_9786_end_0, end_mask = var_9786_end_mask_0, x = v_47_cast_fp16)[name = tensor("op_9786_cast_fp16")]; + tensor var_9790_begin_0 = const()[name = tensor("op_9790_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_9790_end_0 = const()[name = tensor("op_9790_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_9790_end_mask_0 = const()[name = tensor("op_9790_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9790_cast_fp16 = slice_by_index(begin = var_9790_begin_0, end = var_9790_end_0, end_mask = var_9790_end_mask_0, x = v_47_cast_fp16)[name = tensor("op_9790_cast_fp16")]; + tensor var_9794_begin_0 = const()[name = tensor("op_9794_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_9794_end_0 = const()[name = tensor("op_9794_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_9794_end_mask_0 = const()[name = tensor("op_9794_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9794_cast_fp16 = slice_by_index(begin = var_9794_begin_0, end = var_9794_end_0, end_mask = var_9794_end_mask_0, x = v_47_cast_fp16)[name = tensor("op_9794_cast_fp16")]; + tensor var_9798_begin_0 = const()[name = tensor("op_9798_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_9798_end_0 = const()[name = tensor("op_9798_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_9798_end_mask_0 = const()[name = tensor("op_9798_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9798_cast_fp16 = slice_by_index(begin = var_9798_begin_0, end = var_9798_end_0, end_mask = var_9798_end_mask_0, x = v_47_cast_fp16)[name = tensor("op_9798_cast_fp16")]; + tensor var_9802_begin_0 = const()[name = tensor("op_9802_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_9802_end_0 = const()[name = tensor("op_9802_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_9802_end_mask_0 = const()[name = tensor("op_9802_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9802_cast_fp16 = slice_by_index(begin = var_9802_begin_0, end = var_9802_end_0, end_mask = var_9802_end_mask_0, x = v_47_cast_fp16)[name = tensor("op_9802_cast_fp16")]; + tensor var_9806_begin_0 = const()[name = tensor("op_9806_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_9806_end_0 = const()[name = tensor("op_9806_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_9806_end_mask_0 = const()[name = tensor("op_9806_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9806_cast_fp16 = slice_by_index(begin = var_9806_begin_0, end = var_9806_end_0, end_mask = var_9806_end_mask_0, x = v_47_cast_fp16)[name = tensor("op_9806_cast_fp16")]; + tensor var_9810_begin_0 = const()[name = tensor("op_9810_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_9810_end_0 = const()[name = tensor("op_9810_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_9810_end_mask_0 = const()[name = tensor("op_9810_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9810_cast_fp16 = slice_by_index(begin = var_9810_begin_0, end = var_9810_end_0, end_mask = var_9810_end_mask_0, x = v_47_cast_fp16)[name = tensor("op_9810_cast_fp16")]; + tensor var_9814_begin_0 = const()[name = tensor("op_9814_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_9814_end_0 = const()[name = tensor("op_9814_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_9814_end_mask_0 = const()[name = tensor("op_9814_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9814_cast_fp16 = slice_by_index(begin = var_9814_begin_0, end = var_9814_end_0, end_mask = var_9814_end_mask_0, x = v_47_cast_fp16)[name = tensor("op_9814_cast_fp16")]; + tensor var_9818_equation_0 = const()[name = tensor("op_9818_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9818_cast_fp16 = einsum(equation = var_9818_equation_0, values = (var_9660_cast_fp16, var_9577_cast_fp16))[name = tensor("op_9818_cast_fp16")]; + tensor var_9819_to_fp16 = const()[name = tensor("op_9819_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_761_cast_fp16 = mul(x = var_9818_cast_fp16, y = var_9819_to_fp16)[name = tensor("aw_761_cast_fp16")]; + tensor var_9822_equation_0 = const()[name = tensor("op_9822_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9822_cast_fp16 = einsum(equation = var_9822_equation_0, values = (var_9664_cast_fp16, var_9581_cast_fp16))[name = tensor("op_9822_cast_fp16")]; + tensor var_9823_to_fp16 = const()[name = tensor("op_9823_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_763_cast_fp16 = mul(x = var_9822_cast_fp16, y = var_9823_to_fp16)[name = tensor("aw_763_cast_fp16")]; + tensor var_9826_equation_0 = const()[name = tensor("op_9826_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9826_cast_fp16 = einsum(equation = var_9826_equation_0, values = (var_9668_cast_fp16, var_9585_cast_fp16))[name = tensor("op_9826_cast_fp16")]; + tensor var_9827_to_fp16 = const()[name = tensor("op_9827_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_765_cast_fp16 = mul(x = var_9826_cast_fp16, y = var_9827_to_fp16)[name = tensor("aw_765_cast_fp16")]; + tensor var_9830_equation_0 = const()[name = tensor("op_9830_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9830_cast_fp16 = einsum(equation = var_9830_equation_0, values = (var_9672_cast_fp16, var_9589_cast_fp16))[name = tensor("op_9830_cast_fp16")]; + tensor var_9831_to_fp16 = const()[name = tensor("op_9831_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_767_cast_fp16 = mul(x = var_9830_cast_fp16, y = var_9831_to_fp16)[name = tensor("aw_767_cast_fp16")]; + tensor var_9834_equation_0 = const()[name = tensor("op_9834_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9834_cast_fp16 = einsum(equation = var_9834_equation_0, values = (var_9676_cast_fp16, var_9593_cast_fp16))[name = tensor("op_9834_cast_fp16")]; + tensor var_9835_to_fp16 = const()[name = tensor("op_9835_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_769_cast_fp16 = mul(x = var_9834_cast_fp16, y = var_9835_to_fp16)[name = tensor("aw_769_cast_fp16")]; + tensor var_9838_equation_0 = const()[name = tensor("op_9838_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9838_cast_fp16 = einsum(equation = var_9838_equation_0, values = (var_9680_cast_fp16, var_9597_cast_fp16))[name = tensor("op_9838_cast_fp16")]; + tensor var_9839_to_fp16 = const()[name = tensor("op_9839_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_771_cast_fp16 = mul(x = var_9838_cast_fp16, y = var_9839_to_fp16)[name = tensor("aw_771_cast_fp16")]; + tensor var_9842_equation_0 = const()[name = tensor("op_9842_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9842_cast_fp16 = einsum(equation = var_9842_equation_0, values = (var_9684_cast_fp16, var_9601_cast_fp16))[name = tensor("op_9842_cast_fp16")]; + tensor var_9843_to_fp16 = const()[name = tensor("op_9843_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_773_cast_fp16 = mul(x = var_9842_cast_fp16, y = var_9843_to_fp16)[name = tensor("aw_773_cast_fp16")]; + tensor var_9846_equation_0 = const()[name = tensor("op_9846_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9846_cast_fp16 = einsum(equation = var_9846_equation_0, values = (var_9688_cast_fp16, var_9605_cast_fp16))[name = tensor("op_9846_cast_fp16")]; + tensor var_9847_to_fp16 = const()[name = tensor("op_9847_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_775_cast_fp16 = mul(x = var_9846_cast_fp16, y = var_9847_to_fp16)[name = tensor("aw_775_cast_fp16")]; + tensor var_9850_equation_0 = const()[name = tensor("op_9850_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9850_cast_fp16 = einsum(equation = var_9850_equation_0, values = (var_9692_cast_fp16, var_9609_cast_fp16))[name = tensor("op_9850_cast_fp16")]; + tensor var_9851_to_fp16 = const()[name = tensor("op_9851_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_777_cast_fp16 = mul(x = var_9850_cast_fp16, y = var_9851_to_fp16)[name = tensor("aw_777_cast_fp16")]; + tensor var_9854_equation_0 = const()[name = tensor("op_9854_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9854_cast_fp16 = einsum(equation = var_9854_equation_0, values = (var_9696_cast_fp16, var_9613_cast_fp16))[name = tensor("op_9854_cast_fp16")]; + tensor var_9855_to_fp16 = const()[name = tensor("op_9855_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_779_cast_fp16 = mul(x = var_9854_cast_fp16, y = var_9855_to_fp16)[name = tensor("aw_779_cast_fp16")]; + tensor var_9858_equation_0 = const()[name = tensor("op_9858_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9858_cast_fp16 = einsum(equation = var_9858_equation_0, values = (var_9700_cast_fp16, var_9617_cast_fp16))[name = tensor("op_9858_cast_fp16")]; + tensor var_9859_to_fp16 = const()[name = tensor("op_9859_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_781_cast_fp16 = mul(x = var_9858_cast_fp16, y = var_9859_to_fp16)[name = tensor("aw_781_cast_fp16")]; + tensor var_9862_equation_0 = const()[name = tensor("op_9862_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9862_cast_fp16 = einsum(equation = var_9862_equation_0, values = (var_9704_cast_fp16, var_9621_cast_fp16))[name = tensor("op_9862_cast_fp16")]; + tensor var_9863_to_fp16 = const()[name = tensor("op_9863_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_783_cast_fp16 = mul(x = var_9862_cast_fp16, y = var_9863_to_fp16)[name = tensor("aw_783_cast_fp16")]; + tensor var_9866_equation_0 = const()[name = tensor("op_9866_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9866_cast_fp16 = einsum(equation = var_9866_equation_0, values = (var_9708_cast_fp16, var_9625_cast_fp16))[name = tensor("op_9866_cast_fp16")]; + tensor var_9867_to_fp16 = const()[name = tensor("op_9867_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_785_cast_fp16 = mul(x = var_9866_cast_fp16, y = var_9867_to_fp16)[name = tensor("aw_785_cast_fp16")]; + tensor var_9870_equation_0 = const()[name = tensor("op_9870_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9870_cast_fp16 = einsum(equation = var_9870_equation_0, values = (var_9712_cast_fp16, var_9629_cast_fp16))[name = tensor("op_9870_cast_fp16")]; + tensor var_9871_to_fp16 = const()[name = tensor("op_9871_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_787_cast_fp16 = mul(x = var_9870_cast_fp16, y = var_9871_to_fp16)[name = tensor("aw_787_cast_fp16")]; + tensor var_9874_equation_0 = const()[name = tensor("op_9874_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9874_cast_fp16 = einsum(equation = var_9874_equation_0, values = (var_9716_cast_fp16, var_9633_cast_fp16))[name = tensor("op_9874_cast_fp16")]; + tensor var_9875_to_fp16 = const()[name = tensor("op_9875_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_789_cast_fp16 = mul(x = var_9874_cast_fp16, y = var_9875_to_fp16)[name = tensor("aw_789_cast_fp16")]; + tensor var_9878_equation_0 = const()[name = tensor("op_9878_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9878_cast_fp16 = einsum(equation = var_9878_equation_0, values = (var_9720_cast_fp16, var_9637_cast_fp16))[name = tensor("op_9878_cast_fp16")]; + tensor var_9879_to_fp16 = const()[name = tensor("op_9879_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_791_cast_fp16 = mul(x = var_9878_cast_fp16, y = var_9879_to_fp16)[name = tensor("aw_791_cast_fp16")]; + tensor var_9882_equation_0 = const()[name = tensor("op_9882_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9882_cast_fp16 = einsum(equation = var_9882_equation_0, values = (var_9724_cast_fp16, var_9641_cast_fp16))[name = tensor("op_9882_cast_fp16")]; + tensor var_9883_to_fp16 = const()[name = tensor("op_9883_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_793_cast_fp16 = mul(x = var_9882_cast_fp16, y = var_9883_to_fp16)[name = tensor("aw_793_cast_fp16")]; + tensor var_9886_equation_0 = const()[name = tensor("op_9886_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9886_cast_fp16 = einsum(equation = var_9886_equation_0, values = (var_9728_cast_fp16, var_9645_cast_fp16))[name = tensor("op_9886_cast_fp16")]; + tensor var_9887_to_fp16 = const()[name = tensor("op_9887_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_795_cast_fp16 = mul(x = var_9886_cast_fp16, y = var_9887_to_fp16)[name = tensor("aw_795_cast_fp16")]; + tensor var_9890_equation_0 = const()[name = tensor("op_9890_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9890_cast_fp16 = einsum(equation = var_9890_equation_0, values = (var_9732_cast_fp16, var_9649_cast_fp16))[name = tensor("op_9890_cast_fp16")]; + tensor var_9891_to_fp16 = const()[name = tensor("op_9891_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_797_cast_fp16 = mul(x = var_9890_cast_fp16, y = var_9891_to_fp16)[name = tensor("aw_797_cast_fp16")]; + tensor var_9894_equation_0 = const()[name = tensor("op_9894_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9894_cast_fp16 = einsum(equation = var_9894_equation_0, values = (var_9736_cast_fp16, var_9653_cast_fp16))[name = tensor("op_9894_cast_fp16")]; + tensor var_9895_to_fp16 = const()[name = tensor("op_9895_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_799_cast_fp16 = mul(x = var_9894_cast_fp16, y = var_9895_to_fp16)[name = tensor("aw_799_cast_fp16")]; + tensor var_9897_cast_fp16 = softmax(axis = var_2624, x = aw_761_cast_fp16)[name = tensor("op_9897_cast_fp16")]; + tensor var_9898_cast_fp16 = softmax(axis = var_2624, x = aw_763_cast_fp16)[name = tensor("op_9898_cast_fp16")]; + tensor var_9899_cast_fp16 = softmax(axis = var_2624, x = aw_765_cast_fp16)[name = tensor("op_9899_cast_fp16")]; + tensor var_9900_cast_fp16 = softmax(axis = var_2624, x = aw_767_cast_fp16)[name = tensor("op_9900_cast_fp16")]; + tensor var_9901_cast_fp16 = softmax(axis = var_2624, x = aw_769_cast_fp16)[name = tensor("op_9901_cast_fp16")]; + tensor var_9902_cast_fp16 = softmax(axis = var_2624, x = aw_771_cast_fp16)[name = tensor("op_9902_cast_fp16")]; + tensor var_9903_cast_fp16 = softmax(axis = var_2624, x = aw_773_cast_fp16)[name = tensor("op_9903_cast_fp16")]; + tensor var_9904_cast_fp16 = softmax(axis = var_2624, x = aw_775_cast_fp16)[name = tensor("op_9904_cast_fp16")]; + tensor var_9905_cast_fp16 = softmax(axis = var_2624, x = aw_777_cast_fp16)[name = tensor("op_9905_cast_fp16")]; + tensor var_9906_cast_fp16 = softmax(axis = var_2624, x = aw_779_cast_fp16)[name = tensor("op_9906_cast_fp16")]; + tensor var_9907_cast_fp16 = softmax(axis = var_2624, x = aw_781_cast_fp16)[name = tensor("op_9907_cast_fp16")]; + tensor var_9908_cast_fp16 = softmax(axis = var_2624, x = aw_783_cast_fp16)[name = tensor("op_9908_cast_fp16")]; + tensor var_9909_cast_fp16 = softmax(axis = var_2624, x = aw_785_cast_fp16)[name = tensor("op_9909_cast_fp16")]; + tensor var_9910_cast_fp16 = softmax(axis = var_2624, x = aw_787_cast_fp16)[name = tensor("op_9910_cast_fp16")]; + tensor var_9911_cast_fp16 = softmax(axis = var_2624, x = aw_789_cast_fp16)[name = tensor("op_9911_cast_fp16")]; + tensor var_9912_cast_fp16 = softmax(axis = var_2624, x = aw_791_cast_fp16)[name = tensor("op_9912_cast_fp16")]; + tensor var_9913_cast_fp16 = softmax(axis = var_2624, x = aw_793_cast_fp16)[name = tensor("op_9913_cast_fp16")]; + tensor var_9914_cast_fp16 = softmax(axis = var_2624, x = aw_795_cast_fp16)[name = tensor("op_9914_cast_fp16")]; + tensor var_9915_cast_fp16 = softmax(axis = var_2624, x = aw_797_cast_fp16)[name = tensor("op_9915_cast_fp16")]; + tensor var_9916_cast_fp16 = softmax(axis = var_2624, x = aw_799_cast_fp16)[name = tensor("op_9916_cast_fp16")]; + tensor var_9918_equation_0 = const()[name = tensor("op_9918_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9918_cast_fp16 = einsum(equation = var_9918_equation_0, values = (var_9738_cast_fp16, var_9897_cast_fp16))[name = tensor("op_9918_cast_fp16")]; + tensor var_9920_equation_0 = const()[name = tensor("op_9920_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9920_cast_fp16 = einsum(equation = var_9920_equation_0, values = (var_9742_cast_fp16, var_9898_cast_fp16))[name = tensor("op_9920_cast_fp16")]; + tensor var_9922_equation_0 = const()[name = tensor("op_9922_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9922_cast_fp16 = einsum(equation = var_9922_equation_0, values = (var_9746_cast_fp16, var_9899_cast_fp16))[name = tensor("op_9922_cast_fp16")]; + tensor var_9924_equation_0 = const()[name = tensor("op_9924_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9924_cast_fp16 = einsum(equation = var_9924_equation_0, values = (var_9750_cast_fp16, var_9900_cast_fp16))[name = tensor("op_9924_cast_fp16")]; + tensor var_9926_equation_0 = const()[name = tensor("op_9926_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9926_cast_fp16 = einsum(equation = var_9926_equation_0, values = (var_9754_cast_fp16, var_9901_cast_fp16))[name = tensor("op_9926_cast_fp16")]; + tensor var_9928_equation_0 = const()[name = tensor("op_9928_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9928_cast_fp16 = einsum(equation = var_9928_equation_0, values = (var_9758_cast_fp16, var_9902_cast_fp16))[name = tensor("op_9928_cast_fp16")]; + tensor var_9930_equation_0 = const()[name = tensor("op_9930_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9930_cast_fp16 = einsum(equation = var_9930_equation_0, values = (var_9762_cast_fp16, var_9903_cast_fp16))[name = tensor("op_9930_cast_fp16")]; + tensor var_9932_equation_0 = const()[name = tensor("op_9932_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9932_cast_fp16 = einsum(equation = var_9932_equation_0, values = (var_9766_cast_fp16, var_9904_cast_fp16))[name = tensor("op_9932_cast_fp16")]; + tensor var_9934_equation_0 = const()[name = tensor("op_9934_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9934_cast_fp16 = einsum(equation = var_9934_equation_0, values = (var_9770_cast_fp16, var_9905_cast_fp16))[name = tensor("op_9934_cast_fp16")]; + tensor var_9936_equation_0 = const()[name = tensor("op_9936_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9936_cast_fp16 = einsum(equation = var_9936_equation_0, values = (var_9774_cast_fp16, var_9906_cast_fp16))[name = tensor("op_9936_cast_fp16")]; + tensor var_9938_equation_0 = const()[name = tensor("op_9938_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9938_cast_fp16 = einsum(equation = var_9938_equation_0, values = (var_9778_cast_fp16, var_9907_cast_fp16))[name = tensor("op_9938_cast_fp16")]; + tensor var_9940_equation_0 = const()[name = tensor("op_9940_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9940_cast_fp16 = einsum(equation = var_9940_equation_0, values = (var_9782_cast_fp16, var_9908_cast_fp16))[name = tensor("op_9940_cast_fp16")]; + tensor var_9942_equation_0 = const()[name = tensor("op_9942_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9942_cast_fp16 = einsum(equation = var_9942_equation_0, values = (var_9786_cast_fp16, var_9909_cast_fp16))[name = tensor("op_9942_cast_fp16")]; + tensor var_9944_equation_0 = const()[name = tensor("op_9944_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9944_cast_fp16 = einsum(equation = var_9944_equation_0, values = (var_9790_cast_fp16, var_9910_cast_fp16))[name = tensor("op_9944_cast_fp16")]; + tensor var_9946_equation_0 = const()[name = tensor("op_9946_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9946_cast_fp16 = einsum(equation = var_9946_equation_0, values = (var_9794_cast_fp16, var_9911_cast_fp16))[name = tensor("op_9946_cast_fp16")]; + tensor var_9948_equation_0 = const()[name = tensor("op_9948_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9948_cast_fp16 = einsum(equation = var_9948_equation_0, values = (var_9798_cast_fp16, var_9912_cast_fp16))[name = tensor("op_9948_cast_fp16")]; + tensor var_9950_equation_0 = const()[name = tensor("op_9950_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9950_cast_fp16 = einsum(equation = var_9950_equation_0, values = (var_9802_cast_fp16, var_9913_cast_fp16))[name = tensor("op_9950_cast_fp16")]; + tensor var_9952_equation_0 = const()[name = tensor("op_9952_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9952_cast_fp16 = einsum(equation = var_9952_equation_0, values = (var_9806_cast_fp16, var_9914_cast_fp16))[name = tensor("op_9952_cast_fp16")]; + tensor var_9954_equation_0 = const()[name = tensor("op_9954_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9954_cast_fp16 = einsum(equation = var_9954_equation_0, values = (var_9810_cast_fp16, var_9915_cast_fp16))[name = tensor("op_9954_cast_fp16")]; + tensor var_9956_equation_0 = const()[name = tensor("op_9956_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9956_cast_fp16 = einsum(equation = var_9956_equation_0, values = (var_9814_cast_fp16, var_9916_cast_fp16))[name = tensor("op_9956_cast_fp16")]; + tensor input_189_interleave_0 = const()[name = tensor("input_189_interleave_0"), val = tensor(false)]; + tensor input_189_cast_fp16 = concat(axis = var_2624, interleave = input_189_interleave_0, values = (var_9918_cast_fp16, var_9920_cast_fp16, var_9922_cast_fp16, var_9924_cast_fp16, var_9926_cast_fp16, var_9928_cast_fp16, var_9930_cast_fp16, var_9932_cast_fp16, var_9934_cast_fp16, var_9936_cast_fp16, var_9938_cast_fp16, var_9940_cast_fp16, var_9942_cast_fp16, var_9944_cast_fp16, var_9946_cast_fp16, var_9948_cast_fp16, var_9950_cast_fp16, var_9952_cast_fp16, var_9954_cast_fp16, var_9956_cast_fp16))[name = tensor("input_189_cast_fp16")]; + tensor var_9962 = const()[name = tensor("op_9962"), val = tensor([1, 1])]; + tensor var_9964 = const()[name = tensor("op_9964"), val = tensor([1, 1])]; + tensor var_9966_pad_type_0 = const()[name = tensor("op_9966_pad_type_0"), val = tensor("custom")]; + tensor var_9966_pad_0 = const()[name = tensor("op_9966_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_7_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_7_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(711639936)))]; + tensor down_blocks_2_attentions_0_transformer_blocks_7_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_7_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(714916800)))]; + tensor var_9966_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_7_attn2_to_out_0_bias_to_fp16, dilations = var_9964, groups = var_2624, pad = var_9966_pad_0, pad_type = var_9966_pad_type_0, strides = var_9962, weight = down_blocks_2_attentions_0_transformer_blocks_7_attn2_to_out_0_weight_to_fp16, x = input_189_cast_fp16)[name = tensor("op_9966_cast_fp16")]; + tensor inputs_71_cast_fp16 = add(x = var_9966_cast_fp16, y = inputs_69_cast_fp16)[name = tensor("inputs_71_cast_fp16")]; + tensor input_191_axes_0 = const()[name = tensor("input_191_axes_0"), val = tensor([1])]; + tensor input_191_gamma_0_to_fp16 = const()[name = tensor("input_191_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(714919424)))]; + tensor input_191_beta_0_to_fp16 = const()[name = tensor("input_191_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(714922048)))]; + tensor var_9976_to_fp16 = const()[name = tensor("op_9976_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_191_cast_fp16 = layer_norm(axes = input_191_axes_0, beta = input_191_beta_0_to_fp16, epsilon = var_9976_to_fp16, gamma = input_191_gamma_0_to_fp16, x = inputs_71_cast_fp16)[name = tensor("input_191_cast_fp16")]; + tensor var_9992 = const()[name = tensor("op_9992"), val = tensor([1, 1])]; + tensor var_9994 = const()[name = tensor("op_9994"), val = tensor([1, 1])]; + tensor var_9996_pad_type_0 = const()[name = tensor("op_9996_pad_type_0"), val = tensor("custom")]; + tensor var_9996_pad_0 = const()[name = tensor("op_9996_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_7_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_7_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(714924672)))]; + tensor down_blocks_2_attentions_0_transformer_blocks_7_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_7_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(741139136)))]; + tensor var_9996_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_7_ff_net_0_proj_bias_to_fp16, dilations = var_9994, groups = var_2624, pad = var_9996_pad_0, pad_type = var_9996_pad_type_0, strides = var_9992, weight = down_blocks_2_attentions_0_transformer_blocks_7_ff_net_0_proj_weight_to_fp16, x = input_191_cast_fp16)[name = tensor("op_9996_cast_fp16")]; + tensor var_9997_split_sizes_0 = const()[name = tensor("op_9997_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_9997_axis_0 = const()[name = tensor("op_9997_axis_0"), val = tensor(1)]; + tensor var_9997_cast_fp16_0, tensor var_9997_cast_fp16_1 = split(axis = var_9997_axis_0, split_sizes = var_9997_split_sizes_0, x = var_9996_cast_fp16)[name = tensor("op_9997_cast_fp16")]; + tensor var_9999_mode_0 = const()[name = tensor("op_9999_mode_0"), val = tensor("EXACT")]; + tensor var_9999_cast_fp16 = gelu(mode = var_9999_mode_0, x = var_9997_cast_fp16_1)[name = tensor("op_9999_cast_fp16")]; + tensor input_193_cast_fp16 = mul(x = var_9997_cast_fp16_0, y = var_9999_cast_fp16)[name = tensor("input_193_cast_fp16")]; + tensor var_10003 = const()[name = tensor("op_10003"), val = tensor([1, 1])]; + tensor var_10005 = const()[name = tensor("op_10005"), val = tensor([1, 1])]; + tensor var_10007_pad_type_0 = const()[name = tensor("op_10007_pad_type_0"), val = tensor("custom")]; + tensor var_10007_pad_0 = const()[name = tensor("op_10007_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_7_ff_net_2_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_7_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(741159680)))]; + tensor down_blocks_2_attentions_0_transformer_blocks_7_ff_net_2_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_7_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(754266944)))]; + tensor var_10007_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_7_ff_net_2_bias_to_fp16, dilations = var_10005, groups = var_2624, pad = var_10007_pad_0, pad_type = var_10007_pad_type_0, strides = var_10003, weight = down_blocks_2_attentions_0_transformer_blocks_7_ff_net_2_weight_to_fp16, x = input_193_cast_fp16)[name = tensor("op_10007_cast_fp16")]; + tensor inputs_73_cast_fp16 = add(x = var_10007_cast_fp16, y = inputs_71_cast_fp16)[name = tensor("inputs_73_cast_fp16")]; + tensor hidden_states_113_axes_0 = const()[name = tensor("hidden_states_113_axes_0"), val = tensor([1])]; + tensor hidden_states_113_gamma_0_to_fp16 = const()[name = tensor("hidden_states_113_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(754269568)))]; + tensor hidden_states_113_beta_0_to_fp16 = const()[name = tensor("hidden_states_113_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(754272192)))]; + tensor var_10023_to_fp16 = const()[name = tensor("op_10023_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_113_cast_fp16 = layer_norm(axes = hidden_states_113_axes_0, beta = hidden_states_113_beta_0_to_fp16, epsilon = var_10023_to_fp16, gamma = hidden_states_113_gamma_0_to_fp16, x = inputs_73_cast_fp16)[name = tensor("hidden_states_113_cast_fp16")]; + tensor var_10038 = const()[name = tensor("op_10038"), val = tensor([1, 1])]; + tensor var_10040 = const()[name = tensor("op_10040"), val = tensor([1, 1])]; + tensor q_49_pad_type_0 = const()[name = tensor("q_49_pad_type_0"), val = tensor("custom")]; + tensor q_49_pad_0 = const()[name = tensor("q_49_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_8_attn1_to_q_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_8_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(754274816)))]; + tensor q_49_cast_fp16 = conv(dilations = var_10040, groups = var_2624, pad = q_49_pad_0, pad_type = q_49_pad_type_0, strides = var_10038, weight = down_blocks_2_attentions_0_transformer_blocks_8_attn1_to_q_weight_to_fp16, x = hidden_states_113_cast_fp16)[name = tensor("q_49_cast_fp16")]; + tensor var_10044 = const()[name = tensor("op_10044"), val = tensor([1, 1])]; + tensor var_10046 = const()[name = tensor("op_10046"), val = tensor([1, 1])]; + tensor k_97_pad_type_0 = const()[name = tensor("k_97_pad_type_0"), val = tensor("custom")]; + tensor k_97_pad_0 = const()[name = tensor("k_97_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_8_attn1_to_k_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_8_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(757551680)))]; + tensor k_97_cast_fp16 = conv(dilations = var_10046, groups = var_2624, pad = k_97_pad_0, pad_type = k_97_pad_type_0, strides = var_10044, weight = down_blocks_2_attentions_0_transformer_blocks_8_attn1_to_k_weight_to_fp16, x = hidden_states_113_cast_fp16)[name = tensor("k_97_cast_fp16")]; + tensor var_10050 = const()[name = tensor("op_10050"), val = tensor([1, 1])]; + tensor var_10052 = const()[name = tensor("op_10052"), val = tensor([1, 1])]; + tensor v_49_pad_type_0 = const()[name = tensor("v_49_pad_type_0"), val = tensor("custom")]; + tensor v_49_pad_0 = const()[name = tensor("v_49_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_8_attn1_to_v_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_8_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(760828544)))]; + tensor v_49_cast_fp16 = conv(dilations = var_10052, groups = var_2624, pad = v_49_pad_0, pad_type = v_49_pad_type_0, strides = var_10050, weight = down_blocks_2_attentions_0_transformer_blocks_8_attn1_to_v_weight_to_fp16, x = hidden_states_113_cast_fp16)[name = tensor("v_49_cast_fp16")]; + tensor var_10056_begin_0 = const()[name = tensor("op_10056_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10056_end_0 = const()[name = tensor("op_10056_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_10056_end_mask_0 = const()[name = tensor("op_10056_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10056_cast_fp16 = slice_by_index(begin = var_10056_begin_0, end = var_10056_end_0, end_mask = var_10056_end_mask_0, x = q_49_cast_fp16)[name = tensor("op_10056_cast_fp16")]; + tensor var_10060_begin_0 = const()[name = tensor("op_10060_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_10060_end_0 = const()[name = tensor("op_10060_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_10060_end_mask_0 = const()[name = tensor("op_10060_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10060_cast_fp16 = slice_by_index(begin = var_10060_begin_0, end = var_10060_end_0, end_mask = var_10060_end_mask_0, x = q_49_cast_fp16)[name = tensor("op_10060_cast_fp16")]; + tensor var_10064_begin_0 = const()[name = tensor("op_10064_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_10064_end_0 = const()[name = tensor("op_10064_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_10064_end_mask_0 = const()[name = tensor("op_10064_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10064_cast_fp16 = slice_by_index(begin = var_10064_begin_0, end = var_10064_end_0, end_mask = var_10064_end_mask_0, x = q_49_cast_fp16)[name = tensor("op_10064_cast_fp16")]; + tensor var_10068_begin_0 = const()[name = tensor("op_10068_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_10068_end_0 = const()[name = tensor("op_10068_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_10068_end_mask_0 = const()[name = tensor("op_10068_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10068_cast_fp16 = slice_by_index(begin = var_10068_begin_0, end = var_10068_end_0, end_mask = var_10068_end_mask_0, x = q_49_cast_fp16)[name = tensor("op_10068_cast_fp16")]; + tensor var_10072_begin_0 = const()[name = tensor("op_10072_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_10072_end_0 = const()[name = tensor("op_10072_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_10072_end_mask_0 = const()[name = tensor("op_10072_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10072_cast_fp16 = slice_by_index(begin = var_10072_begin_0, end = var_10072_end_0, end_mask = var_10072_end_mask_0, x = q_49_cast_fp16)[name = tensor("op_10072_cast_fp16")]; + tensor var_10076_begin_0 = const()[name = tensor("op_10076_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_10076_end_0 = const()[name = tensor("op_10076_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_10076_end_mask_0 = const()[name = tensor("op_10076_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10076_cast_fp16 = slice_by_index(begin = var_10076_begin_0, end = var_10076_end_0, end_mask = var_10076_end_mask_0, x = q_49_cast_fp16)[name = tensor("op_10076_cast_fp16")]; + tensor var_10080_begin_0 = const()[name = tensor("op_10080_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_10080_end_0 = const()[name = tensor("op_10080_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_10080_end_mask_0 = const()[name = tensor("op_10080_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10080_cast_fp16 = slice_by_index(begin = var_10080_begin_0, end = var_10080_end_0, end_mask = var_10080_end_mask_0, x = q_49_cast_fp16)[name = tensor("op_10080_cast_fp16")]; + tensor var_10084_begin_0 = const()[name = tensor("op_10084_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_10084_end_0 = const()[name = tensor("op_10084_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_10084_end_mask_0 = const()[name = tensor("op_10084_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10084_cast_fp16 = slice_by_index(begin = var_10084_begin_0, end = var_10084_end_0, end_mask = var_10084_end_mask_0, x = q_49_cast_fp16)[name = tensor("op_10084_cast_fp16")]; + tensor var_10088_begin_0 = const()[name = tensor("op_10088_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_10088_end_0 = const()[name = tensor("op_10088_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_10088_end_mask_0 = const()[name = tensor("op_10088_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10088_cast_fp16 = slice_by_index(begin = var_10088_begin_0, end = var_10088_end_0, end_mask = var_10088_end_mask_0, x = q_49_cast_fp16)[name = tensor("op_10088_cast_fp16")]; + tensor var_10092_begin_0 = const()[name = tensor("op_10092_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_10092_end_0 = const()[name = tensor("op_10092_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_10092_end_mask_0 = const()[name = tensor("op_10092_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10092_cast_fp16 = slice_by_index(begin = var_10092_begin_0, end = var_10092_end_0, end_mask = var_10092_end_mask_0, x = q_49_cast_fp16)[name = tensor("op_10092_cast_fp16")]; + tensor var_10096_begin_0 = const()[name = tensor("op_10096_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_10096_end_0 = const()[name = tensor("op_10096_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_10096_end_mask_0 = const()[name = tensor("op_10096_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10096_cast_fp16 = slice_by_index(begin = var_10096_begin_0, end = var_10096_end_0, end_mask = var_10096_end_mask_0, x = q_49_cast_fp16)[name = tensor("op_10096_cast_fp16")]; + tensor var_10100_begin_0 = const()[name = tensor("op_10100_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_10100_end_0 = const()[name = tensor("op_10100_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_10100_end_mask_0 = const()[name = tensor("op_10100_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10100_cast_fp16 = slice_by_index(begin = var_10100_begin_0, end = var_10100_end_0, end_mask = var_10100_end_mask_0, x = q_49_cast_fp16)[name = tensor("op_10100_cast_fp16")]; + tensor var_10104_begin_0 = const()[name = tensor("op_10104_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_10104_end_0 = const()[name = tensor("op_10104_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_10104_end_mask_0 = const()[name = tensor("op_10104_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10104_cast_fp16 = slice_by_index(begin = var_10104_begin_0, end = var_10104_end_0, end_mask = var_10104_end_mask_0, x = q_49_cast_fp16)[name = tensor("op_10104_cast_fp16")]; + tensor var_10108_begin_0 = const()[name = tensor("op_10108_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_10108_end_0 = const()[name = tensor("op_10108_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_10108_end_mask_0 = const()[name = tensor("op_10108_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10108_cast_fp16 = slice_by_index(begin = var_10108_begin_0, end = var_10108_end_0, end_mask = var_10108_end_mask_0, x = q_49_cast_fp16)[name = tensor("op_10108_cast_fp16")]; + tensor var_10112_begin_0 = const()[name = tensor("op_10112_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_10112_end_0 = const()[name = tensor("op_10112_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_10112_end_mask_0 = const()[name = tensor("op_10112_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10112_cast_fp16 = slice_by_index(begin = var_10112_begin_0, end = var_10112_end_0, end_mask = var_10112_end_mask_0, x = q_49_cast_fp16)[name = tensor("op_10112_cast_fp16")]; + tensor var_10116_begin_0 = const()[name = tensor("op_10116_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_10116_end_0 = const()[name = tensor("op_10116_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_10116_end_mask_0 = const()[name = tensor("op_10116_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10116_cast_fp16 = slice_by_index(begin = var_10116_begin_0, end = var_10116_end_0, end_mask = var_10116_end_mask_0, x = q_49_cast_fp16)[name = tensor("op_10116_cast_fp16")]; + tensor var_10120_begin_0 = const()[name = tensor("op_10120_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_10120_end_0 = const()[name = tensor("op_10120_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_10120_end_mask_0 = const()[name = tensor("op_10120_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10120_cast_fp16 = slice_by_index(begin = var_10120_begin_0, end = var_10120_end_0, end_mask = var_10120_end_mask_0, x = q_49_cast_fp16)[name = tensor("op_10120_cast_fp16")]; + tensor var_10124_begin_0 = const()[name = tensor("op_10124_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_10124_end_0 = const()[name = tensor("op_10124_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_10124_end_mask_0 = const()[name = tensor("op_10124_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10124_cast_fp16 = slice_by_index(begin = var_10124_begin_0, end = var_10124_end_0, end_mask = var_10124_end_mask_0, x = q_49_cast_fp16)[name = tensor("op_10124_cast_fp16")]; + tensor var_10128_begin_0 = const()[name = tensor("op_10128_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_10128_end_0 = const()[name = tensor("op_10128_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_10128_end_mask_0 = const()[name = tensor("op_10128_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10128_cast_fp16 = slice_by_index(begin = var_10128_begin_0, end = var_10128_end_0, end_mask = var_10128_end_mask_0, x = q_49_cast_fp16)[name = tensor("op_10128_cast_fp16")]; + tensor var_10132_begin_0 = const()[name = tensor("op_10132_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_10132_end_0 = const()[name = tensor("op_10132_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_10132_end_mask_0 = const()[name = tensor("op_10132_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10132_cast_fp16 = slice_by_index(begin = var_10132_begin_0, end = var_10132_end_0, end_mask = var_10132_end_mask_0, x = q_49_cast_fp16)[name = tensor("op_10132_cast_fp16")]; + tensor k_99_perm_0 = const()[name = tensor("k_99_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_10139_begin_0 = const()[name = tensor("op_10139_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10139_end_0 = const()[name = tensor("op_10139_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_10139_end_mask_0 = const()[name = tensor("op_10139_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_99_cast_fp16 = transpose(perm = k_99_perm_0, x = k_97_cast_fp16)[name = tensor("transpose_115")]; + tensor var_10139_cast_fp16 = slice_by_index(begin = var_10139_begin_0, end = var_10139_end_0, end_mask = var_10139_end_mask_0, x = k_99_cast_fp16)[name = tensor("op_10139_cast_fp16")]; + tensor var_10143_begin_0 = const()[name = tensor("op_10143_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_10143_end_0 = const()[name = tensor("op_10143_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_10143_end_mask_0 = const()[name = tensor("op_10143_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10143_cast_fp16 = slice_by_index(begin = var_10143_begin_0, end = var_10143_end_0, end_mask = var_10143_end_mask_0, x = k_99_cast_fp16)[name = tensor("op_10143_cast_fp16")]; + tensor var_10147_begin_0 = const()[name = tensor("op_10147_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_10147_end_0 = const()[name = tensor("op_10147_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_10147_end_mask_0 = const()[name = tensor("op_10147_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10147_cast_fp16 = slice_by_index(begin = var_10147_begin_0, end = var_10147_end_0, end_mask = var_10147_end_mask_0, x = k_99_cast_fp16)[name = tensor("op_10147_cast_fp16")]; + tensor var_10151_begin_0 = const()[name = tensor("op_10151_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_10151_end_0 = const()[name = tensor("op_10151_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_10151_end_mask_0 = const()[name = tensor("op_10151_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10151_cast_fp16 = slice_by_index(begin = var_10151_begin_0, end = var_10151_end_0, end_mask = var_10151_end_mask_0, x = k_99_cast_fp16)[name = tensor("op_10151_cast_fp16")]; + tensor var_10155_begin_0 = const()[name = tensor("op_10155_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_10155_end_0 = const()[name = tensor("op_10155_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_10155_end_mask_0 = const()[name = tensor("op_10155_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10155_cast_fp16 = slice_by_index(begin = var_10155_begin_0, end = var_10155_end_0, end_mask = var_10155_end_mask_0, x = k_99_cast_fp16)[name = tensor("op_10155_cast_fp16")]; + tensor var_10159_begin_0 = const()[name = tensor("op_10159_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_10159_end_0 = const()[name = tensor("op_10159_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_10159_end_mask_0 = const()[name = tensor("op_10159_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10159_cast_fp16 = slice_by_index(begin = var_10159_begin_0, end = var_10159_end_0, end_mask = var_10159_end_mask_0, x = k_99_cast_fp16)[name = tensor("op_10159_cast_fp16")]; + tensor var_10163_begin_0 = const()[name = tensor("op_10163_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_10163_end_0 = const()[name = tensor("op_10163_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_10163_end_mask_0 = const()[name = tensor("op_10163_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10163_cast_fp16 = slice_by_index(begin = var_10163_begin_0, end = var_10163_end_0, end_mask = var_10163_end_mask_0, x = k_99_cast_fp16)[name = tensor("op_10163_cast_fp16")]; + tensor var_10167_begin_0 = const()[name = tensor("op_10167_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_10167_end_0 = const()[name = tensor("op_10167_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_10167_end_mask_0 = const()[name = tensor("op_10167_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10167_cast_fp16 = slice_by_index(begin = var_10167_begin_0, end = var_10167_end_0, end_mask = var_10167_end_mask_0, x = k_99_cast_fp16)[name = tensor("op_10167_cast_fp16")]; + tensor var_10171_begin_0 = const()[name = tensor("op_10171_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_10171_end_0 = const()[name = tensor("op_10171_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_10171_end_mask_0 = const()[name = tensor("op_10171_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10171_cast_fp16 = slice_by_index(begin = var_10171_begin_0, end = var_10171_end_0, end_mask = var_10171_end_mask_0, x = k_99_cast_fp16)[name = tensor("op_10171_cast_fp16")]; + tensor var_10175_begin_0 = const()[name = tensor("op_10175_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_10175_end_0 = const()[name = tensor("op_10175_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_10175_end_mask_0 = const()[name = tensor("op_10175_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10175_cast_fp16 = slice_by_index(begin = var_10175_begin_0, end = var_10175_end_0, end_mask = var_10175_end_mask_0, x = k_99_cast_fp16)[name = tensor("op_10175_cast_fp16")]; + tensor var_10179_begin_0 = const()[name = tensor("op_10179_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_10179_end_0 = const()[name = tensor("op_10179_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_10179_end_mask_0 = const()[name = tensor("op_10179_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10179_cast_fp16 = slice_by_index(begin = var_10179_begin_0, end = var_10179_end_0, end_mask = var_10179_end_mask_0, x = k_99_cast_fp16)[name = tensor("op_10179_cast_fp16")]; + tensor var_10183_begin_0 = const()[name = tensor("op_10183_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_10183_end_0 = const()[name = tensor("op_10183_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_10183_end_mask_0 = const()[name = tensor("op_10183_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10183_cast_fp16 = slice_by_index(begin = var_10183_begin_0, end = var_10183_end_0, end_mask = var_10183_end_mask_0, x = k_99_cast_fp16)[name = tensor("op_10183_cast_fp16")]; + tensor var_10187_begin_0 = const()[name = tensor("op_10187_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_10187_end_0 = const()[name = tensor("op_10187_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_10187_end_mask_0 = const()[name = tensor("op_10187_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10187_cast_fp16 = slice_by_index(begin = var_10187_begin_0, end = var_10187_end_0, end_mask = var_10187_end_mask_0, x = k_99_cast_fp16)[name = tensor("op_10187_cast_fp16")]; + tensor var_10191_begin_0 = const()[name = tensor("op_10191_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_10191_end_0 = const()[name = tensor("op_10191_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_10191_end_mask_0 = const()[name = tensor("op_10191_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10191_cast_fp16 = slice_by_index(begin = var_10191_begin_0, end = var_10191_end_0, end_mask = var_10191_end_mask_0, x = k_99_cast_fp16)[name = tensor("op_10191_cast_fp16")]; + tensor var_10195_begin_0 = const()[name = tensor("op_10195_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_10195_end_0 = const()[name = tensor("op_10195_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_10195_end_mask_0 = const()[name = tensor("op_10195_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10195_cast_fp16 = slice_by_index(begin = var_10195_begin_0, end = var_10195_end_0, end_mask = var_10195_end_mask_0, x = k_99_cast_fp16)[name = tensor("op_10195_cast_fp16")]; + tensor var_10199_begin_0 = const()[name = tensor("op_10199_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_10199_end_0 = const()[name = tensor("op_10199_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_10199_end_mask_0 = const()[name = tensor("op_10199_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10199_cast_fp16 = slice_by_index(begin = var_10199_begin_0, end = var_10199_end_0, end_mask = var_10199_end_mask_0, x = k_99_cast_fp16)[name = tensor("op_10199_cast_fp16")]; + tensor var_10203_begin_0 = const()[name = tensor("op_10203_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_10203_end_0 = const()[name = tensor("op_10203_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_10203_end_mask_0 = const()[name = tensor("op_10203_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10203_cast_fp16 = slice_by_index(begin = var_10203_begin_0, end = var_10203_end_0, end_mask = var_10203_end_mask_0, x = k_99_cast_fp16)[name = tensor("op_10203_cast_fp16")]; + tensor var_10207_begin_0 = const()[name = tensor("op_10207_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_10207_end_0 = const()[name = tensor("op_10207_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_10207_end_mask_0 = const()[name = tensor("op_10207_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10207_cast_fp16 = slice_by_index(begin = var_10207_begin_0, end = var_10207_end_0, end_mask = var_10207_end_mask_0, x = k_99_cast_fp16)[name = tensor("op_10207_cast_fp16")]; + tensor var_10211_begin_0 = const()[name = tensor("op_10211_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_10211_end_0 = const()[name = tensor("op_10211_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_10211_end_mask_0 = const()[name = tensor("op_10211_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10211_cast_fp16 = slice_by_index(begin = var_10211_begin_0, end = var_10211_end_0, end_mask = var_10211_end_mask_0, x = k_99_cast_fp16)[name = tensor("op_10211_cast_fp16")]; + tensor var_10215_begin_0 = const()[name = tensor("op_10215_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_10215_end_0 = const()[name = tensor("op_10215_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_10215_end_mask_0 = const()[name = tensor("op_10215_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10215_cast_fp16 = slice_by_index(begin = var_10215_begin_0, end = var_10215_end_0, end_mask = var_10215_end_mask_0, x = k_99_cast_fp16)[name = tensor("op_10215_cast_fp16")]; + tensor var_10217_begin_0 = const()[name = tensor("op_10217_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10217_end_0 = const()[name = tensor("op_10217_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_10217_end_mask_0 = const()[name = tensor("op_10217_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10217_cast_fp16 = slice_by_index(begin = var_10217_begin_0, end = var_10217_end_0, end_mask = var_10217_end_mask_0, x = v_49_cast_fp16)[name = tensor("op_10217_cast_fp16")]; + tensor var_10221_begin_0 = const()[name = tensor("op_10221_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_10221_end_0 = const()[name = tensor("op_10221_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_10221_end_mask_0 = const()[name = tensor("op_10221_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10221_cast_fp16 = slice_by_index(begin = var_10221_begin_0, end = var_10221_end_0, end_mask = var_10221_end_mask_0, x = v_49_cast_fp16)[name = tensor("op_10221_cast_fp16")]; + tensor var_10225_begin_0 = const()[name = tensor("op_10225_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_10225_end_0 = const()[name = tensor("op_10225_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_10225_end_mask_0 = const()[name = tensor("op_10225_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10225_cast_fp16 = slice_by_index(begin = var_10225_begin_0, end = var_10225_end_0, end_mask = var_10225_end_mask_0, x = v_49_cast_fp16)[name = tensor("op_10225_cast_fp16")]; + tensor var_10229_begin_0 = const()[name = tensor("op_10229_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_10229_end_0 = const()[name = tensor("op_10229_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_10229_end_mask_0 = const()[name = tensor("op_10229_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10229_cast_fp16 = slice_by_index(begin = var_10229_begin_0, end = var_10229_end_0, end_mask = var_10229_end_mask_0, x = v_49_cast_fp16)[name = tensor("op_10229_cast_fp16")]; + tensor var_10233_begin_0 = const()[name = tensor("op_10233_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_10233_end_0 = const()[name = tensor("op_10233_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_10233_end_mask_0 = const()[name = tensor("op_10233_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10233_cast_fp16 = slice_by_index(begin = var_10233_begin_0, end = var_10233_end_0, end_mask = var_10233_end_mask_0, x = v_49_cast_fp16)[name = tensor("op_10233_cast_fp16")]; + tensor var_10237_begin_0 = const()[name = tensor("op_10237_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_10237_end_0 = const()[name = tensor("op_10237_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_10237_end_mask_0 = const()[name = tensor("op_10237_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10237_cast_fp16 = slice_by_index(begin = var_10237_begin_0, end = var_10237_end_0, end_mask = var_10237_end_mask_0, x = v_49_cast_fp16)[name = tensor("op_10237_cast_fp16")]; + tensor var_10241_begin_0 = const()[name = tensor("op_10241_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_10241_end_0 = const()[name = tensor("op_10241_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_10241_end_mask_0 = const()[name = tensor("op_10241_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10241_cast_fp16 = slice_by_index(begin = var_10241_begin_0, end = var_10241_end_0, end_mask = var_10241_end_mask_0, x = v_49_cast_fp16)[name = tensor("op_10241_cast_fp16")]; + tensor var_10245_begin_0 = const()[name = tensor("op_10245_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_10245_end_0 = const()[name = tensor("op_10245_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_10245_end_mask_0 = const()[name = tensor("op_10245_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10245_cast_fp16 = slice_by_index(begin = var_10245_begin_0, end = var_10245_end_0, end_mask = var_10245_end_mask_0, x = v_49_cast_fp16)[name = tensor("op_10245_cast_fp16")]; + tensor var_10249_begin_0 = const()[name = tensor("op_10249_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_10249_end_0 = const()[name = tensor("op_10249_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_10249_end_mask_0 = const()[name = tensor("op_10249_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10249_cast_fp16 = slice_by_index(begin = var_10249_begin_0, end = var_10249_end_0, end_mask = var_10249_end_mask_0, x = v_49_cast_fp16)[name = tensor("op_10249_cast_fp16")]; + tensor var_10253_begin_0 = const()[name = tensor("op_10253_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_10253_end_0 = const()[name = tensor("op_10253_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_10253_end_mask_0 = const()[name = tensor("op_10253_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10253_cast_fp16 = slice_by_index(begin = var_10253_begin_0, end = var_10253_end_0, end_mask = var_10253_end_mask_0, x = v_49_cast_fp16)[name = tensor("op_10253_cast_fp16")]; + tensor var_10257_begin_0 = const()[name = tensor("op_10257_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_10257_end_0 = const()[name = tensor("op_10257_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_10257_end_mask_0 = const()[name = tensor("op_10257_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10257_cast_fp16 = slice_by_index(begin = var_10257_begin_0, end = var_10257_end_0, end_mask = var_10257_end_mask_0, x = v_49_cast_fp16)[name = tensor("op_10257_cast_fp16")]; + tensor var_10261_begin_0 = const()[name = tensor("op_10261_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_10261_end_0 = const()[name = tensor("op_10261_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_10261_end_mask_0 = const()[name = tensor("op_10261_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10261_cast_fp16 = slice_by_index(begin = var_10261_begin_0, end = var_10261_end_0, end_mask = var_10261_end_mask_0, x = v_49_cast_fp16)[name = tensor("op_10261_cast_fp16")]; + tensor var_10265_begin_0 = const()[name = tensor("op_10265_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_10265_end_0 = const()[name = tensor("op_10265_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_10265_end_mask_0 = const()[name = tensor("op_10265_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10265_cast_fp16 = slice_by_index(begin = var_10265_begin_0, end = var_10265_end_0, end_mask = var_10265_end_mask_0, x = v_49_cast_fp16)[name = tensor("op_10265_cast_fp16")]; + tensor var_10269_begin_0 = const()[name = tensor("op_10269_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_10269_end_0 = const()[name = tensor("op_10269_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_10269_end_mask_0 = const()[name = tensor("op_10269_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10269_cast_fp16 = slice_by_index(begin = var_10269_begin_0, end = var_10269_end_0, end_mask = var_10269_end_mask_0, x = v_49_cast_fp16)[name = tensor("op_10269_cast_fp16")]; + tensor var_10273_begin_0 = const()[name = tensor("op_10273_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_10273_end_0 = const()[name = tensor("op_10273_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_10273_end_mask_0 = const()[name = tensor("op_10273_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10273_cast_fp16 = slice_by_index(begin = var_10273_begin_0, end = var_10273_end_0, end_mask = var_10273_end_mask_0, x = v_49_cast_fp16)[name = tensor("op_10273_cast_fp16")]; + tensor var_10277_begin_0 = const()[name = tensor("op_10277_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_10277_end_0 = const()[name = tensor("op_10277_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_10277_end_mask_0 = const()[name = tensor("op_10277_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10277_cast_fp16 = slice_by_index(begin = var_10277_begin_0, end = var_10277_end_0, end_mask = var_10277_end_mask_0, x = v_49_cast_fp16)[name = tensor("op_10277_cast_fp16")]; + tensor var_10281_begin_0 = const()[name = tensor("op_10281_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_10281_end_0 = const()[name = tensor("op_10281_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_10281_end_mask_0 = const()[name = tensor("op_10281_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10281_cast_fp16 = slice_by_index(begin = var_10281_begin_0, end = var_10281_end_0, end_mask = var_10281_end_mask_0, x = v_49_cast_fp16)[name = tensor("op_10281_cast_fp16")]; + tensor var_10285_begin_0 = const()[name = tensor("op_10285_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_10285_end_0 = const()[name = tensor("op_10285_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_10285_end_mask_0 = const()[name = tensor("op_10285_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10285_cast_fp16 = slice_by_index(begin = var_10285_begin_0, end = var_10285_end_0, end_mask = var_10285_end_mask_0, x = v_49_cast_fp16)[name = tensor("op_10285_cast_fp16")]; + tensor var_10289_begin_0 = const()[name = tensor("op_10289_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_10289_end_0 = const()[name = tensor("op_10289_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_10289_end_mask_0 = const()[name = tensor("op_10289_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10289_cast_fp16 = slice_by_index(begin = var_10289_begin_0, end = var_10289_end_0, end_mask = var_10289_end_mask_0, x = v_49_cast_fp16)[name = tensor("op_10289_cast_fp16")]; + tensor var_10293_begin_0 = const()[name = tensor("op_10293_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_10293_end_0 = const()[name = tensor("op_10293_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_10293_end_mask_0 = const()[name = tensor("op_10293_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10293_cast_fp16 = slice_by_index(begin = var_10293_begin_0, end = var_10293_end_0, end_mask = var_10293_end_mask_0, x = v_49_cast_fp16)[name = tensor("op_10293_cast_fp16")]; + tensor var_10297_equation_0 = const()[name = tensor("op_10297_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10297_cast_fp16 = einsum(equation = var_10297_equation_0, values = (var_10139_cast_fp16, var_10056_cast_fp16))[name = tensor("op_10297_cast_fp16")]; + tensor var_10298_to_fp16 = const()[name = tensor("op_10298_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_801_cast_fp16 = mul(x = var_10297_cast_fp16, y = var_10298_to_fp16)[name = tensor("aw_801_cast_fp16")]; + tensor var_10301_equation_0 = const()[name = tensor("op_10301_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10301_cast_fp16 = einsum(equation = var_10301_equation_0, values = (var_10143_cast_fp16, var_10060_cast_fp16))[name = tensor("op_10301_cast_fp16")]; + tensor var_10302_to_fp16 = const()[name = tensor("op_10302_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_803_cast_fp16 = mul(x = var_10301_cast_fp16, y = var_10302_to_fp16)[name = tensor("aw_803_cast_fp16")]; + tensor var_10305_equation_0 = const()[name = tensor("op_10305_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10305_cast_fp16 = einsum(equation = var_10305_equation_0, values = (var_10147_cast_fp16, var_10064_cast_fp16))[name = tensor("op_10305_cast_fp16")]; + tensor var_10306_to_fp16 = const()[name = tensor("op_10306_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_805_cast_fp16 = mul(x = var_10305_cast_fp16, y = var_10306_to_fp16)[name = tensor("aw_805_cast_fp16")]; + tensor var_10309_equation_0 = const()[name = tensor("op_10309_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10309_cast_fp16 = einsum(equation = var_10309_equation_0, values = (var_10151_cast_fp16, var_10068_cast_fp16))[name = tensor("op_10309_cast_fp16")]; + tensor var_10310_to_fp16 = const()[name = tensor("op_10310_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_807_cast_fp16 = mul(x = var_10309_cast_fp16, y = var_10310_to_fp16)[name = tensor("aw_807_cast_fp16")]; + tensor var_10313_equation_0 = const()[name = tensor("op_10313_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10313_cast_fp16 = einsum(equation = var_10313_equation_0, values = (var_10155_cast_fp16, var_10072_cast_fp16))[name = tensor("op_10313_cast_fp16")]; + tensor var_10314_to_fp16 = const()[name = tensor("op_10314_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_809_cast_fp16 = mul(x = var_10313_cast_fp16, y = var_10314_to_fp16)[name = tensor("aw_809_cast_fp16")]; + tensor var_10317_equation_0 = const()[name = tensor("op_10317_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10317_cast_fp16 = einsum(equation = var_10317_equation_0, values = (var_10159_cast_fp16, var_10076_cast_fp16))[name = tensor("op_10317_cast_fp16")]; + tensor var_10318_to_fp16 = const()[name = tensor("op_10318_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_811_cast_fp16 = mul(x = var_10317_cast_fp16, y = var_10318_to_fp16)[name = tensor("aw_811_cast_fp16")]; + tensor var_10321_equation_0 = const()[name = tensor("op_10321_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10321_cast_fp16 = einsum(equation = var_10321_equation_0, values = (var_10163_cast_fp16, var_10080_cast_fp16))[name = tensor("op_10321_cast_fp16")]; + tensor var_10322_to_fp16 = const()[name = tensor("op_10322_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_813_cast_fp16 = mul(x = var_10321_cast_fp16, y = var_10322_to_fp16)[name = tensor("aw_813_cast_fp16")]; + tensor var_10325_equation_0 = const()[name = tensor("op_10325_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10325_cast_fp16 = einsum(equation = var_10325_equation_0, values = (var_10167_cast_fp16, var_10084_cast_fp16))[name = tensor("op_10325_cast_fp16")]; + tensor var_10326_to_fp16 = const()[name = tensor("op_10326_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_815_cast_fp16 = mul(x = var_10325_cast_fp16, y = var_10326_to_fp16)[name = tensor("aw_815_cast_fp16")]; + tensor var_10329_equation_0 = const()[name = tensor("op_10329_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10329_cast_fp16 = einsum(equation = var_10329_equation_0, values = (var_10171_cast_fp16, var_10088_cast_fp16))[name = tensor("op_10329_cast_fp16")]; + tensor var_10330_to_fp16 = const()[name = tensor("op_10330_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_817_cast_fp16 = mul(x = var_10329_cast_fp16, y = var_10330_to_fp16)[name = tensor("aw_817_cast_fp16")]; + tensor var_10333_equation_0 = const()[name = tensor("op_10333_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10333_cast_fp16 = einsum(equation = var_10333_equation_0, values = (var_10175_cast_fp16, var_10092_cast_fp16))[name = tensor("op_10333_cast_fp16")]; + tensor var_10334_to_fp16 = const()[name = tensor("op_10334_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_819_cast_fp16 = mul(x = var_10333_cast_fp16, y = var_10334_to_fp16)[name = tensor("aw_819_cast_fp16")]; + tensor var_10337_equation_0 = const()[name = tensor("op_10337_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10337_cast_fp16 = einsum(equation = var_10337_equation_0, values = (var_10179_cast_fp16, var_10096_cast_fp16))[name = tensor("op_10337_cast_fp16")]; + tensor var_10338_to_fp16 = const()[name = tensor("op_10338_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_821_cast_fp16 = mul(x = var_10337_cast_fp16, y = var_10338_to_fp16)[name = tensor("aw_821_cast_fp16")]; + tensor var_10341_equation_0 = const()[name = tensor("op_10341_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10341_cast_fp16 = einsum(equation = var_10341_equation_0, values = (var_10183_cast_fp16, var_10100_cast_fp16))[name = tensor("op_10341_cast_fp16")]; + tensor var_10342_to_fp16 = const()[name = tensor("op_10342_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_823_cast_fp16 = mul(x = var_10341_cast_fp16, y = var_10342_to_fp16)[name = tensor("aw_823_cast_fp16")]; + tensor var_10345_equation_0 = const()[name = tensor("op_10345_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10345_cast_fp16 = einsum(equation = var_10345_equation_0, values = (var_10187_cast_fp16, var_10104_cast_fp16))[name = tensor("op_10345_cast_fp16")]; + tensor var_10346_to_fp16 = const()[name = tensor("op_10346_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_825_cast_fp16 = mul(x = var_10345_cast_fp16, y = var_10346_to_fp16)[name = tensor("aw_825_cast_fp16")]; + tensor var_10349_equation_0 = const()[name = tensor("op_10349_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10349_cast_fp16 = einsum(equation = var_10349_equation_0, values = (var_10191_cast_fp16, var_10108_cast_fp16))[name = tensor("op_10349_cast_fp16")]; + tensor var_10350_to_fp16 = const()[name = tensor("op_10350_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_827_cast_fp16 = mul(x = var_10349_cast_fp16, y = var_10350_to_fp16)[name = tensor("aw_827_cast_fp16")]; + tensor var_10353_equation_0 = const()[name = tensor("op_10353_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10353_cast_fp16 = einsum(equation = var_10353_equation_0, values = (var_10195_cast_fp16, var_10112_cast_fp16))[name = tensor("op_10353_cast_fp16")]; + tensor var_10354_to_fp16 = const()[name = tensor("op_10354_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_829_cast_fp16 = mul(x = var_10353_cast_fp16, y = var_10354_to_fp16)[name = tensor("aw_829_cast_fp16")]; + tensor var_10357_equation_0 = const()[name = tensor("op_10357_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10357_cast_fp16 = einsum(equation = var_10357_equation_0, values = (var_10199_cast_fp16, var_10116_cast_fp16))[name = tensor("op_10357_cast_fp16")]; + tensor var_10358_to_fp16 = const()[name = tensor("op_10358_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_831_cast_fp16 = mul(x = var_10357_cast_fp16, y = var_10358_to_fp16)[name = tensor("aw_831_cast_fp16")]; + tensor var_10361_equation_0 = const()[name = tensor("op_10361_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10361_cast_fp16 = einsum(equation = var_10361_equation_0, values = (var_10203_cast_fp16, var_10120_cast_fp16))[name = tensor("op_10361_cast_fp16")]; + tensor var_10362_to_fp16 = const()[name = tensor("op_10362_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_833_cast_fp16 = mul(x = var_10361_cast_fp16, y = var_10362_to_fp16)[name = tensor("aw_833_cast_fp16")]; + tensor var_10365_equation_0 = const()[name = tensor("op_10365_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10365_cast_fp16 = einsum(equation = var_10365_equation_0, values = (var_10207_cast_fp16, var_10124_cast_fp16))[name = tensor("op_10365_cast_fp16")]; + tensor var_10366_to_fp16 = const()[name = tensor("op_10366_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_835_cast_fp16 = mul(x = var_10365_cast_fp16, y = var_10366_to_fp16)[name = tensor("aw_835_cast_fp16")]; + tensor var_10369_equation_0 = const()[name = tensor("op_10369_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10369_cast_fp16 = einsum(equation = var_10369_equation_0, values = (var_10211_cast_fp16, var_10128_cast_fp16))[name = tensor("op_10369_cast_fp16")]; + tensor var_10370_to_fp16 = const()[name = tensor("op_10370_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_837_cast_fp16 = mul(x = var_10369_cast_fp16, y = var_10370_to_fp16)[name = tensor("aw_837_cast_fp16")]; + tensor var_10373_equation_0 = const()[name = tensor("op_10373_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10373_cast_fp16 = einsum(equation = var_10373_equation_0, values = (var_10215_cast_fp16, var_10132_cast_fp16))[name = tensor("op_10373_cast_fp16")]; + tensor var_10374_to_fp16 = const()[name = tensor("op_10374_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_839_cast_fp16 = mul(x = var_10373_cast_fp16, y = var_10374_to_fp16)[name = tensor("aw_839_cast_fp16")]; + tensor var_10376_cast_fp16 = softmax(axis = var_2624, x = aw_801_cast_fp16)[name = tensor("op_10376_cast_fp16")]; + tensor var_10377_cast_fp16 = softmax(axis = var_2624, x = aw_803_cast_fp16)[name = tensor("op_10377_cast_fp16")]; + tensor var_10378_cast_fp16 = softmax(axis = var_2624, x = aw_805_cast_fp16)[name = tensor("op_10378_cast_fp16")]; + tensor var_10379_cast_fp16 = softmax(axis = var_2624, x = aw_807_cast_fp16)[name = tensor("op_10379_cast_fp16")]; + tensor var_10380_cast_fp16 = softmax(axis = var_2624, x = aw_809_cast_fp16)[name = tensor("op_10380_cast_fp16")]; + tensor var_10381_cast_fp16 = softmax(axis = var_2624, x = aw_811_cast_fp16)[name = tensor("op_10381_cast_fp16")]; + tensor var_10382_cast_fp16 = softmax(axis = var_2624, x = aw_813_cast_fp16)[name = tensor("op_10382_cast_fp16")]; + tensor var_10383_cast_fp16 = softmax(axis = var_2624, x = aw_815_cast_fp16)[name = tensor("op_10383_cast_fp16")]; + tensor var_10384_cast_fp16 = softmax(axis = var_2624, x = aw_817_cast_fp16)[name = tensor("op_10384_cast_fp16")]; + tensor var_10385_cast_fp16 = softmax(axis = var_2624, x = aw_819_cast_fp16)[name = tensor("op_10385_cast_fp16")]; + tensor var_10386_cast_fp16 = softmax(axis = var_2624, x = aw_821_cast_fp16)[name = tensor("op_10386_cast_fp16")]; + tensor var_10387_cast_fp16 = softmax(axis = var_2624, x = aw_823_cast_fp16)[name = tensor("op_10387_cast_fp16")]; + tensor var_10388_cast_fp16 = softmax(axis = var_2624, x = aw_825_cast_fp16)[name = tensor("op_10388_cast_fp16")]; + tensor var_10389_cast_fp16 = softmax(axis = var_2624, x = aw_827_cast_fp16)[name = tensor("op_10389_cast_fp16")]; + tensor var_10390_cast_fp16 = softmax(axis = var_2624, x = aw_829_cast_fp16)[name = tensor("op_10390_cast_fp16")]; + tensor var_10391_cast_fp16 = softmax(axis = var_2624, x = aw_831_cast_fp16)[name = tensor("op_10391_cast_fp16")]; + tensor var_10392_cast_fp16 = softmax(axis = var_2624, x = aw_833_cast_fp16)[name = tensor("op_10392_cast_fp16")]; + tensor var_10393_cast_fp16 = softmax(axis = var_2624, x = aw_835_cast_fp16)[name = tensor("op_10393_cast_fp16")]; + tensor var_10394_cast_fp16 = softmax(axis = var_2624, x = aw_837_cast_fp16)[name = tensor("op_10394_cast_fp16")]; + tensor var_10395_cast_fp16 = softmax(axis = var_2624, x = aw_839_cast_fp16)[name = tensor("op_10395_cast_fp16")]; + tensor var_10397_equation_0 = const()[name = tensor("op_10397_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10397_cast_fp16 = einsum(equation = var_10397_equation_0, values = (var_10217_cast_fp16, var_10376_cast_fp16))[name = tensor("op_10397_cast_fp16")]; + tensor var_10399_equation_0 = const()[name = tensor("op_10399_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10399_cast_fp16 = einsum(equation = var_10399_equation_0, values = (var_10221_cast_fp16, var_10377_cast_fp16))[name = tensor("op_10399_cast_fp16")]; + tensor var_10401_equation_0 = const()[name = tensor("op_10401_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10401_cast_fp16 = einsum(equation = var_10401_equation_0, values = (var_10225_cast_fp16, var_10378_cast_fp16))[name = tensor("op_10401_cast_fp16")]; + tensor var_10403_equation_0 = const()[name = tensor("op_10403_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10403_cast_fp16 = einsum(equation = var_10403_equation_0, values = (var_10229_cast_fp16, var_10379_cast_fp16))[name = tensor("op_10403_cast_fp16")]; + tensor var_10405_equation_0 = const()[name = tensor("op_10405_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10405_cast_fp16 = einsum(equation = var_10405_equation_0, values = (var_10233_cast_fp16, var_10380_cast_fp16))[name = tensor("op_10405_cast_fp16")]; + tensor var_10407_equation_0 = const()[name = tensor("op_10407_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10407_cast_fp16 = einsum(equation = var_10407_equation_0, values = (var_10237_cast_fp16, var_10381_cast_fp16))[name = tensor("op_10407_cast_fp16")]; + tensor var_10409_equation_0 = const()[name = tensor("op_10409_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10409_cast_fp16 = einsum(equation = var_10409_equation_0, values = (var_10241_cast_fp16, var_10382_cast_fp16))[name = tensor("op_10409_cast_fp16")]; + tensor var_10411_equation_0 = const()[name = tensor("op_10411_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10411_cast_fp16 = einsum(equation = var_10411_equation_0, values = (var_10245_cast_fp16, var_10383_cast_fp16))[name = tensor("op_10411_cast_fp16")]; + tensor var_10413_equation_0 = const()[name = tensor("op_10413_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10413_cast_fp16 = einsum(equation = var_10413_equation_0, values = (var_10249_cast_fp16, var_10384_cast_fp16))[name = tensor("op_10413_cast_fp16")]; + tensor var_10415_equation_0 = const()[name = tensor("op_10415_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10415_cast_fp16 = einsum(equation = var_10415_equation_0, values = (var_10253_cast_fp16, var_10385_cast_fp16))[name = tensor("op_10415_cast_fp16")]; + tensor var_10417_equation_0 = const()[name = tensor("op_10417_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10417_cast_fp16 = einsum(equation = var_10417_equation_0, values = (var_10257_cast_fp16, var_10386_cast_fp16))[name = tensor("op_10417_cast_fp16")]; + tensor var_10419_equation_0 = const()[name = tensor("op_10419_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10419_cast_fp16 = einsum(equation = var_10419_equation_0, values = (var_10261_cast_fp16, var_10387_cast_fp16))[name = tensor("op_10419_cast_fp16")]; + tensor var_10421_equation_0 = const()[name = tensor("op_10421_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10421_cast_fp16 = einsum(equation = var_10421_equation_0, values = (var_10265_cast_fp16, var_10388_cast_fp16))[name = tensor("op_10421_cast_fp16")]; + tensor var_10423_equation_0 = const()[name = tensor("op_10423_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10423_cast_fp16 = einsum(equation = var_10423_equation_0, values = (var_10269_cast_fp16, var_10389_cast_fp16))[name = tensor("op_10423_cast_fp16")]; + tensor var_10425_equation_0 = const()[name = tensor("op_10425_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10425_cast_fp16 = einsum(equation = var_10425_equation_0, values = (var_10273_cast_fp16, var_10390_cast_fp16))[name = tensor("op_10425_cast_fp16")]; + tensor var_10427_equation_0 = const()[name = tensor("op_10427_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10427_cast_fp16 = einsum(equation = var_10427_equation_0, values = (var_10277_cast_fp16, var_10391_cast_fp16))[name = tensor("op_10427_cast_fp16")]; + tensor var_10429_equation_0 = const()[name = tensor("op_10429_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10429_cast_fp16 = einsum(equation = var_10429_equation_0, values = (var_10281_cast_fp16, var_10392_cast_fp16))[name = tensor("op_10429_cast_fp16")]; + tensor var_10431_equation_0 = const()[name = tensor("op_10431_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10431_cast_fp16 = einsum(equation = var_10431_equation_0, values = (var_10285_cast_fp16, var_10393_cast_fp16))[name = tensor("op_10431_cast_fp16")]; + tensor var_10433_equation_0 = const()[name = tensor("op_10433_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10433_cast_fp16 = einsum(equation = var_10433_equation_0, values = (var_10289_cast_fp16, var_10394_cast_fp16))[name = tensor("op_10433_cast_fp16")]; + tensor var_10435_equation_0 = const()[name = tensor("op_10435_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10435_cast_fp16 = einsum(equation = var_10435_equation_0, values = (var_10293_cast_fp16, var_10395_cast_fp16))[name = tensor("op_10435_cast_fp16")]; + tensor input_195_interleave_0 = const()[name = tensor("input_195_interleave_0"), val = tensor(false)]; + tensor input_195_cast_fp16 = concat(axis = var_2624, interleave = input_195_interleave_0, values = (var_10397_cast_fp16, var_10399_cast_fp16, var_10401_cast_fp16, var_10403_cast_fp16, var_10405_cast_fp16, var_10407_cast_fp16, var_10409_cast_fp16, var_10411_cast_fp16, var_10413_cast_fp16, var_10415_cast_fp16, var_10417_cast_fp16, var_10419_cast_fp16, var_10421_cast_fp16, var_10423_cast_fp16, var_10425_cast_fp16, var_10427_cast_fp16, var_10429_cast_fp16, var_10431_cast_fp16, var_10433_cast_fp16, var_10435_cast_fp16))[name = tensor("input_195_cast_fp16")]; + tensor var_10441 = const()[name = tensor("op_10441"), val = tensor([1, 1])]; + tensor var_10443 = const()[name = tensor("op_10443"), val = tensor([1, 1])]; + tensor var_10445_pad_type_0 = const()[name = tensor("op_10445_pad_type_0"), val = tensor("custom")]; + tensor var_10445_pad_0 = const()[name = tensor("op_10445_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_8_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_8_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(764105408)))]; + tensor down_blocks_2_attentions_0_transformer_blocks_8_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_8_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(767382272)))]; + tensor var_10445_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_8_attn1_to_out_0_bias_to_fp16, dilations = var_10443, groups = var_2624, pad = var_10445_pad_0, pad_type = var_10445_pad_type_0, strides = var_10441, weight = down_blocks_2_attentions_0_transformer_blocks_8_attn1_to_out_0_weight_to_fp16, x = input_195_cast_fp16)[name = tensor("op_10445_cast_fp16")]; + tensor inputs_75_cast_fp16 = add(x = var_10445_cast_fp16, y = inputs_73_cast_fp16)[name = tensor("inputs_75_cast_fp16")]; + tensor hidden_states_115_axes_0 = const()[name = tensor("hidden_states_115_axes_0"), val = tensor([1])]; + tensor hidden_states_115_gamma_0_to_fp16 = const()[name = tensor("hidden_states_115_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(767384896)))]; + tensor hidden_states_115_beta_0_to_fp16 = const()[name = tensor("hidden_states_115_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(767387520)))]; + tensor var_10455_to_fp16 = const()[name = tensor("op_10455_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_115_cast_fp16 = layer_norm(axes = hidden_states_115_axes_0, beta = hidden_states_115_beta_0_to_fp16, epsilon = var_10455_to_fp16, gamma = hidden_states_115_gamma_0_to_fp16, x = inputs_75_cast_fp16)[name = tensor("hidden_states_115_cast_fp16")]; + tensor var_10470 = const()[name = tensor("op_10470"), val = tensor([1, 1])]; + tensor var_10472 = const()[name = tensor("op_10472"), val = tensor([1, 1])]; + tensor q_51_pad_type_0 = const()[name = tensor("q_51_pad_type_0"), val = tensor("custom")]; + tensor q_51_pad_0 = const()[name = tensor("q_51_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_8_attn2_to_q_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_8_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(767390144)))]; + tensor q_51_cast_fp16 = conv(dilations = var_10472, groups = var_2624, pad = q_51_pad_0, pad_type = q_51_pad_type_0, strides = var_10470, weight = down_blocks_2_attentions_0_transformer_blocks_8_attn2_to_q_weight_to_fp16, x = hidden_states_115_cast_fp16)[name = tensor("q_51_cast_fp16")]; + tensor var_10476 = const()[name = tensor("op_10476"), val = tensor([1, 1])]; + tensor var_10478 = const()[name = tensor("op_10478"), val = tensor([1, 1])]; + tensor k_101_pad_type_0 = const()[name = tensor("k_101_pad_type_0"), val = tensor("custom")]; + tensor k_101_pad_0 = const()[name = tensor("k_101_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_8_attn2_to_k_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_8_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(770667008)))]; + tensor k_101_cast_fp16 = conv(dilations = var_10478, groups = var_2624, pad = k_101_pad_0, pad_type = k_101_pad_type_0, strides = var_10476, weight = down_blocks_2_attentions_0_transformer_blocks_8_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_101_cast_fp16")]; + tensor var_10482 = const()[name = tensor("op_10482"), val = tensor([1, 1])]; + tensor var_10484 = const()[name = tensor("op_10484"), val = tensor([1, 1])]; + tensor v_51_pad_type_0 = const()[name = tensor("v_51_pad_type_0"), val = tensor("custom")]; + tensor v_51_pad_0 = const()[name = tensor("v_51_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_8_attn2_to_v_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_8_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(775909952)))]; + tensor v_51_cast_fp16 = conv(dilations = var_10484, groups = var_2624, pad = v_51_pad_0, pad_type = v_51_pad_type_0, strides = var_10482, weight = down_blocks_2_attentions_0_transformer_blocks_8_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_51_cast_fp16")]; + tensor var_10488_begin_0 = const()[name = tensor("op_10488_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10488_end_0 = const()[name = tensor("op_10488_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_10488_end_mask_0 = const()[name = tensor("op_10488_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10488_cast_fp16 = slice_by_index(begin = var_10488_begin_0, end = var_10488_end_0, end_mask = var_10488_end_mask_0, x = q_51_cast_fp16)[name = tensor("op_10488_cast_fp16")]; + tensor var_10492_begin_0 = const()[name = tensor("op_10492_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_10492_end_0 = const()[name = tensor("op_10492_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_10492_end_mask_0 = const()[name = tensor("op_10492_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10492_cast_fp16 = slice_by_index(begin = var_10492_begin_0, end = var_10492_end_0, end_mask = var_10492_end_mask_0, x = q_51_cast_fp16)[name = tensor("op_10492_cast_fp16")]; + tensor var_10496_begin_0 = const()[name = tensor("op_10496_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_10496_end_0 = const()[name = tensor("op_10496_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_10496_end_mask_0 = const()[name = tensor("op_10496_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10496_cast_fp16 = slice_by_index(begin = var_10496_begin_0, end = var_10496_end_0, end_mask = var_10496_end_mask_0, x = q_51_cast_fp16)[name = tensor("op_10496_cast_fp16")]; + tensor var_10500_begin_0 = const()[name = tensor("op_10500_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_10500_end_0 = const()[name = tensor("op_10500_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_10500_end_mask_0 = const()[name = tensor("op_10500_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10500_cast_fp16 = slice_by_index(begin = var_10500_begin_0, end = var_10500_end_0, end_mask = var_10500_end_mask_0, x = q_51_cast_fp16)[name = tensor("op_10500_cast_fp16")]; + tensor var_10504_begin_0 = const()[name = tensor("op_10504_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_10504_end_0 = const()[name = tensor("op_10504_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_10504_end_mask_0 = const()[name = tensor("op_10504_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10504_cast_fp16 = slice_by_index(begin = var_10504_begin_0, end = var_10504_end_0, end_mask = var_10504_end_mask_0, x = q_51_cast_fp16)[name = tensor("op_10504_cast_fp16")]; + tensor var_10508_begin_0 = const()[name = tensor("op_10508_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_10508_end_0 = const()[name = tensor("op_10508_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_10508_end_mask_0 = const()[name = tensor("op_10508_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10508_cast_fp16 = slice_by_index(begin = var_10508_begin_0, end = var_10508_end_0, end_mask = var_10508_end_mask_0, x = q_51_cast_fp16)[name = tensor("op_10508_cast_fp16")]; + tensor var_10512_begin_0 = const()[name = tensor("op_10512_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_10512_end_0 = const()[name = tensor("op_10512_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_10512_end_mask_0 = const()[name = tensor("op_10512_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10512_cast_fp16 = slice_by_index(begin = var_10512_begin_0, end = var_10512_end_0, end_mask = var_10512_end_mask_0, x = q_51_cast_fp16)[name = tensor("op_10512_cast_fp16")]; + tensor var_10516_begin_0 = const()[name = tensor("op_10516_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_10516_end_0 = const()[name = tensor("op_10516_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_10516_end_mask_0 = const()[name = tensor("op_10516_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10516_cast_fp16 = slice_by_index(begin = var_10516_begin_0, end = var_10516_end_0, end_mask = var_10516_end_mask_0, x = q_51_cast_fp16)[name = tensor("op_10516_cast_fp16")]; + tensor var_10520_begin_0 = const()[name = tensor("op_10520_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_10520_end_0 = const()[name = tensor("op_10520_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_10520_end_mask_0 = const()[name = tensor("op_10520_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10520_cast_fp16 = slice_by_index(begin = var_10520_begin_0, end = var_10520_end_0, end_mask = var_10520_end_mask_0, x = q_51_cast_fp16)[name = tensor("op_10520_cast_fp16")]; + tensor var_10524_begin_0 = const()[name = tensor("op_10524_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_10524_end_0 = const()[name = tensor("op_10524_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_10524_end_mask_0 = const()[name = tensor("op_10524_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10524_cast_fp16 = slice_by_index(begin = var_10524_begin_0, end = var_10524_end_0, end_mask = var_10524_end_mask_0, x = q_51_cast_fp16)[name = tensor("op_10524_cast_fp16")]; + tensor var_10528_begin_0 = const()[name = tensor("op_10528_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_10528_end_0 = const()[name = tensor("op_10528_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_10528_end_mask_0 = const()[name = tensor("op_10528_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10528_cast_fp16 = slice_by_index(begin = var_10528_begin_0, end = var_10528_end_0, end_mask = var_10528_end_mask_0, x = q_51_cast_fp16)[name = tensor("op_10528_cast_fp16")]; + tensor var_10532_begin_0 = const()[name = tensor("op_10532_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_10532_end_0 = const()[name = tensor("op_10532_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_10532_end_mask_0 = const()[name = tensor("op_10532_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10532_cast_fp16 = slice_by_index(begin = var_10532_begin_0, end = var_10532_end_0, end_mask = var_10532_end_mask_0, x = q_51_cast_fp16)[name = tensor("op_10532_cast_fp16")]; + tensor var_10536_begin_0 = const()[name = tensor("op_10536_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_10536_end_0 = const()[name = tensor("op_10536_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_10536_end_mask_0 = const()[name = tensor("op_10536_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10536_cast_fp16 = slice_by_index(begin = var_10536_begin_0, end = var_10536_end_0, end_mask = var_10536_end_mask_0, x = q_51_cast_fp16)[name = tensor("op_10536_cast_fp16")]; + tensor var_10540_begin_0 = const()[name = tensor("op_10540_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_10540_end_0 = const()[name = tensor("op_10540_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_10540_end_mask_0 = const()[name = tensor("op_10540_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10540_cast_fp16 = slice_by_index(begin = var_10540_begin_0, end = var_10540_end_0, end_mask = var_10540_end_mask_0, x = q_51_cast_fp16)[name = tensor("op_10540_cast_fp16")]; + tensor var_10544_begin_0 = const()[name = tensor("op_10544_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_10544_end_0 = const()[name = tensor("op_10544_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_10544_end_mask_0 = const()[name = tensor("op_10544_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10544_cast_fp16 = slice_by_index(begin = var_10544_begin_0, end = var_10544_end_0, end_mask = var_10544_end_mask_0, x = q_51_cast_fp16)[name = tensor("op_10544_cast_fp16")]; + tensor var_10548_begin_0 = const()[name = tensor("op_10548_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_10548_end_0 = const()[name = tensor("op_10548_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_10548_end_mask_0 = const()[name = tensor("op_10548_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10548_cast_fp16 = slice_by_index(begin = var_10548_begin_0, end = var_10548_end_0, end_mask = var_10548_end_mask_0, x = q_51_cast_fp16)[name = tensor("op_10548_cast_fp16")]; + tensor var_10552_begin_0 = const()[name = tensor("op_10552_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_10552_end_0 = const()[name = tensor("op_10552_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_10552_end_mask_0 = const()[name = tensor("op_10552_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10552_cast_fp16 = slice_by_index(begin = var_10552_begin_0, end = var_10552_end_0, end_mask = var_10552_end_mask_0, x = q_51_cast_fp16)[name = tensor("op_10552_cast_fp16")]; + tensor var_10556_begin_0 = const()[name = tensor("op_10556_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_10556_end_0 = const()[name = tensor("op_10556_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_10556_end_mask_0 = const()[name = tensor("op_10556_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10556_cast_fp16 = slice_by_index(begin = var_10556_begin_0, end = var_10556_end_0, end_mask = var_10556_end_mask_0, x = q_51_cast_fp16)[name = tensor("op_10556_cast_fp16")]; + tensor var_10560_begin_0 = const()[name = tensor("op_10560_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_10560_end_0 = const()[name = tensor("op_10560_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_10560_end_mask_0 = const()[name = tensor("op_10560_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10560_cast_fp16 = slice_by_index(begin = var_10560_begin_0, end = var_10560_end_0, end_mask = var_10560_end_mask_0, x = q_51_cast_fp16)[name = tensor("op_10560_cast_fp16")]; + tensor var_10564_begin_0 = const()[name = tensor("op_10564_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_10564_end_0 = const()[name = tensor("op_10564_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_10564_end_mask_0 = const()[name = tensor("op_10564_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10564_cast_fp16 = slice_by_index(begin = var_10564_begin_0, end = var_10564_end_0, end_mask = var_10564_end_mask_0, x = q_51_cast_fp16)[name = tensor("op_10564_cast_fp16")]; + tensor k_103_perm_0 = const()[name = tensor("k_103_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_10571_begin_0 = const()[name = tensor("op_10571_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10571_end_0 = const()[name = tensor("op_10571_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_10571_end_mask_0 = const()[name = tensor("op_10571_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_103_cast_fp16 = transpose(perm = k_103_perm_0, x = k_101_cast_fp16)[name = tensor("transpose_114")]; + tensor var_10571_cast_fp16 = slice_by_index(begin = var_10571_begin_0, end = var_10571_end_0, end_mask = var_10571_end_mask_0, x = k_103_cast_fp16)[name = tensor("op_10571_cast_fp16")]; + tensor var_10575_begin_0 = const()[name = tensor("op_10575_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_10575_end_0 = const()[name = tensor("op_10575_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_10575_end_mask_0 = const()[name = tensor("op_10575_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10575_cast_fp16 = slice_by_index(begin = var_10575_begin_0, end = var_10575_end_0, end_mask = var_10575_end_mask_0, x = k_103_cast_fp16)[name = tensor("op_10575_cast_fp16")]; + tensor var_10579_begin_0 = const()[name = tensor("op_10579_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_10579_end_0 = const()[name = tensor("op_10579_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_10579_end_mask_0 = const()[name = tensor("op_10579_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10579_cast_fp16 = slice_by_index(begin = var_10579_begin_0, end = var_10579_end_0, end_mask = var_10579_end_mask_0, x = k_103_cast_fp16)[name = tensor("op_10579_cast_fp16")]; + tensor var_10583_begin_0 = const()[name = tensor("op_10583_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_10583_end_0 = const()[name = tensor("op_10583_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_10583_end_mask_0 = const()[name = tensor("op_10583_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10583_cast_fp16 = slice_by_index(begin = var_10583_begin_0, end = var_10583_end_0, end_mask = var_10583_end_mask_0, x = k_103_cast_fp16)[name = tensor("op_10583_cast_fp16")]; + tensor var_10587_begin_0 = const()[name = tensor("op_10587_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_10587_end_0 = const()[name = tensor("op_10587_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_10587_end_mask_0 = const()[name = tensor("op_10587_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10587_cast_fp16 = slice_by_index(begin = var_10587_begin_0, end = var_10587_end_0, end_mask = var_10587_end_mask_0, x = k_103_cast_fp16)[name = tensor("op_10587_cast_fp16")]; + tensor var_10591_begin_0 = const()[name = tensor("op_10591_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_10591_end_0 = const()[name = tensor("op_10591_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_10591_end_mask_0 = const()[name = tensor("op_10591_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10591_cast_fp16 = slice_by_index(begin = var_10591_begin_0, end = var_10591_end_0, end_mask = var_10591_end_mask_0, x = k_103_cast_fp16)[name = tensor("op_10591_cast_fp16")]; + tensor var_10595_begin_0 = const()[name = tensor("op_10595_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_10595_end_0 = const()[name = tensor("op_10595_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_10595_end_mask_0 = const()[name = tensor("op_10595_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10595_cast_fp16 = slice_by_index(begin = var_10595_begin_0, end = var_10595_end_0, end_mask = var_10595_end_mask_0, x = k_103_cast_fp16)[name = tensor("op_10595_cast_fp16")]; + tensor var_10599_begin_0 = const()[name = tensor("op_10599_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_10599_end_0 = const()[name = tensor("op_10599_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_10599_end_mask_0 = const()[name = tensor("op_10599_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10599_cast_fp16 = slice_by_index(begin = var_10599_begin_0, end = var_10599_end_0, end_mask = var_10599_end_mask_0, x = k_103_cast_fp16)[name = tensor("op_10599_cast_fp16")]; + tensor var_10603_begin_0 = const()[name = tensor("op_10603_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_10603_end_0 = const()[name = tensor("op_10603_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_10603_end_mask_0 = const()[name = tensor("op_10603_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10603_cast_fp16 = slice_by_index(begin = var_10603_begin_0, end = var_10603_end_0, end_mask = var_10603_end_mask_0, x = k_103_cast_fp16)[name = tensor("op_10603_cast_fp16")]; + tensor var_10607_begin_0 = const()[name = tensor("op_10607_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_10607_end_0 = const()[name = tensor("op_10607_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_10607_end_mask_0 = const()[name = tensor("op_10607_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10607_cast_fp16 = slice_by_index(begin = var_10607_begin_0, end = var_10607_end_0, end_mask = var_10607_end_mask_0, x = k_103_cast_fp16)[name = tensor("op_10607_cast_fp16")]; + tensor var_10611_begin_0 = const()[name = tensor("op_10611_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_10611_end_0 = const()[name = tensor("op_10611_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_10611_end_mask_0 = const()[name = tensor("op_10611_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10611_cast_fp16 = slice_by_index(begin = var_10611_begin_0, end = var_10611_end_0, end_mask = var_10611_end_mask_0, x = k_103_cast_fp16)[name = tensor("op_10611_cast_fp16")]; + tensor var_10615_begin_0 = const()[name = tensor("op_10615_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_10615_end_0 = const()[name = tensor("op_10615_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_10615_end_mask_0 = const()[name = tensor("op_10615_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10615_cast_fp16 = slice_by_index(begin = var_10615_begin_0, end = var_10615_end_0, end_mask = var_10615_end_mask_0, x = k_103_cast_fp16)[name = tensor("op_10615_cast_fp16")]; + tensor var_10619_begin_0 = const()[name = tensor("op_10619_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_10619_end_0 = const()[name = tensor("op_10619_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_10619_end_mask_0 = const()[name = tensor("op_10619_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10619_cast_fp16 = slice_by_index(begin = var_10619_begin_0, end = var_10619_end_0, end_mask = var_10619_end_mask_0, x = k_103_cast_fp16)[name = tensor("op_10619_cast_fp16")]; + tensor var_10623_begin_0 = const()[name = tensor("op_10623_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_10623_end_0 = const()[name = tensor("op_10623_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_10623_end_mask_0 = const()[name = tensor("op_10623_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10623_cast_fp16 = slice_by_index(begin = var_10623_begin_0, end = var_10623_end_0, end_mask = var_10623_end_mask_0, x = k_103_cast_fp16)[name = tensor("op_10623_cast_fp16")]; + tensor var_10627_begin_0 = const()[name = tensor("op_10627_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_10627_end_0 = const()[name = tensor("op_10627_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_10627_end_mask_0 = const()[name = tensor("op_10627_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10627_cast_fp16 = slice_by_index(begin = var_10627_begin_0, end = var_10627_end_0, end_mask = var_10627_end_mask_0, x = k_103_cast_fp16)[name = tensor("op_10627_cast_fp16")]; + tensor var_10631_begin_0 = const()[name = tensor("op_10631_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_10631_end_0 = const()[name = tensor("op_10631_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_10631_end_mask_0 = const()[name = tensor("op_10631_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10631_cast_fp16 = slice_by_index(begin = var_10631_begin_0, end = var_10631_end_0, end_mask = var_10631_end_mask_0, x = k_103_cast_fp16)[name = tensor("op_10631_cast_fp16")]; + tensor var_10635_begin_0 = const()[name = tensor("op_10635_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_10635_end_0 = const()[name = tensor("op_10635_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_10635_end_mask_0 = const()[name = tensor("op_10635_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10635_cast_fp16 = slice_by_index(begin = var_10635_begin_0, end = var_10635_end_0, end_mask = var_10635_end_mask_0, x = k_103_cast_fp16)[name = tensor("op_10635_cast_fp16")]; + tensor var_10639_begin_0 = const()[name = tensor("op_10639_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_10639_end_0 = const()[name = tensor("op_10639_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_10639_end_mask_0 = const()[name = tensor("op_10639_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10639_cast_fp16 = slice_by_index(begin = var_10639_begin_0, end = var_10639_end_0, end_mask = var_10639_end_mask_0, x = k_103_cast_fp16)[name = tensor("op_10639_cast_fp16")]; + tensor var_10643_begin_0 = const()[name = tensor("op_10643_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_10643_end_0 = const()[name = tensor("op_10643_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_10643_end_mask_0 = const()[name = tensor("op_10643_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10643_cast_fp16 = slice_by_index(begin = var_10643_begin_0, end = var_10643_end_0, end_mask = var_10643_end_mask_0, x = k_103_cast_fp16)[name = tensor("op_10643_cast_fp16")]; + tensor var_10647_begin_0 = const()[name = tensor("op_10647_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_10647_end_0 = const()[name = tensor("op_10647_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_10647_end_mask_0 = const()[name = tensor("op_10647_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10647_cast_fp16 = slice_by_index(begin = var_10647_begin_0, end = var_10647_end_0, end_mask = var_10647_end_mask_0, x = k_103_cast_fp16)[name = tensor("op_10647_cast_fp16")]; + tensor var_10649_begin_0 = const()[name = tensor("op_10649_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10649_end_0 = const()[name = tensor("op_10649_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_10649_end_mask_0 = const()[name = tensor("op_10649_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10649_cast_fp16 = slice_by_index(begin = var_10649_begin_0, end = var_10649_end_0, end_mask = var_10649_end_mask_0, x = v_51_cast_fp16)[name = tensor("op_10649_cast_fp16")]; + tensor var_10653_begin_0 = const()[name = tensor("op_10653_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_10653_end_0 = const()[name = tensor("op_10653_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_10653_end_mask_0 = const()[name = tensor("op_10653_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10653_cast_fp16 = slice_by_index(begin = var_10653_begin_0, end = var_10653_end_0, end_mask = var_10653_end_mask_0, x = v_51_cast_fp16)[name = tensor("op_10653_cast_fp16")]; + tensor var_10657_begin_0 = const()[name = tensor("op_10657_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_10657_end_0 = const()[name = tensor("op_10657_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_10657_end_mask_0 = const()[name = tensor("op_10657_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10657_cast_fp16 = slice_by_index(begin = var_10657_begin_0, end = var_10657_end_0, end_mask = var_10657_end_mask_0, x = v_51_cast_fp16)[name = tensor("op_10657_cast_fp16")]; + tensor var_10661_begin_0 = const()[name = tensor("op_10661_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_10661_end_0 = const()[name = tensor("op_10661_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_10661_end_mask_0 = const()[name = tensor("op_10661_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10661_cast_fp16 = slice_by_index(begin = var_10661_begin_0, end = var_10661_end_0, end_mask = var_10661_end_mask_0, x = v_51_cast_fp16)[name = tensor("op_10661_cast_fp16")]; + tensor var_10665_begin_0 = const()[name = tensor("op_10665_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_10665_end_0 = const()[name = tensor("op_10665_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_10665_end_mask_0 = const()[name = tensor("op_10665_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10665_cast_fp16 = slice_by_index(begin = var_10665_begin_0, end = var_10665_end_0, end_mask = var_10665_end_mask_0, x = v_51_cast_fp16)[name = tensor("op_10665_cast_fp16")]; + tensor var_10669_begin_0 = const()[name = tensor("op_10669_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_10669_end_0 = const()[name = tensor("op_10669_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_10669_end_mask_0 = const()[name = tensor("op_10669_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10669_cast_fp16 = slice_by_index(begin = var_10669_begin_0, end = var_10669_end_0, end_mask = var_10669_end_mask_0, x = v_51_cast_fp16)[name = tensor("op_10669_cast_fp16")]; + tensor var_10673_begin_0 = const()[name = tensor("op_10673_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_10673_end_0 = const()[name = tensor("op_10673_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_10673_end_mask_0 = const()[name = tensor("op_10673_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10673_cast_fp16 = slice_by_index(begin = var_10673_begin_0, end = var_10673_end_0, end_mask = var_10673_end_mask_0, x = v_51_cast_fp16)[name = tensor("op_10673_cast_fp16")]; + tensor var_10677_begin_0 = const()[name = tensor("op_10677_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_10677_end_0 = const()[name = tensor("op_10677_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_10677_end_mask_0 = const()[name = tensor("op_10677_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10677_cast_fp16 = slice_by_index(begin = var_10677_begin_0, end = var_10677_end_0, end_mask = var_10677_end_mask_0, x = v_51_cast_fp16)[name = tensor("op_10677_cast_fp16")]; + tensor var_10681_begin_0 = const()[name = tensor("op_10681_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_10681_end_0 = const()[name = tensor("op_10681_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_10681_end_mask_0 = const()[name = tensor("op_10681_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10681_cast_fp16 = slice_by_index(begin = var_10681_begin_0, end = var_10681_end_0, end_mask = var_10681_end_mask_0, x = v_51_cast_fp16)[name = tensor("op_10681_cast_fp16")]; + tensor var_10685_begin_0 = const()[name = tensor("op_10685_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_10685_end_0 = const()[name = tensor("op_10685_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_10685_end_mask_0 = const()[name = tensor("op_10685_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10685_cast_fp16 = slice_by_index(begin = var_10685_begin_0, end = var_10685_end_0, end_mask = var_10685_end_mask_0, x = v_51_cast_fp16)[name = tensor("op_10685_cast_fp16")]; + tensor var_10689_begin_0 = const()[name = tensor("op_10689_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_10689_end_0 = const()[name = tensor("op_10689_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_10689_end_mask_0 = const()[name = tensor("op_10689_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10689_cast_fp16 = slice_by_index(begin = var_10689_begin_0, end = var_10689_end_0, end_mask = var_10689_end_mask_0, x = v_51_cast_fp16)[name = tensor("op_10689_cast_fp16")]; + tensor var_10693_begin_0 = const()[name = tensor("op_10693_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_10693_end_0 = const()[name = tensor("op_10693_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_10693_end_mask_0 = const()[name = tensor("op_10693_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10693_cast_fp16 = slice_by_index(begin = var_10693_begin_0, end = var_10693_end_0, end_mask = var_10693_end_mask_0, x = v_51_cast_fp16)[name = tensor("op_10693_cast_fp16")]; + tensor var_10697_begin_0 = const()[name = tensor("op_10697_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_10697_end_0 = const()[name = tensor("op_10697_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_10697_end_mask_0 = const()[name = tensor("op_10697_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10697_cast_fp16 = slice_by_index(begin = var_10697_begin_0, end = var_10697_end_0, end_mask = var_10697_end_mask_0, x = v_51_cast_fp16)[name = tensor("op_10697_cast_fp16")]; + tensor var_10701_begin_0 = const()[name = tensor("op_10701_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_10701_end_0 = const()[name = tensor("op_10701_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_10701_end_mask_0 = const()[name = tensor("op_10701_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10701_cast_fp16 = slice_by_index(begin = var_10701_begin_0, end = var_10701_end_0, end_mask = var_10701_end_mask_0, x = v_51_cast_fp16)[name = tensor("op_10701_cast_fp16")]; + tensor var_10705_begin_0 = const()[name = tensor("op_10705_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_10705_end_0 = const()[name = tensor("op_10705_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_10705_end_mask_0 = const()[name = tensor("op_10705_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10705_cast_fp16 = slice_by_index(begin = var_10705_begin_0, end = var_10705_end_0, end_mask = var_10705_end_mask_0, x = v_51_cast_fp16)[name = tensor("op_10705_cast_fp16")]; + tensor var_10709_begin_0 = const()[name = tensor("op_10709_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_10709_end_0 = const()[name = tensor("op_10709_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_10709_end_mask_0 = const()[name = tensor("op_10709_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10709_cast_fp16 = slice_by_index(begin = var_10709_begin_0, end = var_10709_end_0, end_mask = var_10709_end_mask_0, x = v_51_cast_fp16)[name = tensor("op_10709_cast_fp16")]; + tensor var_10713_begin_0 = const()[name = tensor("op_10713_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_10713_end_0 = const()[name = tensor("op_10713_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_10713_end_mask_0 = const()[name = tensor("op_10713_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10713_cast_fp16 = slice_by_index(begin = var_10713_begin_0, end = var_10713_end_0, end_mask = var_10713_end_mask_0, x = v_51_cast_fp16)[name = tensor("op_10713_cast_fp16")]; + tensor var_10717_begin_0 = const()[name = tensor("op_10717_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_10717_end_0 = const()[name = tensor("op_10717_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_10717_end_mask_0 = const()[name = tensor("op_10717_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10717_cast_fp16 = slice_by_index(begin = var_10717_begin_0, end = var_10717_end_0, end_mask = var_10717_end_mask_0, x = v_51_cast_fp16)[name = tensor("op_10717_cast_fp16")]; + tensor var_10721_begin_0 = const()[name = tensor("op_10721_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_10721_end_0 = const()[name = tensor("op_10721_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_10721_end_mask_0 = const()[name = tensor("op_10721_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10721_cast_fp16 = slice_by_index(begin = var_10721_begin_0, end = var_10721_end_0, end_mask = var_10721_end_mask_0, x = v_51_cast_fp16)[name = tensor("op_10721_cast_fp16")]; + tensor var_10725_begin_0 = const()[name = tensor("op_10725_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_10725_end_0 = const()[name = tensor("op_10725_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_10725_end_mask_0 = const()[name = tensor("op_10725_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10725_cast_fp16 = slice_by_index(begin = var_10725_begin_0, end = var_10725_end_0, end_mask = var_10725_end_mask_0, x = v_51_cast_fp16)[name = tensor("op_10725_cast_fp16")]; + tensor var_10729_equation_0 = const()[name = tensor("op_10729_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10729_cast_fp16 = einsum(equation = var_10729_equation_0, values = (var_10571_cast_fp16, var_10488_cast_fp16))[name = tensor("op_10729_cast_fp16")]; + tensor var_10730_to_fp16 = const()[name = tensor("op_10730_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_841_cast_fp16 = mul(x = var_10729_cast_fp16, y = var_10730_to_fp16)[name = tensor("aw_841_cast_fp16")]; + tensor var_10733_equation_0 = const()[name = tensor("op_10733_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10733_cast_fp16 = einsum(equation = var_10733_equation_0, values = (var_10575_cast_fp16, var_10492_cast_fp16))[name = tensor("op_10733_cast_fp16")]; + tensor var_10734_to_fp16 = const()[name = tensor("op_10734_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_843_cast_fp16 = mul(x = var_10733_cast_fp16, y = var_10734_to_fp16)[name = tensor("aw_843_cast_fp16")]; + tensor var_10737_equation_0 = const()[name = tensor("op_10737_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10737_cast_fp16 = einsum(equation = var_10737_equation_0, values = (var_10579_cast_fp16, var_10496_cast_fp16))[name = tensor("op_10737_cast_fp16")]; + tensor var_10738_to_fp16 = const()[name = tensor("op_10738_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_845_cast_fp16 = mul(x = var_10737_cast_fp16, y = var_10738_to_fp16)[name = tensor("aw_845_cast_fp16")]; + tensor var_10741_equation_0 = const()[name = tensor("op_10741_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10741_cast_fp16 = einsum(equation = var_10741_equation_0, values = (var_10583_cast_fp16, var_10500_cast_fp16))[name = tensor("op_10741_cast_fp16")]; + tensor var_10742_to_fp16 = const()[name = tensor("op_10742_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_847_cast_fp16 = mul(x = var_10741_cast_fp16, y = var_10742_to_fp16)[name = tensor("aw_847_cast_fp16")]; + tensor var_10745_equation_0 = const()[name = tensor("op_10745_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10745_cast_fp16 = einsum(equation = var_10745_equation_0, values = (var_10587_cast_fp16, var_10504_cast_fp16))[name = tensor("op_10745_cast_fp16")]; + tensor var_10746_to_fp16 = const()[name = tensor("op_10746_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_849_cast_fp16 = mul(x = var_10745_cast_fp16, y = var_10746_to_fp16)[name = tensor("aw_849_cast_fp16")]; + tensor var_10749_equation_0 = const()[name = tensor("op_10749_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10749_cast_fp16 = einsum(equation = var_10749_equation_0, values = (var_10591_cast_fp16, var_10508_cast_fp16))[name = tensor("op_10749_cast_fp16")]; + tensor var_10750_to_fp16 = const()[name = tensor("op_10750_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_851_cast_fp16 = mul(x = var_10749_cast_fp16, y = var_10750_to_fp16)[name = tensor("aw_851_cast_fp16")]; + tensor var_10753_equation_0 = const()[name = tensor("op_10753_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10753_cast_fp16 = einsum(equation = var_10753_equation_0, values = (var_10595_cast_fp16, var_10512_cast_fp16))[name = tensor("op_10753_cast_fp16")]; + tensor var_10754_to_fp16 = const()[name = tensor("op_10754_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_853_cast_fp16 = mul(x = var_10753_cast_fp16, y = var_10754_to_fp16)[name = tensor("aw_853_cast_fp16")]; + tensor var_10757_equation_0 = const()[name = tensor("op_10757_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10757_cast_fp16 = einsum(equation = var_10757_equation_0, values = (var_10599_cast_fp16, var_10516_cast_fp16))[name = tensor("op_10757_cast_fp16")]; + tensor var_10758_to_fp16 = const()[name = tensor("op_10758_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_855_cast_fp16 = mul(x = var_10757_cast_fp16, y = var_10758_to_fp16)[name = tensor("aw_855_cast_fp16")]; + tensor var_10761_equation_0 = const()[name = tensor("op_10761_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10761_cast_fp16 = einsum(equation = var_10761_equation_0, values = (var_10603_cast_fp16, var_10520_cast_fp16))[name = tensor("op_10761_cast_fp16")]; + tensor var_10762_to_fp16 = const()[name = tensor("op_10762_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_857_cast_fp16 = mul(x = var_10761_cast_fp16, y = var_10762_to_fp16)[name = tensor("aw_857_cast_fp16")]; + tensor var_10765_equation_0 = const()[name = tensor("op_10765_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10765_cast_fp16 = einsum(equation = var_10765_equation_0, values = (var_10607_cast_fp16, var_10524_cast_fp16))[name = tensor("op_10765_cast_fp16")]; + tensor var_10766_to_fp16 = const()[name = tensor("op_10766_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_859_cast_fp16 = mul(x = var_10765_cast_fp16, y = var_10766_to_fp16)[name = tensor("aw_859_cast_fp16")]; + tensor var_10769_equation_0 = const()[name = tensor("op_10769_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10769_cast_fp16 = einsum(equation = var_10769_equation_0, values = (var_10611_cast_fp16, var_10528_cast_fp16))[name = tensor("op_10769_cast_fp16")]; + tensor var_10770_to_fp16 = const()[name = tensor("op_10770_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_861_cast_fp16 = mul(x = var_10769_cast_fp16, y = var_10770_to_fp16)[name = tensor("aw_861_cast_fp16")]; + tensor var_10773_equation_0 = const()[name = tensor("op_10773_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10773_cast_fp16 = einsum(equation = var_10773_equation_0, values = (var_10615_cast_fp16, var_10532_cast_fp16))[name = tensor("op_10773_cast_fp16")]; + tensor var_10774_to_fp16 = const()[name = tensor("op_10774_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_863_cast_fp16 = mul(x = var_10773_cast_fp16, y = var_10774_to_fp16)[name = tensor("aw_863_cast_fp16")]; + tensor var_10777_equation_0 = const()[name = tensor("op_10777_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10777_cast_fp16 = einsum(equation = var_10777_equation_0, values = (var_10619_cast_fp16, var_10536_cast_fp16))[name = tensor("op_10777_cast_fp16")]; + tensor var_10778_to_fp16 = const()[name = tensor("op_10778_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_865_cast_fp16 = mul(x = var_10777_cast_fp16, y = var_10778_to_fp16)[name = tensor("aw_865_cast_fp16")]; + tensor var_10781_equation_0 = const()[name = tensor("op_10781_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10781_cast_fp16 = einsum(equation = var_10781_equation_0, values = (var_10623_cast_fp16, var_10540_cast_fp16))[name = tensor("op_10781_cast_fp16")]; + tensor var_10782_to_fp16 = const()[name = tensor("op_10782_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_867_cast_fp16 = mul(x = var_10781_cast_fp16, y = var_10782_to_fp16)[name = tensor("aw_867_cast_fp16")]; + tensor var_10785_equation_0 = const()[name = tensor("op_10785_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10785_cast_fp16 = einsum(equation = var_10785_equation_0, values = (var_10627_cast_fp16, var_10544_cast_fp16))[name = tensor("op_10785_cast_fp16")]; + tensor var_10786_to_fp16 = const()[name = tensor("op_10786_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_869_cast_fp16 = mul(x = var_10785_cast_fp16, y = var_10786_to_fp16)[name = tensor("aw_869_cast_fp16")]; + tensor var_10789_equation_0 = const()[name = tensor("op_10789_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10789_cast_fp16 = einsum(equation = var_10789_equation_0, values = (var_10631_cast_fp16, var_10548_cast_fp16))[name = tensor("op_10789_cast_fp16")]; + tensor var_10790_to_fp16 = const()[name = tensor("op_10790_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_871_cast_fp16 = mul(x = var_10789_cast_fp16, y = var_10790_to_fp16)[name = tensor("aw_871_cast_fp16")]; + tensor var_10793_equation_0 = const()[name = tensor("op_10793_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10793_cast_fp16 = einsum(equation = var_10793_equation_0, values = (var_10635_cast_fp16, var_10552_cast_fp16))[name = tensor("op_10793_cast_fp16")]; + tensor var_10794_to_fp16 = const()[name = tensor("op_10794_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_873_cast_fp16 = mul(x = var_10793_cast_fp16, y = var_10794_to_fp16)[name = tensor("aw_873_cast_fp16")]; + tensor var_10797_equation_0 = const()[name = tensor("op_10797_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10797_cast_fp16 = einsum(equation = var_10797_equation_0, values = (var_10639_cast_fp16, var_10556_cast_fp16))[name = tensor("op_10797_cast_fp16")]; + tensor var_10798_to_fp16 = const()[name = tensor("op_10798_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_875_cast_fp16 = mul(x = var_10797_cast_fp16, y = var_10798_to_fp16)[name = tensor("aw_875_cast_fp16")]; + tensor var_10801_equation_0 = const()[name = tensor("op_10801_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10801_cast_fp16 = einsum(equation = var_10801_equation_0, values = (var_10643_cast_fp16, var_10560_cast_fp16))[name = tensor("op_10801_cast_fp16")]; + tensor var_10802_to_fp16 = const()[name = tensor("op_10802_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_877_cast_fp16 = mul(x = var_10801_cast_fp16, y = var_10802_to_fp16)[name = tensor("aw_877_cast_fp16")]; + tensor var_10805_equation_0 = const()[name = tensor("op_10805_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10805_cast_fp16 = einsum(equation = var_10805_equation_0, values = (var_10647_cast_fp16, var_10564_cast_fp16))[name = tensor("op_10805_cast_fp16")]; + tensor var_10806_to_fp16 = const()[name = tensor("op_10806_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_879_cast_fp16 = mul(x = var_10805_cast_fp16, y = var_10806_to_fp16)[name = tensor("aw_879_cast_fp16")]; + tensor var_10808_cast_fp16 = softmax(axis = var_2624, x = aw_841_cast_fp16)[name = tensor("op_10808_cast_fp16")]; + tensor var_10809_cast_fp16 = softmax(axis = var_2624, x = aw_843_cast_fp16)[name = tensor("op_10809_cast_fp16")]; + tensor var_10810_cast_fp16 = softmax(axis = var_2624, x = aw_845_cast_fp16)[name = tensor("op_10810_cast_fp16")]; + tensor var_10811_cast_fp16 = softmax(axis = var_2624, x = aw_847_cast_fp16)[name = tensor("op_10811_cast_fp16")]; + tensor var_10812_cast_fp16 = softmax(axis = var_2624, x = aw_849_cast_fp16)[name = tensor("op_10812_cast_fp16")]; + tensor var_10813_cast_fp16 = softmax(axis = var_2624, x = aw_851_cast_fp16)[name = tensor("op_10813_cast_fp16")]; + tensor var_10814_cast_fp16 = softmax(axis = var_2624, x = aw_853_cast_fp16)[name = tensor("op_10814_cast_fp16")]; + tensor var_10815_cast_fp16 = softmax(axis = var_2624, x = aw_855_cast_fp16)[name = tensor("op_10815_cast_fp16")]; + tensor var_10816_cast_fp16 = softmax(axis = var_2624, x = aw_857_cast_fp16)[name = tensor("op_10816_cast_fp16")]; + tensor var_10817_cast_fp16 = softmax(axis = var_2624, x = aw_859_cast_fp16)[name = tensor("op_10817_cast_fp16")]; + tensor var_10818_cast_fp16 = softmax(axis = var_2624, x = aw_861_cast_fp16)[name = tensor("op_10818_cast_fp16")]; + tensor var_10819_cast_fp16 = softmax(axis = var_2624, x = aw_863_cast_fp16)[name = tensor("op_10819_cast_fp16")]; + tensor var_10820_cast_fp16 = softmax(axis = var_2624, x = aw_865_cast_fp16)[name = tensor("op_10820_cast_fp16")]; + tensor var_10821_cast_fp16 = softmax(axis = var_2624, x = aw_867_cast_fp16)[name = tensor("op_10821_cast_fp16")]; + tensor var_10822_cast_fp16 = softmax(axis = var_2624, x = aw_869_cast_fp16)[name = tensor("op_10822_cast_fp16")]; + tensor var_10823_cast_fp16 = softmax(axis = var_2624, x = aw_871_cast_fp16)[name = tensor("op_10823_cast_fp16")]; + tensor var_10824_cast_fp16 = softmax(axis = var_2624, x = aw_873_cast_fp16)[name = tensor("op_10824_cast_fp16")]; + tensor var_10825_cast_fp16 = softmax(axis = var_2624, x = aw_875_cast_fp16)[name = tensor("op_10825_cast_fp16")]; + tensor var_10826_cast_fp16 = softmax(axis = var_2624, x = aw_877_cast_fp16)[name = tensor("op_10826_cast_fp16")]; + tensor var_10827_cast_fp16 = softmax(axis = var_2624, x = aw_879_cast_fp16)[name = tensor("op_10827_cast_fp16")]; + tensor var_10829_equation_0 = const()[name = tensor("op_10829_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10829_cast_fp16 = einsum(equation = var_10829_equation_0, values = (var_10649_cast_fp16, var_10808_cast_fp16))[name = tensor("op_10829_cast_fp16")]; + tensor var_10831_equation_0 = const()[name = tensor("op_10831_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10831_cast_fp16 = einsum(equation = var_10831_equation_0, values = (var_10653_cast_fp16, var_10809_cast_fp16))[name = tensor("op_10831_cast_fp16")]; + tensor var_10833_equation_0 = const()[name = tensor("op_10833_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10833_cast_fp16 = einsum(equation = var_10833_equation_0, values = (var_10657_cast_fp16, var_10810_cast_fp16))[name = tensor("op_10833_cast_fp16")]; + tensor var_10835_equation_0 = const()[name = tensor("op_10835_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10835_cast_fp16 = einsum(equation = var_10835_equation_0, values = (var_10661_cast_fp16, var_10811_cast_fp16))[name = tensor("op_10835_cast_fp16")]; + tensor var_10837_equation_0 = const()[name = tensor("op_10837_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10837_cast_fp16 = einsum(equation = var_10837_equation_0, values = (var_10665_cast_fp16, var_10812_cast_fp16))[name = tensor("op_10837_cast_fp16")]; + tensor var_10839_equation_0 = const()[name = tensor("op_10839_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10839_cast_fp16 = einsum(equation = var_10839_equation_0, values = (var_10669_cast_fp16, var_10813_cast_fp16))[name = tensor("op_10839_cast_fp16")]; + tensor var_10841_equation_0 = const()[name = tensor("op_10841_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10841_cast_fp16 = einsum(equation = var_10841_equation_0, values = (var_10673_cast_fp16, var_10814_cast_fp16))[name = tensor("op_10841_cast_fp16")]; + tensor var_10843_equation_0 = const()[name = tensor("op_10843_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10843_cast_fp16 = einsum(equation = var_10843_equation_0, values = (var_10677_cast_fp16, var_10815_cast_fp16))[name = tensor("op_10843_cast_fp16")]; + tensor var_10845_equation_0 = const()[name = tensor("op_10845_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10845_cast_fp16 = einsum(equation = var_10845_equation_0, values = (var_10681_cast_fp16, var_10816_cast_fp16))[name = tensor("op_10845_cast_fp16")]; + tensor var_10847_equation_0 = const()[name = tensor("op_10847_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10847_cast_fp16 = einsum(equation = var_10847_equation_0, values = (var_10685_cast_fp16, var_10817_cast_fp16))[name = tensor("op_10847_cast_fp16")]; + tensor var_10849_equation_0 = const()[name = tensor("op_10849_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10849_cast_fp16 = einsum(equation = var_10849_equation_0, values = (var_10689_cast_fp16, var_10818_cast_fp16))[name = tensor("op_10849_cast_fp16")]; + tensor var_10851_equation_0 = const()[name = tensor("op_10851_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10851_cast_fp16 = einsum(equation = var_10851_equation_0, values = (var_10693_cast_fp16, var_10819_cast_fp16))[name = tensor("op_10851_cast_fp16")]; + tensor var_10853_equation_0 = const()[name = tensor("op_10853_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10853_cast_fp16 = einsum(equation = var_10853_equation_0, values = (var_10697_cast_fp16, var_10820_cast_fp16))[name = tensor("op_10853_cast_fp16")]; + tensor var_10855_equation_0 = const()[name = tensor("op_10855_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10855_cast_fp16 = einsum(equation = var_10855_equation_0, values = (var_10701_cast_fp16, var_10821_cast_fp16))[name = tensor("op_10855_cast_fp16")]; + tensor var_10857_equation_0 = const()[name = tensor("op_10857_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10857_cast_fp16 = einsum(equation = var_10857_equation_0, values = (var_10705_cast_fp16, var_10822_cast_fp16))[name = tensor("op_10857_cast_fp16")]; + tensor var_10859_equation_0 = const()[name = tensor("op_10859_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10859_cast_fp16 = einsum(equation = var_10859_equation_0, values = (var_10709_cast_fp16, var_10823_cast_fp16))[name = tensor("op_10859_cast_fp16")]; + tensor var_10861_equation_0 = const()[name = tensor("op_10861_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10861_cast_fp16 = einsum(equation = var_10861_equation_0, values = (var_10713_cast_fp16, var_10824_cast_fp16))[name = tensor("op_10861_cast_fp16")]; + tensor var_10863_equation_0 = const()[name = tensor("op_10863_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10863_cast_fp16 = einsum(equation = var_10863_equation_0, values = (var_10717_cast_fp16, var_10825_cast_fp16))[name = tensor("op_10863_cast_fp16")]; + tensor var_10865_equation_0 = const()[name = tensor("op_10865_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10865_cast_fp16 = einsum(equation = var_10865_equation_0, values = (var_10721_cast_fp16, var_10826_cast_fp16))[name = tensor("op_10865_cast_fp16")]; + tensor var_10867_equation_0 = const()[name = tensor("op_10867_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10867_cast_fp16 = einsum(equation = var_10867_equation_0, values = (var_10725_cast_fp16, var_10827_cast_fp16))[name = tensor("op_10867_cast_fp16")]; + tensor input_197_interleave_0 = const()[name = tensor("input_197_interleave_0"), val = tensor(false)]; + tensor input_197_cast_fp16 = concat(axis = var_2624, interleave = input_197_interleave_0, values = (var_10829_cast_fp16, var_10831_cast_fp16, var_10833_cast_fp16, var_10835_cast_fp16, var_10837_cast_fp16, var_10839_cast_fp16, var_10841_cast_fp16, var_10843_cast_fp16, var_10845_cast_fp16, var_10847_cast_fp16, var_10849_cast_fp16, var_10851_cast_fp16, var_10853_cast_fp16, var_10855_cast_fp16, var_10857_cast_fp16, var_10859_cast_fp16, var_10861_cast_fp16, var_10863_cast_fp16, var_10865_cast_fp16, var_10867_cast_fp16))[name = tensor("input_197_cast_fp16")]; + tensor var_10873 = const()[name = tensor("op_10873"), val = tensor([1, 1])]; + tensor var_10875 = const()[name = tensor("op_10875"), val = tensor([1, 1])]; + tensor var_10877_pad_type_0 = const()[name = tensor("op_10877_pad_type_0"), val = tensor("custom")]; + tensor var_10877_pad_0 = const()[name = tensor("op_10877_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_8_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_8_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(781152896)))]; + tensor down_blocks_2_attentions_0_transformer_blocks_8_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_8_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(784429760)))]; + tensor var_10877_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_8_attn2_to_out_0_bias_to_fp16, dilations = var_10875, groups = var_2624, pad = var_10877_pad_0, pad_type = var_10877_pad_type_0, strides = var_10873, weight = down_blocks_2_attentions_0_transformer_blocks_8_attn2_to_out_0_weight_to_fp16, x = input_197_cast_fp16)[name = tensor("op_10877_cast_fp16")]; + tensor inputs_77_cast_fp16 = add(x = var_10877_cast_fp16, y = inputs_75_cast_fp16)[name = tensor("inputs_77_cast_fp16")]; + tensor input_199_axes_0 = const()[name = tensor("input_199_axes_0"), val = tensor([1])]; + tensor input_199_gamma_0_to_fp16 = const()[name = tensor("input_199_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(784432384)))]; + tensor input_199_beta_0_to_fp16 = const()[name = tensor("input_199_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(784435008)))]; + tensor var_10887_to_fp16 = const()[name = tensor("op_10887_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_199_cast_fp16 = layer_norm(axes = input_199_axes_0, beta = input_199_beta_0_to_fp16, epsilon = var_10887_to_fp16, gamma = input_199_gamma_0_to_fp16, x = inputs_77_cast_fp16)[name = tensor("input_199_cast_fp16")]; + tensor var_10903 = const()[name = tensor("op_10903"), val = tensor([1, 1])]; + tensor var_10905 = const()[name = tensor("op_10905"), val = tensor([1, 1])]; + tensor var_10907_pad_type_0 = const()[name = tensor("op_10907_pad_type_0"), val = tensor("custom")]; + tensor var_10907_pad_0 = const()[name = tensor("op_10907_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_8_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_8_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(784437632)))]; + tensor down_blocks_2_attentions_0_transformer_blocks_8_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_8_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(810652096)))]; + tensor var_10907_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_8_ff_net_0_proj_bias_to_fp16, dilations = var_10905, groups = var_2624, pad = var_10907_pad_0, pad_type = var_10907_pad_type_0, strides = var_10903, weight = down_blocks_2_attentions_0_transformer_blocks_8_ff_net_0_proj_weight_to_fp16, x = input_199_cast_fp16)[name = tensor("op_10907_cast_fp16")]; + tensor var_10908_split_sizes_0 = const()[name = tensor("op_10908_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_10908_axis_0 = const()[name = tensor("op_10908_axis_0"), val = tensor(1)]; + tensor var_10908_cast_fp16_0, tensor var_10908_cast_fp16_1 = split(axis = var_10908_axis_0, split_sizes = var_10908_split_sizes_0, x = var_10907_cast_fp16)[name = tensor("op_10908_cast_fp16")]; + tensor var_10910_mode_0 = const()[name = tensor("op_10910_mode_0"), val = tensor("EXACT")]; + tensor var_10910_cast_fp16 = gelu(mode = var_10910_mode_0, x = var_10908_cast_fp16_1)[name = tensor("op_10910_cast_fp16")]; + tensor input_201_cast_fp16 = mul(x = var_10908_cast_fp16_0, y = var_10910_cast_fp16)[name = tensor("input_201_cast_fp16")]; + tensor var_10914 = const()[name = tensor("op_10914"), val = tensor([1, 1])]; + tensor var_10916 = const()[name = tensor("op_10916"), val = tensor([1, 1])]; + tensor var_10918_pad_type_0 = const()[name = tensor("op_10918_pad_type_0"), val = tensor("custom")]; + tensor var_10918_pad_0 = const()[name = tensor("op_10918_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_8_ff_net_2_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_8_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(810672640)))]; + tensor down_blocks_2_attentions_0_transformer_blocks_8_ff_net_2_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_8_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(823779904)))]; + tensor var_10918_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_8_ff_net_2_bias_to_fp16, dilations = var_10916, groups = var_2624, pad = var_10918_pad_0, pad_type = var_10918_pad_type_0, strides = var_10914, weight = down_blocks_2_attentions_0_transformer_blocks_8_ff_net_2_weight_to_fp16, x = input_201_cast_fp16)[name = tensor("op_10918_cast_fp16")]; + tensor inputs_79_cast_fp16 = add(x = var_10918_cast_fp16, y = inputs_77_cast_fp16)[name = tensor("inputs_79_cast_fp16")]; + tensor hidden_states_119_axes_0 = const()[name = tensor("hidden_states_119_axes_0"), val = tensor([1])]; + tensor hidden_states_119_gamma_0_to_fp16 = const()[name = tensor("hidden_states_119_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(823782528)))]; + tensor hidden_states_119_beta_0_to_fp16 = const()[name = tensor("hidden_states_119_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(823785152)))]; + tensor var_10934_to_fp16 = const()[name = tensor("op_10934_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_119_cast_fp16 = layer_norm(axes = hidden_states_119_axes_0, beta = hidden_states_119_beta_0_to_fp16, epsilon = var_10934_to_fp16, gamma = hidden_states_119_gamma_0_to_fp16, x = inputs_79_cast_fp16)[name = tensor("hidden_states_119_cast_fp16")]; + tensor var_10949 = const()[name = tensor("op_10949"), val = tensor([1, 1])]; + tensor var_10951 = const()[name = tensor("op_10951"), val = tensor([1, 1])]; + tensor q_53_pad_type_0 = const()[name = tensor("q_53_pad_type_0"), val = tensor("custom")]; + tensor q_53_pad_0 = const()[name = tensor("q_53_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_9_attn1_to_q_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_9_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(823787776)))]; + tensor q_53_cast_fp16 = conv(dilations = var_10951, groups = var_2624, pad = q_53_pad_0, pad_type = q_53_pad_type_0, strides = var_10949, weight = down_blocks_2_attentions_0_transformer_blocks_9_attn1_to_q_weight_to_fp16, x = hidden_states_119_cast_fp16)[name = tensor("q_53_cast_fp16")]; + tensor var_10955 = const()[name = tensor("op_10955"), val = tensor([1, 1])]; + tensor var_10957 = const()[name = tensor("op_10957"), val = tensor([1, 1])]; + tensor k_105_pad_type_0 = const()[name = tensor("k_105_pad_type_0"), val = tensor("custom")]; + tensor k_105_pad_0 = const()[name = tensor("k_105_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_9_attn1_to_k_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_9_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(827064640)))]; + tensor k_105_cast_fp16 = conv(dilations = var_10957, groups = var_2624, pad = k_105_pad_0, pad_type = k_105_pad_type_0, strides = var_10955, weight = down_blocks_2_attentions_0_transformer_blocks_9_attn1_to_k_weight_to_fp16, x = hidden_states_119_cast_fp16)[name = tensor("k_105_cast_fp16")]; + tensor var_10961 = const()[name = tensor("op_10961"), val = tensor([1, 1])]; + tensor var_10963 = const()[name = tensor("op_10963"), val = tensor([1, 1])]; + tensor v_53_pad_type_0 = const()[name = tensor("v_53_pad_type_0"), val = tensor("custom")]; + tensor v_53_pad_0 = const()[name = tensor("v_53_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_9_attn1_to_v_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_9_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(830341504)))]; + tensor v_53_cast_fp16 = conv(dilations = var_10963, groups = var_2624, pad = v_53_pad_0, pad_type = v_53_pad_type_0, strides = var_10961, weight = down_blocks_2_attentions_0_transformer_blocks_9_attn1_to_v_weight_to_fp16, x = hidden_states_119_cast_fp16)[name = tensor("v_53_cast_fp16")]; + tensor var_10967_begin_0 = const()[name = tensor("op_10967_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10967_end_0 = const()[name = tensor("op_10967_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_10967_end_mask_0 = const()[name = tensor("op_10967_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10967_cast_fp16 = slice_by_index(begin = var_10967_begin_0, end = var_10967_end_0, end_mask = var_10967_end_mask_0, x = q_53_cast_fp16)[name = tensor("op_10967_cast_fp16")]; + tensor var_10971_begin_0 = const()[name = tensor("op_10971_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_10971_end_0 = const()[name = tensor("op_10971_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_10971_end_mask_0 = const()[name = tensor("op_10971_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10971_cast_fp16 = slice_by_index(begin = var_10971_begin_0, end = var_10971_end_0, end_mask = var_10971_end_mask_0, x = q_53_cast_fp16)[name = tensor("op_10971_cast_fp16")]; + tensor var_10975_begin_0 = const()[name = tensor("op_10975_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_10975_end_0 = const()[name = tensor("op_10975_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_10975_end_mask_0 = const()[name = tensor("op_10975_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10975_cast_fp16 = slice_by_index(begin = var_10975_begin_0, end = var_10975_end_0, end_mask = var_10975_end_mask_0, x = q_53_cast_fp16)[name = tensor("op_10975_cast_fp16")]; + tensor var_10979_begin_0 = const()[name = tensor("op_10979_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_10979_end_0 = const()[name = tensor("op_10979_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_10979_end_mask_0 = const()[name = tensor("op_10979_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10979_cast_fp16 = slice_by_index(begin = var_10979_begin_0, end = var_10979_end_0, end_mask = var_10979_end_mask_0, x = q_53_cast_fp16)[name = tensor("op_10979_cast_fp16")]; + tensor var_10983_begin_0 = const()[name = tensor("op_10983_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_10983_end_0 = const()[name = tensor("op_10983_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_10983_end_mask_0 = const()[name = tensor("op_10983_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10983_cast_fp16 = slice_by_index(begin = var_10983_begin_0, end = var_10983_end_0, end_mask = var_10983_end_mask_0, x = q_53_cast_fp16)[name = tensor("op_10983_cast_fp16")]; + tensor var_10987_begin_0 = const()[name = tensor("op_10987_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_10987_end_0 = const()[name = tensor("op_10987_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_10987_end_mask_0 = const()[name = tensor("op_10987_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10987_cast_fp16 = slice_by_index(begin = var_10987_begin_0, end = var_10987_end_0, end_mask = var_10987_end_mask_0, x = q_53_cast_fp16)[name = tensor("op_10987_cast_fp16")]; + tensor var_10991_begin_0 = const()[name = tensor("op_10991_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_10991_end_0 = const()[name = tensor("op_10991_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_10991_end_mask_0 = const()[name = tensor("op_10991_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10991_cast_fp16 = slice_by_index(begin = var_10991_begin_0, end = var_10991_end_0, end_mask = var_10991_end_mask_0, x = q_53_cast_fp16)[name = tensor("op_10991_cast_fp16")]; + tensor var_10995_begin_0 = const()[name = tensor("op_10995_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_10995_end_0 = const()[name = tensor("op_10995_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_10995_end_mask_0 = const()[name = tensor("op_10995_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10995_cast_fp16 = slice_by_index(begin = var_10995_begin_0, end = var_10995_end_0, end_mask = var_10995_end_mask_0, x = q_53_cast_fp16)[name = tensor("op_10995_cast_fp16")]; + tensor var_10999_begin_0 = const()[name = tensor("op_10999_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_10999_end_0 = const()[name = tensor("op_10999_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_10999_end_mask_0 = const()[name = tensor("op_10999_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10999_cast_fp16 = slice_by_index(begin = var_10999_begin_0, end = var_10999_end_0, end_mask = var_10999_end_mask_0, x = q_53_cast_fp16)[name = tensor("op_10999_cast_fp16")]; + tensor var_11003_begin_0 = const()[name = tensor("op_11003_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_11003_end_0 = const()[name = tensor("op_11003_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_11003_end_mask_0 = const()[name = tensor("op_11003_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11003_cast_fp16 = slice_by_index(begin = var_11003_begin_0, end = var_11003_end_0, end_mask = var_11003_end_mask_0, x = q_53_cast_fp16)[name = tensor("op_11003_cast_fp16")]; + tensor var_11007_begin_0 = const()[name = tensor("op_11007_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_11007_end_0 = const()[name = tensor("op_11007_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_11007_end_mask_0 = const()[name = tensor("op_11007_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11007_cast_fp16 = slice_by_index(begin = var_11007_begin_0, end = var_11007_end_0, end_mask = var_11007_end_mask_0, x = q_53_cast_fp16)[name = tensor("op_11007_cast_fp16")]; + tensor var_11011_begin_0 = const()[name = tensor("op_11011_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_11011_end_0 = const()[name = tensor("op_11011_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_11011_end_mask_0 = const()[name = tensor("op_11011_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11011_cast_fp16 = slice_by_index(begin = var_11011_begin_0, end = var_11011_end_0, end_mask = var_11011_end_mask_0, x = q_53_cast_fp16)[name = tensor("op_11011_cast_fp16")]; + tensor var_11015_begin_0 = const()[name = tensor("op_11015_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_11015_end_0 = const()[name = tensor("op_11015_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_11015_end_mask_0 = const()[name = tensor("op_11015_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11015_cast_fp16 = slice_by_index(begin = var_11015_begin_0, end = var_11015_end_0, end_mask = var_11015_end_mask_0, x = q_53_cast_fp16)[name = tensor("op_11015_cast_fp16")]; + tensor var_11019_begin_0 = const()[name = tensor("op_11019_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_11019_end_0 = const()[name = tensor("op_11019_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_11019_end_mask_0 = const()[name = tensor("op_11019_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11019_cast_fp16 = slice_by_index(begin = var_11019_begin_0, end = var_11019_end_0, end_mask = var_11019_end_mask_0, x = q_53_cast_fp16)[name = tensor("op_11019_cast_fp16")]; + tensor var_11023_begin_0 = const()[name = tensor("op_11023_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_11023_end_0 = const()[name = tensor("op_11023_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_11023_end_mask_0 = const()[name = tensor("op_11023_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11023_cast_fp16 = slice_by_index(begin = var_11023_begin_0, end = var_11023_end_0, end_mask = var_11023_end_mask_0, x = q_53_cast_fp16)[name = tensor("op_11023_cast_fp16")]; + tensor var_11027_begin_0 = const()[name = tensor("op_11027_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_11027_end_0 = const()[name = tensor("op_11027_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_11027_end_mask_0 = const()[name = tensor("op_11027_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11027_cast_fp16 = slice_by_index(begin = var_11027_begin_0, end = var_11027_end_0, end_mask = var_11027_end_mask_0, x = q_53_cast_fp16)[name = tensor("op_11027_cast_fp16")]; + tensor var_11031_begin_0 = const()[name = tensor("op_11031_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_11031_end_0 = const()[name = tensor("op_11031_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_11031_end_mask_0 = const()[name = tensor("op_11031_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11031_cast_fp16 = slice_by_index(begin = var_11031_begin_0, end = var_11031_end_0, end_mask = var_11031_end_mask_0, x = q_53_cast_fp16)[name = tensor("op_11031_cast_fp16")]; + tensor var_11035_begin_0 = const()[name = tensor("op_11035_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_11035_end_0 = const()[name = tensor("op_11035_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_11035_end_mask_0 = const()[name = tensor("op_11035_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11035_cast_fp16 = slice_by_index(begin = var_11035_begin_0, end = var_11035_end_0, end_mask = var_11035_end_mask_0, x = q_53_cast_fp16)[name = tensor("op_11035_cast_fp16")]; + tensor var_11039_begin_0 = const()[name = tensor("op_11039_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_11039_end_0 = const()[name = tensor("op_11039_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_11039_end_mask_0 = const()[name = tensor("op_11039_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11039_cast_fp16 = slice_by_index(begin = var_11039_begin_0, end = var_11039_end_0, end_mask = var_11039_end_mask_0, x = q_53_cast_fp16)[name = tensor("op_11039_cast_fp16")]; + tensor var_11043_begin_0 = const()[name = tensor("op_11043_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_11043_end_0 = const()[name = tensor("op_11043_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_11043_end_mask_0 = const()[name = tensor("op_11043_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11043_cast_fp16 = slice_by_index(begin = var_11043_begin_0, end = var_11043_end_0, end_mask = var_11043_end_mask_0, x = q_53_cast_fp16)[name = tensor("op_11043_cast_fp16")]; + tensor k_107_perm_0 = const()[name = tensor("k_107_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_11050_begin_0 = const()[name = tensor("op_11050_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11050_end_0 = const()[name = tensor("op_11050_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_11050_end_mask_0 = const()[name = tensor("op_11050_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_107_cast_fp16 = transpose(perm = k_107_perm_0, x = k_105_cast_fp16)[name = tensor("transpose_113")]; + tensor var_11050_cast_fp16 = slice_by_index(begin = var_11050_begin_0, end = var_11050_end_0, end_mask = var_11050_end_mask_0, x = k_107_cast_fp16)[name = tensor("op_11050_cast_fp16")]; + tensor var_11054_begin_0 = const()[name = tensor("op_11054_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_11054_end_0 = const()[name = tensor("op_11054_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_11054_end_mask_0 = const()[name = tensor("op_11054_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11054_cast_fp16 = slice_by_index(begin = var_11054_begin_0, end = var_11054_end_0, end_mask = var_11054_end_mask_0, x = k_107_cast_fp16)[name = tensor("op_11054_cast_fp16")]; + tensor var_11058_begin_0 = const()[name = tensor("op_11058_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_11058_end_0 = const()[name = tensor("op_11058_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_11058_end_mask_0 = const()[name = tensor("op_11058_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11058_cast_fp16 = slice_by_index(begin = var_11058_begin_0, end = var_11058_end_0, end_mask = var_11058_end_mask_0, x = k_107_cast_fp16)[name = tensor("op_11058_cast_fp16")]; + tensor var_11062_begin_0 = const()[name = tensor("op_11062_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_11062_end_0 = const()[name = tensor("op_11062_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_11062_end_mask_0 = const()[name = tensor("op_11062_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11062_cast_fp16 = slice_by_index(begin = var_11062_begin_0, end = var_11062_end_0, end_mask = var_11062_end_mask_0, x = k_107_cast_fp16)[name = tensor("op_11062_cast_fp16")]; + tensor var_11066_begin_0 = const()[name = tensor("op_11066_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_11066_end_0 = const()[name = tensor("op_11066_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_11066_end_mask_0 = const()[name = tensor("op_11066_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11066_cast_fp16 = slice_by_index(begin = var_11066_begin_0, end = var_11066_end_0, end_mask = var_11066_end_mask_0, x = k_107_cast_fp16)[name = tensor("op_11066_cast_fp16")]; + tensor var_11070_begin_0 = const()[name = tensor("op_11070_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_11070_end_0 = const()[name = tensor("op_11070_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_11070_end_mask_0 = const()[name = tensor("op_11070_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11070_cast_fp16 = slice_by_index(begin = var_11070_begin_0, end = var_11070_end_0, end_mask = var_11070_end_mask_0, x = k_107_cast_fp16)[name = tensor("op_11070_cast_fp16")]; + tensor var_11074_begin_0 = const()[name = tensor("op_11074_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_11074_end_0 = const()[name = tensor("op_11074_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_11074_end_mask_0 = const()[name = tensor("op_11074_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11074_cast_fp16 = slice_by_index(begin = var_11074_begin_0, end = var_11074_end_0, end_mask = var_11074_end_mask_0, x = k_107_cast_fp16)[name = tensor("op_11074_cast_fp16")]; + tensor var_11078_begin_0 = const()[name = tensor("op_11078_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_11078_end_0 = const()[name = tensor("op_11078_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_11078_end_mask_0 = const()[name = tensor("op_11078_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11078_cast_fp16 = slice_by_index(begin = var_11078_begin_0, end = var_11078_end_0, end_mask = var_11078_end_mask_0, x = k_107_cast_fp16)[name = tensor("op_11078_cast_fp16")]; + tensor var_11082_begin_0 = const()[name = tensor("op_11082_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_11082_end_0 = const()[name = tensor("op_11082_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_11082_end_mask_0 = const()[name = tensor("op_11082_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11082_cast_fp16 = slice_by_index(begin = var_11082_begin_0, end = var_11082_end_0, end_mask = var_11082_end_mask_0, x = k_107_cast_fp16)[name = tensor("op_11082_cast_fp16")]; + tensor var_11086_begin_0 = const()[name = tensor("op_11086_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_11086_end_0 = const()[name = tensor("op_11086_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_11086_end_mask_0 = const()[name = tensor("op_11086_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11086_cast_fp16 = slice_by_index(begin = var_11086_begin_0, end = var_11086_end_0, end_mask = var_11086_end_mask_0, x = k_107_cast_fp16)[name = tensor("op_11086_cast_fp16")]; + tensor var_11090_begin_0 = const()[name = tensor("op_11090_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_11090_end_0 = const()[name = tensor("op_11090_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_11090_end_mask_0 = const()[name = tensor("op_11090_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11090_cast_fp16 = slice_by_index(begin = var_11090_begin_0, end = var_11090_end_0, end_mask = var_11090_end_mask_0, x = k_107_cast_fp16)[name = tensor("op_11090_cast_fp16")]; + tensor var_11094_begin_0 = const()[name = tensor("op_11094_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_11094_end_0 = const()[name = tensor("op_11094_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_11094_end_mask_0 = const()[name = tensor("op_11094_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11094_cast_fp16 = slice_by_index(begin = var_11094_begin_0, end = var_11094_end_0, end_mask = var_11094_end_mask_0, x = k_107_cast_fp16)[name = tensor("op_11094_cast_fp16")]; + tensor var_11098_begin_0 = const()[name = tensor("op_11098_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_11098_end_0 = const()[name = tensor("op_11098_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_11098_end_mask_0 = const()[name = tensor("op_11098_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11098_cast_fp16 = slice_by_index(begin = var_11098_begin_0, end = var_11098_end_0, end_mask = var_11098_end_mask_0, x = k_107_cast_fp16)[name = tensor("op_11098_cast_fp16")]; + tensor var_11102_begin_0 = const()[name = tensor("op_11102_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_11102_end_0 = const()[name = tensor("op_11102_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_11102_end_mask_0 = const()[name = tensor("op_11102_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11102_cast_fp16 = slice_by_index(begin = var_11102_begin_0, end = var_11102_end_0, end_mask = var_11102_end_mask_0, x = k_107_cast_fp16)[name = tensor("op_11102_cast_fp16")]; + tensor var_11106_begin_0 = const()[name = tensor("op_11106_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_11106_end_0 = const()[name = tensor("op_11106_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_11106_end_mask_0 = const()[name = tensor("op_11106_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11106_cast_fp16 = slice_by_index(begin = var_11106_begin_0, end = var_11106_end_0, end_mask = var_11106_end_mask_0, x = k_107_cast_fp16)[name = tensor("op_11106_cast_fp16")]; + tensor var_11110_begin_0 = const()[name = tensor("op_11110_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_11110_end_0 = const()[name = tensor("op_11110_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_11110_end_mask_0 = const()[name = tensor("op_11110_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11110_cast_fp16 = slice_by_index(begin = var_11110_begin_0, end = var_11110_end_0, end_mask = var_11110_end_mask_0, x = k_107_cast_fp16)[name = tensor("op_11110_cast_fp16")]; + tensor var_11114_begin_0 = const()[name = tensor("op_11114_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_11114_end_0 = const()[name = tensor("op_11114_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_11114_end_mask_0 = const()[name = tensor("op_11114_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11114_cast_fp16 = slice_by_index(begin = var_11114_begin_0, end = var_11114_end_0, end_mask = var_11114_end_mask_0, x = k_107_cast_fp16)[name = tensor("op_11114_cast_fp16")]; + tensor var_11118_begin_0 = const()[name = tensor("op_11118_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_11118_end_0 = const()[name = tensor("op_11118_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_11118_end_mask_0 = const()[name = tensor("op_11118_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11118_cast_fp16 = slice_by_index(begin = var_11118_begin_0, end = var_11118_end_0, end_mask = var_11118_end_mask_0, x = k_107_cast_fp16)[name = tensor("op_11118_cast_fp16")]; + tensor var_11122_begin_0 = const()[name = tensor("op_11122_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_11122_end_0 = const()[name = tensor("op_11122_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_11122_end_mask_0 = const()[name = tensor("op_11122_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11122_cast_fp16 = slice_by_index(begin = var_11122_begin_0, end = var_11122_end_0, end_mask = var_11122_end_mask_0, x = k_107_cast_fp16)[name = tensor("op_11122_cast_fp16")]; + tensor var_11126_begin_0 = const()[name = tensor("op_11126_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_11126_end_0 = const()[name = tensor("op_11126_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_11126_end_mask_0 = const()[name = tensor("op_11126_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11126_cast_fp16 = slice_by_index(begin = var_11126_begin_0, end = var_11126_end_0, end_mask = var_11126_end_mask_0, x = k_107_cast_fp16)[name = tensor("op_11126_cast_fp16")]; + tensor var_11128_begin_0 = const()[name = tensor("op_11128_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11128_end_0 = const()[name = tensor("op_11128_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_11128_end_mask_0 = const()[name = tensor("op_11128_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11128_cast_fp16 = slice_by_index(begin = var_11128_begin_0, end = var_11128_end_0, end_mask = var_11128_end_mask_0, x = v_53_cast_fp16)[name = tensor("op_11128_cast_fp16")]; + tensor var_11132_begin_0 = const()[name = tensor("op_11132_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_11132_end_0 = const()[name = tensor("op_11132_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_11132_end_mask_0 = const()[name = tensor("op_11132_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11132_cast_fp16 = slice_by_index(begin = var_11132_begin_0, end = var_11132_end_0, end_mask = var_11132_end_mask_0, x = v_53_cast_fp16)[name = tensor("op_11132_cast_fp16")]; + tensor var_11136_begin_0 = const()[name = tensor("op_11136_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_11136_end_0 = const()[name = tensor("op_11136_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_11136_end_mask_0 = const()[name = tensor("op_11136_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11136_cast_fp16 = slice_by_index(begin = var_11136_begin_0, end = var_11136_end_0, end_mask = var_11136_end_mask_0, x = v_53_cast_fp16)[name = tensor("op_11136_cast_fp16")]; + tensor var_11140_begin_0 = const()[name = tensor("op_11140_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_11140_end_0 = const()[name = tensor("op_11140_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_11140_end_mask_0 = const()[name = tensor("op_11140_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11140_cast_fp16 = slice_by_index(begin = var_11140_begin_0, end = var_11140_end_0, end_mask = var_11140_end_mask_0, x = v_53_cast_fp16)[name = tensor("op_11140_cast_fp16")]; + tensor var_11144_begin_0 = const()[name = tensor("op_11144_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_11144_end_0 = const()[name = tensor("op_11144_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_11144_end_mask_0 = const()[name = tensor("op_11144_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11144_cast_fp16 = slice_by_index(begin = var_11144_begin_0, end = var_11144_end_0, end_mask = var_11144_end_mask_0, x = v_53_cast_fp16)[name = tensor("op_11144_cast_fp16")]; + tensor var_11148_begin_0 = const()[name = tensor("op_11148_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_11148_end_0 = const()[name = tensor("op_11148_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_11148_end_mask_0 = const()[name = tensor("op_11148_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11148_cast_fp16 = slice_by_index(begin = var_11148_begin_0, end = var_11148_end_0, end_mask = var_11148_end_mask_0, x = v_53_cast_fp16)[name = tensor("op_11148_cast_fp16")]; + tensor var_11152_begin_0 = const()[name = tensor("op_11152_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_11152_end_0 = const()[name = tensor("op_11152_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_11152_end_mask_0 = const()[name = tensor("op_11152_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11152_cast_fp16 = slice_by_index(begin = var_11152_begin_0, end = var_11152_end_0, end_mask = var_11152_end_mask_0, x = v_53_cast_fp16)[name = tensor("op_11152_cast_fp16")]; + tensor var_11156_begin_0 = const()[name = tensor("op_11156_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_11156_end_0 = const()[name = tensor("op_11156_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_11156_end_mask_0 = const()[name = tensor("op_11156_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11156_cast_fp16 = slice_by_index(begin = var_11156_begin_0, end = var_11156_end_0, end_mask = var_11156_end_mask_0, x = v_53_cast_fp16)[name = tensor("op_11156_cast_fp16")]; + tensor var_11160_begin_0 = const()[name = tensor("op_11160_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_11160_end_0 = const()[name = tensor("op_11160_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_11160_end_mask_0 = const()[name = tensor("op_11160_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11160_cast_fp16 = slice_by_index(begin = var_11160_begin_0, end = var_11160_end_0, end_mask = var_11160_end_mask_0, x = v_53_cast_fp16)[name = tensor("op_11160_cast_fp16")]; + tensor var_11164_begin_0 = const()[name = tensor("op_11164_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_11164_end_0 = const()[name = tensor("op_11164_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_11164_end_mask_0 = const()[name = tensor("op_11164_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11164_cast_fp16 = slice_by_index(begin = var_11164_begin_0, end = var_11164_end_0, end_mask = var_11164_end_mask_0, x = v_53_cast_fp16)[name = tensor("op_11164_cast_fp16")]; + tensor var_11168_begin_0 = const()[name = tensor("op_11168_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_11168_end_0 = const()[name = tensor("op_11168_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_11168_end_mask_0 = const()[name = tensor("op_11168_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11168_cast_fp16 = slice_by_index(begin = var_11168_begin_0, end = var_11168_end_0, end_mask = var_11168_end_mask_0, x = v_53_cast_fp16)[name = tensor("op_11168_cast_fp16")]; + tensor var_11172_begin_0 = const()[name = tensor("op_11172_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_11172_end_0 = const()[name = tensor("op_11172_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_11172_end_mask_0 = const()[name = tensor("op_11172_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11172_cast_fp16 = slice_by_index(begin = var_11172_begin_0, end = var_11172_end_0, end_mask = var_11172_end_mask_0, x = v_53_cast_fp16)[name = tensor("op_11172_cast_fp16")]; + tensor var_11176_begin_0 = const()[name = tensor("op_11176_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_11176_end_0 = const()[name = tensor("op_11176_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_11176_end_mask_0 = const()[name = tensor("op_11176_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11176_cast_fp16 = slice_by_index(begin = var_11176_begin_0, end = var_11176_end_0, end_mask = var_11176_end_mask_0, x = v_53_cast_fp16)[name = tensor("op_11176_cast_fp16")]; + tensor var_11180_begin_0 = const()[name = tensor("op_11180_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_11180_end_0 = const()[name = tensor("op_11180_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_11180_end_mask_0 = const()[name = tensor("op_11180_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11180_cast_fp16 = slice_by_index(begin = var_11180_begin_0, end = var_11180_end_0, end_mask = var_11180_end_mask_0, x = v_53_cast_fp16)[name = tensor("op_11180_cast_fp16")]; + tensor var_11184_begin_0 = const()[name = tensor("op_11184_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_11184_end_0 = const()[name = tensor("op_11184_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_11184_end_mask_0 = const()[name = tensor("op_11184_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11184_cast_fp16 = slice_by_index(begin = var_11184_begin_0, end = var_11184_end_0, end_mask = var_11184_end_mask_0, x = v_53_cast_fp16)[name = tensor("op_11184_cast_fp16")]; + tensor var_11188_begin_0 = const()[name = tensor("op_11188_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_11188_end_0 = const()[name = tensor("op_11188_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_11188_end_mask_0 = const()[name = tensor("op_11188_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11188_cast_fp16 = slice_by_index(begin = var_11188_begin_0, end = var_11188_end_0, end_mask = var_11188_end_mask_0, x = v_53_cast_fp16)[name = tensor("op_11188_cast_fp16")]; + tensor var_11192_begin_0 = const()[name = tensor("op_11192_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_11192_end_0 = const()[name = tensor("op_11192_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_11192_end_mask_0 = const()[name = tensor("op_11192_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11192_cast_fp16 = slice_by_index(begin = var_11192_begin_0, end = var_11192_end_0, end_mask = var_11192_end_mask_0, x = v_53_cast_fp16)[name = tensor("op_11192_cast_fp16")]; + tensor var_11196_begin_0 = const()[name = tensor("op_11196_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_11196_end_0 = const()[name = tensor("op_11196_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_11196_end_mask_0 = const()[name = tensor("op_11196_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11196_cast_fp16 = slice_by_index(begin = var_11196_begin_0, end = var_11196_end_0, end_mask = var_11196_end_mask_0, x = v_53_cast_fp16)[name = tensor("op_11196_cast_fp16")]; + tensor var_11200_begin_0 = const()[name = tensor("op_11200_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_11200_end_0 = const()[name = tensor("op_11200_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_11200_end_mask_0 = const()[name = tensor("op_11200_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11200_cast_fp16 = slice_by_index(begin = var_11200_begin_0, end = var_11200_end_0, end_mask = var_11200_end_mask_0, x = v_53_cast_fp16)[name = tensor("op_11200_cast_fp16")]; + tensor var_11204_begin_0 = const()[name = tensor("op_11204_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_11204_end_0 = const()[name = tensor("op_11204_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_11204_end_mask_0 = const()[name = tensor("op_11204_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11204_cast_fp16 = slice_by_index(begin = var_11204_begin_0, end = var_11204_end_0, end_mask = var_11204_end_mask_0, x = v_53_cast_fp16)[name = tensor("op_11204_cast_fp16")]; + tensor var_11208_equation_0 = const()[name = tensor("op_11208_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11208_cast_fp16 = einsum(equation = var_11208_equation_0, values = (var_11050_cast_fp16, var_10967_cast_fp16))[name = tensor("op_11208_cast_fp16")]; + tensor var_11209_to_fp16 = const()[name = tensor("op_11209_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_881_cast_fp16 = mul(x = var_11208_cast_fp16, y = var_11209_to_fp16)[name = tensor("aw_881_cast_fp16")]; + tensor var_11212_equation_0 = const()[name = tensor("op_11212_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11212_cast_fp16 = einsum(equation = var_11212_equation_0, values = (var_11054_cast_fp16, var_10971_cast_fp16))[name = tensor("op_11212_cast_fp16")]; + tensor var_11213_to_fp16 = const()[name = tensor("op_11213_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_883_cast_fp16 = mul(x = var_11212_cast_fp16, y = var_11213_to_fp16)[name = tensor("aw_883_cast_fp16")]; + tensor var_11216_equation_0 = const()[name = tensor("op_11216_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11216_cast_fp16 = einsum(equation = var_11216_equation_0, values = (var_11058_cast_fp16, var_10975_cast_fp16))[name = tensor("op_11216_cast_fp16")]; + tensor var_11217_to_fp16 = const()[name = tensor("op_11217_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_885_cast_fp16 = mul(x = var_11216_cast_fp16, y = var_11217_to_fp16)[name = tensor("aw_885_cast_fp16")]; + tensor var_11220_equation_0 = const()[name = tensor("op_11220_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11220_cast_fp16 = einsum(equation = var_11220_equation_0, values = (var_11062_cast_fp16, var_10979_cast_fp16))[name = tensor("op_11220_cast_fp16")]; + tensor var_11221_to_fp16 = const()[name = tensor("op_11221_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_887_cast_fp16 = mul(x = var_11220_cast_fp16, y = var_11221_to_fp16)[name = tensor("aw_887_cast_fp16")]; + tensor var_11224_equation_0 = const()[name = tensor("op_11224_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11224_cast_fp16 = einsum(equation = var_11224_equation_0, values = (var_11066_cast_fp16, var_10983_cast_fp16))[name = tensor("op_11224_cast_fp16")]; + tensor var_11225_to_fp16 = const()[name = tensor("op_11225_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_889_cast_fp16 = mul(x = var_11224_cast_fp16, y = var_11225_to_fp16)[name = tensor("aw_889_cast_fp16")]; + tensor var_11228_equation_0 = const()[name = tensor("op_11228_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11228_cast_fp16 = einsum(equation = var_11228_equation_0, values = (var_11070_cast_fp16, var_10987_cast_fp16))[name = tensor("op_11228_cast_fp16")]; + tensor var_11229_to_fp16 = const()[name = tensor("op_11229_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_891_cast_fp16 = mul(x = var_11228_cast_fp16, y = var_11229_to_fp16)[name = tensor("aw_891_cast_fp16")]; + tensor var_11232_equation_0 = const()[name = tensor("op_11232_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11232_cast_fp16 = einsum(equation = var_11232_equation_0, values = (var_11074_cast_fp16, var_10991_cast_fp16))[name = tensor("op_11232_cast_fp16")]; + tensor var_11233_to_fp16 = const()[name = tensor("op_11233_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_893_cast_fp16 = mul(x = var_11232_cast_fp16, y = var_11233_to_fp16)[name = tensor("aw_893_cast_fp16")]; + tensor var_11236_equation_0 = const()[name = tensor("op_11236_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11236_cast_fp16 = einsum(equation = var_11236_equation_0, values = (var_11078_cast_fp16, var_10995_cast_fp16))[name = tensor("op_11236_cast_fp16")]; + tensor var_11237_to_fp16 = const()[name = tensor("op_11237_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_895_cast_fp16 = mul(x = var_11236_cast_fp16, y = var_11237_to_fp16)[name = tensor("aw_895_cast_fp16")]; + tensor var_11240_equation_0 = const()[name = tensor("op_11240_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11240_cast_fp16 = einsum(equation = var_11240_equation_0, values = (var_11082_cast_fp16, var_10999_cast_fp16))[name = tensor("op_11240_cast_fp16")]; + tensor var_11241_to_fp16 = const()[name = tensor("op_11241_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_897_cast_fp16 = mul(x = var_11240_cast_fp16, y = var_11241_to_fp16)[name = tensor("aw_897_cast_fp16")]; + tensor var_11244_equation_0 = const()[name = tensor("op_11244_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11244_cast_fp16 = einsum(equation = var_11244_equation_0, values = (var_11086_cast_fp16, var_11003_cast_fp16))[name = tensor("op_11244_cast_fp16")]; + tensor var_11245_to_fp16 = const()[name = tensor("op_11245_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_899_cast_fp16 = mul(x = var_11244_cast_fp16, y = var_11245_to_fp16)[name = tensor("aw_899_cast_fp16")]; + tensor var_11248_equation_0 = const()[name = tensor("op_11248_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11248_cast_fp16 = einsum(equation = var_11248_equation_0, values = (var_11090_cast_fp16, var_11007_cast_fp16))[name = tensor("op_11248_cast_fp16")]; + tensor var_11249_to_fp16 = const()[name = tensor("op_11249_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_901_cast_fp16 = mul(x = var_11248_cast_fp16, y = var_11249_to_fp16)[name = tensor("aw_901_cast_fp16")]; + tensor var_11252_equation_0 = const()[name = tensor("op_11252_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11252_cast_fp16 = einsum(equation = var_11252_equation_0, values = (var_11094_cast_fp16, var_11011_cast_fp16))[name = tensor("op_11252_cast_fp16")]; + tensor var_11253_to_fp16 = const()[name = tensor("op_11253_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_903_cast_fp16 = mul(x = var_11252_cast_fp16, y = var_11253_to_fp16)[name = tensor("aw_903_cast_fp16")]; + tensor var_11256_equation_0 = const()[name = tensor("op_11256_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11256_cast_fp16 = einsum(equation = var_11256_equation_0, values = (var_11098_cast_fp16, var_11015_cast_fp16))[name = tensor("op_11256_cast_fp16")]; + tensor var_11257_to_fp16 = const()[name = tensor("op_11257_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_905_cast_fp16 = mul(x = var_11256_cast_fp16, y = var_11257_to_fp16)[name = tensor("aw_905_cast_fp16")]; + tensor var_11260_equation_0 = const()[name = tensor("op_11260_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11260_cast_fp16 = einsum(equation = var_11260_equation_0, values = (var_11102_cast_fp16, var_11019_cast_fp16))[name = tensor("op_11260_cast_fp16")]; + tensor var_11261_to_fp16 = const()[name = tensor("op_11261_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_907_cast_fp16 = mul(x = var_11260_cast_fp16, y = var_11261_to_fp16)[name = tensor("aw_907_cast_fp16")]; + tensor var_11264_equation_0 = const()[name = tensor("op_11264_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11264_cast_fp16 = einsum(equation = var_11264_equation_0, values = (var_11106_cast_fp16, var_11023_cast_fp16))[name = tensor("op_11264_cast_fp16")]; + tensor var_11265_to_fp16 = const()[name = tensor("op_11265_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_909_cast_fp16 = mul(x = var_11264_cast_fp16, y = var_11265_to_fp16)[name = tensor("aw_909_cast_fp16")]; + tensor var_11268_equation_0 = const()[name = tensor("op_11268_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11268_cast_fp16 = einsum(equation = var_11268_equation_0, values = (var_11110_cast_fp16, var_11027_cast_fp16))[name = tensor("op_11268_cast_fp16")]; + tensor var_11269_to_fp16 = const()[name = tensor("op_11269_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_911_cast_fp16 = mul(x = var_11268_cast_fp16, y = var_11269_to_fp16)[name = tensor("aw_911_cast_fp16")]; + tensor var_11272_equation_0 = const()[name = tensor("op_11272_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11272_cast_fp16 = einsum(equation = var_11272_equation_0, values = (var_11114_cast_fp16, var_11031_cast_fp16))[name = tensor("op_11272_cast_fp16")]; + tensor var_11273_to_fp16 = const()[name = tensor("op_11273_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_913_cast_fp16 = mul(x = var_11272_cast_fp16, y = var_11273_to_fp16)[name = tensor("aw_913_cast_fp16")]; + tensor var_11276_equation_0 = const()[name = tensor("op_11276_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11276_cast_fp16 = einsum(equation = var_11276_equation_0, values = (var_11118_cast_fp16, var_11035_cast_fp16))[name = tensor("op_11276_cast_fp16")]; + tensor var_11277_to_fp16 = const()[name = tensor("op_11277_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_915_cast_fp16 = mul(x = var_11276_cast_fp16, y = var_11277_to_fp16)[name = tensor("aw_915_cast_fp16")]; + tensor var_11280_equation_0 = const()[name = tensor("op_11280_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11280_cast_fp16 = einsum(equation = var_11280_equation_0, values = (var_11122_cast_fp16, var_11039_cast_fp16))[name = tensor("op_11280_cast_fp16")]; + tensor var_11281_to_fp16 = const()[name = tensor("op_11281_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_917_cast_fp16 = mul(x = var_11280_cast_fp16, y = var_11281_to_fp16)[name = tensor("aw_917_cast_fp16")]; + tensor var_11284_equation_0 = const()[name = tensor("op_11284_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11284_cast_fp16 = einsum(equation = var_11284_equation_0, values = (var_11126_cast_fp16, var_11043_cast_fp16))[name = tensor("op_11284_cast_fp16")]; + tensor var_11285_to_fp16 = const()[name = tensor("op_11285_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_919_cast_fp16 = mul(x = var_11284_cast_fp16, y = var_11285_to_fp16)[name = tensor("aw_919_cast_fp16")]; + tensor var_11287_cast_fp16 = softmax(axis = var_2624, x = aw_881_cast_fp16)[name = tensor("op_11287_cast_fp16")]; + tensor var_11288_cast_fp16 = softmax(axis = var_2624, x = aw_883_cast_fp16)[name = tensor("op_11288_cast_fp16")]; + tensor var_11289_cast_fp16 = softmax(axis = var_2624, x = aw_885_cast_fp16)[name = tensor("op_11289_cast_fp16")]; + tensor var_11290_cast_fp16 = softmax(axis = var_2624, x = aw_887_cast_fp16)[name = tensor("op_11290_cast_fp16")]; + tensor var_11291_cast_fp16 = softmax(axis = var_2624, x = aw_889_cast_fp16)[name = tensor("op_11291_cast_fp16")]; + tensor var_11292_cast_fp16 = softmax(axis = var_2624, x = aw_891_cast_fp16)[name = tensor("op_11292_cast_fp16")]; + tensor var_11293_cast_fp16 = softmax(axis = var_2624, x = aw_893_cast_fp16)[name = tensor("op_11293_cast_fp16")]; + tensor var_11294_cast_fp16 = softmax(axis = var_2624, x = aw_895_cast_fp16)[name = tensor("op_11294_cast_fp16")]; + tensor var_11295_cast_fp16 = softmax(axis = var_2624, x = aw_897_cast_fp16)[name = tensor("op_11295_cast_fp16")]; + tensor var_11296_cast_fp16 = softmax(axis = var_2624, x = aw_899_cast_fp16)[name = tensor("op_11296_cast_fp16")]; + tensor var_11297_cast_fp16 = softmax(axis = var_2624, x = aw_901_cast_fp16)[name = tensor("op_11297_cast_fp16")]; + tensor var_11298_cast_fp16 = softmax(axis = var_2624, x = aw_903_cast_fp16)[name = tensor("op_11298_cast_fp16")]; + tensor var_11299_cast_fp16 = softmax(axis = var_2624, x = aw_905_cast_fp16)[name = tensor("op_11299_cast_fp16")]; + tensor var_11300_cast_fp16 = softmax(axis = var_2624, x = aw_907_cast_fp16)[name = tensor("op_11300_cast_fp16")]; + tensor var_11301_cast_fp16 = softmax(axis = var_2624, x = aw_909_cast_fp16)[name = tensor("op_11301_cast_fp16")]; + tensor var_11302_cast_fp16 = softmax(axis = var_2624, x = aw_911_cast_fp16)[name = tensor("op_11302_cast_fp16")]; + tensor var_11303_cast_fp16 = softmax(axis = var_2624, x = aw_913_cast_fp16)[name = tensor("op_11303_cast_fp16")]; + tensor var_11304_cast_fp16 = softmax(axis = var_2624, x = aw_915_cast_fp16)[name = tensor("op_11304_cast_fp16")]; + tensor var_11305_cast_fp16 = softmax(axis = var_2624, x = aw_917_cast_fp16)[name = tensor("op_11305_cast_fp16")]; + tensor var_11306_cast_fp16 = softmax(axis = var_2624, x = aw_919_cast_fp16)[name = tensor("op_11306_cast_fp16")]; + tensor var_11308_equation_0 = const()[name = tensor("op_11308_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11308_cast_fp16 = einsum(equation = var_11308_equation_0, values = (var_11128_cast_fp16, var_11287_cast_fp16))[name = tensor("op_11308_cast_fp16")]; + tensor var_11310_equation_0 = const()[name = tensor("op_11310_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11310_cast_fp16 = einsum(equation = var_11310_equation_0, values = (var_11132_cast_fp16, var_11288_cast_fp16))[name = tensor("op_11310_cast_fp16")]; + tensor var_11312_equation_0 = const()[name = tensor("op_11312_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11312_cast_fp16 = einsum(equation = var_11312_equation_0, values = (var_11136_cast_fp16, var_11289_cast_fp16))[name = tensor("op_11312_cast_fp16")]; + tensor var_11314_equation_0 = const()[name = tensor("op_11314_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11314_cast_fp16 = einsum(equation = var_11314_equation_0, values = (var_11140_cast_fp16, var_11290_cast_fp16))[name = tensor("op_11314_cast_fp16")]; + tensor var_11316_equation_0 = const()[name = tensor("op_11316_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11316_cast_fp16 = einsum(equation = var_11316_equation_0, values = (var_11144_cast_fp16, var_11291_cast_fp16))[name = tensor("op_11316_cast_fp16")]; + tensor var_11318_equation_0 = const()[name = tensor("op_11318_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11318_cast_fp16 = einsum(equation = var_11318_equation_0, values = (var_11148_cast_fp16, var_11292_cast_fp16))[name = tensor("op_11318_cast_fp16")]; + tensor var_11320_equation_0 = const()[name = tensor("op_11320_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11320_cast_fp16 = einsum(equation = var_11320_equation_0, values = (var_11152_cast_fp16, var_11293_cast_fp16))[name = tensor("op_11320_cast_fp16")]; + tensor var_11322_equation_0 = const()[name = tensor("op_11322_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11322_cast_fp16 = einsum(equation = var_11322_equation_0, values = (var_11156_cast_fp16, var_11294_cast_fp16))[name = tensor("op_11322_cast_fp16")]; + tensor var_11324_equation_0 = const()[name = tensor("op_11324_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11324_cast_fp16 = einsum(equation = var_11324_equation_0, values = (var_11160_cast_fp16, var_11295_cast_fp16))[name = tensor("op_11324_cast_fp16")]; + tensor var_11326_equation_0 = const()[name = tensor("op_11326_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11326_cast_fp16 = einsum(equation = var_11326_equation_0, values = (var_11164_cast_fp16, var_11296_cast_fp16))[name = tensor("op_11326_cast_fp16")]; + tensor var_11328_equation_0 = const()[name = tensor("op_11328_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11328_cast_fp16 = einsum(equation = var_11328_equation_0, values = (var_11168_cast_fp16, var_11297_cast_fp16))[name = tensor("op_11328_cast_fp16")]; + tensor var_11330_equation_0 = const()[name = tensor("op_11330_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11330_cast_fp16 = einsum(equation = var_11330_equation_0, values = (var_11172_cast_fp16, var_11298_cast_fp16))[name = tensor("op_11330_cast_fp16")]; + tensor var_11332_equation_0 = const()[name = tensor("op_11332_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11332_cast_fp16 = einsum(equation = var_11332_equation_0, values = (var_11176_cast_fp16, var_11299_cast_fp16))[name = tensor("op_11332_cast_fp16")]; + tensor var_11334_equation_0 = const()[name = tensor("op_11334_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11334_cast_fp16 = einsum(equation = var_11334_equation_0, values = (var_11180_cast_fp16, var_11300_cast_fp16))[name = tensor("op_11334_cast_fp16")]; + tensor var_11336_equation_0 = const()[name = tensor("op_11336_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11336_cast_fp16 = einsum(equation = var_11336_equation_0, values = (var_11184_cast_fp16, var_11301_cast_fp16))[name = tensor("op_11336_cast_fp16")]; + tensor var_11338_equation_0 = const()[name = tensor("op_11338_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11338_cast_fp16 = einsum(equation = var_11338_equation_0, values = (var_11188_cast_fp16, var_11302_cast_fp16))[name = tensor("op_11338_cast_fp16")]; + tensor var_11340_equation_0 = const()[name = tensor("op_11340_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11340_cast_fp16 = einsum(equation = var_11340_equation_0, values = (var_11192_cast_fp16, var_11303_cast_fp16))[name = tensor("op_11340_cast_fp16")]; + tensor var_11342_equation_0 = const()[name = tensor("op_11342_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11342_cast_fp16 = einsum(equation = var_11342_equation_0, values = (var_11196_cast_fp16, var_11304_cast_fp16))[name = tensor("op_11342_cast_fp16")]; + tensor var_11344_equation_0 = const()[name = tensor("op_11344_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11344_cast_fp16 = einsum(equation = var_11344_equation_0, values = (var_11200_cast_fp16, var_11305_cast_fp16))[name = tensor("op_11344_cast_fp16")]; + tensor var_11346_equation_0 = const()[name = tensor("op_11346_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11346_cast_fp16 = einsum(equation = var_11346_equation_0, values = (var_11204_cast_fp16, var_11306_cast_fp16))[name = tensor("op_11346_cast_fp16")]; + tensor input_203_interleave_0 = const()[name = tensor("input_203_interleave_0"), val = tensor(false)]; + tensor input_203_cast_fp16 = concat(axis = var_2624, interleave = input_203_interleave_0, values = (var_11308_cast_fp16, var_11310_cast_fp16, var_11312_cast_fp16, var_11314_cast_fp16, var_11316_cast_fp16, var_11318_cast_fp16, var_11320_cast_fp16, var_11322_cast_fp16, var_11324_cast_fp16, var_11326_cast_fp16, var_11328_cast_fp16, var_11330_cast_fp16, var_11332_cast_fp16, var_11334_cast_fp16, var_11336_cast_fp16, var_11338_cast_fp16, var_11340_cast_fp16, var_11342_cast_fp16, var_11344_cast_fp16, var_11346_cast_fp16))[name = tensor("input_203_cast_fp16")]; + tensor var_11352 = const()[name = tensor("op_11352"), val = tensor([1, 1])]; + tensor var_11354 = const()[name = tensor("op_11354"), val = tensor([1, 1])]; + tensor var_11356_pad_type_0 = const()[name = tensor("op_11356_pad_type_0"), val = tensor("custom")]; + tensor var_11356_pad_0 = const()[name = tensor("op_11356_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_9_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_9_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(833618368)))]; + tensor down_blocks_2_attentions_0_transformer_blocks_9_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_9_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(836895232)))]; + tensor var_11356_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_9_attn1_to_out_0_bias_to_fp16, dilations = var_11354, groups = var_2624, pad = var_11356_pad_0, pad_type = var_11356_pad_type_0, strides = var_11352, weight = down_blocks_2_attentions_0_transformer_blocks_9_attn1_to_out_0_weight_to_fp16, x = input_203_cast_fp16)[name = tensor("op_11356_cast_fp16")]; + tensor inputs_81_cast_fp16 = add(x = var_11356_cast_fp16, y = inputs_79_cast_fp16)[name = tensor("inputs_81_cast_fp16")]; + tensor hidden_states_121_axes_0 = const()[name = tensor("hidden_states_121_axes_0"), val = tensor([1])]; + tensor hidden_states_121_gamma_0_to_fp16 = const()[name = tensor("hidden_states_121_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(836897856)))]; + tensor hidden_states_121_beta_0_to_fp16 = const()[name = tensor("hidden_states_121_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(836900480)))]; + tensor var_11366_to_fp16 = const()[name = tensor("op_11366_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_121_cast_fp16 = layer_norm(axes = hidden_states_121_axes_0, beta = hidden_states_121_beta_0_to_fp16, epsilon = var_11366_to_fp16, gamma = hidden_states_121_gamma_0_to_fp16, x = inputs_81_cast_fp16)[name = tensor("hidden_states_121_cast_fp16")]; + tensor var_11381 = const()[name = tensor("op_11381"), val = tensor([1, 1])]; + tensor var_11383 = const()[name = tensor("op_11383"), val = tensor([1, 1])]; + tensor q_55_pad_type_0 = const()[name = tensor("q_55_pad_type_0"), val = tensor("custom")]; + tensor q_55_pad_0 = const()[name = tensor("q_55_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_9_attn2_to_q_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_9_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(836903104)))]; + tensor q_55_cast_fp16 = conv(dilations = var_11383, groups = var_2624, pad = q_55_pad_0, pad_type = q_55_pad_type_0, strides = var_11381, weight = down_blocks_2_attentions_0_transformer_blocks_9_attn2_to_q_weight_to_fp16, x = hidden_states_121_cast_fp16)[name = tensor("q_55_cast_fp16")]; + tensor var_11387 = const()[name = tensor("op_11387"), val = tensor([1, 1])]; + tensor var_11389 = const()[name = tensor("op_11389"), val = tensor([1, 1])]; + tensor k_109_pad_type_0 = const()[name = tensor("k_109_pad_type_0"), val = tensor("custom")]; + tensor k_109_pad_0 = const()[name = tensor("k_109_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_9_attn2_to_k_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_9_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(840179968)))]; + tensor k_109_cast_fp16 = conv(dilations = var_11389, groups = var_2624, pad = k_109_pad_0, pad_type = k_109_pad_type_0, strides = var_11387, weight = down_blocks_2_attentions_0_transformer_blocks_9_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_109_cast_fp16")]; + tensor var_11393 = const()[name = tensor("op_11393"), val = tensor([1, 1])]; + tensor var_11395 = const()[name = tensor("op_11395"), val = tensor([1, 1])]; + tensor v_55_pad_type_0 = const()[name = tensor("v_55_pad_type_0"), val = tensor("custom")]; + tensor v_55_pad_0 = const()[name = tensor("v_55_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_9_attn2_to_v_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_9_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(845422912)))]; + tensor v_55_cast_fp16 = conv(dilations = var_11395, groups = var_2624, pad = v_55_pad_0, pad_type = v_55_pad_type_0, strides = var_11393, weight = down_blocks_2_attentions_0_transformer_blocks_9_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_55_cast_fp16")]; + tensor var_11399_begin_0 = const()[name = tensor("op_11399_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11399_end_0 = const()[name = tensor("op_11399_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_11399_end_mask_0 = const()[name = tensor("op_11399_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11399_cast_fp16 = slice_by_index(begin = var_11399_begin_0, end = var_11399_end_0, end_mask = var_11399_end_mask_0, x = q_55_cast_fp16)[name = tensor("op_11399_cast_fp16")]; + tensor var_11403_begin_0 = const()[name = tensor("op_11403_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_11403_end_0 = const()[name = tensor("op_11403_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_11403_end_mask_0 = const()[name = tensor("op_11403_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11403_cast_fp16 = slice_by_index(begin = var_11403_begin_0, end = var_11403_end_0, end_mask = var_11403_end_mask_0, x = q_55_cast_fp16)[name = tensor("op_11403_cast_fp16")]; + tensor var_11407_begin_0 = const()[name = tensor("op_11407_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_11407_end_0 = const()[name = tensor("op_11407_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_11407_end_mask_0 = const()[name = tensor("op_11407_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11407_cast_fp16 = slice_by_index(begin = var_11407_begin_0, end = var_11407_end_0, end_mask = var_11407_end_mask_0, x = q_55_cast_fp16)[name = tensor("op_11407_cast_fp16")]; + tensor var_11411_begin_0 = const()[name = tensor("op_11411_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_11411_end_0 = const()[name = tensor("op_11411_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_11411_end_mask_0 = const()[name = tensor("op_11411_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11411_cast_fp16 = slice_by_index(begin = var_11411_begin_0, end = var_11411_end_0, end_mask = var_11411_end_mask_0, x = q_55_cast_fp16)[name = tensor("op_11411_cast_fp16")]; + tensor var_11415_begin_0 = const()[name = tensor("op_11415_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_11415_end_0 = const()[name = tensor("op_11415_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_11415_end_mask_0 = const()[name = tensor("op_11415_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11415_cast_fp16 = slice_by_index(begin = var_11415_begin_0, end = var_11415_end_0, end_mask = var_11415_end_mask_0, x = q_55_cast_fp16)[name = tensor("op_11415_cast_fp16")]; + tensor var_11419_begin_0 = const()[name = tensor("op_11419_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_11419_end_0 = const()[name = tensor("op_11419_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_11419_end_mask_0 = const()[name = tensor("op_11419_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11419_cast_fp16 = slice_by_index(begin = var_11419_begin_0, end = var_11419_end_0, end_mask = var_11419_end_mask_0, x = q_55_cast_fp16)[name = tensor("op_11419_cast_fp16")]; + tensor var_11423_begin_0 = const()[name = tensor("op_11423_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_11423_end_0 = const()[name = tensor("op_11423_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_11423_end_mask_0 = const()[name = tensor("op_11423_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11423_cast_fp16 = slice_by_index(begin = var_11423_begin_0, end = var_11423_end_0, end_mask = var_11423_end_mask_0, x = q_55_cast_fp16)[name = tensor("op_11423_cast_fp16")]; + tensor var_11427_begin_0 = const()[name = tensor("op_11427_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_11427_end_0 = const()[name = tensor("op_11427_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_11427_end_mask_0 = const()[name = tensor("op_11427_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11427_cast_fp16 = slice_by_index(begin = var_11427_begin_0, end = var_11427_end_0, end_mask = var_11427_end_mask_0, x = q_55_cast_fp16)[name = tensor("op_11427_cast_fp16")]; + tensor var_11431_begin_0 = const()[name = tensor("op_11431_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_11431_end_0 = const()[name = tensor("op_11431_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_11431_end_mask_0 = const()[name = tensor("op_11431_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11431_cast_fp16 = slice_by_index(begin = var_11431_begin_0, end = var_11431_end_0, end_mask = var_11431_end_mask_0, x = q_55_cast_fp16)[name = tensor("op_11431_cast_fp16")]; + tensor var_11435_begin_0 = const()[name = tensor("op_11435_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_11435_end_0 = const()[name = tensor("op_11435_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_11435_end_mask_0 = const()[name = tensor("op_11435_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11435_cast_fp16 = slice_by_index(begin = var_11435_begin_0, end = var_11435_end_0, end_mask = var_11435_end_mask_0, x = q_55_cast_fp16)[name = tensor("op_11435_cast_fp16")]; + tensor var_11439_begin_0 = const()[name = tensor("op_11439_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_11439_end_0 = const()[name = tensor("op_11439_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_11439_end_mask_0 = const()[name = tensor("op_11439_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11439_cast_fp16 = slice_by_index(begin = var_11439_begin_0, end = var_11439_end_0, end_mask = var_11439_end_mask_0, x = q_55_cast_fp16)[name = tensor("op_11439_cast_fp16")]; + tensor var_11443_begin_0 = const()[name = tensor("op_11443_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_11443_end_0 = const()[name = tensor("op_11443_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_11443_end_mask_0 = const()[name = tensor("op_11443_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11443_cast_fp16 = slice_by_index(begin = var_11443_begin_0, end = var_11443_end_0, end_mask = var_11443_end_mask_0, x = q_55_cast_fp16)[name = tensor("op_11443_cast_fp16")]; + tensor var_11447_begin_0 = const()[name = tensor("op_11447_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_11447_end_0 = const()[name = tensor("op_11447_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_11447_end_mask_0 = const()[name = tensor("op_11447_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11447_cast_fp16 = slice_by_index(begin = var_11447_begin_0, end = var_11447_end_0, end_mask = var_11447_end_mask_0, x = q_55_cast_fp16)[name = tensor("op_11447_cast_fp16")]; + tensor var_11451_begin_0 = const()[name = tensor("op_11451_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_11451_end_0 = const()[name = tensor("op_11451_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_11451_end_mask_0 = const()[name = tensor("op_11451_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11451_cast_fp16 = slice_by_index(begin = var_11451_begin_0, end = var_11451_end_0, end_mask = var_11451_end_mask_0, x = q_55_cast_fp16)[name = tensor("op_11451_cast_fp16")]; + tensor var_11455_begin_0 = const()[name = tensor("op_11455_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_11455_end_0 = const()[name = tensor("op_11455_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_11455_end_mask_0 = const()[name = tensor("op_11455_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11455_cast_fp16 = slice_by_index(begin = var_11455_begin_0, end = var_11455_end_0, end_mask = var_11455_end_mask_0, x = q_55_cast_fp16)[name = tensor("op_11455_cast_fp16")]; + tensor var_11459_begin_0 = const()[name = tensor("op_11459_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_11459_end_0 = const()[name = tensor("op_11459_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_11459_end_mask_0 = const()[name = tensor("op_11459_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11459_cast_fp16 = slice_by_index(begin = var_11459_begin_0, end = var_11459_end_0, end_mask = var_11459_end_mask_0, x = q_55_cast_fp16)[name = tensor("op_11459_cast_fp16")]; + tensor var_11463_begin_0 = const()[name = tensor("op_11463_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_11463_end_0 = const()[name = tensor("op_11463_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_11463_end_mask_0 = const()[name = tensor("op_11463_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11463_cast_fp16 = slice_by_index(begin = var_11463_begin_0, end = var_11463_end_0, end_mask = var_11463_end_mask_0, x = q_55_cast_fp16)[name = tensor("op_11463_cast_fp16")]; + tensor var_11467_begin_0 = const()[name = tensor("op_11467_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_11467_end_0 = const()[name = tensor("op_11467_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_11467_end_mask_0 = const()[name = tensor("op_11467_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11467_cast_fp16 = slice_by_index(begin = var_11467_begin_0, end = var_11467_end_0, end_mask = var_11467_end_mask_0, x = q_55_cast_fp16)[name = tensor("op_11467_cast_fp16")]; + tensor var_11471_begin_0 = const()[name = tensor("op_11471_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_11471_end_0 = const()[name = tensor("op_11471_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_11471_end_mask_0 = const()[name = tensor("op_11471_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11471_cast_fp16 = slice_by_index(begin = var_11471_begin_0, end = var_11471_end_0, end_mask = var_11471_end_mask_0, x = q_55_cast_fp16)[name = tensor("op_11471_cast_fp16")]; + tensor var_11475_begin_0 = const()[name = tensor("op_11475_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_11475_end_0 = const()[name = tensor("op_11475_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_11475_end_mask_0 = const()[name = tensor("op_11475_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11475_cast_fp16 = slice_by_index(begin = var_11475_begin_0, end = var_11475_end_0, end_mask = var_11475_end_mask_0, x = q_55_cast_fp16)[name = tensor("op_11475_cast_fp16")]; + tensor k_111_perm_0 = const()[name = tensor("k_111_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_11482_begin_0 = const()[name = tensor("op_11482_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11482_end_0 = const()[name = tensor("op_11482_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_11482_end_mask_0 = const()[name = tensor("op_11482_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_111_cast_fp16 = transpose(perm = k_111_perm_0, x = k_109_cast_fp16)[name = tensor("transpose_112")]; + tensor var_11482_cast_fp16 = slice_by_index(begin = var_11482_begin_0, end = var_11482_end_0, end_mask = var_11482_end_mask_0, x = k_111_cast_fp16)[name = tensor("op_11482_cast_fp16")]; + tensor var_11486_begin_0 = const()[name = tensor("op_11486_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_11486_end_0 = const()[name = tensor("op_11486_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_11486_end_mask_0 = const()[name = tensor("op_11486_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11486_cast_fp16 = slice_by_index(begin = var_11486_begin_0, end = var_11486_end_0, end_mask = var_11486_end_mask_0, x = k_111_cast_fp16)[name = tensor("op_11486_cast_fp16")]; + tensor var_11490_begin_0 = const()[name = tensor("op_11490_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_11490_end_0 = const()[name = tensor("op_11490_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_11490_end_mask_0 = const()[name = tensor("op_11490_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11490_cast_fp16 = slice_by_index(begin = var_11490_begin_0, end = var_11490_end_0, end_mask = var_11490_end_mask_0, x = k_111_cast_fp16)[name = tensor("op_11490_cast_fp16")]; + tensor var_11494_begin_0 = const()[name = tensor("op_11494_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_11494_end_0 = const()[name = tensor("op_11494_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_11494_end_mask_0 = const()[name = tensor("op_11494_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11494_cast_fp16 = slice_by_index(begin = var_11494_begin_0, end = var_11494_end_0, end_mask = var_11494_end_mask_0, x = k_111_cast_fp16)[name = tensor("op_11494_cast_fp16")]; + tensor var_11498_begin_0 = const()[name = tensor("op_11498_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_11498_end_0 = const()[name = tensor("op_11498_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_11498_end_mask_0 = const()[name = tensor("op_11498_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11498_cast_fp16 = slice_by_index(begin = var_11498_begin_0, end = var_11498_end_0, end_mask = var_11498_end_mask_0, x = k_111_cast_fp16)[name = tensor("op_11498_cast_fp16")]; + tensor var_11502_begin_0 = const()[name = tensor("op_11502_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_11502_end_0 = const()[name = tensor("op_11502_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_11502_end_mask_0 = const()[name = tensor("op_11502_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11502_cast_fp16 = slice_by_index(begin = var_11502_begin_0, end = var_11502_end_0, end_mask = var_11502_end_mask_0, x = k_111_cast_fp16)[name = tensor("op_11502_cast_fp16")]; + tensor var_11506_begin_0 = const()[name = tensor("op_11506_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_11506_end_0 = const()[name = tensor("op_11506_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_11506_end_mask_0 = const()[name = tensor("op_11506_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11506_cast_fp16 = slice_by_index(begin = var_11506_begin_0, end = var_11506_end_0, end_mask = var_11506_end_mask_0, x = k_111_cast_fp16)[name = tensor("op_11506_cast_fp16")]; + tensor var_11510_begin_0 = const()[name = tensor("op_11510_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_11510_end_0 = const()[name = tensor("op_11510_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_11510_end_mask_0 = const()[name = tensor("op_11510_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11510_cast_fp16 = slice_by_index(begin = var_11510_begin_0, end = var_11510_end_0, end_mask = var_11510_end_mask_0, x = k_111_cast_fp16)[name = tensor("op_11510_cast_fp16")]; + tensor var_11514_begin_0 = const()[name = tensor("op_11514_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_11514_end_0 = const()[name = tensor("op_11514_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_11514_end_mask_0 = const()[name = tensor("op_11514_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11514_cast_fp16 = slice_by_index(begin = var_11514_begin_0, end = var_11514_end_0, end_mask = var_11514_end_mask_0, x = k_111_cast_fp16)[name = tensor("op_11514_cast_fp16")]; + tensor var_11518_begin_0 = const()[name = tensor("op_11518_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_11518_end_0 = const()[name = tensor("op_11518_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_11518_end_mask_0 = const()[name = tensor("op_11518_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11518_cast_fp16 = slice_by_index(begin = var_11518_begin_0, end = var_11518_end_0, end_mask = var_11518_end_mask_0, x = k_111_cast_fp16)[name = tensor("op_11518_cast_fp16")]; + tensor var_11522_begin_0 = const()[name = tensor("op_11522_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_11522_end_0 = const()[name = tensor("op_11522_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_11522_end_mask_0 = const()[name = tensor("op_11522_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11522_cast_fp16 = slice_by_index(begin = var_11522_begin_0, end = var_11522_end_0, end_mask = var_11522_end_mask_0, x = k_111_cast_fp16)[name = tensor("op_11522_cast_fp16")]; + tensor var_11526_begin_0 = const()[name = tensor("op_11526_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_11526_end_0 = const()[name = tensor("op_11526_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_11526_end_mask_0 = const()[name = tensor("op_11526_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11526_cast_fp16 = slice_by_index(begin = var_11526_begin_0, end = var_11526_end_0, end_mask = var_11526_end_mask_0, x = k_111_cast_fp16)[name = tensor("op_11526_cast_fp16")]; + tensor var_11530_begin_0 = const()[name = tensor("op_11530_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_11530_end_0 = const()[name = tensor("op_11530_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_11530_end_mask_0 = const()[name = tensor("op_11530_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11530_cast_fp16 = slice_by_index(begin = var_11530_begin_0, end = var_11530_end_0, end_mask = var_11530_end_mask_0, x = k_111_cast_fp16)[name = tensor("op_11530_cast_fp16")]; + tensor var_11534_begin_0 = const()[name = tensor("op_11534_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_11534_end_0 = const()[name = tensor("op_11534_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_11534_end_mask_0 = const()[name = tensor("op_11534_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11534_cast_fp16 = slice_by_index(begin = var_11534_begin_0, end = var_11534_end_0, end_mask = var_11534_end_mask_0, x = k_111_cast_fp16)[name = tensor("op_11534_cast_fp16")]; + tensor var_11538_begin_0 = const()[name = tensor("op_11538_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_11538_end_0 = const()[name = tensor("op_11538_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_11538_end_mask_0 = const()[name = tensor("op_11538_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11538_cast_fp16 = slice_by_index(begin = var_11538_begin_0, end = var_11538_end_0, end_mask = var_11538_end_mask_0, x = k_111_cast_fp16)[name = tensor("op_11538_cast_fp16")]; + tensor var_11542_begin_0 = const()[name = tensor("op_11542_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_11542_end_0 = const()[name = tensor("op_11542_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_11542_end_mask_0 = const()[name = tensor("op_11542_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11542_cast_fp16 = slice_by_index(begin = var_11542_begin_0, end = var_11542_end_0, end_mask = var_11542_end_mask_0, x = k_111_cast_fp16)[name = tensor("op_11542_cast_fp16")]; + tensor var_11546_begin_0 = const()[name = tensor("op_11546_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_11546_end_0 = const()[name = tensor("op_11546_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_11546_end_mask_0 = const()[name = tensor("op_11546_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11546_cast_fp16 = slice_by_index(begin = var_11546_begin_0, end = var_11546_end_0, end_mask = var_11546_end_mask_0, x = k_111_cast_fp16)[name = tensor("op_11546_cast_fp16")]; + tensor var_11550_begin_0 = const()[name = tensor("op_11550_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_11550_end_0 = const()[name = tensor("op_11550_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_11550_end_mask_0 = const()[name = tensor("op_11550_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11550_cast_fp16 = slice_by_index(begin = var_11550_begin_0, end = var_11550_end_0, end_mask = var_11550_end_mask_0, x = k_111_cast_fp16)[name = tensor("op_11550_cast_fp16")]; + tensor var_11554_begin_0 = const()[name = tensor("op_11554_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_11554_end_0 = const()[name = tensor("op_11554_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_11554_end_mask_0 = const()[name = tensor("op_11554_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11554_cast_fp16 = slice_by_index(begin = var_11554_begin_0, end = var_11554_end_0, end_mask = var_11554_end_mask_0, x = k_111_cast_fp16)[name = tensor("op_11554_cast_fp16")]; + tensor var_11558_begin_0 = const()[name = tensor("op_11558_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_11558_end_0 = const()[name = tensor("op_11558_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_11558_end_mask_0 = const()[name = tensor("op_11558_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11558_cast_fp16 = slice_by_index(begin = var_11558_begin_0, end = var_11558_end_0, end_mask = var_11558_end_mask_0, x = k_111_cast_fp16)[name = tensor("op_11558_cast_fp16")]; + tensor var_11560_begin_0 = const()[name = tensor("op_11560_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11560_end_0 = const()[name = tensor("op_11560_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_11560_end_mask_0 = const()[name = tensor("op_11560_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11560_cast_fp16 = slice_by_index(begin = var_11560_begin_0, end = var_11560_end_0, end_mask = var_11560_end_mask_0, x = v_55_cast_fp16)[name = tensor("op_11560_cast_fp16")]; + tensor var_11564_begin_0 = const()[name = tensor("op_11564_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_11564_end_0 = const()[name = tensor("op_11564_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_11564_end_mask_0 = const()[name = tensor("op_11564_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11564_cast_fp16 = slice_by_index(begin = var_11564_begin_0, end = var_11564_end_0, end_mask = var_11564_end_mask_0, x = v_55_cast_fp16)[name = tensor("op_11564_cast_fp16")]; + tensor var_11568_begin_0 = const()[name = tensor("op_11568_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_11568_end_0 = const()[name = tensor("op_11568_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_11568_end_mask_0 = const()[name = tensor("op_11568_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11568_cast_fp16 = slice_by_index(begin = var_11568_begin_0, end = var_11568_end_0, end_mask = var_11568_end_mask_0, x = v_55_cast_fp16)[name = tensor("op_11568_cast_fp16")]; + tensor var_11572_begin_0 = const()[name = tensor("op_11572_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_11572_end_0 = const()[name = tensor("op_11572_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_11572_end_mask_0 = const()[name = tensor("op_11572_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11572_cast_fp16 = slice_by_index(begin = var_11572_begin_0, end = var_11572_end_0, end_mask = var_11572_end_mask_0, x = v_55_cast_fp16)[name = tensor("op_11572_cast_fp16")]; + tensor var_11576_begin_0 = const()[name = tensor("op_11576_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_11576_end_0 = const()[name = tensor("op_11576_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_11576_end_mask_0 = const()[name = tensor("op_11576_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11576_cast_fp16 = slice_by_index(begin = var_11576_begin_0, end = var_11576_end_0, end_mask = var_11576_end_mask_0, x = v_55_cast_fp16)[name = tensor("op_11576_cast_fp16")]; + tensor var_11580_begin_0 = const()[name = tensor("op_11580_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_11580_end_0 = const()[name = tensor("op_11580_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_11580_end_mask_0 = const()[name = tensor("op_11580_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11580_cast_fp16 = slice_by_index(begin = var_11580_begin_0, end = var_11580_end_0, end_mask = var_11580_end_mask_0, x = v_55_cast_fp16)[name = tensor("op_11580_cast_fp16")]; + tensor var_11584_begin_0 = const()[name = tensor("op_11584_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_11584_end_0 = const()[name = tensor("op_11584_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_11584_end_mask_0 = const()[name = tensor("op_11584_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11584_cast_fp16 = slice_by_index(begin = var_11584_begin_0, end = var_11584_end_0, end_mask = var_11584_end_mask_0, x = v_55_cast_fp16)[name = tensor("op_11584_cast_fp16")]; + tensor var_11588_begin_0 = const()[name = tensor("op_11588_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_11588_end_0 = const()[name = tensor("op_11588_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_11588_end_mask_0 = const()[name = tensor("op_11588_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11588_cast_fp16 = slice_by_index(begin = var_11588_begin_0, end = var_11588_end_0, end_mask = var_11588_end_mask_0, x = v_55_cast_fp16)[name = tensor("op_11588_cast_fp16")]; + tensor var_11592_begin_0 = const()[name = tensor("op_11592_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_11592_end_0 = const()[name = tensor("op_11592_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_11592_end_mask_0 = const()[name = tensor("op_11592_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11592_cast_fp16 = slice_by_index(begin = var_11592_begin_0, end = var_11592_end_0, end_mask = var_11592_end_mask_0, x = v_55_cast_fp16)[name = tensor("op_11592_cast_fp16")]; + tensor var_11596_begin_0 = const()[name = tensor("op_11596_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_11596_end_0 = const()[name = tensor("op_11596_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_11596_end_mask_0 = const()[name = tensor("op_11596_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11596_cast_fp16 = slice_by_index(begin = var_11596_begin_0, end = var_11596_end_0, end_mask = var_11596_end_mask_0, x = v_55_cast_fp16)[name = tensor("op_11596_cast_fp16")]; + tensor var_11600_begin_0 = const()[name = tensor("op_11600_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_11600_end_0 = const()[name = tensor("op_11600_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_11600_end_mask_0 = const()[name = tensor("op_11600_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11600_cast_fp16 = slice_by_index(begin = var_11600_begin_0, end = var_11600_end_0, end_mask = var_11600_end_mask_0, x = v_55_cast_fp16)[name = tensor("op_11600_cast_fp16")]; + tensor var_11604_begin_0 = const()[name = tensor("op_11604_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_11604_end_0 = const()[name = tensor("op_11604_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_11604_end_mask_0 = const()[name = tensor("op_11604_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11604_cast_fp16 = slice_by_index(begin = var_11604_begin_0, end = var_11604_end_0, end_mask = var_11604_end_mask_0, x = v_55_cast_fp16)[name = tensor("op_11604_cast_fp16")]; + tensor var_11608_begin_0 = const()[name = tensor("op_11608_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_11608_end_0 = const()[name = tensor("op_11608_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_11608_end_mask_0 = const()[name = tensor("op_11608_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11608_cast_fp16 = slice_by_index(begin = var_11608_begin_0, end = var_11608_end_0, end_mask = var_11608_end_mask_0, x = v_55_cast_fp16)[name = tensor("op_11608_cast_fp16")]; + tensor var_11612_begin_0 = const()[name = tensor("op_11612_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_11612_end_0 = const()[name = tensor("op_11612_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_11612_end_mask_0 = const()[name = tensor("op_11612_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11612_cast_fp16 = slice_by_index(begin = var_11612_begin_0, end = var_11612_end_0, end_mask = var_11612_end_mask_0, x = v_55_cast_fp16)[name = tensor("op_11612_cast_fp16")]; + tensor var_11616_begin_0 = const()[name = tensor("op_11616_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_11616_end_0 = const()[name = tensor("op_11616_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_11616_end_mask_0 = const()[name = tensor("op_11616_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11616_cast_fp16 = slice_by_index(begin = var_11616_begin_0, end = var_11616_end_0, end_mask = var_11616_end_mask_0, x = v_55_cast_fp16)[name = tensor("op_11616_cast_fp16")]; + tensor var_11620_begin_0 = const()[name = tensor("op_11620_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_11620_end_0 = const()[name = tensor("op_11620_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_11620_end_mask_0 = const()[name = tensor("op_11620_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11620_cast_fp16 = slice_by_index(begin = var_11620_begin_0, end = var_11620_end_0, end_mask = var_11620_end_mask_0, x = v_55_cast_fp16)[name = tensor("op_11620_cast_fp16")]; + tensor var_11624_begin_0 = const()[name = tensor("op_11624_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_11624_end_0 = const()[name = tensor("op_11624_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_11624_end_mask_0 = const()[name = tensor("op_11624_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11624_cast_fp16 = slice_by_index(begin = var_11624_begin_0, end = var_11624_end_0, end_mask = var_11624_end_mask_0, x = v_55_cast_fp16)[name = tensor("op_11624_cast_fp16")]; + tensor var_11628_begin_0 = const()[name = tensor("op_11628_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_11628_end_0 = const()[name = tensor("op_11628_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_11628_end_mask_0 = const()[name = tensor("op_11628_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11628_cast_fp16 = slice_by_index(begin = var_11628_begin_0, end = var_11628_end_0, end_mask = var_11628_end_mask_0, x = v_55_cast_fp16)[name = tensor("op_11628_cast_fp16")]; + tensor var_11632_begin_0 = const()[name = tensor("op_11632_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_11632_end_0 = const()[name = tensor("op_11632_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_11632_end_mask_0 = const()[name = tensor("op_11632_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11632_cast_fp16 = slice_by_index(begin = var_11632_begin_0, end = var_11632_end_0, end_mask = var_11632_end_mask_0, x = v_55_cast_fp16)[name = tensor("op_11632_cast_fp16")]; + tensor var_11636_begin_0 = const()[name = tensor("op_11636_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_11636_end_0 = const()[name = tensor("op_11636_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_11636_end_mask_0 = const()[name = tensor("op_11636_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11636_cast_fp16 = slice_by_index(begin = var_11636_begin_0, end = var_11636_end_0, end_mask = var_11636_end_mask_0, x = v_55_cast_fp16)[name = tensor("op_11636_cast_fp16")]; + tensor var_11640_equation_0 = const()[name = tensor("op_11640_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11640_cast_fp16 = einsum(equation = var_11640_equation_0, values = (var_11482_cast_fp16, var_11399_cast_fp16))[name = tensor("op_11640_cast_fp16")]; + tensor var_11641_to_fp16 = const()[name = tensor("op_11641_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_921_cast_fp16 = mul(x = var_11640_cast_fp16, y = var_11641_to_fp16)[name = tensor("aw_921_cast_fp16")]; + tensor var_11644_equation_0 = const()[name = tensor("op_11644_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11644_cast_fp16 = einsum(equation = var_11644_equation_0, values = (var_11486_cast_fp16, var_11403_cast_fp16))[name = tensor("op_11644_cast_fp16")]; + tensor var_11645_to_fp16 = const()[name = tensor("op_11645_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_923_cast_fp16 = mul(x = var_11644_cast_fp16, y = var_11645_to_fp16)[name = tensor("aw_923_cast_fp16")]; + tensor var_11648_equation_0 = const()[name = tensor("op_11648_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11648_cast_fp16 = einsum(equation = var_11648_equation_0, values = (var_11490_cast_fp16, var_11407_cast_fp16))[name = tensor("op_11648_cast_fp16")]; + tensor var_11649_to_fp16 = const()[name = tensor("op_11649_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_925_cast_fp16 = mul(x = var_11648_cast_fp16, y = var_11649_to_fp16)[name = tensor("aw_925_cast_fp16")]; + tensor var_11652_equation_0 = const()[name = tensor("op_11652_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11652_cast_fp16 = einsum(equation = var_11652_equation_0, values = (var_11494_cast_fp16, var_11411_cast_fp16))[name = tensor("op_11652_cast_fp16")]; + tensor var_11653_to_fp16 = const()[name = tensor("op_11653_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_927_cast_fp16 = mul(x = var_11652_cast_fp16, y = var_11653_to_fp16)[name = tensor("aw_927_cast_fp16")]; + tensor var_11656_equation_0 = const()[name = tensor("op_11656_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11656_cast_fp16 = einsum(equation = var_11656_equation_0, values = (var_11498_cast_fp16, var_11415_cast_fp16))[name = tensor("op_11656_cast_fp16")]; + tensor var_11657_to_fp16 = const()[name = tensor("op_11657_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_929_cast_fp16 = mul(x = var_11656_cast_fp16, y = var_11657_to_fp16)[name = tensor("aw_929_cast_fp16")]; + tensor var_11660_equation_0 = const()[name = tensor("op_11660_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11660_cast_fp16 = einsum(equation = var_11660_equation_0, values = (var_11502_cast_fp16, var_11419_cast_fp16))[name = tensor("op_11660_cast_fp16")]; + tensor var_11661_to_fp16 = const()[name = tensor("op_11661_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_931_cast_fp16 = mul(x = var_11660_cast_fp16, y = var_11661_to_fp16)[name = tensor("aw_931_cast_fp16")]; + tensor var_11664_equation_0 = const()[name = tensor("op_11664_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11664_cast_fp16 = einsum(equation = var_11664_equation_0, values = (var_11506_cast_fp16, var_11423_cast_fp16))[name = tensor("op_11664_cast_fp16")]; + tensor var_11665_to_fp16 = const()[name = tensor("op_11665_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_933_cast_fp16 = mul(x = var_11664_cast_fp16, y = var_11665_to_fp16)[name = tensor("aw_933_cast_fp16")]; + tensor var_11668_equation_0 = const()[name = tensor("op_11668_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11668_cast_fp16 = einsum(equation = var_11668_equation_0, values = (var_11510_cast_fp16, var_11427_cast_fp16))[name = tensor("op_11668_cast_fp16")]; + tensor var_11669_to_fp16 = const()[name = tensor("op_11669_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_935_cast_fp16 = mul(x = var_11668_cast_fp16, y = var_11669_to_fp16)[name = tensor("aw_935_cast_fp16")]; + tensor var_11672_equation_0 = const()[name = tensor("op_11672_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11672_cast_fp16 = einsum(equation = var_11672_equation_0, values = (var_11514_cast_fp16, var_11431_cast_fp16))[name = tensor("op_11672_cast_fp16")]; + tensor var_11673_to_fp16 = const()[name = tensor("op_11673_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_937_cast_fp16 = mul(x = var_11672_cast_fp16, y = var_11673_to_fp16)[name = tensor("aw_937_cast_fp16")]; + tensor var_11676_equation_0 = const()[name = tensor("op_11676_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11676_cast_fp16 = einsum(equation = var_11676_equation_0, values = (var_11518_cast_fp16, var_11435_cast_fp16))[name = tensor("op_11676_cast_fp16")]; + tensor var_11677_to_fp16 = const()[name = tensor("op_11677_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_939_cast_fp16 = mul(x = var_11676_cast_fp16, y = var_11677_to_fp16)[name = tensor("aw_939_cast_fp16")]; + tensor var_11680_equation_0 = const()[name = tensor("op_11680_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11680_cast_fp16 = einsum(equation = var_11680_equation_0, values = (var_11522_cast_fp16, var_11439_cast_fp16))[name = tensor("op_11680_cast_fp16")]; + tensor var_11681_to_fp16 = const()[name = tensor("op_11681_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_941_cast_fp16 = mul(x = var_11680_cast_fp16, y = var_11681_to_fp16)[name = tensor("aw_941_cast_fp16")]; + tensor var_11684_equation_0 = const()[name = tensor("op_11684_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11684_cast_fp16 = einsum(equation = var_11684_equation_0, values = (var_11526_cast_fp16, var_11443_cast_fp16))[name = tensor("op_11684_cast_fp16")]; + tensor var_11685_to_fp16 = const()[name = tensor("op_11685_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_943_cast_fp16 = mul(x = var_11684_cast_fp16, y = var_11685_to_fp16)[name = tensor("aw_943_cast_fp16")]; + tensor var_11688_equation_0 = const()[name = tensor("op_11688_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11688_cast_fp16 = einsum(equation = var_11688_equation_0, values = (var_11530_cast_fp16, var_11447_cast_fp16))[name = tensor("op_11688_cast_fp16")]; + tensor var_11689_to_fp16 = const()[name = tensor("op_11689_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_945_cast_fp16 = mul(x = var_11688_cast_fp16, y = var_11689_to_fp16)[name = tensor("aw_945_cast_fp16")]; + tensor var_11692_equation_0 = const()[name = tensor("op_11692_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11692_cast_fp16 = einsum(equation = var_11692_equation_0, values = (var_11534_cast_fp16, var_11451_cast_fp16))[name = tensor("op_11692_cast_fp16")]; + tensor var_11693_to_fp16 = const()[name = tensor("op_11693_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_947_cast_fp16 = mul(x = var_11692_cast_fp16, y = var_11693_to_fp16)[name = tensor("aw_947_cast_fp16")]; + tensor var_11696_equation_0 = const()[name = tensor("op_11696_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11696_cast_fp16 = einsum(equation = var_11696_equation_0, values = (var_11538_cast_fp16, var_11455_cast_fp16))[name = tensor("op_11696_cast_fp16")]; + tensor var_11697_to_fp16 = const()[name = tensor("op_11697_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_949_cast_fp16 = mul(x = var_11696_cast_fp16, y = var_11697_to_fp16)[name = tensor("aw_949_cast_fp16")]; + tensor var_11700_equation_0 = const()[name = tensor("op_11700_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11700_cast_fp16 = einsum(equation = var_11700_equation_0, values = (var_11542_cast_fp16, var_11459_cast_fp16))[name = tensor("op_11700_cast_fp16")]; + tensor var_11701_to_fp16 = const()[name = tensor("op_11701_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_951_cast_fp16 = mul(x = var_11700_cast_fp16, y = var_11701_to_fp16)[name = tensor("aw_951_cast_fp16")]; + tensor var_11704_equation_0 = const()[name = tensor("op_11704_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11704_cast_fp16 = einsum(equation = var_11704_equation_0, values = (var_11546_cast_fp16, var_11463_cast_fp16))[name = tensor("op_11704_cast_fp16")]; + tensor var_11705_to_fp16 = const()[name = tensor("op_11705_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_953_cast_fp16 = mul(x = var_11704_cast_fp16, y = var_11705_to_fp16)[name = tensor("aw_953_cast_fp16")]; + tensor var_11708_equation_0 = const()[name = tensor("op_11708_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11708_cast_fp16 = einsum(equation = var_11708_equation_0, values = (var_11550_cast_fp16, var_11467_cast_fp16))[name = tensor("op_11708_cast_fp16")]; + tensor var_11709_to_fp16 = const()[name = tensor("op_11709_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_955_cast_fp16 = mul(x = var_11708_cast_fp16, y = var_11709_to_fp16)[name = tensor("aw_955_cast_fp16")]; + tensor var_11712_equation_0 = const()[name = tensor("op_11712_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11712_cast_fp16 = einsum(equation = var_11712_equation_0, values = (var_11554_cast_fp16, var_11471_cast_fp16))[name = tensor("op_11712_cast_fp16")]; + tensor var_11713_to_fp16 = const()[name = tensor("op_11713_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_957_cast_fp16 = mul(x = var_11712_cast_fp16, y = var_11713_to_fp16)[name = tensor("aw_957_cast_fp16")]; + tensor var_11716_equation_0 = const()[name = tensor("op_11716_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11716_cast_fp16 = einsum(equation = var_11716_equation_0, values = (var_11558_cast_fp16, var_11475_cast_fp16))[name = tensor("op_11716_cast_fp16")]; + tensor var_11717_to_fp16 = const()[name = tensor("op_11717_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_959_cast_fp16 = mul(x = var_11716_cast_fp16, y = var_11717_to_fp16)[name = tensor("aw_959_cast_fp16")]; + tensor var_11719_cast_fp16 = softmax(axis = var_2624, x = aw_921_cast_fp16)[name = tensor("op_11719_cast_fp16")]; + tensor var_11720_cast_fp16 = softmax(axis = var_2624, x = aw_923_cast_fp16)[name = tensor("op_11720_cast_fp16")]; + tensor var_11721_cast_fp16 = softmax(axis = var_2624, x = aw_925_cast_fp16)[name = tensor("op_11721_cast_fp16")]; + tensor var_11722_cast_fp16 = softmax(axis = var_2624, x = aw_927_cast_fp16)[name = tensor("op_11722_cast_fp16")]; + tensor var_11723_cast_fp16 = softmax(axis = var_2624, x = aw_929_cast_fp16)[name = tensor("op_11723_cast_fp16")]; + tensor var_11724_cast_fp16 = softmax(axis = var_2624, x = aw_931_cast_fp16)[name = tensor("op_11724_cast_fp16")]; + tensor var_11725_cast_fp16 = softmax(axis = var_2624, x = aw_933_cast_fp16)[name = tensor("op_11725_cast_fp16")]; + tensor var_11726_cast_fp16 = softmax(axis = var_2624, x = aw_935_cast_fp16)[name = tensor("op_11726_cast_fp16")]; + tensor var_11727_cast_fp16 = softmax(axis = var_2624, x = aw_937_cast_fp16)[name = tensor("op_11727_cast_fp16")]; + tensor var_11728_cast_fp16 = softmax(axis = var_2624, x = aw_939_cast_fp16)[name = tensor("op_11728_cast_fp16")]; + tensor var_11729_cast_fp16 = softmax(axis = var_2624, x = aw_941_cast_fp16)[name = tensor("op_11729_cast_fp16")]; + tensor var_11730_cast_fp16 = softmax(axis = var_2624, x = aw_943_cast_fp16)[name = tensor("op_11730_cast_fp16")]; + tensor var_11731_cast_fp16 = softmax(axis = var_2624, x = aw_945_cast_fp16)[name = tensor("op_11731_cast_fp16")]; + tensor var_11732_cast_fp16 = softmax(axis = var_2624, x = aw_947_cast_fp16)[name = tensor("op_11732_cast_fp16")]; + tensor var_11733_cast_fp16 = softmax(axis = var_2624, x = aw_949_cast_fp16)[name = tensor("op_11733_cast_fp16")]; + tensor var_11734_cast_fp16 = softmax(axis = var_2624, x = aw_951_cast_fp16)[name = tensor("op_11734_cast_fp16")]; + tensor var_11735_cast_fp16 = softmax(axis = var_2624, x = aw_953_cast_fp16)[name = tensor("op_11735_cast_fp16")]; + tensor var_11736_cast_fp16 = softmax(axis = var_2624, x = aw_955_cast_fp16)[name = tensor("op_11736_cast_fp16")]; + tensor var_11737_cast_fp16 = softmax(axis = var_2624, x = aw_957_cast_fp16)[name = tensor("op_11737_cast_fp16")]; + tensor var_11738_cast_fp16 = softmax(axis = var_2624, x = aw_959_cast_fp16)[name = tensor("op_11738_cast_fp16")]; + tensor var_11740_equation_0 = const()[name = tensor("op_11740_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11740_cast_fp16 = einsum(equation = var_11740_equation_0, values = (var_11560_cast_fp16, var_11719_cast_fp16))[name = tensor("op_11740_cast_fp16")]; + tensor var_11742_equation_0 = const()[name = tensor("op_11742_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11742_cast_fp16 = einsum(equation = var_11742_equation_0, values = (var_11564_cast_fp16, var_11720_cast_fp16))[name = tensor("op_11742_cast_fp16")]; + tensor var_11744_equation_0 = const()[name = tensor("op_11744_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11744_cast_fp16 = einsum(equation = var_11744_equation_0, values = (var_11568_cast_fp16, var_11721_cast_fp16))[name = tensor("op_11744_cast_fp16")]; + tensor var_11746_equation_0 = const()[name = tensor("op_11746_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11746_cast_fp16 = einsum(equation = var_11746_equation_0, values = (var_11572_cast_fp16, var_11722_cast_fp16))[name = tensor("op_11746_cast_fp16")]; + tensor var_11748_equation_0 = const()[name = tensor("op_11748_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11748_cast_fp16 = einsum(equation = var_11748_equation_0, values = (var_11576_cast_fp16, var_11723_cast_fp16))[name = tensor("op_11748_cast_fp16")]; + tensor var_11750_equation_0 = const()[name = tensor("op_11750_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11750_cast_fp16 = einsum(equation = var_11750_equation_0, values = (var_11580_cast_fp16, var_11724_cast_fp16))[name = tensor("op_11750_cast_fp16")]; + tensor var_11752_equation_0 = const()[name = tensor("op_11752_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11752_cast_fp16 = einsum(equation = var_11752_equation_0, values = (var_11584_cast_fp16, var_11725_cast_fp16))[name = tensor("op_11752_cast_fp16")]; + tensor var_11754_equation_0 = const()[name = tensor("op_11754_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11754_cast_fp16 = einsum(equation = var_11754_equation_0, values = (var_11588_cast_fp16, var_11726_cast_fp16))[name = tensor("op_11754_cast_fp16")]; + tensor var_11756_equation_0 = const()[name = tensor("op_11756_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11756_cast_fp16 = einsum(equation = var_11756_equation_0, values = (var_11592_cast_fp16, var_11727_cast_fp16))[name = tensor("op_11756_cast_fp16")]; + tensor var_11758_equation_0 = const()[name = tensor("op_11758_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11758_cast_fp16 = einsum(equation = var_11758_equation_0, values = (var_11596_cast_fp16, var_11728_cast_fp16))[name = tensor("op_11758_cast_fp16")]; + tensor var_11760_equation_0 = const()[name = tensor("op_11760_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11760_cast_fp16 = einsum(equation = var_11760_equation_0, values = (var_11600_cast_fp16, var_11729_cast_fp16))[name = tensor("op_11760_cast_fp16")]; + tensor var_11762_equation_0 = const()[name = tensor("op_11762_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11762_cast_fp16 = einsum(equation = var_11762_equation_0, values = (var_11604_cast_fp16, var_11730_cast_fp16))[name = tensor("op_11762_cast_fp16")]; + tensor var_11764_equation_0 = const()[name = tensor("op_11764_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11764_cast_fp16 = einsum(equation = var_11764_equation_0, values = (var_11608_cast_fp16, var_11731_cast_fp16))[name = tensor("op_11764_cast_fp16")]; + tensor var_11766_equation_0 = const()[name = tensor("op_11766_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11766_cast_fp16 = einsum(equation = var_11766_equation_0, values = (var_11612_cast_fp16, var_11732_cast_fp16))[name = tensor("op_11766_cast_fp16")]; + tensor var_11768_equation_0 = const()[name = tensor("op_11768_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11768_cast_fp16 = einsum(equation = var_11768_equation_0, values = (var_11616_cast_fp16, var_11733_cast_fp16))[name = tensor("op_11768_cast_fp16")]; + tensor var_11770_equation_0 = const()[name = tensor("op_11770_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11770_cast_fp16 = einsum(equation = var_11770_equation_0, values = (var_11620_cast_fp16, var_11734_cast_fp16))[name = tensor("op_11770_cast_fp16")]; + tensor var_11772_equation_0 = const()[name = tensor("op_11772_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11772_cast_fp16 = einsum(equation = var_11772_equation_0, values = (var_11624_cast_fp16, var_11735_cast_fp16))[name = tensor("op_11772_cast_fp16")]; + tensor var_11774_equation_0 = const()[name = tensor("op_11774_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11774_cast_fp16 = einsum(equation = var_11774_equation_0, values = (var_11628_cast_fp16, var_11736_cast_fp16))[name = tensor("op_11774_cast_fp16")]; + tensor var_11776_equation_0 = const()[name = tensor("op_11776_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11776_cast_fp16 = einsum(equation = var_11776_equation_0, values = (var_11632_cast_fp16, var_11737_cast_fp16))[name = tensor("op_11776_cast_fp16")]; + tensor var_11778_equation_0 = const()[name = tensor("op_11778_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11778_cast_fp16 = einsum(equation = var_11778_equation_0, values = (var_11636_cast_fp16, var_11738_cast_fp16))[name = tensor("op_11778_cast_fp16")]; + tensor input_205_interleave_0 = const()[name = tensor("input_205_interleave_0"), val = tensor(false)]; + tensor input_205_cast_fp16 = concat(axis = var_2624, interleave = input_205_interleave_0, values = (var_11740_cast_fp16, var_11742_cast_fp16, var_11744_cast_fp16, var_11746_cast_fp16, var_11748_cast_fp16, var_11750_cast_fp16, var_11752_cast_fp16, var_11754_cast_fp16, var_11756_cast_fp16, var_11758_cast_fp16, var_11760_cast_fp16, var_11762_cast_fp16, var_11764_cast_fp16, var_11766_cast_fp16, var_11768_cast_fp16, var_11770_cast_fp16, var_11772_cast_fp16, var_11774_cast_fp16, var_11776_cast_fp16, var_11778_cast_fp16))[name = tensor("input_205_cast_fp16")]; + tensor var_11784 = const()[name = tensor("op_11784"), val = tensor([1, 1])]; + tensor var_11786 = const()[name = tensor("op_11786"), val = tensor([1, 1])]; + tensor var_11788_pad_type_0 = const()[name = tensor("op_11788_pad_type_0"), val = tensor("custom")]; + tensor var_11788_pad_0 = const()[name = tensor("op_11788_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_9_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_9_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(850665856)))]; + tensor down_blocks_2_attentions_0_transformer_blocks_9_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_9_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(853942720)))]; + tensor var_11788_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_9_attn2_to_out_0_bias_to_fp16, dilations = var_11786, groups = var_2624, pad = var_11788_pad_0, pad_type = var_11788_pad_type_0, strides = var_11784, weight = down_blocks_2_attentions_0_transformer_blocks_9_attn2_to_out_0_weight_to_fp16, x = input_205_cast_fp16)[name = tensor("op_11788_cast_fp16")]; + tensor inputs_83_cast_fp16 = add(x = var_11788_cast_fp16, y = inputs_81_cast_fp16)[name = tensor("inputs_83_cast_fp16")]; + tensor input_207_axes_0 = const()[name = tensor("input_207_axes_0"), val = tensor([1])]; + tensor input_207_gamma_0_to_fp16 = const()[name = tensor("input_207_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(853945344)))]; + tensor input_207_beta_0_to_fp16 = const()[name = tensor("input_207_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(853947968)))]; + tensor var_11798_to_fp16 = const()[name = tensor("op_11798_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_207_cast_fp16 = layer_norm(axes = input_207_axes_0, beta = input_207_beta_0_to_fp16, epsilon = var_11798_to_fp16, gamma = input_207_gamma_0_to_fp16, x = inputs_83_cast_fp16)[name = tensor("input_207_cast_fp16")]; + tensor var_11814 = const()[name = tensor("op_11814"), val = tensor([1, 1])]; + tensor var_11816 = const()[name = tensor("op_11816"), val = tensor([1, 1])]; + tensor var_11818_pad_type_0 = const()[name = tensor("op_11818_pad_type_0"), val = tensor("custom")]; + tensor var_11818_pad_0 = const()[name = tensor("op_11818_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_9_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_9_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(853950592)))]; + tensor down_blocks_2_attentions_0_transformer_blocks_9_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_9_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(880165056)))]; + tensor var_11818_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_9_ff_net_0_proj_bias_to_fp16, dilations = var_11816, groups = var_2624, pad = var_11818_pad_0, pad_type = var_11818_pad_type_0, strides = var_11814, weight = down_blocks_2_attentions_0_transformer_blocks_9_ff_net_0_proj_weight_to_fp16, x = input_207_cast_fp16)[name = tensor("op_11818_cast_fp16")]; + tensor var_11819_split_sizes_0 = const()[name = tensor("op_11819_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_11819_axis_0 = const()[name = tensor("op_11819_axis_0"), val = tensor(1)]; + tensor var_11819_cast_fp16_0, tensor var_11819_cast_fp16_1 = split(axis = var_11819_axis_0, split_sizes = var_11819_split_sizes_0, x = var_11818_cast_fp16)[name = tensor("op_11819_cast_fp16")]; + tensor var_11821_mode_0 = const()[name = tensor("op_11821_mode_0"), val = tensor("EXACT")]; + tensor var_11821_cast_fp16 = gelu(mode = var_11821_mode_0, x = var_11819_cast_fp16_1)[name = tensor("op_11821_cast_fp16")]; + tensor input_209_cast_fp16 = mul(x = var_11819_cast_fp16_0, y = var_11821_cast_fp16)[name = tensor("input_209_cast_fp16")]; + tensor var_11825 = const()[name = tensor("op_11825"), val = tensor([1, 1])]; + tensor var_11827 = const()[name = tensor("op_11827"), val = tensor([1, 1])]; + tensor var_11829_pad_type_0 = const()[name = tensor("op_11829_pad_type_0"), val = tensor("custom")]; + tensor var_11829_pad_0 = const()[name = tensor("op_11829_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_9_ff_net_2_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_9_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(880185600)))]; + tensor down_blocks_2_attentions_0_transformer_blocks_9_ff_net_2_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_9_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(893292864)))]; + tensor var_11829_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_9_ff_net_2_bias_to_fp16, dilations = var_11827, groups = var_2624, pad = var_11829_pad_0, pad_type = var_11829_pad_type_0, strides = var_11825, weight = down_blocks_2_attentions_0_transformer_blocks_9_ff_net_2_weight_to_fp16, x = input_209_cast_fp16)[name = tensor("op_11829_cast_fp16")]; + tensor hidden_states_125_cast_fp16 = add(x = var_11829_cast_fp16, y = inputs_83_cast_fp16)[name = tensor("hidden_states_125_cast_fp16")]; + tensor var_11831 = const()[name = tensor("op_11831"), val = tensor([2, 1280, 32, 32])]; + tensor input_211_cast_fp16 = reshape(shape = var_11831, x = hidden_states_125_cast_fp16)[name = tensor("input_211_cast_fp16")]; + tensor var_11835 = const()[name = tensor("op_11835"), val = tensor([1, 1])]; + tensor var_11837 = const()[name = tensor("op_11837"), val = tensor([1, 1])]; + tensor hidden_states_127_pad_type_0 = const()[name = tensor("hidden_states_127_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_127_pad_0 = const()[name = tensor("hidden_states_127_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_proj_out_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_proj_out_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(893295488)))]; + tensor down_blocks_2_attentions_0_proj_out_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_proj_out_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(896572352)))]; + tensor hidden_states_127_cast_fp16 = conv(bias = down_blocks_2_attentions_0_proj_out_bias_to_fp16, dilations = var_11837, groups = var_2624, pad = hidden_states_127_pad_0, pad_type = hidden_states_127_pad_type_0, strides = var_11835, weight = down_blocks_2_attentions_0_proj_out_weight_to_fp16, x = input_211_cast_fp16)[name = tensor("hidden_states_127_cast_fp16")]; + tensor input_213_cast_fp16 = add(x = hidden_states_127_cast_fp16, y = hidden_states_61_cast_fp16)[name = tensor("input_213_cast_fp16")]; + tensor reshape_52_shape_0 = const()[name = tensor("reshape_52_shape_0"), val = tensor([2, 32, 40, 32, 32])]; + tensor reshape_52_cast_fp16 = reshape(shape = reshape_52_shape_0, x = input_213_cast_fp16)[name = tensor("reshape_52_cast_fp16")]; + tensor reduce_mean_39_axes_0 = const()[name = tensor("reduce_mean_39_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_39_keep_dims_0 = const()[name = tensor("reduce_mean_39_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_39_cast_fp16 = reduce_mean(axes = reduce_mean_39_axes_0, keep_dims = reduce_mean_39_keep_dims_0, x = reshape_52_cast_fp16)[name = tensor("reduce_mean_39_cast_fp16")]; + tensor sub_26_cast_fp16 = sub(x = reshape_52_cast_fp16, y = reduce_mean_39_cast_fp16)[name = tensor("sub_26_cast_fp16")]; + tensor square_13_cast_fp16 = square(x = sub_26_cast_fp16)[name = tensor("square_13_cast_fp16")]; + tensor reduce_mean_41_axes_0 = const()[name = tensor("reduce_mean_41_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_41_keep_dims_0 = const()[name = tensor("reduce_mean_41_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_41_cast_fp16 = reduce_mean(axes = reduce_mean_41_axes_0, keep_dims = reduce_mean_41_keep_dims_0, x = square_13_cast_fp16)[name = tensor("reduce_mean_41_cast_fp16")]; + tensor add_26_y_0_to_fp16 = const()[name = tensor("add_26_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_26_cast_fp16 = add(x = reduce_mean_41_cast_fp16, y = add_26_y_0_to_fp16)[name = tensor("add_26_cast_fp16")]; + tensor sqrt_13_cast_fp16 = sqrt(x = add_26_cast_fp16)[name = tensor("sqrt_13_cast_fp16")]; + tensor real_div_13_cast_fp16 = real_div(x = sub_26_cast_fp16, y = sqrt_13_cast_fp16)[name = tensor("real_div_13_cast_fp16")]; + tensor reshape_53_shape_0 = const()[name = tensor("reshape_53_shape_0"), val = tensor([2, 1280, 32, 32])]; + tensor reshape_53_cast_fp16 = reshape(shape = reshape_53_shape_0, x = real_div_13_cast_fp16)[name = tensor("reshape_53_cast_fp16")]; + tensor add_27_gamma_0_to_fp16 = const()[name = tensor("add_27_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(896574976)))]; + tensor add_27_beta_0_to_fp16 = const()[name = tensor("add_27_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(896577600)))]; + tensor add_27_epsilon_0_to_fp16 = const()[name = tensor("add_27_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_27_cast_fp16 = batch_norm(beta = add_27_beta_0_to_fp16, epsilon = add_27_epsilon_0_to_fp16, gamma = add_27_gamma_0_to_fp16, mean = add_23_mean_0_to_fp16, variance = add_23_variance_0_to_fp16, x = reshape_53_cast_fp16)[name = tensor("add_27_cast_fp16")]; + tensor input_217_cast_fp16 = silu(x = add_27_cast_fp16)[name = tensor("input_217_cast_fp16")]; + tensor var_11852 = const()[name = tensor("op_11852"), val = tensor([1, 1])]; + tensor var_11854 = const()[name = tensor("op_11854"), val = tensor([1, 1])]; + tensor hidden_states_129_pad_type_0 = const()[name = tensor("hidden_states_129_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_129_pad_0 = const()[name = tensor("hidden_states_129_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor down_blocks_2_resnets_1_conv1_weight_to_fp16 = const()[name = tensor("down_blocks_2_resnets_1_conv1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(896580224)))]; + tensor down_blocks_2_resnets_1_conv1_bias_to_fp16 = const()[name = tensor("down_blocks_2_resnets_1_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(926071488)))]; + tensor hidden_states_129_cast_fp16 = conv(bias = down_blocks_2_resnets_1_conv1_bias_to_fp16, dilations = var_11854, groups = var_2624, pad = hidden_states_129_pad_0, pad_type = hidden_states_129_pad_type_0, strides = var_11852, weight = down_blocks_2_resnets_1_conv1_weight_to_fp16, x = input_217_cast_fp16)[name = tensor("hidden_states_129_cast_fp16")]; + tensor var_11860 = const()[name = tensor("op_11860"), val = tensor([1, 1])]; + tensor var_11862 = const()[name = tensor("op_11862"), val = tensor([1, 1])]; + tensor temb_11_pad_type_0 = const()[name = tensor("temb_11_pad_type_0"), val = tensor("custom")]; + tensor temb_11_pad_0 = const()[name = tensor("temb_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_resnets_1_time_emb_proj_weight_to_fp16 = const()[name = tensor("down_blocks_2_resnets_1_time_emb_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(926074112)))]; + tensor down_blocks_2_resnets_1_time_emb_proj_bias_to_fp16 = const()[name = tensor("down_blocks_2_resnets_1_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(929350976)))]; + tensor temb_11_cast_fp16 = conv(bias = down_blocks_2_resnets_1_time_emb_proj_bias_to_fp16, dilations = var_11862, groups = var_2624, pad = temb_11_pad_0, pad_type = temb_11_pad_type_0, strides = var_11860, weight = down_blocks_2_resnets_1_time_emb_proj_weight_to_fp16, x = input_21_cast_fp16)[name = tensor("temb_11_cast_fp16")]; + tensor input_221_cast_fp16 = add(x = hidden_states_129_cast_fp16, y = temb_11_cast_fp16)[name = tensor("input_221_cast_fp16")]; + tensor reshape_56_shape_0 = const()[name = tensor("reshape_56_shape_0"), val = tensor([2, 32, 40, 32, 32])]; + tensor reshape_56_cast_fp16 = reshape(shape = reshape_56_shape_0, x = input_221_cast_fp16)[name = tensor("reshape_56_cast_fp16")]; + tensor reduce_mean_42_axes_0 = const()[name = tensor("reduce_mean_42_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_42_keep_dims_0 = const()[name = tensor("reduce_mean_42_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_42_cast_fp16 = reduce_mean(axes = reduce_mean_42_axes_0, keep_dims = reduce_mean_42_keep_dims_0, x = reshape_56_cast_fp16)[name = tensor("reduce_mean_42_cast_fp16")]; + tensor sub_28_cast_fp16 = sub(x = reshape_56_cast_fp16, y = reduce_mean_42_cast_fp16)[name = tensor("sub_28_cast_fp16")]; + tensor square_14_cast_fp16 = square(x = sub_28_cast_fp16)[name = tensor("square_14_cast_fp16")]; + tensor reduce_mean_44_axes_0 = const()[name = tensor("reduce_mean_44_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_44_keep_dims_0 = const()[name = tensor("reduce_mean_44_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_44_cast_fp16 = reduce_mean(axes = reduce_mean_44_axes_0, keep_dims = reduce_mean_44_keep_dims_0, x = square_14_cast_fp16)[name = tensor("reduce_mean_44_cast_fp16")]; + tensor add_28_y_0_to_fp16 = const()[name = tensor("add_28_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_28_cast_fp16 = add(x = reduce_mean_44_cast_fp16, y = add_28_y_0_to_fp16)[name = tensor("add_28_cast_fp16")]; + tensor sqrt_14_cast_fp16 = sqrt(x = add_28_cast_fp16)[name = tensor("sqrt_14_cast_fp16")]; + tensor real_div_14_cast_fp16 = real_div(x = sub_28_cast_fp16, y = sqrt_14_cast_fp16)[name = tensor("real_div_14_cast_fp16")]; + tensor reshape_57_shape_0 = const()[name = tensor("reshape_57_shape_0"), val = tensor([2, 1280, 32, 32])]; + tensor reshape_57_cast_fp16 = reshape(shape = reshape_57_shape_0, x = real_div_14_cast_fp16)[name = tensor("reshape_57_cast_fp16")]; + tensor add_29_gamma_0_to_fp16 = const()[name = tensor("add_29_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(929353600)))]; + tensor add_29_beta_0_to_fp16 = const()[name = tensor("add_29_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(929356224)))]; + tensor add_29_epsilon_0_to_fp16 = const()[name = tensor("add_29_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_29_cast_fp16 = batch_norm(beta = add_29_beta_0_to_fp16, epsilon = add_29_epsilon_0_to_fp16, gamma = add_29_gamma_0_to_fp16, mean = add_23_mean_0_to_fp16, variance = add_23_variance_0_to_fp16, x = reshape_57_cast_fp16)[name = tensor("add_29_cast_fp16")]; + tensor input_225_cast_fp16 = silu(x = add_29_cast_fp16)[name = tensor("input_225_cast_fp16")]; + tensor var_11872 = const()[name = tensor("op_11872"), val = tensor([1, 1])]; + tensor var_11874 = const()[name = tensor("op_11874"), val = tensor([1, 1])]; + tensor hidden_states_131_pad_type_0 = const()[name = tensor("hidden_states_131_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_131_pad_0 = const()[name = tensor("hidden_states_131_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor down_blocks_2_resnets_1_conv2_weight_to_fp16 = const()[name = tensor("down_blocks_2_resnets_1_conv2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(929358848)))]; + tensor down_blocks_2_resnets_1_conv2_bias_to_fp16 = const()[name = tensor("down_blocks_2_resnets_1_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(958850112)))]; + tensor hidden_states_131_cast_fp16 = conv(bias = down_blocks_2_resnets_1_conv2_bias_to_fp16, dilations = var_11874, groups = var_2624, pad = hidden_states_131_pad_0, pad_type = hidden_states_131_pad_type_0, strides = var_11872, weight = down_blocks_2_resnets_1_conv2_weight_to_fp16, x = input_225_cast_fp16)[name = tensor("hidden_states_131_cast_fp16")]; + tensor hidden_states_133_cast_fp16 = add(x = input_213_cast_fp16, y = hidden_states_131_cast_fp16)[name = tensor("hidden_states_133_cast_fp16")]; + tensor reshape_60_shape_0 = const()[name = tensor("reshape_60_shape_0"), val = tensor([2, 32, 40, 32, 32])]; + tensor reshape_60_cast_fp16 = reshape(shape = reshape_60_shape_0, x = hidden_states_133_cast_fp16)[name = tensor("reshape_60_cast_fp16")]; + tensor reduce_mean_45_axes_0 = const()[name = tensor("reduce_mean_45_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_45_keep_dims_0 = const()[name = tensor("reduce_mean_45_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_45_cast_fp16 = reduce_mean(axes = reduce_mean_45_axes_0, keep_dims = reduce_mean_45_keep_dims_0, x = reshape_60_cast_fp16)[name = tensor("reduce_mean_45_cast_fp16")]; + tensor sub_30_cast_fp16 = sub(x = reshape_60_cast_fp16, y = reduce_mean_45_cast_fp16)[name = tensor("sub_30_cast_fp16")]; + tensor square_15_cast_fp16 = square(x = sub_30_cast_fp16)[name = tensor("square_15_cast_fp16")]; + tensor reduce_mean_47_axes_0 = const()[name = tensor("reduce_mean_47_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_47_keep_dims_0 = const()[name = tensor("reduce_mean_47_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_47_cast_fp16 = reduce_mean(axes = reduce_mean_47_axes_0, keep_dims = reduce_mean_47_keep_dims_0, x = square_15_cast_fp16)[name = tensor("reduce_mean_47_cast_fp16")]; + tensor add_30_y_0_to_fp16 = const()[name = tensor("add_30_y_0_to_fp16"), val = tensor(0x1.1p-20)]; + tensor add_30_cast_fp16 = add(x = reduce_mean_47_cast_fp16, y = add_30_y_0_to_fp16)[name = tensor("add_30_cast_fp16")]; + tensor sqrt_15_cast_fp16 = sqrt(x = add_30_cast_fp16)[name = tensor("sqrt_15_cast_fp16")]; + tensor real_div_15_cast_fp16 = real_div(x = sub_30_cast_fp16, y = sqrt_15_cast_fp16)[name = tensor("real_div_15_cast_fp16")]; + tensor reshape_61_shape_0 = const()[name = tensor("reshape_61_shape_0"), val = tensor([2, 1280, 32, 32])]; + tensor reshape_61_cast_fp16 = reshape(shape = reshape_61_shape_0, x = real_div_15_cast_fp16)[name = tensor("reshape_61_cast_fp16")]; + tensor add_31_gamma_0_to_fp16 = const()[name = tensor("add_31_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(958852736)))]; + tensor add_31_beta_0_to_fp16 = const()[name = tensor("add_31_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(958855360)))]; + tensor add_31_epsilon_0_to_fp16 = const()[name = tensor("add_31_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_31_cast_fp16 = batch_norm(beta = add_31_beta_0_to_fp16, epsilon = add_31_epsilon_0_to_fp16, gamma = add_31_gamma_0_to_fp16, mean = add_23_mean_0_to_fp16, variance = add_23_variance_0_to_fp16, x = reshape_61_cast_fp16)[name = tensor("add_31_cast_fp16")]; + tensor var_11912 = const()[name = tensor("op_11912"), val = tensor([1, 1])]; + tensor var_11914 = const()[name = tensor("op_11914"), val = tensor([1, 1])]; + tensor hidden_states_135_pad_type_0 = const()[name = tensor("hidden_states_135_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_135_pad_0 = const()[name = tensor("hidden_states_135_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_proj_in_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_proj_in_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(958857984)))]; + tensor down_blocks_2_attentions_1_proj_in_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_proj_in_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(962134848)))]; + tensor hidden_states_135_cast_fp16 = conv(bias = down_blocks_2_attentions_1_proj_in_bias_to_fp16, dilations = var_11914, groups = var_2624, pad = hidden_states_135_pad_0, pad_type = hidden_states_135_pad_type_0, strides = var_11912, weight = down_blocks_2_attentions_1_proj_in_weight_to_fp16, x = add_31_cast_fp16)[name = tensor("hidden_states_135_cast_fp16")]; + tensor var_11919 = const()[name = tensor("op_11919"), val = tensor([2, 1280, 1, 1024])]; + tensor inputs_85_cast_fp16 = reshape(shape = var_11919, x = hidden_states_135_cast_fp16)[name = tensor("inputs_85_cast_fp16")]; + tensor hidden_states_137_axes_0 = const()[name = tensor("hidden_states_137_axes_0"), val = tensor([1])]; + tensor hidden_states_137_gamma_0_to_fp16 = const()[name = tensor("hidden_states_137_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(962137472)))]; + tensor hidden_states_137_beta_0_to_fp16 = const()[name = tensor("hidden_states_137_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(962140096)))]; + tensor var_11935_to_fp16 = const()[name = tensor("op_11935_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_137_cast_fp16 = layer_norm(axes = hidden_states_137_axes_0, beta = hidden_states_137_beta_0_to_fp16, epsilon = var_11935_to_fp16, gamma = hidden_states_137_gamma_0_to_fp16, x = inputs_85_cast_fp16)[name = tensor("hidden_states_137_cast_fp16")]; + tensor var_11950 = const()[name = tensor("op_11950"), val = tensor([1, 1])]; + tensor var_11952 = const()[name = tensor("op_11952"), val = tensor([1, 1])]; + tensor q_57_pad_type_0 = const()[name = tensor("q_57_pad_type_0"), val = tensor("custom")]; + tensor q_57_pad_0 = const()[name = tensor("q_57_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(962142720)))]; + tensor q_57_cast_fp16 = conv(dilations = var_11952, groups = var_2624, pad = q_57_pad_0, pad_type = q_57_pad_type_0, strides = var_11950, weight = down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16, x = hidden_states_137_cast_fp16)[name = tensor("q_57_cast_fp16")]; + tensor var_11956 = const()[name = tensor("op_11956"), val = tensor([1, 1])]; + tensor var_11958 = const()[name = tensor("op_11958"), val = tensor([1, 1])]; + tensor k_113_pad_type_0 = const()[name = tensor("k_113_pad_type_0"), val = tensor("custom")]; + tensor k_113_pad_0 = const()[name = tensor("k_113_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(965419584)))]; + tensor k_113_cast_fp16 = conv(dilations = var_11958, groups = var_2624, pad = k_113_pad_0, pad_type = k_113_pad_type_0, strides = var_11956, weight = down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16, x = hidden_states_137_cast_fp16)[name = tensor("k_113_cast_fp16")]; + tensor var_11962 = const()[name = tensor("op_11962"), val = tensor([1, 1])]; + tensor var_11964 = const()[name = tensor("op_11964"), val = tensor([1, 1])]; + tensor v_57_pad_type_0 = const()[name = tensor("v_57_pad_type_0"), val = tensor("custom")]; + tensor v_57_pad_0 = const()[name = tensor("v_57_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(968696448)))]; + tensor v_57_cast_fp16 = conv(dilations = var_11964, groups = var_2624, pad = v_57_pad_0, pad_type = v_57_pad_type_0, strides = var_11962, weight = down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16, x = hidden_states_137_cast_fp16)[name = tensor("v_57_cast_fp16")]; + tensor var_11968_begin_0 = const()[name = tensor("op_11968_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11968_end_0 = const()[name = tensor("op_11968_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_11968_end_mask_0 = const()[name = tensor("op_11968_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11968_cast_fp16 = slice_by_index(begin = var_11968_begin_0, end = var_11968_end_0, end_mask = var_11968_end_mask_0, x = q_57_cast_fp16)[name = tensor("op_11968_cast_fp16")]; + tensor var_11972_begin_0 = const()[name = tensor("op_11972_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_11972_end_0 = const()[name = tensor("op_11972_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_11972_end_mask_0 = const()[name = tensor("op_11972_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11972_cast_fp16 = slice_by_index(begin = var_11972_begin_0, end = var_11972_end_0, end_mask = var_11972_end_mask_0, x = q_57_cast_fp16)[name = tensor("op_11972_cast_fp16")]; + tensor var_11976_begin_0 = const()[name = tensor("op_11976_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_11976_end_0 = const()[name = tensor("op_11976_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_11976_end_mask_0 = const()[name = tensor("op_11976_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11976_cast_fp16 = slice_by_index(begin = var_11976_begin_0, end = var_11976_end_0, end_mask = var_11976_end_mask_0, x = q_57_cast_fp16)[name = tensor("op_11976_cast_fp16")]; + tensor var_11980_begin_0 = const()[name = tensor("op_11980_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_11980_end_0 = const()[name = tensor("op_11980_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_11980_end_mask_0 = const()[name = tensor("op_11980_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11980_cast_fp16 = slice_by_index(begin = var_11980_begin_0, end = var_11980_end_0, end_mask = var_11980_end_mask_0, x = q_57_cast_fp16)[name = tensor("op_11980_cast_fp16")]; + tensor var_11984_begin_0 = const()[name = tensor("op_11984_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_11984_end_0 = const()[name = tensor("op_11984_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_11984_end_mask_0 = const()[name = tensor("op_11984_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11984_cast_fp16 = slice_by_index(begin = var_11984_begin_0, end = var_11984_end_0, end_mask = var_11984_end_mask_0, x = q_57_cast_fp16)[name = tensor("op_11984_cast_fp16")]; + tensor var_11988_begin_0 = const()[name = tensor("op_11988_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_11988_end_0 = const()[name = tensor("op_11988_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_11988_end_mask_0 = const()[name = tensor("op_11988_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11988_cast_fp16 = slice_by_index(begin = var_11988_begin_0, end = var_11988_end_0, end_mask = var_11988_end_mask_0, x = q_57_cast_fp16)[name = tensor("op_11988_cast_fp16")]; + tensor var_11992_begin_0 = const()[name = tensor("op_11992_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_11992_end_0 = const()[name = tensor("op_11992_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_11992_end_mask_0 = const()[name = tensor("op_11992_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11992_cast_fp16 = slice_by_index(begin = var_11992_begin_0, end = var_11992_end_0, end_mask = var_11992_end_mask_0, x = q_57_cast_fp16)[name = tensor("op_11992_cast_fp16")]; + tensor var_11996_begin_0 = const()[name = tensor("op_11996_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_11996_end_0 = const()[name = tensor("op_11996_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_11996_end_mask_0 = const()[name = tensor("op_11996_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11996_cast_fp16 = slice_by_index(begin = var_11996_begin_0, end = var_11996_end_0, end_mask = var_11996_end_mask_0, x = q_57_cast_fp16)[name = tensor("op_11996_cast_fp16")]; + tensor var_12000_begin_0 = const()[name = tensor("op_12000_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_12000_end_0 = const()[name = tensor("op_12000_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_12000_end_mask_0 = const()[name = tensor("op_12000_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12000_cast_fp16 = slice_by_index(begin = var_12000_begin_0, end = var_12000_end_0, end_mask = var_12000_end_mask_0, x = q_57_cast_fp16)[name = tensor("op_12000_cast_fp16")]; + tensor var_12004_begin_0 = const()[name = tensor("op_12004_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_12004_end_0 = const()[name = tensor("op_12004_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_12004_end_mask_0 = const()[name = tensor("op_12004_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12004_cast_fp16 = slice_by_index(begin = var_12004_begin_0, end = var_12004_end_0, end_mask = var_12004_end_mask_0, x = q_57_cast_fp16)[name = tensor("op_12004_cast_fp16")]; + tensor var_12008_begin_0 = const()[name = tensor("op_12008_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_12008_end_0 = const()[name = tensor("op_12008_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_12008_end_mask_0 = const()[name = tensor("op_12008_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12008_cast_fp16 = slice_by_index(begin = var_12008_begin_0, end = var_12008_end_0, end_mask = var_12008_end_mask_0, x = q_57_cast_fp16)[name = tensor("op_12008_cast_fp16")]; + tensor var_12012_begin_0 = const()[name = tensor("op_12012_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_12012_end_0 = const()[name = tensor("op_12012_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_12012_end_mask_0 = const()[name = tensor("op_12012_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12012_cast_fp16 = slice_by_index(begin = var_12012_begin_0, end = var_12012_end_0, end_mask = var_12012_end_mask_0, x = q_57_cast_fp16)[name = tensor("op_12012_cast_fp16")]; + tensor var_12016_begin_0 = const()[name = tensor("op_12016_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_12016_end_0 = const()[name = tensor("op_12016_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_12016_end_mask_0 = const()[name = tensor("op_12016_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12016_cast_fp16 = slice_by_index(begin = var_12016_begin_0, end = var_12016_end_0, end_mask = var_12016_end_mask_0, x = q_57_cast_fp16)[name = tensor("op_12016_cast_fp16")]; + tensor var_12020_begin_0 = const()[name = tensor("op_12020_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_12020_end_0 = const()[name = tensor("op_12020_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_12020_end_mask_0 = const()[name = tensor("op_12020_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12020_cast_fp16 = slice_by_index(begin = var_12020_begin_0, end = var_12020_end_0, end_mask = var_12020_end_mask_0, x = q_57_cast_fp16)[name = tensor("op_12020_cast_fp16")]; + tensor var_12024_begin_0 = const()[name = tensor("op_12024_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_12024_end_0 = const()[name = tensor("op_12024_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_12024_end_mask_0 = const()[name = tensor("op_12024_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12024_cast_fp16 = slice_by_index(begin = var_12024_begin_0, end = var_12024_end_0, end_mask = var_12024_end_mask_0, x = q_57_cast_fp16)[name = tensor("op_12024_cast_fp16")]; + tensor var_12028_begin_0 = const()[name = tensor("op_12028_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_12028_end_0 = const()[name = tensor("op_12028_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_12028_end_mask_0 = const()[name = tensor("op_12028_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12028_cast_fp16 = slice_by_index(begin = var_12028_begin_0, end = var_12028_end_0, end_mask = var_12028_end_mask_0, x = q_57_cast_fp16)[name = tensor("op_12028_cast_fp16")]; + tensor var_12032_begin_0 = const()[name = tensor("op_12032_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_12032_end_0 = const()[name = tensor("op_12032_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_12032_end_mask_0 = const()[name = tensor("op_12032_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12032_cast_fp16 = slice_by_index(begin = var_12032_begin_0, end = var_12032_end_0, end_mask = var_12032_end_mask_0, x = q_57_cast_fp16)[name = tensor("op_12032_cast_fp16")]; + tensor var_12036_begin_0 = const()[name = tensor("op_12036_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_12036_end_0 = const()[name = tensor("op_12036_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_12036_end_mask_0 = const()[name = tensor("op_12036_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12036_cast_fp16 = slice_by_index(begin = var_12036_begin_0, end = var_12036_end_0, end_mask = var_12036_end_mask_0, x = q_57_cast_fp16)[name = tensor("op_12036_cast_fp16")]; + tensor var_12040_begin_0 = const()[name = tensor("op_12040_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_12040_end_0 = const()[name = tensor("op_12040_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_12040_end_mask_0 = const()[name = tensor("op_12040_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12040_cast_fp16 = slice_by_index(begin = var_12040_begin_0, end = var_12040_end_0, end_mask = var_12040_end_mask_0, x = q_57_cast_fp16)[name = tensor("op_12040_cast_fp16")]; + tensor var_12044_begin_0 = const()[name = tensor("op_12044_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_12044_end_0 = const()[name = tensor("op_12044_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_12044_end_mask_0 = const()[name = tensor("op_12044_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12044_cast_fp16 = slice_by_index(begin = var_12044_begin_0, end = var_12044_end_0, end_mask = var_12044_end_mask_0, x = q_57_cast_fp16)[name = tensor("op_12044_cast_fp16")]; + tensor k_115_perm_0 = const()[name = tensor("k_115_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_12051_begin_0 = const()[name = tensor("op_12051_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12051_end_0 = const()[name = tensor("op_12051_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_12051_end_mask_0 = const()[name = tensor("op_12051_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_115_cast_fp16 = transpose(perm = k_115_perm_0, x = k_113_cast_fp16)[name = tensor("transpose_111")]; + tensor var_12051_cast_fp16 = slice_by_index(begin = var_12051_begin_0, end = var_12051_end_0, end_mask = var_12051_end_mask_0, x = k_115_cast_fp16)[name = tensor("op_12051_cast_fp16")]; + tensor var_12055_begin_0 = const()[name = tensor("op_12055_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_12055_end_0 = const()[name = tensor("op_12055_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_12055_end_mask_0 = const()[name = tensor("op_12055_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12055_cast_fp16 = slice_by_index(begin = var_12055_begin_0, end = var_12055_end_0, end_mask = var_12055_end_mask_0, x = k_115_cast_fp16)[name = tensor("op_12055_cast_fp16")]; + tensor var_12059_begin_0 = const()[name = tensor("op_12059_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_12059_end_0 = const()[name = tensor("op_12059_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_12059_end_mask_0 = const()[name = tensor("op_12059_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12059_cast_fp16 = slice_by_index(begin = var_12059_begin_0, end = var_12059_end_0, end_mask = var_12059_end_mask_0, x = k_115_cast_fp16)[name = tensor("op_12059_cast_fp16")]; + tensor var_12063_begin_0 = const()[name = tensor("op_12063_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_12063_end_0 = const()[name = tensor("op_12063_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_12063_end_mask_0 = const()[name = tensor("op_12063_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12063_cast_fp16 = slice_by_index(begin = var_12063_begin_0, end = var_12063_end_0, end_mask = var_12063_end_mask_0, x = k_115_cast_fp16)[name = tensor("op_12063_cast_fp16")]; + tensor var_12067_begin_0 = const()[name = tensor("op_12067_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_12067_end_0 = const()[name = tensor("op_12067_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_12067_end_mask_0 = const()[name = tensor("op_12067_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12067_cast_fp16 = slice_by_index(begin = var_12067_begin_0, end = var_12067_end_0, end_mask = var_12067_end_mask_0, x = k_115_cast_fp16)[name = tensor("op_12067_cast_fp16")]; + tensor var_12071_begin_0 = const()[name = tensor("op_12071_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_12071_end_0 = const()[name = tensor("op_12071_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_12071_end_mask_0 = const()[name = tensor("op_12071_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12071_cast_fp16 = slice_by_index(begin = var_12071_begin_0, end = var_12071_end_0, end_mask = var_12071_end_mask_0, x = k_115_cast_fp16)[name = tensor("op_12071_cast_fp16")]; + tensor var_12075_begin_0 = const()[name = tensor("op_12075_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_12075_end_0 = const()[name = tensor("op_12075_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_12075_end_mask_0 = const()[name = tensor("op_12075_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12075_cast_fp16 = slice_by_index(begin = var_12075_begin_0, end = var_12075_end_0, end_mask = var_12075_end_mask_0, x = k_115_cast_fp16)[name = tensor("op_12075_cast_fp16")]; + tensor var_12079_begin_0 = const()[name = tensor("op_12079_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_12079_end_0 = const()[name = tensor("op_12079_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_12079_end_mask_0 = const()[name = tensor("op_12079_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12079_cast_fp16 = slice_by_index(begin = var_12079_begin_0, end = var_12079_end_0, end_mask = var_12079_end_mask_0, x = k_115_cast_fp16)[name = tensor("op_12079_cast_fp16")]; + tensor var_12083_begin_0 = const()[name = tensor("op_12083_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_12083_end_0 = const()[name = tensor("op_12083_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_12083_end_mask_0 = const()[name = tensor("op_12083_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12083_cast_fp16 = slice_by_index(begin = var_12083_begin_0, end = var_12083_end_0, end_mask = var_12083_end_mask_0, x = k_115_cast_fp16)[name = tensor("op_12083_cast_fp16")]; + tensor var_12087_begin_0 = const()[name = tensor("op_12087_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_12087_end_0 = const()[name = tensor("op_12087_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_12087_end_mask_0 = const()[name = tensor("op_12087_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12087_cast_fp16 = slice_by_index(begin = var_12087_begin_0, end = var_12087_end_0, end_mask = var_12087_end_mask_0, x = k_115_cast_fp16)[name = tensor("op_12087_cast_fp16")]; + tensor var_12091_begin_0 = const()[name = tensor("op_12091_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_12091_end_0 = const()[name = tensor("op_12091_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_12091_end_mask_0 = const()[name = tensor("op_12091_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12091_cast_fp16 = slice_by_index(begin = var_12091_begin_0, end = var_12091_end_0, end_mask = var_12091_end_mask_0, x = k_115_cast_fp16)[name = tensor("op_12091_cast_fp16")]; + tensor var_12095_begin_0 = const()[name = tensor("op_12095_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_12095_end_0 = const()[name = tensor("op_12095_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_12095_end_mask_0 = const()[name = tensor("op_12095_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12095_cast_fp16 = slice_by_index(begin = var_12095_begin_0, end = var_12095_end_0, end_mask = var_12095_end_mask_0, x = k_115_cast_fp16)[name = tensor("op_12095_cast_fp16")]; + tensor var_12099_begin_0 = const()[name = tensor("op_12099_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_12099_end_0 = const()[name = tensor("op_12099_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_12099_end_mask_0 = const()[name = tensor("op_12099_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12099_cast_fp16 = slice_by_index(begin = var_12099_begin_0, end = var_12099_end_0, end_mask = var_12099_end_mask_0, x = k_115_cast_fp16)[name = tensor("op_12099_cast_fp16")]; + tensor var_12103_begin_0 = const()[name = tensor("op_12103_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_12103_end_0 = const()[name = tensor("op_12103_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_12103_end_mask_0 = const()[name = tensor("op_12103_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12103_cast_fp16 = slice_by_index(begin = var_12103_begin_0, end = var_12103_end_0, end_mask = var_12103_end_mask_0, x = k_115_cast_fp16)[name = tensor("op_12103_cast_fp16")]; + tensor var_12107_begin_0 = const()[name = tensor("op_12107_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_12107_end_0 = const()[name = tensor("op_12107_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_12107_end_mask_0 = const()[name = tensor("op_12107_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12107_cast_fp16 = slice_by_index(begin = var_12107_begin_0, end = var_12107_end_0, end_mask = var_12107_end_mask_0, x = k_115_cast_fp16)[name = tensor("op_12107_cast_fp16")]; + tensor var_12111_begin_0 = const()[name = tensor("op_12111_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_12111_end_0 = const()[name = tensor("op_12111_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_12111_end_mask_0 = const()[name = tensor("op_12111_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12111_cast_fp16 = slice_by_index(begin = var_12111_begin_0, end = var_12111_end_0, end_mask = var_12111_end_mask_0, x = k_115_cast_fp16)[name = tensor("op_12111_cast_fp16")]; + tensor var_12115_begin_0 = const()[name = tensor("op_12115_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_12115_end_0 = const()[name = tensor("op_12115_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_12115_end_mask_0 = const()[name = tensor("op_12115_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12115_cast_fp16 = slice_by_index(begin = var_12115_begin_0, end = var_12115_end_0, end_mask = var_12115_end_mask_0, x = k_115_cast_fp16)[name = tensor("op_12115_cast_fp16")]; + tensor var_12119_begin_0 = const()[name = tensor("op_12119_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_12119_end_0 = const()[name = tensor("op_12119_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_12119_end_mask_0 = const()[name = tensor("op_12119_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12119_cast_fp16 = slice_by_index(begin = var_12119_begin_0, end = var_12119_end_0, end_mask = var_12119_end_mask_0, x = k_115_cast_fp16)[name = tensor("op_12119_cast_fp16")]; + tensor var_12123_begin_0 = const()[name = tensor("op_12123_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_12123_end_0 = const()[name = tensor("op_12123_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_12123_end_mask_0 = const()[name = tensor("op_12123_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12123_cast_fp16 = slice_by_index(begin = var_12123_begin_0, end = var_12123_end_0, end_mask = var_12123_end_mask_0, x = k_115_cast_fp16)[name = tensor("op_12123_cast_fp16")]; + tensor var_12127_begin_0 = const()[name = tensor("op_12127_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_12127_end_0 = const()[name = tensor("op_12127_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_12127_end_mask_0 = const()[name = tensor("op_12127_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12127_cast_fp16 = slice_by_index(begin = var_12127_begin_0, end = var_12127_end_0, end_mask = var_12127_end_mask_0, x = k_115_cast_fp16)[name = tensor("op_12127_cast_fp16")]; + tensor var_12129_begin_0 = const()[name = tensor("op_12129_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12129_end_0 = const()[name = tensor("op_12129_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_12129_end_mask_0 = const()[name = tensor("op_12129_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12129_cast_fp16 = slice_by_index(begin = var_12129_begin_0, end = var_12129_end_0, end_mask = var_12129_end_mask_0, x = v_57_cast_fp16)[name = tensor("op_12129_cast_fp16")]; + tensor var_12133_begin_0 = const()[name = tensor("op_12133_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_12133_end_0 = const()[name = tensor("op_12133_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_12133_end_mask_0 = const()[name = tensor("op_12133_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12133_cast_fp16 = slice_by_index(begin = var_12133_begin_0, end = var_12133_end_0, end_mask = var_12133_end_mask_0, x = v_57_cast_fp16)[name = tensor("op_12133_cast_fp16")]; + tensor var_12137_begin_0 = const()[name = tensor("op_12137_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_12137_end_0 = const()[name = tensor("op_12137_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_12137_end_mask_0 = const()[name = tensor("op_12137_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12137_cast_fp16 = slice_by_index(begin = var_12137_begin_0, end = var_12137_end_0, end_mask = var_12137_end_mask_0, x = v_57_cast_fp16)[name = tensor("op_12137_cast_fp16")]; + tensor var_12141_begin_0 = const()[name = tensor("op_12141_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_12141_end_0 = const()[name = tensor("op_12141_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_12141_end_mask_0 = const()[name = tensor("op_12141_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12141_cast_fp16 = slice_by_index(begin = var_12141_begin_0, end = var_12141_end_0, end_mask = var_12141_end_mask_0, x = v_57_cast_fp16)[name = tensor("op_12141_cast_fp16")]; + tensor var_12145_begin_0 = const()[name = tensor("op_12145_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_12145_end_0 = const()[name = tensor("op_12145_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_12145_end_mask_0 = const()[name = tensor("op_12145_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12145_cast_fp16 = slice_by_index(begin = var_12145_begin_0, end = var_12145_end_0, end_mask = var_12145_end_mask_0, x = v_57_cast_fp16)[name = tensor("op_12145_cast_fp16")]; + tensor var_12149_begin_0 = const()[name = tensor("op_12149_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_12149_end_0 = const()[name = tensor("op_12149_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_12149_end_mask_0 = const()[name = tensor("op_12149_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12149_cast_fp16 = slice_by_index(begin = var_12149_begin_0, end = var_12149_end_0, end_mask = var_12149_end_mask_0, x = v_57_cast_fp16)[name = tensor("op_12149_cast_fp16")]; + tensor var_12153_begin_0 = const()[name = tensor("op_12153_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_12153_end_0 = const()[name = tensor("op_12153_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_12153_end_mask_0 = const()[name = tensor("op_12153_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12153_cast_fp16 = slice_by_index(begin = var_12153_begin_0, end = var_12153_end_0, end_mask = var_12153_end_mask_0, x = v_57_cast_fp16)[name = tensor("op_12153_cast_fp16")]; + tensor var_12157_begin_0 = const()[name = tensor("op_12157_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_12157_end_0 = const()[name = tensor("op_12157_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_12157_end_mask_0 = const()[name = tensor("op_12157_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12157_cast_fp16 = slice_by_index(begin = var_12157_begin_0, end = var_12157_end_0, end_mask = var_12157_end_mask_0, x = v_57_cast_fp16)[name = tensor("op_12157_cast_fp16")]; + tensor var_12161_begin_0 = const()[name = tensor("op_12161_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_12161_end_0 = const()[name = tensor("op_12161_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_12161_end_mask_0 = const()[name = tensor("op_12161_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12161_cast_fp16 = slice_by_index(begin = var_12161_begin_0, end = var_12161_end_0, end_mask = var_12161_end_mask_0, x = v_57_cast_fp16)[name = tensor("op_12161_cast_fp16")]; + tensor var_12165_begin_0 = const()[name = tensor("op_12165_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_12165_end_0 = const()[name = tensor("op_12165_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_12165_end_mask_0 = const()[name = tensor("op_12165_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12165_cast_fp16 = slice_by_index(begin = var_12165_begin_0, end = var_12165_end_0, end_mask = var_12165_end_mask_0, x = v_57_cast_fp16)[name = tensor("op_12165_cast_fp16")]; + tensor var_12169_begin_0 = const()[name = tensor("op_12169_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_12169_end_0 = const()[name = tensor("op_12169_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_12169_end_mask_0 = const()[name = tensor("op_12169_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12169_cast_fp16 = slice_by_index(begin = var_12169_begin_0, end = var_12169_end_0, end_mask = var_12169_end_mask_0, x = v_57_cast_fp16)[name = tensor("op_12169_cast_fp16")]; + tensor var_12173_begin_0 = const()[name = tensor("op_12173_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_12173_end_0 = const()[name = tensor("op_12173_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_12173_end_mask_0 = const()[name = tensor("op_12173_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12173_cast_fp16 = slice_by_index(begin = var_12173_begin_0, end = var_12173_end_0, end_mask = var_12173_end_mask_0, x = v_57_cast_fp16)[name = tensor("op_12173_cast_fp16")]; + tensor var_12177_begin_0 = const()[name = tensor("op_12177_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_12177_end_0 = const()[name = tensor("op_12177_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_12177_end_mask_0 = const()[name = tensor("op_12177_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12177_cast_fp16 = slice_by_index(begin = var_12177_begin_0, end = var_12177_end_0, end_mask = var_12177_end_mask_0, x = v_57_cast_fp16)[name = tensor("op_12177_cast_fp16")]; + tensor var_12181_begin_0 = const()[name = tensor("op_12181_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_12181_end_0 = const()[name = tensor("op_12181_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_12181_end_mask_0 = const()[name = tensor("op_12181_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12181_cast_fp16 = slice_by_index(begin = var_12181_begin_0, end = var_12181_end_0, end_mask = var_12181_end_mask_0, x = v_57_cast_fp16)[name = tensor("op_12181_cast_fp16")]; + tensor var_12185_begin_0 = const()[name = tensor("op_12185_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_12185_end_0 = const()[name = tensor("op_12185_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_12185_end_mask_0 = const()[name = tensor("op_12185_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12185_cast_fp16 = slice_by_index(begin = var_12185_begin_0, end = var_12185_end_0, end_mask = var_12185_end_mask_0, x = v_57_cast_fp16)[name = tensor("op_12185_cast_fp16")]; + tensor var_12189_begin_0 = const()[name = tensor("op_12189_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_12189_end_0 = const()[name = tensor("op_12189_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_12189_end_mask_0 = const()[name = tensor("op_12189_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12189_cast_fp16 = slice_by_index(begin = var_12189_begin_0, end = var_12189_end_0, end_mask = var_12189_end_mask_0, x = v_57_cast_fp16)[name = tensor("op_12189_cast_fp16")]; + tensor var_12193_begin_0 = const()[name = tensor("op_12193_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_12193_end_0 = const()[name = tensor("op_12193_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_12193_end_mask_0 = const()[name = tensor("op_12193_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12193_cast_fp16 = slice_by_index(begin = var_12193_begin_0, end = var_12193_end_0, end_mask = var_12193_end_mask_0, x = v_57_cast_fp16)[name = tensor("op_12193_cast_fp16")]; + tensor var_12197_begin_0 = const()[name = tensor("op_12197_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_12197_end_0 = const()[name = tensor("op_12197_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_12197_end_mask_0 = const()[name = tensor("op_12197_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12197_cast_fp16 = slice_by_index(begin = var_12197_begin_0, end = var_12197_end_0, end_mask = var_12197_end_mask_0, x = v_57_cast_fp16)[name = tensor("op_12197_cast_fp16")]; + tensor var_12201_begin_0 = const()[name = tensor("op_12201_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_12201_end_0 = const()[name = tensor("op_12201_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_12201_end_mask_0 = const()[name = tensor("op_12201_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12201_cast_fp16 = slice_by_index(begin = var_12201_begin_0, end = var_12201_end_0, end_mask = var_12201_end_mask_0, x = v_57_cast_fp16)[name = tensor("op_12201_cast_fp16")]; + tensor var_12205_begin_0 = const()[name = tensor("op_12205_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_12205_end_0 = const()[name = tensor("op_12205_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_12205_end_mask_0 = const()[name = tensor("op_12205_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12205_cast_fp16 = slice_by_index(begin = var_12205_begin_0, end = var_12205_end_0, end_mask = var_12205_end_mask_0, x = v_57_cast_fp16)[name = tensor("op_12205_cast_fp16")]; + tensor var_12209_equation_0 = const()[name = tensor("op_12209_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12209_cast_fp16 = einsum(equation = var_12209_equation_0, values = (var_12051_cast_fp16, var_11968_cast_fp16))[name = tensor("op_12209_cast_fp16")]; + tensor var_12210_to_fp16 = const()[name = tensor("op_12210_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_961_cast_fp16 = mul(x = var_12209_cast_fp16, y = var_12210_to_fp16)[name = tensor("aw_961_cast_fp16")]; + tensor var_12213_equation_0 = const()[name = tensor("op_12213_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12213_cast_fp16 = einsum(equation = var_12213_equation_0, values = (var_12055_cast_fp16, var_11972_cast_fp16))[name = tensor("op_12213_cast_fp16")]; + tensor var_12214_to_fp16 = const()[name = tensor("op_12214_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_963_cast_fp16 = mul(x = var_12213_cast_fp16, y = var_12214_to_fp16)[name = tensor("aw_963_cast_fp16")]; + tensor var_12217_equation_0 = const()[name = tensor("op_12217_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12217_cast_fp16 = einsum(equation = var_12217_equation_0, values = (var_12059_cast_fp16, var_11976_cast_fp16))[name = tensor("op_12217_cast_fp16")]; + tensor var_12218_to_fp16 = const()[name = tensor("op_12218_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_965_cast_fp16 = mul(x = var_12217_cast_fp16, y = var_12218_to_fp16)[name = tensor("aw_965_cast_fp16")]; + tensor var_12221_equation_0 = const()[name = tensor("op_12221_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12221_cast_fp16 = einsum(equation = var_12221_equation_0, values = (var_12063_cast_fp16, var_11980_cast_fp16))[name = tensor("op_12221_cast_fp16")]; + tensor var_12222_to_fp16 = const()[name = tensor("op_12222_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_967_cast_fp16 = mul(x = var_12221_cast_fp16, y = var_12222_to_fp16)[name = tensor("aw_967_cast_fp16")]; + tensor var_12225_equation_0 = const()[name = tensor("op_12225_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12225_cast_fp16 = einsum(equation = var_12225_equation_0, values = (var_12067_cast_fp16, var_11984_cast_fp16))[name = tensor("op_12225_cast_fp16")]; + tensor var_12226_to_fp16 = const()[name = tensor("op_12226_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_969_cast_fp16 = mul(x = var_12225_cast_fp16, y = var_12226_to_fp16)[name = tensor("aw_969_cast_fp16")]; + tensor var_12229_equation_0 = const()[name = tensor("op_12229_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12229_cast_fp16 = einsum(equation = var_12229_equation_0, values = (var_12071_cast_fp16, var_11988_cast_fp16))[name = tensor("op_12229_cast_fp16")]; + tensor var_12230_to_fp16 = const()[name = tensor("op_12230_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_971_cast_fp16 = mul(x = var_12229_cast_fp16, y = var_12230_to_fp16)[name = tensor("aw_971_cast_fp16")]; + tensor var_12233_equation_0 = const()[name = tensor("op_12233_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12233_cast_fp16 = einsum(equation = var_12233_equation_0, values = (var_12075_cast_fp16, var_11992_cast_fp16))[name = tensor("op_12233_cast_fp16")]; + tensor var_12234_to_fp16 = const()[name = tensor("op_12234_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_973_cast_fp16 = mul(x = var_12233_cast_fp16, y = var_12234_to_fp16)[name = tensor("aw_973_cast_fp16")]; + tensor var_12237_equation_0 = const()[name = tensor("op_12237_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12237_cast_fp16 = einsum(equation = var_12237_equation_0, values = (var_12079_cast_fp16, var_11996_cast_fp16))[name = tensor("op_12237_cast_fp16")]; + tensor var_12238_to_fp16 = const()[name = tensor("op_12238_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_975_cast_fp16 = mul(x = var_12237_cast_fp16, y = var_12238_to_fp16)[name = tensor("aw_975_cast_fp16")]; + tensor var_12241_equation_0 = const()[name = tensor("op_12241_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12241_cast_fp16 = einsum(equation = var_12241_equation_0, values = (var_12083_cast_fp16, var_12000_cast_fp16))[name = tensor("op_12241_cast_fp16")]; + tensor var_12242_to_fp16 = const()[name = tensor("op_12242_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_977_cast_fp16 = mul(x = var_12241_cast_fp16, y = var_12242_to_fp16)[name = tensor("aw_977_cast_fp16")]; + tensor var_12245_equation_0 = const()[name = tensor("op_12245_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12245_cast_fp16 = einsum(equation = var_12245_equation_0, values = (var_12087_cast_fp16, var_12004_cast_fp16))[name = tensor("op_12245_cast_fp16")]; + tensor var_12246_to_fp16 = const()[name = tensor("op_12246_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_979_cast_fp16 = mul(x = var_12245_cast_fp16, y = var_12246_to_fp16)[name = tensor("aw_979_cast_fp16")]; + tensor var_12249_equation_0 = const()[name = tensor("op_12249_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12249_cast_fp16 = einsum(equation = var_12249_equation_0, values = (var_12091_cast_fp16, var_12008_cast_fp16))[name = tensor("op_12249_cast_fp16")]; + tensor var_12250_to_fp16 = const()[name = tensor("op_12250_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_981_cast_fp16 = mul(x = var_12249_cast_fp16, y = var_12250_to_fp16)[name = tensor("aw_981_cast_fp16")]; + tensor var_12253_equation_0 = const()[name = tensor("op_12253_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12253_cast_fp16 = einsum(equation = var_12253_equation_0, values = (var_12095_cast_fp16, var_12012_cast_fp16))[name = tensor("op_12253_cast_fp16")]; + tensor var_12254_to_fp16 = const()[name = tensor("op_12254_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_983_cast_fp16 = mul(x = var_12253_cast_fp16, y = var_12254_to_fp16)[name = tensor("aw_983_cast_fp16")]; + tensor var_12257_equation_0 = const()[name = tensor("op_12257_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12257_cast_fp16 = einsum(equation = var_12257_equation_0, values = (var_12099_cast_fp16, var_12016_cast_fp16))[name = tensor("op_12257_cast_fp16")]; + tensor var_12258_to_fp16 = const()[name = tensor("op_12258_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_985_cast_fp16 = mul(x = var_12257_cast_fp16, y = var_12258_to_fp16)[name = tensor("aw_985_cast_fp16")]; + tensor var_12261_equation_0 = const()[name = tensor("op_12261_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12261_cast_fp16 = einsum(equation = var_12261_equation_0, values = (var_12103_cast_fp16, var_12020_cast_fp16))[name = tensor("op_12261_cast_fp16")]; + tensor var_12262_to_fp16 = const()[name = tensor("op_12262_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_987_cast_fp16 = mul(x = var_12261_cast_fp16, y = var_12262_to_fp16)[name = tensor("aw_987_cast_fp16")]; + tensor var_12265_equation_0 = const()[name = tensor("op_12265_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12265_cast_fp16 = einsum(equation = var_12265_equation_0, values = (var_12107_cast_fp16, var_12024_cast_fp16))[name = tensor("op_12265_cast_fp16")]; + tensor var_12266_to_fp16 = const()[name = tensor("op_12266_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_989_cast_fp16 = mul(x = var_12265_cast_fp16, y = var_12266_to_fp16)[name = tensor("aw_989_cast_fp16")]; + tensor var_12269_equation_0 = const()[name = tensor("op_12269_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12269_cast_fp16 = einsum(equation = var_12269_equation_0, values = (var_12111_cast_fp16, var_12028_cast_fp16))[name = tensor("op_12269_cast_fp16")]; + tensor var_12270_to_fp16 = const()[name = tensor("op_12270_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_991_cast_fp16 = mul(x = var_12269_cast_fp16, y = var_12270_to_fp16)[name = tensor("aw_991_cast_fp16")]; + tensor var_12273_equation_0 = const()[name = tensor("op_12273_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12273_cast_fp16 = einsum(equation = var_12273_equation_0, values = (var_12115_cast_fp16, var_12032_cast_fp16))[name = tensor("op_12273_cast_fp16")]; + tensor var_12274_to_fp16 = const()[name = tensor("op_12274_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_993_cast_fp16 = mul(x = var_12273_cast_fp16, y = var_12274_to_fp16)[name = tensor("aw_993_cast_fp16")]; + tensor var_12277_equation_0 = const()[name = tensor("op_12277_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12277_cast_fp16 = einsum(equation = var_12277_equation_0, values = (var_12119_cast_fp16, var_12036_cast_fp16))[name = tensor("op_12277_cast_fp16")]; + tensor var_12278_to_fp16 = const()[name = tensor("op_12278_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_995_cast_fp16 = mul(x = var_12277_cast_fp16, y = var_12278_to_fp16)[name = tensor("aw_995_cast_fp16")]; + tensor var_12281_equation_0 = const()[name = tensor("op_12281_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12281_cast_fp16 = einsum(equation = var_12281_equation_0, values = (var_12123_cast_fp16, var_12040_cast_fp16))[name = tensor("op_12281_cast_fp16")]; + tensor var_12282_to_fp16 = const()[name = tensor("op_12282_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_997_cast_fp16 = mul(x = var_12281_cast_fp16, y = var_12282_to_fp16)[name = tensor("aw_997_cast_fp16")]; + tensor var_12285_equation_0 = const()[name = tensor("op_12285_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12285_cast_fp16 = einsum(equation = var_12285_equation_0, values = (var_12127_cast_fp16, var_12044_cast_fp16))[name = tensor("op_12285_cast_fp16")]; + tensor var_12286_to_fp16 = const()[name = tensor("op_12286_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_999_cast_fp16 = mul(x = var_12285_cast_fp16, y = var_12286_to_fp16)[name = tensor("aw_999_cast_fp16")]; + tensor var_12288_cast_fp16 = softmax(axis = var_2624, x = aw_961_cast_fp16)[name = tensor("op_12288_cast_fp16")]; + tensor var_12289_cast_fp16 = softmax(axis = var_2624, x = aw_963_cast_fp16)[name = tensor("op_12289_cast_fp16")]; + tensor var_12290_cast_fp16 = softmax(axis = var_2624, x = aw_965_cast_fp16)[name = tensor("op_12290_cast_fp16")]; + tensor var_12291_cast_fp16 = softmax(axis = var_2624, x = aw_967_cast_fp16)[name = tensor("op_12291_cast_fp16")]; + tensor var_12292_cast_fp16 = softmax(axis = var_2624, x = aw_969_cast_fp16)[name = tensor("op_12292_cast_fp16")]; + tensor var_12293_cast_fp16 = softmax(axis = var_2624, x = aw_971_cast_fp16)[name = tensor("op_12293_cast_fp16")]; + tensor var_12294_cast_fp16 = softmax(axis = var_2624, x = aw_973_cast_fp16)[name = tensor("op_12294_cast_fp16")]; + tensor var_12295_cast_fp16 = softmax(axis = var_2624, x = aw_975_cast_fp16)[name = tensor("op_12295_cast_fp16")]; + tensor var_12296_cast_fp16 = softmax(axis = var_2624, x = aw_977_cast_fp16)[name = tensor("op_12296_cast_fp16")]; + tensor var_12297_cast_fp16 = softmax(axis = var_2624, x = aw_979_cast_fp16)[name = tensor("op_12297_cast_fp16")]; + tensor var_12298_cast_fp16 = softmax(axis = var_2624, x = aw_981_cast_fp16)[name = tensor("op_12298_cast_fp16")]; + tensor var_12299_cast_fp16 = softmax(axis = var_2624, x = aw_983_cast_fp16)[name = tensor("op_12299_cast_fp16")]; + tensor var_12300_cast_fp16 = softmax(axis = var_2624, x = aw_985_cast_fp16)[name = tensor("op_12300_cast_fp16")]; + tensor var_12301_cast_fp16 = softmax(axis = var_2624, x = aw_987_cast_fp16)[name = tensor("op_12301_cast_fp16")]; + tensor var_12302_cast_fp16 = softmax(axis = var_2624, x = aw_989_cast_fp16)[name = tensor("op_12302_cast_fp16")]; + tensor var_12303_cast_fp16 = softmax(axis = var_2624, x = aw_991_cast_fp16)[name = tensor("op_12303_cast_fp16")]; + tensor var_12304_cast_fp16 = softmax(axis = var_2624, x = aw_993_cast_fp16)[name = tensor("op_12304_cast_fp16")]; + tensor var_12305_cast_fp16 = softmax(axis = var_2624, x = aw_995_cast_fp16)[name = tensor("op_12305_cast_fp16")]; + tensor var_12306_cast_fp16 = softmax(axis = var_2624, x = aw_997_cast_fp16)[name = tensor("op_12306_cast_fp16")]; + tensor var_12307_cast_fp16 = softmax(axis = var_2624, x = aw_999_cast_fp16)[name = tensor("op_12307_cast_fp16")]; + tensor var_12309_equation_0 = const()[name = tensor("op_12309_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12309_cast_fp16 = einsum(equation = var_12309_equation_0, values = (var_12129_cast_fp16, var_12288_cast_fp16))[name = tensor("op_12309_cast_fp16")]; + tensor var_12311_equation_0 = const()[name = tensor("op_12311_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12311_cast_fp16 = einsum(equation = var_12311_equation_0, values = (var_12133_cast_fp16, var_12289_cast_fp16))[name = tensor("op_12311_cast_fp16")]; + tensor var_12313_equation_0 = const()[name = tensor("op_12313_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12313_cast_fp16 = einsum(equation = var_12313_equation_0, values = (var_12137_cast_fp16, var_12290_cast_fp16))[name = tensor("op_12313_cast_fp16")]; + tensor var_12315_equation_0 = const()[name = tensor("op_12315_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12315_cast_fp16 = einsum(equation = var_12315_equation_0, values = (var_12141_cast_fp16, var_12291_cast_fp16))[name = tensor("op_12315_cast_fp16")]; + tensor var_12317_equation_0 = const()[name = tensor("op_12317_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12317_cast_fp16 = einsum(equation = var_12317_equation_0, values = (var_12145_cast_fp16, var_12292_cast_fp16))[name = tensor("op_12317_cast_fp16")]; + tensor var_12319_equation_0 = const()[name = tensor("op_12319_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12319_cast_fp16 = einsum(equation = var_12319_equation_0, values = (var_12149_cast_fp16, var_12293_cast_fp16))[name = tensor("op_12319_cast_fp16")]; + tensor var_12321_equation_0 = const()[name = tensor("op_12321_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12321_cast_fp16 = einsum(equation = var_12321_equation_0, values = (var_12153_cast_fp16, var_12294_cast_fp16))[name = tensor("op_12321_cast_fp16")]; + tensor var_12323_equation_0 = const()[name = tensor("op_12323_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12323_cast_fp16 = einsum(equation = var_12323_equation_0, values = (var_12157_cast_fp16, var_12295_cast_fp16))[name = tensor("op_12323_cast_fp16")]; + tensor var_12325_equation_0 = const()[name = tensor("op_12325_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12325_cast_fp16 = einsum(equation = var_12325_equation_0, values = (var_12161_cast_fp16, var_12296_cast_fp16))[name = tensor("op_12325_cast_fp16")]; + tensor var_12327_equation_0 = const()[name = tensor("op_12327_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12327_cast_fp16 = einsum(equation = var_12327_equation_0, values = (var_12165_cast_fp16, var_12297_cast_fp16))[name = tensor("op_12327_cast_fp16")]; + tensor var_12329_equation_0 = const()[name = tensor("op_12329_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12329_cast_fp16 = einsum(equation = var_12329_equation_0, values = (var_12169_cast_fp16, var_12298_cast_fp16))[name = tensor("op_12329_cast_fp16")]; + tensor var_12331_equation_0 = const()[name = tensor("op_12331_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12331_cast_fp16 = einsum(equation = var_12331_equation_0, values = (var_12173_cast_fp16, var_12299_cast_fp16))[name = tensor("op_12331_cast_fp16")]; + tensor var_12333_equation_0 = const()[name = tensor("op_12333_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12333_cast_fp16 = einsum(equation = var_12333_equation_0, values = (var_12177_cast_fp16, var_12300_cast_fp16))[name = tensor("op_12333_cast_fp16")]; + tensor var_12335_equation_0 = const()[name = tensor("op_12335_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12335_cast_fp16 = einsum(equation = var_12335_equation_0, values = (var_12181_cast_fp16, var_12301_cast_fp16))[name = tensor("op_12335_cast_fp16")]; + tensor var_12337_equation_0 = const()[name = tensor("op_12337_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12337_cast_fp16 = einsum(equation = var_12337_equation_0, values = (var_12185_cast_fp16, var_12302_cast_fp16))[name = tensor("op_12337_cast_fp16")]; + tensor var_12339_equation_0 = const()[name = tensor("op_12339_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12339_cast_fp16 = einsum(equation = var_12339_equation_0, values = (var_12189_cast_fp16, var_12303_cast_fp16))[name = tensor("op_12339_cast_fp16")]; + tensor var_12341_equation_0 = const()[name = tensor("op_12341_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12341_cast_fp16 = einsum(equation = var_12341_equation_0, values = (var_12193_cast_fp16, var_12304_cast_fp16))[name = tensor("op_12341_cast_fp16")]; + tensor var_12343_equation_0 = const()[name = tensor("op_12343_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12343_cast_fp16 = einsum(equation = var_12343_equation_0, values = (var_12197_cast_fp16, var_12305_cast_fp16))[name = tensor("op_12343_cast_fp16")]; + tensor var_12345_equation_0 = const()[name = tensor("op_12345_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12345_cast_fp16 = einsum(equation = var_12345_equation_0, values = (var_12201_cast_fp16, var_12306_cast_fp16))[name = tensor("op_12345_cast_fp16")]; + tensor var_12347_equation_0 = const()[name = tensor("op_12347_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12347_cast_fp16 = einsum(equation = var_12347_equation_0, values = (var_12205_cast_fp16, var_12307_cast_fp16))[name = tensor("op_12347_cast_fp16")]; + tensor input_229_interleave_0 = const()[name = tensor("input_229_interleave_0"), val = tensor(false)]; + tensor input_229_cast_fp16 = concat(axis = var_2624, interleave = input_229_interleave_0, values = (var_12309_cast_fp16, var_12311_cast_fp16, var_12313_cast_fp16, var_12315_cast_fp16, var_12317_cast_fp16, var_12319_cast_fp16, var_12321_cast_fp16, var_12323_cast_fp16, var_12325_cast_fp16, var_12327_cast_fp16, var_12329_cast_fp16, var_12331_cast_fp16, var_12333_cast_fp16, var_12335_cast_fp16, var_12337_cast_fp16, var_12339_cast_fp16, var_12341_cast_fp16, var_12343_cast_fp16, var_12345_cast_fp16, var_12347_cast_fp16))[name = tensor("input_229_cast_fp16")]; + tensor var_12353 = const()[name = tensor("op_12353"), val = tensor([1, 1])]; + tensor var_12355 = const()[name = tensor("op_12355"), val = tensor([1, 1])]; + tensor var_12357_pad_type_0 = const()[name = tensor("op_12357_pad_type_0"), val = tensor("custom")]; + tensor var_12357_pad_0 = const()[name = tensor("op_12357_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(971973312)))]; + tensor down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(975250176)))]; + tensor var_12357_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_12355, groups = var_2624, pad = var_12357_pad_0, pad_type = var_12357_pad_type_0, strides = var_12353, weight = down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16, x = input_229_cast_fp16)[name = tensor("op_12357_cast_fp16")]; + tensor inputs_87_cast_fp16 = add(x = var_12357_cast_fp16, y = inputs_85_cast_fp16)[name = tensor("inputs_87_cast_fp16")]; + tensor hidden_states_139_axes_0 = const()[name = tensor("hidden_states_139_axes_0"), val = tensor([1])]; + tensor hidden_states_139_gamma_0_to_fp16 = const()[name = tensor("hidden_states_139_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(975252800)))]; + tensor hidden_states_139_beta_0_to_fp16 = const()[name = tensor("hidden_states_139_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(975255424)))]; + tensor var_12367_to_fp16 = const()[name = tensor("op_12367_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_139_cast_fp16 = layer_norm(axes = hidden_states_139_axes_0, beta = hidden_states_139_beta_0_to_fp16, epsilon = var_12367_to_fp16, gamma = hidden_states_139_gamma_0_to_fp16, x = inputs_87_cast_fp16)[name = tensor("hidden_states_139_cast_fp16")]; + tensor var_12382 = const()[name = tensor("op_12382"), val = tensor([1, 1])]; + tensor var_12384 = const()[name = tensor("op_12384"), val = tensor([1, 1])]; + tensor q_59_pad_type_0 = const()[name = tensor("q_59_pad_type_0"), val = tensor("custom")]; + tensor q_59_pad_0 = const()[name = tensor("q_59_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(975258048)))]; + tensor q_59_cast_fp16 = conv(dilations = var_12384, groups = var_2624, pad = q_59_pad_0, pad_type = q_59_pad_type_0, strides = var_12382, weight = down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16, x = hidden_states_139_cast_fp16)[name = tensor("q_59_cast_fp16")]; + tensor var_12388 = const()[name = tensor("op_12388"), val = tensor([1, 1])]; + tensor var_12390 = const()[name = tensor("op_12390"), val = tensor([1, 1])]; + tensor k_117_pad_type_0 = const()[name = tensor("k_117_pad_type_0"), val = tensor("custom")]; + tensor k_117_pad_0 = const()[name = tensor("k_117_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(978534912)))]; + tensor k_117_cast_fp16 = conv(dilations = var_12390, groups = var_2624, pad = k_117_pad_0, pad_type = k_117_pad_type_0, strides = var_12388, weight = down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_117_cast_fp16")]; + tensor var_12394 = const()[name = tensor("op_12394"), val = tensor([1, 1])]; + tensor var_12396 = const()[name = tensor("op_12396"), val = tensor([1, 1])]; + tensor v_59_pad_type_0 = const()[name = tensor("v_59_pad_type_0"), val = tensor("custom")]; + tensor v_59_pad_0 = const()[name = tensor("v_59_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(983777856)))]; + tensor v_59_cast_fp16 = conv(dilations = var_12396, groups = var_2624, pad = v_59_pad_0, pad_type = v_59_pad_type_0, strides = var_12394, weight = down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_59_cast_fp16")]; + tensor var_12400_begin_0 = const()[name = tensor("op_12400_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12400_end_0 = const()[name = tensor("op_12400_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_12400_end_mask_0 = const()[name = tensor("op_12400_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12400_cast_fp16 = slice_by_index(begin = var_12400_begin_0, end = var_12400_end_0, end_mask = var_12400_end_mask_0, x = q_59_cast_fp16)[name = tensor("op_12400_cast_fp16")]; + tensor var_12404_begin_0 = const()[name = tensor("op_12404_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_12404_end_0 = const()[name = tensor("op_12404_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_12404_end_mask_0 = const()[name = tensor("op_12404_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12404_cast_fp16 = slice_by_index(begin = var_12404_begin_0, end = var_12404_end_0, end_mask = var_12404_end_mask_0, x = q_59_cast_fp16)[name = tensor("op_12404_cast_fp16")]; + tensor var_12408_begin_0 = const()[name = tensor("op_12408_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_12408_end_0 = const()[name = tensor("op_12408_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_12408_end_mask_0 = const()[name = tensor("op_12408_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12408_cast_fp16 = slice_by_index(begin = var_12408_begin_0, end = var_12408_end_0, end_mask = var_12408_end_mask_0, x = q_59_cast_fp16)[name = tensor("op_12408_cast_fp16")]; + tensor var_12412_begin_0 = const()[name = tensor("op_12412_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_12412_end_0 = const()[name = tensor("op_12412_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_12412_end_mask_0 = const()[name = tensor("op_12412_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12412_cast_fp16 = slice_by_index(begin = var_12412_begin_0, end = var_12412_end_0, end_mask = var_12412_end_mask_0, x = q_59_cast_fp16)[name = tensor("op_12412_cast_fp16")]; + tensor var_12416_begin_0 = const()[name = tensor("op_12416_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_12416_end_0 = const()[name = tensor("op_12416_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_12416_end_mask_0 = const()[name = tensor("op_12416_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12416_cast_fp16 = slice_by_index(begin = var_12416_begin_0, end = var_12416_end_0, end_mask = var_12416_end_mask_0, x = q_59_cast_fp16)[name = tensor("op_12416_cast_fp16")]; + tensor var_12420_begin_0 = const()[name = tensor("op_12420_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_12420_end_0 = const()[name = tensor("op_12420_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_12420_end_mask_0 = const()[name = tensor("op_12420_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12420_cast_fp16 = slice_by_index(begin = var_12420_begin_0, end = var_12420_end_0, end_mask = var_12420_end_mask_0, x = q_59_cast_fp16)[name = tensor("op_12420_cast_fp16")]; + tensor var_12424_begin_0 = const()[name = tensor("op_12424_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_12424_end_0 = const()[name = tensor("op_12424_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_12424_end_mask_0 = const()[name = tensor("op_12424_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12424_cast_fp16 = slice_by_index(begin = var_12424_begin_0, end = var_12424_end_0, end_mask = var_12424_end_mask_0, x = q_59_cast_fp16)[name = tensor("op_12424_cast_fp16")]; + tensor var_12428_begin_0 = const()[name = tensor("op_12428_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_12428_end_0 = const()[name = tensor("op_12428_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_12428_end_mask_0 = const()[name = tensor("op_12428_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12428_cast_fp16 = slice_by_index(begin = var_12428_begin_0, end = var_12428_end_0, end_mask = var_12428_end_mask_0, x = q_59_cast_fp16)[name = tensor("op_12428_cast_fp16")]; + tensor var_12432_begin_0 = const()[name = tensor("op_12432_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_12432_end_0 = const()[name = tensor("op_12432_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_12432_end_mask_0 = const()[name = tensor("op_12432_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12432_cast_fp16 = slice_by_index(begin = var_12432_begin_0, end = var_12432_end_0, end_mask = var_12432_end_mask_0, x = q_59_cast_fp16)[name = tensor("op_12432_cast_fp16")]; + tensor var_12436_begin_0 = const()[name = tensor("op_12436_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_12436_end_0 = const()[name = tensor("op_12436_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_12436_end_mask_0 = const()[name = tensor("op_12436_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12436_cast_fp16 = slice_by_index(begin = var_12436_begin_0, end = var_12436_end_0, end_mask = var_12436_end_mask_0, x = q_59_cast_fp16)[name = tensor("op_12436_cast_fp16")]; + tensor var_12440_begin_0 = const()[name = tensor("op_12440_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_12440_end_0 = const()[name = tensor("op_12440_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_12440_end_mask_0 = const()[name = tensor("op_12440_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12440_cast_fp16 = slice_by_index(begin = var_12440_begin_0, end = var_12440_end_0, end_mask = var_12440_end_mask_0, x = q_59_cast_fp16)[name = tensor("op_12440_cast_fp16")]; + tensor var_12444_begin_0 = const()[name = tensor("op_12444_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_12444_end_0 = const()[name = tensor("op_12444_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_12444_end_mask_0 = const()[name = tensor("op_12444_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12444_cast_fp16 = slice_by_index(begin = var_12444_begin_0, end = var_12444_end_0, end_mask = var_12444_end_mask_0, x = q_59_cast_fp16)[name = tensor("op_12444_cast_fp16")]; + tensor var_12448_begin_0 = const()[name = tensor("op_12448_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_12448_end_0 = const()[name = tensor("op_12448_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_12448_end_mask_0 = const()[name = tensor("op_12448_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12448_cast_fp16 = slice_by_index(begin = var_12448_begin_0, end = var_12448_end_0, end_mask = var_12448_end_mask_0, x = q_59_cast_fp16)[name = tensor("op_12448_cast_fp16")]; + tensor var_12452_begin_0 = const()[name = tensor("op_12452_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_12452_end_0 = const()[name = tensor("op_12452_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_12452_end_mask_0 = const()[name = tensor("op_12452_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12452_cast_fp16 = slice_by_index(begin = var_12452_begin_0, end = var_12452_end_0, end_mask = var_12452_end_mask_0, x = q_59_cast_fp16)[name = tensor("op_12452_cast_fp16")]; + tensor var_12456_begin_0 = const()[name = tensor("op_12456_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_12456_end_0 = const()[name = tensor("op_12456_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_12456_end_mask_0 = const()[name = tensor("op_12456_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12456_cast_fp16 = slice_by_index(begin = var_12456_begin_0, end = var_12456_end_0, end_mask = var_12456_end_mask_0, x = q_59_cast_fp16)[name = tensor("op_12456_cast_fp16")]; + tensor var_12460_begin_0 = const()[name = tensor("op_12460_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_12460_end_0 = const()[name = tensor("op_12460_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_12460_end_mask_0 = const()[name = tensor("op_12460_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12460_cast_fp16 = slice_by_index(begin = var_12460_begin_0, end = var_12460_end_0, end_mask = var_12460_end_mask_0, x = q_59_cast_fp16)[name = tensor("op_12460_cast_fp16")]; + tensor var_12464_begin_0 = const()[name = tensor("op_12464_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_12464_end_0 = const()[name = tensor("op_12464_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_12464_end_mask_0 = const()[name = tensor("op_12464_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12464_cast_fp16 = slice_by_index(begin = var_12464_begin_0, end = var_12464_end_0, end_mask = var_12464_end_mask_0, x = q_59_cast_fp16)[name = tensor("op_12464_cast_fp16")]; + tensor var_12468_begin_0 = const()[name = tensor("op_12468_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_12468_end_0 = const()[name = tensor("op_12468_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_12468_end_mask_0 = const()[name = tensor("op_12468_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12468_cast_fp16 = slice_by_index(begin = var_12468_begin_0, end = var_12468_end_0, end_mask = var_12468_end_mask_0, x = q_59_cast_fp16)[name = tensor("op_12468_cast_fp16")]; + tensor var_12472_begin_0 = const()[name = tensor("op_12472_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_12472_end_0 = const()[name = tensor("op_12472_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_12472_end_mask_0 = const()[name = tensor("op_12472_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12472_cast_fp16 = slice_by_index(begin = var_12472_begin_0, end = var_12472_end_0, end_mask = var_12472_end_mask_0, x = q_59_cast_fp16)[name = tensor("op_12472_cast_fp16")]; + tensor var_12476_begin_0 = const()[name = tensor("op_12476_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_12476_end_0 = const()[name = tensor("op_12476_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_12476_end_mask_0 = const()[name = tensor("op_12476_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12476_cast_fp16 = slice_by_index(begin = var_12476_begin_0, end = var_12476_end_0, end_mask = var_12476_end_mask_0, x = q_59_cast_fp16)[name = tensor("op_12476_cast_fp16")]; + tensor k_119_perm_0 = const()[name = tensor("k_119_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_12483_begin_0 = const()[name = tensor("op_12483_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12483_end_0 = const()[name = tensor("op_12483_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_12483_end_mask_0 = const()[name = tensor("op_12483_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_119_cast_fp16 = transpose(perm = k_119_perm_0, x = k_117_cast_fp16)[name = tensor("transpose_110")]; + tensor var_12483_cast_fp16 = slice_by_index(begin = var_12483_begin_0, end = var_12483_end_0, end_mask = var_12483_end_mask_0, x = k_119_cast_fp16)[name = tensor("op_12483_cast_fp16")]; + tensor var_12487_begin_0 = const()[name = tensor("op_12487_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_12487_end_0 = const()[name = tensor("op_12487_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_12487_end_mask_0 = const()[name = tensor("op_12487_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12487_cast_fp16 = slice_by_index(begin = var_12487_begin_0, end = var_12487_end_0, end_mask = var_12487_end_mask_0, x = k_119_cast_fp16)[name = tensor("op_12487_cast_fp16")]; + tensor var_12491_begin_0 = const()[name = tensor("op_12491_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_12491_end_0 = const()[name = tensor("op_12491_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_12491_end_mask_0 = const()[name = tensor("op_12491_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12491_cast_fp16 = slice_by_index(begin = var_12491_begin_0, end = var_12491_end_0, end_mask = var_12491_end_mask_0, x = k_119_cast_fp16)[name = tensor("op_12491_cast_fp16")]; + tensor var_12495_begin_0 = const()[name = tensor("op_12495_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_12495_end_0 = const()[name = tensor("op_12495_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_12495_end_mask_0 = const()[name = tensor("op_12495_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12495_cast_fp16 = slice_by_index(begin = var_12495_begin_0, end = var_12495_end_0, end_mask = var_12495_end_mask_0, x = k_119_cast_fp16)[name = tensor("op_12495_cast_fp16")]; + tensor var_12499_begin_0 = const()[name = tensor("op_12499_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_12499_end_0 = const()[name = tensor("op_12499_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_12499_end_mask_0 = const()[name = tensor("op_12499_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12499_cast_fp16 = slice_by_index(begin = var_12499_begin_0, end = var_12499_end_0, end_mask = var_12499_end_mask_0, x = k_119_cast_fp16)[name = tensor("op_12499_cast_fp16")]; + tensor var_12503_begin_0 = const()[name = tensor("op_12503_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_12503_end_0 = const()[name = tensor("op_12503_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_12503_end_mask_0 = const()[name = tensor("op_12503_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12503_cast_fp16 = slice_by_index(begin = var_12503_begin_0, end = var_12503_end_0, end_mask = var_12503_end_mask_0, x = k_119_cast_fp16)[name = tensor("op_12503_cast_fp16")]; + tensor var_12507_begin_0 = const()[name = tensor("op_12507_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_12507_end_0 = const()[name = tensor("op_12507_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_12507_end_mask_0 = const()[name = tensor("op_12507_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12507_cast_fp16 = slice_by_index(begin = var_12507_begin_0, end = var_12507_end_0, end_mask = var_12507_end_mask_0, x = k_119_cast_fp16)[name = tensor("op_12507_cast_fp16")]; + tensor var_12511_begin_0 = const()[name = tensor("op_12511_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_12511_end_0 = const()[name = tensor("op_12511_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_12511_end_mask_0 = const()[name = tensor("op_12511_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12511_cast_fp16 = slice_by_index(begin = var_12511_begin_0, end = var_12511_end_0, end_mask = var_12511_end_mask_0, x = k_119_cast_fp16)[name = tensor("op_12511_cast_fp16")]; + tensor var_12515_begin_0 = const()[name = tensor("op_12515_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_12515_end_0 = const()[name = tensor("op_12515_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_12515_end_mask_0 = const()[name = tensor("op_12515_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12515_cast_fp16 = slice_by_index(begin = var_12515_begin_0, end = var_12515_end_0, end_mask = var_12515_end_mask_0, x = k_119_cast_fp16)[name = tensor("op_12515_cast_fp16")]; + tensor var_12519_begin_0 = const()[name = tensor("op_12519_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_12519_end_0 = const()[name = tensor("op_12519_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_12519_end_mask_0 = const()[name = tensor("op_12519_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12519_cast_fp16 = slice_by_index(begin = var_12519_begin_0, end = var_12519_end_0, end_mask = var_12519_end_mask_0, x = k_119_cast_fp16)[name = tensor("op_12519_cast_fp16")]; + tensor var_12523_begin_0 = const()[name = tensor("op_12523_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_12523_end_0 = const()[name = tensor("op_12523_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_12523_end_mask_0 = const()[name = tensor("op_12523_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12523_cast_fp16 = slice_by_index(begin = var_12523_begin_0, end = var_12523_end_0, end_mask = var_12523_end_mask_0, x = k_119_cast_fp16)[name = tensor("op_12523_cast_fp16")]; + tensor var_12527_begin_0 = const()[name = tensor("op_12527_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_12527_end_0 = const()[name = tensor("op_12527_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_12527_end_mask_0 = const()[name = tensor("op_12527_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12527_cast_fp16 = slice_by_index(begin = var_12527_begin_0, end = var_12527_end_0, end_mask = var_12527_end_mask_0, x = k_119_cast_fp16)[name = tensor("op_12527_cast_fp16")]; + tensor var_12531_begin_0 = const()[name = tensor("op_12531_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_12531_end_0 = const()[name = tensor("op_12531_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_12531_end_mask_0 = const()[name = tensor("op_12531_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12531_cast_fp16 = slice_by_index(begin = var_12531_begin_0, end = var_12531_end_0, end_mask = var_12531_end_mask_0, x = k_119_cast_fp16)[name = tensor("op_12531_cast_fp16")]; + tensor var_12535_begin_0 = const()[name = tensor("op_12535_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_12535_end_0 = const()[name = tensor("op_12535_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_12535_end_mask_0 = const()[name = tensor("op_12535_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12535_cast_fp16 = slice_by_index(begin = var_12535_begin_0, end = var_12535_end_0, end_mask = var_12535_end_mask_0, x = k_119_cast_fp16)[name = tensor("op_12535_cast_fp16")]; + tensor var_12539_begin_0 = const()[name = tensor("op_12539_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_12539_end_0 = const()[name = tensor("op_12539_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_12539_end_mask_0 = const()[name = tensor("op_12539_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12539_cast_fp16 = slice_by_index(begin = var_12539_begin_0, end = var_12539_end_0, end_mask = var_12539_end_mask_0, x = k_119_cast_fp16)[name = tensor("op_12539_cast_fp16")]; + tensor var_12543_begin_0 = const()[name = tensor("op_12543_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_12543_end_0 = const()[name = tensor("op_12543_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_12543_end_mask_0 = const()[name = tensor("op_12543_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12543_cast_fp16 = slice_by_index(begin = var_12543_begin_0, end = var_12543_end_0, end_mask = var_12543_end_mask_0, x = k_119_cast_fp16)[name = tensor("op_12543_cast_fp16")]; + tensor var_12547_begin_0 = const()[name = tensor("op_12547_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_12547_end_0 = const()[name = tensor("op_12547_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_12547_end_mask_0 = const()[name = tensor("op_12547_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12547_cast_fp16 = slice_by_index(begin = var_12547_begin_0, end = var_12547_end_0, end_mask = var_12547_end_mask_0, x = k_119_cast_fp16)[name = tensor("op_12547_cast_fp16")]; + tensor var_12551_begin_0 = const()[name = tensor("op_12551_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_12551_end_0 = const()[name = tensor("op_12551_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_12551_end_mask_0 = const()[name = tensor("op_12551_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12551_cast_fp16 = slice_by_index(begin = var_12551_begin_0, end = var_12551_end_0, end_mask = var_12551_end_mask_0, x = k_119_cast_fp16)[name = tensor("op_12551_cast_fp16")]; + tensor var_12555_begin_0 = const()[name = tensor("op_12555_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_12555_end_0 = const()[name = tensor("op_12555_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_12555_end_mask_0 = const()[name = tensor("op_12555_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12555_cast_fp16 = slice_by_index(begin = var_12555_begin_0, end = var_12555_end_0, end_mask = var_12555_end_mask_0, x = k_119_cast_fp16)[name = tensor("op_12555_cast_fp16")]; + tensor var_12559_begin_0 = const()[name = tensor("op_12559_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_12559_end_0 = const()[name = tensor("op_12559_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_12559_end_mask_0 = const()[name = tensor("op_12559_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12559_cast_fp16 = slice_by_index(begin = var_12559_begin_0, end = var_12559_end_0, end_mask = var_12559_end_mask_0, x = k_119_cast_fp16)[name = tensor("op_12559_cast_fp16")]; + tensor var_12561_begin_0 = const()[name = tensor("op_12561_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12561_end_0 = const()[name = tensor("op_12561_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_12561_end_mask_0 = const()[name = tensor("op_12561_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12561_cast_fp16 = slice_by_index(begin = var_12561_begin_0, end = var_12561_end_0, end_mask = var_12561_end_mask_0, x = v_59_cast_fp16)[name = tensor("op_12561_cast_fp16")]; + tensor var_12565_begin_0 = const()[name = tensor("op_12565_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_12565_end_0 = const()[name = tensor("op_12565_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_12565_end_mask_0 = const()[name = tensor("op_12565_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12565_cast_fp16 = slice_by_index(begin = var_12565_begin_0, end = var_12565_end_0, end_mask = var_12565_end_mask_0, x = v_59_cast_fp16)[name = tensor("op_12565_cast_fp16")]; + tensor var_12569_begin_0 = const()[name = tensor("op_12569_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_12569_end_0 = const()[name = tensor("op_12569_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_12569_end_mask_0 = const()[name = tensor("op_12569_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12569_cast_fp16 = slice_by_index(begin = var_12569_begin_0, end = var_12569_end_0, end_mask = var_12569_end_mask_0, x = v_59_cast_fp16)[name = tensor("op_12569_cast_fp16")]; + tensor var_12573_begin_0 = const()[name = tensor("op_12573_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_12573_end_0 = const()[name = tensor("op_12573_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_12573_end_mask_0 = const()[name = tensor("op_12573_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12573_cast_fp16 = slice_by_index(begin = var_12573_begin_0, end = var_12573_end_0, end_mask = var_12573_end_mask_0, x = v_59_cast_fp16)[name = tensor("op_12573_cast_fp16")]; + tensor var_12577_begin_0 = const()[name = tensor("op_12577_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_12577_end_0 = const()[name = tensor("op_12577_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_12577_end_mask_0 = const()[name = tensor("op_12577_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12577_cast_fp16 = slice_by_index(begin = var_12577_begin_0, end = var_12577_end_0, end_mask = var_12577_end_mask_0, x = v_59_cast_fp16)[name = tensor("op_12577_cast_fp16")]; + tensor var_12581_begin_0 = const()[name = tensor("op_12581_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_12581_end_0 = const()[name = tensor("op_12581_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_12581_end_mask_0 = const()[name = tensor("op_12581_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12581_cast_fp16 = slice_by_index(begin = var_12581_begin_0, end = var_12581_end_0, end_mask = var_12581_end_mask_0, x = v_59_cast_fp16)[name = tensor("op_12581_cast_fp16")]; + tensor var_12585_begin_0 = const()[name = tensor("op_12585_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_12585_end_0 = const()[name = tensor("op_12585_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_12585_end_mask_0 = const()[name = tensor("op_12585_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12585_cast_fp16 = slice_by_index(begin = var_12585_begin_0, end = var_12585_end_0, end_mask = var_12585_end_mask_0, x = v_59_cast_fp16)[name = tensor("op_12585_cast_fp16")]; + tensor var_12589_begin_0 = const()[name = tensor("op_12589_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_12589_end_0 = const()[name = tensor("op_12589_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_12589_end_mask_0 = const()[name = tensor("op_12589_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12589_cast_fp16 = slice_by_index(begin = var_12589_begin_0, end = var_12589_end_0, end_mask = var_12589_end_mask_0, x = v_59_cast_fp16)[name = tensor("op_12589_cast_fp16")]; + tensor var_12593_begin_0 = const()[name = tensor("op_12593_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_12593_end_0 = const()[name = tensor("op_12593_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_12593_end_mask_0 = const()[name = tensor("op_12593_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12593_cast_fp16 = slice_by_index(begin = var_12593_begin_0, end = var_12593_end_0, end_mask = var_12593_end_mask_0, x = v_59_cast_fp16)[name = tensor("op_12593_cast_fp16")]; + tensor var_12597_begin_0 = const()[name = tensor("op_12597_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_12597_end_0 = const()[name = tensor("op_12597_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_12597_end_mask_0 = const()[name = tensor("op_12597_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12597_cast_fp16 = slice_by_index(begin = var_12597_begin_0, end = var_12597_end_0, end_mask = var_12597_end_mask_0, x = v_59_cast_fp16)[name = tensor("op_12597_cast_fp16")]; + tensor var_12601_begin_0 = const()[name = tensor("op_12601_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_12601_end_0 = const()[name = tensor("op_12601_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_12601_end_mask_0 = const()[name = tensor("op_12601_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12601_cast_fp16 = slice_by_index(begin = var_12601_begin_0, end = var_12601_end_0, end_mask = var_12601_end_mask_0, x = v_59_cast_fp16)[name = tensor("op_12601_cast_fp16")]; + tensor var_12605_begin_0 = const()[name = tensor("op_12605_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_12605_end_0 = const()[name = tensor("op_12605_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_12605_end_mask_0 = const()[name = tensor("op_12605_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12605_cast_fp16 = slice_by_index(begin = var_12605_begin_0, end = var_12605_end_0, end_mask = var_12605_end_mask_0, x = v_59_cast_fp16)[name = tensor("op_12605_cast_fp16")]; + tensor var_12609_begin_0 = const()[name = tensor("op_12609_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_12609_end_0 = const()[name = tensor("op_12609_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_12609_end_mask_0 = const()[name = tensor("op_12609_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12609_cast_fp16 = slice_by_index(begin = var_12609_begin_0, end = var_12609_end_0, end_mask = var_12609_end_mask_0, x = v_59_cast_fp16)[name = tensor("op_12609_cast_fp16")]; + tensor var_12613_begin_0 = const()[name = tensor("op_12613_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_12613_end_0 = const()[name = tensor("op_12613_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_12613_end_mask_0 = const()[name = tensor("op_12613_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12613_cast_fp16 = slice_by_index(begin = var_12613_begin_0, end = var_12613_end_0, end_mask = var_12613_end_mask_0, x = v_59_cast_fp16)[name = tensor("op_12613_cast_fp16")]; + tensor var_12617_begin_0 = const()[name = tensor("op_12617_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_12617_end_0 = const()[name = tensor("op_12617_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_12617_end_mask_0 = const()[name = tensor("op_12617_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12617_cast_fp16 = slice_by_index(begin = var_12617_begin_0, end = var_12617_end_0, end_mask = var_12617_end_mask_0, x = v_59_cast_fp16)[name = tensor("op_12617_cast_fp16")]; + tensor var_12621_begin_0 = const()[name = tensor("op_12621_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_12621_end_0 = const()[name = tensor("op_12621_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_12621_end_mask_0 = const()[name = tensor("op_12621_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12621_cast_fp16 = slice_by_index(begin = var_12621_begin_0, end = var_12621_end_0, end_mask = var_12621_end_mask_0, x = v_59_cast_fp16)[name = tensor("op_12621_cast_fp16")]; + tensor var_12625_begin_0 = const()[name = tensor("op_12625_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_12625_end_0 = const()[name = tensor("op_12625_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_12625_end_mask_0 = const()[name = tensor("op_12625_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12625_cast_fp16 = slice_by_index(begin = var_12625_begin_0, end = var_12625_end_0, end_mask = var_12625_end_mask_0, x = v_59_cast_fp16)[name = tensor("op_12625_cast_fp16")]; + tensor var_12629_begin_0 = const()[name = tensor("op_12629_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_12629_end_0 = const()[name = tensor("op_12629_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_12629_end_mask_0 = const()[name = tensor("op_12629_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12629_cast_fp16 = slice_by_index(begin = var_12629_begin_0, end = var_12629_end_0, end_mask = var_12629_end_mask_0, x = v_59_cast_fp16)[name = tensor("op_12629_cast_fp16")]; + tensor var_12633_begin_0 = const()[name = tensor("op_12633_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_12633_end_0 = const()[name = tensor("op_12633_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_12633_end_mask_0 = const()[name = tensor("op_12633_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12633_cast_fp16 = slice_by_index(begin = var_12633_begin_0, end = var_12633_end_0, end_mask = var_12633_end_mask_0, x = v_59_cast_fp16)[name = tensor("op_12633_cast_fp16")]; + tensor var_12637_begin_0 = const()[name = tensor("op_12637_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_12637_end_0 = const()[name = tensor("op_12637_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_12637_end_mask_0 = const()[name = tensor("op_12637_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12637_cast_fp16 = slice_by_index(begin = var_12637_begin_0, end = var_12637_end_0, end_mask = var_12637_end_mask_0, x = v_59_cast_fp16)[name = tensor("op_12637_cast_fp16")]; + tensor var_12641_equation_0 = const()[name = tensor("op_12641_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12641_cast_fp16 = einsum(equation = var_12641_equation_0, values = (var_12483_cast_fp16, var_12400_cast_fp16))[name = tensor("op_12641_cast_fp16")]; + tensor var_12642_to_fp16 = const()[name = tensor("op_12642_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1001_cast_fp16 = mul(x = var_12641_cast_fp16, y = var_12642_to_fp16)[name = tensor("aw_1001_cast_fp16")]; + tensor var_12645_equation_0 = const()[name = tensor("op_12645_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12645_cast_fp16 = einsum(equation = var_12645_equation_0, values = (var_12487_cast_fp16, var_12404_cast_fp16))[name = tensor("op_12645_cast_fp16")]; + tensor var_12646_to_fp16 = const()[name = tensor("op_12646_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1003_cast_fp16 = mul(x = var_12645_cast_fp16, y = var_12646_to_fp16)[name = tensor("aw_1003_cast_fp16")]; + tensor var_12649_equation_0 = const()[name = tensor("op_12649_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12649_cast_fp16 = einsum(equation = var_12649_equation_0, values = (var_12491_cast_fp16, var_12408_cast_fp16))[name = tensor("op_12649_cast_fp16")]; + tensor var_12650_to_fp16 = const()[name = tensor("op_12650_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1005_cast_fp16 = mul(x = var_12649_cast_fp16, y = var_12650_to_fp16)[name = tensor("aw_1005_cast_fp16")]; + tensor var_12653_equation_0 = const()[name = tensor("op_12653_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12653_cast_fp16 = einsum(equation = var_12653_equation_0, values = (var_12495_cast_fp16, var_12412_cast_fp16))[name = tensor("op_12653_cast_fp16")]; + tensor var_12654_to_fp16 = const()[name = tensor("op_12654_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1007_cast_fp16 = mul(x = var_12653_cast_fp16, y = var_12654_to_fp16)[name = tensor("aw_1007_cast_fp16")]; + tensor var_12657_equation_0 = const()[name = tensor("op_12657_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12657_cast_fp16 = einsum(equation = var_12657_equation_0, values = (var_12499_cast_fp16, var_12416_cast_fp16))[name = tensor("op_12657_cast_fp16")]; + tensor var_12658_to_fp16 = const()[name = tensor("op_12658_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1009_cast_fp16 = mul(x = var_12657_cast_fp16, y = var_12658_to_fp16)[name = tensor("aw_1009_cast_fp16")]; + tensor var_12661_equation_0 = const()[name = tensor("op_12661_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12661_cast_fp16 = einsum(equation = var_12661_equation_0, values = (var_12503_cast_fp16, var_12420_cast_fp16))[name = tensor("op_12661_cast_fp16")]; + tensor var_12662_to_fp16 = const()[name = tensor("op_12662_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1011_cast_fp16 = mul(x = var_12661_cast_fp16, y = var_12662_to_fp16)[name = tensor("aw_1011_cast_fp16")]; + tensor var_12665_equation_0 = const()[name = tensor("op_12665_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12665_cast_fp16 = einsum(equation = var_12665_equation_0, values = (var_12507_cast_fp16, var_12424_cast_fp16))[name = tensor("op_12665_cast_fp16")]; + tensor var_12666_to_fp16 = const()[name = tensor("op_12666_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1013_cast_fp16 = mul(x = var_12665_cast_fp16, y = var_12666_to_fp16)[name = tensor("aw_1013_cast_fp16")]; + tensor var_12669_equation_0 = const()[name = tensor("op_12669_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12669_cast_fp16 = einsum(equation = var_12669_equation_0, values = (var_12511_cast_fp16, var_12428_cast_fp16))[name = tensor("op_12669_cast_fp16")]; + tensor var_12670_to_fp16 = const()[name = tensor("op_12670_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1015_cast_fp16 = mul(x = var_12669_cast_fp16, y = var_12670_to_fp16)[name = tensor("aw_1015_cast_fp16")]; + tensor var_12673_equation_0 = const()[name = tensor("op_12673_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12673_cast_fp16 = einsum(equation = var_12673_equation_0, values = (var_12515_cast_fp16, var_12432_cast_fp16))[name = tensor("op_12673_cast_fp16")]; + tensor var_12674_to_fp16 = const()[name = tensor("op_12674_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1017_cast_fp16 = mul(x = var_12673_cast_fp16, y = var_12674_to_fp16)[name = tensor("aw_1017_cast_fp16")]; + tensor var_12677_equation_0 = const()[name = tensor("op_12677_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12677_cast_fp16 = einsum(equation = var_12677_equation_0, values = (var_12519_cast_fp16, var_12436_cast_fp16))[name = tensor("op_12677_cast_fp16")]; + tensor var_12678_to_fp16 = const()[name = tensor("op_12678_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1019_cast_fp16 = mul(x = var_12677_cast_fp16, y = var_12678_to_fp16)[name = tensor("aw_1019_cast_fp16")]; + tensor var_12681_equation_0 = const()[name = tensor("op_12681_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12681_cast_fp16 = einsum(equation = var_12681_equation_0, values = (var_12523_cast_fp16, var_12440_cast_fp16))[name = tensor("op_12681_cast_fp16")]; + tensor var_12682_to_fp16 = const()[name = tensor("op_12682_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1021_cast_fp16 = mul(x = var_12681_cast_fp16, y = var_12682_to_fp16)[name = tensor("aw_1021_cast_fp16")]; + tensor var_12685_equation_0 = const()[name = tensor("op_12685_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12685_cast_fp16 = einsum(equation = var_12685_equation_0, values = (var_12527_cast_fp16, var_12444_cast_fp16))[name = tensor("op_12685_cast_fp16")]; + tensor var_12686_to_fp16 = const()[name = tensor("op_12686_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1023_cast_fp16 = mul(x = var_12685_cast_fp16, y = var_12686_to_fp16)[name = tensor("aw_1023_cast_fp16")]; + tensor var_12689_equation_0 = const()[name = tensor("op_12689_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12689_cast_fp16 = einsum(equation = var_12689_equation_0, values = (var_12531_cast_fp16, var_12448_cast_fp16))[name = tensor("op_12689_cast_fp16")]; + tensor var_12690_to_fp16 = const()[name = tensor("op_12690_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1025_cast_fp16 = mul(x = var_12689_cast_fp16, y = var_12690_to_fp16)[name = tensor("aw_1025_cast_fp16")]; + tensor var_12693_equation_0 = const()[name = tensor("op_12693_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12693_cast_fp16 = einsum(equation = var_12693_equation_0, values = (var_12535_cast_fp16, var_12452_cast_fp16))[name = tensor("op_12693_cast_fp16")]; + tensor var_12694_to_fp16 = const()[name = tensor("op_12694_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1027_cast_fp16 = mul(x = var_12693_cast_fp16, y = var_12694_to_fp16)[name = tensor("aw_1027_cast_fp16")]; + tensor var_12697_equation_0 = const()[name = tensor("op_12697_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12697_cast_fp16 = einsum(equation = var_12697_equation_0, values = (var_12539_cast_fp16, var_12456_cast_fp16))[name = tensor("op_12697_cast_fp16")]; + tensor var_12698_to_fp16 = const()[name = tensor("op_12698_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1029_cast_fp16 = mul(x = var_12697_cast_fp16, y = var_12698_to_fp16)[name = tensor("aw_1029_cast_fp16")]; + tensor var_12701_equation_0 = const()[name = tensor("op_12701_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12701_cast_fp16 = einsum(equation = var_12701_equation_0, values = (var_12543_cast_fp16, var_12460_cast_fp16))[name = tensor("op_12701_cast_fp16")]; + tensor var_12702_to_fp16 = const()[name = tensor("op_12702_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1031_cast_fp16 = mul(x = var_12701_cast_fp16, y = var_12702_to_fp16)[name = tensor("aw_1031_cast_fp16")]; + tensor var_12705_equation_0 = const()[name = tensor("op_12705_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12705_cast_fp16 = einsum(equation = var_12705_equation_0, values = (var_12547_cast_fp16, var_12464_cast_fp16))[name = tensor("op_12705_cast_fp16")]; + tensor var_12706_to_fp16 = const()[name = tensor("op_12706_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1033_cast_fp16 = mul(x = var_12705_cast_fp16, y = var_12706_to_fp16)[name = tensor("aw_1033_cast_fp16")]; + tensor var_12709_equation_0 = const()[name = tensor("op_12709_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12709_cast_fp16 = einsum(equation = var_12709_equation_0, values = (var_12551_cast_fp16, var_12468_cast_fp16))[name = tensor("op_12709_cast_fp16")]; + tensor var_12710_to_fp16 = const()[name = tensor("op_12710_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1035_cast_fp16 = mul(x = var_12709_cast_fp16, y = var_12710_to_fp16)[name = tensor("aw_1035_cast_fp16")]; + tensor var_12713_equation_0 = const()[name = tensor("op_12713_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12713_cast_fp16 = einsum(equation = var_12713_equation_0, values = (var_12555_cast_fp16, var_12472_cast_fp16))[name = tensor("op_12713_cast_fp16")]; + tensor var_12714_to_fp16 = const()[name = tensor("op_12714_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1037_cast_fp16 = mul(x = var_12713_cast_fp16, y = var_12714_to_fp16)[name = tensor("aw_1037_cast_fp16")]; + tensor var_12717_equation_0 = const()[name = tensor("op_12717_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12717_cast_fp16 = einsum(equation = var_12717_equation_0, values = (var_12559_cast_fp16, var_12476_cast_fp16))[name = tensor("op_12717_cast_fp16")]; + tensor var_12718_to_fp16 = const()[name = tensor("op_12718_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1039_cast_fp16 = mul(x = var_12717_cast_fp16, y = var_12718_to_fp16)[name = tensor("aw_1039_cast_fp16")]; + tensor var_12720_cast_fp16 = softmax(axis = var_2624, x = aw_1001_cast_fp16)[name = tensor("op_12720_cast_fp16")]; + tensor var_12721_cast_fp16 = softmax(axis = var_2624, x = aw_1003_cast_fp16)[name = tensor("op_12721_cast_fp16")]; + tensor var_12722_cast_fp16 = softmax(axis = var_2624, x = aw_1005_cast_fp16)[name = tensor("op_12722_cast_fp16")]; + tensor var_12723_cast_fp16 = softmax(axis = var_2624, x = aw_1007_cast_fp16)[name = tensor("op_12723_cast_fp16")]; + tensor var_12724_cast_fp16 = softmax(axis = var_2624, x = aw_1009_cast_fp16)[name = tensor("op_12724_cast_fp16")]; + tensor var_12725_cast_fp16 = softmax(axis = var_2624, x = aw_1011_cast_fp16)[name = tensor("op_12725_cast_fp16")]; + tensor var_12726_cast_fp16 = softmax(axis = var_2624, x = aw_1013_cast_fp16)[name = tensor("op_12726_cast_fp16")]; + tensor var_12727_cast_fp16 = softmax(axis = var_2624, x = aw_1015_cast_fp16)[name = tensor("op_12727_cast_fp16")]; + tensor var_12728_cast_fp16 = softmax(axis = var_2624, x = aw_1017_cast_fp16)[name = tensor("op_12728_cast_fp16")]; + tensor var_12729_cast_fp16 = softmax(axis = var_2624, x = aw_1019_cast_fp16)[name = tensor("op_12729_cast_fp16")]; + tensor var_12730_cast_fp16 = softmax(axis = var_2624, x = aw_1021_cast_fp16)[name = tensor("op_12730_cast_fp16")]; + tensor var_12731_cast_fp16 = softmax(axis = var_2624, x = aw_1023_cast_fp16)[name = tensor("op_12731_cast_fp16")]; + tensor var_12732_cast_fp16 = softmax(axis = var_2624, x = aw_1025_cast_fp16)[name = tensor("op_12732_cast_fp16")]; + tensor var_12733_cast_fp16 = softmax(axis = var_2624, x = aw_1027_cast_fp16)[name = tensor("op_12733_cast_fp16")]; + tensor var_12734_cast_fp16 = softmax(axis = var_2624, x = aw_1029_cast_fp16)[name = tensor("op_12734_cast_fp16")]; + tensor var_12735_cast_fp16 = softmax(axis = var_2624, x = aw_1031_cast_fp16)[name = tensor("op_12735_cast_fp16")]; + tensor var_12736_cast_fp16 = softmax(axis = var_2624, x = aw_1033_cast_fp16)[name = tensor("op_12736_cast_fp16")]; + tensor var_12737_cast_fp16 = softmax(axis = var_2624, x = aw_1035_cast_fp16)[name = tensor("op_12737_cast_fp16")]; + tensor var_12738_cast_fp16 = softmax(axis = var_2624, x = aw_1037_cast_fp16)[name = tensor("op_12738_cast_fp16")]; + tensor var_12739_cast_fp16 = softmax(axis = var_2624, x = aw_1039_cast_fp16)[name = tensor("op_12739_cast_fp16")]; + tensor var_12741_equation_0 = const()[name = tensor("op_12741_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12741_cast_fp16 = einsum(equation = var_12741_equation_0, values = (var_12561_cast_fp16, var_12720_cast_fp16))[name = tensor("op_12741_cast_fp16")]; + tensor var_12743_equation_0 = const()[name = tensor("op_12743_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12743_cast_fp16 = einsum(equation = var_12743_equation_0, values = (var_12565_cast_fp16, var_12721_cast_fp16))[name = tensor("op_12743_cast_fp16")]; + tensor var_12745_equation_0 = const()[name = tensor("op_12745_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12745_cast_fp16 = einsum(equation = var_12745_equation_0, values = (var_12569_cast_fp16, var_12722_cast_fp16))[name = tensor("op_12745_cast_fp16")]; + tensor var_12747_equation_0 = const()[name = tensor("op_12747_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12747_cast_fp16 = einsum(equation = var_12747_equation_0, values = (var_12573_cast_fp16, var_12723_cast_fp16))[name = tensor("op_12747_cast_fp16")]; + tensor var_12749_equation_0 = const()[name = tensor("op_12749_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12749_cast_fp16 = einsum(equation = var_12749_equation_0, values = (var_12577_cast_fp16, var_12724_cast_fp16))[name = tensor("op_12749_cast_fp16")]; + tensor var_12751_equation_0 = const()[name = tensor("op_12751_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12751_cast_fp16 = einsum(equation = var_12751_equation_0, values = (var_12581_cast_fp16, var_12725_cast_fp16))[name = tensor("op_12751_cast_fp16")]; + tensor var_12753_equation_0 = const()[name = tensor("op_12753_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12753_cast_fp16 = einsum(equation = var_12753_equation_0, values = (var_12585_cast_fp16, var_12726_cast_fp16))[name = tensor("op_12753_cast_fp16")]; + tensor var_12755_equation_0 = const()[name = tensor("op_12755_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12755_cast_fp16 = einsum(equation = var_12755_equation_0, values = (var_12589_cast_fp16, var_12727_cast_fp16))[name = tensor("op_12755_cast_fp16")]; + tensor var_12757_equation_0 = const()[name = tensor("op_12757_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12757_cast_fp16 = einsum(equation = var_12757_equation_0, values = (var_12593_cast_fp16, var_12728_cast_fp16))[name = tensor("op_12757_cast_fp16")]; + tensor var_12759_equation_0 = const()[name = tensor("op_12759_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12759_cast_fp16 = einsum(equation = var_12759_equation_0, values = (var_12597_cast_fp16, var_12729_cast_fp16))[name = tensor("op_12759_cast_fp16")]; + tensor var_12761_equation_0 = const()[name = tensor("op_12761_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12761_cast_fp16 = einsum(equation = var_12761_equation_0, values = (var_12601_cast_fp16, var_12730_cast_fp16))[name = tensor("op_12761_cast_fp16")]; + tensor var_12763_equation_0 = const()[name = tensor("op_12763_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12763_cast_fp16 = einsum(equation = var_12763_equation_0, values = (var_12605_cast_fp16, var_12731_cast_fp16))[name = tensor("op_12763_cast_fp16")]; + tensor var_12765_equation_0 = const()[name = tensor("op_12765_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12765_cast_fp16 = einsum(equation = var_12765_equation_0, values = (var_12609_cast_fp16, var_12732_cast_fp16))[name = tensor("op_12765_cast_fp16")]; + tensor var_12767_equation_0 = const()[name = tensor("op_12767_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12767_cast_fp16 = einsum(equation = var_12767_equation_0, values = (var_12613_cast_fp16, var_12733_cast_fp16))[name = tensor("op_12767_cast_fp16")]; + tensor var_12769_equation_0 = const()[name = tensor("op_12769_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12769_cast_fp16 = einsum(equation = var_12769_equation_0, values = (var_12617_cast_fp16, var_12734_cast_fp16))[name = tensor("op_12769_cast_fp16")]; + tensor var_12771_equation_0 = const()[name = tensor("op_12771_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12771_cast_fp16 = einsum(equation = var_12771_equation_0, values = (var_12621_cast_fp16, var_12735_cast_fp16))[name = tensor("op_12771_cast_fp16")]; + tensor var_12773_equation_0 = const()[name = tensor("op_12773_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12773_cast_fp16 = einsum(equation = var_12773_equation_0, values = (var_12625_cast_fp16, var_12736_cast_fp16))[name = tensor("op_12773_cast_fp16")]; + tensor var_12775_equation_0 = const()[name = tensor("op_12775_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12775_cast_fp16 = einsum(equation = var_12775_equation_0, values = (var_12629_cast_fp16, var_12737_cast_fp16))[name = tensor("op_12775_cast_fp16")]; + tensor var_12777_equation_0 = const()[name = tensor("op_12777_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12777_cast_fp16 = einsum(equation = var_12777_equation_0, values = (var_12633_cast_fp16, var_12738_cast_fp16))[name = tensor("op_12777_cast_fp16")]; + tensor var_12779_equation_0 = const()[name = tensor("op_12779_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12779_cast_fp16 = einsum(equation = var_12779_equation_0, values = (var_12637_cast_fp16, var_12739_cast_fp16))[name = tensor("op_12779_cast_fp16")]; + tensor input_231_interleave_0 = const()[name = tensor("input_231_interleave_0"), val = tensor(false)]; + tensor input_231_cast_fp16 = concat(axis = var_2624, interleave = input_231_interleave_0, values = (var_12741_cast_fp16, var_12743_cast_fp16, var_12745_cast_fp16, var_12747_cast_fp16, var_12749_cast_fp16, var_12751_cast_fp16, var_12753_cast_fp16, var_12755_cast_fp16, var_12757_cast_fp16, var_12759_cast_fp16, var_12761_cast_fp16, var_12763_cast_fp16, var_12765_cast_fp16, var_12767_cast_fp16, var_12769_cast_fp16, var_12771_cast_fp16, var_12773_cast_fp16, var_12775_cast_fp16, var_12777_cast_fp16, var_12779_cast_fp16))[name = tensor("input_231_cast_fp16")]; + tensor var_12785 = const()[name = tensor("op_12785"), val = tensor([1, 1])]; + tensor var_12787 = const()[name = tensor("op_12787"), val = tensor([1, 1])]; + tensor var_12789_pad_type_0 = const()[name = tensor("op_12789_pad_type_0"), val = tensor("custom")]; + tensor var_12789_pad_0 = const()[name = tensor("op_12789_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(989020800)))]; + tensor down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(992297664)))]; + tensor var_12789_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_12787, groups = var_2624, pad = var_12789_pad_0, pad_type = var_12789_pad_type_0, strides = var_12785, weight = down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16, x = input_231_cast_fp16)[name = tensor("op_12789_cast_fp16")]; + tensor inputs_89_cast_fp16 = add(x = var_12789_cast_fp16, y = inputs_87_cast_fp16)[name = tensor("inputs_89_cast_fp16")]; + tensor input_233_axes_0 = const()[name = tensor("input_233_axes_0"), val = tensor([1])]; + tensor input_233_gamma_0_to_fp16 = const()[name = tensor("input_233_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(992300288)))]; + tensor input_233_beta_0_to_fp16 = const()[name = tensor("input_233_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(992302912)))]; + tensor var_12799_to_fp16 = const()[name = tensor("op_12799_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_233_cast_fp16 = layer_norm(axes = input_233_axes_0, beta = input_233_beta_0_to_fp16, epsilon = var_12799_to_fp16, gamma = input_233_gamma_0_to_fp16, x = inputs_89_cast_fp16)[name = tensor("input_233_cast_fp16")]; + tensor var_12815 = const()[name = tensor("op_12815"), val = tensor([1, 1])]; + tensor var_12817 = const()[name = tensor("op_12817"), val = tensor([1, 1])]; + tensor var_12819_pad_type_0 = const()[name = tensor("op_12819_pad_type_0"), val = tensor("custom")]; + tensor var_12819_pad_0 = const()[name = tensor("op_12819_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(992305536)))]; + tensor down_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1018520000)))]; + tensor var_12819_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_12817, groups = var_2624, pad = var_12819_pad_0, pad_type = var_12819_pad_type_0, strides = var_12815, weight = down_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16, x = input_233_cast_fp16)[name = tensor("op_12819_cast_fp16")]; + tensor var_12820_split_sizes_0 = const()[name = tensor("op_12820_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_12820_axis_0 = const()[name = tensor("op_12820_axis_0"), val = tensor(1)]; + tensor var_12820_cast_fp16_0, tensor var_12820_cast_fp16_1 = split(axis = var_12820_axis_0, split_sizes = var_12820_split_sizes_0, x = var_12819_cast_fp16)[name = tensor("op_12820_cast_fp16")]; + tensor var_12822_mode_0 = const()[name = tensor("op_12822_mode_0"), val = tensor("EXACT")]; + tensor var_12822_cast_fp16 = gelu(mode = var_12822_mode_0, x = var_12820_cast_fp16_1)[name = tensor("op_12822_cast_fp16")]; + tensor input_235_cast_fp16 = mul(x = var_12820_cast_fp16_0, y = var_12822_cast_fp16)[name = tensor("input_235_cast_fp16")]; + tensor var_12826 = const()[name = tensor("op_12826"), val = tensor([1, 1])]; + tensor var_12828 = const()[name = tensor("op_12828"), val = tensor([1, 1])]; + tensor var_12830_pad_type_0 = const()[name = tensor("op_12830_pad_type_0"), val = tensor("custom")]; + tensor var_12830_pad_0 = const()[name = tensor("op_12830_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1018540544)))]; + tensor down_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1031647808)))]; + tensor var_12830_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_12828, groups = var_2624, pad = var_12830_pad_0, pad_type = var_12830_pad_type_0, strides = var_12826, weight = down_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16, x = input_235_cast_fp16)[name = tensor("op_12830_cast_fp16")]; + tensor inputs_91_cast_fp16 = add(x = var_12830_cast_fp16, y = inputs_89_cast_fp16)[name = tensor("inputs_91_cast_fp16")]; + tensor hidden_states_143_axes_0 = const()[name = tensor("hidden_states_143_axes_0"), val = tensor([1])]; + tensor hidden_states_143_gamma_0_to_fp16 = const()[name = tensor("hidden_states_143_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1031650432)))]; + tensor hidden_states_143_beta_0_to_fp16 = const()[name = tensor("hidden_states_143_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1031653056)))]; + tensor var_12846_to_fp16 = const()[name = tensor("op_12846_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_143_cast_fp16 = layer_norm(axes = hidden_states_143_axes_0, beta = hidden_states_143_beta_0_to_fp16, epsilon = var_12846_to_fp16, gamma = hidden_states_143_gamma_0_to_fp16, x = inputs_91_cast_fp16)[name = tensor("hidden_states_143_cast_fp16")]; + tensor var_12861 = const()[name = tensor("op_12861"), val = tensor([1, 1])]; + tensor var_12863 = const()[name = tensor("op_12863"), val = tensor([1, 1])]; + tensor q_61_pad_type_0 = const()[name = tensor("q_61_pad_type_0"), val = tensor("custom")]; + tensor q_61_pad_0 = const()[name = tensor("q_61_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_1_attn1_to_q_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_1_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1031655680)))]; + tensor q_61_cast_fp16 = conv(dilations = var_12863, groups = var_2624, pad = q_61_pad_0, pad_type = q_61_pad_type_0, strides = var_12861, weight = down_blocks_2_attentions_1_transformer_blocks_1_attn1_to_q_weight_to_fp16, x = hidden_states_143_cast_fp16)[name = tensor("q_61_cast_fp16")]; + tensor var_12867 = const()[name = tensor("op_12867"), val = tensor([1, 1])]; + tensor var_12869 = const()[name = tensor("op_12869"), val = tensor([1, 1])]; + tensor k_121_pad_type_0 = const()[name = tensor("k_121_pad_type_0"), val = tensor("custom")]; + tensor k_121_pad_0 = const()[name = tensor("k_121_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_1_attn1_to_k_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_1_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1034932544)))]; + tensor k_121_cast_fp16 = conv(dilations = var_12869, groups = var_2624, pad = k_121_pad_0, pad_type = k_121_pad_type_0, strides = var_12867, weight = down_blocks_2_attentions_1_transformer_blocks_1_attn1_to_k_weight_to_fp16, x = hidden_states_143_cast_fp16)[name = tensor("k_121_cast_fp16")]; + tensor var_12873 = const()[name = tensor("op_12873"), val = tensor([1, 1])]; + tensor var_12875 = const()[name = tensor("op_12875"), val = tensor([1, 1])]; + tensor v_61_pad_type_0 = const()[name = tensor("v_61_pad_type_0"), val = tensor("custom")]; + tensor v_61_pad_0 = const()[name = tensor("v_61_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_1_attn1_to_v_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_1_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1038209408)))]; + tensor v_61_cast_fp16 = conv(dilations = var_12875, groups = var_2624, pad = v_61_pad_0, pad_type = v_61_pad_type_0, strides = var_12873, weight = down_blocks_2_attentions_1_transformer_blocks_1_attn1_to_v_weight_to_fp16, x = hidden_states_143_cast_fp16)[name = tensor("v_61_cast_fp16")]; + tensor var_12879_begin_0 = const()[name = tensor("op_12879_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12879_end_0 = const()[name = tensor("op_12879_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_12879_end_mask_0 = const()[name = tensor("op_12879_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12879_cast_fp16 = slice_by_index(begin = var_12879_begin_0, end = var_12879_end_0, end_mask = var_12879_end_mask_0, x = q_61_cast_fp16)[name = tensor("op_12879_cast_fp16")]; + tensor var_12883_begin_0 = const()[name = tensor("op_12883_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_12883_end_0 = const()[name = tensor("op_12883_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_12883_end_mask_0 = const()[name = tensor("op_12883_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12883_cast_fp16 = slice_by_index(begin = var_12883_begin_0, end = var_12883_end_0, end_mask = var_12883_end_mask_0, x = q_61_cast_fp16)[name = tensor("op_12883_cast_fp16")]; + tensor var_12887_begin_0 = const()[name = tensor("op_12887_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_12887_end_0 = const()[name = tensor("op_12887_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_12887_end_mask_0 = const()[name = tensor("op_12887_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12887_cast_fp16 = slice_by_index(begin = var_12887_begin_0, end = var_12887_end_0, end_mask = var_12887_end_mask_0, x = q_61_cast_fp16)[name = tensor("op_12887_cast_fp16")]; + tensor var_12891_begin_0 = const()[name = tensor("op_12891_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_12891_end_0 = const()[name = tensor("op_12891_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_12891_end_mask_0 = const()[name = tensor("op_12891_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12891_cast_fp16 = slice_by_index(begin = var_12891_begin_0, end = var_12891_end_0, end_mask = var_12891_end_mask_0, x = q_61_cast_fp16)[name = tensor("op_12891_cast_fp16")]; + tensor var_12895_begin_0 = const()[name = tensor("op_12895_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_12895_end_0 = const()[name = tensor("op_12895_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_12895_end_mask_0 = const()[name = tensor("op_12895_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12895_cast_fp16 = slice_by_index(begin = var_12895_begin_0, end = var_12895_end_0, end_mask = var_12895_end_mask_0, x = q_61_cast_fp16)[name = tensor("op_12895_cast_fp16")]; + tensor var_12899_begin_0 = const()[name = tensor("op_12899_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_12899_end_0 = const()[name = tensor("op_12899_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_12899_end_mask_0 = const()[name = tensor("op_12899_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12899_cast_fp16 = slice_by_index(begin = var_12899_begin_0, end = var_12899_end_0, end_mask = var_12899_end_mask_0, x = q_61_cast_fp16)[name = tensor("op_12899_cast_fp16")]; + tensor var_12903_begin_0 = const()[name = tensor("op_12903_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_12903_end_0 = const()[name = tensor("op_12903_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_12903_end_mask_0 = const()[name = tensor("op_12903_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12903_cast_fp16 = slice_by_index(begin = var_12903_begin_0, end = var_12903_end_0, end_mask = var_12903_end_mask_0, x = q_61_cast_fp16)[name = tensor("op_12903_cast_fp16")]; + tensor var_12907_begin_0 = const()[name = tensor("op_12907_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_12907_end_0 = const()[name = tensor("op_12907_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_12907_end_mask_0 = const()[name = tensor("op_12907_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12907_cast_fp16 = slice_by_index(begin = var_12907_begin_0, end = var_12907_end_0, end_mask = var_12907_end_mask_0, x = q_61_cast_fp16)[name = tensor("op_12907_cast_fp16")]; + tensor var_12911_begin_0 = const()[name = tensor("op_12911_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_12911_end_0 = const()[name = tensor("op_12911_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_12911_end_mask_0 = const()[name = tensor("op_12911_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12911_cast_fp16 = slice_by_index(begin = var_12911_begin_0, end = var_12911_end_0, end_mask = var_12911_end_mask_0, x = q_61_cast_fp16)[name = tensor("op_12911_cast_fp16")]; + tensor var_12915_begin_0 = const()[name = tensor("op_12915_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_12915_end_0 = const()[name = tensor("op_12915_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_12915_end_mask_0 = const()[name = tensor("op_12915_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12915_cast_fp16 = slice_by_index(begin = var_12915_begin_0, end = var_12915_end_0, end_mask = var_12915_end_mask_0, x = q_61_cast_fp16)[name = tensor("op_12915_cast_fp16")]; + tensor var_12919_begin_0 = const()[name = tensor("op_12919_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_12919_end_0 = const()[name = tensor("op_12919_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_12919_end_mask_0 = const()[name = tensor("op_12919_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12919_cast_fp16 = slice_by_index(begin = var_12919_begin_0, end = var_12919_end_0, end_mask = var_12919_end_mask_0, x = q_61_cast_fp16)[name = tensor("op_12919_cast_fp16")]; + tensor var_12923_begin_0 = const()[name = tensor("op_12923_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_12923_end_0 = const()[name = tensor("op_12923_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_12923_end_mask_0 = const()[name = tensor("op_12923_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12923_cast_fp16 = slice_by_index(begin = var_12923_begin_0, end = var_12923_end_0, end_mask = var_12923_end_mask_0, x = q_61_cast_fp16)[name = tensor("op_12923_cast_fp16")]; + tensor var_12927_begin_0 = const()[name = tensor("op_12927_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_12927_end_0 = const()[name = tensor("op_12927_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_12927_end_mask_0 = const()[name = tensor("op_12927_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12927_cast_fp16 = slice_by_index(begin = var_12927_begin_0, end = var_12927_end_0, end_mask = var_12927_end_mask_0, x = q_61_cast_fp16)[name = tensor("op_12927_cast_fp16")]; + tensor var_12931_begin_0 = const()[name = tensor("op_12931_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_12931_end_0 = const()[name = tensor("op_12931_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_12931_end_mask_0 = const()[name = tensor("op_12931_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12931_cast_fp16 = slice_by_index(begin = var_12931_begin_0, end = var_12931_end_0, end_mask = var_12931_end_mask_0, x = q_61_cast_fp16)[name = tensor("op_12931_cast_fp16")]; + tensor var_12935_begin_0 = const()[name = tensor("op_12935_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_12935_end_0 = const()[name = tensor("op_12935_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_12935_end_mask_0 = const()[name = tensor("op_12935_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12935_cast_fp16 = slice_by_index(begin = var_12935_begin_0, end = var_12935_end_0, end_mask = var_12935_end_mask_0, x = q_61_cast_fp16)[name = tensor("op_12935_cast_fp16")]; + tensor var_12939_begin_0 = const()[name = tensor("op_12939_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_12939_end_0 = const()[name = tensor("op_12939_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_12939_end_mask_0 = const()[name = tensor("op_12939_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12939_cast_fp16 = slice_by_index(begin = var_12939_begin_0, end = var_12939_end_0, end_mask = var_12939_end_mask_0, x = q_61_cast_fp16)[name = tensor("op_12939_cast_fp16")]; + tensor var_12943_begin_0 = const()[name = tensor("op_12943_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_12943_end_0 = const()[name = tensor("op_12943_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_12943_end_mask_0 = const()[name = tensor("op_12943_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12943_cast_fp16 = slice_by_index(begin = var_12943_begin_0, end = var_12943_end_0, end_mask = var_12943_end_mask_0, x = q_61_cast_fp16)[name = tensor("op_12943_cast_fp16")]; + tensor var_12947_begin_0 = const()[name = tensor("op_12947_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_12947_end_0 = const()[name = tensor("op_12947_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_12947_end_mask_0 = const()[name = tensor("op_12947_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12947_cast_fp16 = slice_by_index(begin = var_12947_begin_0, end = var_12947_end_0, end_mask = var_12947_end_mask_0, x = q_61_cast_fp16)[name = tensor("op_12947_cast_fp16")]; + tensor var_12951_begin_0 = const()[name = tensor("op_12951_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_12951_end_0 = const()[name = tensor("op_12951_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_12951_end_mask_0 = const()[name = tensor("op_12951_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12951_cast_fp16 = slice_by_index(begin = var_12951_begin_0, end = var_12951_end_0, end_mask = var_12951_end_mask_0, x = q_61_cast_fp16)[name = tensor("op_12951_cast_fp16")]; + tensor var_12955_begin_0 = const()[name = tensor("op_12955_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_12955_end_0 = const()[name = tensor("op_12955_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_12955_end_mask_0 = const()[name = tensor("op_12955_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12955_cast_fp16 = slice_by_index(begin = var_12955_begin_0, end = var_12955_end_0, end_mask = var_12955_end_mask_0, x = q_61_cast_fp16)[name = tensor("op_12955_cast_fp16")]; + tensor k_123_perm_0 = const()[name = tensor("k_123_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_12962_begin_0 = const()[name = tensor("op_12962_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12962_end_0 = const()[name = tensor("op_12962_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_12962_end_mask_0 = const()[name = tensor("op_12962_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_123_cast_fp16 = transpose(perm = k_123_perm_0, x = k_121_cast_fp16)[name = tensor("transpose_109")]; + tensor var_12962_cast_fp16 = slice_by_index(begin = var_12962_begin_0, end = var_12962_end_0, end_mask = var_12962_end_mask_0, x = k_123_cast_fp16)[name = tensor("op_12962_cast_fp16")]; + tensor var_12966_begin_0 = const()[name = tensor("op_12966_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_12966_end_0 = const()[name = tensor("op_12966_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_12966_end_mask_0 = const()[name = tensor("op_12966_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12966_cast_fp16 = slice_by_index(begin = var_12966_begin_0, end = var_12966_end_0, end_mask = var_12966_end_mask_0, x = k_123_cast_fp16)[name = tensor("op_12966_cast_fp16")]; + tensor var_12970_begin_0 = const()[name = tensor("op_12970_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_12970_end_0 = const()[name = tensor("op_12970_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_12970_end_mask_0 = const()[name = tensor("op_12970_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12970_cast_fp16 = slice_by_index(begin = var_12970_begin_0, end = var_12970_end_0, end_mask = var_12970_end_mask_0, x = k_123_cast_fp16)[name = tensor("op_12970_cast_fp16")]; + tensor var_12974_begin_0 = const()[name = tensor("op_12974_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_12974_end_0 = const()[name = tensor("op_12974_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_12974_end_mask_0 = const()[name = tensor("op_12974_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12974_cast_fp16 = slice_by_index(begin = var_12974_begin_0, end = var_12974_end_0, end_mask = var_12974_end_mask_0, x = k_123_cast_fp16)[name = tensor("op_12974_cast_fp16")]; + tensor var_12978_begin_0 = const()[name = tensor("op_12978_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_12978_end_0 = const()[name = tensor("op_12978_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_12978_end_mask_0 = const()[name = tensor("op_12978_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12978_cast_fp16 = slice_by_index(begin = var_12978_begin_0, end = var_12978_end_0, end_mask = var_12978_end_mask_0, x = k_123_cast_fp16)[name = tensor("op_12978_cast_fp16")]; + tensor var_12982_begin_0 = const()[name = tensor("op_12982_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_12982_end_0 = const()[name = tensor("op_12982_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_12982_end_mask_0 = const()[name = tensor("op_12982_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12982_cast_fp16 = slice_by_index(begin = var_12982_begin_0, end = var_12982_end_0, end_mask = var_12982_end_mask_0, x = k_123_cast_fp16)[name = tensor("op_12982_cast_fp16")]; + tensor var_12986_begin_0 = const()[name = tensor("op_12986_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_12986_end_0 = const()[name = tensor("op_12986_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_12986_end_mask_0 = const()[name = tensor("op_12986_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12986_cast_fp16 = slice_by_index(begin = var_12986_begin_0, end = var_12986_end_0, end_mask = var_12986_end_mask_0, x = k_123_cast_fp16)[name = tensor("op_12986_cast_fp16")]; + tensor var_12990_begin_0 = const()[name = tensor("op_12990_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_12990_end_0 = const()[name = tensor("op_12990_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_12990_end_mask_0 = const()[name = tensor("op_12990_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12990_cast_fp16 = slice_by_index(begin = var_12990_begin_0, end = var_12990_end_0, end_mask = var_12990_end_mask_0, x = k_123_cast_fp16)[name = tensor("op_12990_cast_fp16")]; + tensor var_12994_begin_0 = const()[name = tensor("op_12994_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_12994_end_0 = const()[name = tensor("op_12994_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_12994_end_mask_0 = const()[name = tensor("op_12994_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12994_cast_fp16 = slice_by_index(begin = var_12994_begin_0, end = var_12994_end_0, end_mask = var_12994_end_mask_0, x = k_123_cast_fp16)[name = tensor("op_12994_cast_fp16")]; + tensor var_12998_begin_0 = const()[name = tensor("op_12998_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_12998_end_0 = const()[name = tensor("op_12998_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_12998_end_mask_0 = const()[name = tensor("op_12998_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12998_cast_fp16 = slice_by_index(begin = var_12998_begin_0, end = var_12998_end_0, end_mask = var_12998_end_mask_0, x = k_123_cast_fp16)[name = tensor("op_12998_cast_fp16")]; + tensor var_13002_begin_0 = const()[name = tensor("op_13002_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_13002_end_0 = const()[name = tensor("op_13002_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_13002_end_mask_0 = const()[name = tensor("op_13002_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13002_cast_fp16 = slice_by_index(begin = var_13002_begin_0, end = var_13002_end_0, end_mask = var_13002_end_mask_0, x = k_123_cast_fp16)[name = tensor("op_13002_cast_fp16")]; + tensor var_13006_begin_0 = const()[name = tensor("op_13006_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_13006_end_0 = const()[name = tensor("op_13006_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_13006_end_mask_0 = const()[name = tensor("op_13006_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13006_cast_fp16 = slice_by_index(begin = var_13006_begin_0, end = var_13006_end_0, end_mask = var_13006_end_mask_0, x = k_123_cast_fp16)[name = tensor("op_13006_cast_fp16")]; + tensor var_13010_begin_0 = const()[name = tensor("op_13010_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_13010_end_0 = const()[name = tensor("op_13010_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_13010_end_mask_0 = const()[name = tensor("op_13010_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13010_cast_fp16 = slice_by_index(begin = var_13010_begin_0, end = var_13010_end_0, end_mask = var_13010_end_mask_0, x = k_123_cast_fp16)[name = tensor("op_13010_cast_fp16")]; + tensor var_13014_begin_0 = const()[name = tensor("op_13014_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_13014_end_0 = const()[name = tensor("op_13014_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_13014_end_mask_0 = const()[name = tensor("op_13014_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13014_cast_fp16 = slice_by_index(begin = var_13014_begin_0, end = var_13014_end_0, end_mask = var_13014_end_mask_0, x = k_123_cast_fp16)[name = tensor("op_13014_cast_fp16")]; + tensor var_13018_begin_0 = const()[name = tensor("op_13018_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_13018_end_0 = const()[name = tensor("op_13018_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_13018_end_mask_0 = const()[name = tensor("op_13018_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13018_cast_fp16 = slice_by_index(begin = var_13018_begin_0, end = var_13018_end_0, end_mask = var_13018_end_mask_0, x = k_123_cast_fp16)[name = tensor("op_13018_cast_fp16")]; + tensor var_13022_begin_0 = const()[name = tensor("op_13022_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_13022_end_0 = const()[name = tensor("op_13022_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_13022_end_mask_0 = const()[name = tensor("op_13022_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13022_cast_fp16 = slice_by_index(begin = var_13022_begin_0, end = var_13022_end_0, end_mask = var_13022_end_mask_0, x = k_123_cast_fp16)[name = tensor("op_13022_cast_fp16")]; + tensor var_13026_begin_0 = const()[name = tensor("op_13026_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_13026_end_0 = const()[name = tensor("op_13026_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_13026_end_mask_0 = const()[name = tensor("op_13026_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13026_cast_fp16 = slice_by_index(begin = var_13026_begin_0, end = var_13026_end_0, end_mask = var_13026_end_mask_0, x = k_123_cast_fp16)[name = tensor("op_13026_cast_fp16")]; + tensor var_13030_begin_0 = const()[name = tensor("op_13030_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_13030_end_0 = const()[name = tensor("op_13030_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_13030_end_mask_0 = const()[name = tensor("op_13030_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13030_cast_fp16 = slice_by_index(begin = var_13030_begin_0, end = var_13030_end_0, end_mask = var_13030_end_mask_0, x = k_123_cast_fp16)[name = tensor("op_13030_cast_fp16")]; + tensor var_13034_begin_0 = const()[name = tensor("op_13034_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_13034_end_0 = const()[name = tensor("op_13034_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_13034_end_mask_0 = const()[name = tensor("op_13034_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13034_cast_fp16 = slice_by_index(begin = var_13034_begin_0, end = var_13034_end_0, end_mask = var_13034_end_mask_0, x = k_123_cast_fp16)[name = tensor("op_13034_cast_fp16")]; + tensor var_13038_begin_0 = const()[name = tensor("op_13038_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_13038_end_0 = const()[name = tensor("op_13038_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_13038_end_mask_0 = const()[name = tensor("op_13038_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13038_cast_fp16 = slice_by_index(begin = var_13038_begin_0, end = var_13038_end_0, end_mask = var_13038_end_mask_0, x = k_123_cast_fp16)[name = tensor("op_13038_cast_fp16")]; + tensor var_13040_begin_0 = const()[name = tensor("op_13040_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_13040_end_0 = const()[name = tensor("op_13040_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_13040_end_mask_0 = const()[name = tensor("op_13040_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13040_cast_fp16 = slice_by_index(begin = var_13040_begin_0, end = var_13040_end_0, end_mask = var_13040_end_mask_0, x = v_61_cast_fp16)[name = tensor("op_13040_cast_fp16")]; + tensor var_13044_begin_0 = const()[name = tensor("op_13044_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_13044_end_0 = const()[name = tensor("op_13044_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_13044_end_mask_0 = const()[name = tensor("op_13044_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13044_cast_fp16 = slice_by_index(begin = var_13044_begin_0, end = var_13044_end_0, end_mask = var_13044_end_mask_0, x = v_61_cast_fp16)[name = tensor("op_13044_cast_fp16")]; + tensor var_13048_begin_0 = const()[name = tensor("op_13048_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_13048_end_0 = const()[name = tensor("op_13048_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_13048_end_mask_0 = const()[name = tensor("op_13048_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13048_cast_fp16 = slice_by_index(begin = var_13048_begin_0, end = var_13048_end_0, end_mask = var_13048_end_mask_0, x = v_61_cast_fp16)[name = tensor("op_13048_cast_fp16")]; + tensor var_13052_begin_0 = const()[name = tensor("op_13052_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_13052_end_0 = const()[name = tensor("op_13052_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_13052_end_mask_0 = const()[name = tensor("op_13052_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13052_cast_fp16 = slice_by_index(begin = var_13052_begin_0, end = var_13052_end_0, end_mask = var_13052_end_mask_0, x = v_61_cast_fp16)[name = tensor("op_13052_cast_fp16")]; + tensor var_13056_begin_0 = const()[name = tensor("op_13056_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_13056_end_0 = const()[name = tensor("op_13056_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_13056_end_mask_0 = const()[name = tensor("op_13056_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13056_cast_fp16 = slice_by_index(begin = var_13056_begin_0, end = var_13056_end_0, end_mask = var_13056_end_mask_0, x = v_61_cast_fp16)[name = tensor("op_13056_cast_fp16")]; + tensor var_13060_begin_0 = const()[name = tensor("op_13060_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_13060_end_0 = const()[name = tensor("op_13060_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_13060_end_mask_0 = const()[name = tensor("op_13060_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13060_cast_fp16 = slice_by_index(begin = var_13060_begin_0, end = var_13060_end_0, end_mask = var_13060_end_mask_0, x = v_61_cast_fp16)[name = tensor("op_13060_cast_fp16")]; + tensor var_13064_begin_0 = const()[name = tensor("op_13064_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_13064_end_0 = const()[name = tensor("op_13064_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_13064_end_mask_0 = const()[name = tensor("op_13064_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13064_cast_fp16 = slice_by_index(begin = var_13064_begin_0, end = var_13064_end_0, end_mask = var_13064_end_mask_0, x = v_61_cast_fp16)[name = tensor("op_13064_cast_fp16")]; + tensor var_13068_begin_0 = const()[name = tensor("op_13068_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_13068_end_0 = const()[name = tensor("op_13068_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_13068_end_mask_0 = const()[name = tensor("op_13068_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13068_cast_fp16 = slice_by_index(begin = var_13068_begin_0, end = var_13068_end_0, end_mask = var_13068_end_mask_0, x = v_61_cast_fp16)[name = tensor("op_13068_cast_fp16")]; + tensor var_13072_begin_0 = const()[name = tensor("op_13072_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_13072_end_0 = const()[name = tensor("op_13072_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_13072_end_mask_0 = const()[name = tensor("op_13072_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13072_cast_fp16 = slice_by_index(begin = var_13072_begin_0, end = var_13072_end_0, end_mask = var_13072_end_mask_0, x = v_61_cast_fp16)[name = tensor("op_13072_cast_fp16")]; + tensor var_13076_begin_0 = const()[name = tensor("op_13076_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_13076_end_0 = const()[name = tensor("op_13076_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_13076_end_mask_0 = const()[name = tensor("op_13076_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13076_cast_fp16 = slice_by_index(begin = var_13076_begin_0, end = var_13076_end_0, end_mask = var_13076_end_mask_0, x = v_61_cast_fp16)[name = tensor("op_13076_cast_fp16")]; + tensor var_13080_begin_0 = const()[name = tensor("op_13080_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_13080_end_0 = const()[name = tensor("op_13080_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_13080_end_mask_0 = const()[name = tensor("op_13080_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13080_cast_fp16 = slice_by_index(begin = var_13080_begin_0, end = var_13080_end_0, end_mask = var_13080_end_mask_0, x = v_61_cast_fp16)[name = tensor("op_13080_cast_fp16")]; + tensor var_13084_begin_0 = const()[name = tensor("op_13084_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_13084_end_0 = const()[name = tensor("op_13084_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_13084_end_mask_0 = const()[name = tensor("op_13084_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13084_cast_fp16 = slice_by_index(begin = var_13084_begin_0, end = var_13084_end_0, end_mask = var_13084_end_mask_0, x = v_61_cast_fp16)[name = tensor("op_13084_cast_fp16")]; + tensor var_13088_begin_0 = const()[name = tensor("op_13088_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_13088_end_0 = const()[name = tensor("op_13088_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_13088_end_mask_0 = const()[name = tensor("op_13088_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13088_cast_fp16 = slice_by_index(begin = var_13088_begin_0, end = var_13088_end_0, end_mask = var_13088_end_mask_0, x = v_61_cast_fp16)[name = tensor("op_13088_cast_fp16")]; + tensor var_13092_begin_0 = const()[name = tensor("op_13092_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_13092_end_0 = const()[name = tensor("op_13092_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_13092_end_mask_0 = const()[name = tensor("op_13092_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13092_cast_fp16 = slice_by_index(begin = var_13092_begin_0, end = var_13092_end_0, end_mask = var_13092_end_mask_0, x = v_61_cast_fp16)[name = tensor("op_13092_cast_fp16")]; + tensor var_13096_begin_0 = const()[name = tensor("op_13096_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_13096_end_0 = const()[name = tensor("op_13096_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_13096_end_mask_0 = const()[name = tensor("op_13096_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13096_cast_fp16 = slice_by_index(begin = var_13096_begin_0, end = var_13096_end_0, end_mask = var_13096_end_mask_0, x = v_61_cast_fp16)[name = tensor("op_13096_cast_fp16")]; + tensor var_13100_begin_0 = const()[name = tensor("op_13100_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_13100_end_0 = const()[name = tensor("op_13100_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_13100_end_mask_0 = const()[name = tensor("op_13100_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13100_cast_fp16 = slice_by_index(begin = var_13100_begin_0, end = var_13100_end_0, end_mask = var_13100_end_mask_0, x = v_61_cast_fp16)[name = tensor("op_13100_cast_fp16")]; + tensor var_13104_begin_0 = const()[name = tensor("op_13104_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_13104_end_0 = const()[name = tensor("op_13104_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_13104_end_mask_0 = const()[name = tensor("op_13104_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13104_cast_fp16 = slice_by_index(begin = var_13104_begin_0, end = var_13104_end_0, end_mask = var_13104_end_mask_0, x = v_61_cast_fp16)[name = tensor("op_13104_cast_fp16")]; + tensor var_13108_begin_0 = const()[name = tensor("op_13108_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_13108_end_0 = const()[name = tensor("op_13108_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_13108_end_mask_0 = const()[name = tensor("op_13108_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13108_cast_fp16 = slice_by_index(begin = var_13108_begin_0, end = var_13108_end_0, end_mask = var_13108_end_mask_0, x = v_61_cast_fp16)[name = tensor("op_13108_cast_fp16")]; + tensor var_13112_begin_0 = const()[name = tensor("op_13112_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_13112_end_0 = const()[name = tensor("op_13112_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_13112_end_mask_0 = const()[name = tensor("op_13112_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13112_cast_fp16 = slice_by_index(begin = var_13112_begin_0, end = var_13112_end_0, end_mask = var_13112_end_mask_0, x = v_61_cast_fp16)[name = tensor("op_13112_cast_fp16")]; + tensor var_13116_begin_0 = const()[name = tensor("op_13116_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_13116_end_0 = const()[name = tensor("op_13116_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_13116_end_mask_0 = const()[name = tensor("op_13116_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13116_cast_fp16 = slice_by_index(begin = var_13116_begin_0, end = var_13116_end_0, end_mask = var_13116_end_mask_0, x = v_61_cast_fp16)[name = tensor("op_13116_cast_fp16")]; + tensor var_13120_equation_0 = const()[name = tensor("op_13120_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13120_cast_fp16 = einsum(equation = var_13120_equation_0, values = (var_12962_cast_fp16, var_12879_cast_fp16))[name = tensor("op_13120_cast_fp16")]; + tensor var_13121_to_fp16 = const()[name = tensor("op_13121_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1041_cast_fp16 = mul(x = var_13120_cast_fp16, y = var_13121_to_fp16)[name = tensor("aw_1041_cast_fp16")]; + tensor var_13124_equation_0 = const()[name = tensor("op_13124_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13124_cast_fp16 = einsum(equation = var_13124_equation_0, values = (var_12966_cast_fp16, var_12883_cast_fp16))[name = tensor("op_13124_cast_fp16")]; + tensor var_13125_to_fp16 = const()[name = tensor("op_13125_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1043_cast_fp16 = mul(x = var_13124_cast_fp16, y = var_13125_to_fp16)[name = tensor("aw_1043_cast_fp16")]; + tensor var_13128_equation_0 = const()[name = tensor("op_13128_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13128_cast_fp16 = einsum(equation = var_13128_equation_0, values = (var_12970_cast_fp16, var_12887_cast_fp16))[name = tensor("op_13128_cast_fp16")]; + tensor var_13129_to_fp16 = const()[name = tensor("op_13129_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1045_cast_fp16 = mul(x = var_13128_cast_fp16, y = var_13129_to_fp16)[name = tensor("aw_1045_cast_fp16")]; + tensor var_13132_equation_0 = const()[name = tensor("op_13132_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13132_cast_fp16 = einsum(equation = var_13132_equation_0, values = (var_12974_cast_fp16, var_12891_cast_fp16))[name = tensor("op_13132_cast_fp16")]; + tensor var_13133_to_fp16 = const()[name = tensor("op_13133_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1047_cast_fp16 = mul(x = var_13132_cast_fp16, y = var_13133_to_fp16)[name = tensor("aw_1047_cast_fp16")]; + tensor var_13136_equation_0 = const()[name = tensor("op_13136_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13136_cast_fp16 = einsum(equation = var_13136_equation_0, values = (var_12978_cast_fp16, var_12895_cast_fp16))[name = tensor("op_13136_cast_fp16")]; + tensor var_13137_to_fp16 = const()[name = tensor("op_13137_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1049_cast_fp16 = mul(x = var_13136_cast_fp16, y = var_13137_to_fp16)[name = tensor("aw_1049_cast_fp16")]; + tensor var_13140_equation_0 = const()[name = tensor("op_13140_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13140_cast_fp16 = einsum(equation = var_13140_equation_0, values = (var_12982_cast_fp16, var_12899_cast_fp16))[name = tensor("op_13140_cast_fp16")]; + tensor var_13141_to_fp16 = const()[name = tensor("op_13141_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1051_cast_fp16 = mul(x = var_13140_cast_fp16, y = var_13141_to_fp16)[name = tensor("aw_1051_cast_fp16")]; + tensor var_13144_equation_0 = const()[name = tensor("op_13144_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13144_cast_fp16 = einsum(equation = var_13144_equation_0, values = (var_12986_cast_fp16, var_12903_cast_fp16))[name = tensor("op_13144_cast_fp16")]; + tensor var_13145_to_fp16 = const()[name = tensor("op_13145_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1053_cast_fp16 = mul(x = var_13144_cast_fp16, y = var_13145_to_fp16)[name = tensor("aw_1053_cast_fp16")]; + tensor var_13148_equation_0 = const()[name = tensor("op_13148_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13148_cast_fp16 = einsum(equation = var_13148_equation_0, values = (var_12990_cast_fp16, var_12907_cast_fp16))[name = tensor("op_13148_cast_fp16")]; + tensor var_13149_to_fp16 = const()[name = tensor("op_13149_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1055_cast_fp16 = mul(x = var_13148_cast_fp16, y = var_13149_to_fp16)[name = tensor("aw_1055_cast_fp16")]; + tensor var_13152_equation_0 = const()[name = tensor("op_13152_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13152_cast_fp16 = einsum(equation = var_13152_equation_0, values = (var_12994_cast_fp16, var_12911_cast_fp16))[name = tensor("op_13152_cast_fp16")]; + tensor var_13153_to_fp16 = const()[name = tensor("op_13153_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1057_cast_fp16 = mul(x = var_13152_cast_fp16, y = var_13153_to_fp16)[name = tensor("aw_1057_cast_fp16")]; + tensor var_13156_equation_0 = const()[name = tensor("op_13156_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13156_cast_fp16 = einsum(equation = var_13156_equation_0, values = (var_12998_cast_fp16, var_12915_cast_fp16))[name = tensor("op_13156_cast_fp16")]; + tensor var_13157_to_fp16 = const()[name = tensor("op_13157_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1059_cast_fp16 = mul(x = var_13156_cast_fp16, y = var_13157_to_fp16)[name = tensor("aw_1059_cast_fp16")]; + tensor var_13160_equation_0 = const()[name = tensor("op_13160_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13160_cast_fp16 = einsum(equation = var_13160_equation_0, values = (var_13002_cast_fp16, var_12919_cast_fp16))[name = tensor("op_13160_cast_fp16")]; + tensor var_13161_to_fp16 = const()[name = tensor("op_13161_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1061_cast_fp16 = mul(x = var_13160_cast_fp16, y = var_13161_to_fp16)[name = tensor("aw_1061_cast_fp16")]; + tensor var_13164_equation_0 = const()[name = tensor("op_13164_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13164_cast_fp16 = einsum(equation = var_13164_equation_0, values = (var_13006_cast_fp16, var_12923_cast_fp16))[name = tensor("op_13164_cast_fp16")]; + tensor var_13165_to_fp16 = const()[name = tensor("op_13165_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1063_cast_fp16 = mul(x = var_13164_cast_fp16, y = var_13165_to_fp16)[name = tensor("aw_1063_cast_fp16")]; + tensor var_13168_equation_0 = const()[name = tensor("op_13168_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13168_cast_fp16 = einsum(equation = var_13168_equation_0, values = (var_13010_cast_fp16, var_12927_cast_fp16))[name = tensor("op_13168_cast_fp16")]; + tensor var_13169_to_fp16 = const()[name = tensor("op_13169_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1065_cast_fp16 = mul(x = var_13168_cast_fp16, y = var_13169_to_fp16)[name = tensor("aw_1065_cast_fp16")]; + tensor var_13172_equation_0 = const()[name = tensor("op_13172_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13172_cast_fp16 = einsum(equation = var_13172_equation_0, values = (var_13014_cast_fp16, var_12931_cast_fp16))[name = tensor("op_13172_cast_fp16")]; + tensor var_13173_to_fp16 = const()[name = tensor("op_13173_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1067_cast_fp16 = mul(x = var_13172_cast_fp16, y = var_13173_to_fp16)[name = tensor("aw_1067_cast_fp16")]; + tensor var_13176_equation_0 = const()[name = tensor("op_13176_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13176_cast_fp16 = einsum(equation = var_13176_equation_0, values = (var_13018_cast_fp16, var_12935_cast_fp16))[name = tensor("op_13176_cast_fp16")]; + tensor var_13177_to_fp16 = const()[name = tensor("op_13177_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1069_cast_fp16 = mul(x = var_13176_cast_fp16, y = var_13177_to_fp16)[name = tensor("aw_1069_cast_fp16")]; + tensor var_13180_equation_0 = const()[name = tensor("op_13180_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13180_cast_fp16 = einsum(equation = var_13180_equation_0, values = (var_13022_cast_fp16, var_12939_cast_fp16))[name = tensor("op_13180_cast_fp16")]; + tensor var_13181_to_fp16 = const()[name = tensor("op_13181_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1071_cast_fp16 = mul(x = var_13180_cast_fp16, y = var_13181_to_fp16)[name = tensor("aw_1071_cast_fp16")]; + tensor var_13184_equation_0 = const()[name = tensor("op_13184_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13184_cast_fp16 = einsum(equation = var_13184_equation_0, values = (var_13026_cast_fp16, var_12943_cast_fp16))[name = tensor("op_13184_cast_fp16")]; + tensor var_13185_to_fp16 = const()[name = tensor("op_13185_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1073_cast_fp16 = mul(x = var_13184_cast_fp16, y = var_13185_to_fp16)[name = tensor("aw_1073_cast_fp16")]; + tensor var_13188_equation_0 = const()[name = tensor("op_13188_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13188_cast_fp16 = einsum(equation = var_13188_equation_0, values = (var_13030_cast_fp16, var_12947_cast_fp16))[name = tensor("op_13188_cast_fp16")]; + tensor var_13189_to_fp16 = const()[name = tensor("op_13189_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1075_cast_fp16 = mul(x = var_13188_cast_fp16, y = var_13189_to_fp16)[name = tensor("aw_1075_cast_fp16")]; + tensor var_13192_equation_0 = const()[name = tensor("op_13192_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13192_cast_fp16 = einsum(equation = var_13192_equation_0, values = (var_13034_cast_fp16, var_12951_cast_fp16))[name = tensor("op_13192_cast_fp16")]; + tensor var_13193_to_fp16 = const()[name = tensor("op_13193_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1077_cast_fp16 = mul(x = var_13192_cast_fp16, y = var_13193_to_fp16)[name = tensor("aw_1077_cast_fp16")]; + tensor var_13196_equation_0 = const()[name = tensor("op_13196_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13196_cast_fp16 = einsum(equation = var_13196_equation_0, values = (var_13038_cast_fp16, var_12955_cast_fp16))[name = tensor("op_13196_cast_fp16")]; + tensor var_13197_to_fp16 = const()[name = tensor("op_13197_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1079_cast_fp16 = mul(x = var_13196_cast_fp16, y = var_13197_to_fp16)[name = tensor("aw_1079_cast_fp16")]; + tensor var_13199_cast_fp16 = softmax(axis = var_2624, x = aw_1041_cast_fp16)[name = tensor("op_13199_cast_fp16")]; + tensor var_13200_cast_fp16 = softmax(axis = var_2624, x = aw_1043_cast_fp16)[name = tensor("op_13200_cast_fp16")]; + tensor var_13201_cast_fp16 = softmax(axis = var_2624, x = aw_1045_cast_fp16)[name = tensor("op_13201_cast_fp16")]; + tensor var_13202_cast_fp16 = softmax(axis = var_2624, x = aw_1047_cast_fp16)[name = tensor("op_13202_cast_fp16")]; + tensor var_13203_cast_fp16 = softmax(axis = var_2624, x = aw_1049_cast_fp16)[name = tensor("op_13203_cast_fp16")]; + tensor var_13204_cast_fp16 = softmax(axis = var_2624, x = aw_1051_cast_fp16)[name = tensor("op_13204_cast_fp16")]; + tensor var_13205_cast_fp16 = softmax(axis = var_2624, x = aw_1053_cast_fp16)[name = tensor("op_13205_cast_fp16")]; + tensor var_13206_cast_fp16 = softmax(axis = var_2624, x = aw_1055_cast_fp16)[name = tensor("op_13206_cast_fp16")]; + tensor var_13207_cast_fp16 = softmax(axis = var_2624, x = aw_1057_cast_fp16)[name = tensor("op_13207_cast_fp16")]; + tensor var_13208_cast_fp16 = softmax(axis = var_2624, x = aw_1059_cast_fp16)[name = tensor("op_13208_cast_fp16")]; + tensor var_13209_cast_fp16 = softmax(axis = var_2624, x = aw_1061_cast_fp16)[name = tensor("op_13209_cast_fp16")]; + tensor var_13210_cast_fp16 = softmax(axis = var_2624, x = aw_1063_cast_fp16)[name = tensor("op_13210_cast_fp16")]; + tensor var_13211_cast_fp16 = softmax(axis = var_2624, x = aw_1065_cast_fp16)[name = tensor("op_13211_cast_fp16")]; + tensor var_13212_cast_fp16 = softmax(axis = var_2624, x = aw_1067_cast_fp16)[name = tensor("op_13212_cast_fp16")]; + tensor var_13213_cast_fp16 = softmax(axis = var_2624, x = aw_1069_cast_fp16)[name = tensor("op_13213_cast_fp16")]; + tensor var_13214_cast_fp16 = softmax(axis = var_2624, x = aw_1071_cast_fp16)[name = tensor("op_13214_cast_fp16")]; + tensor var_13215_cast_fp16 = softmax(axis = var_2624, x = aw_1073_cast_fp16)[name = tensor("op_13215_cast_fp16")]; + tensor var_13216_cast_fp16 = softmax(axis = var_2624, x = aw_1075_cast_fp16)[name = tensor("op_13216_cast_fp16")]; + tensor var_13217_cast_fp16 = softmax(axis = var_2624, x = aw_1077_cast_fp16)[name = tensor("op_13217_cast_fp16")]; + tensor var_13218_cast_fp16 = softmax(axis = var_2624, x = aw_1079_cast_fp16)[name = tensor("op_13218_cast_fp16")]; + tensor var_13220_equation_0 = const()[name = tensor("op_13220_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13220_cast_fp16 = einsum(equation = var_13220_equation_0, values = (var_13040_cast_fp16, var_13199_cast_fp16))[name = tensor("op_13220_cast_fp16")]; + tensor var_13222_equation_0 = const()[name = tensor("op_13222_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13222_cast_fp16 = einsum(equation = var_13222_equation_0, values = (var_13044_cast_fp16, var_13200_cast_fp16))[name = tensor("op_13222_cast_fp16")]; + tensor var_13224_equation_0 = const()[name = tensor("op_13224_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13224_cast_fp16 = einsum(equation = var_13224_equation_0, values = (var_13048_cast_fp16, var_13201_cast_fp16))[name = tensor("op_13224_cast_fp16")]; + tensor var_13226_equation_0 = const()[name = tensor("op_13226_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13226_cast_fp16 = einsum(equation = var_13226_equation_0, values = (var_13052_cast_fp16, var_13202_cast_fp16))[name = tensor("op_13226_cast_fp16")]; + tensor var_13228_equation_0 = const()[name = tensor("op_13228_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13228_cast_fp16 = einsum(equation = var_13228_equation_0, values = (var_13056_cast_fp16, var_13203_cast_fp16))[name = tensor("op_13228_cast_fp16")]; + tensor var_13230_equation_0 = const()[name = tensor("op_13230_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13230_cast_fp16 = einsum(equation = var_13230_equation_0, values = (var_13060_cast_fp16, var_13204_cast_fp16))[name = tensor("op_13230_cast_fp16")]; + tensor var_13232_equation_0 = const()[name = tensor("op_13232_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13232_cast_fp16 = einsum(equation = var_13232_equation_0, values = (var_13064_cast_fp16, var_13205_cast_fp16))[name = tensor("op_13232_cast_fp16")]; + tensor var_13234_equation_0 = const()[name = tensor("op_13234_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13234_cast_fp16 = einsum(equation = var_13234_equation_0, values = (var_13068_cast_fp16, var_13206_cast_fp16))[name = tensor("op_13234_cast_fp16")]; + tensor var_13236_equation_0 = const()[name = tensor("op_13236_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13236_cast_fp16 = einsum(equation = var_13236_equation_0, values = (var_13072_cast_fp16, var_13207_cast_fp16))[name = tensor("op_13236_cast_fp16")]; + tensor var_13238_equation_0 = const()[name = tensor("op_13238_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13238_cast_fp16 = einsum(equation = var_13238_equation_0, values = (var_13076_cast_fp16, var_13208_cast_fp16))[name = tensor("op_13238_cast_fp16")]; + tensor var_13240_equation_0 = const()[name = tensor("op_13240_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13240_cast_fp16 = einsum(equation = var_13240_equation_0, values = (var_13080_cast_fp16, var_13209_cast_fp16))[name = tensor("op_13240_cast_fp16")]; + tensor var_13242_equation_0 = const()[name = tensor("op_13242_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13242_cast_fp16 = einsum(equation = var_13242_equation_0, values = (var_13084_cast_fp16, var_13210_cast_fp16))[name = tensor("op_13242_cast_fp16")]; + tensor var_13244_equation_0 = const()[name = tensor("op_13244_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13244_cast_fp16 = einsum(equation = var_13244_equation_0, values = (var_13088_cast_fp16, var_13211_cast_fp16))[name = tensor("op_13244_cast_fp16")]; + tensor var_13246_equation_0 = const()[name = tensor("op_13246_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13246_cast_fp16 = einsum(equation = var_13246_equation_0, values = (var_13092_cast_fp16, var_13212_cast_fp16))[name = tensor("op_13246_cast_fp16")]; + tensor var_13248_equation_0 = const()[name = tensor("op_13248_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13248_cast_fp16 = einsum(equation = var_13248_equation_0, values = (var_13096_cast_fp16, var_13213_cast_fp16))[name = tensor("op_13248_cast_fp16")]; + tensor var_13250_equation_0 = const()[name = tensor("op_13250_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13250_cast_fp16 = einsum(equation = var_13250_equation_0, values = (var_13100_cast_fp16, var_13214_cast_fp16))[name = tensor("op_13250_cast_fp16")]; + tensor var_13252_equation_0 = const()[name = tensor("op_13252_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13252_cast_fp16 = einsum(equation = var_13252_equation_0, values = (var_13104_cast_fp16, var_13215_cast_fp16))[name = tensor("op_13252_cast_fp16")]; + tensor var_13254_equation_0 = const()[name = tensor("op_13254_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13254_cast_fp16 = einsum(equation = var_13254_equation_0, values = (var_13108_cast_fp16, var_13216_cast_fp16))[name = tensor("op_13254_cast_fp16")]; + tensor var_13256_equation_0 = const()[name = tensor("op_13256_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13256_cast_fp16 = einsum(equation = var_13256_equation_0, values = (var_13112_cast_fp16, var_13217_cast_fp16))[name = tensor("op_13256_cast_fp16")]; + tensor var_13258_equation_0 = const()[name = tensor("op_13258_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13258_cast_fp16 = einsum(equation = var_13258_equation_0, values = (var_13116_cast_fp16, var_13218_cast_fp16))[name = tensor("op_13258_cast_fp16")]; + tensor input_237_interleave_0 = const()[name = tensor("input_237_interleave_0"), val = tensor(false)]; + tensor input_237_cast_fp16 = concat(axis = var_2624, interleave = input_237_interleave_0, values = (var_13220_cast_fp16, var_13222_cast_fp16, var_13224_cast_fp16, var_13226_cast_fp16, var_13228_cast_fp16, var_13230_cast_fp16, var_13232_cast_fp16, var_13234_cast_fp16, var_13236_cast_fp16, var_13238_cast_fp16, var_13240_cast_fp16, var_13242_cast_fp16, var_13244_cast_fp16, var_13246_cast_fp16, var_13248_cast_fp16, var_13250_cast_fp16, var_13252_cast_fp16, var_13254_cast_fp16, var_13256_cast_fp16, var_13258_cast_fp16))[name = tensor("input_237_cast_fp16")]; + tensor var_13264 = const()[name = tensor("op_13264"), val = tensor([1, 1])]; + tensor var_13266 = const()[name = tensor("op_13266"), val = tensor([1, 1])]; + tensor var_13268_pad_type_0 = const()[name = tensor("op_13268_pad_type_0"), val = tensor("custom")]; + tensor var_13268_pad_0 = const()[name = tensor("op_13268_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_1_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_1_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1041486272)))]; + tensor down_blocks_2_attentions_1_transformer_blocks_1_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_1_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1044763136)))]; + tensor var_13268_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_1_attn1_to_out_0_bias_to_fp16, dilations = var_13266, groups = var_2624, pad = var_13268_pad_0, pad_type = var_13268_pad_type_0, strides = var_13264, weight = down_blocks_2_attentions_1_transformer_blocks_1_attn1_to_out_0_weight_to_fp16, x = input_237_cast_fp16)[name = tensor("op_13268_cast_fp16")]; + tensor inputs_93_cast_fp16 = add(x = var_13268_cast_fp16, y = inputs_91_cast_fp16)[name = tensor("inputs_93_cast_fp16")]; + tensor hidden_states_145_axes_0 = const()[name = tensor("hidden_states_145_axes_0"), val = tensor([1])]; + tensor hidden_states_145_gamma_0_to_fp16 = const()[name = tensor("hidden_states_145_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1044765760)))]; + tensor hidden_states_145_beta_0_to_fp16 = const()[name = tensor("hidden_states_145_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1044768384)))]; + tensor var_13278_to_fp16 = const()[name = tensor("op_13278_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_145_cast_fp16 = layer_norm(axes = hidden_states_145_axes_0, beta = hidden_states_145_beta_0_to_fp16, epsilon = var_13278_to_fp16, gamma = hidden_states_145_gamma_0_to_fp16, x = inputs_93_cast_fp16)[name = tensor("hidden_states_145_cast_fp16")]; + tensor var_13293 = const()[name = tensor("op_13293"), val = tensor([1, 1])]; + tensor var_13295 = const()[name = tensor("op_13295"), val = tensor([1, 1])]; + tensor q_63_pad_type_0 = const()[name = tensor("q_63_pad_type_0"), val = tensor("custom")]; + tensor q_63_pad_0 = const()[name = tensor("q_63_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_1_attn2_to_q_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_1_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1044771008)))]; + tensor q_63_cast_fp16 = conv(dilations = var_13295, groups = var_2624, pad = q_63_pad_0, pad_type = q_63_pad_type_0, strides = var_13293, weight = down_blocks_2_attentions_1_transformer_blocks_1_attn2_to_q_weight_to_fp16, x = hidden_states_145_cast_fp16)[name = tensor("q_63_cast_fp16")]; + tensor var_13299 = const()[name = tensor("op_13299"), val = tensor([1, 1])]; + tensor var_13301 = const()[name = tensor("op_13301"), val = tensor([1, 1])]; + tensor k_125_pad_type_0 = const()[name = tensor("k_125_pad_type_0"), val = tensor("custom")]; + tensor k_125_pad_0 = const()[name = tensor("k_125_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_1_attn2_to_k_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_1_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1048047872)))]; + tensor k_125_cast_fp16 = conv(dilations = var_13301, groups = var_2624, pad = k_125_pad_0, pad_type = k_125_pad_type_0, strides = var_13299, weight = down_blocks_2_attentions_1_transformer_blocks_1_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_125_cast_fp16")]; + tensor var_13305 = const()[name = tensor("op_13305"), val = tensor([1, 1])]; + tensor var_13307 = const()[name = tensor("op_13307"), val = tensor([1, 1])]; + tensor v_63_pad_type_0 = const()[name = tensor("v_63_pad_type_0"), val = tensor("custom")]; + tensor v_63_pad_0 = const()[name = tensor("v_63_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_1_attn2_to_v_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_1_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1053290816)))]; + tensor v_63_cast_fp16 = conv(dilations = var_13307, groups = var_2624, pad = v_63_pad_0, pad_type = v_63_pad_type_0, strides = var_13305, weight = down_blocks_2_attentions_1_transformer_blocks_1_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_63_cast_fp16")]; + tensor var_13311_begin_0 = const()[name = tensor("op_13311_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_13311_end_0 = const()[name = tensor("op_13311_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_13311_end_mask_0 = const()[name = tensor("op_13311_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13311_cast_fp16 = slice_by_index(begin = var_13311_begin_0, end = var_13311_end_0, end_mask = var_13311_end_mask_0, x = q_63_cast_fp16)[name = tensor("op_13311_cast_fp16")]; + tensor var_13315_begin_0 = const()[name = tensor("op_13315_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_13315_end_0 = const()[name = tensor("op_13315_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_13315_end_mask_0 = const()[name = tensor("op_13315_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13315_cast_fp16 = slice_by_index(begin = var_13315_begin_0, end = var_13315_end_0, end_mask = var_13315_end_mask_0, x = q_63_cast_fp16)[name = tensor("op_13315_cast_fp16")]; + tensor var_13319_begin_0 = const()[name = tensor("op_13319_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_13319_end_0 = const()[name = tensor("op_13319_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_13319_end_mask_0 = const()[name = tensor("op_13319_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13319_cast_fp16 = slice_by_index(begin = var_13319_begin_0, end = var_13319_end_0, end_mask = var_13319_end_mask_0, x = q_63_cast_fp16)[name = tensor("op_13319_cast_fp16")]; + tensor var_13323_begin_0 = const()[name = tensor("op_13323_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_13323_end_0 = const()[name = tensor("op_13323_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_13323_end_mask_0 = const()[name = tensor("op_13323_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13323_cast_fp16 = slice_by_index(begin = var_13323_begin_0, end = var_13323_end_0, end_mask = var_13323_end_mask_0, x = q_63_cast_fp16)[name = tensor("op_13323_cast_fp16")]; + tensor var_13327_begin_0 = const()[name = tensor("op_13327_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_13327_end_0 = const()[name = tensor("op_13327_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_13327_end_mask_0 = const()[name = tensor("op_13327_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13327_cast_fp16 = slice_by_index(begin = var_13327_begin_0, end = var_13327_end_0, end_mask = var_13327_end_mask_0, x = q_63_cast_fp16)[name = tensor("op_13327_cast_fp16")]; + tensor var_13331_begin_0 = const()[name = tensor("op_13331_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_13331_end_0 = const()[name = tensor("op_13331_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_13331_end_mask_0 = const()[name = tensor("op_13331_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13331_cast_fp16 = slice_by_index(begin = var_13331_begin_0, end = var_13331_end_0, end_mask = var_13331_end_mask_0, x = q_63_cast_fp16)[name = tensor("op_13331_cast_fp16")]; + tensor var_13335_begin_0 = const()[name = tensor("op_13335_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_13335_end_0 = const()[name = tensor("op_13335_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_13335_end_mask_0 = const()[name = tensor("op_13335_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13335_cast_fp16 = slice_by_index(begin = var_13335_begin_0, end = var_13335_end_0, end_mask = var_13335_end_mask_0, x = q_63_cast_fp16)[name = tensor("op_13335_cast_fp16")]; + tensor var_13339_begin_0 = const()[name = tensor("op_13339_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_13339_end_0 = const()[name = tensor("op_13339_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_13339_end_mask_0 = const()[name = tensor("op_13339_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13339_cast_fp16 = slice_by_index(begin = var_13339_begin_0, end = var_13339_end_0, end_mask = var_13339_end_mask_0, x = q_63_cast_fp16)[name = tensor("op_13339_cast_fp16")]; + tensor var_13343_begin_0 = const()[name = tensor("op_13343_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_13343_end_0 = const()[name = tensor("op_13343_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_13343_end_mask_0 = const()[name = tensor("op_13343_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13343_cast_fp16 = slice_by_index(begin = var_13343_begin_0, end = var_13343_end_0, end_mask = var_13343_end_mask_0, x = q_63_cast_fp16)[name = tensor("op_13343_cast_fp16")]; + tensor var_13347_begin_0 = const()[name = tensor("op_13347_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_13347_end_0 = const()[name = tensor("op_13347_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_13347_end_mask_0 = const()[name = tensor("op_13347_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13347_cast_fp16 = slice_by_index(begin = var_13347_begin_0, end = var_13347_end_0, end_mask = var_13347_end_mask_0, x = q_63_cast_fp16)[name = tensor("op_13347_cast_fp16")]; + tensor var_13351_begin_0 = const()[name = tensor("op_13351_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_13351_end_0 = const()[name = tensor("op_13351_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_13351_end_mask_0 = const()[name = tensor("op_13351_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13351_cast_fp16 = slice_by_index(begin = var_13351_begin_0, end = var_13351_end_0, end_mask = var_13351_end_mask_0, x = q_63_cast_fp16)[name = tensor("op_13351_cast_fp16")]; + tensor var_13355_begin_0 = const()[name = tensor("op_13355_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_13355_end_0 = const()[name = tensor("op_13355_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_13355_end_mask_0 = const()[name = tensor("op_13355_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13355_cast_fp16 = slice_by_index(begin = var_13355_begin_0, end = var_13355_end_0, end_mask = var_13355_end_mask_0, x = q_63_cast_fp16)[name = tensor("op_13355_cast_fp16")]; + tensor var_13359_begin_0 = const()[name = tensor("op_13359_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_13359_end_0 = const()[name = tensor("op_13359_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_13359_end_mask_0 = const()[name = tensor("op_13359_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13359_cast_fp16 = slice_by_index(begin = var_13359_begin_0, end = var_13359_end_0, end_mask = var_13359_end_mask_0, x = q_63_cast_fp16)[name = tensor("op_13359_cast_fp16")]; + tensor var_13363_begin_0 = const()[name = tensor("op_13363_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_13363_end_0 = const()[name = tensor("op_13363_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_13363_end_mask_0 = const()[name = tensor("op_13363_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13363_cast_fp16 = slice_by_index(begin = var_13363_begin_0, end = var_13363_end_0, end_mask = var_13363_end_mask_0, x = q_63_cast_fp16)[name = tensor("op_13363_cast_fp16")]; + tensor var_13367_begin_0 = const()[name = tensor("op_13367_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_13367_end_0 = const()[name = tensor("op_13367_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_13367_end_mask_0 = const()[name = tensor("op_13367_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13367_cast_fp16 = slice_by_index(begin = var_13367_begin_0, end = var_13367_end_0, end_mask = var_13367_end_mask_0, x = q_63_cast_fp16)[name = tensor("op_13367_cast_fp16")]; + tensor var_13371_begin_0 = const()[name = tensor("op_13371_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_13371_end_0 = const()[name = tensor("op_13371_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_13371_end_mask_0 = const()[name = tensor("op_13371_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13371_cast_fp16 = slice_by_index(begin = var_13371_begin_0, end = var_13371_end_0, end_mask = var_13371_end_mask_0, x = q_63_cast_fp16)[name = tensor("op_13371_cast_fp16")]; + tensor var_13375_begin_0 = const()[name = tensor("op_13375_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_13375_end_0 = const()[name = tensor("op_13375_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_13375_end_mask_0 = const()[name = tensor("op_13375_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13375_cast_fp16 = slice_by_index(begin = var_13375_begin_0, end = var_13375_end_0, end_mask = var_13375_end_mask_0, x = q_63_cast_fp16)[name = tensor("op_13375_cast_fp16")]; + tensor var_13379_begin_0 = const()[name = tensor("op_13379_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_13379_end_0 = const()[name = tensor("op_13379_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_13379_end_mask_0 = const()[name = tensor("op_13379_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13379_cast_fp16 = slice_by_index(begin = var_13379_begin_0, end = var_13379_end_0, end_mask = var_13379_end_mask_0, x = q_63_cast_fp16)[name = tensor("op_13379_cast_fp16")]; + tensor var_13383_begin_0 = const()[name = tensor("op_13383_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_13383_end_0 = const()[name = tensor("op_13383_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_13383_end_mask_0 = const()[name = tensor("op_13383_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13383_cast_fp16 = slice_by_index(begin = var_13383_begin_0, end = var_13383_end_0, end_mask = var_13383_end_mask_0, x = q_63_cast_fp16)[name = tensor("op_13383_cast_fp16")]; + tensor var_13387_begin_0 = const()[name = tensor("op_13387_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_13387_end_0 = const()[name = tensor("op_13387_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_13387_end_mask_0 = const()[name = tensor("op_13387_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13387_cast_fp16 = slice_by_index(begin = var_13387_begin_0, end = var_13387_end_0, end_mask = var_13387_end_mask_0, x = q_63_cast_fp16)[name = tensor("op_13387_cast_fp16")]; + tensor k_127_perm_0 = const()[name = tensor("k_127_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_13394_begin_0 = const()[name = tensor("op_13394_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_13394_end_0 = const()[name = tensor("op_13394_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_13394_end_mask_0 = const()[name = tensor("op_13394_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_127_cast_fp16 = transpose(perm = k_127_perm_0, x = k_125_cast_fp16)[name = tensor("transpose_108")]; + tensor var_13394_cast_fp16 = slice_by_index(begin = var_13394_begin_0, end = var_13394_end_0, end_mask = var_13394_end_mask_0, x = k_127_cast_fp16)[name = tensor("op_13394_cast_fp16")]; + tensor var_13398_begin_0 = const()[name = tensor("op_13398_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_13398_end_0 = const()[name = tensor("op_13398_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_13398_end_mask_0 = const()[name = tensor("op_13398_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13398_cast_fp16 = slice_by_index(begin = var_13398_begin_0, end = var_13398_end_0, end_mask = var_13398_end_mask_0, x = k_127_cast_fp16)[name = tensor("op_13398_cast_fp16")]; + tensor var_13402_begin_0 = const()[name = tensor("op_13402_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_13402_end_0 = const()[name = tensor("op_13402_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_13402_end_mask_0 = const()[name = tensor("op_13402_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13402_cast_fp16 = slice_by_index(begin = var_13402_begin_0, end = var_13402_end_0, end_mask = var_13402_end_mask_0, x = k_127_cast_fp16)[name = tensor("op_13402_cast_fp16")]; + tensor var_13406_begin_0 = const()[name = tensor("op_13406_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_13406_end_0 = const()[name = tensor("op_13406_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_13406_end_mask_0 = const()[name = tensor("op_13406_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13406_cast_fp16 = slice_by_index(begin = var_13406_begin_0, end = var_13406_end_0, end_mask = var_13406_end_mask_0, x = k_127_cast_fp16)[name = tensor("op_13406_cast_fp16")]; + tensor var_13410_begin_0 = const()[name = tensor("op_13410_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_13410_end_0 = const()[name = tensor("op_13410_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_13410_end_mask_0 = const()[name = tensor("op_13410_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13410_cast_fp16 = slice_by_index(begin = var_13410_begin_0, end = var_13410_end_0, end_mask = var_13410_end_mask_0, x = k_127_cast_fp16)[name = tensor("op_13410_cast_fp16")]; + tensor var_13414_begin_0 = const()[name = tensor("op_13414_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_13414_end_0 = const()[name = tensor("op_13414_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_13414_end_mask_0 = const()[name = tensor("op_13414_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13414_cast_fp16 = slice_by_index(begin = var_13414_begin_0, end = var_13414_end_0, end_mask = var_13414_end_mask_0, x = k_127_cast_fp16)[name = tensor("op_13414_cast_fp16")]; + tensor var_13418_begin_0 = const()[name = tensor("op_13418_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_13418_end_0 = const()[name = tensor("op_13418_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_13418_end_mask_0 = const()[name = tensor("op_13418_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13418_cast_fp16 = slice_by_index(begin = var_13418_begin_0, end = var_13418_end_0, end_mask = var_13418_end_mask_0, x = k_127_cast_fp16)[name = tensor("op_13418_cast_fp16")]; + tensor var_13422_begin_0 = const()[name = tensor("op_13422_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_13422_end_0 = const()[name = tensor("op_13422_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_13422_end_mask_0 = const()[name = tensor("op_13422_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13422_cast_fp16 = slice_by_index(begin = var_13422_begin_0, end = var_13422_end_0, end_mask = var_13422_end_mask_0, x = k_127_cast_fp16)[name = tensor("op_13422_cast_fp16")]; + tensor var_13426_begin_0 = const()[name = tensor("op_13426_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_13426_end_0 = const()[name = tensor("op_13426_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_13426_end_mask_0 = const()[name = tensor("op_13426_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13426_cast_fp16 = slice_by_index(begin = var_13426_begin_0, end = var_13426_end_0, end_mask = var_13426_end_mask_0, x = k_127_cast_fp16)[name = tensor("op_13426_cast_fp16")]; + tensor var_13430_begin_0 = const()[name = tensor("op_13430_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_13430_end_0 = const()[name = tensor("op_13430_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_13430_end_mask_0 = const()[name = tensor("op_13430_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13430_cast_fp16 = slice_by_index(begin = var_13430_begin_0, end = var_13430_end_0, end_mask = var_13430_end_mask_0, x = k_127_cast_fp16)[name = tensor("op_13430_cast_fp16")]; + tensor var_13434_begin_0 = const()[name = tensor("op_13434_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_13434_end_0 = const()[name = tensor("op_13434_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_13434_end_mask_0 = const()[name = tensor("op_13434_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13434_cast_fp16 = slice_by_index(begin = var_13434_begin_0, end = var_13434_end_0, end_mask = var_13434_end_mask_0, x = k_127_cast_fp16)[name = tensor("op_13434_cast_fp16")]; + tensor var_13438_begin_0 = const()[name = tensor("op_13438_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_13438_end_0 = const()[name = tensor("op_13438_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_13438_end_mask_0 = const()[name = tensor("op_13438_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13438_cast_fp16 = slice_by_index(begin = var_13438_begin_0, end = var_13438_end_0, end_mask = var_13438_end_mask_0, x = k_127_cast_fp16)[name = tensor("op_13438_cast_fp16")]; + tensor var_13442_begin_0 = const()[name = tensor("op_13442_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_13442_end_0 = const()[name = tensor("op_13442_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_13442_end_mask_0 = const()[name = tensor("op_13442_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13442_cast_fp16 = slice_by_index(begin = var_13442_begin_0, end = var_13442_end_0, end_mask = var_13442_end_mask_0, x = k_127_cast_fp16)[name = tensor("op_13442_cast_fp16")]; + tensor var_13446_begin_0 = const()[name = tensor("op_13446_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_13446_end_0 = const()[name = tensor("op_13446_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_13446_end_mask_0 = const()[name = tensor("op_13446_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13446_cast_fp16 = slice_by_index(begin = var_13446_begin_0, end = var_13446_end_0, end_mask = var_13446_end_mask_0, x = k_127_cast_fp16)[name = tensor("op_13446_cast_fp16")]; + tensor var_13450_begin_0 = const()[name = tensor("op_13450_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_13450_end_0 = const()[name = tensor("op_13450_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_13450_end_mask_0 = const()[name = tensor("op_13450_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13450_cast_fp16 = slice_by_index(begin = var_13450_begin_0, end = var_13450_end_0, end_mask = var_13450_end_mask_0, x = k_127_cast_fp16)[name = tensor("op_13450_cast_fp16")]; + tensor var_13454_begin_0 = const()[name = tensor("op_13454_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_13454_end_0 = const()[name = tensor("op_13454_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_13454_end_mask_0 = const()[name = tensor("op_13454_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13454_cast_fp16 = slice_by_index(begin = var_13454_begin_0, end = var_13454_end_0, end_mask = var_13454_end_mask_0, x = k_127_cast_fp16)[name = tensor("op_13454_cast_fp16")]; + tensor var_13458_begin_0 = const()[name = tensor("op_13458_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_13458_end_0 = const()[name = tensor("op_13458_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_13458_end_mask_0 = const()[name = tensor("op_13458_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13458_cast_fp16 = slice_by_index(begin = var_13458_begin_0, end = var_13458_end_0, end_mask = var_13458_end_mask_0, x = k_127_cast_fp16)[name = tensor("op_13458_cast_fp16")]; + tensor var_13462_begin_0 = const()[name = tensor("op_13462_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_13462_end_0 = const()[name = tensor("op_13462_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_13462_end_mask_0 = const()[name = tensor("op_13462_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13462_cast_fp16 = slice_by_index(begin = var_13462_begin_0, end = var_13462_end_0, end_mask = var_13462_end_mask_0, x = k_127_cast_fp16)[name = tensor("op_13462_cast_fp16")]; + tensor var_13466_begin_0 = const()[name = tensor("op_13466_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_13466_end_0 = const()[name = tensor("op_13466_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_13466_end_mask_0 = const()[name = tensor("op_13466_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13466_cast_fp16 = slice_by_index(begin = var_13466_begin_0, end = var_13466_end_0, end_mask = var_13466_end_mask_0, x = k_127_cast_fp16)[name = tensor("op_13466_cast_fp16")]; + tensor var_13470_begin_0 = const()[name = tensor("op_13470_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_13470_end_0 = const()[name = tensor("op_13470_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_13470_end_mask_0 = const()[name = tensor("op_13470_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13470_cast_fp16 = slice_by_index(begin = var_13470_begin_0, end = var_13470_end_0, end_mask = var_13470_end_mask_0, x = k_127_cast_fp16)[name = tensor("op_13470_cast_fp16")]; + tensor var_13472_begin_0 = const()[name = tensor("op_13472_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_13472_end_0 = const()[name = tensor("op_13472_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_13472_end_mask_0 = const()[name = tensor("op_13472_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13472_cast_fp16 = slice_by_index(begin = var_13472_begin_0, end = var_13472_end_0, end_mask = var_13472_end_mask_0, x = v_63_cast_fp16)[name = tensor("op_13472_cast_fp16")]; + tensor var_13476_begin_0 = const()[name = tensor("op_13476_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_13476_end_0 = const()[name = tensor("op_13476_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_13476_end_mask_0 = const()[name = tensor("op_13476_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13476_cast_fp16 = slice_by_index(begin = var_13476_begin_0, end = var_13476_end_0, end_mask = var_13476_end_mask_0, x = v_63_cast_fp16)[name = tensor("op_13476_cast_fp16")]; + tensor var_13480_begin_0 = const()[name = tensor("op_13480_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_13480_end_0 = const()[name = tensor("op_13480_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_13480_end_mask_0 = const()[name = tensor("op_13480_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13480_cast_fp16 = slice_by_index(begin = var_13480_begin_0, end = var_13480_end_0, end_mask = var_13480_end_mask_0, x = v_63_cast_fp16)[name = tensor("op_13480_cast_fp16")]; + tensor var_13484_begin_0 = const()[name = tensor("op_13484_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_13484_end_0 = const()[name = tensor("op_13484_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_13484_end_mask_0 = const()[name = tensor("op_13484_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13484_cast_fp16 = slice_by_index(begin = var_13484_begin_0, end = var_13484_end_0, end_mask = var_13484_end_mask_0, x = v_63_cast_fp16)[name = tensor("op_13484_cast_fp16")]; + tensor var_13488_begin_0 = const()[name = tensor("op_13488_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_13488_end_0 = const()[name = tensor("op_13488_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_13488_end_mask_0 = const()[name = tensor("op_13488_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13488_cast_fp16 = slice_by_index(begin = var_13488_begin_0, end = var_13488_end_0, end_mask = var_13488_end_mask_0, x = v_63_cast_fp16)[name = tensor("op_13488_cast_fp16")]; + tensor var_13492_begin_0 = const()[name = tensor("op_13492_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_13492_end_0 = const()[name = tensor("op_13492_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_13492_end_mask_0 = const()[name = tensor("op_13492_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13492_cast_fp16 = slice_by_index(begin = var_13492_begin_0, end = var_13492_end_0, end_mask = var_13492_end_mask_0, x = v_63_cast_fp16)[name = tensor("op_13492_cast_fp16")]; + tensor var_13496_begin_0 = const()[name = tensor("op_13496_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_13496_end_0 = const()[name = tensor("op_13496_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_13496_end_mask_0 = const()[name = tensor("op_13496_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13496_cast_fp16 = slice_by_index(begin = var_13496_begin_0, end = var_13496_end_0, end_mask = var_13496_end_mask_0, x = v_63_cast_fp16)[name = tensor("op_13496_cast_fp16")]; + tensor var_13500_begin_0 = const()[name = tensor("op_13500_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_13500_end_0 = const()[name = tensor("op_13500_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_13500_end_mask_0 = const()[name = tensor("op_13500_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13500_cast_fp16 = slice_by_index(begin = var_13500_begin_0, end = var_13500_end_0, end_mask = var_13500_end_mask_0, x = v_63_cast_fp16)[name = tensor("op_13500_cast_fp16")]; + tensor var_13504_begin_0 = const()[name = tensor("op_13504_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_13504_end_0 = const()[name = tensor("op_13504_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_13504_end_mask_0 = const()[name = tensor("op_13504_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13504_cast_fp16 = slice_by_index(begin = var_13504_begin_0, end = var_13504_end_0, end_mask = var_13504_end_mask_0, x = v_63_cast_fp16)[name = tensor("op_13504_cast_fp16")]; + tensor var_13508_begin_0 = const()[name = tensor("op_13508_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_13508_end_0 = const()[name = tensor("op_13508_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_13508_end_mask_0 = const()[name = tensor("op_13508_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13508_cast_fp16 = slice_by_index(begin = var_13508_begin_0, end = var_13508_end_0, end_mask = var_13508_end_mask_0, x = v_63_cast_fp16)[name = tensor("op_13508_cast_fp16")]; + tensor var_13512_begin_0 = const()[name = tensor("op_13512_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_13512_end_0 = const()[name = tensor("op_13512_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_13512_end_mask_0 = const()[name = tensor("op_13512_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13512_cast_fp16 = slice_by_index(begin = var_13512_begin_0, end = var_13512_end_0, end_mask = var_13512_end_mask_0, x = v_63_cast_fp16)[name = tensor("op_13512_cast_fp16")]; + tensor var_13516_begin_0 = const()[name = tensor("op_13516_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_13516_end_0 = const()[name = tensor("op_13516_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_13516_end_mask_0 = const()[name = tensor("op_13516_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13516_cast_fp16 = slice_by_index(begin = var_13516_begin_0, end = var_13516_end_0, end_mask = var_13516_end_mask_0, x = v_63_cast_fp16)[name = tensor("op_13516_cast_fp16")]; + tensor var_13520_begin_0 = const()[name = tensor("op_13520_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_13520_end_0 = const()[name = tensor("op_13520_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_13520_end_mask_0 = const()[name = tensor("op_13520_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13520_cast_fp16 = slice_by_index(begin = var_13520_begin_0, end = var_13520_end_0, end_mask = var_13520_end_mask_0, x = v_63_cast_fp16)[name = tensor("op_13520_cast_fp16")]; + tensor var_13524_begin_0 = const()[name = tensor("op_13524_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_13524_end_0 = const()[name = tensor("op_13524_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_13524_end_mask_0 = const()[name = tensor("op_13524_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13524_cast_fp16 = slice_by_index(begin = var_13524_begin_0, end = var_13524_end_0, end_mask = var_13524_end_mask_0, x = v_63_cast_fp16)[name = tensor("op_13524_cast_fp16")]; + tensor var_13528_begin_0 = const()[name = tensor("op_13528_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_13528_end_0 = const()[name = tensor("op_13528_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_13528_end_mask_0 = const()[name = tensor("op_13528_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13528_cast_fp16 = slice_by_index(begin = var_13528_begin_0, end = var_13528_end_0, end_mask = var_13528_end_mask_0, x = v_63_cast_fp16)[name = tensor("op_13528_cast_fp16")]; + tensor var_13532_begin_0 = const()[name = tensor("op_13532_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_13532_end_0 = const()[name = tensor("op_13532_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_13532_end_mask_0 = const()[name = tensor("op_13532_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13532_cast_fp16 = slice_by_index(begin = var_13532_begin_0, end = var_13532_end_0, end_mask = var_13532_end_mask_0, x = v_63_cast_fp16)[name = tensor("op_13532_cast_fp16")]; + tensor var_13536_begin_0 = const()[name = tensor("op_13536_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_13536_end_0 = const()[name = tensor("op_13536_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_13536_end_mask_0 = const()[name = tensor("op_13536_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13536_cast_fp16 = slice_by_index(begin = var_13536_begin_0, end = var_13536_end_0, end_mask = var_13536_end_mask_0, x = v_63_cast_fp16)[name = tensor("op_13536_cast_fp16")]; + tensor var_13540_begin_0 = const()[name = tensor("op_13540_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_13540_end_0 = const()[name = tensor("op_13540_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_13540_end_mask_0 = const()[name = tensor("op_13540_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13540_cast_fp16 = slice_by_index(begin = var_13540_begin_0, end = var_13540_end_0, end_mask = var_13540_end_mask_0, x = v_63_cast_fp16)[name = tensor("op_13540_cast_fp16")]; + tensor var_13544_begin_0 = const()[name = tensor("op_13544_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_13544_end_0 = const()[name = tensor("op_13544_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_13544_end_mask_0 = const()[name = tensor("op_13544_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13544_cast_fp16 = slice_by_index(begin = var_13544_begin_0, end = var_13544_end_0, end_mask = var_13544_end_mask_0, x = v_63_cast_fp16)[name = tensor("op_13544_cast_fp16")]; + tensor var_13548_begin_0 = const()[name = tensor("op_13548_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_13548_end_0 = const()[name = tensor("op_13548_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_13548_end_mask_0 = const()[name = tensor("op_13548_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13548_cast_fp16 = slice_by_index(begin = var_13548_begin_0, end = var_13548_end_0, end_mask = var_13548_end_mask_0, x = v_63_cast_fp16)[name = tensor("op_13548_cast_fp16")]; + tensor var_13552_equation_0 = const()[name = tensor("op_13552_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13552_cast_fp16 = einsum(equation = var_13552_equation_0, values = (var_13394_cast_fp16, var_13311_cast_fp16))[name = tensor("op_13552_cast_fp16")]; + tensor var_13553_to_fp16 = const()[name = tensor("op_13553_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1081_cast_fp16 = mul(x = var_13552_cast_fp16, y = var_13553_to_fp16)[name = tensor("aw_1081_cast_fp16")]; + tensor var_13556_equation_0 = const()[name = tensor("op_13556_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13556_cast_fp16 = einsum(equation = var_13556_equation_0, values = (var_13398_cast_fp16, var_13315_cast_fp16))[name = tensor("op_13556_cast_fp16")]; + tensor var_13557_to_fp16 = const()[name = tensor("op_13557_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1083_cast_fp16 = mul(x = var_13556_cast_fp16, y = var_13557_to_fp16)[name = tensor("aw_1083_cast_fp16")]; + tensor var_13560_equation_0 = const()[name = tensor("op_13560_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13560_cast_fp16 = einsum(equation = var_13560_equation_0, values = (var_13402_cast_fp16, var_13319_cast_fp16))[name = tensor("op_13560_cast_fp16")]; + tensor var_13561_to_fp16 = const()[name = tensor("op_13561_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1085_cast_fp16 = mul(x = var_13560_cast_fp16, y = var_13561_to_fp16)[name = tensor("aw_1085_cast_fp16")]; + tensor var_13564_equation_0 = const()[name = tensor("op_13564_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13564_cast_fp16 = einsum(equation = var_13564_equation_0, values = (var_13406_cast_fp16, var_13323_cast_fp16))[name = tensor("op_13564_cast_fp16")]; + tensor var_13565_to_fp16 = const()[name = tensor("op_13565_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1087_cast_fp16 = mul(x = var_13564_cast_fp16, y = var_13565_to_fp16)[name = tensor("aw_1087_cast_fp16")]; + tensor var_13568_equation_0 = const()[name = tensor("op_13568_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13568_cast_fp16 = einsum(equation = var_13568_equation_0, values = (var_13410_cast_fp16, var_13327_cast_fp16))[name = tensor("op_13568_cast_fp16")]; + tensor var_13569_to_fp16 = const()[name = tensor("op_13569_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1089_cast_fp16 = mul(x = var_13568_cast_fp16, y = var_13569_to_fp16)[name = tensor("aw_1089_cast_fp16")]; + tensor var_13572_equation_0 = const()[name = tensor("op_13572_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13572_cast_fp16 = einsum(equation = var_13572_equation_0, values = (var_13414_cast_fp16, var_13331_cast_fp16))[name = tensor("op_13572_cast_fp16")]; + tensor var_13573_to_fp16 = const()[name = tensor("op_13573_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1091_cast_fp16 = mul(x = var_13572_cast_fp16, y = var_13573_to_fp16)[name = tensor("aw_1091_cast_fp16")]; + tensor var_13576_equation_0 = const()[name = tensor("op_13576_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13576_cast_fp16 = einsum(equation = var_13576_equation_0, values = (var_13418_cast_fp16, var_13335_cast_fp16))[name = tensor("op_13576_cast_fp16")]; + tensor var_13577_to_fp16 = const()[name = tensor("op_13577_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1093_cast_fp16 = mul(x = var_13576_cast_fp16, y = var_13577_to_fp16)[name = tensor("aw_1093_cast_fp16")]; + tensor var_13580_equation_0 = const()[name = tensor("op_13580_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13580_cast_fp16 = einsum(equation = var_13580_equation_0, values = (var_13422_cast_fp16, var_13339_cast_fp16))[name = tensor("op_13580_cast_fp16")]; + tensor var_13581_to_fp16 = const()[name = tensor("op_13581_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1095_cast_fp16 = mul(x = var_13580_cast_fp16, y = var_13581_to_fp16)[name = tensor("aw_1095_cast_fp16")]; + tensor var_13584_equation_0 = const()[name = tensor("op_13584_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13584_cast_fp16 = einsum(equation = var_13584_equation_0, values = (var_13426_cast_fp16, var_13343_cast_fp16))[name = tensor("op_13584_cast_fp16")]; + tensor var_13585_to_fp16 = const()[name = tensor("op_13585_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1097_cast_fp16 = mul(x = var_13584_cast_fp16, y = var_13585_to_fp16)[name = tensor("aw_1097_cast_fp16")]; + tensor var_13588_equation_0 = const()[name = tensor("op_13588_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13588_cast_fp16 = einsum(equation = var_13588_equation_0, values = (var_13430_cast_fp16, var_13347_cast_fp16))[name = tensor("op_13588_cast_fp16")]; + tensor var_13589_to_fp16 = const()[name = tensor("op_13589_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1099_cast_fp16 = mul(x = var_13588_cast_fp16, y = var_13589_to_fp16)[name = tensor("aw_1099_cast_fp16")]; + tensor var_13592_equation_0 = const()[name = tensor("op_13592_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13592_cast_fp16 = einsum(equation = var_13592_equation_0, values = (var_13434_cast_fp16, var_13351_cast_fp16))[name = tensor("op_13592_cast_fp16")]; + tensor var_13593_to_fp16 = const()[name = tensor("op_13593_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1101_cast_fp16 = mul(x = var_13592_cast_fp16, y = var_13593_to_fp16)[name = tensor("aw_1101_cast_fp16")]; + tensor var_13596_equation_0 = const()[name = tensor("op_13596_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13596_cast_fp16 = einsum(equation = var_13596_equation_0, values = (var_13438_cast_fp16, var_13355_cast_fp16))[name = tensor("op_13596_cast_fp16")]; + tensor var_13597_to_fp16 = const()[name = tensor("op_13597_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1103_cast_fp16 = mul(x = var_13596_cast_fp16, y = var_13597_to_fp16)[name = tensor("aw_1103_cast_fp16")]; + tensor var_13600_equation_0 = const()[name = tensor("op_13600_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13600_cast_fp16 = einsum(equation = var_13600_equation_0, values = (var_13442_cast_fp16, var_13359_cast_fp16))[name = tensor("op_13600_cast_fp16")]; + tensor var_13601_to_fp16 = const()[name = tensor("op_13601_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1105_cast_fp16 = mul(x = var_13600_cast_fp16, y = var_13601_to_fp16)[name = tensor("aw_1105_cast_fp16")]; + tensor var_13604_equation_0 = const()[name = tensor("op_13604_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13604_cast_fp16 = einsum(equation = var_13604_equation_0, values = (var_13446_cast_fp16, var_13363_cast_fp16))[name = tensor("op_13604_cast_fp16")]; + tensor var_13605_to_fp16 = const()[name = tensor("op_13605_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1107_cast_fp16 = mul(x = var_13604_cast_fp16, y = var_13605_to_fp16)[name = tensor("aw_1107_cast_fp16")]; + tensor var_13608_equation_0 = const()[name = tensor("op_13608_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13608_cast_fp16 = einsum(equation = var_13608_equation_0, values = (var_13450_cast_fp16, var_13367_cast_fp16))[name = tensor("op_13608_cast_fp16")]; + tensor var_13609_to_fp16 = const()[name = tensor("op_13609_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1109_cast_fp16 = mul(x = var_13608_cast_fp16, y = var_13609_to_fp16)[name = tensor("aw_1109_cast_fp16")]; + tensor var_13612_equation_0 = const()[name = tensor("op_13612_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13612_cast_fp16 = einsum(equation = var_13612_equation_0, values = (var_13454_cast_fp16, var_13371_cast_fp16))[name = tensor("op_13612_cast_fp16")]; + tensor var_13613_to_fp16 = const()[name = tensor("op_13613_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1111_cast_fp16 = mul(x = var_13612_cast_fp16, y = var_13613_to_fp16)[name = tensor("aw_1111_cast_fp16")]; + tensor var_13616_equation_0 = const()[name = tensor("op_13616_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13616_cast_fp16 = einsum(equation = var_13616_equation_0, values = (var_13458_cast_fp16, var_13375_cast_fp16))[name = tensor("op_13616_cast_fp16")]; + tensor var_13617_to_fp16 = const()[name = tensor("op_13617_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1113_cast_fp16 = mul(x = var_13616_cast_fp16, y = var_13617_to_fp16)[name = tensor("aw_1113_cast_fp16")]; + tensor var_13620_equation_0 = const()[name = tensor("op_13620_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13620_cast_fp16 = einsum(equation = var_13620_equation_0, values = (var_13462_cast_fp16, var_13379_cast_fp16))[name = tensor("op_13620_cast_fp16")]; + tensor var_13621_to_fp16 = const()[name = tensor("op_13621_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1115_cast_fp16 = mul(x = var_13620_cast_fp16, y = var_13621_to_fp16)[name = tensor("aw_1115_cast_fp16")]; + tensor var_13624_equation_0 = const()[name = tensor("op_13624_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13624_cast_fp16 = einsum(equation = var_13624_equation_0, values = (var_13466_cast_fp16, var_13383_cast_fp16))[name = tensor("op_13624_cast_fp16")]; + tensor var_13625_to_fp16 = const()[name = tensor("op_13625_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1117_cast_fp16 = mul(x = var_13624_cast_fp16, y = var_13625_to_fp16)[name = tensor("aw_1117_cast_fp16")]; + tensor var_13628_equation_0 = const()[name = tensor("op_13628_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13628_cast_fp16 = einsum(equation = var_13628_equation_0, values = (var_13470_cast_fp16, var_13387_cast_fp16))[name = tensor("op_13628_cast_fp16")]; + tensor var_13629_to_fp16 = const()[name = tensor("op_13629_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1119_cast_fp16 = mul(x = var_13628_cast_fp16, y = var_13629_to_fp16)[name = tensor("aw_1119_cast_fp16")]; + tensor var_13631_cast_fp16 = softmax(axis = var_2624, x = aw_1081_cast_fp16)[name = tensor("op_13631_cast_fp16")]; + tensor var_13632_cast_fp16 = softmax(axis = var_2624, x = aw_1083_cast_fp16)[name = tensor("op_13632_cast_fp16")]; + tensor var_13633_cast_fp16 = softmax(axis = var_2624, x = aw_1085_cast_fp16)[name = tensor("op_13633_cast_fp16")]; + tensor var_13634_cast_fp16 = softmax(axis = var_2624, x = aw_1087_cast_fp16)[name = tensor("op_13634_cast_fp16")]; + tensor var_13635_cast_fp16 = softmax(axis = var_2624, x = aw_1089_cast_fp16)[name = tensor("op_13635_cast_fp16")]; + tensor var_13636_cast_fp16 = softmax(axis = var_2624, x = aw_1091_cast_fp16)[name = tensor("op_13636_cast_fp16")]; + tensor var_13637_cast_fp16 = softmax(axis = var_2624, x = aw_1093_cast_fp16)[name = tensor("op_13637_cast_fp16")]; + tensor var_13638_cast_fp16 = softmax(axis = var_2624, x = aw_1095_cast_fp16)[name = tensor("op_13638_cast_fp16")]; + tensor var_13639_cast_fp16 = softmax(axis = var_2624, x = aw_1097_cast_fp16)[name = tensor("op_13639_cast_fp16")]; + tensor var_13640_cast_fp16 = softmax(axis = var_2624, x = aw_1099_cast_fp16)[name = tensor("op_13640_cast_fp16")]; + tensor var_13641_cast_fp16 = softmax(axis = var_2624, x = aw_1101_cast_fp16)[name = tensor("op_13641_cast_fp16")]; + tensor var_13642_cast_fp16 = softmax(axis = var_2624, x = aw_1103_cast_fp16)[name = tensor("op_13642_cast_fp16")]; + tensor var_13643_cast_fp16 = softmax(axis = var_2624, x = aw_1105_cast_fp16)[name = tensor("op_13643_cast_fp16")]; + tensor var_13644_cast_fp16 = softmax(axis = var_2624, x = aw_1107_cast_fp16)[name = tensor("op_13644_cast_fp16")]; + tensor var_13645_cast_fp16 = softmax(axis = var_2624, x = aw_1109_cast_fp16)[name = tensor("op_13645_cast_fp16")]; + tensor var_13646_cast_fp16 = softmax(axis = var_2624, x = aw_1111_cast_fp16)[name = tensor("op_13646_cast_fp16")]; + tensor var_13647_cast_fp16 = softmax(axis = var_2624, x = aw_1113_cast_fp16)[name = tensor("op_13647_cast_fp16")]; + tensor var_13648_cast_fp16 = softmax(axis = var_2624, x = aw_1115_cast_fp16)[name = tensor("op_13648_cast_fp16")]; + tensor var_13649_cast_fp16 = softmax(axis = var_2624, x = aw_1117_cast_fp16)[name = tensor("op_13649_cast_fp16")]; + tensor var_13650_cast_fp16 = softmax(axis = var_2624, x = aw_1119_cast_fp16)[name = tensor("op_13650_cast_fp16")]; + tensor var_13652_equation_0 = const()[name = tensor("op_13652_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13652_cast_fp16 = einsum(equation = var_13652_equation_0, values = (var_13472_cast_fp16, var_13631_cast_fp16))[name = tensor("op_13652_cast_fp16")]; + tensor var_13654_equation_0 = const()[name = tensor("op_13654_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13654_cast_fp16 = einsum(equation = var_13654_equation_0, values = (var_13476_cast_fp16, var_13632_cast_fp16))[name = tensor("op_13654_cast_fp16")]; + tensor var_13656_equation_0 = const()[name = tensor("op_13656_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13656_cast_fp16 = einsum(equation = var_13656_equation_0, values = (var_13480_cast_fp16, var_13633_cast_fp16))[name = tensor("op_13656_cast_fp16")]; + tensor var_13658_equation_0 = const()[name = tensor("op_13658_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13658_cast_fp16 = einsum(equation = var_13658_equation_0, values = (var_13484_cast_fp16, var_13634_cast_fp16))[name = tensor("op_13658_cast_fp16")]; + tensor var_13660_equation_0 = const()[name = tensor("op_13660_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13660_cast_fp16 = einsum(equation = var_13660_equation_0, values = (var_13488_cast_fp16, var_13635_cast_fp16))[name = tensor("op_13660_cast_fp16")]; + tensor var_13662_equation_0 = const()[name = tensor("op_13662_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13662_cast_fp16 = einsum(equation = var_13662_equation_0, values = (var_13492_cast_fp16, var_13636_cast_fp16))[name = tensor("op_13662_cast_fp16")]; + tensor var_13664_equation_0 = const()[name = tensor("op_13664_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13664_cast_fp16 = einsum(equation = var_13664_equation_0, values = (var_13496_cast_fp16, var_13637_cast_fp16))[name = tensor("op_13664_cast_fp16")]; + tensor var_13666_equation_0 = const()[name = tensor("op_13666_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13666_cast_fp16 = einsum(equation = var_13666_equation_0, values = (var_13500_cast_fp16, var_13638_cast_fp16))[name = tensor("op_13666_cast_fp16")]; + tensor var_13668_equation_0 = const()[name = tensor("op_13668_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13668_cast_fp16 = einsum(equation = var_13668_equation_0, values = (var_13504_cast_fp16, var_13639_cast_fp16))[name = tensor("op_13668_cast_fp16")]; + tensor var_13670_equation_0 = const()[name = tensor("op_13670_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13670_cast_fp16 = einsum(equation = var_13670_equation_0, values = (var_13508_cast_fp16, var_13640_cast_fp16))[name = tensor("op_13670_cast_fp16")]; + tensor var_13672_equation_0 = const()[name = tensor("op_13672_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13672_cast_fp16 = einsum(equation = var_13672_equation_0, values = (var_13512_cast_fp16, var_13641_cast_fp16))[name = tensor("op_13672_cast_fp16")]; + tensor var_13674_equation_0 = const()[name = tensor("op_13674_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13674_cast_fp16 = einsum(equation = var_13674_equation_0, values = (var_13516_cast_fp16, var_13642_cast_fp16))[name = tensor("op_13674_cast_fp16")]; + tensor var_13676_equation_0 = const()[name = tensor("op_13676_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13676_cast_fp16 = einsum(equation = var_13676_equation_0, values = (var_13520_cast_fp16, var_13643_cast_fp16))[name = tensor("op_13676_cast_fp16")]; + tensor var_13678_equation_0 = const()[name = tensor("op_13678_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13678_cast_fp16 = einsum(equation = var_13678_equation_0, values = (var_13524_cast_fp16, var_13644_cast_fp16))[name = tensor("op_13678_cast_fp16")]; + tensor var_13680_equation_0 = const()[name = tensor("op_13680_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13680_cast_fp16 = einsum(equation = var_13680_equation_0, values = (var_13528_cast_fp16, var_13645_cast_fp16))[name = tensor("op_13680_cast_fp16")]; + tensor var_13682_equation_0 = const()[name = tensor("op_13682_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13682_cast_fp16 = einsum(equation = var_13682_equation_0, values = (var_13532_cast_fp16, var_13646_cast_fp16))[name = tensor("op_13682_cast_fp16")]; + tensor var_13684_equation_0 = const()[name = tensor("op_13684_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13684_cast_fp16 = einsum(equation = var_13684_equation_0, values = (var_13536_cast_fp16, var_13647_cast_fp16))[name = tensor("op_13684_cast_fp16")]; + tensor var_13686_equation_0 = const()[name = tensor("op_13686_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13686_cast_fp16 = einsum(equation = var_13686_equation_0, values = (var_13540_cast_fp16, var_13648_cast_fp16))[name = tensor("op_13686_cast_fp16")]; + tensor var_13688_equation_0 = const()[name = tensor("op_13688_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13688_cast_fp16 = einsum(equation = var_13688_equation_0, values = (var_13544_cast_fp16, var_13649_cast_fp16))[name = tensor("op_13688_cast_fp16")]; + tensor var_13690_equation_0 = const()[name = tensor("op_13690_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13690_cast_fp16 = einsum(equation = var_13690_equation_0, values = (var_13548_cast_fp16, var_13650_cast_fp16))[name = tensor("op_13690_cast_fp16")]; + tensor input_239_interleave_0 = const()[name = tensor("input_239_interleave_0"), val = tensor(false)]; + tensor input_239_cast_fp16 = concat(axis = var_2624, interleave = input_239_interleave_0, values = (var_13652_cast_fp16, var_13654_cast_fp16, var_13656_cast_fp16, var_13658_cast_fp16, var_13660_cast_fp16, var_13662_cast_fp16, var_13664_cast_fp16, var_13666_cast_fp16, var_13668_cast_fp16, var_13670_cast_fp16, var_13672_cast_fp16, var_13674_cast_fp16, var_13676_cast_fp16, var_13678_cast_fp16, var_13680_cast_fp16, var_13682_cast_fp16, var_13684_cast_fp16, var_13686_cast_fp16, var_13688_cast_fp16, var_13690_cast_fp16))[name = tensor("input_239_cast_fp16")]; + tensor var_13696 = const()[name = tensor("op_13696"), val = tensor([1, 1])]; + tensor var_13698 = const()[name = tensor("op_13698"), val = tensor([1, 1])]; + tensor var_13700_pad_type_0 = const()[name = tensor("op_13700_pad_type_0"), val = tensor("custom")]; + tensor var_13700_pad_0 = const()[name = tensor("op_13700_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_1_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_1_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1058533760)))]; + tensor down_blocks_2_attentions_1_transformer_blocks_1_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_1_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1061810624)))]; + tensor var_13700_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_1_attn2_to_out_0_bias_to_fp16, dilations = var_13698, groups = var_2624, pad = var_13700_pad_0, pad_type = var_13700_pad_type_0, strides = var_13696, weight = down_blocks_2_attentions_1_transformer_blocks_1_attn2_to_out_0_weight_to_fp16, x = input_239_cast_fp16)[name = tensor("op_13700_cast_fp16")]; + tensor inputs_95_cast_fp16 = add(x = var_13700_cast_fp16, y = inputs_93_cast_fp16)[name = tensor("inputs_95_cast_fp16")]; + tensor input_241_axes_0 = const()[name = tensor("input_241_axes_0"), val = tensor([1])]; + tensor input_241_gamma_0_to_fp16 = const()[name = tensor("input_241_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1061813248)))]; + tensor input_241_beta_0_to_fp16 = const()[name = tensor("input_241_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1061815872)))]; + tensor var_13710_to_fp16 = const()[name = tensor("op_13710_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_241_cast_fp16 = layer_norm(axes = input_241_axes_0, beta = input_241_beta_0_to_fp16, epsilon = var_13710_to_fp16, gamma = input_241_gamma_0_to_fp16, x = inputs_95_cast_fp16)[name = tensor("input_241_cast_fp16")]; + tensor var_13726 = const()[name = tensor("op_13726"), val = tensor([1, 1])]; + tensor var_13728 = const()[name = tensor("op_13728"), val = tensor([1, 1])]; + tensor var_13730_pad_type_0 = const()[name = tensor("op_13730_pad_type_0"), val = tensor("custom")]; + tensor var_13730_pad_0 = const()[name = tensor("op_13730_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_1_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_1_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1061818496)))]; + tensor down_blocks_2_attentions_1_transformer_blocks_1_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_1_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1088032960)))]; + tensor var_13730_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_1_ff_net_0_proj_bias_to_fp16, dilations = var_13728, groups = var_2624, pad = var_13730_pad_0, pad_type = var_13730_pad_type_0, strides = var_13726, weight = down_blocks_2_attentions_1_transformer_blocks_1_ff_net_0_proj_weight_to_fp16, x = input_241_cast_fp16)[name = tensor("op_13730_cast_fp16")]; + tensor var_13731_split_sizes_0 = const()[name = tensor("op_13731_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_13731_axis_0 = const()[name = tensor("op_13731_axis_0"), val = tensor(1)]; + tensor var_13731_cast_fp16_0, tensor var_13731_cast_fp16_1 = split(axis = var_13731_axis_0, split_sizes = var_13731_split_sizes_0, x = var_13730_cast_fp16)[name = tensor("op_13731_cast_fp16")]; + tensor var_13733_mode_0 = const()[name = tensor("op_13733_mode_0"), val = tensor("EXACT")]; + tensor var_13733_cast_fp16 = gelu(mode = var_13733_mode_0, x = var_13731_cast_fp16_1)[name = tensor("op_13733_cast_fp16")]; + tensor input_243_cast_fp16 = mul(x = var_13731_cast_fp16_0, y = var_13733_cast_fp16)[name = tensor("input_243_cast_fp16")]; + tensor var_13737 = const()[name = tensor("op_13737"), val = tensor([1, 1])]; + tensor var_13739 = const()[name = tensor("op_13739"), val = tensor([1, 1])]; + tensor var_13741_pad_type_0 = const()[name = tensor("op_13741_pad_type_0"), val = tensor("custom")]; + tensor var_13741_pad_0 = const()[name = tensor("op_13741_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_1_ff_net_2_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_1_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1088053504)))]; + tensor down_blocks_2_attentions_1_transformer_blocks_1_ff_net_2_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_1_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1101160768)))]; + tensor var_13741_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_1_ff_net_2_bias_to_fp16, dilations = var_13739, groups = var_2624, pad = var_13741_pad_0, pad_type = var_13741_pad_type_0, strides = var_13737, weight = down_blocks_2_attentions_1_transformer_blocks_1_ff_net_2_weight_to_fp16, x = input_243_cast_fp16)[name = tensor("op_13741_cast_fp16")]; + tensor inputs_97_cast_fp16 = add(x = var_13741_cast_fp16, y = inputs_95_cast_fp16)[name = tensor("inputs_97_cast_fp16")]; + tensor hidden_states_149_axes_0 = const()[name = tensor("hidden_states_149_axes_0"), val = tensor([1])]; + tensor hidden_states_149_gamma_0_to_fp16 = const()[name = tensor("hidden_states_149_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1101163392)))]; + tensor hidden_states_149_beta_0_to_fp16 = const()[name = tensor("hidden_states_149_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1101166016)))]; + tensor var_13757_to_fp16 = const()[name = tensor("op_13757_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_149_cast_fp16 = layer_norm(axes = hidden_states_149_axes_0, beta = hidden_states_149_beta_0_to_fp16, epsilon = var_13757_to_fp16, gamma = hidden_states_149_gamma_0_to_fp16, x = inputs_97_cast_fp16)[name = tensor("hidden_states_149_cast_fp16")]; + tensor var_13772 = const()[name = tensor("op_13772"), val = tensor([1, 1])]; + tensor var_13774 = const()[name = tensor("op_13774"), val = tensor([1, 1])]; + tensor q_65_pad_type_0 = const()[name = tensor("q_65_pad_type_0"), val = tensor("custom")]; + tensor q_65_pad_0 = const()[name = tensor("q_65_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_2_attn1_to_q_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_2_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1101168640)))]; + tensor q_65_cast_fp16 = conv(dilations = var_13774, groups = var_2624, pad = q_65_pad_0, pad_type = q_65_pad_type_0, strides = var_13772, weight = down_blocks_2_attentions_1_transformer_blocks_2_attn1_to_q_weight_to_fp16, x = hidden_states_149_cast_fp16)[name = tensor("q_65_cast_fp16")]; + tensor var_13778 = const()[name = tensor("op_13778"), val = tensor([1, 1])]; + tensor var_13780 = const()[name = tensor("op_13780"), val = tensor([1, 1])]; + tensor k_129_pad_type_0 = const()[name = tensor("k_129_pad_type_0"), val = tensor("custom")]; + tensor k_129_pad_0 = const()[name = tensor("k_129_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_2_attn1_to_k_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_2_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1104445504)))]; + tensor k_129_cast_fp16 = conv(dilations = var_13780, groups = var_2624, pad = k_129_pad_0, pad_type = k_129_pad_type_0, strides = var_13778, weight = down_blocks_2_attentions_1_transformer_blocks_2_attn1_to_k_weight_to_fp16, x = hidden_states_149_cast_fp16)[name = tensor("k_129_cast_fp16")]; + tensor var_13784 = const()[name = tensor("op_13784"), val = tensor([1, 1])]; + tensor var_13786 = const()[name = tensor("op_13786"), val = tensor([1, 1])]; + tensor v_65_pad_type_0 = const()[name = tensor("v_65_pad_type_0"), val = tensor("custom")]; + tensor v_65_pad_0 = const()[name = tensor("v_65_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_2_attn1_to_v_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_2_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1107722368)))]; + tensor v_65_cast_fp16 = conv(dilations = var_13786, groups = var_2624, pad = v_65_pad_0, pad_type = v_65_pad_type_0, strides = var_13784, weight = down_blocks_2_attentions_1_transformer_blocks_2_attn1_to_v_weight_to_fp16, x = hidden_states_149_cast_fp16)[name = tensor("v_65_cast_fp16")]; + tensor var_13790_begin_0 = const()[name = tensor("op_13790_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_13790_end_0 = const()[name = tensor("op_13790_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_13790_end_mask_0 = const()[name = tensor("op_13790_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13790_cast_fp16 = slice_by_index(begin = var_13790_begin_0, end = var_13790_end_0, end_mask = var_13790_end_mask_0, x = q_65_cast_fp16)[name = tensor("op_13790_cast_fp16")]; + tensor var_13794_begin_0 = const()[name = tensor("op_13794_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_13794_end_0 = const()[name = tensor("op_13794_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_13794_end_mask_0 = const()[name = tensor("op_13794_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13794_cast_fp16 = slice_by_index(begin = var_13794_begin_0, end = var_13794_end_0, end_mask = var_13794_end_mask_0, x = q_65_cast_fp16)[name = tensor("op_13794_cast_fp16")]; + tensor var_13798_begin_0 = const()[name = tensor("op_13798_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_13798_end_0 = const()[name = tensor("op_13798_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_13798_end_mask_0 = const()[name = tensor("op_13798_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13798_cast_fp16 = slice_by_index(begin = var_13798_begin_0, end = var_13798_end_0, end_mask = var_13798_end_mask_0, x = q_65_cast_fp16)[name = tensor("op_13798_cast_fp16")]; + tensor var_13802_begin_0 = const()[name = tensor("op_13802_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_13802_end_0 = const()[name = tensor("op_13802_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_13802_end_mask_0 = const()[name = tensor("op_13802_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13802_cast_fp16 = slice_by_index(begin = var_13802_begin_0, end = var_13802_end_0, end_mask = var_13802_end_mask_0, x = q_65_cast_fp16)[name = tensor("op_13802_cast_fp16")]; + tensor var_13806_begin_0 = const()[name = tensor("op_13806_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_13806_end_0 = const()[name = tensor("op_13806_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_13806_end_mask_0 = const()[name = tensor("op_13806_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13806_cast_fp16 = slice_by_index(begin = var_13806_begin_0, end = var_13806_end_0, end_mask = var_13806_end_mask_0, x = q_65_cast_fp16)[name = tensor("op_13806_cast_fp16")]; + tensor var_13810_begin_0 = const()[name = tensor("op_13810_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_13810_end_0 = const()[name = tensor("op_13810_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_13810_end_mask_0 = const()[name = tensor("op_13810_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13810_cast_fp16 = slice_by_index(begin = var_13810_begin_0, end = var_13810_end_0, end_mask = var_13810_end_mask_0, x = q_65_cast_fp16)[name = tensor("op_13810_cast_fp16")]; + tensor var_13814_begin_0 = const()[name = tensor("op_13814_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_13814_end_0 = const()[name = tensor("op_13814_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_13814_end_mask_0 = const()[name = tensor("op_13814_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13814_cast_fp16 = slice_by_index(begin = var_13814_begin_0, end = var_13814_end_0, end_mask = var_13814_end_mask_0, x = q_65_cast_fp16)[name = tensor("op_13814_cast_fp16")]; + tensor var_13818_begin_0 = const()[name = tensor("op_13818_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_13818_end_0 = const()[name = tensor("op_13818_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_13818_end_mask_0 = const()[name = tensor("op_13818_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13818_cast_fp16 = slice_by_index(begin = var_13818_begin_0, end = var_13818_end_0, end_mask = var_13818_end_mask_0, x = q_65_cast_fp16)[name = tensor("op_13818_cast_fp16")]; + tensor var_13822_begin_0 = const()[name = tensor("op_13822_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_13822_end_0 = const()[name = tensor("op_13822_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_13822_end_mask_0 = const()[name = tensor("op_13822_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13822_cast_fp16 = slice_by_index(begin = var_13822_begin_0, end = var_13822_end_0, end_mask = var_13822_end_mask_0, x = q_65_cast_fp16)[name = tensor("op_13822_cast_fp16")]; + tensor var_13826_begin_0 = const()[name = tensor("op_13826_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_13826_end_0 = const()[name = tensor("op_13826_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_13826_end_mask_0 = const()[name = tensor("op_13826_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13826_cast_fp16 = slice_by_index(begin = var_13826_begin_0, end = var_13826_end_0, end_mask = var_13826_end_mask_0, x = q_65_cast_fp16)[name = tensor("op_13826_cast_fp16")]; + tensor var_13830_begin_0 = const()[name = tensor("op_13830_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_13830_end_0 = const()[name = tensor("op_13830_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_13830_end_mask_0 = const()[name = tensor("op_13830_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13830_cast_fp16 = slice_by_index(begin = var_13830_begin_0, end = var_13830_end_0, end_mask = var_13830_end_mask_0, x = q_65_cast_fp16)[name = tensor("op_13830_cast_fp16")]; + tensor var_13834_begin_0 = const()[name = tensor("op_13834_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_13834_end_0 = const()[name = tensor("op_13834_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_13834_end_mask_0 = const()[name = tensor("op_13834_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13834_cast_fp16 = slice_by_index(begin = var_13834_begin_0, end = var_13834_end_0, end_mask = var_13834_end_mask_0, x = q_65_cast_fp16)[name = tensor("op_13834_cast_fp16")]; + tensor var_13838_begin_0 = const()[name = tensor("op_13838_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_13838_end_0 = const()[name = tensor("op_13838_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_13838_end_mask_0 = const()[name = tensor("op_13838_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13838_cast_fp16 = slice_by_index(begin = var_13838_begin_0, end = var_13838_end_0, end_mask = var_13838_end_mask_0, x = q_65_cast_fp16)[name = tensor("op_13838_cast_fp16")]; + tensor var_13842_begin_0 = const()[name = tensor("op_13842_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_13842_end_0 = const()[name = tensor("op_13842_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_13842_end_mask_0 = const()[name = tensor("op_13842_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13842_cast_fp16 = slice_by_index(begin = var_13842_begin_0, end = var_13842_end_0, end_mask = var_13842_end_mask_0, x = q_65_cast_fp16)[name = tensor("op_13842_cast_fp16")]; + tensor var_13846_begin_0 = const()[name = tensor("op_13846_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_13846_end_0 = const()[name = tensor("op_13846_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_13846_end_mask_0 = const()[name = tensor("op_13846_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13846_cast_fp16 = slice_by_index(begin = var_13846_begin_0, end = var_13846_end_0, end_mask = var_13846_end_mask_0, x = q_65_cast_fp16)[name = tensor("op_13846_cast_fp16")]; + tensor var_13850_begin_0 = const()[name = tensor("op_13850_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_13850_end_0 = const()[name = tensor("op_13850_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_13850_end_mask_0 = const()[name = tensor("op_13850_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13850_cast_fp16 = slice_by_index(begin = var_13850_begin_0, end = var_13850_end_0, end_mask = var_13850_end_mask_0, x = q_65_cast_fp16)[name = tensor("op_13850_cast_fp16")]; + tensor var_13854_begin_0 = const()[name = tensor("op_13854_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_13854_end_0 = const()[name = tensor("op_13854_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_13854_end_mask_0 = const()[name = tensor("op_13854_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13854_cast_fp16 = slice_by_index(begin = var_13854_begin_0, end = var_13854_end_0, end_mask = var_13854_end_mask_0, x = q_65_cast_fp16)[name = tensor("op_13854_cast_fp16")]; + tensor var_13858_begin_0 = const()[name = tensor("op_13858_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_13858_end_0 = const()[name = tensor("op_13858_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_13858_end_mask_0 = const()[name = tensor("op_13858_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13858_cast_fp16 = slice_by_index(begin = var_13858_begin_0, end = var_13858_end_0, end_mask = var_13858_end_mask_0, x = q_65_cast_fp16)[name = tensor("op_13858_cast_fp16")]; + tensor var_13862_begin_0 = const()[name = tensor("op_13862_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_13862_end_0 = const()[name = tensor("op_13862_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_13862_end_mask_0 = const()[name = tensor("op_13862_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13862_cast_fp16 = slice_by_index(begin = var_13862_begin_0, end = var_13862_end_0, end_mask = var_13862_end_mask_0, x = q_65_cast_fp16)[name = tensor("op_13862_cast_fp16")]; + tensor var_13866_begin_0 = const()[name = tensor("op_13866_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_13866_end_0 = const()[name = tensor("op_13866_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_13866_end_mask_0 = const()[name = tensor("op_13866_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13866_cast_fp16 = slice_by_index(begin = var_13866_begin_0, end = var_13866_end_0, end_mask = var_13866_end_mask_0, x = q_65_cast_fp16)[name = tensor("op_13866_cast_fp16")]; + tensor k_131_perm_0 = const()[name = tensor("k_131_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_13873_begin_0 = const()[name = tensor("op_13873_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_13873_end_0 = const()[name = tensor("op_13873_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_13873_end_mask_0 = const()[name = tensor("op_13873_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_131_cast_fp16 = transpose(perm = k_131_perm_0, x = k_129_cast_fp16)[name = tensor("transpose_107")]; + tensor var_13873_cast_fp16 = slice_by_index(begin = var_13873_begin_0, end = var_13873_end_0, end_mask = var_13873_end_mask_0, x = k_131_cast_fp16)[name = tensor("op_13873_cast_fp16")]; + tensor var_13877_begin_0 = const()[name = tensor("op_13877_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_13877_end_0 = const()[name = tensor("op_13877_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_13877_end_mask_0 = const()[name = tensor("op_13877_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13877_cast_fp16 = slice_by_index(begin = var_13877_begin_0, end = var_13877_end_0, end_mask = var_13877_end_mask_0, x = k_131_cast_fp16)[name = tensor("op_13877_cast_fp16")]; + tensor var_13881_begin_0 = const()[name = tensor("op_13881_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_13881_end_0 = const()[name = tensor("op_13881_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_13881_end_mask_0 = const()[name = tensor("op_13881_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13881_cast_fp16 = slice_by_index(begin = var_13881_begin_0, end = var_13881_end_0, end_mask = var_13881_end_mask_0, x = k_131_cast_fp16)[name = tensor("op_13881_cast_fp16")]; + tensor var_13885_begin_0 = const()[name = tensor("op_13885_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_13885_end_0 = const()[name = tensor("op_13885_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_13885_end_mask_0 = const()[name = tensor("op_13885_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13885_cast_fp16 = slice_by_index(begin = var_13885_begin_0, end = var_13885_end_0, end_mask = var_13885_end_mask_0, x = k_131_cast_fp16)[name = tensor("op_13885_cast_fp16")]; + tensor var_13889_begin_0 = const()[name = tensor("op_13889_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_13889_end_0 = const()[name = tensor("op_13889_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_13889_end_mask_0 = const()[name = tensor("op_13889_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13889_cast_fp16 = slice_by_index(begin = var_13889_begin_0, end = var_13889_end_0, end_mask = var_13889_end_mask_0, x = k_131_cast_fp16)[name = tensor("op_13889_cast_fp16")]; + tensor var_13893_begin_0 = const()[name = tensor("op_13893_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_13893_end_0 = const()[name = tensor("op_13893_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_13893_end_mask_0 = const()[name = tensor("op_13893_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13893_cast_fp16 = slice_by_index(begin = var_13893_begin_0, end = var_13893_end_0, end_mask = var_13893_end_mask_0, x = k_131_cast_fp16)[name = tensor("op_13893_cast_fp16")]; + tensor var_13897_begin_0 = const()[name = tensor("op_13897_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_13897_end_0 = const()[name = tensor("op_13897_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_13897_end_mask_0 = const()[name = tensor("op_13897_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13897_cast_fp16 = slice_by_index(begin = var_13897_begin_0, end = var_13897_end_0, end_mask = var_13897_end_mask_0, x = k_131_cast_fp16)[name = tensor("op_13897_cast_fp16")]; + tensor var_13901_begin_0 = const()[name = tensor("op_13901_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_13901_end_0 = const()[name = tensor("op_13901_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_13901_end_mask_0 = const()[name = tensor("op_13901_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13901_cast_fp16 = slice_by_index(begin = var_13901_begin_0, end = var_13901_end_0, end_mask = var_13901_end_mask_0, x = k_131_cast_fp16)[name = tensor("op_13901_cast_fp16")]; + tensor var_13905_begin_0 = const()[name = tensor("op_13905_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_13905_end_0 = const()[name = tensor("op_13905_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_13905_end_mask_0 = const()[name = tensor("op_13905_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13905_cast_fp16 = slice_by_index(begin = var_13905_begin_0, end = var_13905_end_0, end_mask = var_13905_end_mask_0, x = k_131_cast_fp16)[name = tensor("op_13905_cast_fp16")]; + tensor var_13909_begin_0 = const()[name = tensor("op_13909_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_13909_end_0 = const()[name = tensor("op_13909_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_13909_end_mask_0 = const()[name = tensor("op_13909_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13909_cast_fp16 = slice_by_index(begin = var_13909_begin_0, end = var_13909_end_0, end_mask = var_13909_end_mask_0, x = k_131_cast_fp16)[name = tensor("op_13909_cast_fp16")]; + tensor var_13913_begin_0 = const()[name = tensor("op_13913_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_13913_end_0 = const()[name = tensor("op_13913_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_13913_end_mask_0 = const()[name = tensor("op_13913_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13913_cast_fp16 = slice_by_index(begin = var_13913_begin_0, end = var_13913_end_0, end_mask = var_13913_end_mask_0, x = k_131_cast_fp16)[name = tensor("op_13913_cast_fp16")]; + tensor var_13917_begin_0 = const()[name = tensor("op_13917_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_13917_end_0 = const()[name = tensor("op_13917_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_13917_end_mask_0 = const()[name = tensor("op_13917_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13917_cast_fp16 = slice_by_index(begin = var_13917_begin_0, end = var_13917_end_0, end_mask = var_13917_end_mask_0, x = k_131_cast_fp16)[name = tensor("op_13917_cast_fp16")]; + tensor var_13921_begin_0 = const()[name = tensor("op_13921_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_13921_end_0 = const()[name = tensor("op_13921_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_13921_end_mask_0 = const()[name = tensor("op_13921_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13921_cast_fp16 = slice_by_index(begin = var_13921_begin_0, end = var_13921_end_0, end_mask = var_13921_end_mask_0, x = k_131_cast_fp16)[name = tensor("op_13921_cast_fp16")]; + tensor var_13925_begin_0 = const()[name = tensor("op_13925_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_13925_end_0 = const()[name = tensor("op_13925_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_13925_end_mask_0 = const()[name = tensor("op_13925_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13925_cast_fp16 = slice_by_index(begin = var_13925_begin_0, end = var_13925_end_0, end_mask = var_13925_end_mask_0, x = k_131_cast_fp16)[name = tensor("op_13925_cast_fp16")]; + tensor var_13929_begin_0 = const()[name = tensor("op_13929_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_13929_end_0 = const()[name = tensor("op_13929_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_13929_end_mask_0 = const()[name = tensor("op_13929_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13929_cast_fp16 = slice_by_index(begin = var_13929_begin_0, end = var_13929_end_0, end_mask = var_13929_end_mask_0, x = k_131_cast_fp16)[name = tensor("op_13929_cast_fp16")]; + tensor var_13933_begin_0 = const()[name = tensor("op_13933_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_13933_end_0 = const()[name = tensor("op_13933_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_13933_end_mask_0 = const()[name = tensor("op_13933_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13933_cast_fp16 = slice_by_index(begin = var_13933_begin_0, end = var_13933_end_0, end_mask = var_13933_end_mask_0, x = k_131_cast_fp16)[name = tensor("op_13933_cast_fp16")]; + tensor var_13937_begin_0 = const()[name = tensor("op_13937_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_13937_end_0 = const()[name = tensor("op_13937_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_13937_end_mask_0 = const()[name = tensor("op_13937_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13937_cast_fp16 = slice_by_index(begin = var_13937_begin_0, end = var_13937_end_0, end_mask = var_13937_end_mask_0, x = k_131_cast_fp16)[name = tensor("op_13937_cast_fp16")]; + tensor var_13941_begin_0 = const()[name = tensor("op_13941_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_13941_end_0 = const()[name = tensor("op_13941_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_13941_end_mask_0 = const()[name = tensor("op_13941_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13941_cast_fp16 = slice_by_index(begin = var_13941_begin_0, end = var_13941_end_0, end_mask = var_13941_end_mask_0, x = k_131_cast_fp16)[name = tensor("op_13941_cast_fp16")]; + tensor var_13945_begin_0 = const()[name = tensor("op_13945_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_13945_end_0 = const()[name = tensor("op_13945_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_13945_end_mask_0 = const()[name = tensor("op_13945_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13945_cast_fp16 = slice_by_index(begin = var_13945_begin_0, end = var_13945_end_0, end_mask = var_13945_end_mask_0, x = k_131_cast_fp16)[name = tensor("op_13945_cast_fp16")]; + tensor var_13949_begin_0 = const()[name = tensor("op_13949_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_13949_end_0 = const()[name = tensor("op_13949_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_13949_end_mask_0 = const()[name = tensor("op_13949_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13949_cast_fp16 = slice_by_index(begin = var_13949_begin_0, end = var_13949_end_0, end_mask = var_13949_end_mask_0, x = k_131_cast_fp16)[name = tensor("op_13949_cast_fp16")]; + tensor var_13951_begin_0 = const()[name = tensor("op_13951_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_13951_end_0 = const()[name = tensor("op_13951_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_13951_end_mask_0 = const()[name = tensor("op_13951_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13951_cast_fp16 = slice_by_index(begin = var_13951_begin_0, end = var_13951_end_0, end_mask = var_13951_end_mask_0, x = v_65_cast_fp16)[name = tensor("op_13951_cast_fp16")]; + tensor var_13955_begin_0 = const()[name = tensor("op_13955_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_13955_end_0 = const()[name = tensor("op_13955_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_13955_end_mask_0 = const()[name = tensor("op_13955_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13955_cast_fp16 = slice_by_index(begin = var_13955_begin_0, end = var_13955_end_0, end_mask = var_13955_end_mask_0, x = v_65_cast_fp16)[name = tensor("op_13955_cast_fp16")]; + tensor var_13959_begin_0 = const()[name = tensor("op_13959_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_13959_end_0 = const()[name = tensor("op_13959_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_13959_end_mask_0 = const()[name = tensor("op_13959_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13959_cast_fp16 = slice_by_index(begin = var_13959_begin_0, end = var_13959_end_0, end_mask = var_13959_end_mask_0, x = v_65_cast_fp16)[name = tensor("op_13959_cast_fp16")]; + tensor var_13963_begin_0 = const()[name = tensor("op_13963_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_13963_end_0 = const()[name = tensor("op_13963_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_13963_end_mask_0 = const()[name = tensor("op_13963_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13963_cast_fp16 = slice_by_index(begin = var_13963_begin_0, end = var_13963_end_0, end_mask = var_13963_end_mask_0, x = v_65_cast_fp16)[name = tensor("op_13963_cast_fp16")]; + tensor var_13967_begin_0 = const()[name = tensor("op_13967_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_13967_end_0 = const()[name = tensor("op_13967_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_13967_end_mask_0 = const()[name = tensor("op_13967_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13967_cast_fp16 = slice_by_index(begin = var_13967_begin_0, end = var_13967_end_0, end_mask = var_13967_end_mask_0, x = v_65_cast_fp16)[name = tensor("op_13967_cast_fp16")]; + tensor var_13971_begin_0 = const()[name = tensor("op_13971_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_13971_end_0 = const()[name = tensor("op_13971_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_13971_end_mask_0 = const()[name = tensor("op_13971_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13971_cast_fp16 = slice_by_index(begin = var_13971_begin_0, end = var_13971_end_0, end_mask = var_13971_end_mask_0, x = v_65_cast_fp16)[name = tensor("op_13971_cast_fp16")]; + tensor var_13975_begin_0 = const()[name = tensor("op_13975_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_13975_end_0 = const()[name = tensor("op_13975_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_13975_end_mask_0 = const()[name = tensor("op_13975_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13975_cast_fp16 = slice_by_index(begin = var_13975_begin_0, end = var_13975_end_0, end_mask = var_13975_end_mask_0, x = v_65_cast_fp16)[name = tensor("op_13975_cast_fp16")]; + tensor var_13979_begin_0 = const()[name = tensor("op_13979_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_13979_end_0 = const()[name = tensor("op_13979_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_13979_end_mask_0 = const()[name = tensor("op_13979_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13979_cast_fp16 = slice_by_index(begin = var_13979_begin_0, end = var_13979_end_0, end_mask = var_13979_end_mask_0, x = v_65_cast_fp16)[name = tensor("op_13979_cast_fp16")]; + tensor var_13983_begin_0 = const()[name = tensor("op_13983_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_13983_end_0 = const()[name = tensor("op_13983_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_13983_end_mask_0 = const()[name = tensor("op_13983_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13983_cast_fp16 = slice_by_index(begin = var_13983_begin_0, end = var_13983_end_0, end_mask = var_13983_end_mask_0, x = v_65_cast_fp16)[name = tensor("op_13983_cast_fp16")]; + tensor var_13987_begin_0 = const()[name = tensor("op_13987_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_13987_end_0 = const()[name = tensor("op_13987_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_13987_end_mask_0 = const()[name = tensor("op_13987_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13987_cast_fp16 = slice_by_index(begin = var_13987_begin_0, end = var_13987_end_0, end_mask = var_13987_end_mask_0, x = v_65_cast_fp16)[name = tensor("op_13987_cast_fp16")]; + tensor var_13991_begin_0 = const()[name = tensor("op_13991_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_13991_end_0 = const()[name = tensor("op_13991_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_13991_end_mask_0 = const()[name = tensor("op_13991_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13991_cast_fp16 = slice_by_index(begin = var_13991_begin_0, end = var_13991_end_0, end_mask = var_13991_end_mask_0, x = v_65_cast_fp16)[name = tensor("op_13991_cast_fp16")]; + tensor var_13995_begin_0 = const()[name = tensor("op_13995_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_13995_end_0 = const()[name = tensor("op_13995_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_13995_end_mask_0 = const()[name = tensor("op_13995_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13995_cast_fp16 = slice_by_index(begin = var_13995_begin_0, end = var_13995_end_0, end_mask = var_13995_end_mask_0, x = v_65_cast_fp16)[name = tensor("op_13995_cast_fp16")]; + tensor var_13999_begin_0 = const()[name = tensor("op_13999_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_13999_end_0 = const()[name = tensor("op_13999_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_13999_end_mask_0 = const()[name = tensor("op_13999_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13999_cast_fp16 = slice_by_index(begin = var_13999_begin_0, end = var_13999_end_0, end_mask = var_13999_end_mask_0, x = v_65_cast_fp16)[name = tensor("op_13999_cast_fp16")]; + tensor var_14003_begin_0 = const()[name = tensor("op_14003_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_14003_end_0 = const()[name = tensor("op_14003_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_14003_end_mask_0 = const()[name = tensor("op_14003_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14003_cast_fp16 = slice_by_index(begin = var_14003_begin_0, end = var_14003_end_0, end_mask = var_14003_end_mask_0, x = v_65_cast_fp16)[name = tensor("op_14003_cast_fp16")]; + tensor var_14007_begin_0 = const()[name = tensor("op_14007_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_14007_end_0 = const()[name = tensor("op_14007_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_14007_end_mask_0 = const()[name = tensor("op_14007_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14007_cast_fp16 = slice_by_index(begin = var_14007_begin_0, end = var_14007_end_0, end_mask = var_14007_end_mask_0, x = v_65_cast_fp16)[name = tensor("op_14007_cast_fp16")]; + tensor var_14011_begin_0 = const()[name = tensor("op_14011_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_14011_end_0 = const()[name = tensor("op_14011_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_14011_end_mask_0 = const()[name = tensor("op_14011_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14011_cast_fp16 = slice_by_index(begin = var_14011_begin_0, end = var_14011_end_0, end_mask = var_14011_end_mask_0, x = v_65_cast_fp16)[name = tensor("op_14011_cast_fp16")]; + tensor var_14015_begin_0 = const()[name = tensor("op_14015_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_14015_end_0 = const()[name = tensor("op_14015_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_14015_end_mask_0 = const()[name = tensor("op_14015_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14015_cast_fp16 = slice_by_index(begin = var_14015_begin_0, end = var_14015_end_0, end_mask = var_14015_end_mask_0, x = v_65_cast_fp16)[name = tensor("op_14015_cast_fp16")]; + tensor var_14019_begin_0 = const()[name = tensor("op_14019_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_14019_end_0 = const()[name = tensor("op_14019_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_14019_end_mask_0 = const()[name = tensor("op_14019_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14019_cast_fp16 = slice_by_index(begin = var_14019_begin_0, end = var_14019_end_0, end_mask = var_14019_end_mask_0, x = v_65_cast_fp16)[name = tensor("op_14019_cast_fp16")]; + tensor var_14023_begin_0 = const()[name = tensor("op_14023_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_14023_end_0 = const()[name = tensor("op_14023_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_14023_end_mask_0 = const()[name = tensor("op_14023_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14023_cast_fp16 = slice_by_index(begin = var_14023_begin_0, end = var_14023_end_0, end_mask = var_14023_end_mask_0, x = v_65_cast_fp16)[name = tensor("op_14023_cast_fp16")]; + tensor var_14027_begin_0 = const()[name = tensor("op_14027_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_14027_end_0 = const()[name = tensor("op_14027_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_14027_end_mask_0 = const()[name = tensor("op_14027_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14027_cast_fp16 = slice_by_index(begin = var_14027_begin_0, end = var_14027_end_0, end_mask = var_14027_end_mask_0, x = v_65_cast_fp16)[name = tensor("op_14027_cast_fp16")]; + tensor var_14031_equation_0 = const()[name = tensor("op_14031_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14031_cast_fp16 = einsum(equation = var_14031_equation_0, values = (var_13873_cast_fp16, var_13790_cast_fp16))[name = tensor("op_14031_cast_fp16")]; + tensor var_14032_to_fp16 = const()[name = tensor("op_14032_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1121_cast_fp16 = mul(x = var_14031_cast_fp16, y = var_14032_to_fp16)[name = tensor("aw_1121_cast_fp16")]; + tensor var_14035_equation_0 = const()[name = tensor("op_14035_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14035_cast_fp16 = einsum(equation = var_14035_equation_0, values = (var_13877_cast_fp16, var_13794_cast_fp16))[name = tensor("op_14035_cast_fp16")]; + tensor var_14036_to_fp16 = const()[name = tensor("op_14036_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1123_cast_fp16 = mul(x = var_14035_cast_fp16, y = var_14036_to_fp16)[name = tensor("aw_1123_cast_fp16")]; + tensor var_14039_equation_0 = const()[name = tensor("op_14039_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14039_cast_fp16 = einsum(equation = var_14039_equation_0, values = (var_13881_cast_fp16, var_13798_cast_fp16))[name = tensor("op_14039_cast_fp16")]; + tensor var_14040_to_fp16 = const()[name = tensor("op_14040_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1125_cast_fp16 = mul(x = var_14039_cast_fp16, y = var_14040_to_fp16)[name = tensor("aw_1125_cast_fp16")]; + tensor var_14043_equation_0 = const()[name = tensor("op_14043_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14043_cast_fp16 = einsum(equation = var_14043_equation_0, values = (var_13885_cast_fp16, var_13802_cast_fp16))[name = tensor("op_14043_cast_fp16")]; + tensor var_14044_to_fp16 = const()[name = tensor("op_14044_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1127_cast_fp16 = mul(x = var_14043_cast_fp16, y = var_14044_to_fp16)[name = tensor("aw_1127_cast_fp16")]; + tensor var_14047_equation_0 = const()[name = tensor("op_14047_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14047_cast_fp16 = einsum(equation = var_14047_equation_0, values = (var_13889_cast_fp16, var_13806_cast_fp16))[name = tensor("op_14047_cast_fp16")]; + tensor var_14048_to_fp16 = const()[name = tensor("op_14048_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1129_cast_fp16 = mul(x = var_14047_cast_fp16, y = var_14048_to_fp16)[name = tensor("aw_1129_cast_fp16")]; + tensor var_14051_equation_0 = const()[name = tensor("op_14051_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14051_cast_fp16 = einsum(equation = var_14051_equation_0, values = (var_13893_cast_fp16, var_13810_cast_fp16))[name = tensor("op_14051_cast_fp16")]; + tensor var_14052_to_fp16 = const()[name = tensor("op_14052_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1131_cast_fp16 = mul(x = var_14051_cast_fp16, y = var_14052_to_fp16)[name = tensor("aw_1131_cast_fp16")]; + tensor var_14055_equation_0 = const()[name = tensor("op_14055_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14055_cast_fp16 = einsum(equation = var_14055_equation_0, values = (var_13897_cast_fp16, var_13814_cast_fp16))[name = tensor("op_14055_cast_fp16")]; + tensor var_14056_to_fp16 = const()[name = tensor("op_14056_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1133_cast_fp16 = mul(x = var_14055_cast_fp16, y = var_14056_to_fp16)[name = tensor("aw_1133_cast_fp16")]; + tensor var_14059_equation_0 = const()[name = tensor("op_14059_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14059_cast_fp16 = einsum(equation = var_14059_equation_0, values = (var_13901_cast_fp16, var_13818_cast_fp16))[name = tensor("op_14059_cast_fp16")]; + tensor var_14060_to_fp16 = const()[name = tensor("op_14060_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1135_cast_fp16 = mul(x = var_14059_cast_fp16, y = var_14060_to_fp16)[name = tensor("aw_1135_cast_fp16")]; + tensor var_14063_equation_0 = const()[name = tensor("op_14063_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14063_cast_fp16 = einsum(equation = var_14063_equation_0, values = (var_13905_cast_fp16, var_13822_cast_fp16))[name = tensor("op_14063_cast_fp16")]; + tensor var_14064_to_fp16 = const()[name = tensor("op_14064_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1137_cast_fp16 = mul(x = var_14063_cast_fp16, y = var_14064_to_fp16)[name = tensor("aw_1137_cast_fp16")]; + tensor var_14067_equation_0 = const()[name = tensor("op_14067_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14067_cast_fp16 = einsum(equation = var_14067_equation_0, values = (var_13909_cast_fp16, var_13826_cast_fp16))[name = tensor("op_14067_cast_fp16")]; + tensor var_14068_to_fp16 = const()[name = tensor("op_14068_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1139_cast_fp16 = mul(x = var_14067_cast_fp16, y = var_14068_to_fp16)[name = tensor("aw_1139_cast_fp16")]; + tensor var_14071_equation_0 = const()[name = tensor("op_14071_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14071_cast_fp16 = einsum(equation = var_14071_equation_0, values = (var_13913_cast_fp16, var_13830_cast_fp16))[name = tensor("op_14071_cast_fp16")]; + tensor var_14072_to_fp16 = const()[name = tensor("op_14072_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1141_cast_fp16 = mul(x = var_14071_cast_fp16, y = var_14072_to_fp16)[name = tensor("aw_1141_cast_fp16")]; + tensor var_14075_equation_0 = const()[name = tensor("op_14075_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14075_cast_fp16 = einsum(equation = var_14075_equation_0, values = (var_13917_cast_fp16, var_13834_cast_fp16))[name = tensor("op_14075_cast_fp16")]; + tensor var_14076_to_fp16 = const()[name = tensor("op_14076_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1143_cast_fp16 = mul(x = var_14075_cast_fp16, y = var_14076_to_fp16)[name = tensor("aw_1143_cast_fp16")]; + tensor var_14079_equation_0 = const()[name = tensor("op_14079_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14079_cast_fp16 = einsum(equation = var_14079_equation_0, values = (var_13921_cast_fp16, var_13838_cast_fp16))[name = tensor("op_14079_cast_fp16")]; + tensor var_14080_to_fp16 = const()[name = tensor("op_14080_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1145_cast_fp16 = mul(x = var_14079_cast_fp16, y = var_14080_to_fp16)[name = tensor("aw_1145_cast_fp16")]; + tensor var_14083_equation_0 = const()[name = tensor("op_14083_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14083_cast_fp16 = einsum(equation = var_14083_equation_0, values = (var_13925_cast_fp16, var_13842_cast_fp16))[name = tensor("op_14083_cast_fp16")]; + tensor var_14084_to_fp16 = const()[name = tensor("op_14084_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1147_cast_fp16 = mul(x = var_14083_cast_fp16, y = var_14084_to_fp16)[name = tensor("aw_1147_cast_fp16")]; + tensor var_14087_equation_0 = const()[name = tensor("op_14087_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14087_cast_fp16 = einsum(equation = var_14087_equation_0, values = (var_13929_cast_fp16, var_13846_cast_fp16))[name = tensor("op_14087_cast_fp16")]; + tensor var_14088_to_fp16 = const()[name = tensor("op_14088_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1149_cast_fp16 = mul(x = var_14087_cast_fp16, y = var_14088_to_fp16)[name = tensor("aw_1149_cast_fp16")]; + tensor var_14091_equation_0 = const()[name = tensor("op_14091_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14091_cast_fp16 = einsum(equation = var_14091_equation_0, values = (var_13933_cast_fp16, var_13850_cast_fp16))[name = tensor("op_14091_cast_fp16")]; + tensor var_14092_to_fp16 = const()[name = tensor("op_14092_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1151_cast_fp16 = mul(x = var_14091_cast_fp16, y = var_14092_to_fp16)[name = tensor("aw_1151_cast_fp16")]; + tensor var_14095_equation_0 = const()[name = tensor("op_14095_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14095_cast_fp16 = einsum(equation = var_14095_equation_0, values = (var_13937_cast_fp16, var_13854_cast_fp16))[name = tensor("op_14095_cast_fp16")]; + tensor var_14096_to_fp16 = const()[name = tensor("op_14096_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1153_cast_fp16 = mul(x = var_14095_cast_fp16, y = var_14096_to_fp16)[name = tensor("aw_1153_cast_fp16")]; + tensor var_14099_equation_0 = const()[name = tensor("op_14099_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14099_cast_fp16 = einsum(equation = var_14099_equation_0, values = (var_13941_cast_fp16, var_13858_cast_fp16))[name = tensor("op_14099_cast_fp16")]; + tensor var_14100_to_fp16 = const()[name = tensor("op_14100_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1155_cast_fp16 = mul(x = var_14099_cast_fp16, y = var_14100_to_fp16)[name = tensor("aw_1155_cast_fp16")]; + tensor var_14103_equation_0 = const()[name = tensor("op_14103_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14103_cast_fp16 = einsum(equation = var_14103_equation_0, values = (var_13945_cast_fp16, var_13862_cast_fp16))[name = tensor("op_14103_cast_fp16")]; + tensor var_14104_to_fp16 = const()[name = tensor("op_14104_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1157_cast_fp16 = mul(x = var_14103_cast_fp16, y = var_14104_to_fp16)[name = tensor("aw_1157_cast_fp16")]; + tensor var_14107_equation_0 = const()[name = tensor("op_14107_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14107_cast_fp16 = einsum(equation = var_14107_equation_0, values = (var_13949_cast_fp16, var_13866_cast_fp16))[name = tensor("op_14107_cast_fp16")]; + tensor var_14108_to_fp16 = const()[name = tensor("op_14108_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1159_cast_fp16 = mul(x = var_14107_cast_fp16, y = var_14108_to_fp16)[name = tensor("aw_1159_cast_fp16")]; + tensor var_14110_cast_fp16 = softmax(axis = var_2624, x = aw_1121_cast_fp16)[name = tensor("op_14110_cast_fp16")]; + tensor var_14111_cast_fp16 = softmax(axis = var_2624, x = aw_1123_cast_fp16)[name = tensor("op_14111_cast_fp16")]; + tensor var_14112_cast_fp16 = softmax(axis = var_2624, x = aw_1125_cast_fp16)[name = tensor("op_14112_cast_fp16")]; + tensor var_14113_cast_fp16 = softmax(axis = var_2624, x = aw_1127_cast_fp16)[name = tensor("op_14113_cast_fp16")]; + tensor var_14114_cast_fp16 = softmax(axis = var_2624, x = aw_1129_cast_fp16)[name = tensor("op_14114_cast_fp16")]; + tensor var_14115_cast_fp16 = softmax(axis = var_2624, x = aw_1131_cast_fp16)[name = tensor("op_14115_cast_fp16")]; + tensor var_14116_cast_fp16 = softmax(axis = var_2624, x = aw_1133_cast_fp16)[name = tensor("op_14116_cast_fp16")]; + tensor var_14117_cast_fp16 = softmax(axis = var_2624, x = aw_1135_cast_fp16)[name = tensor("op_14117_cast_fp16")]; + tensor var_14118_cast_fp16 = softmax(axis = var_2624, x = aw_1137_cast_fp16)[name = tensor("op_14118_cast_fp16")]; + tensor var_14119_cast_fp16 = softmax(axis = var_2624, x = aw_1139_cast_fp16)[name = tensor("op_14119_cast_fp16")]; + tensor var_14120_cast_fp16 = softmax(axis = var_2624, x = aw_1141_cast_fp16)[name = tensor("op_14120_cast_fp16")]; + tensor var_14121_cast_fp16 = softmax(axis = var_2624, x = aw_1143_cast_fp16)[name = tensor("op_14121_cast_fp16")]; + tensor var_14122_cast_fp16 = softmax(axis = var_2624, x = aw_1145_cast_fp16)[name = tensor("op_14122_cast_fp16")]; + tensor var_14123_cast_fp16 = softmax(axis = var_2624, x = aw_1147_cast_fp16)[name = tensor("op_14123_cast_fp16")]; + tensor var_14124_cast_fp16 = softmax(axis = var_2624, x = aw_1149_cast_fp16)[name = tensor("op_14124_cast_fp16")]; + tensor var_14125_cast_fp16 = softmax(axis = var_2624, x = aw_1151_cast_fp16)[name = tensor("op_14125_cast_fp16")]; + tensor var_14126_cast_fp16 = softmax(axis = var_2624, x = aw_1153_cast_fp16)[name = tensor("op_14126_cast_fp16")]; + tensor var_14127_cast_fp16 = softmax(axis = var_2624, x = aw_1155_cast_fp16)[name = tensor("op_14127_cast_fp16")]; + tensor var_14128_cast_fp16 = softmax(axis = var_2624, x = aw_1157_cast_fp16)[name = tensor("op_14128_cast_fp16")]; + tensor var_14129_cast_fp16 = softmax(axis = var_2624, x = aw_1159_cast_fp16)[name = tensor("op_14129_cast_fp16")]; + tensor var_14131_equation_0 = const()[name = tensor("op_14131_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14131_cast_fp16 = einsum(equation = var_14131_equation_0, values = (var_13951_cast_fp16, var_14110_cast_fp16))[name = tensor("op_14131_cast_fp16")]; + tensor var_14133_equation_0 = const()[name = tensor("op_14133_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14133_cast_fp16 = einsum(equation = var_14133_equation_0, values = (var_13955_cast_fp16, var_14111_cast_fp16))[name = tensor("op_14133_cast_fp16")]; + tensor var_14135_equation_0 = const()[name = tensor("op_14135_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14135_cast_fp16 = einsum(equation = var_14135_equation_0, values = (var_13959_cast_fp16, var_14112_cast_fp16))[name = tensor("op_14135_cast_fp16")]; + tensor var_14137_equation_0 = const()[name = tensor("op_14137_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14137_cast_fp16 = einsum(equation = var_14137_equation_0, values = (var_13963_cast_fp16, var_14113_cast_fp16))[name = tensor("op_14137_cast_fp16")]; + tensor var_14139_equation_0 = const()[name = tensor("op_14139_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14139_cast_fp16 = einsum(equation = var_14139_equation_0, values = (var_13967_cast_fp16, var_14114_cast_fp16))[name = tensor("op_14139_cast_fp16")]; + tensor var_14141_equation_0 = const()[name = tensor("op_14141_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14141_cast_fp16 = einsum(equation = var_14141_equation_0, values = (var_13971_cast_fp16, var_14115_cast_fp16))[name = tensor("op_14141_cast_fp16")]; + tensor var_14143_equation_0 = const()[name = tensor("op_14143_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14143_cast_fp16 = einsum(equation = var_14143_equation_0, values = (var_13975_cast_fp16, var_14116_cast_fp16))[name = tensor("op_14143_cast_fp16")]; + tensor var_14145_equation_0 = const()[name = tensor("op_14145_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14145_cast_fp16 = einsum(equation = var_14145_equation_0, values = (var_13979_cast_fp16, var_14117_cast_fp16))[name = tensor("op_14145_cast_fp16")]; + tensor var_14147_equation_0 = const()[name = tensor("op_14147_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14147_cast_fp16 = einsum(equation = var_14147_equation_0, values = (var_13983_cast_fp16, var_14118_cast_fp16))[name = tensor("op_14147_cast_fp16")]; + tensor var_14149_equation_0 = const()[name = tensor("op_14149_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14149_cast_fp16 = einsum(equation = var_14149_equation_0, values = (var_13987_cast_fp16, var_14119_cast_fp16))[name = tensor("op_14149_cast_fp16")]; + tensor var_14151_equation_0 = const()[name = tensor("op_14151_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14151_cast_fp16 = einsum(equation = var_14151_equation_0, values = (var_13991_cast_fp16, var_14120_cast_fp16))[name = tensor("op_14151_cast_fp16")]; + tensor var_14153_equation_0 = const()[name = tensor("op_14153_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14153_cast_fp16 = einsum(equation = var_14153_equation_0, values = (var_13995_cast_fp16, var_14121_cast_fp16))[name = tensor("op_14153_cast_fp16")]; + tensor var_14155_equation_0 = const()[name = tensor("op_14155_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14155_cast_fp16 = einsum(equation = var_14155_equation_0, values = (var_13999_cast_fp16, var_14122_cast_fp16))[name = tensor("op_14155_cast_fp16")]; + tensor var_14157_equation_0 = const()[name = tensor("op_14157_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14157_cast_fp16 = einsum(equation = var_14157_equation_0, values = (var_14003_cast_fp16, var_14123_cast_fp16))[name = tensor("op_14157_cast_fp16")]; + tensor var_14159_equation_0 = const()[name = tensor("op_14159_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14159_cast_fp16 = einsum(equation = var_14159_equation_0, values = (var_14007_cast_fp16, var_14124_cast_fp16))[name = tensor("op_14159_cast_fp16")]; + tensor var_14161_equation_0 = const()[name = tensor("op_14161_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14161_cast_fp16 = einsum(equation = var_14161_equation_0, values = (var_14011_cast_fp16, var_14125_cast_fp16))[name = tensor("op_14161_cast_fp16")]; + tensor var_14163_equation_0 = const()[name = tensor("op_14163_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14163_cast_fp16 = einsum(equation = var_14163_equation_0, values = (var_14015_cast_fp16, var_14126_cast_fp16))[name = tensor("op_14163_cast_fp16")]; + tensor var_14165_equation_0 = const()[name = tensor("op_14165_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14165_cast_fp16 = einsum(equation = var_14165_equation_0, values = (var_14019_cast_fp16, var_14127_cast_fp16))[name = tensor("op_14165_cast_fp16")]; + tensor var_14167_equation_0 = const()[name = tensor("op_14167_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14167_cast_fp16 = einsum(equation = var_14167_equation_0, values = (var_14023_cast_fp16, var_14128_cast_fp16))[name = tensor("op_14167_cast_fp16")]; + tensor var_14169_equation_0 = const()[name = tensor("op_14169_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14169_cast_fp16 = einsum(equation = var_14169_equation_0, values = (var_14027_cast_fp16, var_14129_cast_fp16))[name = tensor("op_14169_cast_fp16")]; + tensor input_245_interleave_0 = const()[name = tensor("input_245_interleave_0"), val = tensor(false)]; + tensor input_245_cast_fp16 = concat(axis = var_2624, interleave = input_245_interleave_0, values = (var_14131_cast_fp16, var_14133_cast_fp16, var_14135_cast_fp16, var_14137_cast_fp16, var_14139_cast_fp16, var_14141_cast_fp16, var_14143_cast_fp16, var_14145_cast_fp16, var_14147_cast_fp16, var_14149_cast_fp16, var_14151_cast_fp16, var_14153_cast_fp16, var_14155_cast_fp16, var_14157_cast_fp16, var_14159_cast_fp16, var_14161_cast_fp16, var_14163_cast_fp16, var_14165_cast_fp16, var_14167_cast_fp16, var_14169_cast_fp16))[name = tensor("input_245_cast_fp16")]; + tensor var_14175 = const()[name = tensor("op_14175"), val = tensor([1, 1])]; + tensor var_14177 = const()[name = tensor("op_14177"), val = tensor([1, 1])]; + tensor var_14179_pad_type_0 = const()[name = tensor("op_14179_pad_type_0"), val = tensor("custom")]; + tensor var_14179_pad_0 = const()[name = tensor("op_14179_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_2_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_2_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1110999232)))]; + tensor down_blocks_2_attentions_1_transformer_blocks_2_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_2_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1114276096)))]; + tensor var_14179_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_2_attn1_to_out_0_bias_to_fp16, dilations = var_14177, groups = var_2624, pad = var_14179_pad_0, pad_type = var_14179_pad_type_0, strides = var_14175, weight = down_blocks_2_attentions_1_transformer_blocks_2_attn1_to_out_0_weight_to_fp16, x = input_245_cast_fp16)[name = tensor("op_14179_cast_fp16")]; + tensor inputs_99_cast_fp16 = add(x = var_14179_cast_fp16, y = inputs_97_cast_fp16)[name = tensor("inputs_99_cast_fp16")]; + tensor hidden_states_151_axes_0 = const()[name = tensor("hidden_states_151_axes_0"), val = tensor([1])]; + tensor hidden_states_151_gamma_0_to_fp16 = const()[name = tensor("hidden_states_151_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1114278720)))]; + tensor hidden_states_151_beta_0_to_fp16 = const()[name = tensor("hidden_states_151_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1114281344)))]; + tensor var_14189_to_fp16 = const()[name = tensor("op_14189_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_151_cast_fp16 = layer_norm(axes = hidden_states_151_axes_0, beta = hidden_states_151_beta_0_to_fp16, epsilon = var_14189_to_fp16, gamma = hidden_states_151_gamma_0_to_fp16, x = inputs_99_cast_fp16)[name = tensor("hidden_states_151_cast_fp16")]; + tensor var_14204 = const()[name = tensor("op_14204"), val = tensor([1, 1])]; + tensor var_14206 = const()[name = tensor("op_14206"), val = tensor([1, 1])]; + tensor q_67_pad_type_0 = const()[name = tensor("q_67_pad_type_0"), val = tensor("custom")]; + tensor q_67_pad_0 = const()[name = tensor("q_67_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_2_attn2_to_q_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_2_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1114283968)))]; + tensor q_67_cast_fp16 = conv(dilations = var_14206, groups = var_2624, pad = q_67_pad_0, pad_type = q_67_pad_type_0, strides = var_14204, weight = down_blocks_2_attentions_1_transformer_blocks_2_attn2_to_q_weight_to_fp16, x = hidden_states_151_cast_fp16)[name = tensor("q_67_cast_fp16")]; + tensor var_14210 = const()[name = tensor("op_14210"), val = tensor([1, 1])]; + tensor var_14212 = const()[name = tensor("op_14212"), val = tensor([1, 1])]; + tensor k_133_pad_type_0 = const()[name = tensor("k_133_pad_type_0"), val = tensor("custom")]; + tensor k_133_pad_0 = const()[name = tensor("k_133_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_2_attn2_to_k_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_2_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1117560832)))]; + tensor k_133_cast_fp16 = conv(dilations = var_14212, groups = var_2624, pad = k_133_pad_0, pad_type = k_133_pad_type_0, strides = var_14210, weight = down_blocks_2_attentions_1_transformer_blocks_2_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_133_cast_fp16")]; + tensor var_14216 = const()[name = tensor("op_14216"), val = tensor([1, 1])]; + tensor var_14218 = const()[name = tensor("op_14218"), val = tensor([1, 1])]; + tensor v_67_pad_type_0 = const()[name = tensor("v_67_pad_type_0"), val = tensor("custom")]; + tensor v_67_pad_0 = const()[name = tensor("v_67_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_2_attn2_to_v_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_2_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1122803776)))]; + tensor v_67_cast_fp16 = conv(dilations = var_14218, groups = var_2624, pad = v_67_pad_0, pad_type = v_67_pad_type_0, strides = var_14216, weight = down_blocks_2_attentions_1_transformer_blocks_2_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_67_cast_fp16")]; + tensor var_14222_begin_0 = const()[name = tensor("op_14222_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14222_end_0 = const()[name = tensor("op_14222_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_14222_end_mask_0 = const()[name = tensor("op_14222_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14222_cast_fp16 = slice_by_index(begin = var_14222_begin_0, end = var_14222_end_0, end_mask = var_14222_end_mask_0, x = q_67_cast_fp16)[name = tensor("op_14222_cast_fp16")]; + tensor var_14226_begin_0 = const()[name = tensor("op_14226_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_14226_end_0 = const()[name = tensor("op_14226_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_14226_end_mask_0 = const()[name = tensor("op_14226_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14226_cast_fp16 = slice_by_index(begin = var_14226_begin_0, end = var_14226_end_0, end_mask = var_14226_end_mask_0, x = q_67_cast_fp16)[name = tensor("op_14226_cast_fp16")]; + tensor var_14230_begin_0 = const()[name = tensor("op_14230_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_14230_end_0 = const()[name = tensor("op_14230_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_14230_end_mask_0 = const()[name = tensor("op_14230_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14230_cast_fp16 = slice_by_index(begin = var_14230_begin_0, end = var_14230_end_0, end_mask = var_14230_end_mask_0, x = q_67_cast_fp16)[name = tensor("op_14230_cast_fp16")]; + tensor var_14234_begin_0 = const()[name = tensor("op_14234_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_14234_end_0 = const()[name = tensor("op_14234_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_14234_end_mask_0 = const()[name = tensor("op_14234_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14234_cast_fp16 = slice_by_index(begin = var_14234_begin_0, end = var_14234_end_0, end_mask = var_14234_end_mask_0, x = q_67_cast_fp16)[name = tensor("op_14234_cast_fp16")]; + tensor var_14238_begin_0 = const()[name = tensor("op_14238_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_14238_end_0 = const()[name = tensor("op_14238_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_14238_end_mask_0 = const()[name = tensor("op_14238_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14238_cast_fp16 = slice_by_index(begin = var_14238_begin_0, end = var_14238_end_0, end_mask = var_14238_end_mask_0, x = q_67_cast_fp16)[name = tensor("op_14238_cast_fp16")]; + tensor var_14242_begin_0 = const()[name = tensor("op_14242_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_14242_end_0 = const()[name = tensor("op_14242_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_14242_end_mask_0 = const()[name = tensor("op_14242_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14242_cast_fp16 = slice_by_index(begin = var_14242_begin_0, end = var_14242_end_0, end_mask = var_14242_end_mask_0, x = q_67_cast_fp16)[name = tensor("op_14242_cast_fp16")]; + tensor var_14246_begin_0 = const()[name = tensor("op_14246_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_14246_end_0 = const()[name = tensor("op_14246_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_14246_end_mask_0 = const()[name = tensor("op_14246_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14246_cast_fp16 = slice_by_index(begin = var_14246_begin_0, end = var_14246_end_0, end_mask = var_14246_end_mask_0, x = q_67_cast_fp16)[name = tensor("op_14246_cast_fp16")]; + tensor var_14250_begin_0 = const()[name = tensor("op_14250_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_14250_end_0 = const()[name = tensor("op_14250_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_14250_end_mask_0 = const()[name = tensor("op_14250_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14250_cast_fp16 = slice_by_index(begin = var_14250_begin_0, end = var_14250_end_0, end_mask = var_14250_end_mask_0, x = q_67_cast_fp16)[name = tensor("op_14250_cast_fp16")]; + tensor var_14254_begin_0 = const()[name = tensor("op_14254_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_14254_end_0 = const()[name = tensor("op_14254_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_14254_end_mask_0 = const()[name = tensor("op_14254_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14254_cast_fp16 = slice_by_index(begin = var_14254_begin_0, end = var_14254_end_0, end_mask = var_14254_end_mask_0, x = q_67_cast_fp16)[name = tensor("op_14254_cast_fp16")]; + tensor var_14258_begin_0 = const()[name = tensor("op_14258_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_14258_end_0 = const()[name = tensor("op_14258_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_14258_end_mask_0 = const()[name = tensor("op_14258_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14258_cast_fp16 = slice_by_index(begin = var_14258_begin_0, end = var_14258_end_0, end_mask = var_14258_end_mask_0, x = q_67_cast_fp16)[name = tensor("op_14258_cast_fp16")]; + tensor var_14262_begin_0 = const()[name = tensor("op_14262_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_14262_end_0 = const()[name = tensor("op_14262_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_14262_end_mask_0 = const()[name = tensor("op_14262_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14262_cast_fp16 = slice_by_index(begin = var_14262_begin_0, end = var_14262_end_0, end_mask = var_14262_end_mask_0, x = q_67_cast_fp16)[name = tensor("op_14262_cast_fp16")]; + tensor var_14266_begin_0 = const()[name = tensor("op_14266_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_14266_end_0 = const()[name = tensor("op_14266_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_14266_end_mask_0 = const()[name = tensor("op_14266_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14266_cast_fp16 = slice_by_index(begin = var_14266_begin_0, end = var_14266_end_0, end_mask = var_14266_end_mask_0, x = q_67_cast_fp16)[name = tensor("op_14266_cast_fp16")]; + tensor var_14270_begin_0 = const()[name = tensor("op_14270_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_14270_end_0 = const()[name = tensor("op_14270_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_14270_end_mask_0 = const()[name = tensor("op_14270_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14270_cast_fp16 = slice_by_index(begin = var_14270_begin_0, end = var_14270_end_0, end_mask = var_14270_end_mask_0, x = q_67_cast_fp16)[name = tensor("op_14270_cast_fp16")]; + tensor var_14274_begin_0 = const()[name = tensor("op_14274_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_14274_end_0 = const()[name = tensor("op_14274_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_14274_end_mask_0 = const()[name = tensor("op_14274_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14274_cast_fp16 = slice_by_index(begin = var_14274_begin_0, end = var_14274_end_0, end_mask = var_14274_end_mask_0, x = q_67_cast_fp16)[name = tensor("op_14274_cast_fp16")]; + tensor var_14278_begin_0 = const()[name = tensor("op_14278_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_14278_end_0 = const()[name = tensor("op_14278_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_14278_end_mask_0 = const()[name = tensor("op_14278_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14278_cast_fp16 = slice_by_index(begin = var_14278_begin_0, end = var_14278_end_0, end_mask = var_14278_end_mask_0, x = q_67_cast_fp16)[name = tensor("op_14278_cast_fp16")]; + tensor var_14282_begin_0 = const()[name = tensor("op_14282_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_14282_end_0 = const()[name = tensor("op_14282_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_14282_end_mask_0 = const()[name = tensor("op_14282_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14282_cast_fp16 = slice_by_index(begin = var_14282_begin_0, end = var_14282_end_0, end_mask = var_14282_end_mask_0, x = q_67_cast_fp16)[name = tensor("op_14282_cast_fp16")]; + tensor var_14286_begin_0 = const()[name = tensor("op_14286_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_14286_end_0 = const()[name = tensor("op_14286_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_14286_end_mask_0 = const()[name = tensor("op_14286_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14286_cast_fp16 = slice_by_index(begin = var_14286_begin_0, end = var_14286_end_0, end_mask = var_14286_end_mask_0, x = q_67_cast_fp16)[name = tensor("op_14286_cast_fp16")]; + tensor var_14290_begin_0 = const()[name = tensor("op_14290_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_14290_end_0 = const()[name = tensor("op_14290_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_14290_end_mask_0 = const()[name = tensor("op_14290_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14290_cast_fp16 = slice_by_index(begin = var_14290_begin_0, end = var_14290_end_0, end_mask = var_14290_end_mask_0, x = q_67_cast_fp16)[name = tensor("op_14290_cast_fp16")]; + tensor var_14294_begin_0 = const()[name = tensor("op_14294_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_14294_end_0 = const()[name = tensor("op_14294_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_14294_end_mask_0 = const()[name = tensor("op_14294_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14294_cast_fp16 = slice_by_index(begin = var_14294_begin_0, end = var_14294_end_0, end_mask = var_14294_end_mask_0, x = q_67_cast_fp16)[name = tensor("op_14294_cast_fp16")]; + tensor var_14298_begin_0 = const()[name = tensor("op_14298_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_14298_end_0 = const()[name = tensor("op_14298_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_14298_end_mask_0 = const()[name = tensor("op_14298_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14298_cast_fp16 = slice_by_index(begin = var_14298_begin_0, end = var_14298_end_0, end_mask = var_14298_end_mask_0, x = q_67_cast_fp16)[name = tensor("op_14298_cast_fp16")]; + tensor k_135_perm_0 = const()[name = tensor("k_135_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_14305_begin_0 = const()[name = tensor("op_14305_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14305_end_0 = const()[name = tensor("op_14305_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_14305_end_mask_0 = const()[name = tensor("op_14305_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_135_cast_fp16 = transpose(perm = k_135_perm_0, x = k_133_cast_fp16)[name = tensor("transpose_106")]; + tensor var_14305_cast_fp16 = slice_by_index(begin = var_14305_begin_0, end = var_14305_end_0, end_mask = var_14305_end_mask_0, x = k_135_cast_fp16)[name = tensor("op_14305_cast_fp16")]; + tensor var_14309_begin_0 = const()[name = tensor("op_14309_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_14309_end_0 = const()[name = tensor("op_14309_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_14309_end_mask_0 = const()[name = tensor("op_14309_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14309_cast_fp16 = slice_by_index(begin = var_14309_begin_0, end = var_14309_end_0, end_mask = var_14309_end_mask_0, x = k_135_cast_fp16)[name = tensor("op_14309_cast_fp16")]; + tensor var_14313_begin_0 = const()[name = tensor("op_14313_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_14313_end_0 = const()[name = tensor("op_14313_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_14313_end_mask_0 = const()[name = tensor("op_14313_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14313_cast_fp16 = slice_by_index(begin = var_14313_begin_0, end = var_14313_end_0, end_mask = var_14313_end_mask_0, x = k_135_cast_fp16)[name = tensor("op_14313_cast_fp16")]; + tensor var_14317_begin_0 = const()[name = tensor("op_14317_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_14317_end_0 = const()[name = tensor("op_14317_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_14317_end_mask_0 = const()[name = tensor("op_14317_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14317_cast_fp16 = slice_by_index(begin = var_14317_begin_0, end = var_14317_end_0, end_mask = var_14317_end_mask_0, x = k_135_cast_fp16)[name = tensor("op_14317_cast_fp16")]; + tensor var_14321_begin_0 = const()[name = tensor("op_14321_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_14321_end_0 = const()[name = tensor("op_14321_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_14321_end_mask_0 = const()[name = tensor("op_14321_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14321_cast_fp16 = slice_by_index(begin = var_14321_begin_0, end = var_14321_end_0, end_mask = var_14321_end_mask_0, x = k_135_cast_fp16)[name = tensor("op_14321_cast_fp16")]; + tensor var_14325_begin_0 = const()[name = tensor("op_14325_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_14325_end_0 = const()[name = tensor("op_14325_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_14325_end_mask_0 = const()[name = tensor("op_14325_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14325_cast_fp16 = slice_by_index(begin = var_14325_begin_0, end = var_14325_end_0, end_mask = var_14325_end_mask_0, x = k_135_cast_fp16)[name = tensor("op_14325_cast_fp16")]; + tensor var_14329_begin_0 = const()[name = tensor("op_14329_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_14329_end_0 = const()[name = tensor("op_14329_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_14329_end_mask_0 = const()[name = tensor("op_14329_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14329_cast_fp16 = slice_by_index(begin = var_14329_begin_0, end = var_14329_end_0, end_mask = var_14329_end_mask_0, x = k_135_cast_fp16)[name = tensor("op_14329_cast_fp16")]; + tensor var_14333_begin_0 = const()[name = tensor("op_14333_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_14333_end_0 = const()[name = tensor("op_14333_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_14333_end_mask_0 = const()[name = tensor("op_14333_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14333_cast_fp16 = slice_by_index(begin = var_14333_begin_0, end = var_14333_end_0, end_mask = var_14333_end_mask_0, x = k_135_cast_fp16)[name = tensor("op_14333_cast_fp16")]; + tensor var_14337_begin_0 = const()[name = tensor("op_14337_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_14337_end_0 = const()[name = tensor("op_14337_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_14337_end_mask_0 = const()[name = tensor("op_14337_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14337_cast_fp16 = slice_by_index(begin = var_14337_begin_0, end = var_14337_end_0, end_mask = var_14337_end_mask_0, x = k_135_cast_fp16)[name = tensor("op_14337_cast_fp16")]; + tensor var_14341_begin_0 = const()[name = tensor("op_14341_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_14341_end_0 = const()[name = tensor("op_14341_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_14341_end_mask_0 = const()[name = tensor("op_14341_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14341_cast_fp16 = slice_by_index(begin = var_14341_begin_0, end = var_14341_end_0, end_mask = var_14341_end_mask_0, x = k_135_cast_fp16)[name = tensor("op_14341_cast_fp16")]; + tensor var_14345_begin_0 = const()[name = tensor("op_14345_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_14345_end_0 = const()[name = tensor("op_14345_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_14345_end_mask_0 = const()[name = tensor("op_14345_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14345_cast_fp16 = slice_by_index(begin = var_14345_begin_0, end = var_14345_end_0, end_mask = var_14345_end_mask_0, x = k_135_cast_fp16)[name = tensor("op_14345_cast_fp16")]; + tensor var_14349_begin_0 = const()[name = tensor("op_14349_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_14349_end_0 = const()[name = tensor("op_14349_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_14349_end_mask_0 = const()[name = tensor("op_14349_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14349_cast_fp16 = slice_by_index(begin = var_14349_begin_0, end = var_14349_end_0, end_mask = var_14349_end_mask_0, x = k_135_cast_fp16)[name = tensor("op_14349_cast_fp16")]; + tensor var_14353_begin_0 = const()[name = tensor("op_14353_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_14353_end_0 = const()[name = tensor("op_14353_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_14353_end_mask_0 = const()[name = tensor("op_14353_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14353_cast_fp16 = slice_by_index(begin = var_14353_begin_0, end = var_14353_end_0, end_mask = var_14353_end_mask_0, x = k_135_cast_fp16)[name = tensor("op_14353_cast_fp16")]; + tensor var_14357_begin_0 = const()[name = tensor("op_14357_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_14357_end_0 = const()[name = tensor("op_14357_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_14357_end_mask_0 = const()[name = tensor("op_14357_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14357_cast_fp16 = slice_by_index(begin = var_14357_begin_0, end = var_14357_end_0, end_mask = var_14357_end_mask_0, x = k_135_cast_fp16)[name = tensor("op_14357_cast_fp16")]; + tensor var_14361_begin_0 = const()[name = tensor("op_14361_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_14361_end_0 = const()[name = tensor("op_14361_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_14361_end_mask_0 = const()[name = tensor("op_14361_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14361_cast_fp16 = slice_by_index(begin = var_14361_begin_0, end = var_14361_end_0, end_mask = var_14361_end_mask_0, x = k_135_cast_fp16)[name = tensor("op_14361_cast_fp16")]; + tensor var_14365_begin_0 = const()[name = tensor("op_14365_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_14365_end_0 = const()[name = tensor("op_14365_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_14365_end_mask_0 = const()[name = tensor("op_14365_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14365_cast_fp16 = slice_by_index(begin = var_14365_begin_0, end = var_14365_end_0, end_mask = var_14365_end_mask_0, x = k_135_cast_fp16)[name = tensor("op_14365_cast_fp16")]; + tensor var_14369_begin_0 = const()[name = tensor("op_14369_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_14369_end_0 = const()[name = tensor("op_14369_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_14369_end_mask_0 = const()[name = tensor("op_14369_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14369_cast_fp16 = slice_by_index(begin = var_14369_begin_0, end = var_14369_end_0, end_mask = var_14369_end_mask_0, x = k_135_cast_fp16)[name = tensor("op_14369_cast_fp16")]; + tensor var_14373_begin_0 = const()[name = tensor("op_14373_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_14373_end_0 = const()[name = tensor("op_14373_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_14373_end_mask_0 = const()[name = tensor("op_14373_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14373_cast_fp16 = slice_by_index(begin = var_14373_begin_0, end = var_14373_end_0, end_mask = var_14373_end_mask_0, x = k_135_cast_fp16)[name = tensor("op_14373_cast_fp16")]; + tensor var_14377_begin_0 = const()[name = tensor("op_14377_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_14377_end_0 = const()[name = tensor("op_14377_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_14377_end_mask_0 = const()[name = tensor("op_14377_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14377_cast_fp16 = slice_by_index(begin = var_14377_begin_0, end = var_14377_end_0, end_mask = var_14377_end_mask_0, x = k_135_cast_fp16)[name = tensor("op_14377_cast_fp16")]; + tensor var_14381_begin_0 = const()[name = tensor("op_14381_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_14381_end_0 = const()[name = tensor("op_14381_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_14381_end_mask_0 = const()[name = tensor("op_14381_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14381_cast_fp16 = slice_by_index(begin = var_14381_begin_0, end = var_14381_end_0, end_mask = var_14381_end_mask_0, x = k_135_cast_fp16)[name = tensor("op_14381_cast_fp16")]; + tensor var_14383_begin_0 = const()[name = tensor("op_14383_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14383_end_0 = const()[name = tensor("op_14383_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_14383_end_mask_0 = const()[name = tensor("op_14383_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14383_cast_fp16 = slice_by_index(begin = var_14383_begin_0, end = var_14383_end_0, end_mask = var_14383_end_mask_0, x = v_67_cast_fp16)[name = tensor("op_14383_cast_fp16")]; + tensor var_14387_begin_0 = const()[name = tensor("op_14387_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_14387_end_0 = const()[name = tensor("op_14387_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_14387_end_mask_0 = const()[name = tensor("op_14387_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14387_cast_fp16 = slice_by_index(begin = var_14387_begin_0, end = var_14387_end_0, end_mask = var_14387_end_mask_0, x = v_67_cast_fp16)[name = tensor("op_14387_cast_fp16")]; + tensor var_14391_begin_0 = const()[name = tensor("op_14391_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_14391_end_0 = const()[name = tensor("op_14391_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_14391_end_mask_0 = const()[name = tensor("op_14391_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14391_cast_fp16 = slice_by_index(begin = var_14391_begin_0, end = var_14391_end_0, end_mask = var_14391_end_mask_0, x = v_67_cast_fp16)[name = tensor("op_14391_cast_fp16")]; + tensor var_14395_begin_0 = const()[name = tensor("op_14395_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_14395_end_0 = const()[name = tensor("op_14395_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_14395_end_mask_0 = const()[name = tensor("op_14395_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14395_cast_fp16 = slice_by_index(begin = var_14395_begin_0, end = var_14395_end_0, end_mask = var_14395_end_mask_0, x = v_67_cast_fp16)[name = tensor("op_14395_cast_fp16")]; + tensor var_14399_begin_0 = const()[name = tensor("op_14399_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_14399_end_0 = const()[name = tensor("op_14399_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_14399_end_mask_0 = const()[name = tensor("op_14399_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14399_cast_fp16 = slice_by_index(begin = var_14399_begin_0, end = var_14399_end_0, end_mask = var_14399_end_mask_0, x = v_67_cast_fp16)[name = tensor("op_14399_cast_fp16")]; + tensor var_14403_begin_0 = const()[name = tensor("op_14403_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_14403_end_0 = const()[name = tensor("op_14403_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_14403_end_mask_0 = const()[name = tensor("op_14403_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14403_cast_fp16 = slice_by_index(begin = var_14403_begin_0, end = var_14403_end_0, end_mask = var_14403_end_mask_0, x = v_67_cast_fp16)[name = tensor("op_14403_cast_fp16")]; + tensor var_14407_begin_0 = const()[name = tensor("op_14407_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_14407_end_0 = const()[name = tensor("op_14407_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_14407_end_mask_0 = const()[name = tensor("op_14407_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14407_cast_fp16 = slice_by_index(begin = var_14407_begin_0, end = var_14407_end_0, end_mask = var_14407_end_mask_0, x = v_67_cast_fp16)[name = tensor("op_14407_cast_fp16")]; + tensor var_14411_begin_0 = const()[name = tensor("op_14411_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_14411_end_0 = const()[name = tensor("op_14411_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_14411_end_mask_0 = const()[name = tensor("op_14411_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14411_cast_fp16 = slice_by_index(begin = var_14411_begin_0, end = var_14411_end_0, end_mask = var_14411_end_mask_0, x = v_67_cast_fp16)[name = tensor("op_14411_cast_fp16")]; + tensor var_14415_begin_0 = const()[name = tensor("op_14415_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_14415_end_0 = const()[name = tensor("op_14415_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_14415_end_mask_0 = const()[name = tensor("op_14415_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14415_cast_fp16 = slice_by_index(begin = var_14415_begin_0, end = var_14415_end_0, end_mask = var_14415_end_mask_0, x = v_67_cast_fp16)[name = tensor("op_14415_cast_fp16")]; + tensor var_14419_begin_0 = const()[name = tensor("op_14419_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_14419_end_0 = const()[name = tensor("op_14419_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_14419_end_mask_0 = const()[name = tensor("op_14419_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14419_cast_fp16 = slice_by_index(begin = var_14419_begin_0, end = var_14419_end_0, end_mask = var_14419_end_mask_0, x = v_67_cast_fp16)[name = tensor("op_14419_cast_fp16")]; + tensor var_14423_begin_0 = const()[name = tensor("op_14423_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_14423_end_0 = const()[name = tensor("op_14423_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_14423_end_mask_0 = const()[name = tensor("op_14423_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14423_cast_fp16 = slice_by_index(begin = var_14423_begin_0, end = var_14423_end_0, end_mask = var_14423_end_mask_0, x = v_67_cast_fp16)[name = tensor("op_14423_cast_fp16")]; + tensor var_14427_begin_0 = const()[name = tensor("op_14427_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_14427_end_0 = const()[name = tensor("op_14427_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_14427_end_mask_0 = const()[name = tensor("op_14427_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14427_cast_fp16 = slice_by_index(begin = var_14427_begin_0, end = var_14427_end_0, end_mask = var_14427_end_mask_0, x = v_67_cast_fp16)[name = tensor("op_14427_cast_fp16")]; + tensor var_14431_begin_0 = const()[name = tensor("op_14431_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_14431_end_0 = const()[name = tensor("op_14431_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_14431_end_mask_0 = const()[name = tensor("op_14431_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14431_cast_fp16 = slice_by_index(begin = var_14431_begin_0, end = var_14431_end_0, end_mask = var_14431_end_mask_0, x = v_67_cast_fp16)[name = tensor("op_14431_cast_fp16")]; + tensor var_14435_begin_0 = const()[name = tensor("op_14435_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_14435_end_0 = const()[name = tensor("op_14435_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_14435_end_mask_0 = const()[name = tensor("op_14435_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14435_cast_fp16 = slice_by_index(begin = var_14435_begin_0, end = var_14435_end_0, end_mask = var_14435_end_mask_0, x = v_67_cast_fp16)[name = tensor("op_14435_cast_fp16")]; + tensor var_14439_begin_0 = const()[name = tensor("op_14439_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_14439_end_0 = const()[name = tensor("op_14439_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_14439_end_mask_0 = const()[name = tensor("op_14439_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14439_cast_fp16 = slice_by_index(begin = var_14439_begin_0, end = var_14439_end_0, end_mask = var_14439_end_mask_0, x = v_67_cast_fp16)[name = tensor("op_14439_cast_fp16")]; + tensor var_14443_begin_0 = const()[name = tensor("op_14443_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_14443_end_0 = const()[name = tensor("op_14443_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_14443_end_mask_0 = const()[name = tensor("op_14443_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14443_cast_fp16 = slice_by_index(begin = var_14443_begin_0, end = var_14443_end_0, end_mask = var_14443_end_mask_0, x = v_67_cast_fp16)[name = tensor("op_14443_cast_fp16")]; + tensor var_14447_begin_0 = const()[name = tensor("op_14447_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_14447_end_0 = const()[name = tensor("op_14447_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_14447_end_mask_0 = const()[name = tensor("op_14447_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14447_cast_fp16 = slice_by_index(begin = var_14447_begin_0, end = var_14447_end_0, end_mask = var_14447_end_mask_0, x = v_67_cast_fp16)[name = tensor("op_14447_cast_fp16")]; + tensor var_14451_begin_0 = const()[name = tensor("op_14451_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_14451_end_0 = const()[name = tensor("op_14451_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_14451_end_mask_0 = const()[name = tensor("op_14451_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14451_cast_fp16 = slice_by_index(begin = var_14451_begin_0, end = var_14451_end_0, end_mask = var_14451_end_mask_0, x = v_67_cast_fp16)[name = tensor("op_14451_cast_fp16")]; + tensor var_14455_begin_0 = const()[name = tensor("op_14455_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_14455_end_0 = const()[name = tensor("op_14455_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_14455_end_mask_0 = const()[name = tensor("op_14455_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14455_cast_fp16 = slice_by_index(begin = var_14455_begin_0, end = var_14455_end_0, end_mask = var_14455_end_mask_0, x = v_67_cast_fp16)[name = tensor("op_14455_cast_fp16")]; + tensor var_14459_begin_0 = const()[name = tensor("op_14459_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_14459_end_0 = const()[name = tensor("op_14459_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_14459_end_mask_0 = const()[name = tensor("op_14459_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14459_cast_fp16 = slice_by_index(begin = var_14459_begin_0, end = var_14459_end_0, end_mask = var_14459_end_mask_0, x = v_67_cast_fp16)[name = tensor("op_14459_cast_fp16")]; + tensor var_14463_equation_0 = const()[name = tensor("op_14463_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14463_cast_fp16 = einsum(equation = var_14463_equation_0, values = (var_14305_cast_fp16, var_14222_cast_fp16))[name = tensor("op_14463_cast_fp16")]; + tensor var_14464_to_fp16 = const()[name = tensor("op_14464_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1161_cast_fp16 = mul(x = var_14463_cast_fp16, y = var_14464_to_fp16)[name = tensor("aw_1161_cast_fp16")]; + tensor var_14467_equation_0 = const()[name = tensor("op_14467_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14467_cast_fp16 = einsum(equation = var_14467_equation_0, values = (var_14309_cast_fp16, var_14226_cast_fp16))[name = tensor("op_14467_cast_fp16")]; + tensor var_14468_to_fp16 = const()[name = tensor("op_14468_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1163_cast_fp16 = mul(x = var_14467_cast_fp16, y = var_14468_to_fp16)[name = tensor("aw_1163_cast_fp16")]; + tensor var_14471_equation_0 = const()[name = tensor("op_14471_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14471_cast_fp16 = einsum(equation = var_14471_equation_0, values = (var_14313_cast_fp16, var_14230_cast_fp16))[name = tensor("op_14471_cast_fp16")]; + tensor var_14472_to_fp16 = const()[name = tensor("op_14472_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1165_cast_fp16 = mul(x = var_14471_cast_fp16, y = var_14472_to_fp16)[name = tensor("aw_1165_cast_fp16")]; + tensor var_14475_equation_0 = const()[name = tensor("op_14475_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14475_cast_fp16 = einsum(equation = var_14475_equation_0, values = (var_14317_cast_fp16, var_14234_cast_fp16))[name = tensor("op_14475_cast_fp16")]; + tensor var_14476_to_fp16 = const()[name = tensor("op_14476_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1167_cast_fp16 = mul(x = var_14475_cast_fp16, y = var_14476_to_fp16)[name = tensor("aw_1167_cast_fp16")]; + tensor var_14479_equation_0 = const()[name = tensor("op_14479_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14479_cast_fp16 = einsum(equation = var_14479_equation_0, values = (var_14321_cast_fp16, var_14238_cast_fp16))[name = tensor("op_14479_cast_fp16")]; + tensor var_14480_to_fp16 = const()[name = tensor("op_14480_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1169_cast_fp16 = mul(x = var_14479_cast_fp16, y = var_14480_to_fp16)[name = tensor("aw_1169_cast_fp16")]; + tensor var_14483_equation_0 = const()[name = tensor("op_14483_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14483_cast_fp16 = einsum(equation = var_14483_equation_0, values = (var_14325_cast_fp16, var_14242_cast_fp16))[name = tensor("op_14483_cast_fp16")]; + tensor var_14484_to_fp16 = const()[name = tensor("op_14484_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1171_cast_fp16 = mul(x = var_14483_cast_fp16, y = var_14484_to_fp16)[name = tensor("aw_1171_cast_fp16")]; + tensor var_14487_equation_0 = const()[name = tensor("op_14487_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14487_cast_fp16 = einsum(equation = var_14487_equation_0, values = (var_14329_cast_fp16, var_14246_cast_fp16))[name = tensor("op_14487_cast_fp16")]; + tensor var_14488_to_fp16 = const()[name = tensor("op_14488_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1173_cast_fp16 = mul(x = var_14487_cast_fp16, y = var_14488_to_fp16)[name = tensor("aw_1173_cast_fp16")]; + tensor var_14491_equation_0 = const()[name = tensor("op_14491_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14491_cast_fp16 = einsum(equation = var_14491_equation_0, values = (var_14333_cast_fp16, var_14250_cast_fp16))[name = tensor("op_14491_cast_fp16")]; + tensor var_14492_to_fp16 = const()[name = tensor("op_14492_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1175_cast_fp16 = mul(x = var_14491_cast_fp16, y = var_14492_to_fp16)[name = tensor("aw_1175_cast_fp16")]; + tensor var_14495_equation_0 = const()[name = tensor("op_14495_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14495_cast_fp16 = einsum(equation = var_14495_equation_0, values = (var_14337_cast_fp16, var_14254_cast_fp16))[name = tensor("op_14495_cast_fp16")]; + tensor var_14496_to_fp16 = const()[name = tensor("op_14496_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1177_cast_fp16 = mul(x = var_14495_cast_fp16, y = var_14496_to_fp16)[name = tensor("aw_1177_cast_fp16")]; + tensor var_14499_equation_0 = const()[name = tensor("op_14499_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14499_cast_fp16 = einsum(equation = var_14499_equation_0, values = (var_14341_cast_fp16, var_14258_cast_fp16))[name = tensor("op_14499_cast_fp16")]; + tensor var_14500_to_fp16 = const()[name = tensor("op_14500_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1179_cast_fp16 = mul(x = var_14499_cast_fp16, y = var_14500_to_fp16)[name = tensor("aw_1179_cast_fp16")]; + tensor var_14503_equation_0 = const()[name = tensor("op_14503_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14503_cast_fp16 = einsum(equation = var_14503_equation_0, values = (var_14345_cast_fp16, var_14262_cast_fp16))[name = tensor("op_14503_cast_fp16")]; + tensor var_14504_to_fp16 = const()[name = tensor("op_14504_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1181_cast_fp16 = mul(x = var_14503_cast_fp16, y = var_14504_to_fp16)[name = tensor("aw_1181_cast_fp16")]; + tensor var_14507_equation_0 = const()[name = tensor("op_14507_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14507_cast_fp16 = einsum(equation = var_14507_equation_0, values = (var_14349_cast_fp16, var_14266_cast_fp16))[name = tensor("op_14507_cast_fp16")]; + tensor var_14508_to_fp16 = const()[name = tensor("op_14508_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1183_cast_fp16 = mul(x = var_14507_cast_fp16, y = var_14508_to_fp16)[name = tensor("aw_1183_cast_fp16")]; + tensor var_14511_equation_0 = const()[name = tensor("op_14511_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14511_cast_fp16 = einsum(equation = var_14511_equation_0, values = (var_14353_cast_fp16, var_14270_cast_fp16))[name = tensor("op_14511_cast_fp16")]; + tensor var_14512_to_fp16 = const()[name = tensor("op_14512_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1185_cast_fp16 = mul(x = var_14511_cast_fp16, y = var_14512_to_fp16)[name = tensor("aw_1185_cast_fp16")]; + tensor var_14515_equation_0 = const()[name = tensor("op_14515_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14515_cast_fp16 = einsum(equation = var_14515_equation_0, values = (var_14357_cast_fp16, var_14274_cast_fp16))[name = tensor("op_14515_cast_fp16")]; + tensor var_14516_to_fp16 = const()[name = tensor("op_14516_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1187_cast_fp16 = mul(x = var_14515_cast_fp16, y = var_14516_to_fp16)[name = tensor("aw_1187_cast_fp16")]; + tensor var_14519_equation_0 = const()[name = tensor("op_14519_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14519_cast_fp16 = einsum(equation = var_14519_equation_0, values = (var_14361_cast_fp16, var_14278_cast_fp16))[name = tensor("op_14519_cast_fp16")]; + tensor var_14520_to_fp16 = const()[name = tensor("op_14520_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1189_cast_fp16 = mul(x = var_14519_cast_fp16, y = var_14520_to_fp16)[name = tensor("aw_1189_cast_fp16")]; + tensor var_14523_equation_0 = const()[name = tensor("op_14523_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14523_cast_fp16 = einsum(equation = var_14523_equation_0, values = (var_14365_cast_fp16, var_14282_cast_fp16))[name = tensor("op_14523_cast_fp16")]; + tensor var_14524_to_fp16 = const()[name = tensor("op_14524_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1191_cast_fp16 = mul(x = var_14523_cast_fp16, y = var_14524_to_fp16)[name = tensor("aw_1191_cast_fp16")]; + tensor var_14527_equation_0 = const()[name = tensor("op_14527_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14527_cast_fp16 = einsum(equation = var_14527_equation_0, values = (var_14369_cast_fp16, var_14286_cast_fp16))[name = tensor("op_14527_cast_fp16")]; + tensor var_14528_to_fp16 = const()[name = tensor("op_14528_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1193_cast_fp16 = mul(x = var_14527_cast_fp16, y = var_14528_to_fp16)[name = tensor("aw_1193_cast_fp16")]; + tensor var_14531_equation_0 = const()[name = tensor("op_14531_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14531_cast_fp16 = einsum(equation = var_14531_equation_0, values = (var_14373_cast_fp16, var_14290_cast_fp16))[name = tensor("op_14531_cast_fp16")]; + tensor var_14532_to_fp16 = const()[name = tensor("op_14532_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1195_cast_fp16 = mul(x = var_14531_cast_fp16, y = var_14532_to_fp16)[name = tensor("aw_1195_cast_fp16")]; + tensor var_14535_equation_0 = const()[name = tensor("op_14535_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14535_cast_fp16 = einsum(equation = var_14535_equation_0, values = (var_14377_cast_fp16, var_14294_cast_fp16))[name = tensor("op_14535_cast_fp16")]; + tensor var_14536_to_fp16 = const()[name = tensor("op_14536_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1197_cast_fp16 = mul(x = var_14535_cast_fp16, y = var_14536_to_fp16)[name = tensor("aw_1197_cast_fp16")]; + tensor var_14539_equation_0 = const()[name = tensor("op_14539_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14539_cast_fp16 = einsum(equation = var_14539_equation_0, values = (var_14381_cast_fp16, var_14298_cast_fp16))[name = tensor("op_14539_cast_fp16")]; + tensor var_14540_to_fp16 = const()[name = tensor("op_14540_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1199_cast_fp16 = mul(x = var_14539_cast_fp16, y = var_14540_to_fp16)[name = tensor("aw_1199_cast_fp16")]; + tensor var_14542_cast_fp16 = softmax(axis = var_2624, x = aw_1161_cast_fp16)[name = tensor("op_14542_cast_fp16")]; + tensor var_14543_cast_fp16 = softmax(axis = var_2624, x = aw_1163_cast_fp16)[name = tensor("op_14543_cast_fp16")]; + tensor var_14544_cast_fp16 = softmax(axis = var_2624, x = aw_1165_cast_fp16)[name = tensor("op_14544_cast_fp16")]; + tensor var_14545_cast_fp16 = softmax(axis = var_2624, x = aw_1167_cast_fp16)[name = tensor("op_14545_cast_fp16")]; + tensor var_14546_cast_fp16 = softmax(axis = var_2624, x = aw_1169_cast_fp16)[name = tensor("op_14546_cast_fp16")]; + tensor var_14547_cast_fp16 = softmax(axis = var_2624, x = aw_1171_cast_fp16)[name = tensor("op_14547_cast_fp16")]; + tensor var_14548_cast_fp16 = softmax(axis = var_2624, x = aw_1173_cast_fp16)[name = tensor("op_14548_cast_fp16")]; + tensor var_14549_cast_fp16 = softmax(axis = var_2624, x = aw_1175_cast_fp16)[name = tensor("op_14549_cast_fp16")]; + tensor var_14550_cast_fp16 = softmax(axis = var_2624, x = aw_1177_cast_fp16)[name = tensor("op_14550_cast_fp16")]; + tensor var_14551_cast_fp16 = softmax(axis = var_2624, x = aw_1179_cast_fp16)[name = tensor("op_14551_cast_fp16")]; + tensor var_14552_cast_fp16 = softmax(axis = var_2624, x = aw_1181_cast_fp16)[name = tensor("op_14552_cast_fp16")]; + tensor var_14553_cast_fp16 = softmax(axis = var_2624, x = aw_1183_cast_fp16)[name = tensor("op_14553_cast_fp16")]; + tensor var_14554_cast_fp16 = softmax(axis = var_2624, x = aw_1185_cast_fp16)[name = tensor("op_14554_cast_fp16")]; + tensor var_14555_cast_fp16 = softmax(axis = var_2624, x = aw_1187_cast_fp16)[name = tensor("op_14555_cast_fp16")]; + tensor var_14556_cast_fp16 = softmax(axis = var_2624, x = aw_1189_cast_fp16)[name = tensor("op_14556_cast_fp16")]; + tensor var_14557_cast_fp16 = softmax(axis = var_2624, x = aw_1191_cast_fp16)[name = tensor("op_14557_cast_fp16")]; + tensor var_14558_cast_fp16 = softmax(axis = var_2624, x = aw_1193_cast_fp16)[name = tensor("op_14558_cast_fp16")]; + tensor var_14559_cast_fp16 = softmax(axis = var_2624, x = aw_1195_cast_fp16)[name = tensor("op_14559_cast_fp16")]; + tensor var_14560_cast_fp16 = softmax(axis = var_2624, x = aw_1197_cast_fp16)[name = tensor("op_14560_cast_fp16")]; + tensor var_14561_cast_fp16 = softmax(axis = var_2624, x = aw_1199_cast_fp16)[name = tensor("op_14561_cast_fp16")]; + tensor var_14563_equation_0 = const()[name = tensor("op_14563_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14563_cast_fp16 = einsum(equation = var_14563_equation_0, values = (var_14383_cast_fp16, var_14542_cast_fp16))[name = tensor("op_14563_cast_fp16")]; + tensor var_14565_equation_0 = const()[name = tensor("op_14565_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14565_cast_fp16 = einsum(equation = var_14565_equation_0, values = (var_14387_cast_fp16, var_14543_cast_fp16))[name = tensor("op_14565_cast_fp16")]; + tensor var_14567_equation_0 = const()[name = tensor("op_14567_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14567_cast_fp16 = einsum(equation = var_14567_equation_0, values = (var_14391_cast_fp16, var_14544_cast_fp16))[name = tensor("op_14567_cast_fp16")]; + tensor var_14569_equation_0 = const()[name = tensor("op_14569_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14569_cast_fp16 = einsum(equation = var_14569_equation_0, values = (var_14395_cast_fp16, var_14545_cast_fp16))[name = tensor("op_14569_cast_fp16")]; + tensor var_14571_equation_0 = const()[name = tensor("op_14571_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14571_cast_fp16 = einsum(equation = var_14571_equation_0, values = (var_14399_cast_fp16, var_14546_cast_fp16))[name = tensor("op_14571_cast_fp16")]; + tensor var_14573_equation_0 = const()[name = tensor("op_14573_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14573_cast_fp16 = einsum(equation = var_14573_equation_0, values = (var_14403_cast_fp16, var_14547_cast_fp16))[name = tensor("op_14573_cast_fp16")]; + tensor var_14575_equation_0 = const()[name = tensor("op_14575_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14575_cast_fp16 = einsum(equation = var_14575_equation_0, values = (var_14407_cast_fp16, var_14548_cast_fp16))[name = tensor("op_14575_cast_fp16")]; + tensor var_14577_equation_0 = const()[name = tensor("op_14577_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14577_cast_fp16 = einsum(equation = var_14577_equation_0, values = (var_14411_cast_fp16, var_14549_cast_fp16))[name = tensor("op_14577_cast_fp16")]; + tensor var_14579_equation_0 = const()[name = tensor("op_14579_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14579_cast_fp16 = einsum(equation = var_14579_equation_0, values = (var_14415_cast_fp16, var_14550_cast_fp16))[name = tensor("op_14579_cast_fp16")]; + tensor var_14581_equation_0 = const()[name = tensor("op_14581_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14581_cast_fp16 = einsum(equation = var_14581_equation_0, values = (var_14419_cast_fp16, var_14551_cast_fp16))[name = tensor("op_14581_cast_fp16")]; + tensor var_14583_equation_0 = const()[name = tensor("op_14583_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14583_cast_fp16 = einsum(equation = var_14583_equation_0, values = (var_14423_cast_fp16, var_14552_cast_fp16))[name = tensor("op_14583_cast_fp16")]; + tensor var_14585_equation_0 = const()[name = tensor("op_14585_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14585_cast_fp16 = einsum(equation = var_14585_equation_0, values = (var_14427_cast_fp16, var_14553_cast_fp16))[name = tensor("op_14585_cast_fp16")]; + tensor var_14587_equation_0 = const()[name = tensor("op_14587_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14587_cast_fp16 = einsum(equation = var_14587_equation_0, values = (var_14431_cast_fp16, var_14554_cast_fp16))[name = tensor("op_14587_cast_fp16")]; + tensor var_14589_equation_0 = const()[name = tensor("op_14589_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14589_cast_fp16 = einsum(equation = var_14589_equation_0, values = (var_14435_cast_fp16, var_14555_cast_fp16))[name = tensor("op_14589_cast_fp16")]; + tensor var_14591_equation_0 = const()[name = tensor("op_14591_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14591_cast_fp16 = einsum(equation = var_14591_equation_0, values = (var_14439_cast_fp16, var_14556_cast_fp16))[name = tensor("op_14591_cast_fp16")]; + tensor var_14593_equation_0 = const()[name = tensor("op_14593_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14593_cast_fp16 = einsum(equation = var_14593_equation_0, values = (var_14443_cast_fp16, var_14557_cast_fp16))[name = tensor("op_14593_cast_fp16")]; + tensor var_14595_equation_0 = const()[name = tensor("op_14595_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14595_cast_fp16 = einsum(equation = var_14595_equation_0, values = (var_14447_cast_fp16, var_14558_cast_fp16))[name = tensor("op_14595_cast_fp16")]; + tensor var_14597_equation_0 = const()[name = tensor("op_14597_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14597_cast_fp16 = einsum(equation = var_14597_equation_0, values = (var_14451_cast_fp16, var_14559_cast_fp16))[name = tensor("op_14597_cast_fp16")]; + tensor var_14599_equation_0 = const()[name = tensor("op_14599_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14599_cast_fp16 = einsum(equation = var_14599_equation_0, values = (var_14455_cast_fp16, var_14560_cast_fp16))[name = tensor("op_14599_cast_fp16")]; + tensor var_14601_equation_0 = const()[name = tensor("op_14601_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14601_cast_fp16 = einsum(equation = var_14601_equation_0, values = (var_14459_cast_fp16, var_14561_cast_fp16))[name = tensor("op_14601_cast_fp16")]; + tensor input_247_interleave_0 = const()[name = tensor("input_247_interleave_0"), val = tensor(false)]; + tensor input_247_cast_fp16 = concat(axis = var_2624, interleave = input_247_interleave_0, values = (var_14563_cast_fp16, var_14565_cast_fp16, var_14567_cast_fp16, var_14569_cast_fp16, var_14571_cast_fp16, var_14573_cast_fp16, var_14575_cast_fp16, var_14577_cast_fp16, var_14579_cast_fp16, var_14581_cast_fp16, var_14583_cast_fp16, var_14585_cast_fp16, var_14587_cast_fp16, var_14589_cast_fp16, var_14591_cast_fp16, var_14593_cast_fp16, var_14595_cast_fp16, var_14597_cast_fp16, var_14599_cast_fp16, var_14601_cast_fp16))[name = tensor("input_247_cast_fp16")]; + tensor var_14607 = const()[name = tensor("op_14607"), val = tensor([1, 1])]; + tensor var_14609 = const()[name = tensor("op_14609"), val = tensor([1, 1])]; + tensor var_14611_pad_type_0 = const()[name = tensor("op_14611_pad_type_0"), val = tensor("custom")]; + tensor var_14611_pad_0 = const()[name = tensor("op_14611_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_2_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_2_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1128046720)))]; + tensor down_blocks_2_attentions_1_transformer_blocks_2_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_2_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1131323584)))]; + tensor var_14611_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_2_attn2_to_out_0_bias_to_fp16, dilations = var_14609, groups = var_2624, pad = var_14611_pad_0, pad_type = var_14611_pad_type_0, strides = var_14607, weight = down_blocks_2_attentions_1_transformer_blocks_2_attn2_to_out_0_weight_to_fp16, x = input_247_cast_fp16)[name = tensor("op_14611_cast_fp16")]; + tensor inputs_101_cast_fp16 = add(x = var_14611_cast_fp16, y = inputs_99_cast_fp16)[name = tensor("inputs_101_cast_fp16")]; + tensor input_249_axes_0 = const()[name = tensor("input_249_axes_0"), val = tensor([1])]; + tensor input_249_gamma_0_to_fp16 = const()[name = tensor("input_249_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1131326208)))]; + tensor input_249_beta_0_to_fp16 = const()[name = tensor("input_249_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1131328832)))]; + tensor var_14621_to_fp16 = const()[name = tensor("op_14621_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_249_cast_fp16 = layer_norm(axes = input_249_axes_0, beta = input_249_beta_0_to_fp16, epsilon = var_14621_to_fp16, gamma = input_249_gamma_0_to_fp16, x = inputs_101_cast_fp16)[name = tensor("input_249_cast_fp16")]; + tensor var_14637 = const()[name = tensor("op_14637"), val = tensor([1, 1])]; + tensor var_14639 = const()[name = tensor("op_14639"), val = tensor([1, 1])]; + tensor var_14641_pad_type_0 = const()[name = tensor("op_14641_pad_type_0"), val = tensor("custom")]; + tensor var_14641_pad_0 = const()[name = tensor("op_14641_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_2_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_2_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1131331456)))]; + tensor down_blocks_2_attentions_1_transformer_blocks_2_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_2_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1157545920)))]; + tensor var_14641_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_2_ff_net_0_proj_bias_to_fp16, dilations = var_14639, groups = var_2624, pad = var_14641_pad_0, pad_type = var_14641_pad_type_0, strides = var_14637, weight = down_blocks_2_attentions_1_transformer_blocks_2_ff_net_0_proj_weight_to_fp16, x = input_249_cast_fp16)[name = tensor("op_14641_cast_fp16")]; + tensor var_14642_split_sizes_0 = const()[name = tensor("op_14642_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_14642_axis_0 = const()[name = tensor("op_14642_axis_0"), val = tensor(1)]; + tensor var_14642_cast_fp16_0, tensor var_14642_cast_fp16_1 = split(axis = var_14642_axis_0, split_sizes = var_14642_split_sizes_0, x = var_14641_cast_fp16)[name = tensor("op_14642_cast_fp16")]; + tensor var_14644_mode_0 = const()[name = tensor("op_14644_mode_0"), val = tensor("EXACT")]; + tensor var_14644_cast_fp16 = gelu(mode = var_14644_mode_0, x = var_14642_cast_fp16_1)[name = tensor("op_14644_cast_fp16")]; + tensor input_251_cast_fp16 = mul(x = var_14642_cast_fp16_0, y = var_14644_cast_fp16)[name = tensor("input_251_cast_fp16")]; + tensor var_14648 = const()[name = tensor("op_14648"), val = tensor([1, 1])]; + tensor var_14650 = const()[name = tensor("op_14650"), val = tensor([1, 1])]; + tensor var_14652_pad_type_0 = const()[name = tensor("op_14652_pad_type_0"), val = tensor("custom")]; + tensor var_14652_pad_0 = const()[name = tensor("op_14652_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_2_ff_net_2_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_2_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1157566464)))]; + tensor down_blocks_2_attentions_1_transformer_blocks_2_ff_net_2_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_2_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1170673728)))]; + tensor var_14652_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_2_ff_net_2_bias_to_fp16, dilations = var_14650, groups = var_2624, pad = var_14652_pad_0, pad_type = var_14652_pad_type_0, strides = var_14648, weight = down_blocks_2_attentions_1_transformer_blocks_2_ff_net_2_weight_to_fp16, x = input_251_cast_fp16)[name = tensor("op_14652_cast_fp16")]; + tensor inputs_103_cast_fp16 = add(x = var_14652_cast_fp16, y = inputs_101_cast_fp16)[name = tensor("inputs_103_cast_fp16")]; + tensor hidden_states_155_axes_0 = const()[name = tensor("hidden_states_155_axes_0"), val = tensor([1])]; + tensor hidden_states_155_gamma_0_to_fp16 = const()[name = tensor("hidden_states_155_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1170676352)))]; + tensor hidden_states_155_beta_0_to_fp16 = const()[name = tensor("hidden_states_155_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1170678976)))]; + tensor var_14668_to_fp16 = const()[name = tensor("op_14668_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_155_cast_fp16 = layer_norm(axes = hidden_states_155_axes_0, beta = hidden_states_155_beta_0_to_fp16, epsilon = var_14668_to_fp16, gamma = hidden_states_155_gamma_0_to_fp16, x = inputs_103_cast_fp16)[name = tensor("hidden_states_155_cast_fp16")]; + tensor var_14683 = const()[name = tensor("op_14683"), val = tensor([1, 1])]; + tensor var_14685 = const()[name = tensor("op_14685"), val = tensor([1, 1])]; + tensor q_69_pad_type_0 = const()[name = tensor("q_69_pad_type_0"), val = tensor("custom")]; + tensor q_69_pad_0 = const()[name = tensor("q_69_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_3_attn1_to_q_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_3_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1170681600)))]; + tensor q_69_cast_fp16 = conv(dilations = var_14685, groups = var_2624, pad = q_69_pad_0, pad_type = q_69_pad_type_0, strides = var_14683, weight = down_blocks_2_attentions_1_transformer_blocks_3_attn1_to_q_weight_to_fp16, x = hidden_states_155_cast_fp16)[name = tensor("q_69_cast_fp16")]; + tensor var_14689 = const()[name = tensor("op_14689"), val = tensor([1, 1])]; + tensor var_14691 = const()[name = tensor("op_14691"), val = tensor([1, 1])]; + tensor k_137_pad_type_0 = const()[name = tensor("k_137_pad_type_0"), val = tensor("custom")]; + tensor k_137_pad_0 = const()[name = tensor("k_137_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_3_attn1_to_k_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_3_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1173958464)))]; + tensor k_137_cast_fp16 = conv(dilations = var_14691, groups = var_2624, pad = k_137_pad_0, pad_type = k_137_pad_type_0, strides = var_14689, weight = down_blocks_2_attentions_1_transformer_blocks_3_attn1_to_k_weight_to_fp16, x = hidden_states_155_cast_fp16)[name = tensor("k_137_cast_fp16")]; + tensor var_14695 = const()[name = tensor("op_14695"), val = tensor([1, 1])]; + tensor var_14697 = const()[name = tensor("op_14697"), val = tensor([1, 1])]; + tensor v_69_pad_type_0 = const()[name = tensor("v_69_pad_type_0"), val = tensor("custom")]; + tensor v_69_pad_0 = const()[name = tensor("v_69_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_3_attn1_to_v_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_3_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1177235328)))]; + tensor v_69_cast_fp16 = conv(dilations = var_14697, groups = var_2624, pad = v_69_pad_0, pad_type = v_69_pad_type_0, strides = var_14695, weight = down_blocks_2_attentions_1_transformer_blocks_3_attn1_to_v_weight_to_fp16, x = hidden_states_155_cast_fp16)[name = tensor("v_69_cast_fp16")]; + tensor var_14701_begin_0 = const()[name = tensor("op_14701_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14701_end_0 = const()[name = tensor("op_14701_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_14701_end_mask_0 = const()[name = tensor("op_14701_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14701_cast_fp16 = slice_by_index(begin = var_14701_begin_0, end = var_14701_end_0, end_mask = var_14701_end_mask_0, x = q_69_cast_fp16)[name = tensor("op_14701_cast_fp16")]; + tensor var_14705_begin_0 = const()[name = tensor("op_14705_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_14705_end_0 = const()[name = tensor("op_14705_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_14705_end_mask_0 = const()[name = tensor("op_14705_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14705_cast_fp16 = slice_by_index(begin = var_14705_begin_0, end = var_14705_end_0, end_mask = var_14705_end_mask_0, x = q_69_cast_fp16)[name = tensor("op_14705_cast_fp16")]; + tensor var_14709_begin_0 = const()[name = tensor("op_14709_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_14709_end_0 = const()[name = tensor("op_14709_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_14709_end_mask_0 = const()[name = tensor("op_14709_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14709_cast_fp16 = slice_by_index(begin = var_14709_begin_0, end = var_14709_end_0, end_mask = var_14709_end_mask_0, x = q_69_cast_fp16)[name = tensor("op_14709_cast_fp16")]; + tensor var_14713_begin_0 = const()[name = tensor("op_14713_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_14713_end_0 = const()[name = tensor("op_14713_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_14713_end_mask_0 = const()[name = tensor("op_14713_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14713_cast_fp16 = slice_by_index(begin = var_14713_begin_0, end = var_14713_end_0, end_mask = var_14713_end_mask_0, x = q_69_cast_fp16)[name = tensor("op_14713_cast_fp16")]; + tensor var_14717_begin_0 = const()[name = tensor("op_14717_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_14717_end_0 = const()[name = tensor("op_14717_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_14717_end_mask_0 = const()[name = tensor("op_14717_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14717_cast_fp16 = slice_by_index(begin = var_14717_begin_0, end = var_14717_end_0, end_mask = var_14717_end_mask_0, x = q_69_cast_fp16)[name = tensor("op_14717_cast_fp16")]; + tensor var_14721_begin_0 = const()[name = tensor("op_14721_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_14721_end_0 = const()[name = tensor("op_14721_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_14721_end_mask_0 = const()[name = tensor("op_14721_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14721_cast_fp16 = slice_by_index(begin = var_14721_begin_0, end = var_14721_end_0, end_mask = var_14721_end_mask_0, x = q_69_cast_fp16)[name = tensor("op_14721_cast_fp16")]; + tensor var_14725_begin_0 = const()[name = tensor("op_14725_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_14725_end_0 = const()[name = tensor("op_14725_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_14725_end_mask_0 = const()[name = tensor("op_14725_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14725_cast_fp16 = slice_by_index(begin = var_14725_begin_0, end = var_14725_end_0, end_mask = var_14725_end_mask_0, x = q_69_cast_fp16)[name = tensor("op_14725_cast_fp16")]; + tensor var_14729_begin_0 = const()[name = tensor("op_14729_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_14729_end_0 = const()[name = tensor("op_14729_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_14729_end_mask_0 = const()[name = tensor("op_14729_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14729_cast_fp16 = slice_by_index(begin = var_14729_begin_0, end = var_14729_end_0, end_mask = var_14729_end_mask_0, x = q_69_cast_fp16)[name = tensor("op_14729_cast_fp16")]; + tensor var_14733_begin_0 = const()[name = tensor("op_14733_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_14733_end_0 = const()[name = tensor("op_14733_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_14733_end_mask_0 = const()[name = tensor("op_14733_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14733_cast_fp16 = slice_by_index(begin = var_14733_begin_0, end = var_14733_end_0, end_mask = var_14733_end_mask_0, x = q_69_cast_fp16)[name = tensor("op_14733_cast_fp16")]; + tensor var_14737_begin_0 = const()[name = tensor("op_14737_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_14737_end_0 = const()[name = tensor("op_14737_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_14737_end_mask_0 = const()[name = tensor("op_14737_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14737_cast_fp16 = slice_by_index(begin = var_14737_begin_0, end = var_14737_end_0, end_mask = var_14737_end_mask_0, x = q_69_cast_fp16)[name = tensor("op_14737_cast_fp16")]; + tensor var_14741_begin_0 = const()[name = tensor("op_14741_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_14741_end_0 = const()[name = tensor("op_14741_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_14741_end_mask_0 = const()[name = tensor("op_14741_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14741_cast_fp16 = slice_by_index(begin = var_14741_begin_0, end = var_14741_end_0, end_mask = var_14741_end_mask_0, x = q_69_cast_fp16)[name = tensor("op_14741_cast_fp16")]; + tensor var_14745_begin_0 = const()[name = tensor("op_14745_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_14745_end_0 = const()[name = tensor("op_14745_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_14745_end_mask_0 = const()[name = tensor("op_14745_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14745_cast_fp16 = slice_by_index(begin = var_14745_begin_0, end = var_14745_end_0, end_mask = var_14745_end_mask_0, x = q_69_cast_fp16)[name = tensor("op_14745_cast_fp16")]; + tensor var_14749_begin_0 = const()[name = tensor("op_14749_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_14749_end_0 = const()[name = tensor("op_14749_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_14749_end_mask_0 = const()[name = tensor("op_14749_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14749_cast_fp16 = slice_by_index(begin = var_14749_begin_0, end = var_14749_end_0, end_mask = var_14749_end_mask_0, x = q_69_cast_fp16)[name = tensor("op_14749_cast_fp16")]; + tensor var_14753_begin_0 = const()[name = tensor("op_14753_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_14753_end_0 = const()[name = tensor("op_14753_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_14753_end_mask_0 = const()[name = tensor("op_14753_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14753_cast_fp16 = slice_by_index(begin = var_14753_begin_0, end = var_14753_end_0, end_mask = var_14753_end_mask_0, x = q_69_cast_fp16)[name = tensor("op_14753_cast_fp16")]; + tensor var_14757_begin_0 = const()[name = tensor("op_14757_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_14757_end_0 = const()[name = tensor("op_14757_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_14757_end_mask_0 = const()[name = tensor("op_14757_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14757_cast_fp16 = slice_by_index(begin = var_14757_begin_0, end = var_14757_end_0, end_mask = var_14757_end_mask_0, x = q_69_cast_fp16)[name = tensor("op_14757_cast_fp16")]; + tensor var_14761_begin_0 = const()[name = tensor("op_14761_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_14761_end_0 = const()[name = tensor("op_14761_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_14761_end_mask_0 = const()[name = tensor("op_14761_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14761_cast_fp16 = slice_by_index(begin = var_14761_begin_0, end = var_14761_end_0, end_mask = var_14761_end_mask_0, x = q_69_cast_fp16)[name = tensor("op_14761_cast_fp16")]; + tensor var_14765_begin_0 = const()[name = tensor("op_14765_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_14765_end_0 = const()[name = tensor("op_14765_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_14765_end_mask_0 = const()[name = tensor("op_14765_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14765_cast_fp16 = slice_by_index(begin = var_14765_begin_0, end = var_14765_end_0, end_mask = var_14765_end_mask_0, x = q_69_cast_fp16)[name = tensor("op_14765_cast_fp16")]; + tensor var_14769_begin_0 = const()[name = tensor("op_14769_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_14769_end_0 = const()[name = tensor("op_14769_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_14769_end_mask_0 = const()[name = tensor("op_14769_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14769_cast_fp16 = slice_by_index(begin = var_14769_begin_0, end = var_14769_end_0, end_mask = var_14769_end_mask_0, x = q_69_cast_fp16)[name = tensor("op_14769_cast_fp16")]; + tensor var_14773_begin_0 = const()[name = tensor("op_14773_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_14773_end_0 = const()[name = tensor("op_14773_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_14773_end_mask_0 = const()[name = tensor("op_14773_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14773_cast_fp16 = slice_by_index(begin = var_14773_begin_0, end = var_14773_end_0, end_mask = var_14773_end_mask_0, x = q_69_cast_fp16)[name = tensor("op_14773_cast_fp16")]; + tensor var_14777_begin_0 = const()[name = tensor("op_14777_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_14777_end_0 = const()[name = tensor("op_14777_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_14777_end_mask_0 = const()[name = tensor("op_14777_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14777_cast_fp16 = slice_by_index(begin = var_14777_begin_0, end = var_14777_end_0, end_mask = var_14777_end_mask_0, x = q_69_cast_fp16)[name = tensor("op_14777_cast_fp16")]; + tensor k_139_perm_0 = const()[name = tensor("k_139_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_14784_begin_0 = const()[name = tensor("op_14784_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14784_end_0 = const()[name = tensor("op_14784_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_14784_end_mask_0 = const()[name = tensor("op_14784_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_139_cast_fp16 = transpose(perm = k_139_perm_0, x = k_137_cast_fp16)[name = tensor("transpose_105")]; + tensor var_14784_cast_fp16 = slice_by_index(begin = var_14784_begin_0, end = var_14784_end_0, end_mask = var_14784_end_mask_0, x = k_139_cast_fp16)[name = tensor("op_14784_cast_fp16")]; + tensor var_14788_begin_0 = const()[name = tensor("op_14788_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_14788_end_0 = const()[name = tensor("op_14788_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_14788_end_mask_0 = const()[name = tensor("op_14788_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14788_cast_fp16 = slice_by_index(begin = var_14788_begin_0, end = var_14788_end_0, end_mask = var_14788_end_mask_0, x = k_139_cast_fp16)[name = tensor("op_14788_cast_fp16")]; + tensor var_14792_begin_0 = const()[name = tensor("op_14792_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_14792_end_0 = const()[name = tensor("op_14792_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_14792_end_mask_0 = const()[name = tensor("op_14792_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14792_cast_fp16 = slice_by_index(begin = var_14792_begin_0, end = var_14792_end_0, end_mask = var_14792_end_mask_0, x = k_139_cast_fp16)[name = tensor("op_14792_cast_fp16")]; + tensor var_14796_begin_0 = const()[name = tensor("op_14796_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_14796_end_0 = const()[name = tensor("op_14796_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_14796_end_mask_0 = const()[name = tensor("op_14796_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14796_cast_fp16 = slice_by_index(begin = var_14796_begin_0, end = var_14796_end_0, end_mask = var_14796_end_mask_0, x = k_139_cast_fp16)[name = tensor("op_14796_cast_fp16")]; + tensor var_14800_begin_0 = const()[name = tensor("op_14800_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_14800_end_0 = const()[name = tensor("op_14800_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_14800_end_mask_0 = const()[name = tensor("op_14800_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14800_cast_fp16 = slice_by_index(begin = var_14800_begin_0, end = var_14800_end_0, end_mask = var_14800_end_mask_0, x = k_139_cast_fp16)[name = tensor("op_14800_cast_fp16")]; + tensor var_14804_begin_0 = const()[name = tensor("op_14804_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_14804_end_0 = const()[name = tensor("op_14804_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_14804_end_mask_0 = const()[name = tensor("op_14804_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14804_cast_fp16 = slice_by_index(begin = var_14804_begin_0, end = var_14804_end_0, end_mask = var_14804_end_mask_0, x = k_139_cast_fp16)[name = tensor("op_14804_cast_fp16")]; + tensor var_14808_begin_0 = const()[name = tensor("op_14808_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_14808_end_0 = const()[name = tensor("op_14808_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_14808_end_mask_0 = const()[name = tensor("op_14808_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14808_cast_fp16 = slice_by_index(begin = var_14808_begin_0, end = var_14808_end_0, end_mask = var_14808_end_mask_0, x = k_139_cast_fp16)[name = tensor("op_14808_cast_fp16")]; + tensor var_14812_begin_0 = const()[name = tensor("op_14812_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_14812_end_0 = const()[name = tensor("op_14812_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_14812_end_mask_0 = const()[name = tensor("op_14812_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14812_cast_fp16 = slice_by_index(begin = var_14812_begin_0, end = var_14812_end_0, end_mask = var_14812_end_mask_0, x = k_139_cast_fp16)[name = tensor("op_14812_cast_fp16")]; + tensor var_14816_begin_0 = const()[name = tensor("op_14816_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_14816_end_0 = const()[name = tensor("op_14816_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_14816_end_mask_0 = const()[name = tensor("op_14816_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14816_cast_fp16 = slice_by_index(begin = var_14816_begin_0, end = var_14816_end_0, end_mask = var_14816_end_mask_0, x = k_139_cast_fp16)[name = tensor("op_14816_cast_fp16")]; + tensor var_14820_begin_0 = const()[name = tensor("op_14820_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_14820_end_0 = const()[name = tensor("op_14820_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_14820_end_mask_0 = const()[name = tensor("op_14820_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14820_cast_fp16 = slice_by_index(begin = var_14820_begin_0, end = var_14820_end_0, end_mask = var_14820_end_mask_0, x = k_139_cast_fp16)[name = tensor("op_14820_cast_fp16")]; + tensor var_14824_begin_0 = const()[name = tensor("op_14824_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_14824_end_0 = const()[name = tensor("op_14824_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_14824_end_mask_0 = const()[name = tensor("op_14824_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14824_cast_fp16 = slice_by_index(begin = var_14824_begin_0, end = var_14824_end_0, end_mask = var_14824_end_mask_0, x = k_139_cast_fp16)[name = tensor("op_14824_cast_fp16")]; + tensor var_14828_begin_0 = const()[name = tensor("op_14828_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_14828_end_0 = const()[name = tensor("op_14828_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_14828_end_mask_0 = const()[name = tensor("op_14828_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14828_cast_fp16 = slice_by_index(begin = var_14828_begin_0, end = var_14828_end_0, end_mask = var_14828_end_mask_0, x = k_139_cast_fp16)[name = tensor("op_14828_cast_fp16")]; + tensor var_14832_begin_0 = const()[name = tensor("op_14832_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_14832_end_0 = const()[name = tensor("op_14832_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_14832_end_mask_0 = const()[name = tensor("op_14832_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14832_cast_fp16 = slice_by_index(begin = var_14832_begin_0, end = var_14832_end_0, end_mask = var_14832_end_mask_0, x = k_139_cast_fp16)[name = tensor("op_14832_cast_fp16")]; + tensor var_14836_begin_0 = const()[name = tensor("op_14836_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_14836_end_0 = const()[name = tensor("op_14836_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_14836_end_mask_0 = const()[name = tensor("op_14836_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14836_cast_fp16 = slice_by_index(begin = var_14836_begin_0, end = var_14836_end_0, end_mask = var_14836_end_mask_0, x = k_139_cast_fp16)[name = tensor("op_14836_cast_fp16")]; + tensor var_14840_begin_0 = const()[name = tensor("op_14840_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_14840_end_0 = const()[name = tensor("op_14840_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_14840_end_mask_0 = const()[name = tensor("op_14840_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14840_cast_fp16 = slice_by_index(begin = var_14840_begin_0, end = var_14840_end_0, end_mask = var_14840_end_mask_0, x = k_139_cast_fp16)[name = tensor("op_14840_cast_fp16")]; + tensor var_14844_begin_0 = const()[name = tensor("op_14844_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_14844_end_0 = const()[name = tensor("op_14844_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_14844_end_mask_0 = const()[name = tensor("op_14844_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14844_cast_fp16 = slice_by_index(begin = var_14844_begin_0, end = var_14844_end_0, end_mask = var_14844_end_mask_0, x = k_139_cast_fp16)[name = tensor("op_14844_cast_fp16")]; + tensor var_14848_begin_0 = const()[name = tensor("op_14848_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_14848_end_0 = const()[name = tensor("op_14848_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_14848_end_mask_0 = const()[name = tensor("op_14848_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14848_cast_fp16 = slice_by_index(begin = var_14848_begin_0, end = var_14848_end_0, end_mask = var_14848_end_mask_0, x = k_139_cast_fp16)[name = tensor("op_14848_cast_fp16")]; + tensor var_14852_begin_0 = const()[name = tensor("op_14852_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_14852_end_0 = const()[name = tensor("op_14852_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_14852_end_mask_0 = const()[name = tensor("op_14852_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14852_cast_fp16 = slice_by_index(begin = var_14852_begin_0, end = var_14852_end_0, end_mask = var_14852_end_mask_0, x = k_139_cast_fp16)[name = tensor("op_14852_cast_fp16")]; + tensor var_14856_begin_0 = const()[name = tensor("op_14856_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_14856_end_0 = const()[name = tensor("op_14856_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_14856_end_mask_0 = const()[name = tensor("op_14856_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14856_cast_fp16 = slice_by_index(begin = var_14856_begin_0, end = var_14856_end_0, end_mask = var_14856_end_mask_0, x = k_139_cast_fp16)[name = tensor("op_14856_cast_fp16")]; + tensor var_14860_begin_0 = const()[name = tensor("op_14860_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_14860_end_0 = const()[name = tensor("op_14860_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_14860_end_mask_0 = const()[name = tensor("op_14860_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14860_cast_fp16 = slice_by_index(begin = var_14860_begin_0, end = var_14860_end_0, end_mask = var_14860_end_mask_0, x = k_139_cast_fp16)[name = tensor("op_14860_cast_fp16")]; + tensor var_14862_begin_0 = const()[name = tensor("op_14862_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14862_end_0 = const()[name = tensor("op_14862_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_14862_end_mask_0 = const()[name = tensor("op_14862_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14862_cast_fp16 = slice_by_index(begin = var_14862_begin_0, end = var_14862_end_0, end_mask = var_14862_end_mask_0, x = v_69_cast_fp16)[name = tensor("op_14862_cast_fp16")]; + tensor var_14866_begin_0 = const()[name = tensor("op_14866_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_14866_end_0 = const()[name = tensor("op_14866_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_14866_end_mask_0 = const()[name = tensor("op_14866_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14866_cast_fp16 = slice_by_index(begin = var_14866_begin_0, end = var_14866_end_0, end_mask = var_14866_end_mask_0, x = v_69_cast_fp16)[name = tensor("op_14866_cast_fp16")]; + tensor var_14870_begin_0 = const()[name = tensor("op_14870_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_14870_end_0 = const()[name = tensor("op_14870_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_14870_end_mask_0 = const()[name = tensor("op_14870_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14870_cast_fp16 = slice_by_index(begin = var_14870_begin_0, end = var_14870_end_0, end_mask = var_14870_end_mask_0, x = v_69_cast_fp16)[name = tensor("op_14870_cast_fp16")]; + tensor var_14874_begin_0 = const()[name = tensor("op_14874_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_14874_end_0 = const()[name = tensor("op_14874_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_14874_end_mask_0 = const()[name = tensor("op_14874_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14874_cast_fp16 = slice_by_index(begin = var_14874_begin_0, end = var_14874_end_0, end_mask = var_14874_end_mask_0, x = v_69_cast_fp16)[name = tensor("op_14874_cast_fp16")]; + tensor var_14878_begin_0 = const()[name = tensor("op_14878_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_14878_end_0 = const()[name = tensor("op_14878_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_14878_end_mask_0 = const()[name = tensor("op_14878_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14878_cast_fp16 = slice_by_index(begin = var_14878_begin_0, end = var_14878_end_0, end_mask = var_14878_end_mask_0, x = v_69_cast_fp16)[name = tensor("op_14878_cast_fp16")]; + tensor var_14882_begin_0 = const()[name = tensor("op_14882_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_14882_end_0 = const()[name = tensor("op_14882_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_14882_end_mask_0 = const()[name = tensor("op_14882_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14882_cast_fp16 = slice_by_index(begin = var_14882_begin_0, end = var_14882_end_0, end_mask = var_14882_end_mask_0, x = v_69_cast_fp16)[name = tensor("op_14882_cast_fp16")]; + tensor var_14886_begin_0 = const()[name = tensor("op_14886_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_14886_end_0 = const()[name = tensor("op_14886_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_14886_end_mask_0 = const()[name = tensor("op_14886_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14886_cast_fp16 = slice_by_index(begin = var_14886_begin_0, end = var_14886_end_0, end_mask = var_14886_end_mask_0, x = v_69_cast_fp16)[name = tensor("op_14886_cast_fp16")]; + tensor var_14890_begin_0 = const()[name = tensor("op_14890_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_14890_end_0 = const()[name = tensor("op_14890_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_14890_end_mask_0 = const()[name = tensor("op_14890_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14890_cast_fp16 = slice_by_index(begin = var_14890_begin_0, end = var_14890_end_0, end_mask = var_14890_end_mask_0, x = v_69_cast_fp16)[name = tensor("op_14890_cast_fp16")]; + tensor var_14894_begin_0 = const()[name = tensor("op_14894_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_14894_end_0 = const()[name = tensor("op_14894_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_14894_end_mask_0 = const()[name = tensor("op_14894_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14894_cast_fp16 = slice_by_index(begin = var_14894_begin_0, end = var_14894_end_0, end_mask = var_14894_end_mask_0, x = v_69_cast_fp16)[name = tensor("op_14894_cast_fp16")]; + tensor var_14898_begin_0 = const()[name = tensor("op_14898_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_14898_end_0 = const()[name = tensor("op_14898_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_14898_end_mask_0 = const()[name = tensor("op_14898_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14898_cast_fp16 = slice_by_index(begin = var_14898_begin_0, end = var_14898_end_0, end_mask = var_14898_end_mask_0, x = v_69_cast_fp16)[name = tensor("op_14898_cast_fp16")]; + tensor var_14902_begin_0 = const()[name = tensor("op_14902_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_14902_end_0 = const()[name = tensor("op_14902_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_14902_end_mask_0 = const()[name = tensor("op_14902_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14902_cast_fp16 = slice_by_index(begin = var_14902_begin_0, end = var_14902_end_0, end_mask = var_14902_end_mask_0, x = v_69_cast_fp16)[name = tensor("op_14902_cast_fp16")]; + tensor var_14906_begin_0 = const()[name = tensor("op_14906_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_14906_end_0 = const()[name = tensor("op_14906_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_14906_end_mask_0 = const()[name = tensor("op_14906_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14906_cast_fp16 = slice_by_index(begin = var_14906_begin_0, end = var_14906_end_0, end_mask = var_14906_end_mask_0, x = v_69_cast_fp16)[name = tensor("op_14906_cast_fp16")]; + tensor var_14910_begin_0 = const()[name = tensor("op_14910_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_14910_end_0 = const()[name = tensor("op_14910_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_14910_end_mask_0 = const()[name = tensor("op_14910_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14910_cast_fp16 = slice_by_index(begin = var_14910_begin_0, end = var_14910_end_0, end_mask = var_14910_end_mask_0, x = v_69_cast_fp16)[name = tensor("op_14910_cast_fp16")]; + tensor var_14914_begin_0 = const()[name = tensor("op_14914_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_14914_end_0 = const()[name = tensor("op_14914_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_14914_end_mask_0 = const()[name = tensor("op_14914_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14914_cast_fp16 = slice_by_index(begin = var_14914_begin_0, end = var_14914_end_0, end_mask = var_14914_end_mask_0, x = v_69_cast_fp16)[name = tensor("op_14914_cast_fp16")]; + tensor var_14918_begin_0 = const()[name = tensor("op_14918_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_14918_end_0 = const()[name = tensor("op_14918_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_14918_end_mask_0 = const()[name = tensor("op_14918_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14918_cast_fp16 = slice_by_index(begin = var_14918_begin_0, end = var_14918_end_0, end_mask = var_14918_end_mask_0, x = v_69_cast_fp16)[name = tensor("op_14918_cast_fp16")]; + tensor var_14922_begin_0 = const()[name = tensor("op_14922_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_14922_end_0 = const()[name = tensor("op_14922_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_14922_end_mask_0 = const()[name = tensor("op_14922_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14922_cast_fp16 = slice_by_index(begin = var_14922_begin_0, end = var_14922_end_0, end_mask = var_14922_end_mask_0, x = v_69_cast_fp16)[name = tensor("op_14922_cast_fp16")]; + tensor var_14926_begin_0 = const()[name = tensor("op_14926_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_14926_end_0 = const()[name = tensor("op_14926_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_14926_end_mask_0 = const()[name = tensor("op_14926_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14926_cast_fp16 = slice_by_index(begin = var_14926_begin_0, end = var_14926_end_0, end_mask = var_14926_end_mask_0, x = v_69_cast_fp16)[name = tensor("op_14926_cast_fp16")]; + tensor var_14930_begin_0 = const()[name = tensor("op_14930_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_14930_end_0 = const()[name = tensor("op_14930_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_14930_end_mask_0 = const()[name = tensor("op_14930_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14930_cast_fp16 = slice_by_index(begin = var_14930_begin_0, end = var_14930_end_0, end_mask = var_14930_end_mask_0, x = v_69_cast_fp16)[name = tensor("op_14930_cast_fp16")]; + tensor var_14934_begin_0 = const()[name = tensor("op_14934_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_14934_end_0 = const()[name = tensor("op_14934_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_14934_end_mask_0 = const()[name = tensor("op_14934_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14934_cast_fp16 = slice_by_index(begin = var_14934_begin_0, end = var_14934_end_0, end_mask = var_14934_end_mask_0, x = v_69_cast_fp16)[name = tensor("op_14934_cast_fp16")]; + tensor var_14938_begin_0 = const()[name = tensor("op_14938_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_14938_end_0 = const()[name = tensor("op_14938_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_14938_end_mask_0 = const()[name = tensor("op_14938_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14938_cast_fp16 = slice_by_index(begin = var_14938_begin_0, end = var_14938_end_0, end_mask = var_14938_end_mask_0, x = v_69_cast_fp16)[name = tensor("op_14938_cast_fp16")]; + tensor var_14942_equation_0 = const()[name = tensor("op_14942_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14942_cast_fp16 = einsum(equation = var_14942_equation_0, values = (var_14784_cast_fp16, var_14701_cast_fp16))[name = tensor("op_14942_cast_fp16")]; + tensor var_14943_to_fp16 = const()[name = tensor("op_14943_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1201_cast_fp16 = mul(x = var_14942_cast_fp16, y = var_14943_to_fp16)[name = tensor("aw_1201_cast_fp16")]; + tensor var_14946_equation_0 = const()[name = tensor("op_14946_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14946_cast_fp16 = einsum(equation = var_14946_equation_0, values = (var_14788_cast_fp16, var_14705_cast_fp16))[name = tensor("op_14946_cast_fp16")]; + tensor var_14947_to_fp16 = const()[name = tensor("op_14947_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1203_cast_fp16 = mul(x = var_14946_cast_fp16, y = var_14947_to_fp16)[name = tensor("aw_1203_cast_fp16")]; + tensor var_14950_equation_0 = const()[name = tensor("op_14950_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14950_cast_fp16 = einsum(equation = var_14950_equation_0, values = (var_14792_cast_fp16, var_14709_cast_fp16))[name = tensor("op_14950_cast_fp16")]; + tensor var_14951_to_fp16 = const()[name = tensor("op_14951_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1205_cast_fp16 = mul(x = var_14950_cast_fp16, y = var_14951_to_fp16)[name = tensor("aw_1205_cast_fp16")]; + tensor var_14954_equation_0 = const()[name = tensor("op_14954_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14954_cast_fp16 = einsum(equation = var_14954_equation_0, values = (var_14796_cast_fp16, var_14713_cast_fp16))[name = tensor("op_14954_cast_fp16")]; + tensor var_14955_to_fp16 = const()[name = tensor("op_14955_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1207_cast_fp16 = mul(x = var_14954_cast_fp16, y = var_14955_to_fp16)[name = tensor("aw_1207_cast_fp16")]; + tensor var_14958_equation_0 = const()[name = tensor("op_14958_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14958_cast_fp16 = einsum(equation = var_14958_equation_0, values = (var_14800_cast_fp16, var_14717_cast_fp16))[name = tensor("op_14958_cast_fp16")]; + tensor var_14959_to_fp16 = const()[name = tensor("op_14959_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1209_cast_fp16 = mul(x = var_14958_cast_fp16, y = var_14959_to_fp16)[name = tensor("aw_1209_cast_fp16")]; + tensor var_14962_equation_0 = const()[name = tensor("op_14962_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14962_cast_fp16 = einsum(equation = var_14962_equation_0, values = (var_14804_cast_fp16, var_14721_cast_fp16))[name = tensor("op_14962_cast_fp16")]; + tensor var_14963_to_fp16 = const()[name = tensor("op_14963_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1211_cast_fp16 = mul(x = var_14962_cast_fp16, y = var_14963_to_fp16)[name = tensor("aw_1211_cast_fp16")]; + tensor var_14966_equation_0 = const()[name = tensor("op_14966_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14966_cast_fp16 = einsum(equation = var_14966_equation_0, values = (var_14808_cast_fp16, var_14725_cast_fp16))[name = tensor("op_14966_cast_fp16")]; + tensor var_14967_to_fp16 = const()[name = tensor("op_14967_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1213_cast_fp16 = mul(x = var_14966_cast_fp16, y = var_14967_to_fp16)[name = tensor("aw_1213_cast_fp16")]; + tensor var_14970_equation_0 = const()[name = tensor("op_14970_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14970_cast_fp16 = einsum(equation = var_14970_equation_0, values = (var_14812_cast_fp16, var_14729_cast_fp16))[name = tensor("op_14970_cast_fp16")]; + tensor var_14971_to_fp16 = const()[name = tensor("op_14971_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1215_cast_fp16 = mul(x = var_14970_cast_fp16, y = var_14971_to_fp16)[name = tensor("aw_1215_cast_fp16")]; + tensor var_14974_equation_0 = const()[name = tensor("op_14974_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14974_cast_fp16 = einsum(equation = var_14974_equation_0, values = (var_14816_cast_fp16, var_14733_cast_fp16))[name = tensor("op_14974_cast_fp16")]; + tensor var_14975_to_fp16 = const()[name = tensor("op_14975_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1217_cast_fp16 = mul(x = var_14974_cast_fp16, y = var_14975_to_fp16)[name = tensor("aw_1217_cast_fp16")]; + tensor var_14978_equation_0 = const()[name = tensor("op_14978_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14978_cast_fp16 = einsum(equation = var_14978_equation_0, values = (var_14820_cast_fp16, var_14737_cast_fp16))[name = tensor("op_14978_cast_fp16")]; + tensor var_14979_to_fp16 = const()[name = tensor("op_14979_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1219_cast_fp16 = mul(x = var_14978_cast_fp16, y = var_14979_to_fp16)[name = tensor("aw_1219_cast_fp16")]; + tensor var_14982_equation_0 = const()[name = tensor("op_14982_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14982_cast_fp16 = einsum(equation = var_14982_equation_0, values = (var_14824_cast_fp16, var_14741_cast_fp16))[name = tensor("op_14982_cast_fp16")]; + tensor var_14983_to_fp16 = const()[name = tensor("op_14983_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1221_cast_fp16 = mul(x = var_14982_cast_fp16, y = var_14983_to_fp16)[name = tensor("aw_1221_cast_fp16")]; + tensor var_14986_equation_0 = const()[name = tensor("op_14986_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14986_cast_fp16 = einsum(equation = var_14986_equation_0, values = (var_14828_cast_fp16, var_14745_cast_fp16))[name = tensor("op_14986_cast_fp16")]; + tensor var_14987_to_fp16 = const()[name = tensor("op_14987_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1223_cast_fp16 = mul(x = var_14986_cast_fp16, y = var_14987_to_fp16)[name = tensor("aw_1223_cast_fp16")]; + tensor var_14990_equation_0 = const()[name = tensor("op_14990_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14990_cast_fp16 = einsum(equation = var_14990_equation_0, values = (var_14832_cast_fp16, var_14749_cast_fp16))[name = tensor("op_14990_cast_fp16")]; + tensor var_14991_to_fp16 = const()[name = tensor("op_14991_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1225_cast_fp16 = mul(x = var_14990_cast_fp16, y = var_14991_to_fp16)[name = tensor("aw_1225_cast_fp16")]; + tensor var_14994_equation_0 = const()[name = tensor("op_14994_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14994_cast_fp16 = einsum(equation = var_14994_equation_0, values = (var_14836_cast_fp16, var_14753_cast_fp16))[name = tensor("op_14994_cast_fp16")]; + tensor var_14995_to_fp16 = const()[name = tensor("op_14995_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1227_cast_fp16 = mul(x = var_14994_cast_fp16, y = var_14995_to_fp16)[name = tensor("aw_1227_cast_fp16")]; + tensor var_14998_equation_0 = const()[name = tensor("op_14998_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14998_cast_fp16 = einsum(equation = var_14998_equation_0, values = (var_14840_cast_fp16, var_14757_cast_fp16))[name = tensor("op_14998_cast_fp16")]; + tensor var_14999_to_fp16 = const()[name = tensor("op_14999_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1229_cast_fp16 = mul(x = var_14998_cast_fp16, y = var_14999_to_fp16)[name = tensor("aw_1229_cast_fp16")]; + tensor var_15002_equation_0 = const()[name = tensor("op_15002_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15002_cast_fp16 = einsum(equation = var_15002_equation_0, values = (var_14844_cast_fp16, var_14761_cast_fp16))[name = tensor("op_15002_cast_fp16")]; + tensor var_15003_to_fp16 = const()[name = tensor("op_15003_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1231_cast_fp16 = mul(x = var_15002_cast_fp16, y = var_15003_to_fp16)[name = tensor("aw_1231_cast_fp16")]; + tensor var_15006_equation_0 = const()[name = tensor("op_15006_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15006_cast_fp16 = einsum(equation = var_15006_equation_0, values = (var_14848_cast_fp16, var_14765_cast_fp16))[name = tensor("op_15006_cast_fp16")]; + tensor var_15007_to_fp16 = const()[name = tensor("op_15007_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1233_cast_fp16 = mul(x = var_15006_cast_fp16, y = var_15007_to_fp16)[name = tensor("aw_1233_cast_fp16")]; + tensor var_15010_equation_0 = const()[name = tensor("op_15010_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15010_cast_fp16 = einsum(equation = var_15010_equation_0, values = (var_14852_cast_fp16, var_14769_cast_fp16))[name = tensor("op_15010_cast_fp16")]; + tensor var_15011_to_fp16 = const()[name = tensor("op_15011_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1235_cast_fp16 = mul(x = var_15010_cast_fp16, y = var_15011_to_fp16)[name = tensor("aw_1235_cast_fp16")]; + tensor var_15014_equation_0 = const()[name = tensor("op_15014_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15014_cast_fp16 = einsum(equation = var_15014_equation_0, values = (var_14856_cast_fp16, var_14773_cast_fp16))[name = tensor("op_15014_cast_fp16")]; + tensor var_15015_to_fp16 = const()[name = tensor("op_15015_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1237_cast_fp16 = mul(x = var_15014_cast_fp16, y = var_15015_to_fp16)[name = tensor("aw_1237_cast_fp16")]; + tensor var_15018_equation_0 = const()[name = tensor("op_15018_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15018_cast_fp16 = einsum(equation = var_15018_equation_0, values = (var_14860_cast_fp16, var_14777_cast_fp16))[name = tensor("op_15018_cast_fp16")]; + tensor var_15019_to_fp16 = const()[name = tensor("op_15019_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1239_cast_fp16 = mul(x = var_15018_cast_fp16, y = var_15019_to_fp16)[name = tensor("aw_1239_cast_fp16")]; + tensor var_15021_cast_fp16 = softmax(axis = var_2624, x = aw_1201_cast_fp16)[name = tensor("op_15021_cast_fp16")]; + tensor var_15022_cast_fp16 = softmax(axis = var_2624, x = aw_1203_cast_fp16)[name = tensor("op_15022_cast_fp16")]; + tensor var_15023_cast_fp16 = softmax(axis = var_2624, x = aw_1205_cast_fp16)[name = tensor("op_15023_cast_fp16")]; + tensor var_15024_cast_fp16 = softmax(axis = var_2624, x = aw_1207_cast_fp16)[name = tensor("op_15024_cast_fp16")]; + tensor var_15025_cast_fp16 = softmax(axis = var_2624, x = aw_1209_cast_fp16)[name = tensor("op_15025_cast_fp16")]; + tensor var_15026_cast_fp16 = softmax(axis = var_2624, x = aw_1211_cast_fp16)[name = tensor("op_15026_cast_fp16")]; + tensor var_15027_cast_fp16 = softmax(axis = var_2624, x = aw_1213_cast_fp16)[name = tensor("op_15027_cast_fp16")]; + tensor var_15028_cast_fp16 = softmax(axis = var_2624, x = aw_1215_cast_fp16)[name = tensor("op_15028_cast_fp16")]; + tensor var_15029_cast_fp16 = softmax(axis = var_2624, x = aw_1217_cast_fp16)[name = tensor("op_15029_cast_fp16")]; + tensor var_15030_cast_fp16 = softmax(axis = var_2624, x = aw_1219_cast_fp16)[name = tensor("op_15030_cast_fp16")]; + tensor var_15031_cast_fp16 = softmax(axis = var_2624, x = aw_1221_cast_fp16)[name = tensor("op_15031_cast_fp16")]; + tensor var_15032_cast_fp16 = softmax(axis = var_2624, x = aw_1223_cast_fp16)[name = tensor("op_15032_cast_fp16")]; + tensor var_15033_cast_fp16 = softmax(axis = var_2624, x = aw_1225_cast_fp16)[name = tensor("op_15033_cast_fp16")]; + tensor var_15034_cast_fp16 = softmax(axis = var_2624, x = aw_1227_cast_fp16)[name = tensor("op_15034_cast_fp16")]; + tensor var_15035_cast_fp16 = softmax(axis = var_2624, x = aw_1229_cast_fp16)[name = tensor("op_15035_cast_fp16")]; + tensor var_15036_cast_fp16 = softmax(axis = var_2624, x = aw_1231_cast_fp16)[name = tensor("op_15036_cast_fp16")]; + tensor var_15037_cast_fp16 = softmax(axis = var_2624, x = aw_1233_cast_fp16)[name = tensor("op_15037_cast_fp16")]; + tensor var_15038_cast_fp16 = softmax(axis = var_2624, x = aw_1235_cast_fp16)[name = tensor("op_15038_cast_fp16")]; + tensor var_15039_cast_fp16 = softmax(axis = var_2624, x = aw_1237_cast_fp16)[name = tensor("op_15039_cast_fp16")]; + tensor var_15040_cast_fp16 = softmax(axis = var_2624, x = aw_1239_cast_fp16)[name = tensor("op_15040_cast_fp16")]; + tensor var_15042_equation_0 = const()[name = tensor("op_15042_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15042_cast_fp16 = einsum(equation = var_15042_equation_0, values = (var_14862_cast_fp16, var_15021_cast_fp16))[name = tensor("op_15042_cast_fp16")]; + tensor var_15044_equation_0 = const()[name = tensor("op_15044_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15044_cast_fp16 = einsum(equation = var_15044_equation_0, values = (var_14866_cast_fp16, var_15022_cast_fp16))[name = tensor("op_15044_cast_fp16")]; + tensor var_15046_equation_0 = const()[name = tensor("op_15046_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15046_cast_fp16 = einsum(equation = var_15046_equation_0, values = (var_14870_cast_fp16, var_15023_cast_fp16))[name = tensor("op_15046_cast_fp16")]; + tensor var_15048_equation_0 = const()[name = tensor("op_15048_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15048_cast_fp16 = einsum(equation = var_15048_equation_0, values = (var_14874_cast_fp16, var_15024_cast_fp16))[name = tensor("op_15048_cast_fp16")]; + tensor var_15050_equation_0 = const()[name = tensor("op_15050_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15050_cast_fp16 = einsum(equation = var_15050_equation_0, values = (var_14878_cast_fp16, var_15025_cast_fp16))[name = tensor("op_15050_cast_fp16")]; + tensor var_15052_equation_0 = const()[name = tensor("op_15052_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15052_cast_fp16 = einsum(equation = var_15052_equation_0, values = (var_14882_cast_fp16, var_15026_cast_fp16))[name = tensor("op_15052_cast_fp16")]; + tensor var_15054_equation_0 = const()[name = tensor("op_15054_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15054_cast_fp16 = einsum(equation = var_15054_equation_0, values = (var_14886_cast_fp16, var_15027_cast_fp16))[name = tensor("op_15054_cast_fp16")]; + tensor var_15056_equation_0 = const()[name = tensor("op_15056_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15056_cast_fp16 = einsum(equation = var_15056_equation_0, values = (var_14890_cast_fp16, var_15028_cast_fp16))[name = tensor("op_15056_cast_fp16")]; + tensor var_15058_equation_0 = const()[name = tensor("op_15058_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15058_cast_fp16 = einsum(equation = var_15058_equation_0, values = (var_14894_cast_fp16, var_15029_cast_fp16))[name = tensor("op_15058_cast_fp16")]; + tensor var_15060_equation_0 = const()[name = tensor("op_15060_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15060_cast_fp16 = einsum(equation = var_15060_equation_0, values = (var_14898_cast_fp16, var_15030_cast_fp16))[name = tensor("op_15060_cast_fp16")]; + tensor var_15062_equation_0 = const()[name = tensor("op_15062_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15062_cast_fp16 = einsum(equation = var_15062_equation_0, values = (var_14902_cast_fp16, var_15031_cast_fp16))[name = tensor("op_15062_cast_fp16")]; + tensor var_15064_equation_0 = const()[name = tensor("op_15064_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15064_cast_fp16 = einsum(equation = var_15064_equation_0, values = (var_14906_cast_fp16, var_15032_cast_fp16))[name = tensor("op_15064_cast_fp16")]; + tensor var_15066_equation_0 = const()[name = tensor("op_15066_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15066_cast_fp16 = einsum(equation = var_15066_equation_0, values = (var_14910_cast_fp16, var_15033_cast_fp16))[name = tensor("op_15066_cast_fp16")]; + tensor var_15068_equation_0 = const()[name = tensor("op_15068_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15068_cast_fp16 = einsum(equation = var_15068_equation_0, values = (var_14914_cast_fp16, var_15034_cast_fp16))[name = tensor("op_15068_cast_fp16")]; + tensor var_15070_equation_0 = const()[name = tensor("op_15070_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15070_cast_fp16 = einsum(equation = var_15070_equation_0, values = (var_14918_cast_fp16, var_15035_cast_fp16))[name = tensor("op_15070_cast_fp16")]; + tensor var_15072_equation_0 = const()[name = tensor("op_15072_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15072_cast_fp16 = einsum(equation = var_15072_equation_0, values = (var_14922_cast_fp16, var_15036_cast_fp16))[name = tensor("op_15072_cast_fp16")]; + tensor var_15074_equation_0 = const()[name = tensor("op_15074_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15074_cast_fp16 = einsum(equation = var_15074_equation_0, values = (var_14926_cast_fp16, var_15037_cast_fp16))[name = tensor("op_15074_cast_fp16")]; + tensor var_15076_equation_0 = const()[name = tensor("op_15076_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15076_cast_fp16 = einsum(equation = var_15076_equation_0, values = (var_14930_cast_fp16, var_15038_cast_fp16))[name = tensor("op_15076_cast_fp16")]; + tensor var_15078_equation_0 = const()[name = tensor("op_15078_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15078_cast_fp16 = einsum(equation = var_15078_equation_0, values = (var_14934_cast_fp16, var_15039_cast_fp16))[name = tensor("op_15078_cast_fp16")]; + tensor var_15080_equation_0 = const()[name = tensor("op_15080_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15080_cast_fp16 = einsum(equation = var_15080_equation_0, values = (var_14938_cast_fp16, var_15040_cast_fp16))[name = tensor("op_15080_cast_fp16")]; + tensor input_253_interleave_0 = const()[name = tensor("input_253_interleave_0"), val = tensor(false)]; + tensor input_253_cast_fp16 = concat(axis = var_2624, interleave = input_253_interleave_0, values = (var_15042_cast_fp16, var_15044_cast_fp16, var_15046_cast_fp16, var_15048_cast_fp16, var_15050_cast_fp16, var_15052_cast_fp16, var_15054_cast_fp16, var_15056_cast_fp16, var_15058_cast_fp16, var_15060_cast_fp16, var_15062_cast_fp16, var_15064_cast_fp16, var_15066_cast_fp16, var_15068_cast_fp16, var_15070_cast_fp16, var_15072_cast_fp16, var_15074_cast_fp16, var_15076_cast_fp16, var_15078_cast_fp16, var_15080_cast_fp16))[name = tensor("input_253_cast_fp16")]; + tensor var_15086 = const()[name = tensor("op_15086"), val = tensor([1, 1])]; + tensor var_15088 = const()[name = tensor("op_15088"), val = tensor([1, 1])]; + tensor var_15090_pad_type_0 = const()[name = tensor("op_15090_pad_type_0"), val = tensor("custom")]; + tensor var_15090_pad_0 = const()[name = tensor("op_15090_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_3_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_3_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1180512192)))]; + tensor down_blocks_2_attentions_1_transformer_blocks_3_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_3_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1183789056)))]; + tensor var_15090_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_3_attn1_to_out_0_bias_to_fp16, dilations = var_15088, groups = var_2624, pad = var_15090_pad_0, pad_type = var_15090_pad_type_0, strides = var_15086, weight = down_blocks_2_attentions_1_transformer_blocks_3_attn1_to_out_0_weight_to_fp16, x = input_253_cast_fp16)[name = tensor("op_15090_cast_fp16")]; + tensor inputs_105_cast_fp16 = add(x = var_15090_cast_fp16, y = inputs_103_cast_fp16)[name = tensor("inputs_105_cast_fp16")]; + tensor hidden_states_157_axes_0 = const()[name = tensor("hidden_states_157_axes_0"), val = tensor([1])]; + tensor hidden_states_157_gamma_0_to_fp16 = const()[name = tensor("hidden_states_157_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1183791680)))]; + tensor hidden_states_157_beta_0_to_fp16 = const()[name = tensor("hidden_states_157_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1183794304)))]; + tensor var_15100_to_fp16 = const()[name = tensor("op_15100_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_157_cast_fp16 = layer_norm(axes = hidden_states_157_axes_0, beta = hidden_states_157_beta_0_to_fp16, epsilon = var_15100_to_fp16, gamma = hidden_states_157_gamma_0_to_fp16, x = inputs_105_cast_fp16)[name = tensor("hidden_states_157_cast_fp16")]; + tensor var_15115 = const()[name = tensor("op_15115"), val = tensor([1, 1])]; + tensor var_15117 = const()[name = tensor("op_15117"), val = tensor([1, 1])]; + tensor q_71_pad_type_0 = const()[name = tensor("q_71_pad_type_0"), val = tensor("custom")]; + tensor q_71_pad_0 = const()[name = tensor("q_71_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_3_attn2_to_q_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_3_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1183796928)))]; + tensor q_71_cast_fp16 = conv(dilations = var_15117, groups = var_2624, pad = q_71_pad_0, pad_type = q_71_pad_type_0, strides = var_15115, weight = down_blocks_2_attentions_1_transformer_blocks_3_attn2_to_q_weight_to_fp16, x = hidden_states_157_cast_fp16)[name = tensor("q_71_cast_fp16")]; + tensor var_15121 = const()[name = tensor("op_15121"), val = tensor([1, 1])]; + tensor var_15123 = const()[name = tensor("op_15123"), val = tensor([1, 1])]; + tensor k_141_pad_type_0 = const()[name = tensor("k_141_pad_type_0"), val = tensor("custom")]; + tensor k_141_pad_0 = const()[name = tensor("k_141_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_3_attn2_to_k_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_3_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1187073792)))]; + tensor k_141_cast_fp16 = conv(dilations = var_15123, groups = var_2624, pad = k_141_pad_0, pad_type = k_141_pad_type_0, strides = var_15121, weight = down_blocks_2_attentions_1_transformer_blocks_3_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_141_cast_fp16")]; + tensor var_15127 = const()[name = tensor("op_15127"), val = tensor([1, 1])]; + tensor var_15129 = const()[name = tensor("op_15129"), val = tensor([1, 1])]; + tensor v_71_pad_type_0 = const()[name = tensor("v_71_pad_type_0"), val = tensor("custom")]; + tensor v_71_pad_0 = const()[name = tensor("v_71_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_3_attn2_to_v_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_3_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1192316736)))]; + tensor v_71_cast_fp16 = conv(dilations = var_15129, groups = var_2624, pad = v_71_pad_0, pad_type = v_71_pad_type_0, strides = var_15127, weight = down_blocks_2_attentions_1_transformer_blocks_3_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_71_cast_fp16")]; + tensor var_15133_begin_0 = const()[name = tensor("op_15133_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15133_end_0 = const()[name = tensor("op_15133_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_15133_end_mask_0 = const()[name = tensor("op_15133_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15133_cast_fp16 = slice_by_index(begin = var_15133_begin_0, end = var_15133_end_0, end_mask = var_15133_end_mask_0, x = q_71_cast_fp16)[name = tensor("op_15133_cast_fp16")]; + tensor var_15137_begin_0 = const()[name = tensor("op_15137_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_15137_end_0 = const()[name = tensor("op_15137_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_15137_end_mask_0 = const()[name = tensor("op_15137_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15137_cast_fp16 = slice_by_index(begin = var_15137_begin_0, end = var_15137_end_0, end_mask = var_15137_end_mask_0, x = q_71_cast_fp16)[name = tensor("op_15137_cast_fp16")]; + tensor var_15141_begin_0 = const()[name = tensor("op_15141_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_15141_end_0 = const()[name = tensor("op_15141_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_15141_end_mask_0 = const()[name = tensor("op_15141_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15141_cast_fp16 = slice_by_index(begin = var_15141_begin_0, end = var_15141_end_0, end_mask = var_15141_end_mask_0, x = q_71_cast_fp16)[name = tensor("op_15141_cast_fp16")]; + tensor var_15145_begin_0 = const()[name = tensor("op_15145_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_15145_end_0 = const()[name = tensor("op_15145_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_15145_end_mask_0 = const()[name = tensor("op_15145_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15145_cast_fp16 = slice_by_index(begin = var_15145_begin_0, end = var_15145_end_0, end_mask = var_15145_end_mask_0, x = q_71_cast_fp16)[name = tensor("op_15145_cast_fp16")]; + tensor var_15149_begin_0 = const()[name = tensor("op_15149_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_15149_end_0 = const()[name = tensor("op_15149_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_15149_end_mask_0 = const()[name = tensor("op_15149_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15149_cast_fp16 = slice_by_index(begin = var_15149_begin_0, end = var_15149_end_0, end_mask = var_15149_end_mask_0, x = q_71_cast_fp16)[name = tensor("op_15149_cast_fp16")]; + tensor var_15153_begin_0 = const()[name = tensor("op_15153_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_15153_end_0 = const()[name = tensor("op_15153_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_15153_end_mask_0 = const()[name = tensor("op_15153_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15153_cast_fp16 = slice_by_index(begin = var_15153_begin_0, end = var_15153_end_0, end_mask = var_15153_end_mask_0, x = q_71_cast_fp16)[name = tensor("op_15153_cast_fp16")]; + tensor var_15157_begin_0 = const()[name = tensor("op_15157_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_15157_end_0 = const()[name = tensor("op_15157_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_15157_end_mask_0 = const()[name = tensor("op_15157_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15157_cast_fp16 = slice_by_index(begin = var_15157_begin_0, end = var_15157_end_0, end_mask = var_15157_end_mask_0, x = q_71_cast_fp16)[name = tensor("op_15157_cast_fp16")]; + tensor var_15161_begin_0 = const()[name = tensor("op_15161_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_15161_end_0 = const()[name = tensor("op_15161_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_15161_end_mask_0 = const()[name = tensor("op_15161_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15161_cast_fp16 = slice_by_index(begin = var_15161_begin_0, end = var_15161_end_0, end_mask = var_15161_end_mask_0, x = q_71_cast_fp16)[name = tensor("op_15161_cast_fp16")]; + tensor var_15165_begin_0 = const()[name = tensor("op_15165_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_15165_end_0 = const()[name = tensor("op_15165_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_15165_end_mask_0 = const()[name = tensor("op_15165_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15165_cast_fp16 = slice_by_index(begin = var_15165_begin_0, end = var_15165_end_0, end_mask = var_15165_end_mask_0, x = q_71_cast_fp16)[name = tensor("op_15165_cast_fp16")]; + tensor var_15169_begin_0 = const()[name = tensor("op_15169_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_15169_end_0 = const()[name = tensor("op_15169_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_15169_end_mask_0 = const()[name = tensor("op_15169_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15169_cast_fp16 = slice_by_index(begin = var_15169_begin_0, end = var_15169_end_0, end_mask = var_15169_end_mask_0, x = q_71_cast_fp16)[name = tensor("op_15169_cast_fp16")]; + tensor var_15173_begin_0 = const()[name = tensor("op_15173_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_15173_end_0 = const()[name = tensor("op_15173_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_15173_end_mask_0 = const()[name = tensor("op_15173_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15173_cast_fp16 = slice_by_index(begin = var_15173_begin_0, end = var_15173_end_0, end_mask = var_15173_end_mask_0, x = q_71_cast_fp16)[name = tensor("op_15173_cast_fp16")]; + tensor var_15177_begin_0 = const()[name = tensor("op_15177_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_15177_end_0 = const()[name = tensor("op_15177_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_15177_end_mask_0 = const()[name = tensor("op_15177_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15177_cast_fp16 = slice_by_index(begin = var_15177_begin_0, end = var_15177_end_0, end_mask = var_15177_end_mask_0, x = q_71_cast_fp16)[name = tensor("op_15177_cast_fp16")]; + tensor var_15181_begin_0 = const()[name = tensor("op_15181_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_15181_end_0 = const()[name = tensor("op_15181_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_15181_end_mask_0 = const()[name = tensor("op_15181_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15181_cast_fp16 = slice_by_index(begin = var_15181_begin_0, end = var_15181_end_0, end_mask = var_15181_end_mask_0, x = q_71_cast_fp16)[name = tensor("op_15181_cast_fp16")]; + tensor var_15185_begin_0 = const()[name = tensor("op_15185_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_15185_end_0 = const()[name = tensor("op_15185_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_15185_end_mask_0 = const()[name = tensor("op_15185_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15185_cast_fp16 = slice_by_index(begin = var_15185_begin_0, end = var_15185_end_0, end_mask = var_15185_end_mask_0, x = q_71_cast_fp16)[name = tensor("op_15185_cast_fp16")]; + tensor var_15189_begin_0 = const()[name = tensor("op_15189_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_15189_end_0 = const()[name = tensor("op_15189_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_15189_end_mask_0 = const()[name = tensor("op_15189_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15189_cast_fp16 = slice_by_index(begin = var_15189_begin_0, end = var_15189_end_0, end_mask = var_15189_end_mask_0, x = q_71_cast_fp16)[name = tensor("op_15189_cast_fp16")]; + tensor var_15193_begin_0 = const()[name = tensor("op_15193_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_15193_end_0 = const()[name = tensor("op_15193_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_15193_end_mask_0 = const()[name = tensor("op_15193_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15193_cast_fp16 = slice_by_index(begin = var_15193_begin_0, end = var_15193_end_0, end_mask = var_15193_end_mask_0, x = q_71_cast_fp16)[name = tensor("op_15193_cast_fp16")]; + tensor var_15197_begin_0 = const()[name = tensor("op_15197_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_15197_end_0 = const()[name = tensor("op_15197_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_15197_end_mask_0 = const()[name = tensor("op_15197_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15197_cast_fp16 = slice_by_index(begin = var_15197_begin_0, end = var_15197_end_0, end_mask = var_15197_end_mask_0, x = q_71_cast_fp16)[name = tensor("op_15197_cast_fp16")]; + tensor var_15201_begin_0 = const()[name = tensor("op_15201_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_15201_end_0 = const()[name = tensor("op_15201_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_15201_end_mask_0 = const()[name = tensor("op_15201_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15201_cast_fp16 = slice_by_index(begin = var_15201_begin_0, end = var_15201_end_0, end_mask = var_15201_end_mask_0, x = q_71_cast_fp16)[name = tensor("op_15201_cast_fp16")]; + tensor var_15205_begin_0 = const()[name = tensor("op_15205_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_15205_end_0 = const()[name = tensor("op_15205_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_15205_end_mask_0 = const()[name = tensor("op_15205_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15205_cast_fp16 = slice_by_index(begin = var_15205_begin_0, end = var_15205_end_0, end_mask = var_15205_end_mask_0, x = q_71_cast_fp16)[name = tensor("op_15205_cast_fp16")]; + tensor var_15209_begin_0 = const()[name = tensor("op_15209_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_15209_end_0 = const()[name = tensor("op_15209_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_15209_end_mask_0 = const()[name = tensor("op_15209_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15209_cast_fp16 = slice_by_index(begin = var_15209_begin_0, end = var_15209_end_0, end_mask = var_15209_end_mask_0, x = q_71_cast_fp16)[name = tensor("op_15209_cast_fp16")]; + tensor k_143_perm_0 = const()[name = tensor("k_143_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_15216_begin_0 = const()[name = tensor("op_15216_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15216_end_0 = const()[name = tensor("op_15216_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_15216_end_mask_0 = const()[name = tensor("op_15216_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_143_cast_fp16 = transpose(perm = k_143_perm_0, x = k_141_cast_fp16)[name = tensor("transpose_104")]; + tensor var_15216_cast_fp16 = slice_by_index(begin = var_15216_begin_0, end = var_15216_end_0, end_mask = var_15216_end_mask_0, x = k_143_cast_fp16)[name = tensor("op_15216_cast_fp16")]; + tensor var_15220_begin_0 = const()[name = tensor("op_15220_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_15220_end_0 = const()[name = tensor("op_15220_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_15220_end_mask_0 = const()[name = tensor("op_15220_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15220_cast_fp16 = slice_by_index(begin = var_15220_begin_0, end = var_15220_end_0, end_mask = var_15220_end_mask_0, x = k_143_cast_fp16)[name = tensor("op_15220_cast_fp16")]; + tensor var_15224_begin_0 = const()[name = tensor("op_15224_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_15224_end_0 = const()[name = tensor("op_15224_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_15224_end_mask_0 = const()[name = tensor("op_15224_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15224_cast_fp16 = slice_by_index(begin = var_15224_begin_0, end = var_15224_end_0, end_mask = var_15224_end_mask_0, x = k_143_cast_fp16)[name = tensor("op_15224_cast_fp16")]; + tensor var_15228_begin_0 = const()[name = tensor("op_15228_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_15228_end_0 = const()[name = tensor("op_15228_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_15228_end_mask_0 = const()[name = tensor("op_15228_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15228_cast_fp16 = slice_by_index(begin = var_15228_begin_0, end = var_15228_end_0, end_mask = var_15228_end_mask_0, x = k_143_cast_fp16)[name = tensor("op_15228_cast_fp16")]; + tensor var_15232_begin_0 = const()[name = tensor("op_15232_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_15232_end_0 = const()[name = tensor("op_15232_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_15232_end_mask_0 = const()[name = tensor("op_15232_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15232_cast_fp16 = slice_by_index(begin = var_15232_begin_0, end = var_15232_end_0, end_mask = var_15232_end_mask_0, x = k_143_cast_fp16)[name = tensor("op_15232_cast_fp16")]; + tensor var_15236_begin_0 = const()[name = tensor("op_15236_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_15236_end_0 = const()[name = tensor("op_15236_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_15236_end_mask_0 = const()[name = tensor("op_15236_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15236_cast_fp16 = slice_by_index(begin = var_15236_begin_0, end = var_15236_end_0, end_mask = var_15236_end_mask_0, x = k_143_cast_fp16)[name = tensor("op_15236_cast_fp16")]; + tensor var_15240_begin_0 = const()[name = tensor("op_15240_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_15240_end_0 = const()[name = tensor("op_15240_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_15240_end_mask_0 = const()[name = tensor("op_15240_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15240_cast_fp16 = slice_by_index(begin = var_15240_begin_0, end = var_15240_end_0, end_mask = var_15240_end_mask_0, x = k_143_cast_fp16)[name = tensor("op_15240_cast_fp16")]; + tensor var_15244_begin_0 = const()[name = tensor("op_15244_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_15244_end_0 = const()[name = tensor("op_15244_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_15244_end_mask_0 = const()[name = tensor("op_15244_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15244_cast_fp16 = slice_by_index(begin = var_15244_begin_0, end = var_15244_end_0, end_mask = var_15244_end_mask_0, x = k_143_cast_fp16)[name = tensor("op_15244_cast_fp16")]; + tensor var_15248_begin_0 = const()[name = tensor("op_15248_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_15248_end_0 = const()[name = tensor("op_15248_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_15248_end_mask_0 = const()[name = tensor("op_15248_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15248_cast_fp16 = slice_by_index(begin = var_15248_begin_0, end = var_15248_end_0, end_mask = var_15248_end_mask_0, x = k_143_cast_fp16)[name = tensor("op_15248_cast_fp16")]; + tensor var_15252_begin_0 = const()[name = tensor("op_15252_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_15252_end_0 = const()[name = tensor("op_15252_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_15252_end_mask_0 = const()[name = tensor("op_15252_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15252_cast_fp16 = slice_by_index(begin = var_15252_begin_0, end = var_15252_end_0, end_mask = var_15252_end_mask_0, x = k_143_cast_fp16)[name = tensor("op_15252_cast_fp16")]; + tensor var_15256_begin_0 = const()[name = tensor("op_15256_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_15256_end_0 = const()[name = tensor("op_15256_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_15256_end_mask_0 = const()[name = tensor("op_15256_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15256_cast_fp16 = slice_by_index(begin = var_15256_begin_0, end = var_15256_end_0, end_mask = var_15256_end_mask_0, x = k_143_cast_fp16)[name = tensor("op_15256_cast_fp16")]; + tensor var_15260_begin_0 = const()[name = tensor("op_15260_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_15260_end_0 = const()[name = tensor("op_15260_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_15260_end_mask_0 = const()[name = tensor("op_15260_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15260_cast_fp16 = slice_by_index(begin = var_15260_begin_0, end = var_15260_end_0, end_mask = var_15260_end_mask_0, x = k_143_cast_fp16)[name = tensor("op_15260_cast_fp16")]; + tensor var_15264_begin_0 = const()[name = tensor("op_15264_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_15264_end_0 = const()[name = tensor("op_15264_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_15264_end_mask_0 = const()[name = tensor("op_15264_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15264_cast_fp16 = slice_by_index(begin = var_15264_begin_0, end = var_15264_end_0, end_mask = var_15264_end_mask_0, x = k_143_cast_fp16)[name = tensor("op_15264_cast_fp16")]; + tensor var_15268_begin_0 = const()[name = tensor("op_15268_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_15268_end_0 = const()[name = tensor("op_15268_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_15268_end_mask_0 = const()[name = tensor("op_15268_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15268_cast_fp16 = slice_by_index(begin = var_15268_begin_0, end = var_15268_end_0, end_mask = var_15268_end_mask_0, x = k_143_cast_fp16)[name = tensor("op_15268_cast_fp16")]; + tensor var_15272_begin_0 = const()[name = tensor("op_15272_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_15272_end_0 = const()[name = tensor("op_15272_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_15272_end_mask_0 = const()[name = tensor("op_15272_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15272_cast_fp16 = slice_by_index(begin = var_15272_begin_0, end = var_15272_end_0, end_mask = var_15272_end_mask_0, x = k_143_cast_fp16)[name = tensor("op_15272_cast_fp16")]; + tensor var_15276_begin_0 = const()[name = tensor("op_15276_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_15276_end_0 = const()[name = tensor("op_15276_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_15276_end_mask_0 = const()[name = tensor("op_15276_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15276_cast_fp16 = slice_by_index(begin = var_15276_begin_0, end = var_15276_end_0, end_mask = var_15276_end_mask_0, x = k_143_cast_fp16)[name = tensor("op_15276_cast_fp16")]; + tensor var_15280_begin_0 = const()[name = tensor("op_15280_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_15280_end_0 = const()[name = tensor("op_15280_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_15280_end_mask_0 = const()[name = tensor("op_15280_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15280_cast_fp16 = slice_by_index(begin = var_15280_begin_0, end = var_15280_end_0, end_mask = var_15280_end_mask_0, x = k_143_cast_fp16)[name = tensor("op_15280_cast_fp16")]; + tensor var_15284_begin_0 = const()[name = tensor("op_15284_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_15284_end_0 = const()[name = tensor("op_15284_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_15284_end_mask_0 = const()[name = tensor("op_15284_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15284_cast_fp16 = slice_by_index(begin = var_15284_begin_0, end = var_15284_end_0, end_mask = var_15284_end_mask_0, x = k_143_cast_fp16)[name = tensor("op_15284_cast_fp16")]; + tensor var_15288_begin_0 = const()[name = tensor("op_15288_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_15288_end_0 = const()[name = tensor("op_15288_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_15288_end_mask_0 = const()[name = tensor("op_15288_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15288_cast_fp16 = slice_by_index(begin = var_15288_begin_0, end = var_15288_end_0, end_mask = var_15288_end_mask_0, x = k_143_cast_fp16)[name = tensor("op_15288_cast_fp16")]; + tensor var_15292_begin_0 = const()[name = tensor("op_15292_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_15292_end_0 = const()[name = tensor("op_15292_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_15292_end_mask_0 = const()[name = tensor("op_15292_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15292_cast_fp16 = slice_by_index(begin = var_15292_begin_0, end = var_15292_end_0, end_mask = var_15292_end_mask_0, x = k_143_cast_fp16)[name = tensor("op_15292_cast_fp16")]; + tensor var_15294_begin_0 = const()[name = tensor("op_15294_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15294_end_0 = const()[name = tensor("op_15294_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_15294_end_mask_0 = const()[name = tensor("op_15294_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15294_cast_fp16 = slice_by_index(begin = var_15294_begin_0, end = var_15294_end_0, end_mask = var_15294_end_mask_0, x = v_71_cast_fp16)[name = tensor("op_15294_cast_fp16")]; + tensor var_15298_begin_0 = const()[name = tensor("op_15298_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_15298_end_0 = const()[name = tensor("op_15298_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_15298_end_mask_0 = const()[name = tensor("op_15298_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15298_cast_fp16 = slice_by_index(begin = var_15298_begin_0, end = var_15298_end_0, end_mask = var_15298_end_mask_0, x = v_71_cast_fp16)[name = tensor("op_15298_cast_fp16")]; + tensor var_15302_begin_0 = const()[name = tensor("op_15302_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_15302_end_0 = const()[name = tensor("op_15302_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_15302_end_mask_0 = const()[name = tensor("op_15302_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15302_cast_fp16 = slice_by_index(begin = var_15302_begin_0, end = var_15302_end_0, end_mask = var_15302_end_mask_0, x = v_71_cast_fp16)[name = tensor("op_15302_cast_fp16")]; + tensor var_15306_begin_0 = const()[name = tensor("op_15306_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_15306_end_0 = const()[name = tensor("op_15306_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_15306_end_mask_0 = const()[name = tensor("op_15306_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15306_cast_fp16 = slice_by_index(begin = var_15306_begin_0, end = var_15306_end_0, end_mask = var_15306_end_mask_0, x = v_71_cast_fp16)[name = tensor("op_15306_cast_fp16")]; + tensor var_15310_begin_0 = const()[name = tensor("op_15310_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_15310_end_0 = const()[name = tensor("op_15310_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_15310_end_mask_0 = const()[name = tensor("op_15310_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15310_cast_fp16 = slice_by_index(begin = var_15310_begin_0, end = var_15310_end_0, end_mask = var_15310_end_mask_0, x = v_71_cast_fp16)[name = tensor("op_15310_cast_fp16")]; + tensor var_15314_begin_0 = const()[name = tensor("op_15314_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_15314_end_0 = const()[name = tensor("op_15314_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_15314_end_mask_0 = const()[name = tensor("op_15314_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15314_cast_fp16 = slice_by_index(begin = var_15314_begin_0, end = var_15314_end_0, end_mask = var_15314_end_mask_0, x = v_71_cast_fp16)[name = tensor("op_15314_cast_fp16")]; + tensor var_15318_begin_0 = const()[name = tensor("op_15318_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_15318_end_0 = const()[name = tensor("op_15318_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_15318_end_mask_0 = const()[name = tensor("op_15318_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15318_cast_fp16 = slice_by_index(begin = var_15318_begin_0, end = var_15318_end_0, end_mask = var_15318_end_mask_0, x = v_71_cast_fp16)[name = tensor("op_15318_cast_fp16")]; + tensor var_15322_begin_0 = const()[name = tensor("op_15322_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_15322_end_0 = const()[name = tensor("op_15322_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_15322_end_mask_0 = const()[name = tensor("op_15322_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15322_cast_fp16 = slice_by_index(begin = var_15322_begin_0, end = var_15322_end_0, end_mask = var_15322_end_mask_0, x = v_71_cast_fp16)[name = tensor("op_15322_cast_fp16")]; + tensor var_15326_begin_0 = const()[name = tensor("op_15326_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_15326_end_0 = const()[name = tensor("op_15326_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_15326_end_mask_0 = const()[name = tensor("op_15326_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15326_cast_fp16 = slice_by_index(begin = var_15326_begin_0, end = var_15326_end_0, end_mask = var_15326_end_mask_0, x = v_71_cast_fp16)[name = tensor("op_15326_cast_fp16")]; + tensor var_15330_begin_0 = const()[name = tensor("op_15330_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_15330_end_0 = const()[name = tensor("op_15330_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_15330_end_mask_0 = const()[name = tensor("op_15330_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15330_cast_fp16 = slice_by_index(begin = var_15330_begin_0, end = var_15330_end_0, end_mask = var_15330_end_mask_0, x = v_71_cast_fp16)[name = tensor("op_15330_cast_fp16")]; + tensor var_15334_begin_0 = const()[name = tensor("op_15334_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_15334_end_0 = const()[name = tensor("op_15334_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_15334_end_mask_0 = const()[name = tensor("op_15334_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15334_cast_fp16 = slice_by_index(begin = var_15334_begin_0, end = var_15334_end_0, end_mask = var_15334_end_mask_0, x = v_71_cast_fp16)[name = tensor("op_15334_cast_fp16")]; + tensor var_15338_begin_0 = const()[name = tensor("op_15338_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_15338_end_0 = const()[name = tensor("op_15338_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_15338_end_mask_0 = const()[name = tensor("op_15338_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15338_cast_fp16 = slice_by_index(begin = var_15338_begin_0, end = var_15338_end_0, end_mask = var_15338_end_mask_0, x = v_71_cast_fp16)[name = tensor("op_15338_cast_fp16")]; + tensor var_15342_begin_0 = const()[name = tensor("op_15342_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_15342_end_0 = const()[name = tensor("op_15342_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_15342_end_mask_0 = const()[name = tensor("op_15342_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15342_cast_fp16 = slice_by_index(begin = var_15342_begin_0, end = var_15342_end_0, end_mask = var_15342_end_mask_0, x = v_71_cast_fp16)[name = tensor("op_15342_cast_fp16")]; + tensor var_15346_begin_0 = const()[name = tensor("op_15346_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_15346_end_0 = const()[name = tensor("op_15346_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_15346_end_mask_0 = const()[name = tensor("op_15346_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15346_cast_fp16 = slice_by_index(begin = var_15346_begin_0, end = var_15346_end_0, end_mask = var_15346_end_mask_0, x = v_71_cast_fp16)[name = tensor("op_15346_cast_fp16")]; + tensor var_15350_begin_0 = const()[name = tensor("op_15350_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_15350_end_0 = const()[name = tensor("op_15350_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_15350_end_mask_0 = const()[name = tensor("op_15350_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15350_cast_fp16 = slice_by_index(begin = var_15350_begin_0, end = var_15350_end_0, end_mask = var_15350_end_mask_0, x = v_71_cast_fp16)[name = tensor("op_15350_cast_fp16")]; + tensor var_15354_begin_0 = const()[name = tensor("op_15354_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_15354_end_0 = const()[name = tensor("op_15354_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_15354_end_mask_0 = const()[name = tensor("op_15354_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15354_cast_fp16 = slice_by_index(begin = var_15354_begin_0, end = var_15354_end_0, end_mask = var_15354_end_mask_0, x = v_71_cast_fp16)[name = tensor("op_15354_cast_fp16")]; + tensor var_15358_begin_0 = const()[name = tensor("op_15358_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_15358_end_0 = const()[name = tensor("op_15358_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_15358_end_mask_0 = const()[name = tensor("op_15358_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15358_cast_fp16 = slice_by_index(begin = var_15358_begin_0, end = var_15358_end_0, end_mask = var_15358_end_mask_0, x = v_71_cast_fp16)[name = tensor("op_15358_cast_fp16")]; + tensor var_15362_begin_0 = const()[name = tensor("op_15362_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_15362_end_0 = const()[name = tensor("op_15362_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_15362_end_mask_0 = const()[name = tensor("op_15362_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15362_cast_fp16 = slice_by_index(begin = var_15362_begin_0, end = var_15362_end_0, end_mask = var_15362_end_mask_0, x = v_71_cast_fp16)[name = tensor("op_15362_cast_fp16")]; + tensor var_15366_begin_0 = const()[name = tensor("op_15366_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_15366_end_0 = const()[name = tensor("op_15366_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_15366_end_mask_0 = const()[name = tensor("op_15366_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15366_cast_fp16 = slice_by_index(begin = var_15366_begin_0, end = var_15366_end_0, end_mask = var_15366_end_mask_0, x = v_71_cast_fp16)[name = tensor("op_15366_cast_fp16")]; + tensor var_15370_begin_0 = const()[name = tensor("op_15370_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_15370_end_0 = const()[name = tensor("op_15370_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_15370_end_mask_0 = const()[name = tensor("op_15370_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15370_cast_fp16 = slice_by_index(begin = var_15370_begin_0, end = var_15370_end_0, end_mask = var_15370_end_mask_0, x = v_71_cast_fp16)[name = tensor("op_15370_cast_fp16")]; + tensor var_15374_equation_0 = const()[name = tensor("op_15374_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15374_cast_fp16 = einsum(equation = var_15374_equation_0, values = (var_15216_cast_fp16, var_15133_cast_fp16))[name = tensor("op_15374_cast_fp16")]; + tensor var_15375_to_fp16 = const()[name = tensor("op_15375_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1241_cast_fp16 = mul(x = var_15374_cast_fp16, y = var_15375_to_fp16)[name = tensor("aw_1241_cast_fp16")]; + tensor var_15378_equation_0 = const()[name = tensor("op_15378_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15378_cast_fp16 = einsum(equation = var_15378_equation_0, values = (var_15220_cast_fp16, var_15137_cast_fp16))[name = tensor("op_15378_cast_fp16")]; + tensor var_15379_to_fp16 = const()[name = tensor("op_15379_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1243_cast_fp16 = mul(x = var_15378_cast_fp16, y = var_15379_to_fp16)[name = tensor("aw_1243_cast_fp16")]; + tensor var_15382_equation_0 = const()[name = tensor("op_15382_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15382_cast_fp16 = einsum(equation = var_15382_equation_0, values = (var_15224_cast_fp16, var_15141_cast_fp16))[name = tensor("op_15382_cast_fp16")]; + tensor var_15383_to_fp16 = const()[name = tensor("op_15383_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1245_cast_fp16 = mul(x = var_15382_cast_fp16, y = var_15383_to_fp16)[name = tensor("aw_1245_cast_fp16")]; + tensor var_15386_equation_0 = const()[name = tensor("op_15386_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15386_cast_fp16 = einsum(equation = var_15386_equation_0, values = (var_15228_cast_fp16, var_15145_cast_fp16))[name = tensor("op_15386_cast_fp16")]; + tensor var_15387_to_fp16 = const()[name = tensor("op_15387_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1247_cast_fp16 = mul(x = var_15386_cast_fp16, y = var_15387_to_fp16)[name = tensor("aw_1247_cast_fp16")]; + tensor var_15390_equation_0 = const()[name = tensor("op_15390_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15390_cast_fp16 = einsum(equation = var_15390_equation_0, values = (var_15232_cast_fp16, var_15149_cast_fp16))[name = tensor("op_15390_cast_fp16")]; + tensor var_15391_to_fp16 = const()[name = tensor("op_15391_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1249_cast_fp16 = mul(x = var_15390_cast_fp16, y = var_15391_to_fp16)[name = tensor("aw_1249_cast_fp16")]; + tensor var_15394_equation_0 = const()[name = tensor("op_15394_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15394_cast_fp16 = einsum(equation = var_15394_equation_0, values = (var_15236_cast_fp16, var_15153_cast_fp16))[name = tensor("op_15394_cast_fp16")]; + tensor var_15395_to_fp16 = const()[name = tensor("op_15395_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1251_cast_fp16 = mul(x = var_15394_cast_fp16, y = var_15395_to_fp16)[name = tensor("aw_1251_cast_fp16")]; + tensor var_15398_equation_0 = const()[name = tensor("op_15398_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15398_cast_fp16 = einsum(equation = var_15398_equation_0, values = (var_15240_cast_fp16, var_15157_cast_fp16))[name = tensor("op_15398_cast_fp16")]; + tensor var_15399_to_fp16 = const()[name = tensor("op_15399_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1253_cast_fp16 = mul(x = var_15398_cast_fp16, y = var_15399_to_fp16)[name = tensor("aw_1253_cast_fp16")]; + tensor var_15402_equation_0 = const()[name = tensor("op_15402_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15402_cast_fp16 = einsum(equation = var_15402_equation_0, values = (var_15244_cast_fp16, var_15161_cast_fp16))[name = tensor("op_15402_cast_fp16")]; + tensor var_15403_to_fp16 = const()[name = tensor("op_15403_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1255_cast_fp16 = mul(x = var_15402_cast_fp16, y = var_15403_to_fp16)[name = tensor("aw_1255_cast_fp16")]; + tensor var_15406_equation_0 = const()[name = tensor("op_15406_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15406_cast_fp16 = einsum(equation = var_15406_equation_0, values = (var_15248_cast_fp16, var_15165_cast_fp16))[name = tensor("op_15406_cast_fp16")]; + tensor var_15407_to_fp16 = const()[name = tensor("op_15407_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1257_cast_fp16 = mul(x = var_15406_cast_fp16, y = var_15407_to_fp16)[name = tensor("aw_1257_cast_fp16")]; + tensor var_15410_equation_0 = const()[name = tensor("op_15410_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15410_cast_fp16 = einsum(equation = var_15410_equation_0, values = (var_15252_cast_fp16, var_15169_cast_fp16))[name = tensor("op_15410_cast_fp16")]; + tensor var_15411_to_fp16 = const()[name = tensor("op_15411_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1259_cast_fp16 = mul(x = var_15410_cast_fp16, y = var_15411_to_fp16)[name = tensor("aw_1259_cast_fp16")]; + tensor var_15414_equation_0 = const()[name = tensor("op_15414_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15414_cast_fp16 = einsum(equation = var_15414_equation_0, values = (var_15256_cast_fp16, var_15173_cast_fp16))[name = tensor("op_15414_cast_fp16")]; + tensor var_15415_to_fp16 = const()[name = tensor("op_15415_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1261_cast_fp16 = mul(x = var_15414_cast_fp16, y = var_15415_to_fp16)[name = tensor("aw_1261_cast_fp16")]; + tensor var_15418_equation_0 = const()[name = tensor("op_15418_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15418_cast_fp16 = einsum(equation = var_15418_equation_0, values = (var_15260_cast_fp16, var_15177_cast_fp16))[name = tensor("op_15418_cast_fp16")]; + tensor var_15419_to_fp16 = const()[name = tensor("op_15419_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1263_cast_fp16 = mul(x = var_15418_cast_fp16, y = var_15419_to_fp16)[name = tensor("aw_1263_cast_fp16")]; + tensor var_15422_equation_0 = const()[name = tensor("op_15422_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15422_cast_fp16 = einsum(equation = var_15422_equation_0, values = (var_15264_cast_fp16, var_15181_cast_fp16))[name = tensor("op_15422_cast_fp16")]; + tensor var_15423_to_fp16 = const()[name = tensor("op_15423_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1265_cast_fp16 = mul(x = var_15422_cast_fp16, y = var_15423_to_fp16)[name = tensor("aw_1265_cast_fp16")]; + tensor var_15426_equation_0 = const()[name = tensor("op_15426_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15426_cast_fp16 = einsum(equation = var_15426_equation_0, values = (var_15268_cast_fp16, var_15185_cast_fp16))[name = tensor("op_15426_cast_fp16")]; + tensor var_15427_to_fp16 = const()[name = tensor("op_15427_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1267_cast_fp16 = mul(x = var_15426_cast_fp16, y = var_15427_to_fp16)[name = tensor("aw_1267_cast_fp16")]; + tensor var_15430_equation_0 = const()[name = tensor("op_15430_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15430_cast_fp16 = einsum(equation = var_15430_equation_0, values = (var_15272_cast_fp16, var_15189_cast_fp16))[name = tensor("op_15430_cast_fp16")]; + tensor var_15431_to_fp16 = const()[name = tensor("op_15431_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1269_cast_fp16 = mul(x = var_15430_cast_fp16, y = var_15431_to_fp16)[name = tensor("aw_1269_cast_fp16")]; + tensor var_15434_equation_0 = const()[name = tensor("op_15434_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15434_cast_fp16 = einsum(equation = var_15434_equation_0, values = (var_15276_cast_fp16, var_15193_cast_fp16))[name = tensor("op_15434_cast_fp16")]; + tensor var_15435_to_fp16 = const()[name = tensor("op_15435_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1271_cast_fp16 = mul(x = var_15434_cast_fp16, y = var_15435_to_fp16)[name = tensor("aw_1271_cast_fp16")]; + tensor var_15438_equation_0 = const()[name = tensor("op_15438_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15438_cast_fp16 = einsum(equation = var_15438_equation_0, values = (var_15280_cast_fp16, var_15197_cast_fp16))[name = tensor("op_15438_cast_fp16")]; + tensor var_15439_to_fp16 = const()[name = tensor("op_15439_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1273_cast_fp16 = mul(x = var_15438_cast_fp16, y = var_15439_to_fp16)[name = tensor("aw_1273_cast_fp16")]; + tensor var_15442_equation_0 = const()[name = tensor("op_15442_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15442_cast_fp16 = einsum(equation = var_15442_equation_0, values = (var_15284_cast_fp16, var_15201_cast_fp16))[name = tensor("op_15442_cast_fp16")]; + tensor var_15443_to_fp16 = const()[name = tensor("op_15443_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1275_cast_fp16 = mul(x = var_15442_cast_fp16, y = var_15443_to_fp16)[name = tensor("aw_1275_cast_fp16")]; + tensor var_15446_equation_0 = const()[name = tensor("op_15446_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15446_cast_fp16 = einsum(equation = var_15446_equation_0, values = (var_15288_cast_fp16, var_15205_cast_fp16))[name = tensor("op_15446_cast_fp16")]; + tensor var_15447_to_fp16 = const()[name = tensor("op_15447_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1277_cast_fp16 = mul(x = var_15446_cast_fp16, y = var_15447_to_fp16)[name = tensor("aw_1277_cast_fp16")]; + tensor var_15450_equation_0 = const()[name = tensor("op_15450_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15450_cast_fp16 = einsum(equation = var_15450_equation_0, values = (var_15292_cast_fp16, var_15209_cast_fp16))[name = tensor("op_15450_cast_fp16")]; + tensor var_15451_to_fp16 = const()[name = tensor("op_15451_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1279_cast_fp16 = mul(x = var_15450_cast_fp16, y = var_15451_to_fp16)[name = tensor("aw_1279_cast_fp16")]; + tensor var_15453_cast_fp16 = softmax(axis = var_2624, x = aw_1241_cast_fp16)[name = tensor("op_15453_cast_fp16")]; + tensor var_15454_cast_fp16 = softmax(axis = var_2624, x = aw_1243_cast_fp16)[name = tensor("op_15454_cast_fp16")]; + tensor var_15455_cast_fp16 = softmax(axis = var_2624, x = aw_1245_cast_fp16)[name = tensor("op_15455_cast_fp16")]; + tensor var_15456_cast_fp16 = softmax(axis = var_2624, x = aw_1247_cast_fp16)[name = tensor("op_15456_cast_fp16")]; + tensor var_15457_cast_fp16 = softmax(axis = var_2624, x = aw_1249_cast_fp16)[name = tensor("op_15457_cast_fp16")]; + tensor var_15458_cast_fp16 = softmax(axis = var_2624, x = aw_1251_cast_fp16)[name = tensor("op_15458_cast_fp16")]; + tensor var_15459_cast_fp16 = softmax(axis = var_2624, x = aw_1253_cast_fp16)[name = tensor("op_15459_cast_fp16")]; + tensor var_15460_cast_fp16 = softmax(axis = var_2624, x = aw_1255_cast_fp16)[name = tensor("op_15460_cast_fp16")]; + tensor var_15461_cast_fp16 = softmax(axis = var_2624, x = aw_1257_cast_fp16)[name = tensor("op_15461_cast_fp16")]; + tensor var_15462_cast_fp16 = softmax(axis = var_2624, x = aw_1259_cast_fp16)[name = tensor("op_15462_cast_fp16")]; + tensor var_15463_cast_fp16 = softmax(axis = var_2624, x = aw_1261_cast_fp16)[name = tensor("op_15463_cast_fp16")]; + tensor var_15464_cast_fp16 = softmax(axis = var_2624, x = aw_1263_cast_fp16)[name = tensor("op_15464_cast_fp16")]; + tensor var_15465_cast_fp16 = softmax(axis = var_2624, x = aw_1265_cast_fp16)[name = tensor("op_15465_cast_fp16")]; + tensor var_15466_cast_fp16 = softmax(axis = var_2624, x = aw_1267_cast_fp16)[name = tensor("op_15466_cast_fp16")]; + tensor var_15467_cast_fp16 = softmax(axis = var_2624, x = aw_1269_cast_fp16)[name = tensor("op_15467_cast_fp16")]; + tensor var_15468_cast_fp16 = softmax(axis = var_2624, x = aw_1271_cast_fp16)[name = tensor("op_15468_cast_fp16")]; + tensor var_15469_cast_fp16 = softmax(axis = var_2624, x = aw_1273_cast_fp16)[name = tensor("op_15469_cast_fp16")]; + tensor var_15470_cast_fp16 = softmax(axis = var_2624, x = aw_1275_cast_fp16)[name = tensor("op_15470_cast_fp16")]; + tensor var_15471_cast_fp16 = softmax(axis = var_2624, x = aw_1277_cast_fp16)[name = tensor("op_15471_cast_fp16")]; + tensor var_15472_cast_fp16 = softmax(axis = var_2624, x = aw_1279_cast_fp16)[name = tensor("op_15472_cast_fp16")]; + tensor var_15474_equation_0 = const()[name = tensor("op_15474_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15474_cast_fp16 = einsum(equation = var_15474_equation_0, values = (var_15294_cast_fp16, var_15453_cast_fp16))[name = tensor("op_15474_cast_fp16")]; + tensor var_15476_equation_0 = const()[name = tensor("op_15476_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15476_cast_fp16 = einsum(equation = var_15476_equation_0, values = (var_15298_cast_fp16, var_15454_cast_fp16))[name = tensor("op_15476_cast_fp16")]; + tensor var_15478_equation_0 = const()[name = tensor("op_15478_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15478_cast_fp16 = einsum(equation = var_15478_equation_0, values = (var_15302_cast_fp16, var_15455_cast_fp16))[name = tensor("op_15478_cast_fp16")]; + tensor var_15480_equation_0 = const()[name = tensor("op_15480_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15480_cast_fp16 = einsum(equation = var_15480_equation_0, values = (var_15306_cast_fp16, var_15456_cast_fp16))[name = tensor("op_15480_cast_fp16")]; + tensor var_15482_equation_0 = const()[name = tensor("op_15482_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15482_cast_fp16 = einsum(equation = var_15482_equation_0, values = (var_15310_cast_fp16, var_15457_cast_fp16))[name = tensor("op_15482_cast_fp16")]; + tensor var_15484_equation_0 = const()[name = tensor("op_15484_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15484_cast_fp16 = einsum(equation = var_15484_equation_0, values = (var_15314_cast_fp16, var_15458_cast_fp16))[name = tensor("op_15484_cast_fp16")]; + tensor var_15486_equation_0 = const()[name = tensor("op_15486_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15486_cast_fp16 = einsum(equation = var_15486_equation_0, values = (var_15318_cast_fp16, var_15459_cast_fp16))[name = tensor("op_15486_cast_fp16")]; + tensor var_15488_equation_0 = const()[name = tensor("op_15488_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15488_cast_fp16 = einsum(equation = var_15488_equation_0, values = (var_15322_cast_fp16, var_15460_cast_fp16))[name = tensor("op_15488_cast_fp16")]; + tensor var_15490_equation_0 = const()[name = tensor("op_15490_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15490_cast_fp16 = einsum(equation = var_15490_equation_0, values = (var_15326_cast_fp16, var_15461_cast_fp16))[name = tensor("op_15490_cast_fp16")]; + tensor var_15492_equation_0 = const()[name = tensor("op_15492_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15492_cast_fp16 = einsum(equation = var_15492_equation_0, values = (var_15330_cast_fp16, var_15462_cast_fp16))[name = tensor("op_15492_cast_fp16")]; + tensor var_15494_equation_0 = const()[name = tensor("op_15494_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15494_cast_fp16 = einsum(equation = var_15494_equation_0, values = (var_15334_cast_fp16, var_15463_cast_fp16))[name = tensor("op_15494_cast_fp16")]; + tensor var_15496_equation_0 = const()[name = tensor("op_15496_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15496_cast_fp16 = einsum(equation = var_15496_equation_0, values = (var_15338_cast_fp16, var_15464_cast_fp16))[name = tensor("op_15496_cast_fp16")]; + tensor var_15498_equation_0 = const()[name = tensor("op_15498_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15498_cast_fp16 = einsum(equation = var_15498_equation_0, values = (var_15342_cast_fp16, var_15465_cast_fp16))[name = tensor("op_15498_cast_fp16")]; + tensor var_15500_equation_0 = const()[name = tensor("op_15500_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15500_cast_fp16 = einsum(equation = var_15500_equation_0, values = (var_15346_cast_fp16, var_15466_cast_fp16))[name = tensor("op_15500_cast_fp16")]; + tensor var_15502_equation_0 = const()[name = tensor("op_15502_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15502_cast_fp16 = einsum(equation = var_15502_equation_0, values = (var_15350_cast_fp16, var_15467_cast_fp16))[name = tensor("op_15502_cast_fp16")]; + tensor var_15504_equation_0 = const()[name = tensor("op_15504_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15504_cast_fp16 = einsum(equation = var_15504_equation_0, values = (var_15354_cast_fp16, var_15468_cast_fp16))[name = tensor("op_15504_cast_fp16")]; + tensor var_15506_equation_0 = const()[name = tensor("op_15506_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15506_cast_fp16 = einsum(equation = var_15506_equation_0, values = (var_15358_cast_fp16, var_15469_cast_fp16))[name = tensor("op_15506_cast_fp16")]; + tensor var_15508_equation_0 = const()[name = tensor("op_15508_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15508_cast_fp16 = einsum(equation = var_15508_equation_0, values = (var_15362_cast_fp16, var_15470_cast_fp16))[name = tensor("op_15508_cast_fp16")]; + tensor var_15510_equation_0 = const()[name = tensor("op_15510_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15510_cast_fp16 = einsum(equation = var_15510_equation_0, values = (var_15366_cast_fp16, var_15471_cast_fp16))[name = tensor("op_15510_cast_fp16")]; + tensor var_15512_equation_0 = const()[name = tensor("op_15512_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15512_cast_fp16 = einsum(equation = var_15512_equation_0, values = (var_15370_cast_fp16, var_15472_cast_fp16))[name = tensor("op_15512_cast_fp16")]; + tensor input_255_interleave_0 = const()[name = tensor("input_255_interleave_0"), val = tensor(false)]; + tensor input_255_cast_fp16 = concat(axis = var_2624, interleave = input_255_interleave_0, values = (var_15474_cast_fp16, var_15476_cast_fp16, var_15478_cast_fp16, var_15480_cast_fp16, var_15482_cast_fp16, var_15484_cast_fp16, var_15486_cast_fp16, var_15488_cast_fp16, var_15490_cast_fp16, var_15492_cast_fp16, var_15494_cast_fp16, var_15496_cast_fp16, var_15498_cast_fp16, var_15500_cast_fp16, var_15502_cast_fp16, var_15504_cast_fp16, var_15506_cast_fp16, var_15508_cast_fp16, var_15510_cast_fp16, var_15512_cast_fp16))[name = tensor("input_255_cast_fp16")]; + tensor var_15518 = const()[name = tensor("op_15518"), val = tensor([1, 1])]; + tensor var_15520 = const()[name = tensor("op_15520"), val = tensor([1, 1])]; + tensor var_15522_pad_type_0 = const()[name = tensor("op_15522_pad_type_0"), val = tensor("custom")]; + tensor var_15522_pad_0 = const()[name = tensor("op_15522_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_3_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_3_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1197559680)))]; + tensor down_blocks_2_attentions_1_transformer_blocks_3_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_3_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1200836544)))]; + tensor var_15522_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_3_attn2_to_out_0_bias_to_fp16, dilations = var_15520, groups = var_2624, pad = var_15522_pad_0, pad_type = var_15522_pad_type_0, strides = var_15518, weight = down_blocks_2_attentions_1_transformer_blocks_3_attn2_to_out_0_weight_to_fp16, x = input_255_cast_fp16)[name = tensor("op_15522_cast_fp16")]; + tensor inputs_107_cast_fp16 = add(x = var_15522_cast_fp16, y = inputs_105_cast_fp16)[name = tensor("inputs_107_cast_fp16")]; + tensor input_257_axes_0 = const()[name = tensor("input_257_axes_0"), val = tensor([1])]; + tensor input_257_gamma_0_to_fp16 = const()[name = tensor("input_257_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1200839168)))]; + tensor input_257_beta_0_to_fp16 = const()[name = tensor("input_257_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1200841792)))]; + tensor var_15532_to_fp16 = const()[name = tensor("op_15532_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_257_cast_fp16 = layer_norm(axes = input_257_axes_0, beta = input_257_beta_0_to_fp16, epsilon = var_15532_to_fp16, gamma = input_257_gamma_0_to_fp16, x = inputs_107_cast_fp16)[name = tensor("input_257_cast_fp16")]; + tensor var_15548 = const()[name = tensor("op_15548"), val = tensor([1, 1])]; + tensor var_15550 = const()[name = tensor("op_15550"), val = tensor([1, 1])]; + tensor var_15552_pad_type_0 = const()[name = tensor("op_15552_pad_type_0"), val = tensor("custom")]; + tensor var_15552_pad_0 = const()[name = tensor("op_15552_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_3_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_3_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1200844416)))]; + tensor down_blocks_2_attentions_1_transformer_blocks_3_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_3_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1227058880)))]; + tensor var_15552_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_3_ff_net_0_proj_bias_to_fp16, dilations = var_15550, groups = var_2624, pad = var_15552_pad_0, pad_type = var_15552_pad_type_0, strides = var_15548, weight = down_blocks_2_attentions_1_transformer_blocks_3_ff_net_0_proj_weight_to_fp16, x = input_257_cast_fp16)[name = tensor("op_15552_cast_fp16")]; + tensor var_15553_split_sizes_0 = const()[name = tensor("op_15553_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_15553_axis_0 = const()[name = tensor("op_15553_axis_0"), val = tensor(1)]; + tensor var_15553_cast_fp16_0, tensor var_15553_cast_fp16_1 = split(axis = var_15553_axis_0, split_sizes = var_15553_split_sizes_0, x = var_15552_cast_fp16)[name = tensor("op_15553_cast_fp16")]; + tensor var_15555_mode_0 = const()[name = tensor("op_15555_mode_0"), val = tensor("EXACT")]; + tensor var_15555_cast_fp16 = gelu(mode = var_15555_mode_0, x = var_15553_cast_fp16_1)[name = tensor("op_15555_cast_fp16")]; + tensor input_259_cast_fp16 = mul(x = var_15553_cast_fp16_0, y = var_15555_cast_fp16)[name = tensor("input_259_cast_fp16")]; + tensor var_15559 = const()[name = tensor("op_15559"), val = tensor([1, 1])]; + tensor var_15561 = const()[name = tensor("op_15561"), val = tensor([1, 1])]; + tensor var_15563_pad_type_0 = const()[name = tensor("op_15563_pad_type_0"), val = tensor("custom")]; + tensor var_15563_pad_0 = const()[name = tensor("op_15563_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_3_ff_net_2_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_3_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1227079424)))]; + tensor down_blocks_2_attentions_1_transformer_blocks_3_ff_net_2_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_3_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1240186688)))]; + tensor var_15563_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_3_ff_net_2_bias_to_fp16, dilations = var_15561, groups = var_2624, pad = var_15563_pad_0, pad_type = var_15563_pad_type_0, strides = var_15559, weight = down_blocks_2_attentions_1_transformer_blocks_3_ff_net_2_weight_to_fp16, x = input_259_cast_fp16)[name = tensor("op_15563_cast_fp16")]; + tensor inputs_109_cast_fp16 = add(x = var_15563_cast_fp16, y = inputs_107_cast_fp16)[name = tensor("inputs_109_cast_fp16")]; + tensor hidden_states_161_axes_0 = const()[name = tensor("hidden_states_161_axes_0"), val = tensor([1])]; + tensor hidden_states_161_gamma_0_to_fp16 = const()[name = tensor("hidden_states_161_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1240189312)))]; + tensor hidden_states_161_beta_0_to_fp16 = const()[name = tensor("hidden_states_161_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1240191936)))]; + tensor var_15579_to_fp16 = const()[name = tensor("op_15579_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_161_cast_fp16 = layer_norm(axes = hidden_states_161_axes_0, beta = hidden_states_161_beta_0_to_fp16, epsilon = var_15579_to_fp16, gamma = hidden_states_161_gamma_0_to_fp16, x = inputs_109_cast_fp16)[name = tensor("hidden_states_161_cast_fp16")]; + tensor var_15594 = const()[name = tensor("op_15594"), val = tensor([1, 1])]; + tensor var_15596 = const()[name = tensor("op_15596"), val = tensor([1, 1])]; + tensor q_73_pad_type_0 = const()[name = tensor("q_73_pad_type_0"), val = tensor("custom")]; + tensor q_73_pad_0 = const()[name = tensor("q_73_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_4_attn1_to_q_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_4_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1240194560)))]; + tensor q_73_cast_fp16 = conv(dilations = var_15596, groups = var_2624, pad = q_73_pad_0, pad_type = q_73_pad_type_0, strides = var_15594, weight = down_blocks_2_attentions_1_transformer_blocks_4_attn1_to_q_weight_to_fp16, x = hidden_states_161_cast_fp16)[name = tensor("q_73_cast_fp16")]; + tensor var_15600 = const()[name = tensor("op_15600"), val = tensor([1, 1])]; + tensor var_15602 = const()[name = tensor("op_15602"), val = tensor([1, 1])]; + tensor k_145_pad_type_0 = const()[name = tensor("k_145_pad_type_0"), val = tensor("custom")]; + tensor k_145_pad_0 = const()[name = tensor("k_145_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_4_attn1_to_k_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_4_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1243471424)))]; + tensor k_145_cast_fp16 = conv(dilations = var_15602, groups = var_2624, pad = k_145_pad_0, pad_type = k_145_pad_type_0, strides = var_15600, weight = down_blocks_2_attentions_1_transformer_blocks_4_attn1_to_k_weight_to_fp16, x = hidden_states_161_cast_fp16)[name = tensor("k_145_cast_fp16")]; + tensor var_15606 = const()[name = tensor("op_15606"), val = tensor([1, 1])]; + tensor var_15608 = const()[name = tensor("op_15608"), val = tensor([1, 1])]; + tensor v_73_pad_type_0 = const()[name = tensor("v_73_pad_type_0"), val = tensor("custom")]; + tensor v_73_pad_0 = const()[name = tensor("v_73_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_4_attn1_to_v_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_4_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1246748288)))]; + tensor v_73_cast_fp16 = conv(dilations = var_15608, groups = var_2624, pad = v_73_pad_0, pad_type = v_73_pad_type_0, strides = var_15606, weight = down_blocks_2_attentions_1_transformer_blocks_4_attn1_to_v_weight_to_fp16, x = hidden_states_161_cast_fp16)[name = tensor("v_73_cast_fp16")]; + tensor var_15612_begin_0 = const()[name = tensor("op_15612_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15612_end_0 = const()[name = tensor("op_15612_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_15612_end_mask_0 = const()[name = tensor("op_15612_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15612_cast_fp16 = slice_by_index(begin = var_15612_begin_0, end = var_15612_end_0, end_mask = var_15612_end_mask_0, x = q_73_cast_fp16)[name = tensor("op_15612_cast_fp16")]; + tensor var_15616_begin_0 = const()[name = tensor("op_15616_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_15616_end_0 = const()[name = tensor("op_15616_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_15616_end_mask_0 = const()[name = tensor("op_15616_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15616_cast_fp16 = slice_by_index(begin = var_15616_begin_0, end = var_15616_end_0, end_mask = var_15616_end_mask_0, x = q_73_cast_fp16)[name = tensor("op_15616_cast_fp16")]; + tensor var_15620_begin_0 = const()[name = tensor("op_15620_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_15620_end_0 = const()[name = tensor("op_15620_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_15620_end_mask_0 = const()[name = tensor("op_15620_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15620_cast_fp16 = slice_by_index(begin = var_15620_begin_0, end = var_15620_end_0, end_mask = var_15620_end_mask_0, x = q_73_cast_fp16)[name = tensor("op_15620_cast_fp16")]; + tensor var_15624_begin_0 = const()[name = tensor("op_15624_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_15624_end_0 = const()[name = tensor("op_15624_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_15624_end_mask_0 = const()[name = tensor("op_15624_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15624_cast_fp16 = slice_by_index(begin = var_15624_begin_0, end = var_15624_end_0, end_mask = var_15624_end_mask_0, x = q_73_cast_fp16)[name = tensor("op_15624_cast_fp16")]; + tensor var_15628_begin_0 = const()[name = tensor("op_15628_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_15628_end_0 = const()[name = tensor("op_15628_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_15628_end_mask_0 = const()[name = tensor("op_15628_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15628_cast_fp16 = slice_by_index(begin = var_15628_begin_0, end = var_15628_end_0, end_mask = var_15628_end_mask_0, x = q_73_cast_fp16)[name = tensor("op_15628_cast_fp16")]; + tensor var_15632_begin_0 = const()[name = tensor("op_15632_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_15632_end_0 = const()[name = tensor("op_15632_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_15632_end_mask_0 = const()[name = tensor("op_15632_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15632_cast_fp16 = slice_by_index(begin = var_15632_begin_0, end = var_15632_end_0, end_mask = var_15632_end_mask_0, x = q_73_cast_fp16)[name = tensor("op_15632_cast_fp16")]; + tensor var_15636_begin_0 = const()[name = tensor("op_15636_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_15636_end_0 = const()[name = tensor("op_15636_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_15636_end_mask_0 = const()[name = tensor("op_15636_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15636_cast_fp16 = slice_by_index(begin = var_15636_begin_0, end = var_15636_end_0, end_mask = var_15636_end_mask_0, x = q_73_cast_fp16)[name = tensor("op_15636_cast_fp16")]; + tensor var_15640_begin_0 = const()[name = tensor("op_15640_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_15640_end_0 = const()[name = tensor("op_15640_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_15640_end_mask_0 = const()[name = tensor("op_15640_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15640_cast_fp16 = slice_by_index(begin = var_15640_begin_0, end = var_15640_end_0, end_mask = var_15640_end_mask_0, x = q_73_cast_fp16)[name = tensor("op_15640_cast_fp16")]; + tensor var_15644_begin_0 = const()[name = tensor("op_15644_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_15644_end_0 = const()[name = tensor("op_15644_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_15644_end_mask_0 = const()[name = tensor("op_15644_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15644_cast_fp16 = slice_by_index(begin = var_15644_begin_0, end = var_15644_end_0, end_mask = var_15644_end_mask_0, x = q_73_cast_fp16)[name = tensor("op_15644_cast_fp16")]; + tensor var_15648_begin_0 = const()[name = tensor("op_15648_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_15648_end_0 = const()[name = tensor("op_15648_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_15648_end_mask_0 = const()[name = tensor("op_15648_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15648_cast_fp16 = slice_by_index(begin = var_15648_begin_0, end = var_15648_end_0, end_mask = var_15648_end_mask_0, x = q_73_cast_fp16)[name = tensor("op_15648_cast_fp16")]; + tensor var_15652_begin_0 = const()[name = tensor("op_15652_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_15652_end_0 = const()[name = tensor("op_15652_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_15652_end_mask_0 = const()[name = tensor("op_15652_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15652_cast_fp16 = slice_by_index(begin = var_15652_begin_0, end = var_15652_end_0, end_mask = var_15652_end_mask_0, x = q_73_cast_fp16)[name = tensor("op_15652_cast_fp16")]; + tensor var_15656_begin_0 = const()[name = tensor("op_15656_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_15656_end_0 = const()[name = tensor("op_15656_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_15656_end_mask_0 = const()[name = tensor("op_15656_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15656_cast_fp16 = slice_by_index(begin = var_15656_begin_0, end = var_15656_end_0, end_mask = var_15656_end_mask_0, x = q_73_cast_fp16)[name = tensor("op_15656_cast_fp16")]; + tensor var_15660_begin_0 = const()[name = tensor("op_15660_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_15660_end_0 = const()[name = tensor("op_15660_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_15660_end_mask_0 = const()[name = tensor("op_15660_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15660_cast_fp16 = slice_by_index(begin = var_15660_begin_0, end = var_15660_end_0, end_mask = var_15660_end_mask_0, x = q_73_cast_fp16)[name = tensor("op_15660_cast_fp16")]; + tensor var_15664_begin_0 = const()[name = tensor("op_15664_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_15664_end_0 = const()[name = tensor("op_15664_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_15664_end_mask_0 = const()[name = tensor("op_15664_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15664_cast_fp16 = slice_by_index(begin = var_15664_begin_0, end = var_15664_end_0, end_mask = var_15664_end_mask_0, x = q_73_cast_fp16)[name = tensor("op_15664_cast_fp16")]; + tensor var_15668_begin_0 = const()[name = tensor("op_15668_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_15668_end_0 = const()[name = tensor("op_15668_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_15668_end_mask_0 = const()[name = tensor("op_15668_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15668_cast_fp16 = slice_by_index(begin = var_15668_begin_0, end = var_15668_end_0, end_mask = var_15668_end_mask_0, x = q_73_cast_fp16)[name = tensor("op_15668_cast_fp16")]; + tensor var_15672_begin_0 = const()[name = tensor("op_15672_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_15672_end_0 = const()[name = tensor("op_15672_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_15672_end_mask_0 = const()[name = tensor("op_15672_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15672_cast_fp16 = slice_by_index(begin = var_15672_begin_0, end = var_15672_end_0, end_mask = var_15672_end_mask_0, x = q_73_cast_fp16)[name = tensor("op_15672_cast_fp16")]; + tensor var_15676_begin_0 = const()[name = tensor("op_15676_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_15676_end_0 = const()[name = tensor("op_15676_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_15676_end_mask_0 = const()[name = tensor("op_15676_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15676_cast_fp16 = slice_by_index(begin = var_15676_begin_0, end = var_15676_end_0, end_mask = var_15676_end_mask_0, x = q_73_cast_fp16)[name = tensor("op_15676_cast_fp16")]; + tensor var_15680_begin_0 = const()[name = tensor("op_15680_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_15680_end_0 = const()[name = tensor("op_15680_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_15680_end_mask_0 = const()[name = tensor("op_15680_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15680_cast_fp16 = slice_by_index(begin = var_15680_begin_0, end = var_15680_end_0, end_mask = var_15680_end_mask_0, x = q_73_cast_fp16)[name = tensor("op_15680_cast_fp16")]; + tensor var_15684_begin_0 = const()[name = tensor("op_15684_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_15684_end_0 = const()[name = tensor("op_15684_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_15684_end_mask_0 = const()[name = tensor("op_15684_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15684_cast_fp16 = slice_by_index(begin = var_15684_begin_0, end = var_15684_end_0, end_mask = var_15684_end_mask_0, x = q_73_cast_fp16)[name = tensor("op_15684_cast_fp16")]; + tensor var_15688_begin_0 = const()[name = tensor("op_15688_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_15688_end_0 = const()[name = tensor("op_15688_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_15688_end_mask_0 = const()[name = tensor("op_15688_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15688_cast_fp16 = slice_by_index(begin = var_15688_begin_0, end = var_15688_end_0, end_mask = var_15688_end_mask_0, x = q_73_cast_fp16)[name = tensor("op_15688_cast_fp16")]; + tensor k_147_perm_0 = const()[name = tensor("k_147_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_15695_begin_0 = const()[name = tensor("op_15695_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15695_end_0 = const()[name = tensor("op_15695_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_15695_end_mask_0 = const()[name = tensor("op_15695_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_147_cast_fp16 = transpose(perm = k_147_perm_0, x = k_145_cast_fp16)[name = tensor("transpose_103")]; + tensor var_15695_cast_fp16 = slice_by_index(begin = var_15695_begin_0, end = var_15695_end_0, end_mask = var_15695_end_mask_0, x = k_147_cast_fp16)[name = tensor("op_15695_cast_fp16")]; + tensor var_15699_begin_0 = const()[name = tensor("op_15699_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_15699_end_0 = const()[name = tensor("op_15699_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_15699_end_mask_0 = const()[name = tensor("op_15699_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15699_cast_fp16 = slice_by_index(begin = var_15699_begin_0, end = var_15699_end_0, end_mask = var_15699_end_mask_0, x = k_147_cast_fp16)[name = tensor("op_15699_cast_fp16")]; + tensor var_15703_begin_0 = const()[name = tensor("op_15703_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_15703_end_0 = const()[name = tensor("op_15703_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_15703_end_mask_0 = const()[name = tensor("op_15703_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15703_cast_fp16 = slice_by_index(begin = var_15703_begin_0, end = var_15703_end_0, end_mask = var_15703_end_mask_0, x = k_147_cast_fp16)[name = tensor("op_15703_cast_fp16")]; + tensor var_15707_begin_0 = const()[name = tensor("op_15707_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_15707_end_0 = const()[name = tensor("op_15707_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_15707_end_mask_0 = const()[name = tensor("op_15707_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15707_cast_fp16 = slice_by_index(begin = var_15707_begin_0, end = var_15707_end_0, end_mask = var_15707_end_mask_0, x = k_147_cast_fp16)[name = tensor("op_15707_cast_fp16")]; + tensor var_15711_begin_0 = const()[name = tensor("op_15711_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_15711_end_0 = const()[name = tensor("op_15711_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_15711_end_mask_0 = const()[name = tensor("op_15711_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15711_cast_fp16 = slice_by_index(begin = var_15711_begin_0, end = var_15711_end_0, end_mask = var_15711_end_mask_0, x = k_147_cast_fp16)[name = tensor("op_15711_cast_fp16")]; + tensor var_15715_begin_0 = const()[name = tensor("op_15715_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_15715_end_0 = const()[name = tensor("op_15715_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_15715_end_mask_0 = const()[name = tensor("op_15715_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15715_cast_fp16 = slice_by_index(begin = var_15715_begin_0, end = var_15715_end_0, end_mask = var_15715_end_mask_0, x = k_147_cast_fp16)[name = tensor("op_15715_cast_fp16")]; + tensor var_15719_begin_0 = const()[name = tensor("op_15719_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_15719_end_0 = const()[name = tensor("op_15719_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_15719_end_mask_0 = const()[name = tensor("op_15719_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15719_cast_fp16 = slice_by_index(begin = var_15719_begin_0, end = var_15719_end_0, end_mask = var_15719_end_mask_0, x = k_147_cast_fp16)[name = tensor("op_15719_cast_fp16")]; + tensor var_15723_begin_0 = const()[name = tensor("op_15723_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_15723_end_0 = const()[name = tensor("op_15723_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_15723_end_mask_0 = const()[name = tensor("op_15723_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15723_cast_fp16 = slice_by_index(begin = var_15723_begin_0, end = var_15723_end_0, end_mask = var_15723_end_mask_0, x = k_147_cast_fp16)[name = tensor("op_15723_cast_fp16")]; + tensor var_15727_begin_0 = const()[name = tensor("op_15727_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_15727_end_0 = const()[name = tensor("op_15727_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_15727_end_mask_0 = const()[name = tensor("op_15727_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15727_cast_fp16 = slice_by_index(begin = var_15727_begin_0, end = var_15727_end_0, end_mask = var_15727_end_mask_0, x = k_147_cast_fp16)[name = tensor("op_15727_cast_fp16")]; + tensor var_15731_begin_0 = const()[name = tensor("op_15731_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_15731_end_0 = const()[name = tensor("op_15731_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_15731_end_mask_0 = const()[name = tensor("op_15731_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15731_cast_fp16 = slice_by_index(begin = var_15731_begin_0, end = var_15731_end_0, end_mask = var_15731_end_mask_0, x = k_147_cast_fp16)[name = tensor("op_15731_cast_fp16")]; + tensor var_15735_begin_0 = const()[name = tensor("op_15735_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_15735_end_0 = const()[name = tensor("op_15735_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_15735_end_mask_0 = const()[name = tensor("op_15735_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15735_cast_fp16 = slice_by_index(begin = var_15735_begin_0, end = var_15735_end_0, end_mask = var_15735_end_mask_0, x = k_147_cast_fp16)[name = tensor("op_15735_cast_fp16")]; + tensor var_15739_begin_0 = const()[name = tensor("op_15739_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_15739_end_0 = const()[name = tensor("op_15739_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_15739_end_mask_0 = const()[name = tensor("op_15739_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15739_cast_fp16 = slice_by_index(begin = var_15739_begin_0, end = var_15739_end_0, end_mask = var_15739_end_mask_0, x = k_147_cast_fp16)[name = tensor("op_15739_cast_fp16")]; + tensor var_15743_begin_0 = const()[name = tensor("op_15743_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_15743_end_0 = const()[name = tensor("op_15743_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_15743_end_mask_0 = const()[name = tensor("op_15743_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15743_cast_fp16 = slice_by_index(begin = var_15743_begin_0, end = var_15743_end_0, end_mask = var_15743_end_mask_0, x = k_147_cast_fp16)[name = tensor("op_15743_cast_fp16")]; + tensor var_15747_begin_0 = const()[name = tensor("op_15747_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_15747_end_0 = const()[name = tensor("op_15747_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_15747_end_mask_0 = const()[name = tensor("op_15747_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15747_cast_fp16 = slice_by_index(begin = var_15747_begin_0, end = var_15747_end_0, end_mask = var_15747_end_mask_0, x = k_147_cast_fp16)[name = tensor("op_15747_cast_fp16")]; + tensor var_15751_begin_0 = const()[name = tensor("op_15751_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_15751_end_0 = const()[name = tensor("op_15751_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_15751_end_mask_0 = const()[name = tensor("op_15751_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15751_cast_fp16 = slice_by_index(begin = var_15751_begin_0, end = var_15751_end_0, end_mask = var_15751_end_mask_0, x = k_147_cast_fp16)[name = tensor("op_15751_cast_fp16")]; + tensor var_15755_begin_0 = const()[name = tensor("op_15755_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_15755_end_0 = const()[name = tensor("op_15755_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_15755_end_mask_0 = const()[name = tensor("op_15755_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15755_cast_fp16 = slice_by_index(begin = var_15755_begin_0, end = var_15755_end_0, end_mask = var_15755_end_mask_0, x = k_147_cast_fp16)[name = tensor("op_15755_cast_fp16")]; + tensor var_15759_begin_0 = const()[name = tensor("op_15759_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_15759_end_0 = const()[name = tensor("op_15759_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_15759_end_mask_0 = const()[name = tensor("op_15759_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15759_cast_fp16 = slice_by_index(begin = var_15759_begin_0, end = var_15759_end_0, end_mask = var_15759_end_mask_0, x = k_147_cast_fp16)[name = tensor("op_15759_cast_fp16")]; + tensor var_15763_begin_0 = const()[name = tensor("op_15763_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_15763_end_0 = const()[name = tensor("op_15763_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_15763_end_mask_0 = const()[name = tensor("op_15763_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15763_cast_fp16 = slice_by_index(begin = var_15763_begin_0, end = var_15763_end_0, end_mask = var_15763_end_mask_0, x = k_147_cast_fp16)[name = tensor("op_15763_cast_fp16")]; + tensor var_15767_begin_0 = const()[name = tensor("op_15767_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_15767_end_0 = const()[name = tensor("op_15767_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_15767_end_mask_0 = const()[name = tensor("op_15767_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15767_cast_fp16 = slice_by_index(begin = var_15767_begin_0, end = var_15767_end_0, end_mask = var_15767_end_mask_0, x = k_147_cast_fp16)[name = tensor("op_15767_cast_fp16")]; + tensor var_15771_begin_0 = const()[name = tensor("op_15771_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_15771_end_0 = const()[name = tensor("op_15771_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_15771_end_mask_0 = const()[name = tensor("op_15771_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15771_cast_fp16 = slice_by_index(begin = var_15771_begin_0, end = var_15771_end_0, end_mask = var_15771_end_mask_0, x = k_147_cast_fp16)[name = tensor("op_15771_cast_fp16")]; + tensor var_15773_begin_0 = const()[name = tensor("op_15773_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15773_end_0 = const()[name = tensor("op_15773_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_15773_end_mask_0 = const()[name = tensor("op_15773_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15773_cast_fp16 = slice_by_index(begin = var_15773_begin_0, end = var_15773_end_0, end_mask = var_15773_end_mask_0, x = v_73_cast_fp16)[name = tensor("op_15773_cast_fp16")]; + tensor var_15777_begin_0 = const()[name = tensor("op_15777_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_15777_end_0 = const()[name = tensor("op_15777_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_15777_end_mask_0 = const()[name = tensor("op_15777_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15777_cast_fp16 = slice_by_index(begin = var_15777_begin_0, end = var_15777_end_0, end_mask = var_15777_end_mask_0, x = v_73_cast_fp16)[name = tensor("op_15777_cast_fp16")]; + tensor var_15781_begin_0 = const()[name = tensor("op_15781_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_15781_end_0 = const()[name = tensor("op_15781_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_15781_end_mask_0 = const()[name = tensor("op_15781_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15781_cast_fp16 = slice_by_index(begin = var_15781_begin_0, end = var_15781_end_0, end_mask = var_15781_end_mask_0, x = v_73_cast_fp16)[name = tensor("op_15781_cast_fp16")]; + tensor var_15785_begin_0 = const()[name = tensor("op_15785_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_15785_end_0 = const()[name = tensor("op_15785_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_15785_end_mask_0 = const()[name = tensor("op_15785_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15785_cast_fp16 = slice_by_index(begin = var_15785_begin_0, end = var_15785_end_0, end_mask = var_15785_end_mask_0, x = v_73_cast_fp16)[name = tensor("op_15785_cast_fp16")]; + tensor var_15789_begin_0 = const()[name = tensor("op_15789_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_15789_end_0 = const()[name = tensor("op_15789_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_15789_end_mask_0 = const()[name = tensor("op_15789_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15789_cast_fp16 = slice_by_index(begin = var_15789_begin_0, end = var_15789_end_0, end_mask = var_15789_end_mask_0, x = v_73_cast_fp16)[name = tensor("op_15789_cast_fp16")]; + tensor var_15793_begin_0 = const()[name = tensor("op_15793_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_15793_end_0 = const()[name = tensor("op_15793_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_15793_end_mask_0 = const()[name = tensor("op_15793_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15793_cast_fp16 = slice_by_index(begin = var_15793_begin_0, end = var_15793_end_0, end_mask = var_15793_end_mask_0, x = v_73_cast_fp16)[name = tensor("op_15793_cast_fp16")]; + tensor var_15797_begin_0 = const()[name = tensor("op_15797_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_15797_end_0 = const()[name = tensor("op_15797_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_15797_end_mask_0 = const()[name = tensor("op_15797_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15797_cast_fp16 = slice_by_index(begin = var_15797_begin_0, end = var_15797_end_0, end_mask = var_15797_end_mask_0, x = v_73_cast_fp16)[name = tensor("op_15797_cast_fp16")]; + tensor var_15801_begin_0 = const()[name = tensor("op_15801_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_15801_end_0 = const()[name = tensor("op_15801_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_15801_end_mask_0 = const()[name = tensor("op_15801_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15801_cast_fp16 = slice_by_index(begin = var_15801_begin_0, end = var_15801_end_0, end_mask = var_15801_end_mask_0, x = v_73_cast_fp16)[name = tensor("op_15801_cast_fp16")]; + tensor var_15805_begin_0 = const()[name = tensor("op_15805_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_15805_end_0 = const()[name = tensor("op_15805_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_15805_end_mask_0 = const()[name = tensor("op_15805_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15805_cast_fp16 = slice_by_index(begin = var_15805_begin_0, end = var_15805_end_0, end_mask = var_15805_end_mask_0, x = v_73_cast_fp16)[name = tensor("op_15805_cast_fp16")]; + tensor var_15809_begin_0 = const()[name = tensor("op_15809_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_15809_end_0 = const()[name = tensor("op_15809_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_15809_end_mask_0 = const()[name = tensor("op_15809_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15809_cast_fp16 = slice_by_index(begin = var_15809_begin_0, end = var_15809_end_0, end_mask = var_15809_end_mask_0, x = v_73_cast_fp16)[name = tensor("op_15809_cast_fp16")]; + tensor var_15813_begin_0 = const()[name = tensor("op_15813_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_15813_end_0 = const()[name = tensor("op_15813_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_15813_end_mask_0 = const()[name = tensor("op_15813_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15813_cast_fp16 = slice_by_index(begin = var_15813_begin_0, end = var_15813_end_0, end_mask = var_15813_end_mask_0, x = v_73_cast_fp16)[name = tensor("op_15813_cast_fp16")]; + tensor var_15817_begin_0 = const()[name = tensor("op_15817_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_15817_end_0 = const()[name = tensor("op_15817_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_15817_end_mask_0 = const()[name = tensor("op_15817_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15817_cast_fp16 = slice_by_index(begin = var_15817_begin_0, end = var_15817_end_0, end_mask = var_15817_end_mask_0, x = v_73_cast_fp16)[name = tensor("op_15817_cast_fp16")]; + tensor var_15821_begin_0 = const()[name = tensor("op_15821_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_15821_end_0 = const()[name = tensor("op_15821_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_15821_end_mask_0 = const()[name = tensor("op_15821_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15821_cast_fp16 = slice_by_index(begin = var_15821_begin_0, end = var_15821_end_0, end_mask = var_15821_end_mask_0, x = v_73_cast_fp16)[name = tensor("op_15821_cast_fp16")]; + tensor var_15825_begin_0 = const()[name = tensor("op_15825_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_15825_end_0 = const()[name = tensor("op_15825_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_15825_end_mask_0 = const()[name = tensor("op_15825_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15825_cast_fp16 = slice_by_index(begin = var_15825_begin_0, end = var_15825_end_0, end_mask = var_15825_end_mask_0, x = v_73_cast_fp16)[name = tensor("op_15825_cast_fp16")]; + tensor var_15829_begin_0 = const()[name = tensor("op_15829_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_15829_end_0 = const()[name = tensor("op_15829_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_15829_end_mask_0 = const()[name = tensor("op_15829_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15829_cast_fp16 = slice_by_index(begin = var_15829_begin_0, end = var_15829_end_0, end_mask = var_15829_end_mask_0, x = v_73_cast_fp16)[name = tensor("op_15829_cast_fp16")]; + tensor var_15833_begin_0 = const()[name = tensor("op_15833_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_15833_end_0 = const()[name = tensor("op_15833_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_15833_end_mask_0 = const()[name = tensor("op_15833_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15833_cast_fp16 = slice_by_index(begin = var_15833_begin_0, end = var_15833_end_0, end_mask = var_15833_end_mask_0, x = v_73_cast_fp16)[name = tensor("op_15833_cast_fp16")]; + tensor var_15837_begin_0 = const()[name = tensor("op_15837_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_15837_end_0 = const()[name = tensor("op_15837_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_15837_end_mask_0 = const()[name = tensor("op_15837_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15837_cast_fp16 = slice_by_index(begin = var_15837_begin_0, end = var_15837_end_0, end_mask = var_15837_end_mask_0, x = v_73_cast_fp16)[name = tensor("op_15837_cast_fp16")]; + tensor var_15841_begin_0 = const()[name = tensor("op_15841_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_15841_end_0 = const()[name = tensor("op_15841_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_15841_end_mask_0 = const()[name = tensor("op_15841_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15841_cast_fp16 = slice_by_index(begin = var_15841_begin_0, end = var_15841_end_0, end_mask = var_15841_end_mask_0, x = v_73_cast_fp16)[name = tensor("op_15841_cast_fp16")]; + tensor var_15845_begin_0 = const()[name = tensor("op_15845_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_15845_end_0 = const()[name = tensor("op_15845_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_15845_end_mask_0 = const()[name = tensor("op_15845_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15845_cast_fp16 = slice_by_index(begin = var_15845_begin_0, end = var_15845_end_0, end_mask = var_15845_end_mask_0, x = v_73_cast_fp16)[name = tensor("op_15845_cast_fp16")]; + tensor var_15849_begin_0 = const()[name = tensor("op_15849_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_15849_end_0 = const()[name = tensor("op_15849_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_15849_end_mask_0 = const()[name = tensor("op_15849_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15849_cast_fp16 = slice_by_index(begin = var_15849_begin_0, end = var_15849_end_0, end_mask = var_15849_end_mask_0, x = v_73_cast_fp16)[name = tensor("op_15849_cast_fp16")]; + tensor var_15853_equation_0 = const()[name = tensor("op_15853_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15853_cast_fp16 = einsum(equation = var_15853_equation_0, values = (var_15695_cast_fp16, var_15612_cast_fp16))[name = tensor("op_15853_cast_fp16")]; + tensor var_15854_to_fp16 = const()[name = tensor("op_15854_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1281_cast_fp16 = mul(x = var_15853_cast_fp16, y = var_15854_to_fp16)[name = tensor("aw_1281_cast_fp16")]; + tensor var_15857_equation_0 = const()[name = tensor("op_15857_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15857_cast_fp16 = einsum(equation = var_15857_equation_0, values = (var_15699_cast_fp16, var_15616_cast_fp16))[name = tensor("op_15857_cast_fp16")]; + tensor var_15858_to_fp16 = const()[name = tensor("op_15858_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1283_cast_fp16 = mul(x = var_15857_cast_fp16, y = var_15858_to_fp16)[name = tensor("aw_1283_cast_fp16")]; + tensor var_15861_equation_0 = const()[name = tensor("op_15861_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15861_cast_fp16 = einsum(equation = var_15861_equation_0, values = (var_15703_cast_fp16, var_15620_cast_fp16))[name = tensor("op_15861_cast_fp16")]; + tensor var_15862_to_fp16 = const()[name = tensor("op_15862_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1285_cast_fp16 = mul(x = var_15861_cast_fp16, y = var_15862_to_fp16)[name = tensor("aw_1285_cast_fp16")]; + tensor var_15865_equation_0 = const()[name = tensor("op_15865_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15865_cast_fp16 = einsum(equation = var_15865_equation_0, values = (var_15707_cast_fp16, var_15624_cast_fp16))[name = tensor("op_15865_cast_fp16")]; + tensor var_15866_to_fp16 = const()[name = tensor("op_15866_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1287_cast_fp16 = mul(x = var_15865_cast_fp16, y = var_15866_to_fp16)[name = tensor("aw_1287_cast_fp16")]; + tensor var_15869_equation_0 = const()[name = tensor("op_15869_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15869_cast_fp16 = einsum(equation = var_15869_equation_0, values = (var_15711_cast_fp16, var_15628_cast_fp16))[name = tensor("op_15869_cast_fp16")]; + tensor var_15870_to_fp16 = const()[name = tensor("op_15870_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1289_cast_fp16 = mul(x = var_15869_cast_fp16, y = var_15870_to_fp16)[name = tensor("aw_1289_cast_fp16")]; + tensor var_15873_equation_0 = const()[name = tensor("op_15873_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15873_cast_fp16 = einsum(equation = var_15873_equation_0, values = (var_15715_cast_fp16, var_15632_cast_fp16))[name = tensor("op_15873_cast_fp16")]; + tensor var_15874_to_fp16 = const()[name = tensor("op_15874_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1291_cast_fp16 = mul(x = var_15873_cast_fp16, y = var_15874_to_fp16)[name = tensor("aw_1291_cast_fp16")]; + tensor var_15877_equation_0 = const()[name = tensor("op_15877_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15877_cast_fp16 = einsum(equation = var_15877_equation_0, values = (var_15719_cast_fp16, var_15636_cast_fp16))[name = tensor("op_15877_cast_fp16")]; + tensor var_15878_to_fp16 = const()[name = tensor("op_15878_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1293_cast_fp16 = mul(x = var_15877_cast_fp16, y = var_15878_to_fp16)[name = tensor("aw_1293_cast_fp16")]; + tensor var_15881_equation_0 = const()[name = tensor("op_15881_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15881_cast_fp16 = einsum(equation = var_15881_equation_0, values = (var_15723_cast_fp16, var_15640_cast_fp16))[name = tensor("op_15881_cast_fp16")]; + tensor var_15882_to_fp16 = const()[name = tensor("op_15882_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1295_cast_fp16 = mul(x = var_15881_cast_fp16, y = var_15882_to_fp16)[name = tensor("aw_1295_cast_fp16")]; + tensor var_15885_equation_0 = const()[name = tensor("op_15885_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15885_cast_fp16 = einsum(equation = var_15885_equation_0, values = (var_15727_cast_fp16, var_15644_cast_fp16))[name = tensor("op_15885_cast_fp16")]; + tensor var_15886_to_fp16 = const()[name = tensor("op_15886_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1297_cast_fp16 = mul(x = var_15885_cast_fp16, y = var_15886_to_fp16)[name = tensor("aw_1297_cast_fp16")]; + tensor var_15889_equation_0 = const()[name = tensor("op_15889_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15889_cast_fp16 = einsum(equation = var_15889_equation_0, values = (var_15731_cast_fp16, var_15648_cast_fp16))[name = tensor("op_15889_cast_fp16")]; + tensor var_15890_to_fp16 = const()[name = tensor("op_15890_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1299_cast_fp16 = mul(x = var_15889_cast_fp16, y = var_15890_to_fp16)[name = tensor("aw_1299_cast_fp16")]; + tensor var_15893_equation_0 = const()[name = tensor("op_15893_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15893_cast_fp16 = einsum(equation = var_15893_equation_0, values = (var_15735_cast_fp16, var_15652_cast_fp16))[name = tensor("op_15893_cast_fp16")]; + tensor var_15894_to_fp16 = const()[name = tensor("op_15894_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1301_cast_fp16 = mul(x = var_15893_cast_fp16, y = var_15894_to_fp16)[name = tensor("aw_1301_cast_fp16")]; + tensor var_15897_equation_0 = const()[name = tensor("op_15897_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15897_cast_fp16 = einsum(equation = var_15897_equation_0, values = (var_15739_cast_fp16, var_15656_cast_fp16))[name = tensor("op_15897_cast_fp16")]; + tensor var_15898_to_fp16 = const()[name = tensor("op_15898_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1303_cast_fp16 = mul(x = var_15897_cast_fp16, y = var_15898_to_fp16)[name = tensor("aw_1303_cast_fp16")]; + tensor var_15901_equation_0 = const()[name = tensor("op_15901_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15901_cast_fp16 = einsum(equation = var_15901_equation_0, values = (var_15743_cast_fp16, var_15660_cast_fp16))[name = tensor("op_15901_cast_fp16")]; + tensor var_15902_to_fp16 = const()[name = tensor("op_15902_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1305_cast_fp16 = mul(x = var_15901_cast_fp16, y = var_15902_to_fp16)[name = tensor("aw_1305_cast_fp16")]; + tensor var_15905_equation_0 = const()[name = tensor("op_15905_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15905_cast_fp16 = einsum(equation = var_15905_equation_0, values = (var_15747_cast_fp16, var_15664_cast_fp16))[name = tensor("op_15905_cast_fp16")]; + tensor var_15906_to_fp16 = const()[name = tensor("op_15906_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1307_cast_fp16 = mul(x = var_15905_cast_fp16, y = var_15906_to_fp16)[name = tensor("aw_1307_cast_fp16")]; + tensor var_15909_equation_0 = const()[name = tensor("op_15909_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15909_cast_fp16 = einsum(equation = var_15909_equation_0, values = (var_15751_cast_fp16, var_15668_cast_fp16))[name = tensor("op_15909_cast_fp16")]; + tensor var_15910_to_fp16 = const()[name = tensor("op_15910_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1309_cast_fp16 = mul(x = var_15909_cast_fp16, y = var_15910_to_fp16)[name = tensor("aw_1309_cast_fp16")]; + tensor var_15913_equation_0 = const()[name = tensor("op_15913_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15913_cast_fp16 = einsum(equation = var_15913_equation_0, values = (var_15755_cast_fp16, var_15672_cast_fp16))[name = tensor("op_15913_cast_fp16")]; + tensor var_15914_to_fp16 = const()[name = tensor("op_15914_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1311_cast_fp16 = mul(x = var_15913_cast_fp16, y = var_15914_to_fp16)[name = tensor("aw_1311_cast_fp16")]; + tensor var_15917_equation_0 = const()[name = tensor("op_15917_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15917_cast_fp16 = einsum(equation = var_15917_equation_0, values = (var_15759_cast_fp16, var_15676_cast_fp16))[name = tensor("op_15917_cast_fp16")]; + tensor var_15918_to_fp16 = const()[name = tensor("op_15918_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1313_cast_fp16 = mul(x = var_15917_cast_fp16, y = var_15918_to_fp16)[name = tensor("aw_1313_cast_fp16")]; + tensor var_15921_equation_0 = const()[name = tensor("op_15921_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15921_cast_fp16 = einsum(equation = var_15921_equation_0, values = (var_15763_cast_fp16, var_15680_cast_fp16))[name = tensor("op_15921_cast_fp16")]; + tensor var_15922_to_fp16 = const()[name = tensor("op_15922_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1315_cast_fp16 = mul(x = var_15921_cast_fp16, y = var_15922_to_fp16)[name = tensor("aw_1315_cast_fp16")]; + tensor var_15925_equation_0 = const()[name = tensor("op_15925_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15925_cast_fp16 = einsum(equation = var_15925_equation_0, values = (var_15767_cast_fp16, var_15684_cast_fp16))[name = tensor("op_15925_cast_fp16")]; + tensor var_15926_to_fp16 = const()[name = tensor("op_15926_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1317_cast_fp16 = mul(x = var_15925_cast_fp16, y = var_15926_to_fp16)[name = tensor("aw_1317_cast_fp16")]; + tensor var_15929_equation_0 = const()[name = tensor("op_15929_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15929_cast_fp16 = einsum(equation = var_15929_equation_0, values = (var_15771_cast_fp16, var_15688_cast_fp16))[name = tensor("op_15929_cast_fp16")]; + tensor var_15930_to_fp16 = const()[name = tensor("op_15930_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1319_cast_fp16 = mul(x = var_15929_cast_fp16, y = var_15930_to_fp16)[name = tensor("aw_1319_cast_fp16")]; + tensor var_15932_cast_fp16 = softmax(axis = var_2624, x = aw_1281_cast_fp16)[name = tensor("op_15932_cast_fp16")]; + tensor var_15933_cast_fp16 = softmax(axis = var_2624, x = aw_1283_cast_fp16)[name = tensor("op_15933_cast_fp16")]; + tensor var_15934_cast_fp16 = softmax(axis = var_2624, x = aw_1285_cast_fp16)[name = tensor("op_15934_cast_fp16")]; + tensor var_15935_cast_fp16 = softmax(axis = var_2624, x = aw_1287_cast_fp16)[name = tensor("op_15935_cast_fp16")]; + tensor var_15936_cast_fp16 = softmax(axis = var_2624, x = aw_1289_cast_fp16)[name = tensor("op_15936_cast_fp16")]; + tensor var_15937_cast_fp16 = softmax(axis = var_2624, x = aw_1291_cast_fp16)[name = tensor("op_15937_cast_fp16")]; + tensor var_15938_cast_fp16 = softmax(axis = var_2624, x = aw_1293_cast_fp16)[name = tensor("op_15938_cast_fp16")]; + tensor var_15939_cast_fp16 = softmax(axis = var_2624, x = aw_1295_cast_fp16)[name = tensor("op_15939_cast_fp16")]; + tensor var_15940_cast_fp16 = softmax(axis = var_2624, x = aw_1297_cast_fp16)[name = tensor("op_15940_cast_fp16")]; + tensor var_15941_cast_fp16 = softmax(axis = var_2624, x = aw_1299_cast_fp16)[name = tensor("op_15941_cast_fp16")]; + tensor var_15942_cast_fp16 = softmax(axis = var_2624, x = aw_1301_cast_fp16)[name = tensor("op_15942_cast_fp16")]; + tensor var_15943_cast_fp16 = softmax(axis = var_2624, x = aw_1303_cast_fp16)[name = tensor("op_15943_cast_fp16")]; + tensor var_15944_cast_fp16 = softmax(axis = var_2624, x = aw_1305_cast_fp16)[name = tensor("op_15944_cast_fp16")]; + tensor var_15945_cast_fp16 = softmax(axis = var_2624, x = aw_1307_cast_fp16)[name = tensor("op_15945_cast_fp16")]; + tensor var_15946_cast_fp16 = softmax(axis = var_2624, x = aw_1309_cast_fp16)[name = tensor("op_15946_cast_fp16")]; + tensor var_15947_cast_fp16 = softmax(axis = var_2624, x = aw_1311_cast_fp16)[name = tensor("op_15947_cast_fp16")]; + tensor var_15948_cast_fp16 = softmax(axis = var_2624, x = aw_1313_cast_fp16)[name = tensor("op_15948_cast_fp16")]; + tensor var_15949_cast_fp16 = softmax(axis = var_2624, x = aw_1315_cast_fp16)[name = tensor("op_15949_cast_fp16")]; + tensor var_15950_cast_fp16 = softmax(axis = var_2624, x = aw_1317_cast_fp16)[name = tensor("op_15950_cast_fp16")]; + tensor var_15951_cast_fp16 = softmax(axis = var_2624, x = aw_1319_cast_fp16)[name = tensor("op_15951_cast_fp16")]; + tensor var_15953_equation_0 = const()[name = tensor("op_15953_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15953_cast_fp16 = einsum(equation = var_15953_equation_0, values = (var_15773_cast_fp16, var_15932_cast_fp16))[name = tensor("op_15953_cast_fp16")]; + tensor var_15955_equation_0 = const()[name = tensor("op_15955_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15955_cast_fp16 = einsum(equation = var_15955_equation_0, values = (var_15777_cast_fp16, var_15933_cast_fp16))[name = tensor("op_15955_cast_fp16")]; + tensor var_15957_equation_0 = const()[name = tensor("op_15957_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15957_cast_fp16 = einsum(equation = var_15957_equation_0, values = (var_15781_cast_fp16, var_15934_cast_fp16))[name = tensor("op_15957_cast_fp16")]; + tensor var_15959_equation_0 = const()[name = tensor("op_15959_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15959_cast_fp16 = einsum(equation = var_15959_equation_0, values = (var_15785_cast_fp16, var_15935_cast_fp16))[name = tensor("op_15959_cast_fp16")]; + tensor var_15961_equation_0 = const()[name = tensor("op_15961_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15961_cast_fp16 = einsum(equation = var_15961_equation_0, values = (var_15789_cast_fp16, var_15936_cast_fp16))[name = tensor("op_15961_cast_fp16")]; + tensor var_15963_equation_0 = const()[name = tensor("op_15963_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15963_cast_fp16 = einsum(equation = var_15963_equation_0, values = (var_15793_cast_fp16, var_15937_cast_fp16))[name = tensor("op_15963_cast_fp16")]; + tensor var_15965_equation_0 = const()[name = tensor("op_15965_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15965_cast_fp16 = einsum(equation = var_15965_equation_0, values = (var_15797_cast_fp16, var_15938_cast_fp16))[name = tensor("op_15965_cast_fp16")]; + tensor var_15967_equation_0 = const()[name = tensor("op_15967_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15967_cast_fp16 = einsum(equation = var_15967_equation_0, values = (var_15801_cast_fp16, var_15939_cast_fp16))[name = tensor("op_15967_cast_fp16")]; + tensor var_15969_equation_0 = const()[name = tensor("op_15969_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15969_cast_fp16 = einsum(equation = var_15969_equation_0, values = (var_15805_cast_fp16, var_15940_cast_fp16))[name = tensor("op_15969_cast_fp16")]; + tensor var_15971_equation_0 = const()[name = tensor("op_15971_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15971_cast_fp16 = einsum(equation = var_15971_equation_0, values = (var_15809_cast_fp16, var_15941_cast_fp16))[name = tensor("op_15971_cast_fp16")]; + tensor var_15973_equation_0 = const()[name = tensor("op_15973_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15973_cast_fp16 = einsum(equation = var_15973_equation_0, values = (var_15813_cast_fp16, var_15942_cast_fp16))[name = tensor("op_15973_cast_fp16")]; + tensor var_15975_equation_0 = const()[name = tensor("op_15975_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15975_cast_fp16 = einsum(equation = var_15975_equation_0, values = (var_15817_cast_fp16, var_15943_cast_fp16))[name = tensor("op_15975_cast_fp16")]; + tensor var_15977_equation_0 = const()[name = tensor("op_15977_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15977_cast_fp16 = einsum(equation = var_15977_equation_0, values = (var_15821_cast_fp16, var_15944_cast_fp16))[name = tensor("op_15977_cast_fp16")]; + tensor var_15979_equation_0 = const()[name = tensor("op_15979_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15979_cast_fp16 = einsum(equation = var_15979_equation_0, values = (var_15825_cast_fp16, var_15945_cast_fp16))[name = tensor("op_15979_cast_fp16")]; + tensor var_15981_equation_0 = const()[name = tensor("op_15981_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15981_cast_fp16 = einsum(equation = var_15981_equation_0, values = (var_15829_cast_fp16, var_15946_cast_fp16))[name = tensor("op_15981_cast_fp16")]; + tensor var_15983_equation_0 = const()[name = tensor("op_15983_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15983_cast_fp16 = einsum(equation = var_15983_equation_0, values = (var_15833_cast_fp16, var_15947_cast_fp16))[name = tensor("op_15983_cast_fp16")]; + tensor var_15985_equation_0 = const()[name = tensor("op_15985_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15985_cast_fp16 = einsum(equation = var_15985_equation_0, values = (var_15837_cast_fp16, var_15948_cast_fp16))[name = tensor("op_15985_cast_fp16")]; + tensor var_15987_equation_0 = const()[name = tensor("op_15987_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15987_cast_fp16 = einsum(equation = var_15987_equation_0, values = (var_15841_cast_fp16, var_15949_cast_fp16))[name = tensor("op_15987_cast_fp16")]; + tensor var_15989_equation_0 = const()[name = tensor("op_15989_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15989_cast_fp16 = einsum(equation = var_15989_equation_0, values = (var_15845_cast_fp16, var_15950_cast_fp16))[name = tensor("op_15989_cast_fp16")]; + tensor var_15991_equation_0 = const()[name = tensor("op_15991_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15991_cast_fp16 = einsum(equation = var_15991_equation_0, values = (var_15849_cast_fp16, var_15951_cast_fp16))[name = tensor("op_15991_cast_fp16")]; + tensor input_261_interleave_0 = const()[name = tensor("input_261_interleave_0"), val = tensor(false)]; + tensor input_261_cast_fp16 = concat(axis = var_2624, interleave = input_261_interleave_0, values = (var_15953_cast_fp16, var_15955_cast_fp16, var_15957_cast_fp16, var_15959_cast_fp16, var_15961_cast_fp16, var_15963_cast_fp16, var_15965_cast_fp16, var_15967_cast_fp16, var_15969_cast_fp16, var_15971_cast_fp16, var_15973_cast_fp16, var_15975_cast_fp16, var_15977_cast_fp16, var_15979_cast_fp16, var_15981_cast_fp16, var_15983_cast_fp16, var_15985_cast_fp16, var_15987_cast_fp16, var_15989_cast_fp16, var_15991_cast_fp16))[name = tensor("input_261_cast_fp16")]; + tensor var_15997 = const()[name = tensor("op_15997"), val = tensor([1, 1])]; + tensor var_15999 = const()[name = tensor("op_15999"), val = tensor([1, 1])]; + tensor var_16001_pad_type_0 = const()[name = tensor("op_16001_pad_type_0"), val = tensor("custom")]; + tensor var_16001_pad_0 = const()[name = tensor("op_16001_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_4_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_4_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1250025152)))]; + tensor down_blocks_2_attentions_1_transformer_blocks_4_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_4_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1253302016)))]; + tensor var_16001_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_4_attn1_to_out_0_bias_to_fp16, dilations = var_15999, groups = var_2624, pad = var_16001_pad_0, pad_type = var_16001_pad_type_0, strides = var_15997, weight = down_blocks_2_attentions_1_transformer_blocks_4_attn1_to_out_0_weight_to_fp16, x = input_261_cast_fp16)[name = tensor("op_16001_cast_fp16")]; + tensor inputs_111_cast_fp16 = add(x = var_16001_cast_fp16, y = inputs_109_cast_fp16)[name = tensor("inputs_111_cast_fp16")]; + tensor hidden_states_163_axes_0 = const()[name = tensor("hidden_states_163_axes_0"), val = tensor([1])]; + tensor hidden_states_163_gamma_0_to_fp16 = const()[name = tensor("hidden_states_163_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1253304640)))]; + tensor hidden_states_163_beta_0_to_fp16 = const()[name = tensor("hidden_states_163_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1253307264)))]; + tensor var_16011_to_fp16 = const()[name = tensor("op_16011_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_163_cast_fp16 = layer_norm(axes = hidden_states_163_axes_0, beta = hidden_states_163_beta_0_to_fp16, epsilon = var_16011_to_fp16, gamma = hidden_states_163_gamma_0_to_fp16, x = inputs_111_cast_fp16)[name = tensor("hidden_states_163_cast_fp16")]; + tensor var_16026 = const()[name = tensor("op_16026"), val = tensor([1, 1])]; + tensor var_16028 = const()[name = tensor("op_16028"), val = tensor([1, 1])]; + tensor q_75_pad_type_0 = const()[name = tensor("q_75_pad_type_0"), val = tensor("custom")]; + tensor q_75_pad_0 = const()[name = tensor("q_75_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_4_attn2_to_q_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_4_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1253309888)))]; + tensor q_75_cast_fp16 = conv(dilations = var_16028, groups = var_2624, pad = q_75_pad_0, pad_type = q_75_pad_type_0, strides = var_16026, weight = down_blocks_2_attentions_1_transformer_blocks_4_attn2_to_q_weight_to_fp16, x = hidden_states_163_cast_fp16)[name = tensor("q_75_cast_fp16")]; + tensor var_16032 = const()[name = tensor("op_16032"), val = tensor([1, 1])]; + tensor var_16034 = const()[name = tensor("op_16034"), val = tensor([1, 1])]; + tensor k_149_pad_type_0 = const()[name = tensor("k_149_pad_type_0"), val = tensor("custom")]; + tensor k_149_pad_0 = const()[name = tensor("k_149_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_4_attn2_to_k_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_4_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1256586752)))]; + tensor k_149_cast_fp16 = conv(dilations = var_16034, groups = var_2624, pad = k_149_pad_0, pad_type = k_149_pad_type_0, strides = var_16032, weight = down_blocks_2_attentions_1_transformer_blocks_4_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_149_cast_fp16")]; + tensor var_16038 = const()[name = tensor("op_16038"), val = tensor([1, 1])]; + tensor var_16040 = const()[name = tensor("op_16040"), val = tensor([1, 1])]; + tensor v_75_pad_type_0 = const()[name = tensor("v_75_pad_type_0"), val = tensor("custom")]; + tensor v_75_pad_0 = const()[name = tensor("v_75_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_4_attn2_to_v_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_4_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1261829696)))]; + tensor v_75_cast_fp16 = conv(dilations = var_16040, groups = var_2624, pad = v_75_pad_0, pad_type = v_75_pad_type_0, strides = var_16038, weight = down_blocks_2_attentions_1_transformer_blocks_4_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_75_cast_fp16")]; + tensor var_16044_begin_0 = const()[name = tensor("op_16044_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_16044_end_0 = const()[name = tensor("op_16044_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_16044_end_mask_0 = const()[name = tensor("op_16044_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16044_cast_fp16 = slice_by_index(begin = var_16044_begin_0, end = var_16044_end_0, end_mask = var_16044_end_mask_0, x = q_75_cast_fp16)[name = tensor("op_16044_cast_fp16")]; + tensor var_16048_begin_0 = const()[name = tensor("op_16048_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_16048_end_0 = const()[name = tensor("op_16048_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_16048_end_mask_0 = const()[name = tensor("op_16048_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16048_cast_fp16 = slice_by_index(begin = var_16048_begin_0, end = var_16048_end_0, end_mask = var_16048_end_mask_0, x = q_75_cast_fp16)[name = tensor("op_16048_cast_fp16")]; + tensor var_16052_begin_0 = const()[name = tensor("op_16052_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_16052_end_0 = const()[name = tensor("op_16052_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_16052_end_mask_0 = const()[name = tensor("op_16052_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16052_cast_fp16 = slice_by_index(begin = var_16052_begin_0, end = var_16052_end_0, end_mask = var_16052_end_mask_0, x = q_75_cast_fp16)[name = tensor("op_16052_cast_fp16")]; + tensor var_16056_begin_0 = const()[name = tensor("op_16056_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_16056_end_0 = const()[name = tensor("op_16056_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_16056_end_mask_0 = const()[name = tensor("op_16056_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16056_cast_fp16 = slice_by_index(begin = var_16056_begin_0, end = var_16056_end_0, end_mask = var_16056_end_mask_0, x = q_75_cast_fp16)[name = tensor("op_16056_cast_fp16")]; + tensor var_16060_begin_0 = const()[name = tensor("op_16060_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_16060_end_0 = const()[name = tensor("op_16060_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_16060_end_mask_0 = const()[name = tensor("op_16060_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16060_cast_fp16 = slice_by_index(begin = var_16060_begin_0, end = var_16060_end_0, end_mask = var_16060_end_mask_0, x = q_75_cast_fp16)[name = tensor("op_16060_cast_fp16")]; + tensor var_16064_begin_0 = const()[name = tensor("op_16064_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_16064_end_0 = const()[name = tensor("op_16064_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_16064_end_mask_0 = const()[name = tensor("op_16064_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16064_cast_fp16 = slice_by_index(begin = var_16064_begin_0, end = var_16064_end_0, end_mask = var_16064_end_mask_0, x = q_75_cast_fp16)[name = tensor("op_16064_cast_fp16")]; + tensor var_16068_begin_0 = const()[name = tensor("op_16068_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_16068_end_0 = const()[name = tensor("op_16068_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_16068_end_mask_0 = const()[name = tensor("op_16068_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16068_cast_fp16 = slice_by_index(begin = var_16068_begin_0, end = var_16068_end_0, end_mask = var_16068_end_mask_0, x = q_75_cast_fp16)[name = tensor("op_16068_cast_fp16")]; + tensor var_16072_begin_0 = const()[name = tensor("op_16072_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_16072_end_0 = const()[name = tensor("op_16072_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_16072_end_mask_0 = const()[name = tensor("op_16072_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16072_cast_fp16 = slice_by_index(begin = var_16072_begin_0, end = var_16072_end_0, end_mask = var_16072_end_mask_0, x = q_75_cast_fp16)[name = tensor("op_16072_cast_fp16")]; + tensor var_16076_begin_0 = const()[name = tensor("op_16076_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_16076_end_0 = const()[name = tensor("op_16076_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_16076_end_mask_0 = const()[name = tensor("op_16076_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16076_cast_fp16 = slice_by_index(begin = var_16076_begin_0, end = var_16076_end_0, end_mask = var_16076_end_mask_0, x = q_75_cast_fp16)[name = tensor("op_16076_cast_fp16")]; + tensor var_16080_begin_0 = const()[name = tensor("op_16080_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_16080_end_0 = const()[name = tensor("op_16080_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_16080_end_mask_0 = const()[name = tensor("op_16080_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16080_cast_fp16 = slice_by_index(begin = var_16080_begin_0, end = var_16080_end_0, end_mask = var_16080_end_mask_0, x = q_75_cast_fp16)[name = tensor("op_16080_cast_fp16")]; + tensor var_16084_begin_0 = const()[name = tensor("op_16084_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_16084_end_0 = const()[name = tensor("op_16084_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_16084_end_mask_0 = const()[name = tensor("op_16084_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16084_cast_fp16 = slice_by_index(begin = var_16084_begin_0, end = var_16084_end_0, end_mask = var_16084_end_mask_0, x = q_75_cast_fp16)[name = tensor("op_16084_cast_fp16")]; + tensor var_16088_begin_0 = const()[name = tensor("op_16088_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_16088_end_0 = const()[name = tensor("op_16088_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_16088_end_mask_0 = const()[name = tensor("op_16088_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16088_cast_fp16 = slice_by_index(begin = var_16088_begin_0, end = var_16088_end_0, end_mask = var_16088_end_mask_0, x = q_75_cast_fp16)[name = tensor("op_16088_cast_fp16")]; + tensor var_16092_begin_0 = const()[name = tensor("op_16092_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_16092_end_0 = const()[name = tensor("op_16092_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_16092_end_mask_0 = const()[name = tensor("op_16092_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16092_cast_fp16 = slice_by_index(begin = var_16092_begin_0, end = var_16092_end_0, end_mask = var_16092_end_mask_0, x = q_75_cast_fp16)[name = tensor("op_16092_cast_fp16")]; + tensor var_16096_begin_0 = const()[name = tensor("op_16096_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_16096_end_0 = const()[name = tensor("op_16096_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_16096_end_mask_0 = const()[name = tensor("op_16096_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16096_cast_fp16 = slice_by_index(begin = var_16096_begin_0, end = var_16096_end_0, end_mask = var_16096_end_mask_0, x = q_75_cast_fp16)[name = tensor("op_16096_cast_fp16")]; + tensor var_16100_begin_0 = const()[name = tensor("op_16100_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_16100_end_0 = const()[name = tensor("op_16100_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_16100_end_mask_0 = const()[name = tensor("op_16100_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16100_cast_fp16 = slice_by_index(begin = var_16100_begin_0, end = var_16100_end_0, end_mask = var_16100_end_mask_0, x = q_75_cast_fp16)[name = tensor("op_16100_cast_fp16")]; + tensor var_16104_begin_0 = const()[name = tensor("op_16104_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_16104_end_0 = const()[name = tensor("op_16104_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_16104_end_mask_0 = const()[name = tensor("op_16104_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16104_cast_fp16 = slice_by_index(begin = var_16104_begin_0, end = var_16104_end_0, end_mask = var_16104_end_mask_0, x = q_75_cast_fp16)[name = tensor("op_16104_cast_fp16")]; + tensor var_16108_begin_0 = const()[name = tensor("op_16108_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_16108_end_0 = const()[name = tensor("op_16108_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_16108_end_mask_0 = const()[name = tensor("op_16108_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16108_cast_fp16 = slice_by_index(begin = var_16108_begin_0, end = var_16108_end_0, end_mask = var_16108_end_mask_0, x = q_75_cast_fp16)[name = tensor("op_16108_cast_fp16")]; + tensor var_16112_begin_0 = const()[name = tensor("op_16112_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_16112_end_0 = const()[name = tensor("op_16112_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_16112_end_mask_0 = const()[name = tensor("op_16112_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16112_cast_fp16 = slice_by_index(begin = var_16112_begin_0, end = var_16112_end_0, end_mask = var_16112_end_mask_0, x = q_75_cast_fp16)[name = tensor("op_16112_cast_fp16")]; + tensor var_16116_begin_0 = const()[name = tensor("op_16116_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_16116_end_0 = const()[name = tensor("op_16116_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_16116_end_mask_0 = const()[name = tensor("op_16116_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16116_cast_fp16 = slice_by_index(begin = var_16116_begin_0, end = var_16116_end_0, end_mask = var_16116_end_mask_0, x = q_75_cast_fp16)[name = tensor("op_16116_cast_fp16")]; + tensor var_16120_begin_0 = const()[name = tensor("op_16120_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_16120_end_0 = const()[name = tensor("op_16120_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_16120_end_mask_0 = const()[name = tensor("op_16120_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16120_cast_fp16 = slice_by_index(begin = var_16120_begin_0, end = var_16120_end_0, end_mask = var_16120_end_mask_0, x = q_75_cast_fp16)[name = tensor("op_16120_cast_fp16")]; + tensor k_151_perm_0 = const()[name = tensor("k_151_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_16127_begin_0 = const()[name = tensor("op_16127_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_16127_end_0 = const()[name = tensor("op_16127_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_16127_end_mask_0 = const()[name = tensor("op_16127_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_151_cast_fp16 = transpose(perm = k_151_perm_0, x = k_149_cast_fp16)[name = tensor("transpose_102")]; + tensor var_16127_cast_fp16 = slice_by_index(begin = var_16127_begin_0, end = var_16127_end_0, end_mask = var_16127_end_mask_0, x = k_151_cast_fp16)[name = tensor("op_16127_cast_fp16")]; + tensor var_16131_begin_0 = const()[name = tensor("op_16131_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_16131_end_0 = const()[name = tensor("op_16131_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_16131_end_mask_0 = const()[name = tensor("op_16131_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16131_cast_fp16 = slice_by_index(begin = var_16131_begin_0, end = var_16131_end_0, end_mask = var_16131_end_mask_0, x = k_151_cast_fp16)[name = tensor("op_16131_cast_fp16")]; + tensor var_16135_begin_0 = const()[name = tensor("op_16135_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_16135_end_0 = const()[name = tensor("op_16135_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_16135_end_mask_0 = const()[name = tensor("op_16135_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16135_cast_fp16 = slice_by_index(begin = var_16135_begin_0, end = var_16135_end_0, end_mask = var_16135_end_mask_0, x = k_151_cast_fp16)[name = tensor("op_16135_cast_fp16")]; + tensor var_16139_begin_0 = const()[name = tensor("op_16139_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_16139_end_0 = const()[name = tensor("op_16139_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_16139_end_mask_0 = const()[name = tensor("op_16139_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16139_cast_fp16 = slice_by_index(begin = var_16139_begin_0, end = var_16139_end_0, end_mask = var_16139_end_mask_0, x = k_151_cast_fp16)[name = tensor("op_16139_cast_fp16")]; + tensor var_16143_begin_0 = const()[name = tensor("op_16143_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_16143_end_0 = const()[name = tensor("op_16143_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_16143_end_mask_0 = const()[name = tensor("op_16143_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16143_cast_fp16 = slice_by_index(begin = var_16143_begin_0, end = var_16143_end_0, end_mask = var_16143_end_mask_0, x = k_151_cast_fp16)[name = tensor("op_16143_cast_fp16")]; + tensor var_16147_begin_0 = const()[name = tensor("op_16147_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_16147_end_0 = const()[name = tensor("op_16147_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_16147_end_mask_0 = const()[name = tensor("op_16147_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16147_cast_fp16 = slice_by_index(begin = var_16147_begin_0, end = var_16147_end_0, end_mask = var_16147_end_mask_0, x = k_151_cast_fp16)[name = tensor("op_16147_cast_fp16")]; + tensor var_16151_begin_0 = const()[name = tensor("op_16151_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_16151_end_0 = const()[name = tensor("op_16151_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_16151_end_mask_0 = const()[name = tensor("op_16151_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16151_cast_fp16 = slice_by_index(begin = var_16151_begin_0, end = var_16151_end_0, end_mask = var_16151_end_mask_0, x = k_151_cast_fp16)[name = tensor("op_16151_cast_fp16")]; + tensor var_16155_begin_0 = const()[name = tensor("op_16155_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_16155_end_0 = const()[name = tensor("op_16155_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_16155_end_mask_0 = const()[name = tensor("op_16155_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16155_cast_fp16 = slice_by_index(begin = var_16155_begin_0, end = var_16155_end_0, end_mask = var_16155_end_mask_0, x = k_151_cast_fp16)[name = tensor("op_16155_cast_fp16")]; + tensor var_16159_begin_0 = const()[name = tensor("op_16159_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_16159_end_0 = const()[name = tensor("op_16159_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_16159_end_mask_0 = const()[name = tensor("op_16159_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16159_cast_fp16 = slice_by_index(begin = var_16159_begin_0, end = var_16159_end_0, end_mask = var_16159_end_mask_0, x = k_151_cast_fp16)[name = tensor("op_16159_cast_fp16")]; + tensor var_16163_begin_0 = const()[name = tensor("op_16163_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_16163_end_0 = const()[name = tensor("op_16163_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_16163_end_mask_0 = const()[name = tensor("op_16163_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16163_cast_fp16 = slice_by_index(begin = var_16163_begin_0, end = var_16163_end_0, end_mask = var_16163_end_mask_0, x = k_151_cast_fp16)[name = tensor("op_16163_cast_fp16")]; + tensor var_16167_begin_0 = const()[name = tensor("op_16167_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_16167_end_0 = const()[name = tensor("op_16167_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_16167_end_mask_0 = const()[name = tensor("op_16167_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16167_cast_fp16 = slice_by_index(begin = var_16167_begin_0, end = var_16167_end_0, end_mask = var_16167_end_mask_0, x = k_151_cast_fp16)[name = tensor("op_16167_cast_fp16")]; + tensor var_16171_begin_0 = const()[name = tensor("op_16171_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_16171_end_0 = const()[name = tensor("op_16171_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_16171_end_mask_0 = const()[name = tensor("op_16171_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16171_cast_fp16 = slice_by_index(begin = var_16171_begin_0, end = var_16171_end_0, end_mask = var_16171_end_mask_0, x = k_151_cast_fp16)[name = tensor("op_16171_cast_fp16")]; + tensor var_16175_begin_0 = const()[name = tensor("op_16175_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_16175_end_0 = const()[name = tensor("op_16175_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_16175_end_mask_0 = const()[name = tensor("op_16175_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16175_cast_fp16 = slice_by_index(begin = var_16175_begin_0, end = var_16175_end_0, end_mask = var_16175_end_mask_0, x = k_151_cast_fp16)[name = tensor("op_16175_cast_fp16")]; + tensor var_16179_begin_0 = const()[name = tensor("op_16179_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_16179_end_0 = const()[name = tensor("op_16179_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_16179_end_mask_0 = const()[name = tensor("op_16179_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16179_cast_fp16 = slice_by_index(begin = var_16179_begin_0, end = var_16179_end_0, end_mask = var_16179_end_mask_0, x = k_151_cast_fp16)[name = tensor("op_16179_cast_fp16")]; + tensor var_16183_begin_0 = const()[name = tensor("op_16183_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_16183_end_0 = const()[name = tensor("op_16183_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_16183_end_mask_0 = const()[name = tensor("op_16183_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16183_cast_fp16 = slice_by_index(begin = var_16183_begin_0, end = var_16183_end_0, end_mask = var_16183_end_mask_0, x = k_151_cast_fp16)[name = tensor("op_16183_cast_fp16")]; + tensor var_16187_begin_0 = const()[name = tensor("op_16187_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_16187_end_0 = const()[name = tensor("op_16187_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_16187_end_mask_0 = const()[name = tensor("op_16187_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16187_cast_fp16 = slice_by_index(begin = var_16187_begin_0, end = var_16187_end_0, end_mask = var_16187_end_mask_0, x = k_151_cast_fp16)[name = tensor("op_16187_cast_fp16")]; + tensor var_16191_begin_0 = const()[name = tensor("op_16191_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_16191_end_0 = const()[name = tensor("op_16191_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_16191_end_mask_0 = const()[name = tensor("op_16191_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16191_cast_fp16 = slice_by_index(begin = var_16191_begin_0, end = var_16191_end_0, end_mask = var_16191_end_mask_0, x = k_151_cast_fp16)[name = tensor("op_16191_cast_fp16")]; + tensor var_16195_begin_0 = const()[name = tensor("op_16195_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_16195_end_0 = const()[name = tensor("op_16195_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_16195_end_mask_0 = const()[name = tensor("op_16195_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16195_cast_fp16 = slice_by_index(begin = var_16195_begin_0, end = var_16195_end_0, end_mask = var_16195_end_mask_0, x = k_151_cast_fp16)[name = tensor("op_16195_cast_fp16")]; + tensor var_16199_begin_0 = const()[name = tensor("op_16199_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_16199_end_0 = const()[name = tensor("op_16199_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_16199_end_mask_0 = const()[name = tensor("op_16199_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16199_cast_fp16 = slice_by_index(begin = var_16199_begin_0, end = var_16199_end_0, end_mask = var_16199_end_mask_0, x = k_151_cast_fp16)[name = tensor("op_16199_cast_fp16")]; + tensor var_16203_begin_0 = const()[name = tensor("op_16203_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_16203_end_0 = const()[name = tensor("op_16203_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_16203_end_mask_0 = const()[name = tensor("op_16203_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16203_cast_fp16 = slice_by_index(begin = var_16203_begin_0, end = var_16203_end_0, end_mask = var_16203_end_mask_0, x = k_151_cast_fp16)[name = tensor("op_16203_cast_fp16")]; + tensor var_16205_begin_0 = const()[name = tensor("op_16205_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_16205_end_0 = const()[name = tensor("op_16205_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_16205_end_mask_0 = const()[name = tensor("op_16205_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16205_cast_fp16 = slice_by_index(begin = var_16205_begin_0, end = var_16205_end_0, end_mask = var_16205_end_mask_0, x = v_75_cast_fp16)[name = tensor("op_16205_cast_fp16")]; + tensor var_16209_begin_0 = const()[name = tensor("op_16209_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_16209_end_0 = const()[name = tensor("op_16209_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_16209_end_mask_0 = const()[name = tensor("op_16209_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16209_cast_fp16 = slice_by_index(begin = var_16209_begin_0, end = var_16209_end_0, end_mask = var_16209_end_mask_0, x = v_75_cast_fp16)[name = tensor("op_16209_cast_fp16")]; + tensor var_16213_begin_0 = const()[name = tensor("op_16213_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_16213_end_0 = const()[name = tensor("op_16213_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_16213_end_mask_0 = const()[name = tensor("op_16213_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16213_cast_fp16 = slice_by_index(begin = var_16213_begin_0, end = var_16213_end_0, end_mask = var_16213_end_mask_0, x = v_75_cast_fp16)[name = tensor("op_16213_cast_fp16")]; + tensor var_16217_begin_0 = const()[name = tensor("op_16217_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_16217_end_0 = const()[name = tensor("op_16217_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_16217_end_mask_0 = const()[name = tensor("op_16217_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16217_cast_fp16 = slice_by_index(begin = var_16217_begin_0, end = var_16217_end_0, end_mask = var_16217_end_mask_0, x = v_75_cast_fp16)[name = tensor("op_16217_cast_fp16")]; + tensor var_16221_begin_0 = const()[name = tensor("op_16221_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_16221_end_0 = const()[name = tensor("op_16221_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_16221_end_mask_0 = const()[name = tensor("op_16221_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16221_cast_fp16 = slice_by_index(begin = var_16221_begin_0, end = var_16221_end_0, end_mask = var_16221_end_mask_0, x = v_75_cast_fp16)[name = tensor("op_16221_cast_fp16")]; + tensor var_16225_begin_0 = const()[name = tensor("op_16225_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_16225_end_0 = const()[name = tensor("op_16225_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_16225_end_mask_0 = const()[name = tensor("op_16225_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16225_cast_fp16 = slice_by_index(begin = var_16225_begin_0, end = var_16225_end_0, end_mask = var_16225_end_mask_0, x = v_75_cast_fp16)[name = tensor("op_16225_cast_fp16")]; + tensor var_16229_begin_0 = const()[name = tensor("op_16229_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_16229_end_0 = const()[name = tensor("op_16229_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_16229_end_mask_0 = const()[name = tensor("op_16229_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16229_cast_fp16 = slice_by_index(begin = var_16229_begin_0, end = var_16229_end_0, end_mask = var_16229_end_mask_0, x = v_75_cast_fp16)[name = tensor("op_16229_cast_fp16")]; + tensor var_16233_begin_0 = const()[name = tensor("op_16233_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_16233_end_0 = const()[name = tensor("op_16233_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_16233_end_mask_0 = const()[name = tensor("op_16233_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16233_cast_fp16 = slice_by_index(begin = var_16233_begin_0, end = var_16233_end_0, end_mask = var_16233_end_mask_0, x = v_75_cast_fp16)[name = tensor("op_16233_cast_fp16")]; + tensor var_16237_begin_0 = const()[name = tensor("op_16237_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_16237_end_0 = const()[name = tensor("op_16237_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_16237_end_mask_0 = const()[name = tensor("op_16237_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16237_cast_fp16 = slice_by_index(begin = var_16237_begin_0, end = var_16237_end_0, end_mask = var_16237_end_mask_0, x = v_75_cast_fp16)[name = tensor("op_16237_cast_fp16")]; + tensor var_16241_begin_0 = const()[name = tensor("op_16241_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_16241_end_0 = const()[name = tensor("op_16241_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_16241_end_mask_0 = const()[name = tensor("op_16241_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16241_cast_fp16 = slice_by_index(begin = var_16241_begin_0, end = var_16241_end_0, end_mask = var_16241_end_mask_0, x = v_75_cast_fp16)[name = tensor("op_16241_cast_fp16")]; + tensor var_16245_begin_0 = const()[name = tensor("op_16245_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_16245_end_0 = const()[name = tensor("op_16245_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_16245_end_mask_0 = const()[name = tensor("op_16245_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16245_cast_fp16 = slice_by_index(begin = var_16245_begin_0, end = var_16245_end_0, end_mask = var_16245_end_mask_0, x = v_75_cast_fp16)[name = tensor("op_16245_cast_fp16")]; + tensor var_16249_begin_0 = const()[name = tensor("op_16249_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_16249_end_0 = const()[name = tensor("op_16249_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_16249_end_mask_0 = const()[name = tensor("op_16249_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16249_cast_fp16 = slice_by_index(begin = var_16249_begin_0, end = var_16249_end_0, end_mask = var_16249_end_mask_0, x = v_75_cast_fp16)[name = tensor("op_16249_cast_fp16")]; + tensor var_16253_begin_0 = const()[name = tensor("op_16253_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_16253_end_0 = const()[name = tensor("op_16253_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_16253_end_mask_0 = const()[name = tensor("op_16253_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16253_cast_fp16 = slice_by_index(begin = var_16253_begin_0, end = var_16253_end_0, end_mask = var_16253_end_mask_0, x = v_75_cast_fp16)[name = tensor("op_16253_cast_fp16")]; + tensor var_16257_begin_0 = const()[name = tensor("op_16257_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_16257_end_0 = const()[name = tensor("op_16257_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_16257_end_mask_0 = const()[name = tensor("op_16257_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16257_cast_fp16 = slice_by_index(begin = var_16257_begin_0, end = var_16257_end_0, end_mask = var_16257_end_mask_0, x = v_75_cast_fp16)[name = tensor("op_16257_cast_fp16")]; + tensor var_16261_begin_0 = const()[name = tensor("op_16261_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_16261_end_0 = const()[name = tensor("op_16261_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_16261_end_mask_0 = const()[name = tensor("op_16261_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16261_cast_fp16 = slice_by_index(begin = var_16261_begin_0, end = var_16261_end_0, end_mask = var_16261_end_mask_0, x = v_75_cast_fp16)[name = tensor("op_16261_cast_fp16")]; + tensor var_16265_begin_0 = const()[name = tensor("op_16265_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_16265_end_0 = const()[name = tensor("op_16265_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_16265_end_mask_0 = const()[name = tensor("op_16265_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16265_cast_fp16 = slice_by_index(begin = var_16265_begin_0, end = var_16265_end_0, end_mask = var_16265_end_mask_0, x = v_75_cast_fp16)[name = tensor("op_16265_cast_fp16")]; + tensor var_16269_begin_0 = const()[name = tensor("op_16269_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_16269_end_0 = const()[name = tensor("op_16269_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_16269_end_mask_0 = const()[name = tensor("op_16269_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16269_cast_fp16 = slice_by_index(begin = var_16269_begin_0, end = var_16269_end_0, end_mask = var_16269_end_mask_0, x = v_75_cast_fp16)[name = tensor("op_16269_cast_fp16")]; + tensor var_16273_begin_0 = const()[name = tensor("op_16273_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_16273_end_0 = const()[name = tensor("op_16273_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_16273_end_mask_0 = const()[name = tensor("op_16273_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16273_cast_fp16 = slice_by_index(begin = var_16273_begin_0, end = var_16273_end_0, end_mask = var_16273_end_mask_0, x = v_75_cast_fp16)[name = tensor("op_16273_cast_fp16")]; + tensor var_16277_begin_0 = const()[name = tensor("op_16277_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_16277_end_0 = const()[name = tensor("op_16277_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_16277_end_mask_0 = const()[name = tensor("op_16277_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16277_cast_fp16 = slice_by_index(begin = var_16277_begin_0, end = var_16277_end_0, end_mask = var_16277_end_mask_0, x = v_75_cast_fp16)[name = tensor("op_16277_cast_fp16")]; + tensor var_16281_begin_0 = const()[name = tensor("op_16281_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_16281_end_0 = const()[name = tensor("op_16281_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_16281_end_mask_0 = const()[name = tensor("op_16281_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16281_cast_fp16 = slice_by_index(begin = var_16281_begin_0, end = var_16281_end_0, end_mask = var_16281_end_mask_0, x = v_75_cast_fp16)[name = tensor("op_16281_cast_fp16")]; + tensor var_16285_equation_0 = const()[name = tensor("op_16285_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16285_cast_fp16 = einsum(equation = var_16285_equation_0, values = (var_16127_cast_fp16, var_16044_cast_fp16))[name = tensor("op_16285_cast_fp16")]; + tensor var_16286_to_fp16 = const()[name = tensor("op_16286_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1321_cast_fp16 = mul(x = var_16285_cast_fp16, y = var_16286_to_fp16)[name = tensor("aw_1321_cast_fp16")]; + tensor var_16289_equation_0 = const()[name = tensor("op_16289_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16289_cast_fp16 = einsum(equation = var_16289_equation_0, values = (var_16131_cast_fp16, var_16048_cast_fp16))[name = tensor("op_16289_cast_fp16")]; + tensor var_16290_to_fp16 = const()[name = tensor("op_16290_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1323_cast_fp16 = mul(x = var_16289_cast_fp16, y = var_16290_to_fp16)[name = tensor("aw_1323_cast_fp16")]; + tensor var_16293_equation_0 = const()[name = tensor("op_16293_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16293_cast_fp16 = einsum(equation = var_16293_equation_0, values = (var_16135_cast_fp16, var_16052_cast_fp16))[name = tensor("op_16293_cast_fp16")]; + tensor var_16294_to_fp16 = const()[name = tensor("op_16294_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1325_cast_fp16 = mul(x = var_16293_cast_fp16, y = var_16294_to_fp16)[name = tensor("aw_1325_cast_fp16")]; + tensor var_16297_equation_0 = const()[name = tensor("op_16297_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16297_cast_fp16 = einsum(equation = var_16297_equation_0, values = (var_16139_cast_fp16, var_16056_cast_fp16))[name = tensor("op_16297_cast_fp16")]; + tensor var_16298_to_fp16 = const()[name = tensor("op_16298_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1327_cast_fp16 = mul(x = var_16297_cast_fp16, y = var_16298_to_fp16)[name = tensor("aw_1327_cast_fp16")]; + tensor var_16301_equation_0 = const()[name = tensor("op_16301_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16301_cast_fp16 = einsum(equation = var_16301_equation_0, values = (var_16143_cast_fp16, var_16060_cast_fp16))[name = tensor("op_16301_cast_fp16")]; + tensor var_16302_to_fp16 = const()[name = tensor("op_16302_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1329_cast_fp16 = mul(x = var_16301_cast_fp16, y = var_16302_to_fp16)[name = tensor("aw_1329_cast_fp16")]; + tensor var_16305_equation_0 = const()[name = tensor("op_16305_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16305_cast_fp16 = einsum(equation = var_16305_equation_0, values = (var_16147_cast_fp16, var_16064_cast_fp16))[name = tensor("op_16305_cast_fp16")]; + tensor var_16306_to_fp16 = const()[name = tensor("op_16306_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1331_cast_fp16 = mul(x = var_16305_cast_fp16, y = var_16306_to_fp16)[name = tensor("aw_1331_cast_fp16")]; + tensor var_16309_equation_0 = const()[name = tensor("op_16309_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16309_cast_fp16 = einsum(equation = var_16309_equation_0, values = (var_16151_cast_fp16, var_16068_cast_fp16))[name = tensor("op_16309_cast_fp16")]; + tensor var_16310_to_fp16 = const()[name = tensor("op_16310_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1333_cast_fp16 = mul(x = var_16309_cast_fp16, y = var_16310_to_fp16)[name = tensor("aw_1333_cast_fp16")]; + tensor var_16313_equation_0 = const()[name = tensor("op_16313_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16313_cast_fp16 = einsum(equation = var_16313_equation_0, values = (var_16155_cast_fp16, var_16072_cast_fp16))[name = tensor("op_16313_cast_fp16")]; + tensor var_16314_to_fp16 = const()[name = tensor("op_16314_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1335_cast_fp16 = mul(x = var_16313_cast_fp16, y = var_16314_to_fp16)[name = tensor("aw_1335_cast_fp16")]; + tensor var_16317_equation_0 = const()[name = tensor("op_16317_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16317_cast_fp16 = einsum(equation = var_16317_equation_0, values = (var_16159_cast_fp16, var_16076_cast_fp16))[name = tensor("op_16317_cast_fp16")]; + tensor var_16318_to_fp16 = const()[name = tensor("op_16318_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1337_cast_fp16 = mul(x = var_16317_cast_fp16, y = var_16318_to_fp16)[name = tensor("aw_1337_cast_fp16")]; + tensor var_16321_equation_0 = const()[name = tensor("op_16321_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16321_cast_fp16 = einsum(equation = var_16321_equation_0, values = (var_16163_cast_fp16, var_16080_cast_fp16))[name = tensor("op_16321_cast_fp16")]; + tensor var_16322_to_fp16 = const()[name = tensor("op_16322_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1339_cast_fp16 = mul(x = var_16321_cast_fp16, y = var_16322_to_fp16)[name = tensor("aw_1339_cast_fp16")]; + tensor var_16325_equation_0 = const()[name = tensor("op_16325_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16325_cast_fp16 = einsum(equation = var_16325_equation_0, values = (var_16167_cast_fp16, var_16084_cast_fp16))[name = tensor("op_16325_cast_fp16")]; + tensor var_16326_to_fp16 = const()[name = tensor("op_16326_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1341_cast_fp16 = mul(x = var_16325_cast_fp16, y = var_16326_to_fp16)[name = tensor("aw_1341_cast_fp16")]; + tensor var_16329_equation_0 = const()[name = tensor("op_16329_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16329_cast_fp16 = einsum(equation = var_16329_equation_0, values = (var_16171_cast_fp16, var_16088_cast_fp16))[name = tensor("op_16329_cast_fp16")]; + tensor var_16330_to_fp16 = const()[name = tensor("op_16330_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1343_cast_fp16 = mul(x = var_16329_cast_fp16, y = var_16330_to_fp16)[name = tensor("aw_1343_cast_fp16")]; + tensor var_16333_equation_0 = const()[name = tensor("op_16333_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16333_cast_fp16 = einsum(equation = var_16333_equation_0, values = (var_16175_cast_fp16, var_16092_cast_fp16))[name = tensor("op_16333_cast_fp16")]; + tensor var_16334_to_fp16 = const()[name = tensor("op_16334_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1345_cast_fp16 = mul(x = var_16333_cast_fp16, y = var_16334_to_fp16)[name = tensor("aw_1345_cast_fp16")]; + tensor var_16337_equation_0 = const()[name = tensor("op_16337_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16337_cast_fp16 = einsum(equation = var_16337_equation_0, values = (var_16179_cast_fp16, var_16096_cast_fp16))[name = tensor("op_16337_cast_fp16")]; + tensor var_16338_to_fp16 = const()[name = tensor("op_16338_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1347_cast_fp16 = mul(x = var_16337_cast_fp16, y = var_16338_to_fp16)[name = tensor("aw_1347_cast_fp16")]; + tensor var_16341_equation_0 = const()[name = tensor("op_16341_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16341_cast_fp16 = einsum(equation = var_16341_equation_0, values = (var_16183_cast_fp16, var_16100_cast_fp16))[name = tensor("op_16341_cast_fp16")]; + tensor var_16342_to_fp16 = const()[name = tensor("op_16342_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1349_cast_fp16 = mul(x = var_16341_cast_fp16, y = var_16342_to_fp16)[name = tensor("aw_1349_cast_fp16")]; + tensor var_16345_equation_0 = const()[name = tensor("op_16345_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16345_cast_fp16 = einsum(equation = var_16345_equation_0, values = (var_16187_cast_fp16, var_16104_cast_fp16))[name = tensor("op_16345_cast_fp16")]; + tensor var_16346_to_fp16 = const()[name = tensor("op_16346_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1351_cast_fp16 = mul(x = var_16345_cast_fp16, y = var_16346_to_fp16)[name = tensor("aw_1351_cast_fp16")]; + tensor var_16349_equation_0 = const()[name = tensor("op_16349_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16349_cast_fp16 = einsum(equation = var_16349_equation_0, values = (var_16191_cast_fp16, var_16108_cast_fp16))[name = tensor("op_16349_cast_fp16")]; + tensor var_16350_to_fp16 = const()[name = tensor("op_16350_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1353_cast_fp16 = mul(x = var_16349_cast_fp16, y = var_16350_to_fp16)[name = tensor("aw_1353_cast_fp16")]; + tensor var_16353_equation_0 = const()[name = tensor("op_16353_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16353_cast_fp16 = einsum(equation = var_16353_equation_0, values = (var_16195_cast_fp16, var_16112_cast_fp16))[name = tensor("op_16353_cast_fp16")]; + tensor var_16354_to_fp16 = const()[name = tensor("op_16354_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1355_cast_fp16 = mul(x = var_16353_cast_fp16, y = var_16354_to_fp16)[name = tensor("aw_1355_cast_fp16")]; + tensor var_16357_equation_0 = const()[name = tensor("op_16357_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16357_cast_fp16 = einsum(equation = var_16357_equation_0, values = (var_16199_cast_fp16, var_16116_cast_fp16))[name = tensor("op_16357_cast_fp16")]; + tensor var_16358_to_fp16 = const()[name = tensor("op_16358_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1357_cast_fp16 = mul(x = var_16357_cast_fp16, y = var_16358_to_fp16)[name = tensor("aw_1357_cast_fp16")]; + tensor var_16361_equation_0 = const()[name = tensor("op_16361_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16361_cast_fp16 = einsum(equation = var_16361_equation_0, values = (var_16203_cast_fp16, var_16120_cast_fp16))[name = tensor("op_16361_cast_fp16")]; + tensor var_16362_to_fp16 = const()[name = tensor("op_16362_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1359_cast_fp16 = mul(x = var_16361_cast_fp16, y = var_16362_to_fp16)[name = tensor("aw_1359_cast_fp16")]; + tensor var_16364_cast_fp16 = softmax(axis = var_2624, x = aw_1321_cast_fp16)[name = tensor("op_16364_cast_fp16")]; + tensor var_16365_cast_fp16 = softmax(axis = var_2624, x = aw_1323_cast_fp16)[name = tensor("op_16365_cast_fp16")]; + tensor var_16366_cast_fp16 = softmax(axis = var_2624, x = aw_1325_cast_fp16)[name = tensor("op_16366_cast_fp16")]; + tensor var_16367_cast_fp16 = softmax(axis = var_2624, x = aw_1327_cast_fp16)[name = tensor("op_16367_cast_fp16")]; + tensor var_16368_cast_fp16 = softmax(axis = var_2624, x = aw_1329_cast_fp16)[name = tensor("op_16368_cast_fp16")]; + tensor var_16369_cast_fp16 = softmax(axis = var_2624, x = aw_1331_cast_fp16)[name = tensor("op_16369_cast_fp16")]; + tensor var_16370_cast_fp16 = softmax(axis = var_2624, x = aw_1333_cast_fp16)[name = tensor("op_16370_cast_fp16")]; + tensor var_16371_cast_fp16 = softmax(axis = var_2624, x = aw_1335_cast_fp16)[name = tensor("op_16371_cast_fp16")]; + tensor var_16372_cast_fp16 = softmax(axis = var_2624, x = aw_1337_cast_fp16)[name = tensor("op_16372_cast_fp16")]; + tensor var_16373_cast_fp16 = softmax(axis = var_2624, x = aw_1339_cast_fp16)[name = tensor("op_16373_cast_fp16")]; + tensor var_16374_cast_fp16 = softmax(axis = var_2624, x = aw_1341_cast_fp16)[name = tensor("op_16374_cast_fp16")]; + tensor var_16375_cast_fp16 = softmax(axis = var_2624, x = aw_1343_cast_fp16)[name = tensor("op_16375_cast_fp16")]; + tensor var_16376_cast_fp16 = softmax(axis = var_2624, x = aw_1345_cast_fp16)[name = tensor("op_16376_cast_fp16")]; + tensor var_16377_cast_fp16 = softmax(axis = var_2624, x = aw_1347_cast_fp16)[name = tensor("op_16377_cast_fp16")]; + tensor var_16378_cast_fp16 = softmax(axis = var_2624, x = aw_1349_cast_fp16)[name = tensor("op_16378_cast_fp16")]; + tensor var_16379_cast_fp16 = softmax(axis = var_2624, x = aw_1351_cast_fp16)[name = tensor("op_16379_cast_fp16")]; + tensor var_16380_cast_fp16 = softmax(axis = var_2624, x = aw_1353_cast_fp16)[name = tensor("op_16380_cast_fp16")]; + tensor var_16381_cast_fp16 = softmax(axis = var_2624, x = aw_1355_cast_fp16)[name = tensor("op_16381_cast_fp16")]; + tensor var_16382_cast_fp16 = softmax(axis = var_2624, x = aw_1357_cast_fp16)[name = tensor("op_16382_cast_fp16")]; + tensor var_16383_cast_fp16 = softmax(axis = var_2624, x = aw_1359_cast_fp16)[name = tensor("op_16383_cast_fp16")]; + tensor var_16385_equation_0 = const()[name = tensor("op_16385_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16385_cast_fp16 = einsum(equation = var_16385_equation_0, values = (var_16205_cast_fp16, var_16364_cast_fp16))[name = tensor("op_16385_cast_fp16")]; + tensor var_16387_equation_0 = const()[name = tensor("op_16387_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16387_cast_fp16 = einsum(equation = var_16387_equation_0, values = (var_16209_cast_fp16, var_16365_cast_fp16))[name = tensor("op_16387_cast_fp16")]; + tensor var_16389_equation_0 = const()[name = tensor("op_16389_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16389_cast_fp16 = einsum(equation = var_16389_equation_0, values = (var_16213_cast_fp16, var_16366_cast_fp16))[name = tensor("op_16389_cast_fp16")]; + tensor var_16391_equation_0 = const()[name = tensor("op_16391_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16391_cast_fp16 = einsum(equation = var_16391_equation_0, values = (var_16217_cast_fp16, var_16367_cast_fp16))[name = tensor("op_16391_cast_fp16")]; + tensor var_16393_equation_0 = const()[name = tensor("op_16393_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16393_cast_fp16 = einsum(equation = var_16393_equation_0, values = (var_16221_cast_fp16, var_16368_cast_fp16))[name = tensor("op_16393_cast_fp16")]; + tensor var_16395_equation_0 = const()[name = tensor("op_16395_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16395_cast_fp16 = einsum(equation = var_16395_equation_0, values = (var_16225_cast_fp16, var_16369_cast_fp16))[name = tensor("op_16395_cast_fp16")]; + tensor var_16397_equation_0 = const()[name = tensor("op_16397_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16397_cast_fp16 = einsum(equation = var_16397_equation_0, values = (var_16229_cast_fp16, var_16370_cast_fp16))[name = tensor("op_16397_cast_fp16")]; + tensor var_16399_equation_0 = const()[name = tensor("op_16399_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16399_cast_fp16 = einsum(equation = var_16399_equation_0, values = (var_16233_cast_fp16, var_16371_cast_fp16))[name = tensor("op_16399_cast_fp16")]; + tensor var_16401_equation_0 = const()[name = tensor("op_16401_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16401_cast_fp16 = einsum(equation = var_16401_equation_0, values = (var_16237_cast_fp16, var_16372_cast_fp16))[name = tensor("op_16401_cast_fp16")]; + tensor var_16403_equation_0 = const()[name = tensor("op_16403_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16403_cast_fp16 = einsum(equation = var_16403_equation_0, values = (var_16241_cast_fp16, var_16373_cast_fp16))[name = tensor("op_16403_cast_fp16")]; + tensor var_16405_equation_0 = const()[name = tensor("op_16405_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16405_cast_fp16 = einsum(equation = var_16405_equation_0, values = (var_16245_cast_fp16, var_16374_cast_fp16))[name = tensor("op_16405_cast_fp16")]; + tensor var_16407_equation_0 = const()[name = tensor("op_16407_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16407_cast_fp16 = einsum(equation = var_16407_equation_0, values = (var_16249_cast_fp16, var_16375_cast_fp16))[name = tensor("op_16407_cast_fp16")]; + tensor var_16409_equation_0 = const()[name = tensor("op_16409_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16409_cast_fp16 = einsum(equation = var_16409_equation_0, values = (var_16253_cast_fp16, var_16376_cast_fp16))[name = tensor("op_16409_cast_fp16")]; + tensor var_16411_equation_0 = const()[name = tensor("op_16411_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16411_cast_fp16 = einsum(equation = var_16411_equation_0, values = (var_16257_cast_fp16, var_16377_cast_fp16))[name = tensor("op_16411_cast_fp16")]; + tensor var_16413_equation_0 = const()[name = tensor("op_16413_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16413_cast_fp16 = einsum(equation = var_16413_equation_0, values = (var_16261_cast_fp16, var_16378_cast_fp16))[name = tensor("op_16413_cast_fp16")]; + tensor var_16415_equation_0 = const()[name = tensor("op_16415_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16415_cast_fp16 = einsum(equation = var_16415_equation_0, values = (var_16265_cast_fp16, var_16379_cast_fp16))[name = tensor("op_16415_cast_fp16")]; + tensor var_16417_equation_0 = const()[name = tensor("op_16417_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16417_cast_fp16 = einsum(equation = var_16417_equation_0, values = (var_16269_cast_fp16, var_16380_cast_fp16))[name = tensor("op_16417_cast_fp16")]; + tensor var_16419_equation_0 = const()[name = tensor("op_16419_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16419_cast_fp16 = einsum(equation = var_16419_equation_0, values = (var_16273_cast_fp16, var_16381_cast_fp16))[name = tensor("op_16419_cast_fp16")]; + tensor var_16421_equation_0 = const()[name = tensor("op_16421_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16421_cast_fp16 = einsum(equation = var_16421_equation_0, values = (var_16277_cast_fp16, var_16382_cast_fp16))[name = tensor("op_16421_cast_fp16")]; + tensor var_16423_equation_0 = const()[name = tensor("op_16423_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16423_cast_fp16 = einsum(equation = var_16423_equation_0, values = (var_16281_cast_fp16, var_16383_cast_fp16))[name = tensor("op_16423_cast_fp16")]; + tensor input_263_interleave_0 = const()[name = tensor("input_263_interleave_0"), val = tensor(false)]; + tensor input_263_cast_fp16 = concat(axis = var_2624, interleave = input_263_interleave_0, values = (var_16385_cast_fp16, var_16387_cast_fp16, var_16389_cast_fp16, var_16391_cast_fp16, var_16393_cast_fp16, var_16395_cast_fp16, var_16397_cast_fp16, var_16399_cast_fp16, var_16401_cast_fp16, var_16403_cast_fp16, var_16405_cast_fp16, var_16407_cast_fp16, var_16409_cast_fp16, var_16411_cast_fp16, var_16413_cast_fp16, var_16415_cast_fp16, var_16417_cast_fp16, var_16419_cast_fp16, var_16421_cast_fp16, var_16423_cast_fp16))[name = tensor("input_263_cast_fp16")]; + tensor var_16429 = const()[name = tensor("op_16429"), val = tensor([1, 1])]; + tensor var_16431 = const()[name = tensor("op_16431"), val = tensor([1, 1])]; + tensor var_16433_pad_type_0 = const()[name = tensor("op_16433_pad_type_0"), val = tensor("custom")]; + tensor var_16433_pad_0 = const()[name = tensor("op_16433_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_4_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_4_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1267072640)))]; + tensor down_blocks_2_attentions_1_transformer_blocks_4_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_4_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1270349504)))]; + tensor var_16433_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_4_attn2_to_out_0_bias_to_fp16, dilations = var_16431, groups = var_2624, pad = var_16433_pad_0, pad_type = var_16433_pad_type_0, strides = var_16429, weight = down_blocks_2_attentions_1_transformer_blocks_4_attn2_to_out_0_weight_to_fp16, x = input_263_cast_fp16)[name = tensor("op_16433_cast_fp16")]; + tensor inputs_113_cast_fp16 = add(x = var_16433_cast_fp16, y = inputs_111_cast_fp16)[name = tensor("inputs_113_cast_fp16")]; + tensor input_265_axes_0 = const()[name = tensor("input_265_axes_0"), val = tensor([1])]; + tensor input_265_gamma_0_to_fp16 = const()[name = tensor("input_265_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1270352128)))]; + tensor input_265_beta_0_to_fp16 = const()[name = tensor("input_265_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1270354752)))]; + tensor var_16443_to_fp16 = const()[name = tensor("op_16443_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_265_cast_fp16 = layer_norm(axes = input_265_axes_0, beta = input_265_beta_0_to_fp16, epsilon = var_16443_to_fp16, gamma = input_265_gamma_0_to_fp16, x = inputs_113_cast_fp16)[name = tensor("input_265_cast_fp16")]; + tensor var_16459 = const()[name = tensor("op_16459"), val = tensor([1, 1])]; + tensor var_16461 = const()[name = tensor("op_16461"), val = tensor([1, 1])]; + tensor var_16463_pad_type_0 = const()[name = tensor("op_16463_pad_type_0"), val = tensor("custom")]; + tensor var_16463_pad_0 = const()[name = tensor("op_16463_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_4_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_4_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1270357376)))]; + tensor down_blocks_2_attentions_1_transformer_blocks_4_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_4_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1296571840)))]; + tensor var_16463_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_4_ff_net_0_proj_bias_to_fp16, dilations = var_16461, groups = var_2624, pad = var_16463_pad_0, pad_type = var_16463_pad_type_0, strides = var_16459, weight = down_blocks_2_attentions_1_transformer_blocks_4_ff_net_0_proj_weight_to_fp16, x = input_265_cast_fp16)[name = tensor("op_16463_cast_fp16")]; + tensor var_16464_split_sizes_0 = const()[name = tensor("op_16464_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_16464_axis_0 = const()[name = tensor("op_16464_axis_0"), val = tensor(1)]; + tensor var_16464_cast_fp16_0, tensor var_16464_cast_fp16_1 = split(axis = var_16464_axis_0, split_sizes = var_16464_split_sizes_0, x = var_16463_cast_fp16)[name = tensor("op_16464_cast_fp16")]; + tensor var_16466_mode_0 = const()[name = tensor("op_16466_mode_0"), val = tensor("EXACT")]; + tensor var_16466_cast_fp16 = gelu(mode = var_16466_mode_0, x = var_16464_cast_fp16_1)[name = tensor("op_16466_cast_fp16")]; + tensor input_267_cast_fp16 = mul(x = var_16464_cast_fp16_0, y = var_16466_cast_fp16)[name = tensor("input_267_cast_fp16")]; + tensor var_16470 = const()[name = tensor("op_16470"), val = tensor([1, 1])]; + tensor var_16472 = const()[name = tensor("op_16472"), val = tensor([1, 1])]; + tensor var_16474_pad_type_0 = const()[name = tensor("op_16474_pad_type_0"), val = tensor("custom")]; + tensor var_16474_pad_0 = const()[name = tensor("op_16474_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_4_ff_net_2_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_4_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1296592384)))]; + tensor down_blocks_2_attentions_1_transformer_blocks_4_ff_net_2_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_4_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1309699648)))]; + tensor var_16474_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_4_ff_net_2_bias_to_fp16, dilations = var_16472, groups = var_2624, pad = var_16474_pad_0, pad_type = var_16474_pad_type_0, strides = var_16470, weight = down_blocks_2_attentions_1_transformer_blocks_4_ff_net_2_weight_to_fp16, x = input_267_cast_fp16)[name = tensor("op_16474_cast_fp16")]; + tensor inputs_115_cast_fp16 = add(x = var_16474_cast_fp16, y = inputs_113_cast_fp16)[name = tensor("inputs_115_cast_fp16")]; + tensor hidden_states_167_axes_0 = const()[name = tensor("hidden_states_167_axes_0"), val = tensor([1])]; + tensor hidden_states_167_gamma_0_to_fp16 = const()[name = tensor("hidden_states_167_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1309702272)))]; + tensor hidden_states_167_beta_0_to_fp16 = const()[name = tensor("hidden_states_167_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1309704896)))]; + tensor var_16490_to_fp16 = const()[name = tensor("op_16490_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_167_cast_fp16 = layer_norm(axes = hidden_states_167_axes_0, beta = hidden_states_167_beta_0_to_fp16, epsilon = var_16490_to_fp16, gamma = hidden_states_167_gamma_0_to_fp16, x = inputs_115_cast_fp16)[name = tensor("hidden_states_167_cast_fp16")]; + tensor var_16505 = const()[name = tensor("op_16505"), val = tensor([1, 1])]; + tensor var_16507 = const()[name = tensor("op_16507"), val = tensor([1, 1])]; + tensor q_77_pad_type_0 = const()[name = tensor("q_77_pad_type_0"), val = tensor("custom")]; + tensor q_77_pad_0 = const()[name = tensor("q_77_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_5_attn1_to_q_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_5_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1309707520)))]; + tensor q_77_cast_fp16 = conv(dilations = var_16507, groups = var_2624, pad = q_77_pad_0, pad_type = q_77_pad_type_0, strides = var_16505, weight = down_blocks_2_attentions_1_transformer_blocks_5_attn1_to_q_weight_to_fp16, x = hidden_states_167_cast_fp16)[name = tensor("q_77_cast_fp16")]; + tensor var_16511 = const()[name = tensor("op_16511"), val = tensor([1, 1])]; + tensor var_16513 = const()[name = tensor("op_16513"), val = tensor([1, 1])]; + tensor k_153_pad_type_0 = const()[name = tensor("k_153_pad_type_0"), val = tensor("custom")]; + tensor k_153_pad_0 = const()[name = tensor("k_153_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_5_attn1_to_k_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_5_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1312984384)))]; + tensor k_153_cast_fp16 = conv(dilations = var_16513, groups = var_2624, pad = k_153_pad_0, pad_type = k_153_pad_type_0, strides = var_16511, weight = down_blocks_2_attentions_1_transformer_blocks_5_attn1_to_k_weight_to_fp16, x = hidden_states_167_cast_fp16)[name = tensor("k_153_cast_fp16")]; + tensor var_16517 = const()[name = tensor("op_16517"), val = tensor([1, 1])]; + tensor var_16519 = const()[name = tensor("op_16519"), val = tensor([1, 1])]; + tensor v_77_pad_type_0 = const()[name = tensor("v_77_pad_type_0"), val = tensor("custom")]; + tensor v_77_pad_0 = const()[name = tensor("v_77_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_5_attn1_to_v_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_5_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1316261248)))]; + tensor v_77_cast_fp16 = conv(dilations = var_16519, groups = var_2624, pad = v_77_pad_0, pad_type = v_77_pad_type_0, strides = var_16517, weight = down_blocks_2_attentions_1_transformer_blocks_5_attn1_to_v_weight_to_fp16, x = hidden_states_167_cast_fp16)[name = tensor("v_77_cast_fp16")]; + tensor var_16523_begin_0 = const()[name = tensor("op_16523_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_16523_end_0 = const()[name = tensor("op_16523_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_16523_end_mask_0 = const()[name = tensor("op_16523_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16523_cast_fp16 = slice_by_index(begin = var_16523_begin_0, end = var_16523_end_0, end_mask = var_16523_end_mask_0, x = q_77_cast_fp16)[name = tensor("op_16523_cast_fp16")]; + tensor var_16527_begin_0 = const()[name = tensor("op_16527_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_16527_end_0 = const()[name = tensor("op_16527_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_16527_end_mask_0 = const()[name = tensor("op_16527_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16527_cast_fp16 = slice_by_index(begin = var_16527_begin_0, end = var_16527_end_0, end_mask = var_16527_end_mask_0, x = q_77_cast_fp16)[name = tensor("op_16527_cast_fp16")]; + tensor var_16531_begin_0 = const()[name = tensor("op_16531_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_16531_end_0 = const()[name = tensor("op_16531_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_16531_end_mask_0 = const()[name = tensor("op_16531_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16531_cast_fp16 = slice_by_index(begin = var_16531_begin_0, end = var_16531_end_0, end_mask = var_16531_end_mask_0, x = q_77_cast_fp16)[name = tensor("op_16531_cast_fp16")]; + tensor var_16535_begin_0 = const()[name = tensor("op_16535_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_16535_end_0 = const()[name = tensor("op_16535_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_16535_end_mask_0 = const()[name = tensor("op_16535_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16535_cast_fp16 = slice_by_index(begin = var_16535_begin_0, end = var_16535_end_0, end_mask = var_16535_end_mask_0, x = q_77_cast_fp16)[name = tensor("op_16535_cast_fp16")]; + tensor var_16539_begin_0 = const()[name = tensor("op_16539_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_16539_end_0 = const()[name = tensor("op_16539_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_16539_end_mask_0 = const()[name = tensor("op_16539_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16539_cast_fp16 = slice_by_index(begin = var_16539_begin_0, end = var_16539_end_0, end_mask = var_16539_end_mask_0, x = q_77_cast_fp16)[name = tensor("op_16539_cast_fp16")]; + tensor var_16543_begin_0 = const()[name = tensor("op_16543_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_16543_end_0 = const()[name = tensor("op_16543_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_16543_end_mask_0 = const()[name = tensor("op_16543_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16543_cast_fp16 = slice_by_index(begin = var_16543_begin_0, end = var_16543_end_0, end_mask = var_16543_end_mask_0, x = q_77_cast_fp16)[name = tensor("op_16543_cast_fp16")]; + tensor var_16547_begin_0 = const()[name = tensor("op_16547_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_16547_end_0 = const()[name = tensor("op_16547_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_16547_end_mask_0 = const()[name = tensor("op_16547_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16547_cast_fp16 = slice_by_index(begin = var_16547_begin_0, end = var_16547_end_0, end_mask = var_16547_end_mask_0, x = q_77_cast_fp16)[name = tensor("op_16547_cast_fp16")]; + tensor var_16551_begin_0 = const()[name = tensor("op_16551_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_16551_end_0 = const()[name = tensor("op_16551_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_16551_end_mask_0 = const()[name = tensor("op_16551_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16551_cast_fp16 = slice_by_index(begin = var_16551_begin_0, end = var_16551_end_0, end_mask = var_16551_end_mask_0, x = q_77_cast_fp16)[name = tensor("op_16551_cast_fp16")]; + tensor var_16555_begin_0 = const()[name = tensor("op_16555_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_16555_end_0 = const()[name = tensor("op_16555_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_16555_end_mask_0 = const()[name = tensor("op_16555_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16555_cast_fp16 = slice_by_index(begin = var_16555_begin_0, end = var_16555_end_0, end_mask = var_16555_end_mask_0, x = q_77_cast_fp16)[name = tensor("op_16555_cast_fp16")]; + tensor var_16559_begin_0 = const()[name = tensor("op_16559_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_16559_end_0 = const()[name = tensor("op_16559_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_16559_end_mask_0 = const()[name = tensor("op_16559_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16559_cast_fp16 = slice_by_index(begin = var_16559_begin_0, end = var_16559_end_0, end_mask = var_16559_end_mask_0, x = q_77_cast_fp16)[name = tensor("op_16559_cast_fp16")]; + tensor var_16563_begin_0 = const()[name = tensor("op_16563_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_16563_end_0 = const()[name = tensor("op_16563_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_16563_end_mask_0 = const()[name = tensor("op_16563_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16563_cast_fp16 = slice_by_index(begin = var_16563_begin_0, end = var_16563_end_0, end_mask = var_16563_end_mask_0, x = q_77_cast_fp16)[name = tensor("op_16563_cast_fp16")]; + tensor var_16567_begin_0 = const()[name = tensor("op_16567_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_16567_end_0 = const()[name = tensor("op_16567_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_16567_end_mask_0 = const()[name = tensor("op_16567_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16567_cast_fp16 = slice_by_index(begin = var_16567_begin_0, end = var_16567_end_0, end_mask = var_16567_end_mask_0, x = q_77_cast_fp16)[name = tensor("op_16567_cast_fp16")]; + tensor var_16571_begin_0 = const()[name = tensor("op_16571_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_16571_end_0 = const()[name = tensor("op_16571_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_16571_end_mask_0 = const()[name = tensor("op_16571_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16571_cast_fp16 = slice_by_index(begin = var_16571_begin_0, end = var_16571_end_0, end_mask = var_16571_end_mask_0, x = q_77_cast_fp16)[name = tensor("op_16571_cast_fp16")]; + tensor var_16575_begin_0 = const()[name = tensor("op_16575_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_16575_end_0 = const()[name = tensor("op_16575_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_16575_end_mask_0 = const()[name = tensor("op_16575_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16575_cast_fp16 = slice_by_index(begin = var_16575_begin_0, end = var_16575_end_0, end_mask = var_16575_end_mask_0, x = q_77_cast_fp16)[name = tensor("op_16575_cast_fp16")]; + tensor var_16579_begin_0 = const()[name = tensor("op_16579_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_16579_end_0 = const()[name = tensor("op_16579_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_16579_end_mask_0 = const()[name = tensor("op_16579_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16579_cast_fp16 = slice_by_index(begin = var_16579_begin_0, end = var_16579_end_0, end_mask = var_16579_end_mask_0, x = q_77_cast_fp16)[name = tensor("op_16579_cast_fp16")]; + tensor var_16583_begin_0 = const()[name = tensor("op_16583_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_16583_end_0 = const()[name = tensor("op_16583_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_16583_end_mask_0 = const()[name = tensor("op_16583_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16583_cast_fp16 = slice_by_index(begin = var_16583_begin_0, end = var_16583_end_0, end_mask = var_16583_end_mask_0, x = q_77_cast_fp16)[name = tensor("op_16583_cast_fp16")]; + tensor var_16587_begin_0 = const()[name = tensor("op_16587_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_16587_end_0 = const()[name = tensor("op_16587_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_16587_end_mask_0 = const()[name = tensor("op_16587_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16587_cast_fp16 = slice_by_index(begin = var_16587_begin_0, end = var_16587_end_0, end_mask = var_16587_end_mask_0, x = q_77_cast_fp16)[name = tensor("op_16587_cast_fp16")]; + tensor var_16591_begin_0 = const()[name = tensor("op_16591_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_16591_end_0 = const()[name = tensor("op_16591_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_16591_end_mask_0 = const()[name = tensor("op_16591_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16591_cast_fp16 = slice_by_index(begin = var_16591_begin_0, end = var_16591_end_0, end_mask = var_16591_end_mask_0, x = q_77_cast_fp16)[name = tensor("op_16591_cast_fp16")]; + tensor var_16595_begin_0 = const()[name = tensor("op_16595_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_16595_end_0 = const()[name = tensor("op_16595_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_16595_end_mask_0 = const()[name = tensor("op_16595_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16595_cast_fp16 = slice_by_index(begin = var_16595_begin_0, end = var_16595_end_0, end_mask = var_16595_end_mask_0, x = q_77_cast_fp16)[name = tensor("op_16595_cast_fp16")]; + tensor var_16599_begin_0 = const()[name = tensor("op_16599_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_16599_end_0 = const()[name = tensor("op_16599_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_16599_end_mask_0 = const()[name = tensor("op_16599_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16599_cast_fp16 = slice_by_index(begin = var_16599_begin_0, end = var_16599_end_0, end_mask = var_16599_end_mask_0, x = q_77_cast_fp16)[name = tensor("op_16599_cast_fp16")]; + tensor k_155_perm_0 = const()[name = tensor("k_155_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_16606_begin_0 = const()[name = tensor("op_16606_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_16606_end_0 = const()[name = tensor("op_16606_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_16606_end_mask_0 = const()[name = tensor("op_16606_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_155_cast_fp16 = transpose(perm = k_155_perm_0, x = k_153_cast_fp16)[name = tensor("transpose_101")]; + tensor var_16606_cast_fp16 = slice_by_index(begin = var_16606_begin_0, end = var_16606_end_0, end_mask = var_16606_end_mask_0, x = k_155_cast_fp16)[name = tensor("op_16606_cast_fp16")]; + tensor var_16610_begin_0 = const()[name = tensor("op_16610_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_16610_end_0 = const()[name = tensor("op_16610_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_16610_end_mask_0 = const()[name = tensor("op_16610_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16610_cast_fp16 = slice_by_index(begin = var_16610_begin_0, end = var_16610_end_0, end_mask = var_16610_end_mask_0, x = k_155_cast_fp16)[name = tensor("op_16610_cast_fp16")]; + tensor var_16614_begin_0 = const()[name = tensor("op_16614_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_16614_end_0 = const()[name = tensor("op_16614_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_16614_end_mask_0 = const()[name = tensor("op_16614_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16614_cast_fp16 = slice_by_index(begin = var_16614_begin_0, end = var_16614_end_0, end_mask = var_16614_end_mask_0, x = k_155_cast_fp16)[name = tensor("op_16614_cast_fp16")]; + tensor var_16618_begin_0 = const()[name = tensor("op_16618_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_16618_end_0 = const()[name = tensor("op_16618_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_16618_end_mask_0 = const()[name = tensor("op_16618_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16618_cast_fp16 = slice_by_index(begin = var_16618_begin_0, end = var_16618_end_0, end_mask = var_16618_end_mask_0, x = k_155_cast_fp16)[name = tensor("op_16618_cast_fp16")]; + tensor var_16622_begin_0 = const()[name = tensor("op_16622_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_16622_end_0 = const()[name = tensor("op_16622_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_16622_end_mask_0 = const()[name = tensor("op_16622_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16622_cast_fp16 = slice_by_index(begin = var_16622_begin_0, end = var_16622_end_0, end_mask = var_16622_end_mask_0, x = k_155_cast_fp16)[name = tensor("op_16622_cast_fp16")]; + tensor var_16626_begin_0 = const()[name = tensor("op_16626_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_16626_end_0 = const()[name = tensor("op_16626_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_16626_end_mask_0 = const()[name = tensor("op_16626_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16626_cast_fp16 = slice_by_index(begin = var_16626_begin_0, end = var_16626_end_0, end_mask = var_16626_end_mask_0, x = k_155_cast_fp16)[name = tensor("op_16626_cast_fp16")]; + tensor var_16630_begin_0 = const()[name = tensor("op_16630_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_16630_end_0 = const()[name = tensor("op_16630_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_16630_end_mask_0 = const()[name = tensor("op_16630_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16630_cast_fp16 = slice_by_index(begin = var_16630_begin_0, end = var_16630_end_0, end_mask = var_16630_end_mask_0, x = k_155_cast_fp16)[name = tensor("op_16630_cast_fp16")]; + tensor var_16634_begin_0 = const()[name = tensor("op_16634_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_16634_end_0 = const()[name = tensor("op_16634_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_16634_end_mask_0 = const()[name = tensor("op_16634_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16634_cast_fp16 = slice_by_index(begin = var_16634_begin_0, end = var_16634_end_0, end_mask = var_16634_end_mask_0, x = k_155_cast_fp16)[name = tensor("op_16634_cast_fp16")]; + tensor var_16638_begin_0 = const()[name = tensor("op_16638_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_16638_end_0 = const()[name = tensor("op_16638_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_16638_end_mask_0 = const()[name = tensor("op_16638_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16638_cast_fp16 = slice_by_index(begin = var_16638_begin_0, end = var_16638_end_0, end_mask = var_16638_end_mask_0, x = k_155_cast_fp16)[name = tensor("op_16638_cast_fp16")]; + tensor var_16642_begin_0 = const()[name = tensor("op_16642_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_16642_end_0 = const()[name = tensor("op_16642_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_16642_end_mask_0 = const()[name = tensor("op_16642_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16642_cast_fp16 = slice_by_index(begin = var_16642_begin_0, end = var_16642_end_0, end_mask = var_16642_end_mask_0, x = k_155_cast_fp16)[name = tensor("op_16642_cast_fp16")]; + tensor var_16646_begin_0 = const()[name = tensor("op_16646_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_16646_end_0 = const()[name = tensor("op_16646_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_16646_end_mask_0 = const()[name = tensor("op_16646_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16646_cast_fp16 = slice_by_index(begin = var_16646_begin_0, end = var_16646_end_0, end_mask = var_16646_end_mask_0, x = k_155_cast_fp16)[name = tensor("op_16646_cast_fp16")]; + tensor var_16650_begin_0 = const()[name = tensor("op_16650_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_16650_end_0 = const()[name = tensor("op_16650_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_16650_end_mask_0 = const()[name = tensor("op_16650_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16650_cast_fp16 = slice_by_index(begin = var_16650_begin_0, end = var_16650_end_0, end_mask = var_16650_end_mask_0, x = k_155_cast_fp16)[name = tensor("op_16650_cast_fp16")]; + tensor var_16654_begin_0 = const()[name = tensor("op_16654_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_16654_end_0 = const()[name = tensor("op_16654_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_16654_end_mask_0 = const()[name = tensor("op_16654_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16654_cast_fp16 = slice_by_index(begin = var_16654_begin_0, end = var_16654_end_0, end_mask = var_16654_end_mask_0, x = k_155_cast_fp16)[name = tensor("op_16654_cast_fp16")]; + tensor var_16658_begin_0 = const()[name = tensor("op_16658_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_16658_end_0 = const()[name = tensor("op_16658_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_16658_end_mask_0 = const()[name = tensor("op_16658_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16658_cast_fp16 = slice_by_index(begin = var_16658_begin_0, end = var_16658_end_0, end_mask = var_16658_end_mask_0, x = k_155_cast_fp16)[name = tensor("op_16658_cast_fp16")]; + tensor var_16662_begin_0 = const()[name = tensor("op_16662_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_16662_end_0 = const()[name = tensor("op_16662_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_16662_end_mask_0 = const()[name = tensor("op_16662_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16662_cast_fp16 = slice_by_index(begin = var_16662_begin_0, end = var_16662_end_0, end_mask = var_16662_end_mask_0, x = k_155_cast_fp16)[name = tensor("op_16662_cast_fp16")]; + tensor var_16666_begin_0 = const()[name = tensor("op_16666_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_16666_end_0 = const()[name = tensor("op_16666_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_16666_end_mask_0 = const()[name = tensor("op_16666_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16666_cast_fp16 = slice_by_index(begin = var_16666_begin_0, end = var_16666_end_0, end_mask = var_16666_end_mask_0, x = k_155_cast_fp16)[name = tensor("op_16666_cast_fp16")]; + tensor var_16670_begin_0 = const()[name = tensor("op_16670_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_16670_end_0 = const()[name = tensor("op_16670_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_16670_end_mask_0 = const()[name = tensor("op_16670_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16670_cast_fp16 = slice_by_index(begin = var_16670_begin_0, end = var_16670_end_0, end_mask = var_16670_end_mask_0, x = k_155_cast_fp16)[name = tensor("op_16670_cast_fp16")]; + tensor var_16674_begin_0 = const()[name = tensor("op_16674_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_16674_end_0 = const()[name = tensor("op_16674_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_16674_end_mask_0 = const()[name = tensor("op_16674_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16674_cast_fp16 = slice_by_index(begin = var_16674_begin_0, end = var_16674_end_0, end_mask = var_16674_end_mask_0, x = k_155_cast_fp16)[name = tensor("op_16674_cast_fp16")]; + tensor var_16678_begin_0 = const()[name = tensor("op_16678_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_16678_end_0 = const()[name = tensor("op_16678_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_16678_end_mask_0 = const()[name = tensor("op_16678_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16678_cast_fp16 = slice_by_index(begin = var_16678_begin_0, end = var_16678_end_0, end_mask = var_16678_end_mask_0, x = k_155_cast_fp16)[name = tensor("op_16678_cast_fp16")]; + tensor var_16682_begin_0 = const()[name = tensor("op_16682_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_16682_end_0 = const()[name = tensor("op_16682_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_16682_end_mask_0 = const()[name = tensor("op_16682_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16682_cast_fp16 = slice_by_index(begin = var_16682_begin_0, end = var_16682_end_0, end_mask = var_16682_end_mask_0, x = k_155_cast_fp16)[name = tensor("op_16682_cast_fp16")]; + tensor var_16684_begin_0 = const()[name = tensor("op_16684_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_16684_end_0 = const()[name = tensor("op_16684_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_16684_end_mask_0 = const()[name = tensor("op_16684_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16684_cast_fp16 = slice_by_index(begin = var_16684_begin_0, end = var_16684_end_0, end_mask = var_16684_end_mask_0, x = v_77_cast_fp16)[name = tensor("op_16684_cast_fp16")]; + tensor var_16688_begin_0 = const()[name = tensor("op_16688_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_16688_end_0 = const()[name = tensor("op_16688_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_16688_end_mask_0 = const()[name = tensor("op_16688_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16688_cast_fp16 = slice_by_index(begin = var_16688_begin_0, end = var_16688_end_0, end_mask = var_16688_end_mask_0, x = v_77_cast_fp16)[name = tensor("op_16688_cast_fp16")]; + tensor var_16692_begin_0 = const()[name = tensor("op_16692_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_16692_end_0 = const()[name = tensor("op_16692_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_16692_end_mask_0 = const()[name = tensor("op_16692_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16692_cast_fp16 = slice_by_index(begin = var_16692_begin_0, end = var_16692_end_0, end_mask = var_16692_end_mask_0, x = v_77_cast_fp16)[name = tensor("op_16692_cast_fp16")]; + tensor var_16696_begin_0 = const()[name = tensor("op_16696_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_16696_end_0 = const()[name = tensor("op_16696_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_16696_end_mask_0 = const()[name = tensor("op_16696_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16696_cast_fp16 = slice_by_index(begin = var_16696_begin_0, end = var_16696_end_0, end_mask = var_16696_end_mask_0, x = v_77_cast_fp16)[name = tensor("op_16696_cast_fp16")]; + tensor var_16700_begin_0 = const()[name = tensor("op_16700_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_16700_end_0 = const()[name = tensor("op_16700_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_16700_end_mask_0 = const()[name = tensor("op_16700_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16700_cast_fp16 = slice_by_index(begin = var_16700_begin_0, end = var_16700_end_0, end_mask = var_16700_end_mask_0, x = v_77_cast_fp16)[name = tensor("op_16700_cast_fp16")]; + tensor var_16704_begin_0 = const()[name = tensor("op_16704_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_16704_end_0 = const()[name = tensor("op_16704_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_16704_end_mask_0 = const()[name = tensor("op_16704_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16704_cast_fp16 = slice_by_index(begin = var_16704_begin_0, end = var_16704_end_0, end_mask = var_16704_end_mask_0, x = v_77_cast_fp16)[name = tensor("op_16704_cast_fp16")]; + tensor var_16708_begin_0 = const()[name = tensor("op_16708_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_16708_end_0 = const()[name = tensor("op_16708_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_16708_end_mask_0 = const()[name = tensor("op_16708_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16708_cast_fp16 = slice_by_index(begin = var_16708_begin_0, end = var_16708_end_0, end_mask = var_16708_end_mask_0, x = v_77_cast_fp16)[name = tensor("op_16708_cast_fp16")]; + tensor var_16712_begin_0 = const()[name = tensor("op_16712_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_16712_end_0 = const()[name = tensor("op_16712_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_16712_end_mask_0 = const()[name = tensor("op_16712_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16712_cast_fp16 = slice_by_index(begin = var_16712_begin_0, end = var_16712_end_0, end_mask = var_16712_end_mask_0, x = v_77_cast_fp16)[name = tensor("op_16712_cast_fp16")]; + tensor var_16716_begin_0 = const()[name = tensor("op_16716_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_16716_end_0 = const()[name = tensor("op_16716_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_16716_end_mask_0 = const()[name = tensor("op_16716_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16716_cast_fp16 = slice_by_index(begin = var_16716_begin_0, end = var_16716_end_0, end_mask = var_16716_end_mask_0, x = v_77_cast_fp16)[name = tensor("op_16716_cast_fp16")]; + tensor var_16720_begin_0 = const()[name = tensor("op_16720_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_16720_end_0 = const()[name = tensor("op_16720_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_16720_end_mask_0 = const()[name = tensor("op_16720_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16720_cast_fp16 = slice_by_index(begin = var_16720_begin_0, end = var_16720_end_0, end_mask = var_16720_end_mask_0, x = v_77_cast_fp16)[name = tensor("op_16720_cast_fp16")]; + tensor var_16724_begin_0 = const()[name = tensor("op_16724_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_16724_end_0 = const()[name = tensor("op_16724_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_16724_end_mask_0 = const()[name = tensor("op_16724_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16724_cast_fp16 = slice_by_index(begin = var_16724_begin_0, end = var_16724_end_0, end_mask = var_16724_end_mask_0, x = v_77_cast_fp16)[name = tensor("op_16724_cast_fp16")]; + tensor var_16728_begin_0 = const()[name = tensor("op_16728_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_16728_end_0 = const()[name = tensor("op_16728_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_16728_end_mask_0 = const()[name = tensor("op_16728_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16728_cast_fp16 = slice_by_index(begin = var_16728_begin_0, end = var_16728_end_0, end_mask = var_16728_end_mask_0, x = v_77_cast_fp16)[name = tensor("op_16728_cast_fp16")]; + tensor var_16732_begin_0 = const()[name = tensor("op_16732_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_16732_end_0 = const()[name = tensor("op_16732_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_16732_end_mask_0 = const()[name = tensor("op_16732_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16732_cast_fp16 = slice_by_index(begin = var_16732_begin_0, end = var_16732_end_0, end_mask = var_16732_end_mask_0, x = v_77_cast_fp16)[name = tensor("op_16732_cast_fp16")]; + tensor var_16736_begin_0 = const()[name = tensor("op_16736_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_16736_end_0 = const()[name = tensor("op_16736_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_16736_end_mask_0 = const()[name = tensor("op_16736_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16736_cast_fp16 = slice_by_index(begin = var_16736_begin_0, end = var_16736_end_0, end_mask = var_16736_end_mask_0, x = v_77_cast_fp16)[name = tensor("op_16736_cast_fp16")]; + tensor var_16740_begin_0 = const()[name = tensor("op_16740_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_16740_end_0 = const()[name = tensor("op_16740_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_16740_end_mask_0 = const()[name = tensor("op_16740_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16740_cast_fp16 = slice_by_index(begin = var_16740_begin_0, end = var_16740_end_0, end_mask = var_16740_end_mask_0, x = v_77_cast_fp16)[name = tensor("op_16740_cast_fp16")]; + tensor var_16744_begin_0 = const()[name = tensor("op_16744_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_16744_end_0 = const()[name = tensor("op_16744_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_16744_end_mask_0 = const()[name = tensor("op_16744_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16744_cast_fp16 = slice_by_index(begin = var_16744_begin_0, end = var_16744_end_0, end_mask = var_16744_end_mask_0, x = v_77_cast_fp16)[name = tensor("op_16744_cast_fp16")]; + tensor var_16748_begin_0 = const()[name = tensor("op_16748_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_16748_end_0 = const()[name = tensor("op_16748_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_16748_end_mask_0 = const()[name = tensor("op_16748_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16748_cast_fp16 = slice_by_index(begin = var_16748_begin_0, end = var_16748_end_0, end_mask = var_16748_end_mask_0, x = v_77_cast_fp16)[name = tensor("op_16748_cast_fp16")]; + tensor var_16752_begin_0 = const()[name = tensor("op_16752_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_16752_end_0 = const()[name = tensor("op_16752_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_16752_end_mask_0 = const()[name = tensor("op_16752_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16752_cast_fp16 = slice_by_index(begin = var_16752_begin_0, end = var_16752_end_0, end_mask = var_16752_end_mask_0, x = v_77_cast_fp16)[name = tensor("op_16752_cast_fp16")]; + tensor var_16756_begin_0 = const()[name = tensor("op_16756_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_16756_end_0 = const()[name = tensor("op_16756_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_16756_end_mask_0 = const()[name = tensor("op_16756_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16756_cast_fp16 = slice_by_index(begin = var_16756_begin_0, end = var_16756_end_0, end_mask = var_16756_end_mask_0, x = v_77_cast_fp16)[name = tensor("op_16756_cast_fp16")]; + tensor var_16760_begin_0 = const()[name = tensor("op_16760_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_16760_end_0 = const()[name = tensor("op_16760_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_16760_end_mask_0 = const()[name = tensor("op_16760_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16760_cast_fp16 = slice_by_index(begin = var_16760_begin_0, end = var_16760_end_0, end_mask = var_16760_end_mask_0, x = v_77_cast_fp16)[name = tensor("op_16760_cast_fp16")]; + tensor var_16764_equation_0 = const()[name = tensor("op_16764_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16764_cast_fp16 = einsum(equation = var_16764_equation_0, values = (var_16606_cast_fp16, var_16523_cast_fp16))[name = tensor("op_16764_cast_fp16")]; + tensor var_16765_to_fp16 = const()[name = tensor("op_16765_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1361_cast_fp16 = mul(x = var_16764_cast_fp16, y = var_16765_to_fp16)[name = tensor("aw_1361_cast_fp16")]; + tensor var_16768_equation_0 = const()[name = tensor("op_16768_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16768_cast_fp16 = einsum(equation = var_16768_equation_0, values = (var_16610_cast_fp16, var_16527_cast_fp16))[name = tensor("op_16768_cast_fp16")]; + tensor var_16769_to_fp16 = const()[name = tensor("op_16769_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1363_cast_fp16 = mul(x = var_16768_cast_fp16, y = var_16769_to_fp16)[name = tensor("aw_1363_cast_fp16")]; + tensor var_16772_equation_0 = const()[name = tensor("op_16772_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16772_cast_fp16 = einsum(equation = var_16772_equation_0, values = (var_16614_cast_fp16, var_16531_cast_fp16))[name = tensor("op_16772_cast_fp16")]; + tensor var_16773_to_fp16 = const()[name = tensor("op_16773_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1365_cast_fp16 = mul(x = var_16772_cast_fp16, y = var_16773_to_fp16)[name = tensor("aw_1365_cast_fp16")]; + tensor var_16776_equation_0 = const()[name = tensor("op_16776_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16776_cast_fp16 = einsum(equation = var_16776_equation_0, values = (var_16618_cast_fp16, var_16535_cast_fp16))[name = tensor("op_16776_cast_fp16")]; + tensor var_16777_to_fp16 = const()[name = tensor("op_16777_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1367_cast_fp16 = mul(x = var_16776_cast_fp16, y = var_16777_to_fp16)[name = tensor("aw_1367_cast_fp16")]; + tensor var_16780_equation_0 = const()[name = tensor("op_16780_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16780_cast_fp16 = einsum(equation = var_16780_equation_0, values = (var_16622_cast_fp16, var_16539_cast_fp16))[name = tensor("op_16780_cast_fp16")]; + tensor var_16781_to_fp16 = const()[name = tensor("op_16781_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1369_cast_fp16 = mul(x = var_16780_cast_fp16, y = var_16781_to_fp16)[name = tensor("aw_1369_cast_fp16")]; + tensor var_16784_equation_0 = const()[name = tensor("op_16784_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16784_cast_fp16 = einsum(equation = var_16784_equation_0, values = (var_16626_cast_fp16, var_16543_cast_fp16))[name = tensor("op_16784_cast_fp16")]; + tensor var_16785_to_fp16 = const()[name = tensor("op_16785_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1371_cast_fp16 = mul(x = var_16784_cast_fp16, y = var_16785_to_fp16)[name = tensor("aw_1371_cast_fp16")]; + tensor var_16788_equation_0 = const()[name = tensor("op_16788_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16788_cast_fp16 = einsum(equation = var_16788_equation_0, values = (var_16630_cast_fp16, var_16547_cast_fp16))[name = tensor("op_16788_cast_fp16")]; + tensor var_16789_to_fp16 = const()[name = tensor("op_16789_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1373_cast_fp16 = mul(x = var_16788_cast_fp16, y = var_16789_to_fp16)[name = tensor("aw_1373_cast_fp16")]; + tensor var_16792_equation_0 = const()[name = tensor("op_16792_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16792_cast_fp16 = einsum(equation = var_16792_equation_0, values = (var_16634_cast_fp16, var_16551_cast_fp16))[name = tensor("op_16792_cast_fp16")]; + tensor var_16793_to_fp16 = const()[name = tensor("op_16793_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1375_cast_fp16 = mul(x = var_16792_cast_fp16, y = var_16793_to_fp16)[name = tensor("aw_1375_cast_fp16")]; + tensor var_16796_equation_0 = const()[name = tensor("op_16796_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16796_cast_fp16 = einsum(equation = var_16796_equation_0, values = (var_16638_cast_fp16, var_16555_cast_fp16))[name = tensor("op_16796_cast_fp16")]; + tensor var_16797_to_fp16 = const()[name = tensor("op_16797_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1377_cast_fp16 = mul(x = var_16796_cast_fp16, y = var_16797_to_fp16)[name = tensor("aw_1377_cast_fp16")]; + tensor var_16800_equation_0 = const()[name = tensor("op_16800_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16800_cast_fp16 = einsum(equation = var_16800_equation_0, values = (var_16642_cast_fp16, var_16559_cast_fp16))[name = tensor("op_16800_cast_fp16")]; + tensor var_16801_to_fp16 = const()[name = tensor("op_16801_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1379_cast_fp16 = mul(x = var_16800_cast_fp16, y = var_16801_to_fp16)[name = tensor("aw_1379_cast_fp16")]; + tensor var_16804_equation_0 = const()[name = tensor("op_16804_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16804_cast_fp16 = einsum(equation = var_16804_equation_0, values = (var_16646_cast_fp16, var_16563_cast_fp16))[name = tensor("op_16804_cast_fp16")]; + tensor var_16805_to_fp16 = const()[name = tensor("op_16805_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1381_cast_fp16 = mul(x = var_16804_cast_fp16, y = var_16805_to_fp16)[name = tensor("aw_1381_cast_fp16")]; + tensor var_16808_equation_0 = const()[name = tensor("op_16808_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16808_cast_fp16 = einsum(equation = var_16808_equation_0, values = (var_16650_cast_fp16, var_16567_cast_fp16))[name = tensor("op_16808_cast_fp16")]; + tensor var_16809_to_fp16 = const()[name = tensor("op_16809_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1383_cast_fp16 = mul(x = var_16808_cast_fp16, y = var_16809_to_fp16)[name = tensor("aw_1383_cast_fp16")]; + tensor var_16812_equation_0 = const()[name = tensor("op_16812_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16812_cast_fp16 = einsum(equation = var_16812_equation_0, values = (var_16654_cast_fp16, var_16571_cast_fp16))[name = tensor("op_16812_cast_fp16")]; + tensor var_16813_to_fp16 = const()[name = tensor("op_16813_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1385_cast_fp16 = mul(x = var_16812_cast_fp16, y = var_16813_to_fp16)[name = tensor("aw_1385_cast_fp16")]; + tensor var_16816_equation_0 = const()[name = tensor("op_16816_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16816_cast_fp16 = einsum(equation = var_16816_equation_0, values = (var_16658_cast_fp16, var_16575_cast_fp16))[name = tensor("op_16816_cast_fp16")]; + tensor var_16817_to_fp16 = const()[name = tensor("op_16817_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1387_cast_fp16 = mul(x = var_16816_cast_fp16, y = var_16817_to_fp16)[name = tensor("aw_1387_cast_fp16")]; + tensor var_16820_equation_0 = const()[name = tensor("op_16820_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16820_cast_fp16 = einsum(equation = var_16820_equation_0, values = (var_16662_cast_fp16, var_16579_cast_fp16))[name = tensor("op_16820_cast_fp16")]; + tensor var_16821_to_fp16 = const()[name = tensor("op_16821_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1389_cast_fp16 = mul(x = var_16820_cast_fp16, y = var_16821_to_fp16)[name = tensor("aw_1389_cast_fp16")]; + tensor var_16824_equation_0 = const()[name = tensor("op_16824_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16824_cast_fp16 = einsum(equation = var_16824_equation_0, values = (var_16666_cast_fp16, var_16583_cast_fp16))[name = tensor("op_16824_cast_fp16")]; + tensor var_16825_to_fp16 = const()[name = tensor("op_16825_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1391_cast_fp16 = mul(x = var_16824_cast_fp16, y = var_16825_to_fp16)[name = tensor("aw_1391_cast_fp16")]; + tensor var_16828_equation_0 = const()[name = tensor("op_16828_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16828_cast_fp16 = einsum(equation = var_16828_equation_0, values = (var_16670_cast_fp16, var_16587_cast_fp16))[name = tensor("op_16828_cast_fp16")]; + tensor var_16829_to_fp16 = const()[name = tensor("op_16829_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1393_cast_fp16 = mul(x = var_16828_cast_fp16, y = var_16829_to_fp16)[name = tensor("aw_1393_cast_fp16")]; + tensor var_16832_equation_0 = const()[name = tensor("op_16832_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16832_cast_fp16 = einsum(equation = var_16832_equation_0, values = (var_16674_cast_fp16, var_16591_cast_fp16))[name = tensor("op_16832_cast_fp16")]; + tensor var_16833_to_fp16 = const()[name = tensor("op_16833_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1395_cast_fp16 = mul(x = var_16832_cast_fp16, y = var_16833_to_fp16)[name = tensor("aw_1395_cast_fp16")]; + tensor var_16836_equation_0 = const()[name = tensor("op_16836_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16836_cast_fp16 = einsum(equation = var_16836_equation_0, values = (var_16678_cast_fp16, var_16595_cast_fp16))[name = tensor("op_16836_cast_fp16")]; + tensor var_16837_to_fp16 = const()[name = tensor("op_16837_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1397_cast_fp16 = mul(x = var_16836_cast_fp16, y = var_16837_to_fp16)[name = tensor("aw_1397_cast_fp16")]; + tensor var_16840_equation_0 = const()[name = tensor("op_16840_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16840_cast_fp16 = einsum(equation = var_16840_equation_0, values = (var_16682_cast_fp16, var_16599_cast_fp16))[name = tensor("op_16840_cast_fp16")]; + tensor var_16841_to_fp16 = const()[name = tensor("op_16841_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1399_cast_fp16 = mul(x = var_16840_cast_fp16, y = var_16841_to_fp16)[name = tensor("aw_1399_cast_fp16")]; + tensor var_16843_cast_fp16 = softmax(axis = var_2624, x = aw_1361_cast_fp16)[name = tensor("op_16843_cast_fp16")]; + tensor var_16844_cast_fp16 = softmax(axis = var_2624, x = aw_1363_cast_fp16)[name = tensor("op_16844_cast_fp16")]; + tensor var_16845_cast_fp16 = softmax(axis = var_2624, x = aw_1365_cast_fp16)[name = tensor("op_16845_cast_fp16")]; + tensor var_16846_cast_fp16 = softmax(axis = var_2624, x = aw_1367_cast_fp16)[name = tensor("op_16846_cast_fp16")]; + tensor var_16847_cast_fp16 = softmax(axis = var_2624, x = aw_1369_cast_fp16)[name = tensor("op_16847_cast_fp16")]; + tensor var_16848_cast_fp16 = softmax(axis = var_2624, x = aw_1371_cast_fp16)[name = tensor("op_16848_cast_fp16")]; + tensor var_16849_cast_fp16 = softmax(axis = var_2624, x = aw_1373_cast_fp16)[name = tensor("op_16849_cast_fp16")]; + tensor var_16850_cast_fp16 = softmax(axis = var_2624, x = aw_1375_cast_fp16)[name = tensor("op_16850_cast_fp16")]; + tensor var_16851_cast_fp16 = softmax(axis = var_2624, x = aw_1377_cast_fp16)[name = tensor("op_16851_cast_fp16")]; + tensor var_16852_cast_fp16 = softmax(axis = var_2624, x = aw_1379_cast_fp16)[name = tensor("op_16852_cast_fp16")]; + tensor var_16853_cast_fp16 = softmax(axis = var_2624, x = aw_1381_cast_fp16)[name = tensor("op_16853_cast_fp16")]; + tensor var_16854_cast_fp16 = softmax(axis = var_2624, x = aw_1383_cast_fp16)[name = tensor("op_16854_cast_fp16")]; + tensor var_16855_cast_fp16 = softmax(axis = var_2624, x = aw_1385_cast_fp16)[name = tensor("op_16855_cast_fp16")]; + tensor var_16856_cast_fp16 = softmax(axis = var_2624, x = aw_1387_cast_fp16)[name = tensor("op_16856_cast_fp16")]; + tensor var_16857_cast_fp16 = softmax(axis = var_2624, x = aw_1389_cast_fp16)[name = tensor("op_16857_cast_fp16")]; + tensor var_16858_cast_fp16 = softmax(axis = var_2624, x = aw_1391_cast_fp16)[name = tensor("op_16858_cast_fp16")]; + tensor var_16859_cast_fp16 = softmax(axis = var_2624, x = aw_1393_cast_fp16)[name = tensor("op_16859_cast_fp16")]; + tensor var_16860_cast_fp16 = softmax(axis = var_2624, x = aw_1395_cast_fp16)[name = tensor("op_16860_cast_fp16")]; + tensor var_16861_cast_fp16 = softmax(axis = var_2624, x = aw_1397_cast_fp16)[name = tensor("op_16861_cast_fp16")]; + tensor var_16862_cast_fp16 = softmax(axis = var_2624, x = aw_1399_cast_fp16)[name = tensor("op_16862_cast_fp16")]; + tensor var_16864_equation_0 = const()[name = tensor("op_16864_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16864_cast_fp16 = einsum(equation = var_16864_equation_0, values = (var_16684_cast_fp16, var_16843_cast_fp16))[name = tensor("op_16864_cast_fp16")]; + tensor var_16866_equation_0 = const()[name = tensor("op_16866_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16866_cast_fp16 = einsum(equation = var_16866_equation_0, values = (var_16688_cast_fp16, var_16844_cast_fp16))[name = tensor("op_16866_cast_fp16")]; + tensor var_16868_equation_0 = const()[name = tensor("op_16868_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16868_cast_fp16 = einsum(equation = var_16868_equation_0, values = (var_16692_cast_fp16, var_16845_cast_fp16))[name = tensor("op_16868_cast_fp16")]; + tensor var_16870_equation_0 = const()[name = tensor("op_16870_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16870_cast_fp16 = einsum(equation = var_16870_equation_0, values = (var_16696_cast_fp16, var_16846_cast_fp16))[name = tensor("op_16870_cast_fp16")]; + tensor var_16872_equation_0 = const()[name = tensor("op_16872_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16872_cast_fp16 = einsum(equation = var_16872_equation_0, values = (var_16700_cast_fp16, var_16847_cast_fp16))[name = tensor("op_16872_cast_fp16")]; + tensor var_16874_equation_0 = const()[name = tensor("op_16874_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16874_cast_fp16 = einsum(equation = var_16874_equation_0, values = (var_16704_cast_fp16, var_16848_cast_fp16))[name = tensor("op_16874_cast_fp16")]; + tensor var_16876_equation_0 = const()[name = tensor("op_16876_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16876_cast_fp16 = einsum(equation = var_16876_equation_0, values = (var_16708_cast_fp16, var_16849_cast_fp16))[name = tensor("op_16876_cast_fp16")]; + tensor var_16878_equation_0 = const()[name = tensor("op_16878_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16878_cast_fp16 = einsum(equation = var_16878_equation_0, values = (var_16712_cast_fp16, var_16850_cast_fp16))[name = tensor("op_16878_cast_fp16")]; + tensor var_16880_equation_0 = const()[name = tensor("op_16880_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16880_cast_fp16 = einsum(equation = var_16880_equation_0, values = (var_16716_cast_fp16, var_16851_cast_fp16))[name = tensor("op_16880_cast_fp16")]; + tensor var_16882_equation_0 = const()[name = tensor("op_16882_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16882_cast_fp16 = einsum(equation = var_16882_equation_0, values = (var_16720_cast_fp16, var_16852_cast_fp16))[name = tensor("op_16882_cast_fp16")]; + tensor var_16884_equation_0 = const()[name = tensor("op_16884_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16884_cast_fp16 = einsum(equation = var_16884_equation_0, values = (var_16724_cast_fp16, var_16853_cast_fp16))[name = tensor("op_16884_cast_fp16")]; + tensor var_16886_equation_0 = const()[name = tensor("op_16886_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16886_cast_fp16 = einsum(equation = var_16886_equation_0, values = (var_16728_cast_fp16, var_16854_cast_fp16))[name = tensor("op_16886_cast_fp16")]; + tensor var_16888_equation_0 = const()[name = tensor("op_16888_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16888_cast_fp16 = einsum(equation = var_16888_equation_0, values = (var_16732_cast_fp16, var_16855_cast_fp16))[name = tensor("op_16888_cast_fp16")]; + tensor var_16890_equation_0 = const()[name = tensor("op_16890_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16890_cast_fp16 = einsum(equation = var_16890_equation_0, values = (var_16736_cast_fp16, var_16856_cast_fp16))[name = tensor("op_16890_cast_fp16")]; + tensor var_16892_equation_0 = const()[name = tensor("op_16892_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16892_cast_fp16 = einsum(equation = var_16892_equation_0, values = (var_16740_cast_fp16, var_16857_cast_fp16))[name = tensor("op_16892_cast_fp16")]; + tensor var_16894_equation_0 = const()[name = tensor("op_16894_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16894_cast_fp16 = einsum(equation = var_16894_equation_0, values = (var_16744_cast_fp16, var_16858_cast_fp16))[name = tensor("op_16894_cast_fp16")]; + tensor var_16896_equation_0 = const()[name = tensor("op_16896_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16896_cast_fp16 = einsum(equation = var_16896_equation_0, values = (var_16748_cast_fp16, var_16859_cast_fp16))[name = tensor("op_16896_cast_fp16")]; + tensor var_16898_equation_0 = const()[name = tensor("op_16898_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16898_cast_fp16 = einsum(equation = var_16898_equation_0, values = (var_16752_cast_fp16, var_16860_cast_fp16))[name = tensor("op_16898_cast_fp16")]; + tensor var_16900_equation_0 = const()[name = tensor("op_16900_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16900_cast_fp16 = einsum(equation = var_16900_equation_0, values = (var_16756_cast_fp16, var_16861_cast_fp16))[name = tensor("op_16900_cast_fp16")]; + tensor var_16902_equation_0 = const()[name = tensor("op_16902_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16902_cast_fp16 = einsum(equation = var_16902_equation_0, values = (var_16760_cast_fp16, var_16862_cast_fp16))[name = tensor("op_16902_cast_fp16")]; + tensor input_269_interleave_0 = const()[name = tensor("input_269_interleave_0"), val = tensor(false)]; + tensor input_269_cast_fp16 = concat(axis = var_2624, interleave = input_269_interleave_0, values = (var_16864_cast_fp16, var_16866_cast_fp16, var_16868_cast_fp16, var_16870_cast_fp16, var_16872_cast_fp16, var_16874_cast_fp16, var_16876_cast_fp16, var_16878_cast_fp16, var_16880_cast_fp16, var_16882_cast_fp16, var_16884_cast_fp16, var_16886_cast_fp16, var_16888_cast_fp16, var_16890_cast_fp16, var_16892_cast_fp16, var_16894_cast_fp16, var_16896_cast_fp16, var_16898_cast_fp16, var_16900_cast_fp16, var_16902_cast_fp16))[name = tensor("input_269_cast_fp16")]; + tensor var_16908 = const()[name = tensor("op_16908"), val = tensor([1, 1])]; + tensor var_16910 = const()[name = tensor("op_16910"), val = tensor([1, 1])]; + tensor var_16912_pad_type_0 = const()[name = tensor("op_16912_pad_type_0"), val = tensor("custom")]; + tensor var_16912_pad_0 = const()[name = tensor("op_16912_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_5_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_5_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1319538112)))]; + tensor down_blocks_2_attentions_1_transformer_blocks_5_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_5_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1322814976)))]; + tensor var_16912_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_5_attn1_to_out_0_bias_to_fp16, dilations = var_16910, groups = var_2624, pad = var_16912_pad_0, pad_type = var_16912_pad_type_0, strides = var_16908, weight = down_blocks_2_attentions_1_transformer_blocks_5_attn1_to_out_0_weight_to_fp16, x = input_269_cast_fp16)[name = tensor("op_16912_cast_fp16")]; + tensor inputs_117_cast_fp16 = add(x = var_16912_cast_fp16, y = inputs_115_cast_fp16)[name = tensor("inputs_117_cast_fp16")]; + tensor hidden_states_169_axes_0 = const()[name = tensor("hidden_states_169_axes_0"), val = tensor([1])]; + tensor hidden_states_169_gamma_0_to_fp16 = const()[name = tensor("hidden_states_169_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1322817600)))]; + tensor hidden_states_169_beta_0_to_fp16 = const()[name = tensor("hidden_states_169_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1322820224)))]; + tensor var_16922_to_fp16 = const()[name = tensor("op_16922_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_169_cast_fp16 = layer_norm(axes = hidden_states_169_axes_0, beta = hidden_states_169_beta_0_to_fp16, epsilon = var_16922_to_fp16, gamma = hidden_states_169_gamma_0_to_fp16, x = inputs_117_cast_fp16)[name = tensor("hidden_states_169_cast_fp16")]; + tensor var_16937 = const()[name = tensor("op_16937"), val = tensor([1, 1])]; + tensor var_16939 = const()[name = tensor("op_16939"), val = tensor([1, 1])]; + tensor q_79_pad_type_0 = const()[name = tensor("q_79_pad_type_0"), val = tensor("custom")]; + tensor q_79_pad_0 = const()[name = tensor("q_79_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_5_attn2_to_q_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_5_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1322822848)))]; + tensor q_79_cast_fp16 = conv(dilations = var_16939, groups = var_2624, pad = q_79_pad_0, pad_type = q_79_pad_type_0, strides = var_16937, weight = down_blocks_2_attentions_1_transformer_blocks_5_attn2_to_q_weight_to_fp16, x = hidden_states_169_cast_fp16)[name = tensor("q_79_cast_fp16")]; + tensor var_16943 = const()[name = tensor("op_16943"), val = tensor([1, 1])]; + tensor var_16945 = const()[name = tensor("op_16945"), val = tensor([1, 1])]; + tensor k_157_pad_type_0 = const()[name = tensor("k_157_pad_type_0"), val = tensor("custom")]; + tensor k_157_pad_0 = const()[name = tensor("k_157_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_5_attn2_to_k_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_5_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1326099712)))]; + tensor k_157_cast_fp16 = conv(dilations = var_16945, groups = var_2624, pad = k_157_pad_0, pad_type = k_157_pad_type_0, strides = var_16943, weight = down_blocks_2_attentions_1_transformer_blocks_5_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_157_cast_fp16")]; + tensor var_16949 = const()[name = tensor("op_16949"), val = tensor([1, 1])]; + tensor var_16951 = const()[name = tensor("op_16951"), val = tensor([1, 1])]; + tensor v_79_pad_type_0 = const()[name = tensor("v_79_pad_type_0"), val = tensor("custom")]; + tensor v_79_pad_0 = const()[name = tensor("v_79_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_5_attn2_to_v_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_5_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1331342656)))]; + tensor v_79_cast_fp16 = conv(dilations = var_16951, groups = var_2624, pad = v_79_pad_0, pad_type = v_79_pad_type_0, strides = var_16949, weight = down_blocks_2_attentions_1_transformer_blocks_5_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_79_cast_fp16")]; + tensor var_16955_begin_0 = const()[name = tensor("op_16955_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_16955_end_0 = const()[name = tensor("op_16955_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_16955_end_mask_0 = const()[name = tensor("op_16955_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16955_cast_fp16 = slice_by_index(begin = var_16955_begin_0, end = var_16955_end_0, end_mask = var_16955_end_mask_0, x = q_79_cast_fp16)[name = tensor("op_16955_cast_fp16")]; + tensor var_16959_begin_0 = const()[name = tensor("op_16959_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_16959_end_0 = const()[name = tensor("op_16959_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_16959_end_mask_0 = const()[name = tensor("op_16959_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16959_cast_fp16 = slice_by_index(begin = var_16959_begin_0, end = var_16959_end_0, end_mask = var_16959_end_mask_0, x = q_79_cast_fp16)[name = tensor("op_16959_cast_fp16")]; + tensor var_16963_begin_0 = const()[name = tensor("op_16963_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_16963_end_0 = const()[name = tensor("op_16963_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_16963_end_mask_0 = const()[name = tensor("op_16963_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16963_cast_fp16 = slice_by_index(begin = var_16963_begin_0, end = var_16963_end_0, end_mask = var_16963_end_mask_0, x = q_79_cast_fp16)[name = tensor("op_16963_cast_fp16")]; + tensor var_16967_begin_0 = const()[name = tensor("op_16967_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_16967_end_0 = const()[name = tensor("op_16967_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_16967_end_mask_0 = const()[name = tensor("op_16967_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16967_cast_fp16 = slice_by_index(begin = var_16967_begin_0, end = var_16967_end_0, end_mask = var_16967_end_mask_0, x = q_79_cast_fp16)[name = tensor("op_16967_cast_fp16")]; + tensor var_16971_begin_0 = const()[name = tensor("op_16971_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_16971_end_0 = const()[name = tensor("op_16971_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_16971_end_mask_0 = const()[name = tensor("op_16971_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16971_cast_fp16 = slice_by_index(begin = var_16971_begin_0, end = var_16971_end_0, end_mask = var_16971_end_mask_0, x = q_79_cast_fp16)[name = tensor("op_16971_cast_fp16")]; + tensor var_16975_begin_0 = const()[name = tensor("op_16975_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_16975_end_0 = const()[name = tensor("op_16975_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_16975_end_mask_0 = const()[name = tensor("op_16975_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16975_cast_fp16 = slice_by_index(begin = var_16975_begin_0, end = var_16975_end_0, end_mask = var_16975_end_mask_0, x = q_79_cast_fp16)[name = tensor("op_16975_cast_fp16")]; + tensor var_16979_begin_0 = const()[name = tensor("op_16979_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_16979_end_0 = const()[name = tensor("op_16979_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_16979_end_mask_0 = const()[name = tensor("op_16979_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16979_cast_fp16 = slice_by_index(begin = var_16979_begin_0, end = var_16979_end_0, end_mask = var_16979_end_mask_0, x = q_79_cast_fp16)[name = tensor("op_16979_cast_fp16")]; + tensor var_16983_begin_0 = const()[name = tensor("op_16983_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_16983_end_0 = const()[name = tensor("op_16983_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_16983_end_mask_0 = const()[name = tensor("op_16983_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16983_cast_fp16 = slice_by_index(begin = var_16983_begin_0, end = var_16983_end_0, end_mask = var_16983_end_mask_0, x = q_79_cast_fp16)[name = tensor("op_16983_cast_fp16")]; + tensor var_16987_begin_0 = const()[name = tensor("op_16987_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_16987_end_0 = const()[name = tensor("op_16987_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_16987_end_mask_0 = const()[name = tensor("op_16987_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16987_cast_fp16 = slice_by_index(begin = var_16987_begin_0, end = var_16987_end_0, end_mask = var_16987_end_mask_0, x = q_79_cast_fp16)[name = tensor("op_16987_cast_fp16")]; + tensor var_16991_begin_0 = const()[name = tensor("op_16991_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_16991_end_0 = const()[name = tensor("op_16991_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_16991_end_mask_0 = const()[name = tensor("op_16991_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16991_cast_fp16 = slice_by_index(begin = var_16991_begin_0, end = var_16991_end_0, end_mask = var_16991_end_mask_0, x = q_79_cast_fp16)[name = tensor("op_16991_cast_fp16")]; + tensor var_16995_begin_0 = const()[name = tensor("op_16995_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_16995_end_0 = const()[name = tensor("op_16995_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_16995_end_mask_0 = const()[name = tensor("op_16995_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16995_cast_fp16 = slice_by_index(begin = var_16995_begin_0, end = var_16995_end_0, end_mask = var_16995_end_mask_0, x = q_79_cast_fp16)[name = tensor("op_16995_cast_fp16")]; + tensor var_16999_begin_0 = const()[name = tensor("op_16999_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_16999_end_0 = const()[name = tensor("op_16999_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_16999_end_mask_0 = const()[name = tensor("op_16999_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16999_cast_fp16 = slice_by_index(begin = var_16999_begin_0, end = var_16999_end_0, end_mask = var_16999_end_mask_0, x = q_79_cast_fp16)[name = tensor("op_16999_cast_fp16")]; + tensor var_17003_begin_0 = const()[name = tensor("op_17003_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_17003_end_0 = const()[name = tensor("op_17003_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_17003_end_mask_0 = const()[name = tensor("op_17003_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17003_cast_fp16 = slice_by_index(begin = var_17003_begin_0, end = var_17003_end_0, end_mask = var_17003_end_mask_0, x = q_79_cast_fp16)[name = tensor("op_17003_cast_fp16")]; + tensor var_17007_begin_0 = const()[name = tensor("op_17007_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_17007_end_0 = const()[name = tensor("op_17007_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_17007_end_mask_0 = const()[name = tensor("op_17007_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17007_cast_fp16 = slice_by_index(begin = var_17007_begin_0, end = var_17007_end_0, end_mask = var_17007_end_mask_0, x = q_79_cast_fp16)[name = tensor("op_17007_cast_fp16")]; + tensor var_17011_begin_0 = const()[name = tensor("op_17011_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_17011_end_0 = const()[name = tensor("op_17011_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_17011_end_mask_0 = const()[name = tensor("op_17011_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17011_cast_fp16 = slice_by_index(begin = var_17011_begin_0, end = var_17011_end_0, end_mask = var_17011_end_mask_0, x = q_79_cast_fp16)[name = tensor("op_17011_cast_fp16")]; + tensor var_17015_begin_0 = const()[name = tensor("op_17015_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_17015_end_0 = const()[name = tensor("op_17015_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_17015_end_mask_0 = const()[name = tensor("op_17015_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17015_cast_fp16 = slice_by_index(begin = var_17015_begin_0, end = var_17015_end_0, end_mask = var_17015_end_mask_0, x = q_79_cast_fp16)[name = tensor("op_17015_cast_fp16")]; + tensor var_17019_begin_0 = const()[name = tensor("op_17019_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_17019_end_0 = const()[name = tensor("op_17019_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_17019_end_mask_0 = const()[name = tensor("op_17019_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17019_cast_fp16 = slice_by_index(begin = var_17019_begin_0, end = var_17019_end_0, end_mask = var_17019_end_mask_0, x = q_79_cast_fp16)[name = tensor("op_17019_cast_fp16")]; + tensor var_17023_begin_0 = const()[name = tensor("op_17023_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_17023_end_0 = const()[name = tensor("op_17023_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_17023_end_mask_0 = const()[name = tensor("op_17023_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17023_cast_fp16 = slice_by_index(begin = var_17023_begin_0, end = var_17023_end_0, end_mask = var_17023_end_mask_0, x = q_79_cast_fp16)[name = tensor("op_17023_cast_fp16")]; + tensor var_17027_begin_0 = const()[name = tensor("op_17027_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_17027_end_0 = const()[name = tensor("op_17027_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_17027_end_mask_0 = const()[name = tensor("op_17027_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17027_cast_fp16 = slice_by_index(begin = var_17027_begin_0, end = var_17027_end_0, end_mask = var_17027_end_mask_0, x = q_79_cast_fp16)[name = tensor("op_17027_cast_fp16")]; + tensor var_17031_begin_0 = const()[name = tensor("op_17031_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_17031_end_0 = const()[name = tensor("op_17031_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_17031_end_mask_0 = const()[name = tensor("op_17031_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17031_cast_fp16 = slice_by_index(begin = var_17031_begin_0, end = var_17031_end_0, end_mask = var_17031_end_mask_0, x = q_79_cast_fp16)[name = tensor("op_17031_cast_fp16")]; + tensor k_159_perm_0 = const()[name = tensor("k_159_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_17038_begin_0 = const()[name = tensor("op_17038_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17038_end_0 = const()[name = tensor("op_17038_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_17038_end_mask_0 = const()[name = tensor("op_17038_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_159_cast_fp16 = transpose(perm = k_159_perm_0, x = k_157_cast_fp16)[name = tensor("transpose_100")]; + tensor var_17038_cast_fp16 = slice_by_index(begin = var_17038_begin_0, end = var_17038_end_0, end_mask = var_17038_end_mask_0, x = k_159_cast_fp16)[name = tensor("op_17038_cast_fp16")]; + tensor var_17042_begin_0 = const()[name = tensor("op_17042_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_17042_end_0 = const()[name = tensor("op_17042_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_17042_end_mask_0 = const()[name = tensor("op_17042_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17042_cast_fp16 = slice_by_index(begin = var_17042_begin_0, end = var_17042_end_0, end_mask = var_17042_end_mask_0, x = k_159_cast_fp16)[name = tensor("op_17042_cast_fp16")]; + tensor var_17046_begin_0 = const()[name = tensor("op_17046_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_17046_end_0 = const()[name = tensor("op_17046_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_17046_end_mask_0 = const()[name = tensor("op_17046_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17046_cast_fp16 = slice_by_index(begin = var_17046_begin_0, end = var_17046_end_0, end_mask = var_17046_end_mask_0, x = k_159_cast_fp16)[name = tensor("op_17046_cast_fp16")]; + tensor var_17050_begin_0 = const()[name = tensor("op_17050_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_17050_end_0 = const()[name = tensor("op_17050_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_17050_end_mask_0 = const()[name = tensor("op_17050_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17050_cast_fp16 = slice_by_index(begin = var_17050_begin_0, end = var_17050_end_0, end_mask = var_17050_end_mask_0, x = k_159_cast_fp16)[name = tensor("op_17050_cast_fp16")]; + tensor var_17054_begin_0 = const()[name = tensor("op_17054_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_17054_end_0 = const()[name = tensor("op_17054_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_17054_end_mask_0 = const()[name = tensor("op_17054_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17054_cast_fp16 = slice_by_index(begin = var_17054_begin_0, end = var_17054_end_0, end_mask = var_17054_end_mask_0, x = k_159_cast_fp16)[name = tensor("op_17054_cast_fp16")]; + tensor var_17058_begin_0 = const()[name = tensor("op_17058_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_17058_end_0 = const()[name = tensor("op_17058_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_17058_end_mask_0 = const()[name = tensor("op_17058_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17058_cast_fp16 = slice_by_index(begin = var_17058_begin_0, end = var_17058_end_0, end_mask = var_17058_end_mask_0, x = k_159_cast_fp16)[name = tensor("op_17058_cast_fp16")]; + tensor var_17062_begin_0 = const()[name = tensor("op_17062_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_17062_end_0 = const()[name = tensor("op_17062_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_17062_end_mask_0 = const()[name = tensor("op_17062_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17062_cast_fp16 = slice_by_index(begin = var_17062_begin_0, end = var_17062_end_0, end_mask = var_17062_end_mask_0, x = k_159_cast_fp16)[name = tensor("op_17062_cast_fp16")]; + tensor var_17066_begin_0 = const()[name = tensor("op_17066_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_17066_end_0 = const()[name = tensor("op_17066_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_17066_end_mask_0 = const()[name = tensor("op_17066_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17066_cast_fp16 = slice_by_index(begin = var_17066_begin_0, end = var_17066_end_0, end_mask = var_17066_end_mask_0, x = k_159_cast_fp16)[name = tensor("op_17066_cast_fp16")]; + tensor var_17070_begin_0 = const()[name = tensor("op_17070_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_17070_end_0 = const()[name = tensor("op_17070_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_17070_end_mask_0 = const()[name = tensor("op_17070_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17070_cast_fp16 = slice_by_index(begin = var_17070_begin_0, end = var_17070_end_0, end_mask = var_17070_end_mask_0, x = k_159_cast_fp16)[name = tensor("op_17070_cast_fp16")]; + tensor var_17074_begin_0 = const()[name = tensor("op_17074_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_17074_end_0 = const()[name = tensor("op_17074_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_17074_end_mask_0 = const()[name = tensor("op_17074_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17074_cast_fp16 = slice_by_index(begin = var_17074_begin_0, end = var_17074_end_0, end_mask = var_17074_end_mask_0, x = k_159_cast_fp16)[name = tensor("op_17074_cast_fp16")]; + tensor var_17078_begin_0 = const()[name = tensor("op_17078_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_17078_end_0 = const()[name = tensor("op_17078_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_17078_end_mask_0 = const()[name = tensor("op_17078_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17078_cast_fp16 = slice_by_index(begin = var_17078_begin_0, end = var_17078_end_0, end_mask = var_17078_end_mask_0, x = k_159_cast_fp16)[name = tensor("op_17078_cast_fp16")]; + tensor var_17082_begin_0 = const()[name = tensor("op_17082_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_17082_end_0 = const()[name = tensor("op_17082_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_17082_end_mask_0 = const()[name = tensor("op_17082_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17082_cast_fp16 = slice_by_index(begin = var_17082_begin_0, end = var_17082_end_0, end_mask = var_17082_end_mask_0, x = k_159_cast_fp16)[name = tensor("op_17082_cast_fp16")]; + tensor var_17086_begin_0 = const()[name = tensor("op_17086_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_17086_end_0 = const()[name = tensor("op_17086_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_17086_end_mask_0 = const()[name = tensor("op_17086_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17086_cast_fp16 = slice_by_index(begin = var_17086_begin_0, end = var_17086_end_0, end_mask = var_17086_end_mask_0, x = k_159_cast_fp16)[name = tensor("op_17086_cast_fp16")]; + tensor var_17090_begin_0 = const()[name = tensor("op_17090_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_17090_end_0 = const()[name = tensor("op_17090_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_17090_end_mask_0 = const()[name = tensor("op_17090_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17090_cast_fp16 = slice_by_index(begin = var_17090_begin_0, end = var_17090_end_0, end_mask = var_17090_end_mask_0, x = k_159_cast_fp16)[name = tensor("op_17090_cast_fp16")]; + tensor var_17094_begin_0 = const()[name = tensor("op_17094_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_17094_end_0 = const()[name = tensor("op_17094_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_17094_end_mask_0 = const()[name = tensor("op_17094_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17094_cast_fp16 = slice_by_index(begin = var_17094_begin_0, end = var_17094_end_0, end_mask = var_17094_end_mask_0, x = k_159_cast_fp16)[name = tensor("op_17094_cast_fp16")]; + tensor var_17098_begin_0 = const()[name = tensor("op_17098_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_17098_end_0 = const()[name = tensor("op_17098_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_17098_end_mask_0 = const()[name = tensor("op_17098_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17098_cast_fp16 = slice_by_index(begin = var_17098_begin_0, end = var_17098_end_0, end_mask = var_17098_end_mask_0, x = k_159_cast_fp16)[name = tensor("op_17098_cast_fp16")]; + tensor var_17102_begin_0 = const()[name = tensor("op_17102_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_17102_end_0 = const()[name = tensor("op_17102_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_17102_end_mask_0 = const()[name = tensor("op_17102_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17102_cast_fp16 = slice_by_index(begin = var_17102_begin_0, end = var_17102_end_0, end_mask = var_17102_end_mask_0, x = k_159_cast_fp16)[name = tensor("op_17102_cast_fp16")]; + tensor var_17106_begin_0 = const()[name = tensor("op_17106_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_17106_end_0 = const()[name = tensor("op_17106_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_17106_end_mask_0 = const()[name = tensor("op_17106_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17106_cast_fp16 = slice_by_index(begin = var_17106_begin_0, end = var_17106_end_0, end_mask = var_17106_end_mask_0, x = k_159_cast_fp16)[name = tensor("op_17106_cast_fp16")]; + tensor var_17110_begin_0 = const()[name = tensor("op_17110_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_17110_end_0 = const()[name = tensor("op_17110_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_17110_end_mask_0 = const()[name = tensor("op_17110_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17110_cast_fp16 = slice_by_index(begin = var_17110_begin_0, end = var_17110_end_0, end_mask = var_17110_end_mask_0, x = k_159_cast_fp16)[name = tensor("op_17110_cast_fp16")]; + tensor var_17114_begin_0 = const()[name = tensor("op_17114_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_17114_end_0 = const()[name = tensor("op_17114_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_17114_end_mask_0 = const()[name = tensor("op_17114_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17114_cast_fp16 = slice_by_index(begin = var_17114_begin_0, end = var_17114_end_0, end_mask = var_17114_end_mask_0, x = k_159_cast_fp16)[name = tensor("op_17114_cast_fp16")]; + tensor var_17116_begin_0 = const()[name = tensor("op_17116_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17116_end_0 = const()[name = tensor("op_17116_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_17116_end_mask_0 = const()[name = tensor("op_17116_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17116_cast_fp16 = slice_by_index(begin = var_17116_begin_0, end = var_17116_end_0, end_mask = var_17116_end_mask_0, x = v_79_cast_fp16)[name = tensor("op_17116_cast_fp16")]; + tensor var_17120_begin_0 = const()[name = tensor("op_17120_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_17120_end_0 = const()[name = tensor("op_17120_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_17120_end_mask_0 = const()[name = tensor("op_17120_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17120_cast_fp16 = slice_by_index(begin = var_17120_begin_0, end = var_17120_end_0, end_mask = var_17120_end_mask_0, x = v_79_cast_fp16)[name = tensor("op_17120_cast_fp16")]; + tensor var_17124_begin_0 = const()[name = tensor("op_17124_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_17124_end_0 = const()[name = tensor("op_17124_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_17124_end_mask_0 = const()[name = tensor("op_17124_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17124_cast_fp16 = slice_by_index(begin = var_17124_begin_0, end = var_17124_end_0, end_mask = var_17124_end_mask_0, x = v_79_cast_fp16)[name = tensor("op_17124_cast_fp16")]; + tensor var_17128_begin_0 = const()[name = tensor("op_17128_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_17128_end_0 = const()[name = tensor("op_17128_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_17128_end_mask_0 = const()[name = tensor("op_17128_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17128_cast_fp16 = slice_by_index(begin = var_17128_begin_0, end = var_17128_end_0, end_mask = var_17128_end_mask_0, x = v_79_cast_fp16)[name = tensor("op_17128_cast_fp16")]; + tensor var_17132_begin_0 = const()[name = tensor("op_17132_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_17132_end_0 = const()[name = tensor("op_17132_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_17132_end_mask_0 = const()[name = tensor("op_17132_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17132_cast_fp16 = slice_by_index(begin = var_17132_begin_0, end = var_17132_end_0, end_mask = var_17132_end_mask_0, x = v_79_cast_fp16)[name = tensor("op_17132_cast_fp16")]; + tensor var_17136_begin_0 = const()[name = tensor("op_17136_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_17136_end_0 = const()[name = tensor("op_17136_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_17136_end_mask_0 = const()[name = tensor("op_17136_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17136_cast_fp16 = slice_by_index(begin = var_17136_begin_0, end = var_17136_end_0, end_mask = var_17136_end_mask_0, x = v_79_cast_fp16)[name = tensor("op_17136_cast_fp16")]; + tensor var_17140_begin_0 = const()[name = tensor("op_17140_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_17140_end_0 = const()[name = tensor("op_17140_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_17140_end_mask_0 = const()[name = tensor("op_17140_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17140_cast_fp16 = slice_by_index(begin = var_17140_begin_0, end = var_17140_end_0, end_mask = var_17140_end_mask_0, x = v_79_cast_fp16)[name = tensor("op_17140_cast_fp16")]; + tensor var_17144_begin_0 = const()[name = tensor("op_17144_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_17144_end_0 = const()[name = tensor("op_17144_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_17144_end_mask_0 = const()[name = tensor("op_17144_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17144_cast_fp16 = slice_by_index(begin = var_17144_begin_0, end = var_17144_end_0, end_mask = var_17144_end_mask_0, x = v_79_cast_fp16)[name = tensor("op_17144_cast_fp16")]; + tensor var_17148_begin_0 = const()[name = tensor("op_17148_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_17148_end_0 = const()[name = tensor("op_17148_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_17148_end_mask_0 = const()[name = tensor("op_17148_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17148_cast_fp16 = slice_by_index(begin = var_17148_begin_0, end = var_17148_end_0, end_mask = var_17148_end_mask_0, x = v_79_cast_fp16)[name = tensor("op_17148_cast_fp16")]; + tensor var_17152_begin_0 = const()[name = tensor("op_17152_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_17152_end_0 = const()[name = tensor("op_17152_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_17152_end_mask_0 = const()[name = tensor("op_17152_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17152_cast_fp16 = slice_by_index(begin = var_17152_begin_0, end = var_17152_end_0, end_mask = var_17152_end_mask_0, x = v_79_cast_fp16)[name = tensor("op_17152_cast_fp16")]; + tensor var_17156_begin_0 = const()[name = tensor("op_17156_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_17156_end_0 = const()[name = tensor("op_17156_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_17156_end_mask_0 = const()[name = tensor("op_17156_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17156_cast_fp16 = slice_by_index(begin = var_17156_begin_0, end = var_17156_end_0, end_mask = var_17156_end_mask_0, x = v_79_cast_fp16)[name = tensor("op_17156_cast_fp16")]; + tensor var_17160_begin_0 = const()[name = tensor("op_17160_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_17160_end_0 = const()[name = tensor("op_17160_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_17160_end_mask_0 = const()[name = tensor("op_17160_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17160_cast_fp16 = slice_by_index(begin = var_17160_begin_0, end = var_17160_end_0, end_mask = var_17160_end_mask_0, x = v_79_cast_fp16)[name = tensor("op_17160_cast_fp16")]; + tensor var_17164_begin_0 = const()[name = tensor("op_17164_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_17164_end_0 = const()[name = tensor("op_17164_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_17164_end_mask_0 = const()[name = tensor("op_17164_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17164_cast_fp16 = slice_by_index(begin = var_17164_begin_0, end = var_17164_end_0, end_mask = var_17164_end_mask_0, x = v_79_cast_fp16)[name = tensor("op_17164_cast_fp16")]; + tensor var_17168_begin_0 = const()[name = tensor("op_17168_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_17168_end_0 = const()[name = tensor("op_17168_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_17168_end_mask_0 = const()[name = tensor("op_17168_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17168_cast_fp16 = slice_by_index(begin = var_17168_begin_0, end = var_17168_end_0, end_mask = var_17168_end_mask_0, x = v_79_cast_fp16)[name = tensor("op_17168_cast_fp16")]; + tensor var_17172_begin_0 = const()[name = tensor("op_17172_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_17172_end_0 = const()[name = tensor("op_17172_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_17172_end_mask_0 = const()[name = tensor("op_17172_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17172_cast_fp16 = slice_by_index(begin = var_17172_begin_0, end = var_17172_end_0, end_mask = var_17172_end_mask_0, x = v_79_cast_fp16)[name = tensor("op_17172_cast_fp16")]; + tensor var_17176_begin_0 = const()[name = tensor("op_17176_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_17176_end_0 = const()[name = tensor("op_17176_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_17176_end_mask_0 = const()[name = tensor("op_17176_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17176_cast_fp16 = slice_by_index(begin = var_17176_begin_0, end = var_17176_end_0, end_mask = var_17176_end_mask_0, x = v_79_cast_fp16)[name = tensor("op_17176_cast_fp16")]; + tensor var_17180_begin_0 = const()[name = tensor("op_17180_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_17180_end_0 = const()[name = tensor("op_17180_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_17180_end_mask_0 = const()[name = tensor("op_17180_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17180_cast_fp16 = slice_by_index(begin = var_17180_begin_0, end = var_17180_end_0, end_mask = var_17180_end_mask_0, x = v_79_cast_fp16)[name = tensor("op_17180_cast_fp16")]; + tensor var_17184_begin_0 = const()[name = tensor("op_17184_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_17184_end_0 = const()[name = tensor("op_17184_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_17184_end_mask_0 = const()[name = tensor("op_17184_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17184_cast_fp16 = slice_by_index(begin = var_17184_begin_0, end = var_17184_end_0, end_mask = var_17184_end_mask_0, x = v_79_cast_fp16)[name = tensor("op_17184_cast_fp16")]; + tensor var_17188_begin_0 = const()[name = tensor("op_17188_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_17188_end_0 = const()[name = tensor("op_17188_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_17188_end_mask_0 = const()[name = tensor("op_17188_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17188_cast_fp16 = slice_by_index(begin = var_17188_begin_0, end = var_17188_end_0, end_mask = var_17188_end_mask_0, x = v_79_cast_fp16)[name = tensor("op_17188_cast_fp16")]; + tensor var_17192_begin_0 = const()[name = tensor("op_17192_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_17192_end_0 = const()[name = tensor("op_17192_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_17192_end_mask_0 = const()[name = tensor("op_17192_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17192_cast_fp16 = slice_by_index(begin = var_17192_begin_0, end = var_17192_end_0, end_mask = var_17192_end_mask_0, x = v_79_cast_fp16)[name = tensor("op_17192_cast_fp16")]; + tensor var_17196_equation_0 = const()[name = tensor("op_17196_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17196_cast_fp16 = einsum(equation = var_17196_equation_0, values = (var_17038_cast_fp16, var_16955_cast_fp16))[name = tensor("op_17196_cast_fp16")]; + tensor var_17197_to_fp16 = const()[name = tensor("op_17197_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1401_cast_fp16 = mul(x = var_17196_cast_fp16, y = var_17197_to_fp16)[name = tensor("aw_1401_cast_fp16")]; + tensor var_17200_equation_0 = const()[name = tensor("op_17200_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17200_cast_fp16 = einsum(equation = var_17200_equation_0, values = (var_17042_cast_fp16, var_16959_cast_fp16))[name = tensor("op_17200_cast_fp16")]; + tensor var_17201_to_fp16 = const()[name = tensor("op_17201_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1403_cast_fp16 = mul(x = var_17200_cast_fp16, y = var_17201_to_fp16)[name = tensor("aw_1403_cast_fp16")]; + tensor var_17204_equation_0 = const()[name = tensor("op_17204_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17204_cast_fp16 = einsum(equation = var_17204_equation_0, values = (var_17046_cast_fp16, var_16963_cast_fp16))[name = tensor("op_17204_cast_fp16")]; + tensor var_17205_to_fp16 = const()[name = tensor("op_17205_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1405_cast_fp16 = mul(x = var_17204_cast_fp16, y = var_17205_to_fp16)[name = tensor("aw_1405_cast_fp16")]; + tensor var_17208_equation_0 = const()[name = tensor("op_17208_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17208_cast_fp16 = einsum(equation = var_17208_equation_0, values = (var_17050_cast_fp16, var_16967_cast_fp16))[name = tensor("op_17208_cast_fp16")]; + tensor var_17209_to_fp16 = const()[name = tensor("op_17209_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1407_cast_fp16 = mul(x = var_17208_cast_fp16, y = var_17209_to_fp16)[name = tensor("aw_1407_cast_fp16")]; + tensor var_17212_equation_0 = const()[name = tensor("op_17212_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17212_cast_fp16 = einsum(equation = var_17212_equation_0, values = (var_17054_cast_fp16, var_16971_cast_fp16))[name = tensor("op_17212_cast_fp16")]; + tensor var_17213_to_fp16 = const()[name = tensor("op_17213_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1409_cast_fp16 = mul(x = var_17212_cast_fp16, y = var_17213_to_fp16)[name = tensor("aw_1409_cast_fp16")]; + tensor var_17216_equation_0 = const()[name = tensor("op_17216_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17216_cast_fp16 = einsum(equation = var_17216_equation_0, values = (var_17058_cast_fp16, var_16975_cast_fp16))[name = tensor("op_17216_cast_fp16")]; + tensor var_17217_to_fp16 = const()[name = tensor("op_17217_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1411_cast_fp16 = mul(x = var_17216_cast_fp16, y = var_17217_to_fp16)[name = tensor("aw_1411_cast_fp16")]; + tensor var_17220_equation_0 = const()[name = tensor("op_17220_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17220_cast_fp16 = einsum(equation = var_17220_equation_0, values = (var_17062_cast_fp16, var_16979_cast_fp16))[name = tensor("op_17220_cast_fp16")]; + tensor var_17221_to_fp16 = const()[name = tensor("op_17221_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1413_cast_fp16 = mul(x = var_17220_cast_fp16, y = var_17221_to_fp16)[name = tensor("aw_1413_cast_fp16")]; + tensor var_17224_equation_0 = const()[name = tensor("op_17224_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17224_cast_fp16 = einsum(equation = var_17224_equation_0, values = (var_17066_cast_fp16, var_16983_cast_fp16))[name = tensor("op_17224_cast_fp16")]; + tensor var_17225_to_fp16 = const()[name = tensor("op_17225_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1415_cast_fp16 = mul(x = var_17224_cast_fp16, y = var_17225_to_fp16)[name = tensor("aw_1415_cast_fp16")]; + tensor var_17228_equation_0 = const()[name = tensor("op_17228_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17228_cast_fp16 = einsum(equation = var_17228_equation_0, values = (var_17070_cast_fp16, var_16987_cast_fp16))[name = tensor("op_17228_cast_fp16")]; + tensor var_17229_to_fp16 = const()[name = tensor("op_17229_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1417_cast_fp16 = mul(x = var_17228_cast_fp16, y = var_17229_to_fp16)[name = tensor("aw_1417_cast_fp16")]; + tensor var_17232_equation_0 = const()[name = tensor("op_17232_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17232_cast_fp16 = einsum(equation = var_17232_equation_0, values = (var_17074_cast_fp16, var_16991_cast_fp16))[name = tensor("op_17232_cast_fp16")]; + tensor var_17233_to_fp16 = const()[name = tensor("op_17233_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1419_cast_fp16 = mul(x = var_17232_cast_fp16, y = var_17233_to_fp16)[name = tensor("aw_1419_cast_fp16")]; + tensor var_17236_equation_0 = const()[name = tensor("op_17236_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17236_cast_fp16 = einsum(equation = var_17236_equation_0, values = (var_17078_cast_fp16, var_16995_cast_fp16))[name = tensor("op_17236_cast_fp16")]; + tensor var_17237_to_fp16 = const()[name = tensor("op_17237_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1421_cast_fp16 = mul(x = var_17236_cast_fp16, y = var_17237_to_fp16)[name = tensor("aw_1421_cast_fp16")]; + tensor var_17240_equation_0 = const()[name = tensor("op_17240_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17240_cast_fp16 = einsum(equation = var_17240_equation_0, values = (var_17082_cast_fp16, var_16999_cast_fp16))[name = tensor("op_17240_cast_fp16")]; + tensor var_17241_to_fp16 = const()[name = tensor("op_17241_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1423_cast_fp16 = mul(x = var_17240_cast_fp16, y = var_17241_to_fp16)[name = tensor("aw_1423_cast_fp16")]; + tensor var_17244_equation_0 = const()[name = tensor("op_17244_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17244_cast_fp16 = einsum(equation = var_17244_equation_0, values = (var_17086_cast_fp16, var_17003_cast_fp16))[name = tensor("op_17244_cast_fp16")]; + tensor var_17245_to_fp16 = const()[name = tensor("op_17245_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1425_cast_fp16 = mul(x = var_17244_cast_fp16, y = var_17245_to_fp16)[name = tensor("aw_1425_cast_fp16")]; + tensor var_17248_equation_0 = const()[name = tensor("op_17248_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17248_cast_fp16 = einsum(equation = var_17248_equation_0, values = (var_17090_cast_fp16, var_17007_cast_fp16))[name = tensor("op_17248_cast_fp16")]; + tensor var_17249_to_fp16 = const()[name = tensor("op_17249_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1427_cast_fp16 = mul(x = var_17248_cast_fp16, y = var_17249_to_fp16)[name = tensor("aw_1427_cast_fp16")]; + tensor var_17252_equation_0 = const()[name = tensor("op_17252_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17252_cast_fp16 = einsum(equation = var_17252_equation_0, values = (var_17094_cast_fp16, var_17011_cast_fp16))[name = tensor("op_17252_cast_fp16")]; + tensor var_17253_to_fp16 = const()[name = tensor("op_17253_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1429_cast_fp16 = mul(x = var_17252_cast_fp16, y = var_17253_to_fp16)[name = tensor("aw_1429_cast_fp16")]; + tensor var_17256_equation_0 = const()[name = tensor("op_17256_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17256_cast_fp16 = einsum(equation = var_17256_equation_0, values = (var_17098_cast_fp16, var_17015_cast_fp16))[name = tensor("op_17256_cast_fp16")]; + tensor var_17257_to_fp16 = const()[name = tensor("op_17257_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1431_cast_fp16 = mul(x = var_17256_cast_fp16, y = var_17257_to_fp16)[name = tensor("aw_1431_cast_fp16")]; + tensor var_17260_equation_0 = const()[name = tensor("op_17260_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17260_cast_fp16 = einsum(equation = var_17260_equation_0, values = (var_17102_cast_fp16, var_17019_cast_fp16))[name = tensor("op_17260_cast_fp16")]; + tensor var_17261_to_fp16 = const()[name = tensor("op_17261_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1433_cast_fp16 = mul(x = var_17260_cast_fp16, y = var_17261_to_fp16)[name = tensor("aw_1433_cast_fp16")]; + tensor var_17264_equation_0 = const()[name = tensor("op_17264_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17264_cast_fp16 = einsum(equation = var_17264_equation_0, values = (var_17106_cast_fp16, var_17023_cast_fp16))[name = tensor("op_17264_cast_fp16")]; + tensor var_17265_to_fp16 = const()[name = tensor("op_17265_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1435_cast_fp16 = mul(x = var_17264_cast_fp16, y = var_17265_to_fp16)[name = tensor("aw_1435_cast_fp16")]; + tensor var_17268_equation_0 = const()[name = tensor("op_17268_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17268_cast_fp16 = einsum(equation = var_17268_equation_0, values = (var_17110_cast_fp16, var_17027_cast_fp16))[name = tensor("op_17268_cast_fp16")]; + tensor var_17269_to_fp16 = const()[name = tensor("op_17269_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1437_cast_fp16 = mul(x = var_17268_cast_fp16, y = var_17269_to_fp16)[name = tensor("aw_1437_cast_fp16")]; + tensor var_17272_equation_0 = const()[name = tensor("op_17272_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17272_cast_fp16 = einsum(equation = var_17272_equation_0, values = (var_17114_cast_fp16, var_17031_cast_fp16))[name = tensor("op_17272_cast_fp16")]; + tensor var_17273_to_fp16 = const()[name = tensor("op_17273_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1439_cast_fp16 = mul(x = var_17272_cast_fp16, y = var_17273_to_fp16)[name = tensor("aw_1439_cast_fp16")]; + tensor var_17275_cast_fp16 = softmax(axis = var_2624, x = aw_1401_cast_fp16)[name = tensor("op_17275_cast_fp16")]; + tensor var_17276_cast_fp16 = softmax(axis = var_2624, x = aw_1403_cast_fp16)[name = tensor("op_17276_cast_fp16")]; + tensor var_17277_cast_fp16 = softmax(axis = var_2624, x = aw_1405_cast_fp16)[name = tensor("op_17277_cast_fp16")]; + tensor var_17278_cast_fp16 = softmax(axis = var_2624, x = aw_1407_cast_fp16)[name = tensor("op_17278_cast_fp16")]; + tensor var_17279_cast_fp16 = softmax(axis = var_2624, x = aw_1409_cast_fp16)[name = tensor("op_17279_cast_fp16")]; + tensor var_17280_cast_fp16 = softmax(axis = var_2624, x = aw_1411_cast_fp16)[name = tensor("op_17280_cast_fp16")]; + tensor var_17281_cast_fp16 = softmax(axis = var_2624, x = aw_1413_cast_fp16)[name = tensor("op_17281_cast_fp16")]; + tensor var_17282_cast_fp16 = softmax(axis = var_2624, x = aw_1415_cast_fp16)[name = tensor("op_17282_cast_fp16")]; + tensor var_17283_cast_fp16 = softmax(axis = var_2624, x = aw_1417_cast_fp16)[name = tensor("op_17283_cast_fp16")]; + tensor var_17284_cast_fp16 = softmax(axis = var_2624, x = aw_1419_cast_fp16)[name = tensor("op_17284_cast_fp16")]; + tensor var_17285_cast_fp16 = softmax(axis = var_2624, x = aw_1421_cast_fp16)[name = tensor("op_17285_cast_fp16")]; + tensor var_17286_cast_fp16 = softmax(axis = var_2624, x = aw_1423_cast_fp16)[name = tensor("op_17286_cast_fp16")]; + tensor var_17287_cast_fp16 = softmax(axis = var_2624, x = aw_1425_cast_fp16)[name = tensor("op_17287_cast_fp16")]; + tensor var_17288_cast_fp16 = softmax(axis = var_2624, x = aw_1427_cast_fp16)[name = tensor("op_17288_cast_fp16")]; + tensor var_17289_cast_fp16 = softmax(axis = var_2624, x = aw_1429_cast_fp16)[name = tensor("op_17289_cast_fp16")]; + tensor var_17290_cast_fp16 = softmax(axis = var_2624, x = aw_1431_cast_fp16)[name = tensor("op_17290_cast_fp16")]; + tensor var_17291_cast_fp16 = softmax(axis = var_2624, x = aw_1433_cast_fp16)[name = tensor("op_17291_cast_fp16")]; + tensor var_17292_cast_fp16 = softmax(axis = var_2624, x = aw_1435_cast_fp16)[name = tensor("op_17292_cast_fp16")]; + tensor var_17293_cast_fp16 = softmax(axis = var_2624, x = aw_1437_cast_fp16)[name = tensor("op_17293_cast_fp16")]; + tensor var_17294_cast_fp16 = softmax(axis = var_2624, x = aw_1439_cast_fp16)[name = tensor("op_17294_cast_fp16")]; + tensor var_17296_equation_0 = const()[name = tensor("op_17296_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17296_cast_fp16 = einsum(equation = var_17296_equation_0, values = (var_17116_cast_fp16, var_17275_cast_fp16))[name = tensor("op_17296_cast_fp16")]; + tensor var_17298_equation_0 = const()[name = tensor("op_17298_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17298_cast_fp16 = einsum(equation = var_17298_equation_0, values = (var_17120_cast_fp16, var_17276_cast_fp16))[name = tensor("op_17298_cast_fp16")]; + tensor var_17300_equation_0 = const()[name = tensor("op_17300_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17300_cast_fp16 = einsum(equation = var_17300_equation_0, values = (var_17124_cast_fp16, var_17277_cast_fp16))[name = tensor("op_17300_cast_fp16")]; + tensor var_17302_equation_0 = const()[name = tensor("op_17302_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17302_cast_fp16 = einsum(equation = var_17302_equation_0, values = (var_17128_cast_fp16, var_17278_cast_fp16))[name = tensor("op_17302_cast_fp16")]; + tensor var_17304_equation_0 = const()[name = tensor("op_17304_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17304_cast_fp16 = einsum(equation = var_17304_equation_0, values = (var_17132_cast_fp16, var_17279_cast_fp16))[name = tensor("op_17304_cast_fp16")]; + tensor var_17306_equation_0 = const()[name = tensor("op_17306_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17306_cast_fp16 = einsum(equation = var_17306_equation_0, values = (var_17136_cast_fp16, var_17280_cast_fp16))[name = tensor("op_17306_cast_fp16")]; + tensor var_17308_equation_0 = const()[name = tensor("op_17308_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17308_cast_fp16 = einsum(equation = var_17308_equation_0, values = (var_17140_cast_fp16, var_17281_cast_fp16))[name = tensor("op_17308_cast_fp16")]; + tensor var_17310_equation_0 = const()[name = tensor("op_17310_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17310_cast_fp16 = einsum(equation = var_17310_equation_0, values = (var_17144_cast_fp16, var_17282_cast_fp16))[name = tensor("op_17310_cast_fp16")]; + tensor var_17312_equation_0 = const()[name = tensor("op_17312_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17312_cast_fp16 = einsum(equation = var_17312_equation_0, values = (var_17148_cast_fp16, var_17283_cast_fp16))[name = tensor("op_17312_cast_fp16")]; + tensor var_17314_equation_0 = const()[name = tensor("op_17314_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17314_cast_fp16 = einsum(equation = var_17314_equation_0, values = (var_17152_cast_fp16, var_17284_cast_fp16))[name = tensor("op_17314_cast_fp16")]; + tensor var_17316_equation_0 = const()[name = tensor("op_17316_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17316_cast_fp16 = einsum(equation = var_17316_equation_0, values = (var_17156_cast_fp16, var_17285_cast_fp16))[name = tensor("op_17316_cast_fp16")]; + tensor var_17318_equation_0 = const()[name = tensor("op_17318_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17318_cast_fp16 = einsum(equation = var_17318_equation_0, values = (var_17160_cast_fp16, var_17286_cast_fp16))[name = tensor("op_17318_cast_fp16")]; + tensor var_17320_equation_0 = const()[name = tensor("op_17320_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17320_cast_fp16 = einsum(equation = var_17320_equation_0, values = (var_17164_cast_fp16, var_17287_cast_fp16))[name = tensor("op_17320_cast_fp16")]; + tensor var_17322_equation_0 = const()[name = tensor("op_17322_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17322_cast_fp16 = einsum(equation = var_17322_equation_0, values = (var_17168_cast_fp16, var_17288_cast_fp16))[name = tensor("op_17322_cast_fp16")]; + tensor var_17324_equation_0 = const()[name = tensor("op_17324_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17324_cast_fp16 = einsum(equation = var_17324_equation_0, values = (var_17172_cast_fp16, var_17289_cast_fp16))[name = tensor("op_17324_cast_fp16")]; + tensor var_17326_equation_0 = const()[name = tensor("op_17326_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17326_cast_fp16 = einsum(equation = var_17326_equation_0, values = (var_17176_cast_fp16, var_17290_cast_fp16))[name = tensor("op_17326_cast_fp16")]; + tensor var_17328_equation_0 = const()[name = tensor("op_17328_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17328_cast_fp16 = einsum(equation = var_17328_equation_0, values = (var_17180_cast_fp16, var_17291_cast_fp16))[name = tensor("op_17328_cast_fp16")]; + tensor var_17330_equation_0 = const()[name = tensor("op_17330_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17330_cast_fp16 = einsum(equation = var_17330_equation_0, values = (var_17184_cast_fp16, var_17292_cast_fp16))[name = tensor("op_17330_cast_fp16")]; + tensor var_17332_equation_0 = const()[name = tensor("op_17332_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17332_cast_fp16 = einsum(equation = var_17332_equation_0, values = (var_17188_cast_fp16, var_17293_cast_fp16))[name = tensor("op_17332_cast_fp16")]; + tensor var_17334_equation_0 = const()[name = tensor("op_17334_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17334_cast_fp16 = einsum(equation = var_17334_equation_0, values = (var_17192_cast_fp16, var_17294_cast_fp16))[name = tensor("op_17334_cast_fp16")]; + tensor input_271_interleave_0 = const()[name = tensor("input_271_interleave_0"), val = tensor(false)]; + tensor input_271_cast_fp16 = concat(axis = var_2624, interleave = input_271_interleave_0, values = (var_17296_cast_fp16, var_17298_cast_fp16, var_17300_cast_fp16, var_17302_cast_fp16, var_17304_cast_fp16, var_17306_cast_fp16, var_17308_cast_fp16, var_17310_cast_fp16, var_17312_cast_fp16, var_17314_cast_fp16, var_17316_cast_fp16, var_17318_cast_fp16, var_17320_cast_fp16, var_17322_cast_fp16, var_17324_cast_fp16, var_17326_cast_fp16, var_17328_cast_fp16, var_17330_cast_fp16, var_17332_cast_fp16, var_17334_cast_fp16))[name = tensor("input_271_cast_fp16")]; + tensor var_17340 = const()[name = tensor("op_17340"), val = tensor([1, 1])]; + tensor var_17342 = const()[name = tensor("op_17342"), val = tensor([1, 1])]; + tensor var_17344_pad_type_0 = const()[name = tensor("op_17344_pad_type_0"), val = tensor("custom")]; + tensor var_17344_pad_0 = const()[name = tensor("op_17344_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_5_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_5_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1336585600)))]; + tensor down_blocks_2_attentions_1_transformer_blocks_5_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_5_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1339862464)))]; + tensor var_17344_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_5_attn2_to_out_0_bias_to_fp16, dilations = var_17342, groups = var_2624, pad = var_17344_pad_0, pad_type = var_17344_pad_type_0, strides = var_17340, weight = down_blocks_2_attentions_1_transformer_blocks_5_attn2_to_out_0_weight_to_fp16, x = input_271_cast_fp16)[name = tensor("op_17344_cast_fp16")]; + tensor inputs_119_cast_fp16 = add(x = var_17344_cast_fp16, y = inputs_117_cast_fp16)[name = tensor("inputs_119_cast_fp16")]; + tensor input_273_axes_0 = const()[name = tensor("input_273_axes_0"), val = tensor([1])]; + tensor input_273_gamma_0_to_fp16 = const()[name = tensor("input_273_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1339865088)))]; + tensor input_273_beta_0_to_fp16 = const()[name = tensor("input_273_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1339867712)))]; + tensor var_17354_to_fp16 = const()[name = tensor("op_17354_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_273_cast_fp16 = layer_norm(axes = input_273_axes_0, beta = input_273_beta_0_to_fp16, epsilon = var_17354_to_fp16, gamma = input_273_gamma_0_to_fp16, x = inputs_119_cast_fp16)[name = tensor("input_273_cast_fp16")]; + tensor var_17370 = const()[name = tensor("op_17370"), val = tensor([1, 1])]; + tensor var_17372 = const()[name = tensor("op_17372"), val = tensor([1, 1])]; + tensor var_17374_pad_type_0 = const()[name = tensor("op_17374_pad_type_0"), val = tensor("custom")]; + tensor var_17374_pad_0 = const()[name = tensor("op_17374_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_5_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_5_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1339870336)))]; + tensor down_blocks_2_attentions_1_transformer_blocks_5_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_5_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1366084800)))]; + tensor var_17374_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_5_ff_net_0_proj_bias_to_fp16, dilations = var_17372, groups = var_2624, pad = var_17374_pad_0, pad_type = var_17374_pad_type_0, strides = var_17370, weight = down_blocks_2_attentions_1_transformer_blocks_5_ff_net_0_proj_weight_to_fp16, x = input_273_cast_fp16)[name = tensor("op_17374_cast_fp16")]; + tensor var_17375_split_sizes_0 = const()[name = tensor("op_17375_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_17375_axis_0 = const()[name = tensor("op_17375_axis_0"), val = tensor(1)]; + tensor var_17375_cast_fp16_0, tensor var_17375_cast_fp16_1 = split(axis = var_17375_axis_0, split_sizes = var_17375_split_sizes_0, x = var_17374_cast_fp16)[name = tensor("op_17375_cast_fp16")]; + tensor var_17377_mode_0 = const()[name = tensor("op_17377_mode_0"), val = tensor("EXACT")]; + tensor var_17377_cast_fp16 = gelu(mode = var_17377_mode_0, x = var_17375_cast_fp16_1)[name = tensor("op_17377_cast_fp16")]; + tensor input_275_cast_fp16 = mul(x = var_17375_cast_fp16_0, y = var_17377_cast_fp16)[name = tensor("input_275_cast_fp16")]; + tensor var_17381 = const()[name = tensor("op_17381"), val = tensor([1, 1])]; + tensor var_17383 = const()[name = tensor("op_17383"), val = tensor([1, 1])]; + tensor var_17385_pad_type_0 = const()[name = tensor("op_17385_pad_type_0"), val = tensor("custom")]; + tensor var_17385_pad_0 = const()[name = tensor("op_17385_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_5_ff_net_2_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_5_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1366105344)))]; + tensor down_blocks_2_attentions_1_transformer_blocks_5_ff_net_2_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_5_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1379212608)))]; + tensor var_17385_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_5_ff_net_2_bias_to_fp16, dilations = var_17383, groups = var_2624, pad = var_17385_pad_0, pad_type = var_17385_pad_type_0, strides = var_17381, weight = down_blocks_2_attentions_1_transformer_blocks_5_ff_net_2_weight_to_fp16, x = input_275_cast_fp16)[name = tensor("op_17385_cast_fp16")]; + tensor inputs_121_cast_fp16 = add(x = var_17385_cast_fp16, y = inputs_119_cast_fp16)[name = tensor("inputs_121_cast_fp16")]; + tensor hidden_states_173_axes_0 = const()[name = tensor("hidden_states_173_axes_0"), val = tensor([1])]; + tensor hidden_states_173_gamma_0_to_fp16 = const()[name = tensor("hidden_states_173_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1379215232)))]; + tensor hidden_states_173_beta_0_to_fp16 = const()[name = tensor("hidden_states_173_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1379217856)))]; + tensor var_17401_to_fp16 = const()[name = tensor("op_17401_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_173_cast_fp16 = layer_norm(axes = hidden_states_173_axes_0, beta = hidden_states_173_beta_0_to_fp16, epsilon = var_17401_to_fp16, gamma = hidden_states_173_gamma_0_to_fp16, x = inputs_121_cast_fp16)[name = tensor("hidden_states_173_cast_fp16")]; + tensor var_17416 = const()[name = tensor("op_17416"), val = tensor([1, 1])]; + tensor var_17418 = const()[name = tensor("op_17418"), val = tensor([1, 1])]; + tensor q_81_pad_type_0 = const()[name = tensor("q_81_pad_type_0"), val = tensor("custom")]; + tensor q_81_pad_0 = const()[name = tensor("q_81_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_6_attn1_to_q_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_6_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1379220480)))]; + tensor q_81_cast_fp16 = conv(dilations = var_17418, groups = var_2624, pad = q_81_pad_0, pad_type = q_81_pad_type_0, strides = var_17416, weight = down_blocks_2_attentions_1_transformer_blocks_6_attn1_to_q_weight_to_fp16, x = hidden_states_173_cast_fp16)[name = tensor("q_81_cast_fp16")]; + tensor var_17422 = const()[name = tensor("op_17422"), val = tensor([1, 1])]; + tensor var_17424 = const()[name = tensor("op_17424"), val = tensor([1, 1])]; + tensor k_161_pad_type_0 = const()[name = tensor("k_161_pad_type_0"), val = tensor("custom")]; + tensor k_161_pad_0 = const()[name = tensor("k_161_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_6_attn1_to_k_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_6_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1382497344)))]; + tensor k_161_cast_fp16 = conv(dilations = var_17424, groups = var_2624, pad = k_161_pad_0, pad_type = k_161_pad_type_0, strides = var_17422, weight = down_blocks_2_attentions_1_transformer_blocks_6_attn1_to_k_weight_to_fp16, x = hidden_states_173_cast_fp16)[name = tensor("k_161_cast_fp16")]; + tensor var_17428 = const()[name = tensor("op_17428"), val = tensor([1, 1])]; + tensor var_17430 = const()[name = tensor("op_17430"), val = tensor([1, 1])]; + tensor v_81_pad_type_0 = const()[name = tensor("v_81_pad_type_0"), val = tensor("custom")]; + tensor v_81_pad_0 = const()[name = tensor("v_81_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_6_attn1_to_v_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_6_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1385774208)))]; + tensor v_81_cast_fp16 = conv(dilations = var_17430, groups = var_2624, pad = v_81_pad_0, pad_type = v_81_pad_type_0, strides = var_17428, weight = down_blocks_2_attentions_1_transformer_blocks_6_attn1_to_v_weight_to_fp16, x = hidden_states_173_cast_fp16)[name = tensor("v_81_cast_fp16")]; + tensor var_17434_begin_0 = const()[name = tensor("op_17434_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17434_end_0 = const()[name = tensor("op_17434_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_17434_end_mask_0 = const()[name = tensor("op_17434_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17434_cast_fp16 = slice_by_index(begin = var_17434_begin_0, end = var_17434_end_0, end_mask = var_17434_end_mask_0, x = q_81_cast_fp16)[name = tensor("op_17434_cast_fp16")]; + tensor var_17438_begin_0 = const()[name = tensor("op_17438_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_17438_end_0 = const()[name = tensor("op_17438_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_17438_end_mask_0 = const()[name = tensor("op_17438_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17438_cast_fp16 = slice_by_index(begin = var_17438_begin_0, end = var_17438_end_0, end_mask = var_17438_end_mask_0, x = q_81_cast_fp16)[name = tensor("op_17438_cast_fp16")]; + tensor var_17442_begin_0 = const()[name = tensor("op_17442_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_17442_end_0 = const()[name = tensor("op_17442_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_17442_end_mask_0 = const()[name = tensor("op_17442_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17442_cast_fp16 = slice_by_index(begin = var_17442_begin_0, end = var_17442_end_0, end_mask = var_17442_end_mask_0, x = q_81_cast_fp16)[name = tensor("op_17442_cast_fp16")]; + tensor var_17446_begin_0 = const()[name = tensor("op_17446_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_17446_end_0 = const()[name = tensor("op_17446_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_17446_end_mask_0 = const()[name = tensor("op_17446_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17446_cast_fp16 = slice_by_index(begin = var_17446_begin_0, end = var_17446_end_0, end_mask = var_17446_end_mask_0, x = q_81_cast_fp16)[name = tensor("op_17446_cast_fp16")]; + tensor var_17450_begin_0 = const()[name = tensor("op_17450_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_17450_end_0 = const()[name = tensor("op_17450_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_17450_end_mask_0 = const()[name = tensor("op_17450_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17450_cast_fp16 = slice_by_index(begin = var_17450_begin_0, end = var_17450_end_0, end_mask = var_17450_end_mask_0, x = q_81_cast_fp16)[name = tensor("op_17450_cast_fp16")]; + tensor var_17454_begin_0 = const()[name = tensor("op_17454_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_17454_end_0 = const()[name = tensor("op_17454_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_17454_end_mask_0 = const()[name = tensor("op_17454_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17454_cast_fp16 = slice_by_index(begin = var_17454_begin_0, end = var_17454_end_0, end_mask = var_17454_end_mask_0, x = q_81_cast_fp16)[name = tensor("op_17454_cast_fp16")]; + tensor var_17458_begin_0 = const()[name = tensor("op_17458_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_17458_end_0 = const()[name = tensor("op_17458_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_17458_end_mask_0 = const()[name = tensor("op_17458_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17458_cast_fp16 = slice_by_index(begin = var_17458_begin_0, end = var_17458_end_0, end_mask = var_17458_end_mask_0, x = q_81_cast_fp16)[name = tensor("op_17458_cast_fp16")]; + tensor var_17462_begin_0 = const()[name = tensor("op_17462_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_17462_end_0 = const()[name = tensor("op_17462_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_17462_end_mask_0 = const()[name = tensor("op_17462_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17462_cast_fp16 = slice_by_index(begin = var_17462_begin_0, end = var_17462_end_0, end_mask = var_17462_end_mask_0, x = q_81_cast_fp16)[name = tensor("op_17462_cast_fp16")]; + tensor var_17466_begin_0 = const()[name = tensor("op_17466_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_17466_end_0 = const()[name = tensor("op_17466_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_17466_end_mask_0 = const()[name = tensor("op_17466_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17466_cast_fp16 = slice_by_index(begin = var_17466_begin_0, end = var_17466_end_0, end_mask = var_17466_end_mask_0, x = q_81_cast_fp16)[name = tensor("op_17466_cast_fp16")]; + tensor var_17470_begin_0 = const()[name = tensor("op_17470_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_17470_end_0 = const()[name = tensor("op_17470_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_17470_end_mask_0 = const()[name = tensor("op_17470_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17470_cast_fp16 = slice_by_index(begin = var_17470_begin_0, end = var_17470_end_0, end_mask = var_17470_end_mask_0, x = q_81_cast_fp16)[name = tensor("op_17470_cast_fp16")]; + tensor var_17474_begin_0 = const()[name = tensor("op_17474_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_17474_end_0 = const()[name = tensor("op_17474_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_17474_end_mask_0 = const()[name = tensor("op_17474_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17474_cast_fp16 = slice_by_index(begin = var_17474_begin_0, end = var_17474_end_0, end_mask = var_17474_end_mask_0, x = q_81_cast_fp16)[name = tensor("op_17474_cast_fp16")]; + tensor var_17478_begin_0 = const()[name = tensor("op_17478_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_17478_end_0 = const()[name = tensor("op_17478_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_17478_end_mask_0 = const()[name = tensor("op_17478_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17478_cast_fp16 = slice_by_index(begin = var_17478_begin_0, end = var_17478_end_0, end_mask = var_17478_end_mask_0, x = q_81_cast_fp16)[name = tensor("op_17478_cast_fp16")]; + tensor var_17482_begin_0 = const()[name = tensor("op_17482_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_17482_end_0 = const()[name = tensor("op_17482_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_17482_end_mask_0 = const()[name = tensor("op_17482_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17482_cast_fp16 = slice_by_index(begin = var_17482_begin_0, end = var_17482_end_0, end_mask = var_17482_end_mask_0, x = q_81_cast_fp16)[name = tensor("op_17482_cast_fp16")]; + tensor var_17486_begin_0 = const()[name = tensor("op_17486_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_17486_end_0 = const()[name = tensor("op_17486_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_17486_end_mask_0 = const()[name = tensor("op_17486_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17486_cast_fp16 = slice_by_index(begin = var_17486_begin_0, end = var_17486_end_0, end_mask = var_17486_end_mask_0, x = q_81_cast_fp16)[name = tensor("op_17486_cast_fp16")]; + tensor var_17490_begin_0 = const()[name = tensor("op_17490_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_17490_end_0 = const()[name = tensor("op_17490_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_17490_end_mask_0 = const()[name = tensor("op_17490_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17490_cast_fp16 = slice_by_index(begin = var_17490_begin_0, end = var_17490_end_0, end_mask = var_17490_end_mask_0, x = q_81_cast_fp16)[name = tensor("op_17490_cast_fp16")]; + tensor var_17494_begin_0 = const()[name = tensor("op_17494_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_17494_end_0 = const()[name = tensor("op_17494_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_17494_end_mask_0 = const()[name = tensor("op_17494_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17494_cast_fp16 = slice_by_index(begin = var_17494_begin_0, end = var_17494_end_0, end_mask = var_17494_end_mask_0, x = q_81_cast_fp16)[name = tensor("op_17494_cast_fp16")]; + tensor var_17498_begin_0 = const()[name = tensor("op_17498_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_17498_end_0 = const()[name = tensor("op_17498_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_17498_end_mask_0 = const()[name = tensor("op_17498_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17498_cast_fp16 = slice_by_index(begin = var_17498_begin_0, end = var_17498_end_0, end_mask = var_17498_end_mask_0, x = q_81_cast_fp16)[name = tensor("op_17498_cast_fp16")]; + tensor var_17502_begin_0 = const()[name = tensor("op_17502_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_17502_end_0 = const()[name = tensor("op_17502_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_17502_end_mask_0 = const()[name = tensor("op_17502_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17502_cast_fp16 = slice_by_index(begin = var_17502_begin_0, end = var_17502_end_0, end_mask = var_17502_end_mask_0, x = q_81_cast_fp16)[name = tensor("op_17502_cast_fp16")]; + tensor var_17506_begin_0 = const()[name = tensor("op_17506_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_17506_end_0 = const()[name = tensor("op_17506_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_17506_end_mask_0 = const()[name = tensor("op_17506_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17506_cast_fp16 = slice_by_index(begin = var_17506_begin_0, end = var_17506_end_0, end_mask = var_17506_end_mask_0, x = q_81_cast_fp16)[name = tensor("op_17506_cast_fp16")]; + tensor var_17510_begin_0 = const()[name = tensor("op_17510_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_17510_end_0 = const()[name = tensor("op_17510_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_17510_end_mask_0 = const()[name = tensor("op_17510_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17510_cast_fp16 = slice_by_index(begin = var_17510_begin_0, end = var_17510_end_0, end_mask = var_17510_end_mask_0, x = q_81_cast_fp16)[name = tensor("op_17510_cast_fp16")]; + tensor k_163_perm_0 = const()[name = tensor("k_163_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_17517_begin_0 = const()[name = tensor("op_17517_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17517_end_0 = const()[name = tensor("op_17517_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_17517_end_mask_0 = const()[name = tensor("op_17517_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_163_cast_fp16 = transpose(perm = k_163_perm_0, x = k_161_cast_fp16)[name = tensor("transpose_99")]; + tensor var_17517_cast_fp16 = slice_by_index(begin = var_17517_begin_0, end = var_17517_end_0, end_mask = var_17517_end_mask_0, x = k_163_cast_fp16)[name = tensor("op_17517_cast_fp16")]; + tensor var_17521_begin_0 = const()[name = tensor("op_17521_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_17521_end_0 = const()[name = tensor("op_17521_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_17521_end_mask_0 = const()[name = tensor("op_17521_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17521_cast_fp16 = slice_by_index(begin = var_17521_begin_0, end = var_17521_end_0, end_mask = var_17521_end_mask_0, x = k_163_cast_fp16)[name = tensor("op_17521_cast_fp16")]; + tensor var_17525_begin_0 = const()[name = tensor("op_17525_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_17525_end_0 = const()[name = tensor("op_17525_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_17525_end_mask_0 = const()[name = tensor("op_17525_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17525_cast_fp16 = slice_by_index(begin = var_17525_begin_0, end = var_17525_end_0, end_mask = var_17525_end_mask_0, x = k_163_cast_fp16)[name = tensor("op_17525_cast_fp16")]; + tensor var_17529_begin_0 = const()[name = tensor("op_17529_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_17529_end_0 = const()[name = tensor("op_17529_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_17529_end_mask_0 = const()[name = tensor("op_17529_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17529_cast_fp16 = slice_by_index(begin = var_17529_begin_0, end = var_17529_end_0, end_mask = var_17529_end_mask_0, x = k_163_cast_fp16)[name = tensor("op_17529_cast_fp16")]; + tensor var_17533_begin_0 = const()[name = tensor("op_17533_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_17533_end_0 = const()[name = tensor("op_17533_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_17533_end_mask_0 = const()[name = tensor("op_17533_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17533_cast_fp16 = slice_by_index(begin = var_17533_begin_0, end = var_17533_end_0, end_mask = var_17533_end_mask_0, x = k_163_cast_fp16)[name = tensor("op_17533_cast_fp16")]; + tensor var_17537_begin_0 = const()[name = tensor("op_17537_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_17537_end_0 = const()[name = tensor("op_17537_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_17537_end_mask_0 = const()[name = tensor("op_17537_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17537_cast_fp16 = slice_by_index(begin = var_17537_begin_0, end = var_17537_end_0, end_mask = var_17537_end_mask_0, x = k_163_cast_fp16)[name = tensor("op_17537_cast_fp16")]; + tensor var_17541_begin_0 = const()[name = tensor("op_17541_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_17541_end_0 = const()[name = tensor("op_17541_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_17541_end_mask_0 = const()[name = tensor("op_17541_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17541_cast_fp16 = slice_by_index(begin = var_17541_begin_0, end = var_17541_end_0, end_mask = var_17541_end_mask_0, x = k_163_cast_fp16)[name = tensor("op_17541_cast_fp16")]; + tensor var_17545_begin_0 = const()[name = tensor("op_17545_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_17545_end_0 = const()[name = tensor("op_17545_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_17545_end_mask_0 = const()[name = tensor("op_17545_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17545_cast_fp16 = slice_by_index(begin = var_17545_begin_0, end = var_17545_end_0, end_mask = var_17545_end_mask_0, x = k_163_cast_fp16)[name = tensor("op_17545_cast_fp16")]; + tensor var_17549_begin_0 = const()[name = tensor("op_17549_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_17549_end_0 = const()[name = tensor("op_17549_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_17549_end_mask_0 = const()[name = tensor("op_17549_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17549_cast_fp16 = slice_by_index(begin = var_17549_begin_0, end = var_17549_end_0, end_mask = var_17549_end_mask_0, x = k_163_cast_fp16)[name = tensor("op_17549_cast_fp16")]; + tensor var_17553_begin_0 = const()[name = tensor("op_17553_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_17553_end_0 = const()[name = tensor("op_17553_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_17553_end_mask_0 = const()[name = tensor("op_17553_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17553_cast_fp16 = slice_by_index(begin = var_17553_begin_0, end = var_17553_end_0, end_mask = var_17553_end_mask_0, x = k_163_cast_fp16)[name = tensor("op_17553_cast_fp16")]; + tensor var_17557_begin_0 = const()[name = tensor("op_17557_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_17557_end_0 = const()[name = tensor("op_17557_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_17557_end_mask_0 = const()[name = tensor("op_17557_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17557_cast_fp16 = slice_by_index(begin = var_17557_begin_0, end = var_17557_end_0, end_mask = var_17557_end_mask_0, x = k_163_cast_fp16)[name = tensor("op_17557_cast_fp16")]; + tensor var_17561_begin_0 = const()[name = tensor("op_17561_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_17561_end_0 = const()[name = tensor("op_17561_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_17561_end_mask_0 = const()[name = tensor("op_17561_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17561_cast_fp16 = slice_by_index(begin = var_17561_begin_0, end = var_17561_end_0, end_mask = var_17561_end_mask_0, x = k_163_cast_fp16)[name = tensor("op_17561_cast_fp16")]; + tensor var_17565_begin_0 = const()[name = tensor("op_17565_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_17565_end_0 = const()[name = tensor("op_17565_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_17565_end_mask_0 = const()[name = tensor("op_17565_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17565_cast_fp16 = slice_by_index(begin = var_17565_begin_0, end = var_17565_end_0, end_mask = var_17565_end_mask_0, x = k_163_cast_fp16)[name = tensor("op_17565_cast_fp16")]; + tensor var_17569_begin_0 = const()[name = tensor("op_17569_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_17569_end_0 = const()[name = tensor("op_17569_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_17569_end_mask_0 = const()[name = tensor("op_17569_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17569_cast_fp16 = slice_by_index(begin = var_17569_begin_0, end = var_17569_end_0, end_mask = var_17569_end_mask_0, x = k_163_cast_fp16)[name = tensor("op_17569_cast_fp16")]; + tensor var_17573_begin_0 = const()[name = tensor("op_17573_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_17573_end_0 = const()[name = tensor("op_17573_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_17573_end_mask_0 = const()[name = tensor("op_17573_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17573_cast_fp16 = slice_by_index(begin = var_17573_begin_0, end = var_17573_end_0, end_mask = var_17573_end_mask_0, x = k_163_cast_fp16)[name = tensor("op_17573_cast_fp16")]; + tensor var_17577_begin_0 = const()[name = tensor("op_17577_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_17577_end_0 = const()[name = tensor("op_17577_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_17577_end_mask_0 = const()[name = tensor("op_17577_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17577_cast_fp16 = slice_by_index(begin = var_17577_begin_0, end = var_17577_end_0, end_mask = var_17577_end_mask_0, x = k_163_cast_fp16)[name = tensor("op_17577_cast_fp16")]; + tensor var_17581_begin_0 = const()[name = tensor("op_17581_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_17581_end_0 = const()[name = tensor("op_17581_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_17581_end_mask_0 = const()[name = tensor("op_17581_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17581_cast_fp16 = slice_by_index(begin = var_17581_begin_0, end = var_17581_end_0, end_mask = var_17581_end_mask_0, x = k_163_cast_fp16)[name = tensor("op_17581_cast_fp16")]; + tensor var_17585_begin_0 = const()[name = tensor("op_17585_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_17585_end_0 = const()[name = tensor("op_17585_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_17585_end_mask_0 = const()[name = tensor("op_17585_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17585_cast_fp16 = slice_by_index(begin = var_17585_begin_0, end = var_17585_end_0, end_mask = var_17585_end_mask_0, x = k_163_cast_fp16)[name = tensor("op_17585_cast_fp16")]; + tensor var_17589_begin_0 = const()[name = tensor("op_17589_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_17589_end_0 = const()[name = tensor("op_17589_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_17589_end_mask_0 = const()[name = tensor("op_17589_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17589_cast_fp16 = slice_by_index(begin = var_17589_begin_0, end = var_17589_end_0, end_mask = var_17589_end_mask_0, x = k_163_cast_fp16)[name = tensor("op_17589_cast_fp16")]; + tensor var_17593_begin_0 = const()[name = tensor("op_17593_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_17593_end_0 = const()[name = tensor("op_17593_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_17593_end_mask_0 = const()[name = tensor("op_17593_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17593_cast_fp16 = slice_by_index(begin = var_17593_begin_0, end = var_17593_end_0, end_mask = var_17593_end_mask_0, x = k_163_cast_fp16)[name = tensor("op_17593_cast_fp16")]; + tensor var_17595_begin_0 = const()[name = tensor("op_17595_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17595_end_0 = const()[name = tensor("op_17595_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_17595_end_mask_0 = const()[name = tensor("op_17595_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17595_cast_fp16 = slice_by_index(begin = var_17595_begin_0, end = var_17595_end_0, end_mask = var_17595_end_mask_0, x = v_81_cast_fp16)[name = tensor("op_17595_cast_fp16")]; + tensor var_17599_begin_0 = const()[name = tensor("op_17599_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_17599_end_0 = const()[name = tensor("op_17599_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_17599_end_mask_0 = const()[name = tensor("op_17599_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17599_cast_fp16 = slice_by_index(begin = var_17599_begin_0, end = var_17599_end_0, end_mask = var_17599_end_mask_0, x = v_81_cast_fp16)[name = tensor("op_17599_cast_fp16")]; + tensor var_17603_begin_0 = const()[name = tensor("op_17603_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_17603_end_0 = const()[name = tensor("op_17603_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_17603_end_mask_0 = const()[name = tensor("op_17603_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17603_cast_fp16 = slice_by_index(begin = var_17603_begin_0, end = var_17603_end_0, end_mask = var_17603_end_mask_0, x = v_81_cast_fp16)[name = tensor("op_17603_cast_fp16")]; + tensor var_17607_begin_0 = const()[name = tensor("op_17607_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_17607_end_0 = const()[name = tensor("op_17607_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_17607_end_mask_0 = const()[name = tensor("op_17607_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17607_cast_fp16 = slice_by_index(begin = var_17607_begin_0, end = var_17607_end_0, end_mask = var_17607_end_mask_0, x = v_81_cast_fp16)[name = tensor("op_17607_cast_fp16")]; + tensor var_17611_begin_0 = const()[name = tensor("op_17611_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_17611_end_0 = const()[name = tensor("op_17611_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_17611_end_mask_0 = const()[name = tensor("op_17611_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17611_cast_fp16 = slice_by_index(begin = var_17611_begin_0, end = var_17611_end_0, end_mask = var_17611_end_mask_0, x = v_81_cast_fp16)[name = tensor("op_17611_cast_fp16")]; + tensor var_17615_begin_0 = const()[name = tensor("op_17615_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_17615_end_0 = const()[name = tensor("op_17615_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_17615_end_mask_0 = const()[name = tensor("op_17615_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17615_cast_fp16 = slice_by_index(begin = var_17615_begin_0, end = var_17615_end_0, end_mask = var_17615_end_mask_0, x = v_81_cast_fp16)[name = tensor("op_17615_cast_fp16")]; + tensor var_17619_begin_0 = const()[name = tensor("op_17619_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_17619_end_0 = const()[name = tensor("op_17619_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_17619_end_mask_0 = const()[name = tensor("op_17619_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17619_cast_fp16 = slice_by_index(begin = var_17619_begin_0, end = var_17619_end_0, end_mask = var_17619_end_mask_0, x = v_81_cast_fp16)[name = tensor("op_17619_cast_fp16")]; + tensor var_17623_begin_0 = const()[name = tensor("op_17623_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_17623_end_0 = const()[name = tensor("op_17623_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_17623_end_mask_0 = const()[name = tensor("op_17623_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17623_cast_fp16 = slice_by_index(begin = var_17623_begin_0, end = var_17623_end_0, end_mask = var_17623_end_mask_0, x = v_81_cast_fp16)[name = tensor("op_17623_cast_fp16")]; + tensor var_17627_begin_0 = const()[name = tensor("op_17627_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_17627_end_0 = const()[name = tensor("op_17627_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_17627_end_mask_0 = const()[name = tensor("op_17627_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17627_cast_fp16 = slice_by_index(begin = var_17627_begin_0, end = var_17627_end_0, end_mask = var_17627_end_mask_0, x = v_81_cast_fp16)[name = tensor("op_17627_cast_fp16")]; + tensor var_17631_begin_0 = const()[name = tensor("op_17631_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_17631_end_0 = const()[name = tensor("op_17631_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_17631_end_mask_0 = const()[name = tensor("op_17631_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17631_cast_fp16 = slice_by_index(begin = var_17631_begin_0, end = var_17631_end_0, end_mask = var_17631_end_mask_0, x = v_81_cast_fp16)[name = tensor("op_17631_cast_fp16")]; + tensor var_17635_begin_0 = const()[name = tensor("op_17635_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_17635_end_0 = const()[name = tensor("op_17635_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_17635_end_mask_0 = const()[name = tensor("op_17635_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17635_cast_fp16 = slice_by_index(begin = var_17635_begin_0, end = var_17635_end_0, end_mask = var_17635_end_mask_0, x = v_81_cast_fp16)[name = tensor("op_17635_cast_fp16")]; + tensor var_17639_begin_0 = const()[name = tensor("op_17639_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_17639_end_0 = const()[name = tensor("op_17639_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_17639_end_mask_0 = const()[name = tensor("op_17639_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17639_cast_fp16 = slice_by_index(begin = var_17639_begin_0, end = var_17639_end_0, end_mask = var_17639_end_mask_0, x = v_81_cast_fp16)[name = tensor("op_17639_cast_fp16")]; + tensor var_17643_begin_0 = const()[name = tensor("op_17643_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_17643_end_0 = const()[name = tensor("op_17643_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_17643_end_mask_0 = const()[name = tensor("op_17643_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17643_cast_fp16 = slice_by_index(begin = var_17643_begin_0, end = var_17643_end_0, end_mask = var_17643_end_mask_0, x = v_81_cast_fp16)[name = tensor("op_17643_cast_fp16")]; + tensor var_17647_begin_0 = const()[name = tensor("op_17647_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_17647_end_0 = const()[name = tensor("op_17647_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_17647_end_mask_0 = const()[name = tensor("op_17647_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17647_cast_fp16 = slice_by_index(begin = var_17647_begin_0, end = var_17647_end_0, end_mask = var_17647_end_mask_0, x = v_81_cast_fp16)[name = tensor("op_17647_cast_fp16")]; + tensor var_17651_begin_0 = const()[name = tensor("op_17651_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_17651_end_0 = const()[name = tensor("op_17651_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_17651_end_mask_0 = const()[name = tensor("op_17651_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17651_cast_fp16 = slice_by_index(begin = var_17651_begin_0, end = var_17651_end_0, end_mask = var_17651_end_mask_0, x = v_81_cast_fp16)[name = tensor("op_17651_cast_fp16")]; + tensor var_17655_begin_0 = const()[name = tensor("op_17655_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_17655_end_0 = const()[name = tensor("op_17655_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_17655_end_mask_0 = const()[name = tensor("op_17655_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17655_cast_fp16 = slice_by_index(begin = var_17655_begin_0, end = var_17655_end_0, end_mask = var_17655_end_mask_0, x = v_81_cast_fp16)[name = tensor("op_17655_cast_fp16")]; + tensor var_17659_begin_0 = const()[name = tensor("op_17659_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_17659_end_0 = const()[name = tensor("op_17659_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_17659_end_mask_0 = const()[name = tensor("op_17659_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17659_cast_fp16 = slice_by_index(begin = var_17659_begin_0, end = var_17659_end_0, end_mask = var_17659_end_mask_0, x = v_81_cast_fp16)[name = tensor("op_17659_cast_fp16")]; + tensor var_17663_begin_0 = const()[name = tensor("op_17663_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_17663_end_0 = const()[name = tensor("op_17663_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_17663_end_mask_0 = const()[name = tensor("op_17663_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17663_cast_fp16 = slice_by_index(begin = var_17663_begin_0, end = var_17663_end_0, end_mask = var_17663_end_mask_0, x = v_81_cast_fp16)[name = tensor("op_17663_cast_fp16")]; + tensor var_17667_begin_0 = const()[name = tensor("op_17667_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_17667_end_0 = const()[name = tensor("op_17667_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_17667_end_mask_0 = const()[name = tensor("op_17667_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17667_cast_fp16 = slice_by_index(begin = var_17667_begin_0, end = var_17667_end_0, end_mask = var_17667_end_mask_0, x = v_81_cast_fp16)[name = tensor("op_17667_cast_fp16")]; + tensor var_17671_begin_0 = const()[name = tensor("op_17671_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_17671_end_0 = const()[name = tensor("op_17671_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_17671_end_mask_0 = const()[name = tensor("op_17671_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17671_cast_fp16 = slice_by_index(begin = var_17671_begin_0, end = var_17671_end_0, end_mask = var_17671_end_mask_0, x = v_81_cast_fp16)[name = tensor("op_17671_cast_fp16")]; + tensor var_17675_equation_0 = const()[name = tensor("op_17675_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17675_cast_fp16 = einsum(equation = var_17675_equation_0, values = (var_17517_cast_fp16, var_17434_cast_fp16))[name = tensor("op_17675_cast_fp16")]; + tensor var_17676_to_fp16 = const()[name = tensor("op_17676_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1441_cast_fp16 = mul(x = var_17675_cast_fp16, y = var_17676_to_fp16)[name = tensor("aw_1441_cast_fp16")]; + tensor var_17679_equation_0 = const()[name = tensor("op_17679_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17679_cast_fp16 = einsum(equation = var_17679_equation_0, values = (var_17521_cast_fp16, var_17438_cast_fp16))[name = tensor("op_17679_cast_fp16")]; + tensor var_17680_to_fp16 = const()[name = tensor("op_17680_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1443_cast_fp16 = mul(x = var_17679_cast_fp16, y = var_17680_to_fp16)[name = tensor("aw_1443_cast_fp16")]; + tensor var_17683_equation_0 = const()[name = tensor("op_17683_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17683_cast_fp16 = einsum(equation = var_17683_equation_0, values = (var_17525_cast_fp16, var_17442_cast_fp16))[name = tensor("op_17683_cast_fp16")]; + tensor var_17684_to_fp16 = const()[name = tensor("op_17684_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1445_cast_fp16 = mul(x = var_17683_cast_fp16, y = var_17684_to_fp16)[name = tensor("aw_1445_cast_fp16")]; + tensor var_17687_equation_0 = const()[name = tensor("op_17687_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17687_cast_fp16 = einsum(equation = var_17687_equation_0, values = (var_17529_cast_fp16, var_17446_cast_fp16))[name = tensor("op_17687_cast_fp16")]; + tensor var_17688_to_fp16 = const()[name = tensor("op_17688_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1447_cast_fp16 = mul(x = var_17687_cast_fp16, y = var_17688_to_fp16)[name = tensor("aw_1447_cast_fp16")]; + tensor var_17691_equation_0 = const()[name = tensor("op_17691_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17691_cast_fp16 = einsum(equation = var_17691_equation_0, values = (var_17533_cast_fp16, var_17450_cast_fp16))[name = tensor("op_17691_cast_fp16")]; + tensor var_17692_to_fp16 = const()[name = tensor("op_17692_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1449_cast_fp16 = mul(x = var_17691_cast_fp16, y = var_17692_to_fp16)[name = tensor("aw_1449_cast_fp16")]; + tensor var_17695_equation_0 = const()[name = tensor("op_17695_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17695_cast_fp16 = einsum(equation = var_17695_equation_0, values = (var_17537_cast_fp16, var_17454_cast_fp16))[name = tensor("op_17695_cast_fp16")]; + tensor var_17696_to_fp16 = const()[name = tensor("op_17696_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1451_cast_fp16 = mul(x = var_17695_cast_fp16, y = var_17696_to_fp16)[name = tensor("aw_1451_cast_fp16")]; + tensor var_17699_equation_0 = const()[name = tensor("op_17699_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17699_cast_fp16 = einsum(equation = var_17699_equation_0, values = (var_17541_cast_fp16, var_17458_cast_fp16))[name = tensor("op_17699_cast_fp16")]; + tensor var_17700_to_fp16 = const()[name = tensor("op_17700_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1453_cast_fp16 = mul(x = var_17699_cast_fp16, y = var_17700_to_fp16)[name = tensor("aw_1453_cast_fp16")]; + tensor var_17703_equation_0 = const()[name = tensor("op_17703_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17703_cast_fp16 = einsum(equation = var_17703_equation_0, values = (var_17545_cast_fp16, var_17462_cast_fp16))[name = tensor("op_17703_cast_fp16")]; + tensor var_17704_to_fp16 = const()[name = tensor("op_17704_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1455_cast_fp16 = mul(x = var_17703_cast_fp16, y = var_17704_to_fp16)[name = tensor("aw_1455_cast_fp16")]; + tensor var_17707_equation_0 = const()[name = tensor("op_17707_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17707_cast_fp16 = einsum(equation = var_17707_equation_0, values = (var_17549_cast_fp16, var_17466_cast_fp16))[name = tensor("op_17707_cast_fp16")]; + tensor var_17708_to_fp16 = const()[name = tensor("op_17708_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1457_cast_fp16 = mul(x = var_17707_cast_fp16, y = var_17708_to_fp16)[name = tensor("aw_1457_cast_fp16")]; + tensor var_17711_equation_0 = const()[name = tensor("op_17711_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17711_cast_fp16 = einsum(equation = var_17711_equation_0, values = (var_17553_cast_fp16, var_17470_cast_fp16))[name = tensor("op_17711_cast_fp16")]; + tensor var_17712_to_fp16 = const()[name = tensor("op_17712_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1459_cast_fp16 = mul(x = var_17711_cast_fp16, y = var_17712_to_fp16)[name = tensor("aw_1459_cast_fp16")]; + tensor var_17715_equation_0 = const()[name = tensor("op_17715_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17715_cast_fp16 = einsum(equation = var_17715_equation_0, values = (var_17557_cast_fp16, var_17474_cast_fp16))[name = tensor("op_17715_cast_fp16")]; + tensor var_17716_to_fp16 = const()[name = tensor("op_17716_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1461_cast_fp16 = mul(x = var_17715_cast_fp16, y = var_17716_to_fp16)[name = tensor("aw_1461_cast_fp16")]; + tensor var_17719_equation_0 = const()[name = tensor("op_17719_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17719_cast_fp16 = einsum(equation = var_17719_equation_0, values = (var_17561_cast_fp16, var_17478_cast_fp16))[name = tensor("op_17719_cast_fp16")]; + tensor var_17720_to_fp16 = const()[name = tensor("op_17720_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1463_cast_fp16 = mul(x = var_17719_cast_fp16, y = var_17720_to_fp16)[name = tensor("aw_1463_cast_fp16")]; + tensor var_17723_equation_0 = const()[name = tensor("op_17723_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17723_cast_fp16 = einsum(equation = var_17723_equation_0, values = (var_17565_cast_fp16, var_17482_cast_fp16))[name = tensor("op_17723_cast_fp16")]; + tensor var_17724_to_fp16 = const()[name = tensor("op_17724_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1465_cast_fp16 = mul(x = var_17723_cast_fp16, y = var_17724_to_fp16)[name = tensor("aw_1465_cast_fp16")]; + tensor var_17727_equation_0 = const()[name = tensor("op_17727_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17727_cast_fp16 = einsum(equation = var_17727_equation_0, values = (var_17569_cast_fp16, var_17486_cast_fp16))[name = tensor("op_17727_cast_fp16")]; + tensor var_17728_to_fp16 = const()[name = tensor("op_17728_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1467_cast_fp16 = mul(x = var_17727_cast_fp16, y = var_17728_to_fp16)[name = tensor("aw_1467_cast_fp16")]; + tensor var_17731_equation_0 = const()[name = tensor("op_17731_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17731_cast_fp16 = einsum(equation = var_17731_equation_0, values = (var_17573_cast_fp16, var_17490_cast_fp16))[name = tensor("op_17731_cast_fp16")]; + tensor var_17732_to_fp16 = const()[name = tensor("op_17732_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1469_cast_fp16 = mul(x = var_17731_cast_fp16, y = var_17732_to_fp16)[name = tensor("aw_1469_cast_fp16")]; + tensor var_17735_equation_0 = const()[name = tensor("op_17735_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17735_cast_fp16 = einsum(equation = var_17735_equation_0, values = (var_17577_cast_fp16, var_17494_cast_fp16))[name = tensor("op_17735_cast_fp16")]; + tensor var_17736_to_fp16 = const()[name = tensor("op_17736_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1471_cast_fp16 = mul(x = var_17735_cast_fp16, y = var_17736_to_fp16)[name = tensor("aw_1471_cast_fp16")]; + tensor var_17739_equation_0 = const()[name = tensor("op_17739_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17739_cast_fp16 = einsum(equation = var_17739_equation_0, values = (var_17581_cast_fp16, var_17498_cast_fp16))[name = tensor("op_17739_cast_fp16")]; + tensor var_17740_to_fp16 = const()[name = tensor("op_17740_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1473_cast_fp16 = mul(x = var_17739_cast_fp16, y = var_17740_to_fp16)[name = tensor("aw_1473_cast_fp16")]; + tensor var_17743_equation_0 = const()[name = tensor("op_17743_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17743_cast_fp16 = einsum(equation = var_17743_equation_0, values = (var_17585_cast_fp16, var_17502_cast_fp16))[name = tensor("op_17743_cast_fp16")]; + tensor var_17744_to_fp16 = const()[name = tensor("op_17744_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1475_cast_fp16 = mul(x = var_17743_cast_fp16, y = var_17744_to_fp16)[name = tensor("aw_1475_cast_fp16")]; + tensor var_17747_equation_0 = const()[name = tensor("op_17747_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17747_cast_fp16 = einsum(equation = var_17747_equation_0, values = (var_17589_cast_fp16, var_17506_cast_fp16))[name = tensor("op_17747_cast_fp16")]; + tensor var_17748_to_fp16 = const()[name = tensor("op_17748_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1477_cast_fp16 = mul(x = var_17747_cast_fp16, y = var_17748_to_fp16)[name = tensor("aw_1477_cast_fp16")]; + tensor var_17751_equation_0 = const()[name = tensor("op_17751_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17751_cast_fp16 = einsum(equation = var_17751_equation_0, values = (var_17593_cast_fp16, var_17510_cast_fp16))[name = tensor("op_17751_cast_fp16")]; + tensor var_17752_to_fp16 = const()[name = tensor("op_17752_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1479_cast_fp16 = mul(x = var_17751_cast_fp16, y = var_17752_to_fp16)[name = tensor("aw_1479_cast_fp16")]; + tensor var_17754_cast_fp16 = softmax(axis = var_2624, x = aw_1441_cast_fp16)[name = tensor("op_17754_cast_fp16")]; + tensor var_17755_cast_fp16 = softmax(axis = var_2624, x = aw_1443_cast_fp16)[name = tensor("op_17755_cast_fp16")]; + tensor var_17756_cast_fp16 = softmax(axis = var_2624, x = aw_1445_cast_fp16)[name = tensor("op_17756_cast_fp16")]; + tensor var_17757_cast_fp16 = softmax(axis = var_2624, x = aw_1447_cast_fp16)[name = tensor("op_17757_cast_fp16")]; + tensor var_17758_cast_fp16 = softmax(axis = var_2624, x = aw_1449_cast_fp16)[name = tensor("op_17758_cast_fp16")]; + tensor var_17759_cast_fp16 = softmax(axis = var_2624, x = aw_1451_cast_fp16)[name = tensor("op_17759_cast_fp16")]; + tensor var_17760_cast_fp16 = softmax(axis = var_2624, x = aw_1453_cast_fp16)[name = tensor("op_17760_cast_fp16")]; + tensor var_17761_cast_fp16 = softmax(axis = var_2624, x = aw_1455_cast_fp16)[name = tensor("op_17761_cast_fp16")]; + tensor var_17762_cast_fp16 = softmax(axis = var_2624, x = aw_1457_cast_fp16)[name = tensor("op_17762_cast_fp16")]; + tensor var_17763_cast_fp16 = softmax(axis = var_2624, x = aw_1459_cast_fp16)[name = tensor("op_17763_cast_fp16")]; + tensor var_17764_cast_fp16 = softmax(axis = var_2624, x = aw_1461_cast_fp16)[name = tensor("op_17764_cast_fp16")]; + tensor var_17765_cast_fp16 = softmax(axis = var_2624, x = aw_1463_cast_fp16)[name = tensor("op_17765_cast_fp16")]; + tensor var_17766_cast_fp16 = softmax(axis = var_2624, x = aw_1465_cast_fp16)[name = tensor("op_17766_cast_fp16")]; + tensor var_17767_cast_fp16 = softmax(axis = var_2624, x = aw_1467_cast_fp16)[name = tensor("op_17767_cast_fp16")]; + tensor var_17768_cast_fp16 = softmax(axis = var_2624, x = aw_1469_cast_fp16)[name = tensor("op_17768_cast_fp16")]; + tensor var_17769_cast_fp16 = softmax(axis = var_2624, x = aw_1471_cast_fp16)[name = tensor("op_17769_cast_fp16")]; + tensor var_17770_cast_fp16 = softmax(axis = var_2624, x = aw_1473_cast_fp16)[name = tensor("op_17770_cast_fp16")]; + tensor var_17771_cast_fp16 = softmax(axis = var_2624, x = aw_1475_cast_fp16)[name = tensor("op_17771_cast_fp16")]; + tensor var_17772_cast_fp16 = softmax(axis = var_2624, x = aw_1477_cast_fp16)[name = tensor("op_17772_cast_fp16")]; + tensor var_17773_cast_fp16 = softmax(axis = var_2624, x = aw_1479_cast_fp16)[name = tensor("op_17773_cast_fp16")]; + tensor var_17775_equation_0 = const()[name = tensor("op_17775_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17775_cast_fp16 = einsum(equation = var_17775_equation_0, values = (var_17595_cast_fp16, var_17754_cast_fp16))[name = tensor("op_17775_cast_fp16")]; + tensor var_17777_equation_0 = const()[name = tensor("op_17777_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17777_cast_fp16 = einsum(equation = var_17777_equation_0, values = (var_17599_cast_fp16, var_17755_cast_fp16))[name = tensor("op_17777_cast_fp16")]; + tensor var_17779_equation_0 = const()[name = tensor("op_17779_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17779_cast_fp16 = einsum(equation = var_17779_equation_0, values = (var_17603_cast_fp16, var_17756_cast_fp16))[name = tensor("op_17779_cast_fp16")]; + tensor var_17781_equation_0 = const()[name = tensor("op_17781_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17781_cast_fp16 = einsum(equation = var_17781_equation_0, values = (var_17607_cast_fp16, var_17757_cast_fp16))[name = tensor("op_17781_cast_fp16")]; + tensor var_17783_equation_0 = const()[name = tensor("op_17783_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17783_cast_fp16 = einsum(equation = var_17783_equation_0, values = (var_17611_cast_fp16, var_17758_cast_fp16))[name = tensor("op_17783_cast_fp16")]; + tensor var_17785_equation_0 = const()[name = tensor("op_17785_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17785_cast_fp16 = einsum(equation = var_17785_equation_0, values = (var_17615_cast_fp16, var_17759_cast_fp16))[name = tensor("op_17785_cast_fp16")]; + tensor var_17787_equation_0 = const()[name = tensor("op_17787_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17787_cast_fp16 = einsum(equation = var_17787_equation_0, values = (var_17619_cast_fp16, var_17760_cast_fp16))[name = tensor("op_17787_cast_fp16")]; + tensor var_17789_equation_0 = const()[name = tensor("op_17789_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17789_cast_fp16 = einsum(equation = var_17789_equation_0, values = (var_17623_cast_fp16, var_17761_cast_fp16))[name = tensor("op_17789_cast_fp16")]; + tensor var_17791_equation_0 = const()[name = tensor("op_17791_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17791_cast_fp16 = einsum(equation = var_17791_equation_0, values = (var_17627_cast_fp16, var_17762_cast_fp16))[name = tensor("op_17791_cast_fp16")]; + tensor var_17793_equation_0 = const()[name = tensor("op_17793_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17793_cast_fp16 = einsum(equation = var_17793_equation_0, values = (var_17631_cast_fp16, var_17763_cast_fp16))[name = tensor("op_17793_cast_fp16")]; + tensor var_17795_equation_0 = const()[name = tensor("op_17795_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17795_cast_fp16 = einsum(equation = var_17795_equation_0, values = (var_17635_cast_fp16, var_17764_cast_fp16))[name = tensor("op_17795_cast_fp16")]; + tensor var_17797_equation_0 = const()[name = tensor("op_17797_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17797_cast_fp16 = einsum(equation = var_17797_equation_0, values = (var_17639_cast_fp16, var_17765_cast_fp16))[name = tensor("op_17797_cast_fp16")]; + tensor var_17799_equation_0 = const()[name = tensor("op_17799_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17799_cast_fp16 = einsum(equation = var_17799_equation_0, values = (var_17643_cast_fp16, var_17766_cast_fp16))[name = tensor("op_17799_cast_fp16")]; + tensor var_17801_equation_0 = const()[name = tensor("op_17801_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17801_cast_fp16 = einsum(equation = var_17801_equation_0, values = (var_17647_cast_fp16, var_17767_cast_fp16))[name = tensor("op_17801_cast_fp16")]; + tensor var_17803_equation_0 = const()[name = tensor("op_17803_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17803_cast_fp16 = einsum(equation = var_17803_equation_0, values = (var_17651_cast_fp16, var_17768_cast_fp16))[name = tensor("op_17803_cast_fp16")]; + tensor var_17805_equation_0 = const()[name = tensor("op_17805_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17805_cast_fp16 = einsum(equation = var_17805_equation_0, values = (var_17655_cast_fp16, var_17769_cast_fp16))[name = tensor("op_17805_cast_fp16")]; + tensor var_17807_equation_0 = const()[name = tensor("op_17807_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17807_cast_fp16 = einsum(equation = var_17807_equation_0, values = (var_17659_cast_fp16, var_17770_cast_fp16))[name = tensor("op_17807_cast_fp16")]; + tensor var_17809_equation_0 = const()[name = tensor("op_17809_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17809_cast_fp16 = einsum(equation = var_17809_equation_0, values = (var_17663_cast_fp16, var_17771_cast_fp16))[name = tensor("op_17809_cast_fp16")]; + tensor var_17811_equation_0 = const()[name = tensor("op_17811_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17811_cast_fp16 = einsum(equation = var_17811_equation_0, values = (var_17667_cast_fp16, var_17772_cast_fp16))[name = tensor("op_17811_cast_fp16")]; + tensor var_17813_equation_0 = const()[name = tensor("op_17813_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17813_cast_fp16 = einsum(equation = var_17813_equation_0, values = (var_17671_cast_fp16, var_17773_cast_fp16))[name = tensor("op_17813_cast_fp16")]; + tensor input_277_interleave_0 = const()[name = tensor("input_277_interleave_0"), val = tensor(false)]; + tensor input_277_cast_fp16 = concat(axis = var_2624, interleave = input_277_interleave_0, values = (var_17775_cast_fp16, var_17777_cast_fp16, var_17779_cast_fp16, var_17781_cast_fp16, var_17783_cast_fp16, var_17785_cast_fp16, var_17787_cast_fp16, var_17789_cast_fp16, var_17791_cast_fp16, var_17793_cast_fp16, var_17795_cast_fp16, var_17797_cast_fp16, var_17799_cast_fp16, var_17801_cast_fp16, var_17803_cast_fp16, var_17805_cast_fp16, var_17807_cast_fp16, var_17809_cast_fp16, var_17811_cast_fp16, var_17813_cast_fp16))[name = tensor("input_277_cast_fp16")]; + tensor var_17819 = const()[name = tensor("op_17819"), val = tensor([1, 1])]; + tensor var_17821 = const()[name = tensor("op_17821"), val = tensor([1, 1])]; + tensor var_17823_pad_type_0 = const()[name = tensor("op_17823_pad_type_0"), val = tensor("custom")]; + tensor var_17823_pad_0 = const()[name = tensor("op_17823_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_6_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_6_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1389051072)))]; + tensor down_blocks_2_attentions_1_transformer_blocks_6_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_6_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1392327936)))]; + tensor var_17823_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_6_attn1_to_out_0_bias_to_fp16, dilations = var_17821, groups = var_2624, pad = var_17823_pad_0, pad_type = var_17823_pad_type_0, strides = var_17819, weight = down_blocks_2_attentions_1_transformer_blocks_6_attn1_to_out_0_weight_to_fp16, x = input_277_cast_fp16)[name = tensor("op_17823_cast_fp16")]; + tensor inputs_123_cast_fp16 = add(x = var_17823_cast_fp16, y = inputs_121_cast_fp16)[name = tensor("inputs_123_cast_fp16")]; + tensor hidden_states_175_axes_0 = const()[name = tensor("hidden_states_175_axes_0"), val = tensor([1])]; + tensor hidden_states_175_gamma_0_to_fp16 = const()[name = tensor("hidden_states_175_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1392330560)))]; + tensor hidden_states_175_beta_0_to_fp16 = const()[name = tensor("hidden_states_175_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1392333184)))]; + tensor var_17833_to_fp16 = const()[name = tensor("op_17833_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_175_cast_fp16 = layer_norm(axes = hidden_states_175_axes_0, beta = hidden_states_175_beta_0_to_fp16, epsilon = var_17833_to_fp16, gamma = hidden_states_175_gamma_0_to_fp16, x = inputs_123_cast_fp16)[name = tensor("hidden_states_175_cast_fp16")]; + tensor var_17848 = const()[name = tensor("op_17848"), val = tensor([1, 1])]; + tensor var_17850 = const()[name = tensor("op_17850"), val = tensor([1, 1])]; + tensor q_83_pad_type_0 = const()[name = tensor("q_83_pad_type_0"), val = tensor("custom")]; + tensor q_83_pad_0 = const()[name = tensor("q_83_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_6_attn2_to_q_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_6_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1392335808)))]; + tensor q_83_cast_fp16 = conv(dilations = var_17850, groups = var_2624, pad = q_83_pad_0, pad_type = q_83_pad_type_0, strides = var_17848, weight = down_blocks_2_attentions_1_transformer_blocks_6_attn2_to_q_weight_to_fp16, x = hidden_states_175_cast_fp16)[name = tensor("q_83_cast_fp16")]; + tensor var_17854 = const()[name = tensor("op_17854"), val = tensor([1, 1])]; + tensor var_17856 = const()[name = tensor("op_17856"), val = tensor([1, 1])]; + tensor k_165_pad_type_0 = const()[name = tensor("k_165_pad_type_0"), val = tensor("custom")]; + tensor k_165_pad_0 = const()[name = tensor("k_165_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_6_attn2_to_k_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_6_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1395612672)))]; + tensor k_165_cast_fp16 = conv(dilations = var_17856, groups = var_2624, pad = k_165_pad_0, pad_type = k_165_pad_type_0, strides = var_17854, weight = down_blocks_2_attentions_1_transformer_blocks_6_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_165_cast_fp16")]; + tensor var_17860 = const()[name = tensor("op_17860"), val = tensor([1, 1])]; + tensor var_17862 = const()[name = tensor("op_17862"), val = tensor([1, 1])]; + tensor v_83_pad_type_0 = const()[name = tensor("v_83_pad_type_0"), val = tensor("custom")]; + tensor v_83_pad_0 = const()[name = tensor("v_83_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_6_attn2_to_v_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_6_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1400855616)))]; + tensor v_83_cast_fp16 = conv(dilations = var_17862, groups = var_2624, pad = v_83_pad_0, pad_type = v_83_pad_type_0, strides = var_17860, weight = down_blocks_2_attentions_1_transformer_blocks_6_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_83_cast_fp16")]; + tensor var_17866_begin_0 = const()[name = tensor("op_17866_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17866_end_0 = const()[name = tensor("op_17866_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_17866_end_mask_0 = const()[name = tensor("op_17866_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17866_cast_fp16 = slice_by_index(begin = var_17866_begin_0, end = var_17866_end_0, end_mask = var_17866_end_mask_0, x = q_83_cast_fp16)[name = tensor("op_17866_cast_fp16")]; + tensor var_17870_begin_0 = const()[name = tensor("op_17870_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_17870_end_0 = const()[name = tensor("op_17870_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_17870_end_mask_0 = const()[name = tensor("op_17870_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17870_cast_fp16 = slice_by_index(begin = var_17870_begin_0, end = var_17870_end_0, end_mask = var_17870_end_mask_0, x = q_83_cast_fp16)[name = tensor("op_17870_cast_fp16")]; + tensor var_17874_begin_0 = const()[name = tensor("op_17874_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_17874_end_0 = const()[name = tensor("op_17874_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_17874_end_mask_0 = const()[name = tensor("op_17874_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17874_cast_fp16 = slice_by_index(begin = var_17874_begin_0, end = var_17874_end_0, end_mask = var_17874_end_mask_0, x = q_83_cast_fp16)[name = tensor("op_17874_cast_fp16")]; + tensor var_17878_begin_0 = const()[name = tensor("op_17878_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_17878_end_0 = const()[name = tensor("op_17878_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_17878_end_mask_0 = const()[name = tensor("op_17878_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17878_cast_fp16 = slice_by_index(begin = var_17878_begin_0, end = var_17878_end_0, end_mask = var_17878_end_mask_0, x = q_83_cast_fp16)[name = tensor("op_17878_cast_fp16")]; + tensor var_17882_begin_0 = const()[name = tensor("op_17882_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_17882_end_0 = const()[name = tensor("op_17882_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_17882_end_mask_0 = const()[name = tensor("op_17882_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17882_cast_fp16 = slice_by_index(begin = var_17882_begin_0, end = var_17882_end_0, end_mask = var_17882_end_mask_0, x = q_83_cast_fp16)[name = tensor("op_17882_cast_fp16")]; + tensor var_17886_begin_0 = const()[name = tensor("op_17886_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_17886_end_0 = const()[name = tensor("op_17886_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_17886_end_mask_0 = const()[name = tensor("op_17886_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17886_cast_fp16 = slice_by_index(begin = var_17886_begin_0, end = var_17886_end_0, end_mask = var_17886_end_mask_0, x = q_83_cast_fp16)[name = tensor("op_17886_cast_fp16")]; + tensor var_17890_begin_0 = const()[name = tensor("op_17890_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_17890_end_0 = const()[name = tensor("op_17890_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_17890_end_mask_0 = const()[name = tensor("op_17890_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17890_cast_fp16 = slice_by_index(begin = var_17890_begin_0, end = var_17890_end_0, end_mask = var_17890_end_mask_0, x = q_83_cast_fp16)[name = tensor("op_17890_cast_fp16")]; + tensor var_17894_begin_0 = const()[name = tensor("op_17894_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_17894_end_0 = const()[name = tensor("op_17894_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_17894_end_mask_0 = const()[name = tensor("op_17894_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17894_cast_fp16 = slice_by_index(begin = var_17894_begin_0, end = var_17894_end_0, end_mask = var_17894_end_mask_0, x = q_83_cast_fp16)[name = tensor("op_17894_cast_fp16")]; + tensor var_17898_begin_0 = const()[name = tensor("op_17898_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_17898_end_0 = const()[name = tensor("op_17898_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_17898_end_mask_0 = const()[name = tensor("op_17898_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17898_cast_fp16 = slice_by_index(begin = var_17898_begin_0, end = var_17898_end_0, end_mask = var_17898_end_mask_0, x = q_83_cast_fp16)[name = tensor("op_17898_cast_fp16")]; + tensor var_17902_begin_0 = const()[name = tensor("op_17902_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_17902_end_0 = const()[name = tensor("op_17902_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_17902_end_mask_0 = const()[name = tensor("op_17902_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17902_cast_fp16 = slice_by_index(begin = var_17902_begin_0, end = var_17902_end_0, end_mask = var_17902_end_mask_0, x = q_83_cast_fp16)[name = tensor("op_17902_cast_fp16")]; + tensor var_17906_begin_0 = const()[name = tensor("op_17906_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_17906_end_0 = const()[name = tensor("op_17906_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_17906_end_mask_0 = const()[name = tensor("op_17906_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17906_cast_fp16 = slice_by_index(begin = var_17906_begin_0, end = var_17906_end_0, end_mask = var_17906_end_mask_0, x = q_83_cast_fp16)[name = tensor("op_17906_cast_fp16")]; + tensor var_17910_begin_0 = const()[name = tensor("op_17910_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_17910_end_0 = const()[name = tensor("op_17910_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_17910_end_mask_0 = const()[name = tensor("op_17910_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17910_cast_fp16 = slice_by_index(begin = var_17910_begin_0, end = var_17910_end_0, end_mask = var_17910_end_mask_0, x = q_83_cast_fp16)[name = tensor("op_17910_cast_fp16")]; + tensor var_17914_begin_0 = const()[name = tensor("op_17914_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_17914_end_0 = const()[name = tensor("op_17914_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_17914_end_mask_0 = const()[name = tensor("op_17914_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17914_cast_fp16 = slice_by_index(begin = var_17914_begin_0, end = var_17914_end_0, end_mask = var_17914_end_mask_0, x = q_83_cast_fp16)[name = tensor("op_17914_cast_fp16")]; + tensor var_17918_begin_0 = const()[name = tensor("op_17918_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_17918_end_0 = const()[name = tensor("op_17918_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_17918_end_mask_0 = const()[name = tensor("op_17918_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17918_cast_fp16 = slice_by_index(begin = var_17918_begin_0, end = var_17918_end_0, end_mask = var_17918_end_mask_0, x = q_83_cast_fp16)[name = tensor("op_17918_cast_fp16")]; + tensor var_17922_begin_0 = const()[name = tensor("op_17922_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_17922_end_0 = const()[name = tensor("op_17922_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_17922_end_mask_0 = const()[name = tensor("op_17922_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17922_cast_fp16 = slice_by_index(begin = var_17922_begin_0, end = var_17922_end_0, end_mask = var_17922_end_mask_0, x = q_83_cast_fp16)[name = tensor("op_17922_cast_fp16")]; + tensor var_17926_begin_0 = const()[name = tensor("op_17926_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_17926_end_0 = const()[name = tensor("op_17926_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_17926_end_mask_0 = const()[name = tensor("op_17926_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17926_cast_fp16 = slice_by_index(begin = var_17926_begin_0, end = var_17926_end_0, end_mask = var_17926_end_mask_0, x = q_83_cast_fp16)[name = tensor("op_17926_cast_fp16")]; + tensor var_17930_begin_0 = const()[name = tensor("op_17930_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_17930_end_0 = const()[name = tensor("op_17930_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_17930_end_mask_0 = const()[name = tensor("op_17930_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17930_cast_fp16 = slice_by_index(begin = var_17930_begin_0, end = var_17930_end_0, end_mask = var_17930_end_mask_0, x = q_83_cast_fp16)[name = tensor("op_17930_cast_fp16")]; + tensor var_17934_begin_0 = const()[name = tensor("op_17934_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_17934_end_0 = const()[name = tensor("op_17934_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_17934_end_mask_0 = const()[name = tensor("op_17934_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17934_cast_fp16 = slice_by_index(begin = var_17934_begin_0, end = var_17934_end_0, end_mask = var_17934_end_mask_0, x = q_83_cast_fp16)[name = tensor("op_17934_cast_fp16")]; + tensor var_17938_begin_0 = const()[name = tensor("op_17938_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_17938_end_0 = const()[name = tensor("op_17938_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_17938_end_mask_0 = const()[name = tensor("op_17938_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17938_cast_fp16 = slice_by_index(begin = var_17938_begin_0, end = var_17938_end_0, end_mask = var_17938_end_mask_0, x = q_83_cast_fp16)[name = tensor("op_17938_cast_fp16")]; + tensor var_17942_begin_0 = const()[name = tensor("op_17942_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_17942_end_0 = const()[name = tensor("op_17942_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_17942_end_mask_0 = const()[name = tensor("op_17942_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17942_cast_fp16 = slice_by_index(begin = var_17942_begin_0, end = var_17942_end_0, end_mask = var_17942_end_mask_0, x = q_83_cast_fp16)[name = tensor("op_17942_cast_fp16")]; + tensor k_167_perm_0 = const()[name = tensor("k_167_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_17949_begin_0 = const()[name = tensor("op_17949_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17949_end_0 = const()[name = tensor("op_17949_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_17949_end_mask_0 = const()[name = tensor("op_17949_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_167_cast_fp16 = transpose(perm = k_167_perm_0, x = k_165_cast_fp16)[name = tensor("transpose_98")]; + tensor var_17949_cast_fp16 = slice_by_index(begin = var_17949_begin_0, end = var_17949_end_0, end_mask = var_17949_end_mask_0, x = k_167_cast_fp16)[name = tensor("op_17949_cast_fp16")]; + tensor var_17953_begin_0 = const()[name = tensor("op_17953_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_17953_end_0 = const()[name = tensor("op_17953_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_17953_end_mask_0 = const()[name = tensor("op_17953_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17953_cast_fp16 = slice_by_index(begin = var_17953_begin_0, end = var_17953_end_0, end_mask = var_17953_end_mask_0, x = k_167_cast_fp16)[name = tensor("op_17953_cast_fp16")]; + tensor var_17957_begin_0 = const()[name = tensor("op_17957_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_17957_end_0 = const()[name = tensor("op_17957_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_17957_end_mask_0 = const()[name = tensor("op_17957_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17957_cast_fp16 = slice_by_index(begin = var_17957_begin_0, end = var_17957_end_0, end_mask = var_17957_end_mask_0, x = k_167_cast_fp16)[name = tensor("op_17957_cast_fp16")]; + tensor var_17961_begin_0 = const()[name = tensor("op_17961_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_17961_end_0 = const()[name = tensor("op_17961_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_17961_end_mask_0 = const()[name = tensor("op_17961_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17961_cast_fp16 = slice_by_index(begin = var_17961_begin_0, end = var_17961_end_0, end_mask = var_17961_end_mask_0, x = k_167_cast_fp16)[name = tensor("op_17961_cast_fp16")]; + tensor var_17965_begin_0 = const()[name = tensor("op_17965_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_17965_end_0 = const()[name = tensor("op_17965_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_17965_end_mask_0 = const()[name = tensor("op_17965_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17965_cast_fp16 = slice_by_index(begin = var_17965_begin_0, end = var_17965_end_0, end_mask = var_17965_end_mask_0, x = k_167_cast_fp16)[name = tensor("op_17965_cast_fp16")]; + tensor var_17969_begin_0 = const()[name = tensor("op_17969_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_17969_end_0 = const()[name = tensor("op_17969_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_17969_end_mask_0 = const()[name = tensor("op_17969_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17969_cast_fp16 = slice_by_index(begin = var_17969_begin_0, end = var_17969_end_0, end_mask = var_17969_end_mask_0, x = k_167_cast_fp16)[name = tensor("op_17969_cast_fp16")]; + tensor var_17973_begin_0 = const()[name = tensor("op_17973_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_17973_end_0 = const()[name = tensor("op_17973_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_17973_end_mask_0 = const()[name = tensor("op_17973_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17973_cast_fp16 = slice_by_index(begin = var_17973_begin_0, end = var_17973_end_0, end_mask = var_17973_end_mask_0, x = k_167_cast_fp16)[name = tensor("op_17973_cast_fp16")]; + tensor var_17977_begin_0 = const()[name = tensor("op_17977_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_17977_end_0 = const()[name = tensor("op_17977_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_17977_end_mask_0 = const()[name = tensor("op_17977_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17977_cast_fp16 = slice_by_index(begin = var_17977_begin_0, end = var_17977_end_0, end_mask = var_17977_end_mask_0, x = k_167_cast_fp16)[name = tensor("op_17977_cast_fp16")]; + tensor var_17981_begin_0 = const()[name = tensor("op_17981_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_17981_end_0 = const()[name = tensor("op_17981_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_17981_end_mask_0 = const()[name = tensor("op_17981_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17981_cast_fp16 = slice_by_index(begin = var_17981_begin_0, end = var_17981_end_0, end_mask = var_17981_end_mask_0, x = k_167_cast_fp16)[name = tensor("op_17981_cast_fp16")]; + tensor var_17985_begin_0 = const()[name = tensor("op_17985_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_17985_end_0 = const()[name = tensor("op_17985_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_17985_end_mask_0 = const()[name = tensor("op_17985_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17985_cast_fp16 = slice_by_index(begin = var_17985_begin_0, end = var_17985_end_0, end_mask = var_17985_end_mask_0, x = k_167_cast_fp16)[name = tensor("op_17985_cast_fp16")]; + tensor var_17989_begin_0 = const()[name = tensor("op_17989_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_17989_end_0 = const()[name = tensor("op_17989_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_17989_end_mask_0 = const()[name = tensor("op_17989_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17989_cast_fp16 = slice_by_index(begin = var_17989_begin_0, end = var_17989_end_0, end_mask = var_17989_end_mask_0, x = k_167_cast_fp16)[name = tensor("op_17989_cast_fp16")]; + tensor var_17993_begin_0 = const()[name = tensor("op_17993_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_17993_end_0 = const()[name = tensor("op_17993_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_17993_end_mask_0 = const()[name = tensor("op_17993_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17993_cast_fp16 = slice_by_index(begin = var_17993_begin_0, end = var_17993_end_0, end_mask = var_17993_end_mask_0, x = k_167_cast_fp16)[name = tensor("op_17993_cast_fp16")]; + tensor var_17997_begin_0 = const()[name = tensor("op_17997_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_17997_end_0 = const()[name = tensor("op_17997_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_17997_end_mask_0 = const()[name = tensor("op_17997_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17997_cast_fp16 = slice_by_index(begin = var_17997_begin_0, end = var_17997_end_0, end_mask = var_17997_end_mask_0, x = k_167_cast_fp16)[name = tensor("op_17997_cast_fp16")]; + tensor var_18001_begin_0 = const()[name = tensor("op_18001_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_18001_end_0 = const()[name = tensor("op_18001_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_18001_end_mask_0 = const()[name = tensor("op_18001_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18001_cast_fp16 = slice_by_index(begin = var_18001_begin_0, end = var_18001_end_0, end_mask = var_18001_end_mask_0, x = k_167_cast_fp16)[name = tensor("op_18001_cast_fp16")]; + tensor var_18005_begin_0 = const()[name = tensor("op_18005_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_18005_end_0 = const()[name = tensor("op_18005_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_18005_end_mask_0 = const()[name = tensor("op_18005_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18005_cast_fp16 = slice_by_index(begin = var_18005_begin_0, end = var_18005_end_0, end_mask = var_18005_end_mask_0, x = k_167_cast_fp16)[name = tensor("op_18005_cast_fp16")]; + tensor var_18009_begin_0 = const()[name = tensor("op_18009_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_18009_end_0 = const()[name = tensor("op_18009_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_18009_end_mask_0 = const()[name = tensor("op_18009_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18009_cast_fp16 = slice_by_index(begin = var_18009_begin_0, end = var_18009_end_0, end_mask = var_18009_end_mask_0, x = k_167_cast_fp16)[name = tensor("op_18009_cast_fp16")]; + tensor var_18013_begin_0 = const()[name = tensor("op_18013_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_18013_end_0 = const()[name = tensor("op_18013_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_18013_end_mask_0 = const()[name = tensor("op_18013_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18013_cast_fp16 = slice_by_index(begin = var_18013_begin_0, end = var_18013_end_0, end_mask = var_18013_end_mask_0, x = k_167_cast_fp16)[name = tensor("op_18013_cast_fp16")]; + tensor var_18017_begin_0 = const()[name = tensor("op_18017_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_18017_end_0 = const()[name = tensor("op_18017_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_18017_end_mask_0 = const()[name = tensor("op_18017_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18017_cast_fp16 = slice_by_index(begin = var_18017_begin_0, end = var_18017_end_0, end_mask = var_18017_end_mask_0, x = k_167_cast_fp16)[name = tensor("op_18017_cast_fp16")]; + tensor var_18021_begin_0 = const()[name = tensor("op_18021_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_18021_end_0 = const()[name = tensor("op_18021_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_18021_end_mask_0 = const()[name = tensor("op_18021_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18021_cast_fp16 = slice_by_index(begin = var_18021_begin_0, end = var_18021_end_0, end_mask = var_18021_end_mask_0, x = k_167_cast_fp16)[name = tensor("op_18021_cast_fp16")]; + tensor var_18025_begin_0 = const()[name = tensor("op_18025_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_18025_end_0 = const()[name = tensor("op_18025_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_18025_end_mask_0 = const()[name = tensor("op_18025_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18025_cast_fp16 = slice_by_index(begin = var_18025_begin_0, end = var_18025_end_0, end_mask = var_18025_end_mask_0, x = k_167_cast_fp16)[name = tensor("op_18025_cast_fp16")]; + tensor var_18027_begin_0 = const()[name = tensor("op_18027_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_18027_end_0 = const()[name = tensor("op_18027_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_18027_end_mask_0 = const()[name = tensor("op_18027_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18027_cast_fp16 = slice_by_index(begin = var_18027_begin_0, end = var_18027_end_0, end_mask = var_18027_end_mask_0, x = v_83_cast_fp16)[name = tensor("op_18027_cast_fp16")]; + tensor var_18031_begin_0 = const()[name = tensor("op_18031_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_18031_end_0 = const()[name = tensor("op_18031_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_18031_end_mask_0 = const()[name = tensor("op_18031_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18031_cast_fp16 = slice_by_index(begin = var_18031_begin_0, end = var_18031_end_0, end_mask = var_18031_end_mask_0, x = v_83_cast_fp16)[name = tensor("op_18031_cast_fp16")]; + tensor var_18035_begin_0 = const()[name = tensor("op_18035_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_18035_end_0 = const()[name = tensor("op_18035_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_18035_end_mask_0 = const()[name = tensor("op_18035_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18035_cast_fp16 = slice_by_index(begin = var_18035_begin_0, end = var_18035_end_0, end_mask = var_18035_end_mask_0, x = v_83_cast_fp16)[name = tensor("op_18035_cast_fp16")]; + tensor var_18039_begin_0 = const()[name = tensor("op_18039_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_18039_end_0 = const()[name = tensor("op_18039_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_18039_end_mask_0 = const()[name = tensor("op_18039_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18039_cast_fp16 = slice_by_index(begin = var_18039_begin_0, end = var_18039_end_0, end_mask = var_18039_end_mask_0, x = v_83_cast_fp16)[name = tensor("op_18039_cast_fp16")]; + tensor var_18043_begin_0 = const()[name = tensor("op_18043_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_18043_end_0 = const()[name = tensor("op_18043_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_18043_end_mask_0 = const()[name = tensor("op_18043_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18043_cast_fp16 = slice_by_index(begin = var_18043_begin_0, end = var_18043_end_0, end_mask = var_18043_end_mask_0, x = v_83_cast_fp16)[name = tensor("op_18043_cast_fp16")]; + tensor var_18047_begin_0 = const()[name = tensor("op_18047_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_18047_end_0 = const()[name = tensor("op_18047_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_18047_end_mask_0 = const()[name = tensor("op_18047_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18047_cast_fp16 = slice_by_index(begin = var_18047_begin_0, end = var_18047_end_0, end_mask = var_18047_end_mask_0, x = v_83_cast_fp16)[name = tensor("op_18047_cast_fp16")]; + tensor var_18051_begin_0 = const()[name = tensor("op_18051_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_18051_end_0 = const()[name = tensor("op_18051_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_18051_end_mask_0 = const()[name = tensor("op_18051_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18051_cast_fp16 = slice_by_index(begin = var_18051_begin_0, end = var_18051_end_0, end_mask = var_18051_end_mask_0, x = v_83_cast_fp16)[name = tensor("op_18051_cast_fp16")]; + tensor var_18055_begin_0 = const()[name = tensor("op_18055_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_18055_end_0 = const()[name = tensor("op_18055_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_18055_end_mask_0 = const()[name = tensor("op_18055_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18055_cast_fp16 = slice_by_index(begin = var_18055_begin_0, end = var_18055_end_0, end_mask = var_18055_end_mask_0, x = v_83_cast_fp16)[name = tensor("op_18055_cast_fp16")]; + tensor var_18059_begin_0 = const()[name = tensor("op_18059_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_18059_end_0 = const()[name = tensor("op_18059_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_18059_end_mask_0 = const()[name = tensor("op_18059_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18059_cast_fp16 = slice_by_index(begin = var_18059_begin_0, end = var_18059_end_0, end_mask = var_18059_end_mask_0, x = v_83_cast_fp16)[name = tensor("op_18059_cast_fp16")]; + tensor var_18063_begin_0 = const()[name = tensor("op_18063_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_18063_end_0 = const()[name = tensor("op_18063_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_18063_end_mask_0 = const()[name = tensor("op_18063_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18063_cast_fp16 = slice_by_index(begin = var_18063_begin_0, end = var_18063_end_0, end_mask = var_18063_end_mask_0, x = v_83_cast_fp16)[name = tensor("op_18063_cast_fp16")]; + tensor var_18067_begin_0 = const()[name = tensor("op_18067_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_18067_end_0 = const()[name = tensor("op_18067_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_18067_end_mask_0 = const()[name = tensor("op_18067_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18067_cast_fp16 = slice_by_index(begin = var_18067_begin_0, end = var_18067_end_0, end_mask = var_18067_end_mask_0, x = v_83_cast_fp16)[name = tensor("op_18067_cast_fp16")]; + tensor var_18071_begin_0 = const()[name = tensor("op_18071_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_18071_end_0 = const()[name = tensor("op_18071_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_18071_end_mask_0 = const()[name = tensor("op_18071_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18071_cast_fp16 = slice_by_index(begin = var_18071_begin_0, end = var_18071_end_0, end_mask = var_18071_end_mask_0, x = v_83_cast_fp16)[name = tensor("op_18071_cast_fp16")]; + tensor var_18075_begin_0 = const()[name = tensor("op_18075_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_18075_end_0 = const()[name = tensor("op_18075_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_18075_end_mask_0 = const()[name = tensor("op_18075_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18075_cast_fp16 = slice_by_index(begin = var_18075_begin_0, end = var_18075_end_0, end_mask = var_18075_end_mask_0, x = v_83_cast_fp16)[name = tensor("op_18075_cast_fp16")]; + tensor var_18079_begin_0 = const()[name = tensor("op_18079_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_18079_end_0 = const()[name = tensor("op_18079_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_18079_end_mask_0 = const()[name = tensor("op_18079_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18079_cast_fp16 = slice_by_index(begin = var_18079_begin_0, end = var_18079_end_0, end_mask = var_18079_end_mask_0, x = v_83_cast_fp16)[name = tensor("op_18079_cast_fp16")]; + tensor var_18083_begin_0 = const()[name = tensor("op_18083_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_18083_end_0 = const()[name = tensor("op_18083_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_18083_end_mask_0 = const()[name = tensor("op_18083_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18083_cast_fp16 = slice_by_index(begin = var_18083_begin_0, end = var_18083_end_0, end_mask = var_18083_end_mask_0, x = v_83_cast_fp16)[name = tensor("op_18083_cast_fp16")]; + tensor var_18087_begin_0 = const()[name = tensor("op_18087_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_18087_end_0 = const()[name = tensor("op_18087_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_18087_end_mask_0 = const()[name = tensor("op_18087_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18087_cast_fp16 = slice_by_index(begin = var_18087_begin_0, end = var_18087_end_0, end_mask = var_18087_end_mask_0, x = v_83_cast_fp16)[name = tensor("op_18087_cast_fp16")]; + tensor var_18091_begin_0 = const()[name = tensor("op_18091_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_18091_end_0 = const()[name = tensor("op_18091_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_18091_end_mask_0 = const()[name = tensor("op_18091_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18091_cast_fp16 = slice_by_index(begin = var_18091_begin_0, end = var_18091_end_0, end_mask = var_18091_end_mask_0, x = v_83_cast_fp16)[name = tensor("op_18091_cast_fp16")]; + tensor var_18095_begin_0 = const()[name = tensor("op_18095_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_18095_end_0 = const()[name = tensor("op_18095_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_18095_end_mask_0 = const()[name = tensor("op_18095_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18095_cast_fp16 = slice_by_index(begin = var_18095_begin_0, end = var_18095_end_0, end_mask = var_18095_end_mask_0, x = v_83_cast_fp16)[name = tensor("op_18095_cast_fp16")]; + tensor var_18099_begin_0 = const()[name = tensor("op_18099_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_18099_end_0 = const()[name = tensor("op_18099_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_18099_end_mask_0 = const()[name = tensor("op_18099_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18099_cast_fp16 = slice_by_index(begin = var_18099_begin_0, end = var_18099_end_0, end_mask = var_18099_end_mask_0, x = v_83_cast_fp16)[name = tensor("op_18099_cast_fp16")]; + tensor var_18103_begin_0 = const()[name = tensor("op_18103_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_18103_end_0 = const()[name = tensor("op_18103_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_18103_end_mask_0 = const()[name = tensor("op_18103_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18103_cast_fp16 = slice_by_index(begin = var_18103_begin_0, end = var_18103_end_0, end_mask = var_18103_end_mask_0, x = v_83_cast_fp16)[name = tensor("op_18103_cast_fp16")]; + tensor var_18107_equation_0 = const()[name = tensor("op_18107_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18107_cast_fp16 = einsum(equation = var_18107_equation_0, values = (var_17949_cast_fp16, var_17866_cast_fp16))[name = tensor("op_18107_cast_fp16")]; + tensor var_18108_to_fp16 = const()[name = tensor("op_18108_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1481_cast_fp16 = mul(x = var_18107_cast_fp16, y = var_18108_to_fp16)[name = tensor("aw_1481_cast_fp16")]; + tensor var_18111_equation_0 = const()[name = tensor("op_18111_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18111_cast_fp16 = einsum(equation = var_18111_equation_0, values = (var_17953_cast_fp16, var_17870_cast_fp16))[name = tensor("op_18111_cast_fp16")]; + tensor var_18112_to_fp16 = const()[name = tensor("op_18112_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1483_cast_fp16 = mul(x = var_18111_cast_fp16, y = var_18112_to_fp16)[name = tensor("aw_1483_cast_fp16")]; + tensor var_18115_equation_0 = const()[name = tensor("op_18115_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18115_cast_fp16 = einsum(equation = var_18115_equation_0, values = (var_17957_cast_fp16, var_17874_cast_fp16))[name = tensor("op_18115_cast_fp16")]; + tensor var_18116_to_fp16 = const()[name = tensor("op_18116_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1485_cast_fp16 = mul(x = var_18115_cast_fp16, y = var_18116_to_fp16)[name = tensor("aw_1485_cast_fp16")]; + tensor var_18119_equation_0 = const()[name = tensor("op_18119_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18119_cast_fp16 = einsum(equation = var_18119_equation_0, values = (var_17961_cast_fp16, var_17878_cast_fp16))[name = tensor("op_18119_cast_fp16")]; + tensor var_18120_to_fp16 = const()[name = tensor("op_18120_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1487_cast_fp16 = mul(x = var_18119_cast_fp16, y = var_18120_to_fp16)[name = tensor("aw_1487_cast_fp16")]; + tensor var_18123_equation_0 = const()[name = tensor("op_18123_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18123_cast_fp16 = einsum(equation = var_18123_equation_0, values = (var_17965_cast_fp16, var_17882_cast_fp16))[name = tensor("op_18123_cast_fp16")]; + tensor var_18124_to_fp16 = const()[name = tensor("op_18124_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1489_cast_fp16 = mul(x = var_18123_cast_fp16, y = var_18124_to_fp16)[name = tensor("aw_1489_cast_fp16")]; + tensor var_18127_equation_0 = const()[name = tensor("op_18127_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18127_cast_fp16 = einsum(equation = var_18127_equation_0, values = (var_17969_cast_fp16, var_17886_cast_fp16))[name = tensor("op_18127_cast_fp16")]; + tensor var_18128_to_fp16 = const()[name = tensor("op_18128_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1491_cast_fp16 = mul(x = var_18127_cast_fp16, y = var_18128_to_fp16)[name = tensor("aw_1491_cast_fp16")]; + tensor var_18131_equation_0 = const()[name = tensor("op_18131_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18131_cast_fp16 = einsum(equation = var_18131_equation_0, values = (var_17973_cast_fp16, var_17890_cast_fp16))[name = tensor("op_18131_cast_fp16")]; + tensor var_18132_to_fp16 = const()[name = tensor("op_18132_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1493_cast_fp16 = mul(x = var_18131_cast_fp16, y = var_18132_to_fp16)[name = tensor("aw_1493_cast_fp16")]; + tensor var_18135_equation_0 = const()[name = tensor("op_18135_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18135_cast_fp16 = einsum(equation = var_18135_equation_0, values = (var_17977_cast_fp16, var_17894_cast_fp16))[name = tensor("op_18135_cast_fp16")]; + tensor var_18136_to_fp16 = const()[name = tensor("op_18136_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1495_cast_fp16 = mul(x = var_18135_cast_fp16, y = var_18136_to_fp16)[name = tensor("aw_1495_cast_fp16")]; + tensor var_18139_equation_0 = const()[name = tensor("op_18139_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18139_cast_fp16 = einsum(equation = var_18139_equation_0, values = (var_17981_cast_fp16, var_17898_cast_fp16))[name = tensor("op_18139_cast_fp16")]; + tensor var_18140_to_fp16 = const()[name = tensor("op_18140_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1497_cast_fp16 = mul(x = var_18139_cast_fp16, y = var_18140_to_fp16)[name = tensor("aw_1497_cast_fp16")]; + tensor var_18143_equation_0 = const()[name = tensor("op_18143_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18143_cast_fp16 = einsum(equation = var_18143_equation_0, values = (var_17985_cast_fp16, var_17902_cast_fp16))[name = tensor("op_18143_cast_fp16")]; + tensor var_18144_to_fp16 = const()[name = tensor("op_18144_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1499_cast_fp16 = mul(x = var_18143_cast_fp16, y = var_18144_to_fp16)[name = tensor("aw_1499_cast_fp16")]; + tensor var_18147_equation_0 = const()[name = tensor("op_18147_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18147_cast_fp16 = einsum(equation = var_18147_equation_0, values = (var_17989_cast_fp16, var_17906_cast_fp16))[name = tensor("op_18147_cast_fp16")]; + tensor var_18148_to_fp16 = const()[name = tensor("op_18148_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1501_cast_fp16 = mul(x = var_18147_cast_fp16, y = var_18148_to_fp16)[name = tensor("aw_1501_cast_fp16")]; + tensor var_18151_equation_0 = const()[name = tensor("op_18151_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18151_cast_fp16 = einsum(equation = var_18151_equation_0, values = (var_17993_cast_fp16, var_17910_cast_fp16))[name = tensor("op_18151_cast_fp16")]; + tensor var_18152_to_fp16 = const()[name = tensor("op_18152_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1503_cast_fp16 = mul(x = var_18151_cast_fp16, y = var_18152_to_fp16)[name = tensor("aw_1503_cast_fp16")]; + tensor var_18155_equation_0 = const()[name = tensor("op_18155_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18155_cast_fp16 = einsum(equation = var_18155_equation_0, values = (var_17997_cast_fp16, var_17914_cast_fp16))[name = tensor("op_18155_cast_fp16")]; + tensor var_18156_to_fp16 = const()[name = tensor("op_18156_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1505_cast_fp16 = mul(x = var_18155_cast_fp16, y = var_18156_to_fp16)[name = tensor("aw_1505_cast_fp16")]; + tensor var_18159_equation_0 = const()[name = tensor("op_18159_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18159_cast_fp16 = einsum(equation = var_18159_equation_0, values = (var_18001_cast_fp16, var_17918_cast_fp16))[name = tensor("op_18159_cast_fp16")]; + tensor var_18160_to_fp16 = const()[name = tensor("op_18160_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1507_cast_fp16 = mul(x = var_18159_cast_fp16, y = var_18160_to_fp16)[name = tensor("aw_1507_cast_fp16")]; + tensor var_18163_equation_0 = const()[name = tensor("op_18163_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18163_cast_fp16 = einsum(equation = var_18163_equation_0, values = (var_18005_cast_fp16, var_17922_cast_fp16))[name = tensor("op_18163_cast_fp16")]; + tensor var_18164_to_fp16 = const()[name = tensor("op_18164_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1509_cast_fp16 = mul(x = var_18163_cast_fp16, y = var_18164_to_fp16)[name = tensor("aw_1509_cast_fp16")]; + tensor var_18167_equation_0 = const()[name = tensor("op_18167_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18167_cast_fp16 = einsum(equation = var_18167_equation_0, values = (var_18009_cast_fp16, var_17926_cast_fp16))[name = tensor("op_18167_cast_fp16")]; + tensor var_18168_to_fp16 = const()[name = tensor("op_18168_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1511_cast_fp16 = mul(x = var_18167_cast_fp16, y = var_18168_to_fp16)[name = tensor("aw_1511_cast_fp16")]; + tensor var_18171_equation_0 = const()[name = tensor("op_18171_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18171_cast_fp16 = einsum(equation = var_18171_equation_0, values = (var_18013_cast_fp16, var_17930_cast_fp16))[name = tensor("op_18171_cast_fp16")]; + tensor var_18172_to_fp16 = const()[name = tensor("op_18172_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1513_cast_fp16 = mul(x = var_18171_cast_fp16, y = var_18172_to_fp16)[name = tensor("aw_1513_cast_fp16")]; + tensor var_18175_equation_0 = const()[name = tensor("op_18175_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18175_cast_fp16 = einsum(equation = var_18175_equation_0, values = (var_18017_cast_fp16, var_17934_cast_fp16))[name = tensor("op_18175_cast_fp16")]; + tensor var_18176_to_fp16 = const()[name = tensor("op_18176_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1515_cast_fp16 = mul(x = var_18175_cast_fp16, y = var_18176_to_fp16)[name = tensor("aw_1515_cast_fp16")]; + tensor var_18179_equation_0 = const()[name = tensor("op_18179_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18179_cast_fp16 = einsum(equation = var_18179_equation_0, values = (var_18021_cast_fp16, var_17938_cast_fp16))[name = tensor("op_18179_cast_fp16")]; + tensor var_18180_to_fp16 = const()[name = tensor("op_18180_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1517_cast_fp16 = mul(x = var_18179_cast_fp16, y = var_18180_to_fp16)[name = tensor("aw_1517_cast_fp16")]; + tensor var_18183_equation_0 = const()[name = tensor("op_18183_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18183_cast_fp16 = einsum(equation = var_18183_equation_0, values = (var_18025_cast_fp16, var_17942_cast_fp16))[name = tensor("op_18183_cast_fp16")]; + tensor var_18184_to_fp16 = const()[name = tensor("op_18184_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1519_cast_fp16 = mul(x = var_18183_cast_fp16, y = var_18184_to_fp16)[name = tensor("aw_1519_cast_fp16")]; + tensor var_18186_cast_fp16 = softmax(axis = var_2624, x = aw_1481_cast_fp16)[name = tensor("op_18186_cast_fp16")]; + tensor var_18187_cast_fp16 = softmax(axis = var_2624, x = aw_1483_cast_fp16)[name = tensor("op_18187_cast_fp16")]; + tensor var_18188_cast_fp16 = softmax(axis = var_2624, x = aw_1485_cast_fp16)[name = tensor("op_18188_cast_fp16")]; + tensor var_18189_cast_fp16 = softmax(axis = var_2624, x = aw_1487_cast_fp16)[name = tensor("op_18189_cast_fp16")]; + tensor var_18190_cast_fp16 = softmax(axis = var_2624, x = aw_1489_cast_fp16)[name = tensor("op_18190_cast_fp16")]; + tensor var_18191_cast_fp16 = softmax(axis = var_2624, x = aw_1491_cast_fp16)[name = tensor("op_18191_cast_fp16")]; + tensor var_18192_cast_fp16 = softmax(axis = var_2624, x = aw_1493_cast_fp16)[name = tensor("op_18192_cast_fp16")]; + tensor var_18193_cast_fp16 = softmax(axis = var_2624, x = aw_1495_cast_fp16)[name = tensor("op_18193_cast_fp16")]; + tensor var_18194_cast_fp16 = softmax(axis = var_2624, x = aw_1497_cast_fp16)[name = tensor("op_18194_cast_fp16")]; + tensor var_18195_cast_fp16 = softmax(axis = var_2624, x = aw_1499_cast_fp16)[name = tensor("op_18195_cast_fp16")]; + tensor var_18196_cast_fp16 = softmax(axis = var_2624, x = aw_1501_cast_fp16)[name = tensor("op_18196_cast_fp16")]; + tensor var_18197_cast_fp16 = softmax(axis = var_2624, x = aw_1503_cast_fp16)[name = tensor("op_18197_cast_fp16")]; + tensor var_18198_cast_fp16 = softmax(axis = var_2624, x = aw_1505_cast_fp16)[name = tensor("op_18198_cast_fp16")]; + tensor var_18199_cast_fp16 = softmax(axis = var_2624, x = aw_1507_cast_fp16)[name = tensor("op_18199_cast_fp16")]; + tensor var_18200_cast_fp16 = softmax(axis = var_2624, x = aw_1509_cast_fp16)[name = tensor("op_18200_cast_fp16")]; + tensor var_18201_cast_fp16 = softmax(axis = var_2624, x = aw_1511_cast_fp16)[name = tensor("op_18201_cast_fp16")]; + tensor var_18202_cast_fp16 = softmax(axis = var_2624, x = aw_1513_cast_fp16)[name = tensor("op_18202_cast_fp16")]; + tensor var_18203_cast_fp16 = softmax(axis = var_2624, x = aw_1515_cast_fp16)[name = tensor("op_18203_cast_fp16")]; + tensor var_18204_cast_fp16 = softmax(axis = var_2624, x = aw_1517_cast_fp16)[name = tensor("op_18204_cast_fp16")]; + tensor var_18205_cast_fp16 = softmax(axis = var_2624, x = aw_1519_cast_fp16)[name = tensor("op_18205_cast_fp16")]; + tensor var_18207_equation_0 = const()[name = tensor("op_18207_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18207_cast_fp16 = einsum(equation = var_18207_equation_0, values = (var_18027_cast_fp16, var_18186_cast_fp16))[name = tensor("op_18207_cast_fp16")]; + tensor var_18209_equation_0 = const()[name = tensor("op_18209_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18209_cast_fp16 = einsum(equation = var_18209_equation_0, values = (var_18031_cast_fp16, var_18187_cast_fp16))[name = tensor("op_18209_cast_fp16")]; + tensor var_18211_equation_0 = const()[name = tensor("op_18211_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18211_cast_fp16 = einsum(equation = var_18211_equation_0, values = (var_18035_cast_fp16, var_18188_cast_fp16))[name = tensor("op_18211_cast_fp16")]; + tensor var_18213_equation_0 = const()[name = tensor("op_18213_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18213_cast_fp16 = einsum(equation = var_18213_equation_0, values = (var_18039_cast_fp16, var_18189_cast_fp16))[name = tensor("op_18213_cast_fp16")]; + tensor var_18215_equation_0 = const()[name = tensor("op_18215_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18215_cast_fp16 = einsum(equation = var_18215_equation_0, values = (var_18043_cast_fp16, var_18190_cast_fp16))[name = tensor("op_18215_cast_fp16")]; + tensor var_18217_equation_0 = const()[name = tensor("op_18217_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18217_cast_fp16 = einsum(equation = var_18217_equation_0, values = (var_18047_cast_fp16, var_18191_cast_fp16))[name = tensor("op_18217_cast_fp16")]; + tensor var_18219_equation_0 = const()[name = tensor("op_18219_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18219_cast_fp16 = einsum(equation = var_18219_equation_0, values = (var_18051_cast_fp16, var_18192_cast_fp16))[name = tensor("op_18219_cast_fp16")]; + tensor var_18221_equation_0 = const()[name = tensor("op_18221_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18221_cast_fp16 = einsum(equation = var_18221_equation_0, values = (var_18055_cast_fp16, var_18193_cast_fp16))[name = tensor("op_18221_cast_fp16")]; + tensor var_18223_equation_0 = const()[name = tensor("op_18223_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18223_cast_fp16 = einsum(equation = var_18223_equation_0, values = (var_18059_cast_fp16, var_18194_cast_fp16))[name = tensor("op_18223_cast_fp16")]; + tensor var_18225_equation_0 = const()[name = tensor("op_18225_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18225_cast_fp16 = einsum(equation = var_18225_equation_0, values = (var_18063_cast_fp16, var_18195_cast_fp16))[name = tensor("op_18225_cast_fp16")]; + tensor var_18227_equation_0 = const()[name = tensor("op_18227_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18227_cast_fp16 = einsum(equation = var_18227_equation_0, values = (var_18067_cast_fp16, var_18196_cast_fp16))[name = tensor("op_18227_cast_fp16")]; + tensor var_18229_equation_0 = const()[name = tensor("op_18229_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18229_cast_fp16 = einsum(equation = var_18229_equation_0, values = (var_18071_cast_fp16, var_18197_cast_fp16))[name = tensor("op_18229_cast_fp16")]; + tensor var_18231_equation_0 = const()[name = tensor("op_18231_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18231_cast_fp16 = einsum(equation = var_18231_equation_0, values = (var_18075_cast_fp16, var_18198_cast_fp16))[name = tensor("op_18231_cast_fp16")]; + tensor var_18233_equation_0 = const()[name = tensor("op_18233_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18233_cast_fp16 = einsum(equation = var_18233_equation_0, values = (var_18079_cast_fp16, var_18199_cast_fp16))[name = tensor("op_18233_cast_fp16")]; + tensor var_18235_equation_0 = const()[name = tensor("op_18235_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18235_cast_fp16 = einsum(equation = var_18235_equation_0, values = (var_18083_cast_fp16, var_18200_cast_fp16))[name = tensor("op_18235_cast_fp16")]; + tensor var_18237_equation_0 = const()[name = tensor("op_18237_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18237_cast_fp16 = einsum(equation = var_18237_equation_0, values = (var_18087_cast_fp16, var_18201_cast_fp16))[name = tensor("op_18237_cast_fp16")]; + tensor var_18239_equation_0 = const()[name = tensor("op_18239_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18239_cast_fp16 = einsum(equation = var_18239_equation_0, values = (var_18091_cast_fp16, var_18202_cast_fp16))[name = tensor("op_18239_cast_fp16")]; + tensor var_18241_equation_0 = const()[name = tensor("op_18241_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18241_cast_fp16 = einsum(equation = var_18241_equation_0, values = (var_18095_cast_fp16, var_18203_cast_fp16))[name = tensor("op_18241_cast_fp16")]; + tensor var_18243_equation_0 = const()[name = tensor("op_18243_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18243_cast_fp16 = einsum(equation = var_18243_equation_0, values = (var_18099_cast_fp16, var_18204_cast_fp16))[name = tensor("op_18243_cast_fp16")]; + tensor var_18245_equation_0 = const()[name = tensor("op_18245_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18245_cast_fp16 = einsum(equation = var_18245_equation_0, values = (var_18103_cast_fp16, var_18205_cast_fp16))[name = tensor("op_18245_cast_fp16")]; + tensor input_279_interleave_0 = const()[name = tensor("input_279_interleave_0"), val = tensor(false)]; + tensor input_279_cast_fp16 = concat(axis = var_2624, interleave = input_279_interleave_0, values = (var_18207_cast_fp16, var_18209_cast_fp16, var_18211_cast_fp16, var_18213_cast_fp16, var_18215_cast_fp16, var_18217_cast_fp16, var_18219_cast_fp16, var_18221_cast_fp16, var_18223_cast_fp16, var_18225_cast_fp16, var_18227_cast_fp16, var_18229_cast_fp16, var_18231_cast_fp16, var_18233_cast_fp16, var_18235_cast_fp16, var_18237_cast_fp16, var_18239_cast_fp16, var_18241_cast_fp16, var_18243_cast_fp16, var_18245_cast_fp16))[name = tensor("input_279_cast_fp16")]; + tensor var_18251 = const()[name = tensor("op_18251"), val = tensor([1, 1])]; + tensor var_18253 = const()[name = tensor("op_18253"), val = tensor([1, 1])]; + tensor var_18255_pad_type_0 = const()[name = tensor("op_18255_pad_type_0"), val = tensor("custom")]; + tensor var_18255_pad_0 = const()[name = tensor("op_18255_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_6_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_6_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1406098560)))]; + tensor down_blocks_2_attentions_1_transformer_blocks_6_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_6_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1409375424)))]; + tensor var_18255_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_6_attn2_to_out_0_bias_to_fp16, dilations = var_18253, groups = var_2624, pad = var_18255_pad_0, pad_type = var_18255_pad_type_0, strides = var_18251, weight = down_blocks_2_attentions_1_transformer_blocks_6_attn2_to_out_0_weight_to_fp16, x = input_279_cast_fp16)[name = tensor("op_18255_cast_fp16")]; + tensor inputs_125_cast_fp16 = add(x = var_18255_cast_fp16, y = inputs_123_cast_fp16)[name = tensor("inputs_125_cast_fp16")]; + tensor input_281_axes_0 = const()[name = tensor("input_281_axes_0"), val = tensor([1])]; + tensor input_281_gamma_0_to_fp16 = const()[name = tensor("input_281_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1409378048)))]; + tensor input_281_beta_0_to_fp16 = const()[name = tensor("input_281_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1409380672)))]; + tensor var_18265_to_fp16 = const()[name = tensor("op_18265_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_281_cast_fp16 = layer_norm(axes = input_281_axes_0, beta = input_281_beta_0_to_fp16, epsilon = var_18265_to_fp16, gamma = input_281_gamma_0_to_fp16, x = inputs_125_cast_fp16)[name = tensor("input_281_cast_fp16")]; + tensor var_18281 = const()[name = tensor("op_18281"), val = tensor([1, 1])]; + tensor var_18283 = const()[name = tensor("op_18283"), val = tensor([1, 1])]; + tensor var_18285_pad_type_0 = const()[name = tensor("op_18285_pad_type_0"), val = tensor("custom")]; + tensor var_18285_pad_0 = const()[name = tensor("op_18285_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_6_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_6_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1409383296)))]; + tensor down_blocks_2_attentions_1_transformer_blocks_6_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_6_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1435597760)))]; + tensor var_18285_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_6_ff_net_0_proj_bias_to_fp16, dilations = var_18283, groups = var_2624, pad = var_18285_pad_0, pad_type = var_18285_pad_type_0, strides = var_18281, weight = down_blocks_2_attentions_1_transformer_blocks_6_ff_net_0_proj_weight_to_fp16, x = input_281_cast_fp16)[name = tensor("op_18285_cast_fp16")]; + tensor var_18286_split_sizes_0 = const()[name = tensor("op_18286_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_18286_axis_0 = const()[name = tensor("op_18286_axis_0"), val = tensor(1)]; + tensor var_18286_cast_fp16_0, tensor var_18286_cast_fp16_1 = split(axis = var_18286_axis_0, split_sizes = var_18286_split_sizes_0, x = var_18285_cast_fp16)[name = tensor("op_18286_cast_fp16")]; + tensor var_18288_mode_0 = const()[name = tensor("op_18288_mode_0"), val = tensor("EXACT")]; + tensor var_18288_cast_fp16 = gelu(mode = var_18288_mode_0, x = var_18286_cast_fp16_1)[name = tensor("op_18288_cast_fp16")]; + tensor input_283_cast_fp16 = mul(x = var_18286_cast_fp16_0, y = var_18288_cast_fp16)[name = tensor("input_283_cast_fp16")]; + tensor var_18292 = const()[name = tensor("op_18292"), val = tensor([1, 1])]; + tensor var_18294 = const()[name = tensor("op_18294"), val = tensor([1, 1])]; + tensor var_18296_pad_type_0 = const()[name = tensor("op_18296_pad_type_0"), val = tensor("custom")]; + tensor var_18296_pad_0 = const()[name = tensor("op_18296_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_6_ff_net_2_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_6_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1435618304)))]; + tensor down_blocks_2_attentions_1_transformer_blocks_6_ff_net_2_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_6_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1448725568)))]; + tensor var_18296_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_6_ff_net_2_bias_to_fp16, dilations = var_18294, groups = var_2624, pad = var_18296_pad_0, pad_type = var_18296_pad_type_0, strides = var_18292, weight = down_blocks_2_attentions_1_transformer_blocks_6_ff_net_2_weight_to_fp16, x = input_283_cast_fp16)[name = tensor("op_18296_cast_fp16")]; + tensor inputs_127_cast_fp16 = add(x = var_18296_cast_fp16, y = inputs_125_cast_fp16)[name = tensor("inputs_127_cast_fp16")]; + tensor hidden_states_179_axes_0 = const()[name = tensor("hidden_states_179_axes_0"), val = tensor([1])]; + tensor hidden_states_179_gamma_0_to_fp16 = const()[name = tensor("hidden_states_179_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1448728192)))]; + tensor hidden_states_179_beta_0_to_fp16 = const()[name = tensor("hidden_states_179_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1448730816)))]; + tensor var_18312_to_fp16 = const()[name = tensor("op_18312_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_179_cast_fp16 = layer_norm(axes = hidden_states_179_axes_0, beta = hidden_states_179_beta_0_to_fp16, epsilon = var_18312_to_fp16, gamma = hidden_states_179_gamma_0_to_fp16, x = inputs_127_cast_fp16)[name = tensor("hidden_states_179_cast_fp16")]; + tensor var_18327 = const()[name = tensor("op_18327"), val = tensor([1, 1])]; + tensor var_18329 = const()[name = tensor("op_18329"), val = tensor([1, 1])]; + tensor q_85_pad_type_0 = const()[name = tensor("q_85_pad_type_0"), val = tensor("custom")]; + tensor q_85_pad_0 = const()[name = tensor("q_85_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_7_attn1_to_q_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_7_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1448733440)))]; + tensor q_85_cast_fp16 = conv(dilations = var_18329, groups = var_2624, pad = q_85_pad_0, pad_type = q_85_pad_type_0, strides = var_18327, weight = down_blocks_2_attentions_1_transformer_blocks_7_attn1_to_q_weight_to_fp16, x = hidden_states_179_cast_fp16)[name = tensor("q_85_cast_fp16")]; + tensor var_18333 = const()[name = tensor("op_18333"), val = tensor([1, 1])]; + tensor var_18335 = const()[name = tensor("op_18335"), val = tensor([1, 1])]; + tensor k_169_pad_type_0 = const()[name = tensor("k_169_pad_type_0"), val = tensor("custom")]; + tensor k_169_pad_0 = const()[name = tensor("k_169_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_7_attn1_to_k_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_7_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1452010304)))]; + tensor k_169_cast_fp16 = conv(dilations = var_18335, groups = var_2624, pad = k_169_pad_0, pad_type = k_169_pad_type_0, strides = var_18333, weight = down_blocks_2_attentions_1_transformer_blocks_7_attn1_to_k_weight_to_fp16, x = hidden_states_179_cast_fp16)[name = tensor("k_169_cast_fp16")]; + tensor var_18339 = const()[name = tensor("op_18339"), val = tensor([1, 1])]; + tensor var_18341 = const()[name = tensor("op_18341"), val = tensor([1, 1])]; + tensor v_85_pad_type_0 = const()[name = tensor("v_85_pad_type_0"), val = tensor("custom")]; + tensor v_85_pad_0 = const()[name = tensor("v_85_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_7_attn1_to_v_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_7_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1455287168)))]; + tensor v_85_cast_fp16 = conv(dilations = var_18341, groups = var_2624, pad = v_85_pad_0, pad_type = v_85_pad_type_0, strides = var_18339, weight = down_blocks_2_attentions_1_transformer_blocks_7_attn1_to_v_weight_to_fp16, x = hidden_states_179_cast_fp16)[name = tensor("v_85_cast_fp16")]; + tensor var_18345_begin_0 = const()[name = tensor("op_18345_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_18345_end_0 = const()[name = tensor("op_18345_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_18345_end_mask_0 = const()[name = tensor("op_18345_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18345_cast_fp16 = slice_by_index(begin = var_18345_begin_0, end = var_18345_end_0, end_mask = var_18345_end_mask_0, x = q_85_cast_fp16)[name = tensor("op_18345_cast_fp16")]; + tensor var_18349_begin_0 = const()[name = tensor("op_18349_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_18349_end_0 = const()[name = tensor("op_18349_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_18349_end_mask_0 = const()[name = tensor("op_18349_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18349_cast_fp16 = slice_by_index(begin = var_18349_begin_0, end = var_18349_end_0, end_mask = var_18349_end_mask_0, x = q_85_cast_fp16)[name = tensor("op_18349_cast_fp16")]; + tensor var_18353_begin_0 = const()[name = tensor("op_18353_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_18353_end_0 = const()[name = tensor("op_18353_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_18353_end_mask_0 = const()[name = tensor("op_18353_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18353_cast_fp16 = slice_by_index(begin = var_18353_begin_0, end = var_18353_end_0, end_mask = var_18353_end_mask_0, x = q_85_cast_fp16)[name = tensor("op_18353_cast_fp16")]; + tensor var_18357_begin_0 = const()[name = tensor("op_18357_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_18357_end_0 = const()[name = tensor("op_18357_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_18357_end_mask_0 = const()[name = tensor("op_18357_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18357_cast_fp16 = slice_by_index(begin = var_18357_begin_0, end = var_18357_end_0, end_mask = var_18357_end_mask_0, x = q_85_cast_fp16)[name = tensor("op_18357_cast_fp16")]; + tensor var_18361_begin_0 = const()[name = tensor("op_18361_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_18361_end_0 = const()[name = tensor("op_18361_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_18361_end_mask_0 = const()[name = tensor("op_18361_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18361_cast_fp16 = slice_by_index(begin = var_18361_begin_0, end = var_18361_end_0, end_mask = var_18361_end_mask_0, x = q_85_cast_fp16)[name = tensor("op_18361_cast_fp16")]; + tensor var_18365_begin_0 = const()[name = tensor("op_18365_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_18365_end_0 = const()[name = tensor("op_18365_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_18365_end_mask_0 = const()[name = tensor("op_18365_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18365_cast_fp16 = slice_by_index(begin = var_18365_begin_0, end = var_18365_end_0, end_mask = var_18365_end_mask_0, x = q_85_cast_fp16)[name = tensor("op_18365_cast_fp16")]; + tensor var_18369_begin_0 = const()[name = tensor("op_18369_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_18369_end_0 = const()[name = tensor("op_18369_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_18369_end_mask_0 = const()[name = tensor("op_18369_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18369_cast_fp16 = slice_by_index(begin = var_18369_begin_0, end = var_18369_end_0, end_mask = var_18369_end_mask_0, x = q_85_cast_fp16)[name = tensor("op_18369_cast_fp16")]; + tensor var_18373_begin_0 = const()[name = tensor("op_18373_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_18373_end_0 = const()[name = tensor("op_18373_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_18373_end_mask_0 = const()[name = tensor("op_18373_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18373_cast_fp16 = slice_by_index(begin = var_18373_begin_0, end = var_18373_end_0, end_mask = var_18373_end_mask_0, x = q_85_cast_fp16)[name = tensor("op_18373_cast_fp16")]; + tensor var_18377_begin_0 = const()[name = tensor("op_18377_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_18377_end_0 = const()[name = tensor("op_18377_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_18377_end_mask_0 = const()[name = tensor("op_18377_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18377_cast_fp16 = slice_by_index(begin = var_18377_begin_0, end = var_18377_end_0, end_mask = var_18377_end_mask_0, x = q_85_cast_fp16)[name = tensor("op_18377_cast_fp16")]; + tensor var_18381_begin_0 = const()[name = tensor("op_18381_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_18381_end_0 = const()[name = tensor("op_18381_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_18381_end_mask_0 = const()[name = tensor("op_18381_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18381_cast_fp16 = slice_by_index(begin = var_18381_begin_0, end = var_18381_end_0, end_mask = var_18381_end_mask_0, x = q_85_cast_fp16)[name = tensor("op_18381_cast_fp16")]; + tensor var_18385_begin_0 = const()[name = tensor("op_18385_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_18385_end_0 = const()[name = tensor("op_18385_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_18385_end_mask_0 = const()[name = tensor("op_18385_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18385_cast_fp16 = slice_by_index(begin = var_18385_begin_0, end = var_18385_end_0, end_mask = var_18385_end_mask_0, x = q_85_cast_fp16)[name = tensor("op_18385_cast_fp16")]; + tensor var_18389_begin_0 = const()[name = tensor("op_18389_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_18389_end_0 = const()[name = tensor("op_18389_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_18389_end_mask_0 = const()[name = tensor("op_18389_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18389_cast_fp16 = slice_by_index(begin = var_18389_begin_0, end = var_18389_end_0, end_mask = var_18389_end_mask_0, x = q_85_cast_fp16)[name = tensor("op_18389_cast_fp16")]; + tensor var_18393_begin_0 = const()[name = tensor("op_18393_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_18393_end_0 = const()[name = tensor("op_18393_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_18393_end_mask_0 = const()[name = tensor("op_18393_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18393_cast_fp16 = slice_by_index(begin = var_18393_begin_0, end = var_18393_end_0, end_mask = var_18393_end_mask_0, x = q_85_cast_fp16)[name = tensor("op_18393_cast_fp16")]; + tensor var_18397_begin_0 = const()[name = tensor("op_18397_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_18397_end_0 = const()[name = tensor("op_18397_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_18397_end_mask_0 = const()[name = tensor("op_18397_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18397_cast_fp16 = slice_by_index(begin = var_18397_begin_0, end = var_18397_end_0, end_mask = var_18397_end_mask_0, x = q_85_cast_fp16)[name = tensor("op_18397_cast_fp16")]; + tensor var_18401_begin_0 = const()[name = tensor("op_18401_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_18401_end_0 = const()[name = tensor("op_18401_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_18401_end_mask_0 = const()[name = tensor("op_18401_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18401_cast_fp16 = slice_by_index(begin = var_18401_begin_0, end = var_18401_end_0, end_mask = var_18401_end_mask_0, x = q_85_cast_fp16)[name = tensor("op_18401_cast_fp16")]; + tensor var_18405_begin_0 = const()[name = tensor("op_18405_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_18405_end_0 = const()[name = tensor("op_18405_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_18405_end_mask_0 = const()[name = tensor("op_18405_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18405_cast_fp16 = slice_by_index(begin = var_18405_begin_0, end = var_18405_end_0, end_mask = var_18405_end_mask_0, x = q_85_cast_fp16)[name = tensor("op_18405_cast_fp16")]; + tensor var_18409_begin_0 = const()[name = tensor("op_18409_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_18409_end_0 = const()[name = tensor("op_18409_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_18409_end_mask_0 = const()[name = tensor("op_18409_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18409_cast_fp16 = slice_by_index(begin = var_18409_begin_0, end = var_18409_end_0, end_mask = var_18409_end_mask_0, x = q_85_cast_fp16)[name = tensor("op_18409_cast_fp16")]; + tensor var_18413_begin_0 = const()[name = tensor("op_18413_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_18413_end_0 = const()[name = tensor("op_18413_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_18413_end_mask_0 = const()[name = tensor("op_18413_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18413_cast_fp16 = slice_by_index(begin = var_18413_begin_0, end = var_18413_end_0, end_mask = var_18413_end_mask_0, x = q_85_cast_fp16)[name = tensor("op_18413_cast_fp16")]; + tensor var_18417_begin_0 = const()[name = tensor("op_18417_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_18417_end_0 = const()[name = tensor("op_18417_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_18417_end_mask_0 = const()[name = tensor("op_18417_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18417_cast_fp16 = slice_by_index(begin = var_18417_begin_0, end = var_18417_end_0, end_mask = var_18417_end_mask_0, x = q_85_cast_fp16)[name = tensor("op_18417_cast_fp16")]; + tensor var_18421_begin_0 = const()[name = tensor("op_18421_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_18421_end_0 = const()[name = tensor("op_18421_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_18421_end_mask_0 = const()[name = tensor("op_18421_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18421_cast_fp16 = slice_by_index(begin = var_18421_begin_0, end = var_18421_end_0, end_mask = var_18421_end_mask_0, x = q_85_cast_fp16)[name = tensor("op_18421_cast_fp16")]; + tensor k_171_perm_0 = const()[name = tensor("k_171_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_18428_begin_0 = const()[name = tensor("op_18428_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_18428_end_0 = const()[name = tensor("op_18428_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_18428_end_mask_0 = const()[name = tensor("op_18428_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_171_cast_fp16 = transpose(perm = k_171_perm_0, x = k_169_cast_fp16)[name = tensor("transpose_97")]; + tensor var_18428_cast_fp16 = slice_by_index(begin = var_18428_begin_0, end = var_18428_end_0, end_mask = var_18428_end_mask_0, x = k_171_cast_fp16)[name = tensor("op_18428_cast_fp16")]; + tensor var_18432_begin_0 = const()[name = tensor("op_18432_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_18432_end_0 = const()[name = tensor("op_18432_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_18432_end_mask_0 = const()[name = tensor("op_18432_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18432_cast_fp16 = slice_by_index(begin = var_18432_begin_0, end = var_18432_end_0, end_mask = var_18432_end_mask_0, x = k_171_cast_fp16)[name = tensor("op_18432_cast_fp16")]; + tensor var_18436_begin_0 = const()[name = tensor("op_18436_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_18436_end_0 = const()[name = tensor("op_18436_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_18436_end_mask_0 = const()[name = tensor("op_18436_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18436_cast_fp16 = slice_by_index(begin = var_18436_begin_0, end = var_18436_end_0, end_mask = var_18436_end_mask_0, x = k_171_cast_fp16)[name = tensor("op_18436_cast_fp16")]; + tensor var_18440_begin_0 = const()[name = tensor("op_18440_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_18440_end_0 = const()[name = tensor("op_18440_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_18440_end_mask_0 = const()[name = tensor("op_18440_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18440_cast_fp16 = slice_by_index(begin = var_18440_begin_0, end = var_18440_end_0, end_mask = var_18440_end_mask_0, x = k_171_cast_fp16)[name = tensor("op_18440_cast_fp16")]; + tensor var_18444_begin_0 = const()[name = tensor("op_18444_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_18444_end_0 = const()[name = tensor("op_18444_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_18444_end_mask_0 = const()[name = tensor("op_18444_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18444_cast_fp16 = slice_by_index(begin = var_18444_begin_0, end = var_18444_end_0, end_mask = var_18444_end_mask_0, x = k_171_cast_fp16)[name = tensor("op_18444_cast_fp16")]; + tensor var_18448_begin_0 = const()[name = tensor("op_18448_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_18448_end_0 = const()[name = tensor("op_18448_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_18448_end_mask_0 = const()[name = tensor("op_18448_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18448_cast_fp16 = slice_by_index(begin = var_18448_begin_0, end = var_18448_end_0, end_mask = var_18448_end_mask_0, x = k_171_cast_fp16)[name = tensor("op_18448_cast_fp16")]; + tensor var_18452_begin_0 = const()[name = tensor("op_18452_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_18452_end_0 = const()[name = tensor("op_18452_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_18452_end_mask_0 = const()[name = tensor("op_18452_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18452_cast_fp16 = slice_by_index(begin = var_18452_begin_0, end = var_18452_end_0, end_mask = var_18452_end_mask_0, x = k_171_cast_fp16)[name = tensor("op_18452_cast_fp16")]; + tensor var_18456_begin_0 = const()[name = tensor("op_18456_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_18456_end_0 = const()[name = tensor("op_18456_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_18456_end_mask_0 = const()[name = tensor("op_18456_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18456_cast_fp16 = slice_by_index(begin = var_18456_begin_0, end = var_18456_end_0, end_mask = var_18456_end_mask_0, x = k_171_cast_fp16)[name = tensor("op_18456_cast_fp16")]; + tensor var_18460_begin_0 = const()[name = tensor("op_18460_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_18460_end_0 = const()[name = tensor("op_18460_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_18460_end_mask_0 = const()[name = tensor("op_18460_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18460_cast_fp16 = slice_by_index(begin = var_18460_begin_0, end = var_18460_end_0, end_mask = var_18460_end_mask_0, x = k_171_cast_fp16)[name = tensor("op_18460_cast_fp16")]; + tensor var_18464_begin_0 = const()[name = tensor("op_18464_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_18464_end_0 = const()[name = tensor("op_18464_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_18464_end_mask_0 = const()[name = tensor("op_18464_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18464_cast_fp16 = slice_by_index(begin = var_18464_begin_0, end = var_18464_end_0, end_mask = var_18464_end_mask_0, x = k_171_cast_fp16)[name = tensor("op_18464_cast_fp16")]; + tensor var_18468_begin_0 = const()[name = tensor("op_18468_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_18468_end_0 = const()[name = tensor("op_18468_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_18468_end_mask_0 = const()[name = tensor("op_18468_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18468_cast_fp16 = slice_by_index(begin = var_18468_begin_0, end = var_18468_end_0, end_mask = var_18468_end_mask_0, x = k_171_cast_fp16)[name = tensor("op_18468_cast_fp16")]; + tensor var_18472_begin_0 = const()[name = tensor("op_18472_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_18472_end_0 = const()[name = tensor("op_18472_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_18472_end_mask_0 = const()[name = tensor("op_18472_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18472_cast_fp16 = slice_by_index(begin = var_18472_begin_0, end = var_18472_end_0, end_mask = var_18472_end_mask_0, x = k_171_cast_fp16)[name = tensor("op_18472_cast_fp16")]; + tensor var_18476_begin_0 = const()[name = tensor("op_18476_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_18476_end_0 = const()[name = tensor("op_18476_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_18476_end_mask_0 = const()[name = tensor("op_18476_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18476_cast_fp16 = slice_by_index(begin = var_18476_begin_0, end = var_18476_end_0, end_mask = var_18476_end_mask_0, x = k_171_cast_fp16)[name = tensor("op_18476_cast_fp16")]; + tensor var_18480_begin_0 = const()[name = tensor("op_18480_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_18480_end_0 = const()[name = tensor("op_18480_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_18480_end_mask_0 = const()[name = tensor("op_18480_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18480_cast_fp16 = slice_by_index(begin = var_18480_begin_0, end = var_18480_end_0, end_mask = var_18480_end_mask_0, x = k_171_cast_fp16)[name = tensor("op_18480_cast_fp16")]; + tensor var_18484_begin_0 = const()[name = tensor("op_18484_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_18484_end_0 = const()[name = tensor("op_18484_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_18484_end_mask_0 = const()[name = tensor("op_18484_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18484_cast_fp16 = slice_by_index(begin = var_18484_begin_0, end = var_18484_end_0, end_mask = var_18484_end_mask_0, x = k_171_cast_fp16)[name = tensor("op_18484_cast_fp16")]; + tensor var_18488_begin_0 = const()[name = tensor("op_18488_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_18488_end_0 = const()[name = tensor("op_18488_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_18488_end_mask_0 = const()[name = tensor("op_18488_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18488_cast_fp16 = slice_by_index(begin = var_18488_begin_0, end = var_18488_end_0, end_mask = var_18488_end_mask_0, x = k_171_cast_fp16)[name = tensor("op_18488_cast_fp16")]; + tensor var_18492_begin_0 = const()[name = tensor("op_18492_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_18492_end_0 = const()[name = tensor("op_18492_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_18492_end_mask_0 = const()[name = tensor("op_18492_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18492_cast_fp16 = slice_by_index(begin = var_18492_begin_0, end = var_18492_end_0, end_mask = var_18492_end_mask_0, x = k_171_cast_fp16)[name = tensor("op_18492_cast_fp16")]; + tensor var_18496_begin_0 = const()[name = tensor("op_18496_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_18496_end_0 = const()[name = tensor("op_18496_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_18496_end_mask_0 = const()[name = tensor("op_18496_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18496_cast_fp16 = slice_by_index(begin = var_18496_begin_0, end = var_18496_end_0, end_mask = var_18496_end_mask_0, x = k_171_cast_fp16)[name = tensor("op_18496_cast_fp16")]; + tensor var_18500_begin_0 = const()[name = tensor("op_18500_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_18500_end_0 = const()[name = tensor("op_18500_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_18500_end_mask_0 = const()[name = tensor("op_18500_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18500_cast_fp16 = slice_by_index(begin = var_18500_begin_0, end = var_18500_end_0, end_mask = var_18500_end_mask_0, x = k_171_cast_fp16)[name = tensor("op_18500_cast_fp16")]; + tensor var_18504_begin_0 = const()[name = tensor("op_18504_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_18504_end_0 = const()[name = tensor("op_18504_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_18504_end_mask_0 = const()[name = tensor("op_18504_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18504_cast_fp16 = slice_by_index(begin = var_18504_begin_0, end = var_18504_end_0, end_mask = var_18504_end_mask_0, x = k_171_cast_fp16)[name = tensor("op_18504_cast_fp16")]; + tensor var_18506_begin_0 = const()[name = tensor("op_18506_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_18506_end_0 = const()[name = tensor("op_18506_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_18506_end_mask_0 = const()[name = tensor("op_18506_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18506_cast_fp16 = slice_by_index(begin = var_18506_begin_0, end = var_18506_end_0, end_mask = var_18506_end_mask_0, x = v_85_cast_fp16)[name = tensor("op_18506_cast_fp16")]; + tensor var_18510_begin_0 = const()[name = tensor("op_18510_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_18510_end_0 = const()[name = tensor("op_18510_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_18510_end_mask_0 = const()[name = tensor("op_18510_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18510_cast_fp16 = slice_by_index(begin = var_18510_begin_0, end = var_18510_end_0, end_mask = var_18510_end_mask_0, x = v_85_cast_fp16)[name = tensor("op_18510_cast_fp16")]; + tensor var_18514_begin_0 = const()[name = tensor("op_18514_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_18514_end_0 = const()[name = tensor("op_18514_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_18514_end_mask_0 = const()[name = tensor("op_18514_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18514_cast_fp16 = slice_by_index(begin = var_18514_begin_0, end = var_18514_end_0, end_mask = var_18514_end_mask_0, x = v_85_cast_fp16)[name = tensor("op_18514_cast_fp16")]; + tensor var_18518_begin_0 = const()[name = tensor("op_18518_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_18518_end_0 = const()[name = tensor("op_18518_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_18518_end_mask_0 = const()[name = tensor("op_18518_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18518_cast_fp16 = slice_by_index(begin = var_18518_begin_0, end = var_18518_end_0, end_mask = var_18518_end_mask_0, x = v_85_cast_fp16)[name = tensor("op_18518_cast_fp16")]; + tensor var_18522_begin_0 = const()[name = tensor("op_18522_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_18522_end_0 = const()[name = tensor("op_18522_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_18522_end_mask_0 = const()[name = tensor("op_18522_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18522_cast_fp16 = slice_by_index(begin = var_18522_begin_0, end = var_18522_end_0, end_mask = var_18522_end_mask_0, x = v_85_cast_fp16)[name = tensor("op_18522_cast_fp16")]; + tensor var_18526_begin_0 = const()[name = tensor("op_18526_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_18526_end_0 = const()[name = tensor("op_18526_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_18526_end_mask_0 = const()[name = tensor("op_18526_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18526_cast_fp16 = slice_by_index(begin = var_18526_begin_0, end = var_18526_end_0, end_mask = var_18526_end_mask_0, x = v_85_cast_fp16)[name = tensor("op_18526_cast_fp16")]; + tensor var_18530_begin_0 = const()[name = tensor("op_18530_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_18530_end_0 = const()[name = tensor("op_18530_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_18530_end_mask_0 = const()[name = tensor("op_18530_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18530_cast_fp16 = slice_by_index(begin = var_18530_begin_0, end = var_18530_end_0, end_mask = var_18530_end_mask_0, x = v_85_cast_fp16)[name = tensor("op_18530_cast_fp16")]; + tensor var_18534_begin_0 = const()[name = tensor("op_18534_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_18534_end_0 = const()[name = tensor("op_18534_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_18534_end_mask_0 = const()[name = tensor("op_18534_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18534_cast_fp16 = slice_by_index(begin = var_18534_begin_0, end = var_18534_end_0, end_mask = var_18534_end_mask_0, x = v_85_cast_fp16)[name = tensor("op_18534_cast_fp16")]; + tensor var_18538_begin_0 = const()[name = tensor("op_18538_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_18538_end_0 = const()[name = tensor("op_18538_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_18538_end_mask_0 = const()[name = tensor("op_18538_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18538_cast_fp16 = slice_by_index(begin = var_18538_begin_0, end = var_18538_end_0, end_mask = var_18538_end_mask_0, x = v_85_cast_fp16)[name = tensor("op_18538_cast_fp16")]; + tensor var_18542_begin_0 = const()[name = tensor("op_18542_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_18542_end_0 = const()[name = tensor("op_18542_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_18542_end_mask_0 = const()[name = tensor("op_18542_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18542_cast_fp16 = slice_by_index(begin = var_18542_begin_0, end = var_18542_end_0, end_mask = var_18542_end_mask_0, x = v_85_cast_fp16)[name = tensor("op_18542_cast_fp16")]; + tensor var_18546_begin_0 = const()[name = tensor("op_18546_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_18546_end_0 = const()[name = tensor("op_18546_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_18546_end_mask_0 = const()[name = tensor("op_18546_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18546_cast_fp16 = slice_by_index(begin = var_18546_begin_0, end = var_18546_end_0, end_mask = var_18546_end_mask_0, x = v_85_cast_fp16)[name = tensor("op_18546_cast_fp16")]; + tensor var_18550_begin_0 = const()[name = tensor("op_18550_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_18550_end_0 = const()[name = tensor("op_18550_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_18550_end_mask_0 = const()[name = tensor("op_18550_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18550_cast_fp16 = slice_by_index(begin = var_18550_begin_0, end = var_18550_end_0, end_mask = var_18550_end_mask_0, x = v_85_cast_fp16)[name = tensor("op_18550_cast_fp16")]; + tensor var_18554_begin_0 = const()[name = tensor("op_18554_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_18554_end_0 = const()[name = tensor("op_18554_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_18554_end_mask_0 = const()[name = tensor("op_18554_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18554_cast_fp16 = slice_by_index(begin = var_18554_begin_0, end = var_18554_end_0, end_mask = var_18554_end_mask_0, x = v_85_cast_fp16)[name = tensor("op_18554_cast_fp16")]; + tensor var_18558_begin_0 = const()[name = tensor("op_18558_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_18558_end_0 = const()[name = tensor("op_18558_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_18558_end_mask_0 = const()[name = tensor("op_18558_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18558_cast_fp16 = slice_by_index(begin = var_18558_begin_0, end = var_18558_end_0, end_mask = var_18558_end_mask_0, x = v_85_cast_fp16)[name = tensor("op_18558_cast_fp16")]; + tensor var_18562_begin_0 = const()[name = tensor("op_18562_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_18562_end_0 = const()[name = tensor("op_18562_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_18562_end_mask_0 = const()[name = tensor("op_18562_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18562_cast_fp16 = slice_by_index(begin = var_18562_begin_0, end = var_18562_end_0, end_mask = var_18562_end_mask_0, x = v_85_cast_fp16)[name = tensor("op_18562_cast_fp16")]; + tensor var_18566_begin_0 = const()[name = tensor("op_18566_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_18566_end_0 = const()[name = tensor("op_18566_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_18566_end_mask_0 = const()[name = tensor("op_18566_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18566_cast_fp16 = slice_by_index(begin = var_18566_begin_0, end = var_18566_end_0, end_mask = var_18566_end_mask_0, x = v_85_cast_fp16)[name = tensor("op_18566_cast_fp16")]; + tensor var_18570_begin_0 = const()[name = tensor("op_18570_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_18570_end_0 = const()[name = tensor("op_18570_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_18570_end_mask_0 = const()[name = tensor("op_18570_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18570_cast_fp16 = slice_by_index(begin = var_18570_begin_0, end = var_18570_end_0, end_mask = var_18570_end_mask_0, x = v_85_cast_fp16)[name = tensor("op_18570_cast_fp16")]; + tensor var_18574_begin_0 = const()[name = tensor("op_18574_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_18574_end_0 = const()[name = tensor("op_18574_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_18574_end_mask_0 = const()[name = tensor("op_18574_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18574_cast_fp16 = slice_by_index(begin = var_18574_begin_0, end = var_18574_end_0, end_mask = var_18574_end_mask_0, x = v_85_cast_fp16)[name = tensor("op_18574_cast_fp16")]; + tensor var_18578_begin_0 = const()[name = tensor("op_18578_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_18578_end_0 = const()[name = tensor("op_18578_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_18578_end_mask_0 = const()[name = tensor("op_18578_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18578_cast_fp16 = slice_by_index(begin = var_18578_begin_0, end = var_18578_end_0, end_mask = var_18578_end_mask_0, x = v_85_cast_fp16)[name = tensor("op_18578_cast_fp16")]; + tensor var_18582_begin_0 = const()[name = tensor("op_18582_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_18582_end_0 = const()[name = tensor("op_18582_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_18582_end_mask_0 = const()[name = tensor("op_18582_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18582_cast_fp16 = slice_by_index(begin = var_18582_begin_0, end = var_18582_end_0, end_mask = var_18582_end_mask_0, x = v_85_cast_fp16)[name = tensor("op_18582_cast_fp16")]; + tensor var_18586_equation_0 = const()[name = tensor("op_18586_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18586_cast_fp16 = einsum(equation = var_18586_equation_0, values = (var_18428_cast_fp16, var_18345_cast_fp16))[name = tensor("op_18586_cast_fp16")]; + tensor var_18587_to_fp16 = const()[name = tensor("op_18587_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1521_cast_fp16 = mul(x = var_18586_cast_fp16, y = var_18587_to_fp16)[name = tensor("aw_1521_cast_fp16")]; + tensor var_18590_equation_0 = const()[name = tensor("op_18590_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18590_cast_fp16 = einsum(equation = var_18590_equation_0, values = (var_18432_cast_fp16, var_18349_cast_fp16))[name = tensor("op_18590_cast_fp16")]; + tensor var_18591_to_fp16 = const()[name = tensor("op_18591_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1523_cast_fp16 = mul(x = var_18590_cast_fp16, y = var_18591_to_fp16)[name = tensor("aw_1523_cast_fp16")]; + tensor var_18594_equation_0 = const()[name = tensor("op_18594_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18594_cast_fp16 = einsum(equation = var_18594_equation_0, values = (var_18436_cast_fp16, var_18353_cast_fp16))[name = tensor("op_18594_cast_fp16")]; + tensor var_18595_to_fp16 = const()[name = tensor("op_18595_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1525_cast_fp16 = mul(x = var_18594_cast_fp16, y = var_18595_to_fp16)[name = tensor("aw_1525_cast_fp16")]; + tensor var_18598_equation_0 = const()[name = tensor("op_18598_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18598_cast_fp16 = einsum(equation = var_18598_equation_0, values = (var_18440_cast_fp16, var_18357_cast_fp16))[name = tensor("op_18598_cast_fp16")]; + tensor var_18599_to_fp16 = const()[name = tensor("op_18599_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1527_cast_fp16 = mul(x = var_18598_cast_fp16, y = var_18599_to_fp16)[name = tensor("aw_1527_cast_fp16")]; + tensor var_18602_equation_0 = const()[name = tensor("op_18602_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18602_cast_fp16 = einsum(equation = var_18602_equation_0, values = (var_18444_cast_fp16, var_18361_cast_fp16))[name = tensor("op_18602_cast_fp16")]; + tensor var_18603_to_fp16 = const()[name = tensor("op_18603_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1529_cast_fp16 = mul(x = var_18602_cast_fp16, y = var_18603_to_fp16)[name = tensor("aw_1529_cast_fp16")]; + tensor var_18606_equation_0 = const()[name = tensor("op_18606_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18606_cast_fp16 = einsum(equation = var_18606_equation_0, values = (var_18448_cast_fp16, var_18365_cast_fp16))[name = tensor("op_18606_cast_fp16")]; + tensor var_18607_to_fp16 = const()[name = tensor("op_18607_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1531_cast_fp16 = mul(x = var_18606_cast_fp16, y = var_18607_to_fp16)[name = tensor("aw_1531_cast_fp16")]; + tensor var_18610_equation_0 = const()[name = tensor("op_18610_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18610_cast_fp16 = einsum(equation = var_18610_equation_0, values = (var_18452_cast_fp16, var_18369_cast_fp16))[name = tensor("op_18610_cast_fp16")]; + tensor var_18611_to_fp16 = const()[name = tensor("op_18611_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1533_cast_fp16 = mul(x = var_18610_cast_fp16, y = var_18611_to_fp16)[name = tensor("aw_1533_cast_fp16")]; + tensor var_18614_equation_0 = const()[name = tensor("op_18614_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18614_cast_fp16 = einsum(equation = var_18614_equation_0, values = (var_18456_cast_fp16, var_18373_cast_fp16))[name = tensor("op_18614_cast_fp16")]; + tensor var_18615_to_fp16 = const()[name = tensor("op_18615_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1535_cast_fp16 = mul(x = var_18614_cast_fp16, y = var_18615_to_fp16)[name = tensor("aw_1535_cast_fp16")]; + tensor var_18618_equation_0 = const()[name = tensor("op_18618_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18618_cast_fp16 = einsum(equation = var_18618_equation_0, values = (var_18460_cast_fp16, var_18377_cast_fp16))[name = tensor("op_18618_cast_fp16")]; + tensor var_18619_to_fp16 = const()[name = tensor("op_18619_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1537_cast_fp16 = mul(x = var_18618_cast_fp16, y = var_18619_to_fp16)[name = tensor("aw_1537_cast_fp16")]; + tensor var_18622_equation_0 = const()[name = tensor("op_18622_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18622_cast_fp16 = einsum(equation = var_18622_equation_0, values = (var_18464_cast_fp16, var_18381_cast_fp16))[name = tensor("op_18622_cast_fp16")]; + tensor var_18623_to_fp16 = const()[name = tensor("op_18623_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1539_cast_fp16 = mul(x = var_18622_cast_fp16, y = var_18623_to_fp16)[name = tensor("aw_1539_cast_fp16")]; + tensor var_18626_equation_0 = const()[name = tensor("op_18626_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18626_cast_fp16 = einsum(equation = var_18626_equation_0, values = (var_18468_cast_fp16, var_18385_cast_fp16))[name = tensor("op_18626_cast_fp16")]; + tensor var_18627_to_fp16 = const()[name = tensor("op_18627_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1541_cast_fp16 = mul(x = var_18626_cast_fp16, y = var_18627_to_fp16)[name = tensor("aw_1541_cast_fp16")]; + tensor var_18630_equation_0 = const()[name = tensor("op_18630_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18630_cast_fp16 = einsum(equation = var_18630_equation_0, values = (var_18472_cast_fp16, var_18389_cast_fp16))[name = tensor("op_18630_cast_fp16")]; + tensor var_18631_to_fp16 = const()[name = tensor("op_18631_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1543_cast_fp16 = mul(x = var_18630_cast_fp16, y = var_18631_to_fp16)[name = tensor("aw_1543_cast_fp16")]; + tensor var_18634_equation_0 = const()[name = tensor("op_18634_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18634_cast_fp16 = einsum(equation = var_18634_equation_0, values = (var_18476_cast_fp16, var_18393_cast_fp16))[name = tensor("op_18634_cast_fp16")]; + tensor var_18635_to_fp16 = const()[name = tensor("op_18635_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1545_cast_fp16 = mul(x = var_18634_cast_fp16, y = var_18635_to_fp16)[name = tensor("aw_1545_cast_fp16")]; + tensor var_18638_equation_0 = const()[name = tensor("op_18638_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18638_cast_fp16 = einsum(equation = var_18638_equation_0, values = (var_18480_cast_fp16, var_18397_cast_fp16))[name = tensor("op_18638_cast_fp16")]; + tensor var_18639_to_fp16 = const()[name = tensor("op_18639_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1547_cast_fp16 = mul(x = var_18638_cast_fp16, y = var_18639_to_fp16)[name = tensor("aw_1547_cast_fp16")]; + tensor var_18642_equation_0 = const()[name = tensor("op_18642_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18642_cast_fp16 = einsum(equation = var_18642_equation_0, values = (var_18484_cast_fp16, var_18401_cast_fp16))[name = tensor("op_18642_cast_fp16")]; + tensor var_18643_to_fp16 = const()[name = tensor("op_18643_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1549_cast_fp16 = mul(x = var_18642_cast_fp16, y = var_18643_to_fp16)[name = tensor("aw_1549_cast_fp16")]; + tensor var_18646_equation_0 = const()[name = tensor("op_18646_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18646_cast_fp16 = einsum(equation = var_18646_equation_0, values = (var_18488_cast_fp16, var_18405_cast_fp16))[name = tensor("op_18646_cast_fp16")]; + tensor var_18647_to_fp16 = const()[name = tensor("op_18647_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1551_cast_fp16 = mul(x = var_18646_cast_fp16, y = var_18647_to_fp16)[name = tensor("aw_1551_cast_fp16")]; + tensor var_18650_equation_0 = const()[name = tensor("op_18650_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18650_cast_fp16 = einsum(equation = var_18650_equation_0, values = (var_18492_cast_fp16, var_18409_cast_fp16))[name = tensor("op_18650_cast_fp16")]; + tensor var_18651_to_fp16 = const()[name = tensor("op_18651_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1553_cast_fp16 = mul(x = var_18650_cast_fp16, y = var_18651_to_fp16)[name = tensor("aw_1553_cast_fp16")]; + tensor var_18654_equation_0 = const()[name = tensor("op_18654_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18654_cast_fp16 = einsum(equation = var_18654_equation_0, values = (var_18496_cast_fp16, var_18413_cast_fp16))[name = tensor("op_18654_cast_fp16")]; + tensor var_18655_to_fp16 = const()[name = tensor("op_18655_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1555_cast_fp16 = mul(x = var_18654_cast_fp16, y = var_18655_to_fp16)[name = tensor("aw_1555_cast_fp16")]; + tensor var_18658_equation_0 = const()[name = tensor("op_18658_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18658_cast_fp16 = einsum(equation = var_18658_equation_0, values = (var_18500_cast_fp16, var_18417_cast_fp16))[name = tensor("op_18658_cast_fp16")]; + tensor var_18659_to_fp16 = const()[name = tensor("op_18659_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1557_cast_fp16 = mul(x = var_18658_cast_fp16, y = var_18659_to_fp16)[name = tensor("aw_1557_cast_fp16")]; + tensor var_18662_equation_0 = const()[name = tensor("op_18662_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_18662_cast_fp16 = einsum(equation = var_18662_equation_0, values = (var_18504_cast_fp16, var_18421_cast_fp16))[name = tensor("op_18662_cast_fp16")]; + tensor var_18663_to_fp16 = const()[name = tensor("op_18663_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1559_cast_fp16 = mul(x = var_18662_cast_fp16, y = var_18663_to_fp16)[name = tensor("aw_1559_cast_fp16")]; + tensor var_18665_cast_fp16 = softmax(axis = var_2624, x = aw_1521_cast_fp16)[name = tensor("op_18665_cast_fp16")]; + tensor var_18666_cast_fp16 = softmax(axis = var_2624, x = aw_1523_cast_fp16)[name = tensor("op_18666_cast_fp16")]; + tensor var_18667_cast_fp16 = softmax(axis = var_2624, x = aw_1525_cast_fp16)[name = tensor("op_18667_cast_fp16")]; + tensor var_18668_cast_fp16 = softmax(axis = var_2624, x = aw_1527_cast_fp16)[name = tensor("op_18668_cast_fp16")]; + tensor var_18669_cast_fp16 = softmax(axis = var_2624, x = aw_1529_cast_fp16)[name = tensor("op_18669_cast_fp16")]; + tensor var_18670_cast_fp16 = softmax(axis = var_2624, x = aw_1531_cast_fp16)[name = tensor("op_18670_cast_fp16")]; + tensor var_18671_cast_fp16 = softmax(axis = var_2624, x = aw_1533_cast_fp16)[name = tensor("op_18671_cast_fp16")]; + tensor var_18672_cast_fp16 = softmax(axis = var_2624, x = aw_1535_cast_fp16)[name = tensor("op_18672_cast_fp16")]; + tensor var_18673_cast_fp16 = softmax(axis = var_2624, x = aw_1537_cast_fp16)[name = tensor("op_18673_cast_fp16")]; + tensor var_18674_cast_fp16 = softmax(axis = var_2624, x = aw_1539_cast_fp16)[name = tensor("op_18674_cast_fp16")]; + tensor var_18675_cast_fp16 = softmax(axis = var_2624, x = aw_1541_cast_fp16)[name = tensor("op_18675_cast_fp16")]; + tensor var_18676_cast_fp16 = softmax(axis = var_2624, x = aw_1543_cast_fp16)[name = tensor("op_18676_cast_fp16")]; + tensor var_18677_cast_fp16 = softmax(axis = var_2624, x = aw_1545_cast_fp16)[name = tensor("op_18677_cast_fp16")]; + tensor var_18678_cast_fp16 = softmax(axis = var_2624, x = aw_1547_cast_fp16)[name = tensor("op_18678_cast_fp16")]; + tensor var_18679_cast_fp16 = softmax(axis = var_2624, x = aw_1549_cast_fp16)[name = tensor("op_18679_cast_fp16")]; + tensor var_18680_cast_fp16 = softmax(axis = var_2624, x = aw_1551_cast_fp16)[name = tensor("op_18680_cast_fp16")]; + tensor var_18681_cast_fp16 = softmax(axis = var_2624, x = aw_1553_cast_fp16)[name = tensor("op_18681_cast_fp16")]; + tensor var_18682_cast_fp16 = softmax(axis = var_2624, x = aw_1555_cast_fp16)[name = tensor("op_18682_cast_fp16")]; + tensor var_18683_cast_fp16 = softmax(axis = var_2624, x = aw_1557_cast_fp16)[name = tensor("op_18683_cast_fp16")]; + tensor var_18684_cast_fp16 = softmax(axis = var_2624, x = aw_1559_cast_fp16)[name = tensor("op_18684_cast_fp16")]; + tensor var_18686_equation_0 = const()[name = tensor("op_18686_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18686_cast_fp16 = einsum(equation = var_18686_equation_0, values = (var_18506_cast_fp16, var_18665_cast_fp16))[name = tensor("op_18686_cast_fp16")]; + tensor var_18688_equation_0 = const()[name = tensor("op_18688_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18688_cast_fp16 = einsum(equation = var_18688_equation_0, values = (var_18510_cast_fp16, var_18666_cast_fp16))[name = tensor("op_18688_cast_fp16")]; + tensor var_18690_equation_0 = const()[name = tensor("op_18690_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18690_cast_fp16 = einsum(equation = var_18690_equation_0, values = (var_18514_cast_fp16, var_18667_cast_fp16))[name = tensor("op_18690_cast_fp16")]; + tensor var_18692_equation_0 = const()[name = tensor("op_18692_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18692_cast_fp16 = einsum(equation = var_18692_equation_0, values = (var_18518_cast_fp16, var_18668_cast_fp16))[name = tensor("op_18692_cast_fp16")]; + tensor var_18694_equation_0 = const()[name = tensor("op_18694_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18694_cast_fp16 = einsum(equation = var_18694_equation_0, values = (var_18522_cast_fp16, var_18669_cast_fp16))[name = tensor("op_18694_cast_fp16")]; + tensor var_18696_equation_0 = const()[name = tensor("op_18696_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18696_cast_fp16 = einsum(equation = var_18696_equation_0, values = (var_18526_cast_fp16, var_18670_cast_fp16))[name = tensor("op_18696_cast_fp16")]; + tensor var_18698_equation_0 = const()[name = tensor("op_18698_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18698_cast_fp16 = einsum(equation = var_18698_equation_0, values = (var_18530_cast_fp16, var_18671_cast_fp16))[name = tensor("op_18698_cast_fp16")]; + tensor var_18700_equation_0 = const()[name = tensor("op_18700_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18700_cast_fp16 = einsum(equation = var_18700_equation_0, values = (var_18534_cast_fp16, var_18672_cast_fp16))[name = tensor("op_18700_cast_fp16")]; + tensor var_18702_equation_0 = const()[name = tensor("op_18702_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18702_cast_fp16 = einsum(equation = var_18702_equation_0, values = (var_18538_cast_fp16, var_18673_cast_fp16))[name = tensor("op_18702_cast_fp16")]; + tensor var_18704_equation_0 = const()[name = tensor("op_18704_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18704_cast_fp16 = einsum(equation = var_18704_equation_0, values = (var_18542_cast_fp16, var_18674_cast_fp16))[name = tensor("op_18704_cast_fp16")]; + tensor var_18706_equation_0 = const()[name = tensor("op_18706_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18706_cast_fp16 = einsum(equation = var_18706_equation_0, values = (var_18546_cast_fp16, var_18675_cast_fp16))[name = tensor("op_18706_cast_fp16")]; + tensor var_18708_equation_0 = const()[name = tensor("op_18708_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18708_cast_fp16 = einsum(equation = var_18708_equation_0, values = (var_18550_cast_fp16, var_18676_cast_fp16))[name = tensor("op_18708_cast_fp16")]; + tensor var_18710_equation_0 = const()[name = tensor("op_18710_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18710_cast_fp16 = einsum(equation = var_18710_equation_0, values = (var_18554_cast_fp16, var_18677_cast_fp16))[name = tensor("op_18710_cast_fp16")]; + tensor var_18712_equation_0 = const()[name = tensor("op_18712_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18712_cast_fp16 = einsum(equation = var_18712_equation_0, values = (var_18558_cast_fp16, var_18678_cast_fp16))[name = tensor("op_18712_cast_fp16")]; + tensor var_18714_equation_0 = const()[name = tensor("op_18714_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18714_cast_fp16 = einsum(equation = var_18714_equation_0, values = (var_18562_cast_fp16, var_18679_cast_fp16))[name = tensor("op_18714_cast_fp16")]; + tensor var_18716_equation_0 = const()[name = tensor("op_18716_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18716_cast_fp16 = einsum(equation = var_18716_equation_0, values = (var_18566_cast_fp16, var_18680_cast_fp16))[name = tensor("op_18716_cast_fp16")]; + tensor var_18718_equation_0 = const()[name = tensor("op_18718_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18718_cast_fp16 = einsum(equation = var_18718_equation_0, values = (var_18570_cast_fp16, var_18681_cast_fp16))[name = tensor("op_18718_cast_fp16")]; + tensor var_18720_equation_0 = const()[name = tensor("op_18720_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18720_cast_fp16 = einsum(equation = var_18720_equation_0, values = (var_18574_cast_fp16, var_18682_cast_fp16))[name = tensor("op_18720_cast_fp16")]; + tensor var_18722_equation_0 = const()[name = tensor("op_18722_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18722_cast_fp16 = einsum(equation = var_18722_equation_0, values = (var_18578_cast_fp16, var_18683_cast_fp16))[name = tensor("op_18722_cast_fp16")]; + tensor var_18724_equation_0 = const()[name = tensor("op_18724_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18724_cast_fp16 = einsum(equation = var_18724_equation_0, values = (var_18582_cast_fp16, var_18684_cast_fp16))[name = tensor("op_18724_cast_fp16")]; + tensor input_285_interleave_0 = const()[name = tensor("input_285_interleave_0"), val = tensor(false)]; + tensor input_285_cast_fp16 = concat(axis = var_2624, interleave = input_285_interleave_0, values = (var_18686_cast_fp16, var_18688_cast_fp16, var_18690_cast_fp16, var_18692_cast_fp16, var_18694_cast_fp16, var_18696_cast_fp16, var_18698_cast_fp16, var_18700_cast_fp16, var_18702_cast_fp16, var_18704_cast_fp16, var_18706_cast_fp16, var_18708_cast_fp16, var_18710_cast_fp16, var_18712_cast_fp16, var_18714_cast_fp16, var_18716_cast_fp16, var_18718_cast_fp16, var_18720_cast_fp16, var_18722_cast_fp16, var_18724_cast_fp16))[name = tensor("input_285_cast_fp16")]; + tensor var_18730 = const()[name = tensor("op_18730"), val = tensor([1, 1])]; + tensor var_18732 = const()[name = tensor("op_18732"), val = tensor([1, 1])]; + tensor var_18734_pad_type_0 = const()[name = tensor("op_18734_pad_type_0"), val = tensor("custom")]; + tensor var_18734_pad_0 = const()[name = tensor("op_18734_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_7_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_7_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1458564032)))]; + tensor down_blocks_2_attentions_1_transformer_blocks_7_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_7_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1461840896)))]; + tensor var_18734_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_7_attn1_to_out_0_bias_to_fp16, dilations = var_18732, groups = var_2624, pad = var_18734_pad_0, pad_type = var_18734_pad_type_0, strides = var_18730, weight = down_blocks_2_attentions_1_transformer_blocks_7_attn1_to_out_0_weight_to_fp16, x = input_285_cast_fp16)[name = tensor("op_18734_cast_fp16")]; + tensor inputs_129_cast_fp16 = add(x = var_18734_cast_fp16, y = inputs_127_cast_fp16)[name = tensor("inputs_129_cast_fp16")]; + tensor hidden_states_181_axes_0 = const()[name = tensor("hidden_states_181_axes_0"), val = tensor([1])]; + tensor hidden_states_181_gamma_0_to_fp16 = const()[name = tensor("hidden_states_181_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1461843520)))]; + tensor hidden_states_181_beta_0_to_fp16 = const()[name = tensor("hidden_states_181_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1461846144)))]; + tensor var_18744_to_fp16 = const()[name = tensor("op_18744_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_181_cast_fp16 = layer_norm(axes = hidden_states_181_axes_0, beta = hidden_states_181_beta_0_to_fp16, epsilon = var_18744_to_fp16, gamma = hidden_states_181_gamma_0_to_fp16, x = inputs_129_cast_fp16)[name = tensor("hidden_states_181_cast_fp16")]; + tensor var_18759 = const()[name = tensor("op_18759"), val = tensor([1, 1])]; + tensor var_18761 = const()[name = tensor("op_18761"), val = tensor([1, 1])]; + tensor q_87_pad_type_0 = const()[name = tensor("q_87_pad_type_0"), val = tensor("custom")]; + tensor q_87_pad_0 = const()[name = tensor("q_87_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_7_attn2_to_q_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_7_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1461848768)))]; + tensor q_87_cast_fp16 = conv(dilations = var_18761, groups = var_2624, pad = q_87_pad_0, pad_type = q_87_pad_type_0, strides = var_18759, weight = down_blocks_2_attentions_1_transformer_blocks_7_attn2_to_q_weight_to_fp16, x = hidden_states_181_cast_fp16)[name = tensor("q_87_cast_fp16")]; + tensor var_18765 = const()[name = tensor("op_18765"), val = tensor([1, 1])]; + tensor var_18767 = const()[name = tensor("op_18767"), val = tensor([1, 1])]; + tensor k_173_pad_type_0 = const()[name = tensor("k_173_pad_type_0"), val = tensor("custom")]; + tensor k_173_pad_0 = const()[name = tensor("k_173_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_7_attn2_to_k_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_7_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1465125632)))]; + tensor k_173_cast_fp16 = conv(dilations = var_18767, groups = var_2624, pad = k_173_pad_0, pad_type = k_173_pad_type_0, strides = var_18765, weight = down_blocks_2_attentions_1_transformer_blocks_7_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_173_cast_fp16")]; + tensor var_18771 = const()[name = tensor("op_18771"), val = tensor([1, 1])]; + tensor var_18773 = const()[name = tensor("op_18773"), val = tensor([1, 1])]; + tensor v_87_pad_type_0 = const()[name = tensor("v_87_pad_type_0"), val = tensor("custom")]; + tensor v_87_pad_0 = const()[name = tensor("v_87_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_7_attn2_to_v_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_7_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1470368576)))]; + tensor v_87_cast_fp16 = conv(dilations = var_18773, groups = var_2624, pad = v_87_pad_0, pad_type = v_87_pad_type_0, strides = var_18771, weight = down_blocks_2_attentions_1_transformer_blocks_7_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_87_cast_fp16")]; + tensor var_18777_begin_0 = const()[name = tensor("op_18777_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_18777_end_0 = const()[name = tensor("op_18777_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_18777_end_mask_0 = const()[name = tensor("op_18777_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18777_cast_fp16 = slice_by_index(begin = var_18777_begin_0, end = var_18777_end_0, end_mask = var_18777_end_mask_0, x = q_87_cast_fp16)[name = tensor("op_18777_cast_fp16")]; + tensor var_18781_begin_0 = const()[name = tensor("op_18781_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_18781_end_0 = const()[name = tensor("op_18781_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_18781_end_mask_0 = const()[name = tensor("op_18781_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18781_cast_fp16 = slice_by_index(begin = var_18781_begin_0, end = var_18781_end_0, end_mask = var_18781_end_mask_0, x = q_87_cast_fp16)[name = tensor("op_18781_cast_fp16")]; + tensor var_18785_begin_0 = const()[name = tensor("op_18785_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_18785_end_0 = const()[name = tensor("op_18785_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_18785_end_mask_0 = const()[name = tensor("op_18785_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18785_cast_fp16 = slice_by_index(begin = var_18785_begin_0, end = var_18785_end_0, end_mask = var_18785_end_mask_0, x = q_87_cast_fp16)[name = tensor("op_18785_cast_fp16")]; + tensor var_18789_begin_0 = const()[name = tensor("op_18789_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_18789_end_0 = const()[name = tensor("op_18789_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_18789_end_mask_0 = const()[name = tensor("op_18789_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18789_cast_fp16 = slice_by_index(begin = var_18789_begin_0, end = var_18789_end_0, end_mask = var_18789_end_mask_0, x = q_87_cast_fp16)[name = tensor("op_18789_cast_fp16")]; + tensor var_18793_begin_0 = const()[name = tensor("op_18793_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_18793_end_0 = const()[name = tensor("op_18793_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_18793_end_mask_0 = const()[name = tensor("op_18793_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18793_cast_fp16 = slice_by_index(begin = var_18793_begin_0, end = var_18793_end_0, end_mask = var_18793_end_mask_0, x = q_87_cast_fp16)[name = tensor("op_18793_cast_fp16")]; + tensor var_18797_begin_0 = const()[name = tensor("op_18797_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_18797_end_0 = const()[name = tensor("op_18797_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_18797_end_mask_0 = const()[name = tensor("op_18797_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18797_cast_fp16 = slice_by_index(begin = var_18797_begin_0, end = var_18797_end_0, end_mask = var_18797_end_mask_0, x = q_87_cast_fp16)[name = tensor("op_18797_cast_fp16")]; + tensor var_18801_begin_0 = const()[name = tensor("op_18801_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_18801_end_0 = const()[name = tensor("op_18801_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_18801_end_mask_0 = const()[name = tensor("op_18801_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18801_cast_fp16 = slice_by_index(begin = var_18801_begin_0, end = var_18801_end_0, end_mask = var_18801_end_mask_0, x = q_87_cast_fp16)[name = tensor("op_18801_cast_fp16")]; + tensor var_18805_begin_0 = const()[name = tensor("op_18805_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_18805_end_0 = const()[name = tensor("op_18805_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_18805_end_mask_0 = const()[name = tensor("op_18805_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18805_cast_fp16 = slice_by_index(begin = var_18805_begin_0, end = var_18805_end_0, end_mask = var_18805_end_mask_0, x = q_87_cast_fp16)[name = tensor("op_18805_cast_fp16")]; + tensor var_18809_begin_0 = const()[name = tensor("op_18809_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_18809_end_0 = const()[name = tensor("op_18809_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_18809_end_mask_0 = const()[name = tensor("op_18809_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18809_cast_fp16 = slice_by_index(begin = var_18809_begin_0, end = var_18809_end_0, end_mask = var_18809_end_mask_0, x = q_87_cast_fp16)[name = tensor("op_18809_cast_fp16")]; + tensor var_18813_begin_0 = const()[name = tensor("op_18813_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_18813_end_0 = const()[name = tensor("op_18813_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_18813_end_mask_0 = const()[name = tensor("op_18813_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18813_cast_fp16 = slice_by_index(begin = var_18813_begin_0, end = var_18813_end_0, end_mask = var_18813_end_mask_0, x = q_87_cast_fp16)[name = tensor("op_18813_cast_fp16")]; + tensor var_18817_begin_0 = const()[name = tensor("op_18817_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_18817_end_0 = const()[name = tensor("op_18817_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_18817_end_mask_0 = const()[name = tensor("op_18817_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18817_cast_fp16 = slice_by_index(begin = var_18817_begin_0, end = var_18817_end_0, end_mask = var_18817_end_mask_0, x = q_87_cast_fp16)[name = tensor("op_18817_cast_fp16")]; + tensor var_18821_begin_0 = const()[name = tensor("op_18821_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_18821_end_0 = const()[name = tensor("op_18821_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_18821_end_mask_0 = const()[name = tensor("op_18821_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18821_cast_fp16 = slice_by_index(begin = var_18821_begin_0, end = var_18821_end_0, end_mask = var_18821_end_mask_0, x = q_87_cast_fp16)[name = tensor("op_18821_cast_fp16")]; + tensor var_18825_begin_0 = const()[name = tensor("op_18825_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_18825_end_0 = const()[name = tensor("op_18825_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_18825_end_mask_0 = const()[name = tensor("op_18825_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18825_cast_fp16 = slice_by_index(begin = var_18825_begin_0, end = var_18825_end_0, end_mask = var_18825_end_mask_0, x = q_87_cast_fp16)[name = tensor("op_18825_cast_fp16")]; + tensor var_18829_begin_0 = const()[name = tensor("op_18829_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_18829_end_0 = const()[name = tensor("op_18829_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_18829_end_mask_0 = const()[name = tensor("op_18829_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18829_cast_fp16 = slice_by_index(begin = var_18829_begin_0, end = var_18829_end_0, end_mask = var_18829_end_mask_0, x = q_87_cast_fp16)[name = tensor("op_18829_cast_fp16")]; + tensor var_18833_begin_0 = const()[name = tensor("op_18833_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_18833_end_0 = const()[name = tensor("op_18833_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_18833_end_mask_0 = const()[name = tensor("op_18833_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18833_cast_fp16 = slice_by_index(begin = var_18833_begin_0, end = var_18833_end_0, end_mask = var_18833_end_mask_0, x = q_87_cast_fp16)[name = tensor("op_18833_cast_fp16")]; + tensor var_18837_begin_0 = const()[name = tensor("op_18837_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_18837_end_0 = const()[name = tensor("op_18837_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_18837_end_mask_0 = const()[name = tensor("op_18837_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18837_cast_fp16 = slice_by_index(begin = var_18837_begin_0, end = var_18837_end_0, end_mask = var_18837_end_mask_0, x = q_87_cast_fp16)[name = tensor("op_18837_cast_fp16")]; + tensor var_18841_begin_0 = const()[name = tensor("op_18841_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_18841_end_0 = const()[name = tensor("op_18841_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_18841_end_mask_0 = const()[name = tensor("op_18841_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18841_cast_fp16 = slice_by_index(begin = var_18841_begin_0, end = var_18841_end_0, end_mask = var_18841_end_mask_0, x = q_87_cast_fp16)[name = tensor("op_18841_cast_fp16")]; + tensor var_18845_begin_0 = const()[name = tensor("op_18845_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_18845_end_0 = const()[name = tensor("op_18845_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_18845_end_mask_0 = const()[name = tensor("op_18845_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18845_cast_fp16 = slice_by_index(begin = var_18845_begin_0, end = var_18845_end_0, end_mask = var_18845_end_mask_0, x = q_87_cast_fp16)[name = tensor("op_18845_cast_fp16")]; + tensor var_18849_begin_0 = const()[name = tensor("op_18849_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_18849_end_0 = const()[name = tensor("op_18849_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_18849_end_mask_0 = const()[name = tensor("op_18849_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18849_cast_fp16 = slice_by_index(begin = var_18849_begin_0, end = var_18849_end_0, end_mask = var_18849_end_mask_0, x = q_87_cast_fp16)[name = tensor("op_18849_cast_fp16")]; + tensor var_18853_begin_0 = const()[name = tensor("op_18853_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_18853_end_0 = const()[name = tensor("op_18853_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_18853_end_mask_0 = const()[name = tensor("op_18853_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18853_cast_fp16 = slice_by_index(begin = var_18853_begin_0, end = var_18853_end_0, end_mask = var_18853_end_mask_0, x = q_87_cast_fp16)[name = tensor("op_18853_cast_fp16")]; + tensor k_175_perm_0 = const()[name = tensor("k_175_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_18860_begin_0 = const()[name = tensor("op_18860_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_18860_end_0 = const()[name = tensor("op_18860_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_18860_end_mask_0 = const()[name = tensor("op_18860_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_175_cast_fp16 = transpose(perm = k_175_perm_0, x = k_173_cast_fp16)[name = tensor("transpose_96")]; + tensor var_18860_cast_fp16 = slice_by_index(begin = var_18860_begin_0, end = var_18860_end_0, end_mask = var_18860_end_mask_0, x = k_175_cast_fp16)[name = tensor("op_18860_cast_fp16")]; + tensor var_18864_begin_0 = const()[name = tensor("op_18864_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_18864_end_0 = const()[name = tensor("op_18864_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_18864_end_mask_0 = const()[name = tensor("op_18864_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18864_cast_fp16 = slice_by_index(begin = var_18864_begin_0, end = var_18864_end_0, end_mask = var_18864_end_mask_0, x = k_175_cast_fp16)[name = tensor("op_18864_cast_fp16")]; + tensor var_18868_begin_0 = const()[name = tensor("op_18868_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_18868_end_0 = const()[name = tensor("op_18868_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_18868_end_mask_0 = const()[name = tensor("op_18868_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18868_cast_fp16 = slice_by_index(begin = var_18868_begin_0, end = var_18868_end_0, end_mask = var_18868_end_mask_0, x = k_175_cast_fp16)[name = tensor("op_18868_cast_fp16")]; + tensor var_18872_begin_0 = const()[name = tensor("op_18872_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_18872_end_0 = const()[name = tensor("op_18872_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_18872_end_mask_0 = const()[name = tensor("op_18872_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18872_cast_fp16 = slice_by_index(begin = var_18872_begin_0, end = var_18872_end_0, end_mask = var_18872_end_mask_0, x = k_175_cast_fp16)[name = tensor("op_18872_cast_fp16")]; + tensor var_18876_begin_0 = const()[name = tensor("op_18876_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_18876_end_0 = const()[name = tensor("op_18876_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_18876_end_mask_0 = const()[name = tensor("op_18876_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18876_cast_fp16 = slice_by_index(begin = var_18876_begin_0, end = var_18876_end_0, end_mask = var_18876_end_mask_0, x = k_175_cast_fp16)[name = tensor("op_18876_cast_fp16")]; + tensor var_18880_begin_0 = const()[name = tensor("op_18880_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_18880_end_0 = const()[name = tensor("op_18880_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_18880_end_mask_0 = const()[name = tensor("op_18880_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18880_cast_fp16 = slice_by_index(begin = var_18880_begin_0, end = var_18880_end_0, end_mask = var_18880_end_mask_0, x = k_175_cast_fp16)[name = tensor("op_18880_cast_fp16")]; + tensor var_18884_begin_0 = const()[name = tensor("op_18884_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_18884_end_0 = const()[name = tensor("op_18884_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_18884_end_mask_0 = const()[name = tensor("op_18884_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18884_cast_fp16 = slice_by_index(begin = var_18884_begin_0, end = var_18884_end_0, end_mask = var_18884_end_mask_0, x = k_175_cast_fp16)[name = tensor("op_18884_cast_fp16")]; + tensor var_18888_begin_0 = const()[name = tensor("op_18888_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_18888_end_0 = const()[name = tensor("op_18888_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_18888_end_mask_0 = const()[name = tensor("op_18888_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18888_cast_fp16 = slice_by_index(begin = var_18888_begin_0, end = var_18888_end_0, end_mask = var_18888_end_mask_0, x = k_175_cast_fp16)[name = tensor("op_18888_cast_fp16")]; + tensor var_18892_begin_0 = const()[name = tensor("op_18892_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_18892_end_0 = const()[name = tensor("op_18892_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_18892_end_mask_0 = const()[name = tensor("op_18892_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18892_cast_fp16 = slice_by_index(begin = var_18892_begin_0, end = var_18892_end_0, end_mask = var_18892_end_mask_0, x = k_175_cast_fp16)[name = tensor("op_18892_cast_fp16")]; + tensor var_18896_begin_0 = const()[name = tensor("op_18896_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_18896_end_0 = const()[name = tensor("op_18896_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_18896_end_mask_0 = const()[name = tensor("op_18896_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18896_cast_fp16 = slice_by_index(begin = var_18896_begin_0, end = var_18896_end_0, end_mask = var_18896_end_mask_0, x = k_175_cast_fp16)[name = tensor("op_18896_cast_fp16")]; + tensor var_18900_begin_0 = const()[name = tensor("op_18900_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_18900_end_0 = const()[name = tensor("op_18900_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_18900_end_mask_0 = const()[name = tensor("op_18900_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18900_cast_fp16 = slice_by_index(begin = var_18900_begin_0, end = var_18900_end_0, end_mask = var_18900_end_mask_0, x = k_175_cast_fp16)[name = tensor("op_18900_cast_fp16")]; + tensor var_18904_begin_0 = const()[name = tensor("op_18904_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_18904_end_0 = const()[name = tensor("op_18904_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_18904_end_mask_0 = const()[name = tensor("op_18904_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18904_cast_fp16 = slice_by_index(begin = var_18904_begin_0, end = var_18904_end_0, end_mask = var_18904_end_mask_0, x = k_175_cast_fp16)[name = tensor("op_18904_cast_fp16")]; + tensor var_18908_begin_0 = const()[name = tensor("op_18908_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_18908_end_0 = const()[name = tensor("op_18908_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_18908_end_mask_0 = const()[name = tensor("op_18908_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18908_cast_fp16 = slice_by_index(begin = var_18908_begin_0, end = var_18908_end_0, end_mask = var_18908_end_mask_0, x = k_175_cast_fp16)[name = tensor("op_18908_cast_fp16")]; + tensor var_18912_begin_0 = const()[name = tensor("op_18912_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_18912_end_0 = const()[name = tensor("op_18912_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_18912_end_mask_0 = const()[name = tensor("op_18912_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18912_cast_fp16 = slice_by_index(begin = var_18912_begin_0, end = var_18912_end_0, end_mask = var_18912_end_mask_0, x = k_175_cast_fp16)[name = tensor("op_18912_cast_fp16")]; + tensor var_18916_begin_0 = const()[name = tensor("op_18916_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_18916_end_0 = const()[name = tensor("op_18916_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_18916_end_mask_0 = const()[name = tensor("op_18916_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18916_cast_fp16 = slice_by_index(begin = var_18916_begin_0, end = var_18916_end_0, end_mask = var_18916_end_mask_0, x = k_175_cast_fp16)[name = tensor("op_18916_cast_fp16")]; + tensor var_18920_begin_0 = const()[name = tensor("op_18920_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_18920_end_0 = const()[name = tensor("op_18920_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_18920_end_mask_0 = const()[name = tensor("op_18920_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18920_cast_fp16 = slice_by_index(begin = var_18920_begin_0, end = var_18920_end_0, end_mask = var_18920_end_mask_0, x = k_175_cast_fp16)[name = tensor("op_18920_cast_fp16")]; + tensor var_18924_begin_0 = const()[name = tensor("op_18924_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_18924_end_0 = const()[name = tensor("op_18924_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_18924_end_mask_0 = const()[name = tensor("op_18924_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18924_cast_fp16 = slice_by_index(begin = var_18924_begin_0, end = var_18924_end_0, end_mask = var_18924_end_mask_0, x = k_175_cast_fp16)[name = tensor("op_18924_cast_fp16")]; + tensor var_18928_begin_0 = const()[name = tensor("op_18928_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_18928_end_0 = const()[name = tensor("op_18928_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_18928_end_mask_0 = const()[name = tensor("op_18928_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18928_cast_fp16 = slice_by_index(begin = var_18928_begin_0, end = var_18928_end_0, end_mask = var_18928_end_mask_0, x = k_175_cast_fp16)[name = tensor("op_18928_cast_fp16")]; + tensor var_18932_begin_0 = const()[name = tensor("op_18932_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_18932_end_0 = const()[name = tensor("op_18932_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_18932_end_mask_0 = const()[name = tensor("op_18932_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18932_cast_fp16 = slice_by_index(begin = var_18932_begin_0, end = var_18932_end_0, end_mask = var_18932_end_mask_0, x = k_175_cast_fp16)[name = tensor("op_18932_cast_fp16")]; + tensor var_18936_begin_0 = const()[name = tensor("op_18936_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_18936_end_0 = const()[name = tensor("op_18936_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_18936_end_mask_0 = const()[name = tensor("op_18936_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18936_cast_fp16 = slice_by_index(begin = var_18936_begin_0, end = var_18936_end_0, end_mask = var_18936_end_mask_0, x = k_175_cast_fp16)[name = tensor("op_18936_cast_fp16")]; + tensor var_18938_begin_0 = const()[name = tensor("op_18938_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_18938_end_0 = const()[name = tensor("op_18938_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_18938_end_mask_0 = const()[name = tensor("op_18938_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18938_cast_fp16 = slice_by_index(begin = var_18938_begin_0, end = var_18938_end_0, end_mask = var_18938_end_mask_0, x = v_87_cast_fp16)[name = tensor("op_18938_cast_fp16")]; + tensor var_18942_begin_0 = const()[name = tensor("op_18942_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_18942_end_0 = const()[name = tensor("op_18942_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_18942_end_mask_0 = const()[name = tensor("op_18942_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18942_cast_fp16 = slice_by_index(begin = var_18942_begin_0, end = var_18942_end_0, end_mask = var_18942_end_mask_0, x = v_87_cast_fp16)[name = tensor("op_18942_cast_fp16")]; + tensor var_18946_begin_0 = const()[name = tensor("op_18946_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_18946_end_0 = const()[name = tensor("op_18946_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_18946_end_mask_0 = const()[name = tensor("op_18946_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18946_cast_fp16 = slice_by_index(begin = var_18946_begin_0, end = var_18946_end_0, end_mask = var_18946_end_mask_0, x = v_87_cast_fp16)[name = tensor("op_18946_cast_fp16")]; + tensor var_18950_begin_0 = const()[name = tensor("op_18950_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_18950_end_0 = const()[name = tensor("op_18950_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_18950_end_mask_0 = const()[name = tensor("op_18950_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18950_cast_fp16 = slice_by_index(begin = var_18950_begin_0, end = var_18950_end_0, end_mask = var_18950_end_mask_0, x = v_87_cast_fp16)[name = tensor("op_18950_cast_fp16")]; + tensor var_18954_begin_0 = const()[name = tensor("op_18954_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_18954_end_0 = const()[name = tensor("op_18954_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_18954_end_mask_0 = const()[name = tensor("op_18954_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18954_cast_fp16 = slice_by_index(begin = var_18954_begin_0, end = var_18954_end_0, end_mask = var_18954_end_mask_0, x = v_87_cast_fp16)[name = tensor("op_18954_cast_fp16")]; + tensor var_18958_begin_0 = const()[name = tensor("op_18958_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_18958_end_0 = const()[name = tensor("op_18958_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_18958_end_mask_0 = const()[name = tensor("op_18958_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18958_cast_fp16 = slice_by_index(begin = var_18958_begin_0, end = var_18958_end_0, end_mask = var_18958_end_mask_0, x = v_87_cast_fp16)[name = tensor("op_18958_cast_fp16")]; + tensor var_18962_begin_0 = const()[name = tensor("op_18962_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_18962_end_0 = const()[name = tensor("op_18962_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_18962_end_mask_0 = const()[name = tensor("op_18962_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18962_cast_fp16 = slice_by_index(begin = var_18962_begin_0, end = var_18962_end_0, end_mask = var_18962_end_mask_0, x = v_87_cast_fp16)[name = tensor("op_18962_cast_fp16")]; + tensor var_18966_begin_0 = const()[name = tensor("op_18966_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_18966_end_0 = const()[name = tensor("op_18966_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_18966_end_mask_0 = const()[name = tensor("op_18966_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18966_cast_fp16 = slice_by_index(begin = var_18966_begin_0, end = var_18966_end_0, end_mask = var_18966_end_mask_0, x = v_87_cast_fp16)[name = tensor("op_18966_cast_fp16")]; + tensor var_18970_begin_0 = const()[name = tensor("op_18970_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_18970_end_0 = const()[name = tensor("op_18970_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_18970_end_mask_0 = const()[name = tensor("op_18970_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18970_cast_fp16 = slice_by_index(begin = var_18970_begin_0, end = var_18970_end_0, end_mask = var_18970_end_mask_0, x = v_87_cast_fp16)[name = tensor("op_18970_cast_fp16")]; + tensor var_18974_begin_0 = const()[name = tensor("op_18974_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_18974_end_0 = const()[name = tensor("op_18974_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_18974_end_mask_0 = const()[name = tensor("op_18974_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18974_cast_fp16 = slice_by_index(begin = var_18974_begin_0, end = var_18974_end_0, end_mask = var_18974_end_mask_0, x = v_87_cast_fp16)[name = tensor("op_18974_cast_fp16")]; + tensor var_18978_begin_0 = const()[name = tensor("op_18978_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_18978_end_0 = const()[name = tensor("op_18978_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_18978_end_mask_0 = const()[name = tensor("op_18978_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18978_cast_fp16 = slice_by_index(begin = var_18978_begin_0, end = var_18978_end_0, end_mask = var_18978_end_mask_0, x = v_87_cast_fp16)[name = tensor("op_18978_cast_fp16")]; + tensor var_18982_begin_0 = const()[name = tensor("op_18982_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_18982_end_0 = const()[name = tensor("op_18982_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_18982_end_mask_0 = const()[name = tensor("op_18982_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18982_cast_fp16 = slice_by_index(begin = var_18982_begin_0, end = var_18982_end_0, end_mask = var_18982_end_mask_0, x = v_87_cast_fp16)[name = tensor("op_18982_cast_fp16")]; + tensor var_18986_begin_0 = const()[name = tensor("op_18986_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_18986_end_0 = const()[name = tensor("op_18986_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_18986_end_mask_0 = const()[name = tensor("op_18986_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18986_cast_fp16 = slice_by_index(begin = var_18986_begin_0, end = var_18986_end_0, end_mask = var_18986_end_mask_0, x = v_87_cast_fp16)[name = tensor("op_18986_cast_fp16")]; + tensor var_18990_begin_0 = const()[name = tensor("op_18990_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_18990_end_0 = const()[name = tensor("op_18990_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_18990_end_mask_0 = const()[name = tensor("op_18990_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18990_cast_fp16 = slice_by_index(begin = var_18990_begin_0, end = var_18990_end_0, end_mask = var_18990_end_mask_0, x = v_87_cast_fp16)[name = tensor("op_18990_cast_fp16")]; + tensor var_18994_begin_0 = const()[name = tensor("op_18994_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_18994_end_0 = const()[name = tensor("op_18994_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_18994_end_mask_0 = const()[name = tensor("op_18994_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18994_cast_fp16 = slice_by_index(begin = var_18994_begin_0, end = var_18994_end_0, end_mask = var_18994_end_mask_0, x = v_87_cast_fp16)[name = tensor("op_18994_cast_fp16")]; + tensor var_18998_begin_0 = const()[name = tensor("op_18998_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_18998_end_0 = const()[name = tensor("op_18998_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_18998_end_mask_0 = const()[name = tensor("op_18998_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18998_cast_fp16 = slice_by_index(begin = var_18998_begin_0, end = var_18998_end_0, end_mask = var_18998_end_mask_0, x = v_87_cast_fp16)[name = tensor("op_18998_cast_fp16")]; + tensor var_19002_begin_0 = const()[name = tensor("op_19002_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_19002_end_0 = const()[name = tensor("op_19002_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_19002_end_mask_0 = const()[name = tensor("op_19002_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19002_cast_fp16 = slice_by_index(begin = var_19002_begin_0, end = var_19002_end_0, end_mask = var_19002_end_mask_0, x = v_87_cast_fp16)[name = tensor("op_19002_cast_fp16")]; + tensor var_19006_begin_0 = const()[name = tensor("op_19006_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_19006_end_0 = const()[name = tensor("op_19006_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_19006_end_mask_0 = const()[name = tensor("op_19006_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19006_cast_fp16 = slice_by_index(begin = var_19006_begin_0, end = var_19006_end_0, end_mask = var_19006_end_mask_0, x = v_87_cast_fp16)[name = tensor("op_19006_cast_fp16")]; + tensor var_19010_begin_0 = const()[name = tensor("op_19010_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_19010_end_0 = const()[name = tensor("op_19010_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_19010_end_mask_0 = const()[name = tensor("op_19010_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19010_cast_fp16 = slice_by_index(begin = var_19010_begin_0, end = var_19010_end_0, end_mask = var_19010_end_mask_0, x = v_87_cast_fp16)[name = tensor("op_19010_cast_fp16")]; + tensor var_19014_begin_0 = const()[name = tensor("op_19014_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_19014_end_0 = const()[name = tensor("op_19014_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_19014_end_mask_0 = const()[name = tensor("op_19014_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19014_cast_fp16 = slice_by_index(begin = var_19014_begin_0, end = var_19014_end_0, end_mask = var_19014_end_mask_0, x = v_87_cast_fp16)[name = tensor("op_19014_cast_fp16")]; + tensor var_19018_equation_0 = const()[name = tensor("op_19018_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19018_cast_fp16 = einsum(equation = var_19018_equation_0, values = (var_18860_cast_fp16, var_18777_cast_fp16))[name = tensor("op_19018_cast_fp16")]; + tensor var_19019_to_fp16 = const()[name = tensor("op_19019_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1561_cast_fp16 = mul(x = var_19018_cast_fp16, y = var_19019_to_fp16)[name = tensor("aw_1561_cast_fp16")]; + tensor var_19022_equation_0 = const()[name = tensor("op_19022_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19022_cast_fp16 = einsum(equation = var_19022_equation_0, values = (var_18864_cast_fp16, var_18781_cast_fp16))[name = tensor("op_19022_cast_fp16")]; + tensor var_19023_to_fp16 = const()[name = tensor("op_19023_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1563_cast_fp16 = mul(x = var_19022_cast_fp16, y = var_19023_to_fp16)[name = tensor("aw_1563_cast_fp16")]; + tensor var_19026_equation_0 = const()[name = tensor("op_19026_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19026_cast_fp16 = einsum(equation = var_19026_equation_0, values = (var_18868_cast_fp16, var_18785_cast_fp16))[name = tensor("op_19026_cast_fp16")]; + tensor var_19027_to_fp16 = const()[name = tensor("op_19027_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1565_cast_fp16 = mul(x = var_19026_cast_fp16, y = var_19027_to_fp16)[name = tensor("aw_1565_cast_fp16")]; + tensor var_19030_equation_0 = const()[name = tensor("op_19030_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19030_cast_fp16 = einsum(equation = var_19030_equation_0, values = (var_18872_cast_fp16, var_18789_cast_fp16))[name = tensor("op_19030_cast_fp16")]; + tensor var_19031_to_fp16 = const()[name = tensor("op_19031_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1567_cast_fp16 = mul(x = var_19030_cast_fp16, y = var_19031_to_fp16)[name = tensor("aw_1567_cast_fp16")]; + tensor var_19034_equation_0 = const()[name = tensor("op_19034_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19034_cast_fp16 = einsum(equation = var_19034_equation_0, values = (var_18876_cast_fp16, var_18793_cast_fp16))[name = tensor("op_19034_cast_fp16")]; + tensor var_19035_to_fp16 = const()[name = tensor("op_19035_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1569_cast_fp16 = mul(x = var_19034_cast_fp16, y = var_19035_to_fp16)[name = tensor("aw_1569_cast_fp16")]; + tensor var_19038_equation_0 = const()[name = tensor("op_19038_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19038_cast_fp16 = einsum(equation = var_19038_equation_0, values = (var_18880_cast_fp16, var_18797_cast_fp16))[name = tensor("op_19038_cast_fp16")]; + tensor var_19039_to_fp16 = const()[name = tensor("op_19039_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1571_cast_fp16 = mul(x = var_19038_cast_fp16, y = var_19039_to_fp16)[name = tensor("aw_1571_cast_fp16")]; + tensor var_19042_equation_0 = const()[name = tensor("op_19042_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19042_cast_fp16 = einsum(equation = var_19042_equation_0, values = (var_18884_cast_fp16, var_18801_cast_fp16))[name = tensor("op_19042_cast_fp16")]; + tensor var_19043_to_fp16 = const()[name = tensor("op_19043_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1573_cast_fp16 = mul(x = var_19042_cast_fp16, y = var_19043_to_fp16)[name = tensor("aw_1573_cast_fp16")]; + tensor var_19046_equation_0 = const()[name = tensor("op_19046_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19046_cast_fp16 = einsum(equation = var_19046_equation_0, values = (var_18888_cast_fp16, var_18805_cast_fp16))[name = tensor("op_19046_cast_fp16")]; + tensor var_19047_to_fp16 = const()[name = tensor("op_19047_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1575_cast_fp16 = mul(x = var_19046_cast_fp16, y = var_19047_to_fp16)[name = tensor("aw_1575_cast_fp16")]; + tensor var_19050_equation_0 = const()[name = tensor("op_19050_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19050_cast_fp16 = einsum(equation = var_19050_equation_0, values = (var_18892_cast_fp16, var_18809_cast_fp16))[name = tensor("op_19050_cast_fp16")]; + tensor var_19051_to_fp16 = const()[name = tensor("op_19051_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1577_cast_fp16 = mul(x = var_19050_cast_fp16, y = var_19051_to_fp16)[name = tensor("aw_1577_cast_fp16")]; + tensor var_19054_equation_0 = const()[name = tensor("op_19054_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19054_cast_fp16 = einsum(equation = var_19054_equation_0, values = (var_18896_cast_fp16, var_18813_cast_fp16))[name = tensor("op_19054_cast_fp16")]; + tensor var_19055_to_fp16 = const()[name = tensor("op_19055_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1579_cast_fp16 = mul(x = var_19054_cast_fp16, y = var_19055_to_fp16)[name = tensor("aw_1579_cast_fp16")]; + tensor var_19058_equation_0 = const()[name = tensor("op_19058_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19058_cast_fp16 = einsum(equation = var_19058_equation_0, values = (var_18900_cast_fp16, var_18817_cast_fp16))[name = tensor("op_19058_cast_fp16")]; + tensor var_19059_to_fp16 = const()[name = tensor("op_19059_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1581_cast_fp16 = mul(x = var_19058_cast_fp16, y = var_19059_to_fp16)[name = tensor("aw_1581_cast_fp16")]; + tensor var_19062_equation_0 = const()[name = tensor("op_19062_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19062_cast_fp16 = einsum(equation = var_19062_equation_0, values = (var_18904_cast_fp16, var_18821_cast_fp16))[name = tensor("op_19062_cast_fp16")]; + tensor var_19063_to_fp16 = const()[name = tensor("op_19063_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1583_cast_fp16 = mul(x = var_19062_cast_fp16, y = var_19063_to_fp16)[name = tensor("aw_1583_cast_fp16")]; + tensor var_19066_equation_0 = const()[name = tensor("op_19066_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19066_cast_fp16 = einsum(equation = var_19066_equation_0, values = (var_18908_cast_fp16, var_18825_cast_fp16))[name = tensor("op_19066_cast_fp16")]; + tensor var_19067_to_fp16 = const()[name = tensor("op_19067_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1585_cast_fp16 = mul(x = var_19066_cast_fp16, y = var_19067_to_fp16)[name = tensor("aw_1585_cast_fp16")]; + tensor var_19070_equation_0 = const()[name = tensor("op_19070_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19070_cast_fp16 = einsum(equation = var_19070_equation_0, values = (var_18912_cast_fp16, var_18829_cast_fp16))[name = tensor("op_19070_cast_fp16")]; + tensor var_19071_to_fp16 = const()[name = tensor("op_19071_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1587_cast_fp16 = mul(x = var_19070_cast_fp16, y = var_19071_to_fp16)[name = tensor("aw_1587_cast_fp16")]; + tensor var_19074_equation_0 = const()[name = tensor("op_19074_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19074_cast_fp16 = einsum(equation = var_19074_equation_0, values = (var_18916_cast_fp16, var_18833_cast_fp16))[name = tensor("op_19074_cast_fp16")]; + tensor var_19075_to_fp16 = const()[name = tensor("op_19075_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1589_cast_fp16 = mul(x = var_19074_cast_fp16, y = var_19075_to_fp16)[name = tensor("aw_1589_cast_fp16")]; + tensor var_19078_equation_0 = const()[name = tensor("op_19078_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19078_cast_fp16 = einsum(equation = var_19078_equation_0, values = (var_18920_cast_fp16, var_18837_cast_fp16))[name = tensor("op_19078_cast_fp16")]; + tensor var_19079_to_fp16 = const()[name = tensor("op_19079_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1591_cast_fp16 = mul(x = var_19078_cast_fp16, y = var_19079_to_fp16)[name = tensor("aw_1591_cast_fp16")]; + tensor var_19082_equation_0 = const()[name = tensor("op_19082_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19082_cast_fp16 = einsum(equation = var_19082_equation_0, values = (var_18924_cast_fp16, var_18841_cast_fp16))[name = tensor("op_19082_cast_fp16")]; + tensor var_19083_to_fp16 = const()[name = tensor("op_19083_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1593_cast_fp16 = mul(x = var_19082_cast_fp16, y = var_19083_to_fp16)[name = tensor("aw_1593_cast_fp16")]; + tensor var_19086_equation_0 = const()[name = tensor("op_19086_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19086_cast_fp16 = einsum(equation = var_19086_equation_0, values = (var_18928_cast_fp16, var_18845_cast_fp16))[name = tensor("op_19086_cast_fp16")]; + tensor var_19087_to_fp16 = const()[name = tensor("op_19087_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1595_cast_fp16 = mul(x = var_19086_cast_fp16, y = var_19087_to_fp16)[name = tensor("aw_1595_cast_fp16")]; + tensor var_19090_equation_0 = const()[name = tensor("op_19090_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19090_cast_fp16 = einsum(equation = var_19090_equation_0, values = (var_18932_cast_fp16, var_18849_cast_fp16))[name = tensor("op_19090_cast_fp16")]; + tensor var_19091_to_fp16 = const()[name = tensor("op_19091_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1597_cast_fp16 = mul(x = var_19090_cast_fp16, y = var_19091_to_fp16)[name = tensor("aw_1597_cast_fp16")]; + tensor var_19094_equation_0 = const()[name = tensor("op_19094_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19094_cast_fp16 = einsum(equation = var_19094_equation_0, values = (var_18936_cast_fp16, var_18853_cast_fp16))[name = tensor("op_19094_cast_fp16")]; + tensor var_19095_to_fp16 = const()[name = tensor("op_19095_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1599_cast_fp16 = mul(x = var_19094_cast_fp16, y = var_19095_to_fp16)[name = tensor("aw_1599_cast_fp16")]; + tensor var_19097_cast_fp16 = softmax(axis = var_2624, x = aw_1561_cast_fp16)[name = tensor("op_19097_cast_fp16")]; + tensor var_19098_cast_fp16 = softmax(axis = var_2624, x = aw_1563_cast_fp16)[name = tensor("op_19098_cast_fp16")]; + tensor var_19099_cast_fp16 = softmax(axis = var_2624, x = aw_1565_cast_fp16)[name = tensor("op_19099_cast_fp16")]; + tensor var_19100_cast_fp16 = softmax(axis = var_2624, x = aw_1567_cast_fp16)[name = tensor("op_19100_cast_fp16")]; + tensor var_19101_cast_fp16 = softmax(axis = var_2624, x = aw_1569_cast_fp16)[name = tensor("op_19101_cast_fp16")]; + tensor var_19102_cast_fp16 = softmax(axis = var_2624, x = aw_1571_cast_fp16)[name = tensor("op_19102_cast_fp16")]; + tensor var_19103_cast_fp16 = softmax(axis = var_2624, x = aw_1573_cast_fp16)[name = tensor("op_19103_cast_fp16")]; + tensor var_19104_cast_fp16 = softmax(axis = var_2624, x = aw_1575_cast_fp16)[name = tensor("op_19104_cast_fp16")]; + tensor var_19105_cast_fp16 = softmax(axis = var_2624, x = aw_1577_cast_fp16)[name = tensor("op_19105_cast_fp16")]; + tensor var_19106_cast_fp16 = softmax(axis = var_2624, x = aw_1579_cast_fp16)[name = tensor("op_19106_cast_fp16")]; + tensor var_19107_cast_fp16 = softmax(axis = var_2624, x = aw_1581_cast_fp16)[name = tensor("op_19107_cast_fp16")]; + tensor var_19108_cast_fp16 = softmax(axis = var_2624, x = aw_1583_cast_fp16)[name = tensor("op_19108_cast_fp16")]; + tensor var_19109_cast_fp16 = softmax(axis = var_2624, x = aw_1585_cast_fp16)[name = tensor("op_19109_cast_fp16")]; + tensor var_19110_cast_fp16 = softmax(axis = var_2624, x = aw_1587_cast_fp16)[name = tensor("op_19110_cast_fp16")]; + tensor var_19111_cast_fp16 = softmax(axis = var_2624, x = aw_1589_cast_fp16)[name = tensor("op_19111_cast_fp16")]; + tensor var_19112_cast_fp16 = softmax(axis = var_2624, x = aw_1591_cast_fp16)[name = tensor("op_19112_cast_fp16")]; + tensor var_19113_cast_fp16 = softmax(axis = var_2624, x = aw_1593_cast_fp16)[name = tensor("op_19113_cast_fp16")]; + tensor var_19114_cast_fp16 = softmax(axis = var_2624, x = aw_1595_cast_fp16)[name = tensor("op_19114_cast_fp16")]; + tensor var_19115_cast_fp16 = softmax(axis = var_2624, x = aw_1597_cast_fp16)[name = tensor("op_19115_cast_fp16")]; + tensor var_19116_cast_fp16 = softmax(axis = var_2624, x = aw_1599_cast_fp16)[name = tensor("op_19116_cast_fp16")]; + tensor var_19118_equation_0 = const()[name = tensor("op_19118_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19118_cast_fp16 = einsum(equation = var_19118_equation_0, values = (var_18938_cast_fp16, var_19097_cast_fp16))[name = tensor("op_19118_cast_fp16")]; + tensor var_19120_equation_0 = const()[name = tensor("op_19120_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19120_cast_fp16 = einsum(equation = var_19120_equation_0, values = (var_18942_cast_fp16, var_19098_cast_fp16))[name = tensor("op_19120_cast_fp16")]; + tensor var_19122_equation_0 = const()[name = tensor("op_19122_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19122_cast_fp16 = einsum(equation = var_19122_equation_0, values = (var_18946_cast_fp16, var_19099_cast_fp16))[name = tensor("op_19122_cast_fp16")]; + tensor var_19124_equation_0 = const()[name = tensor("op_19124_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19124_cast_fp16 = einsum(equation = var_19124_equation_0, values = (var_18950_cast_fp16, var_19100_cast_fp16))[name = tensor("op_19124_cast_fp16")]; + tensor var_19126_equation_0 = const()[name = tensor("op_19126_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19126_cast_fp16 = einsum(equation = var_19126_equation_0, values = (var_18954_cast_fp16, var_19101_cast_fp16))[name = tensor("op_19126_cast_fp16")]; + tensor var_19128_equation_0 = const()[name = tensor("op_19128_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19128_cast_fp16 = einsum(equation = var_19128_equation_0, values = (var_18958_cast_fp16, var_19102_cast_fp16))[name = tensor("op_19128_cast_fp16")]; + tensor var_19130_equation_0 = const()[name = tensor("op_19130_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19130_cast_fp16 = einsum(equation = var_19130_equation_0, values = (var_18962_cast_fp16, var_19103_cast_fp16))[name = tensor("op_19130_cast_fp16")]; + tensor var_19132_equation_0 = const()[name = tensor("op_19132_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19132_cast_fp16 = einsum(equation = var_19132_equation_0, values = (var_18966_cast_fp16, var_19104_cast_fp16))[name = tensor("op_19132_cast_fp16")]; + tensor var_19134_equation_0 = const()[name = tensor("op_19134_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19134_cast_fp16 = einsum(equation = var_19134_equation_0, values = (var_18970_cast_fp16, var_19105_cast_fp16))[name = tensor("op_19134_cast_fp16")]; + tensor var_19136_equation_0 = const()[name = tensor("op_19136_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19136_cast_fp16 = einsum(equation = var_19136_equation_0, values = (var_18974_cast_fp16, var_19106_cast_fp16))[name = tensor("op_19136_cast_fp16")]; + tensor var_19138_equation_0 = const()[name = tensor("op_19138_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19138_cast_fp16 = einsum(equation = var_19138_equation_0, values = (var_18978_cast_fp16, var_19107_cast_fp16))[name = tensor("op_19138_cast_fp16")]; + tensor var_19140_equation_0 = const()[name = tensor("op_19140_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19140_cast_fp16 = einsum(equation = var_19140_equation_0, values = (var_18982_cast_fp16, var_19108_cast_fp16))[name = tensor("op_19140_cast_fp16")]; + tensor var_19142_equation_0 = const()[name = tensor("op_19142_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19142_cast_fp16 = einsum(equation = var_19142_equation_0, values = (var_18986_cast_fp16, var_19109_cast_fp16))[name = tensor("op_19142_cast_fp16")]; + tensor var_19144_equation_0 = const()[name = tensor("op_19144_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19144_cast_fp16 = einsum(equation = var_19144_equation_0, values = (var_18990_cast_fp16, var_19110_cast_fp16))[name = tensor("op_19144_cast_fp16")]; + tensor var_19146_equation_0 = const()[name = tensor("op_19146_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19146_cast_fp16 = einsum(equation = var_19146_equation_0, values = (var_18994_cast_fp16, var_19111_cast_fp16))[name = tensor("op_19146_cast_fp16")]; + tensor var_19148_equation_0 = const()[name = tensor("op_19148_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19148_cast_fp16 = einsum(equation = var_19148_equation_0, values = (var_18998_cast_fp16, var_19112_cast_fp16))[name = tensor("op_19148_cast_fp16")]; + tensor var_19150_equation_0 = const()[name = tensor("op_19150_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19150_cast_fp16 = einsum(equation = var_19150_equation_0, values = (var_19002_cast_fp16, var_19113_cast_fp16))[name = tensor("op_19150_cast_fp16")]; + tensor var_19152_equation_0 = const()[name = tensor("op_19152_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19152_cast_fp16 = einsum(equation = var_19152_equation_0, values = (var_19006_cast_fp16, var_19114_cast_fp16))[name = tensor("op_19152_cast_fp16")]; + tensor var_19154_equation_0 = const()[name = tensor("op_19154_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19154_cast_fp16 = einsum(equation = var_19154_equation_0, values = (var_19010_cast_fp16, var_19115_cast_fp16))[name = tensor("op_19154_cast_fp16")]; + tensor var_19156_equation_0 = const()[name = tensor("op_19156_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19156_cast_fp16 = einsum(equation = var_19156_equation_0, values = (var_19014_cast_fp16, var_19116_cast_fp16))[name = tensor("op_19156_cast_fp16")]; + tensor input_287_interleave_0 = const()[name = tensor("input_287_interleave_0"), val = tensor(false)]; + tensor input_287_cast_fp16 = concat(axis = var_2624, interleave = input_287_interleave_0, values = (var_19118_cast_fp16, var_19120_cast_fp16, var_19122_cast_fp16, var_19124_cast_fp16, var_19126_cast_fp16, var_19128_cast_fp16, var_19130_cast_fp16, var_19132_cast_fp16, var_19134_cast_fp16, var_19136_cast_fp16, var_19138_cast_fp16, var_19140_cast_fp16, var_19142_cast_fp16, var_19144_cast_fp16, var_19146_cast_fp16, var_19148_cast_fp16, var_19150_cast_fp16, var_19152_cast_fp16, var_19154_cast_fp16, var_19156_cast_fp16))[name = tensor("input_287_cast_fp16")]; + tensor var_19162 = const()[name = tensor("op_19162"), val = tensor([1, 1])]; + tensor var_19164 = const()[name = tensor("op_19164"), val = tensor([1, 1])]; + tensor var_19166_pad_type_0 = const()[name = tensor("op_19166_pad_type_0"), val = tensor("custom")]; + tensor var_19166_pad_0 = const()[name = tensor("op_19166_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_7_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_7_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1475611520)))]; + tensor down_blocks_2_attentions_1_transformer_blocks_7_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_7_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1478888384)))]; + tensor var_19166_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_7_attn2_to_out_0_bias_to_fp16, dilations = var_19164, groups = var_2624, pad = var_19166_pad_0, pad_type = var_19166_pad_type_0, strides = var_19162, weight = down_blocks_2_attentions_1_transformer_blocks_7_attn2_to_out_0_weight_to_fp16, x = input_287_cast_fp16)[name = tensor("op_19166_cast_fp16")]; + tensor inputs_131_cast_fp16 = add(x = var_19166_cast_fp16, y = inputs_129_cast_fp16)[name = tensor("inputs_131_cast_fp16")]; + tensor input_289_axes_0 = const()[name = tensor("input_289_axes_0"), val = tensor([1])]; + tensor input_289_gamma_0_to_fp16 = const()[name = tensor("input_289_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1478891008)))]; + tensor input_289_beta_0_to_fp16 = const()[name = tensor("input_289_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1478893632)))]; + tensor var_19176_to_fp16 = const()[name = tensor("op_19176_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_289_cast_fp16 = layer_norm(axes = input_289_axes_0, beta = input_289_beta_0_to_fp16, epsilon = var_19176_to_fp16, gamma = input_289_gamma_0_to_fp16, x = inputs_131_cast_fp16)[name = tensor("input_289_cast_fp16")]; + tensor var_19192 = const()[name = tensor("op_19192"), val = tensor([1, 1])]; + tensor var_19194 = const()[name = tensor("op_19194"), val = tensor([1, 1])]; + tensor var_19196_pad_type_0 = const()[name = tensor("op_19196_pad_type_0"), val = tensor("custom")]; + tensor var_19196_pad_0 = const()[name = tensor("op_19196_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_7_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_7_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1478896256)))]; + tensor down_blocks_2_attentions_1_transformer_blocks_7_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_7_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1505110720)))]; + tensor var_19196_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_7_ff_net_0_proj_bias_to_fp16, dilations = var_19194, groups = var_2624, pad = var_19196_pad_0, pad_type = var_19196_pad_type_0, strides = var_19192, weight = down_blocks_2_attentions_1_transformer_blocks_7_ff_net_0_proj_weight_to_fp16, x = input_289_cast_fp16)[name = tensor("op_19196_cast_fp16")]; + tensor var_19197_split_sizes_0 = const()[name = tensor("op_19197_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_19197_axis_0 = const()[name = tensor("op_19197_axis_0"), val = tensor(1)]; + tensor var_19197_cast_fp16_0, tensor var_19197_cast_fp16_1 = split(axis = var_19197_axis_0, split_sizes = var_19197_split_sizes_0, x = var_19196_cast_fp16)[name = tensor("op_19197_cast_fp16")]; + tensor var_19199_mode_0 = const()[name = tensor("op_19199_mode_0"), val = tensor("EXACT")]; + tensor var_19199_cast_fp16 = gelu(mode = var_19199_mode_0, x = var_19197_cast_fp16_1)[name = tensor("op_19199_cast_fp16")]; + tensor input_291_cast_fp16 = mul(x = var_19197_cast_fp16_0, y = var_19199_cast_fp16)[name = tensor("input_291_cast_fp16")]; + tensor var_19203 = const()[name = tensor("op_19203"), val = tensor([1, 1])]; + tensor var_19205 = const()[name = tensor("op_19205"), val = tensor([1, 1])]; + tensor var_19207_pad_type_0 = const()[name = tensor("op_19207_pad_type_0"), val = tensor("custom")]; + tensor var_19207_pad_0 = const()[name = tensor("op_19207_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_7_ff_net_2_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_7_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1505131264)))]; + tensor down_blocks_2_attentions_1_transformer_blocks_7_ff_net_2_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_7_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1518238528)))]; + tensor var_19207_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_7_ff_net_2_bias_to_fp16, dilations = var_19205, groups = var_2624, pad = var_19207_pad_0, pad_type = var_19207_pad_type_0, strides = var_19203, weight = down_blocks_2_attentions_1_transformer_blocks_7_ff_net_2_weight_to_fp16, x = input_291_cast_fp16)[name = tensor("op_19207_cast_fp16")]; + tensor inputs_133_cast_fp16 = add(x = var_19207_cast_fp16, y = inputs_131_cast_fp16)[name = tensor("inputs_133_cast_fp16")]; + tensor hidden_states_185_axes_0 = const()[name = tensor("hidden_states_185_axes_0"), val = tensor([1])]; + tensor hidden_states_185_gamma_0_to_fp16 = const()[name = tensor("hidden_states_185_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1518241152)))]; + tensor hidden_states_185_beta_0_to_fp16 = const()[name = tensor("hidden_states_185_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1518243776)))]; + tensor var_19223_to_fp16 = const()[name = tensor("op_19223_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_185_cast_fp16 = layer_norm(axes = hidden_states_185_axes_0, beta = hidden_states_185_beta_0_to_fp16, epsilon = var_19223_to_fp16, gamma = hidden_states_185_gamma_0_to_fp16, x = inputs_133_cast_fp16)[name = tensor("hidden_states_185_cast_fp16")]; + tensor var_19238 = const()[name = tensor("op_19238"), val = tensor([1, 1])]; + tensor var_19240 = const()[name = tensor("op_19240"), val = tensor([1, 1])]; + tensor q_89_pad_type_0 = const()[name = tensor("q_89_pad_type_0"), val = tensor("custom")]; + tensor q_89_pad_0 = const()[name = tensor("q_89_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_8_attn1_to_q_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_8_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1518246400)))]; + tensor q_89_cast_fp16 = conv(dilations = var_19240, groups = var_2624, pad = q_89_pad_0, pad_type = q_89_pad_type_0, strides = var_19238, weight = down_blocks_2_attentions_1_transformer_blocks_8_attn1_to_q_weight_to_fp16, x = hidden_states_185_cast_fp16)[name = tensor("q_89_cast_fp16")]; + tensor var_19244 = const()[name = tensor("op_19244"), val = tensor([1, 1])]; + tensor var_19246 = const()[name = tensor("op_19246"), val = tensor([1, 1])]; + tensor k_177_pad_type_0 = const()[name = tensor("k_177_pad_type_0"), val = tensor("custom")]; + tensor k_177_pad_0 = const()[name = tensor("k_177_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_8_attn1_to_k_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_8_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1521523264)))]; + tensor k_177_cast_fp16 = conv(dilations = var_19246, groups = var_2624, pad = k_177_pad_0, pad_type = k_177_pad_type_0, strides = var_19244, weight = down_blocks_2_attentions_1_transformer_blocks_8_attn1_to_k_weight_to_fp16, x = hidden_states_185_cast_fp16)[name = tensor("k_177_cast_fp16")]; + tensor var_19250 = const()[name = tensor("op_19250"), val = tensor([1, 1])]; + tensor var_19252 = const()[name = tensor("op_19252"), val = tensor([1, 1])]; + tensor v_89_pad_type_0 = const()[name = tensor("v_89_pad_type_0"), val = tensor("custom")]; + tensor v_89_pad_0 = const()[name = tensor("v_89_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_8_attn1_to_v_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_8_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1524800128)))]; + tensor v_89_cast_fp16 = conv(dilations = var_19252, groups = var_2624, pad = v_89_pad_0, pad_type = v_89_pad_type_0, strides = var_19250, weight = down_blocks_2_attentions_1_transformer_blocks_8_attn1_to_v_weight_to_fp16, x = hidden_states_185_cast_fp16)[name = tensor("v_89_cast_fp16")]; + tensor var_19256_begin_0 = const()[name = tensor("op_19256_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_19256_end_0 = const()[name = tensor("op_19256_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_19256_end_mask_0 = const()[name = tensor("op_19256_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19256_cast_fp16 = slice_by_index(begin = var_19256_begin_0, end = var_19256_end_0, end_mask = var_19256_end_mask_0, x = q_89_cast_fp16)[name = tensor("op_19256_cast_fp16")]; + tensor var_19260_begin_0 = const()[name = tensor("op_19260_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_19260_end_0 = const()[name = tensor("op_19260_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_19260_end_mask_0 = const()[name = tensor("op_19260_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19260_cast_fp16 = slice_by_index(begin = var_19260_begin_0, end = var_19260_end_0, end_mask = var_19260_end_mask_0, x = q_89_cast_fp16)[name = tensor("op_19260_cast_fp16")]; + tensor var_19264_begin_0 = const()[name = tensor("op_19264_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_19264_end_0 = const()[name = tensor("op_19264_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_19264_end_mask_0 = const()[name = tensor("op_19264_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19264_cast_fp16 = slice_by_index(begin = var_19264_begin_0, end = var_19264_end_0, end_mask = var_19264_end_mask_0, x = q_89_cast_fp16)[name = tensor("op_19264_cast_fp16")]; + tensor var_19268_begin_0 = const()[name = tensor("op_19268_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_19268_end_0 = const()[name = tensor("op_19268_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_19268_end_mask_0 = const()[name = tensor("op_19268_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19268_cast_fp16 = slice_by_index(begin = var_19268_begin_0, end = var_19268_end_0, end_mask = var_19268_end_mask_0, x = q_89_cast_fp16)[name = tensor("op_19268_cast_fp16")]; + tensor var_19272_begin_0 = const()[name = tensor("op_19272_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_19272_end_0 = const()[name = tensor("op_19272_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_19272_end_mask_0 = const()[name = tensor("op_19272_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19272_cast_fp16 = slice_by_index(begin = var_19272_begin_0, end = var_19272_end_0, end_mask = var_19272_end_mask_0, x = q_89_cast_fp16)[name = tensor("op_19272_cast_fp16")]; + tensor var_19276_begin_0 = const()[name = tensor("op_19276_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_19276_end_0 = const()[name = tensor("op_19276_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_19276_end_mask_0 = const()[name = tensor("op_19276_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19276_cast_fp16 = slice_by_index(begin = var_19276_begin_0, end = var_19276_end_0, end_mask = var_19276_end_mask_0, x = q_89_cast_fp16)[name = tensor("op_19276_cast_fp16")]; + tensor var_19280_begin_0 = const()[name = tensor("op_19280_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_19280_end_0 = const()[name = tensor("op_19280_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_19280_end_mask_0 = const()[name = tensor("op_19280_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19280_cast_fp16 = slice_by_index(begin = var_19280_begin_0, end = var_19280_end_0, end_mask = var_19280_end_mask_0, x = q_89_cast_fp16)[name = tensor("op_19280_cast_fp16")]; + tensor var_19284_begin_0 = const()[name = tensor("op_19284_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_19284_end_0 = const()[name = tensor("op_19284_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_19284_end_mask_0 = const()[name = tensor("op_19284_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19284_cast_fp16 = slice_by_index(begin = var_19284_begin_0, end = var_19284_end_0, end_mask = var_19284_end_mask_0, x = q_89_cast_fp16)[name = tensor("op_19284_cast_fp16")]; + tensor var_19288_begin_0 = const()[name = tensor("op_19288_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_19288_end_0 = const()[name = tensor("op_19288_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_19288_end_mask_0 = const()[name = tensor("op_19288_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19288_cast_fp16 = slice_by_index(begin = var_19288_begin_0, end = var_19288_end_0, end_mask = var_19288_end_mask_0, x = q_89_cast_fp16)[name = tensor("op_19288_cast_fp16")]; + tensor var_19292_begin_0 = const()[name = tensor("op_19292_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_19292_end_0 = const()[name = tensor("op_19292_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_19292_end_mask_0 = const()[name = tensor("op_19292_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19292_cast_fp16 = slice_by_index(begin = var_19292_begin_0, end = var_19292_end_0, end_mask = var_19292_end_mask_0, x = q_89_cast_fp16)[name = tensor("op_19292_cast_fp16")]; + tensor var_19296_begin_0 = const()[name = tensor("op_19296_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_19296_end_0 = const()[name = tensor("op_19296_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_19296_end_mask_0 = const()[name = tensor("op_19296_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19296_cast_fp16 = slice_by_index(begin = var_19296_begin_0, end = var_19296_end_0, end_mask = var_19296_end_mask_0, x = q_89_cast_fp16)[name = tensor("op_19296_cast_fp16")]; + tensor var_19300_begin_0 = const()[name = tensor("op_19300_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_19300_end_0 = const()[name = tensor("op_19300_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_19300_end_mask_0 = const()[name = tensor("op_19300_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19300_cast_fp16 = slice_by_index(begin = var_19300_begin_0, end = var_19300_end_0, end_mask = var_19300_end_mask_0, x = q_89_cast_fp16)[name = tensor("op_19300_cast_fp16")]; + tensor var_19304_begin_0 = const()[name = tensor("op_19304_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_19304_end_0 = const()[name = tensor("op_19304_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_19304_end_mask_0 = const()[name = tensor("op_19304_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19304_cast_fp16 = slice_by_index(begin = var_19304_begin_0, end = var_19304_end_0, end_mask = var_19304_end_mask_0, x = q_89_cast_fp16)[name = tensor("op_19304_cast_fp16")]; + tensor var_19308_begin_0 = const()[name = tensor("op_19308_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_19308_end_0 = const()[name = tensor("op_19308_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_19308_end_mask_0 = const()[name = tensor("op_19308_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19308_cast_fp16 = slice_by_index(begin = var_19308_begin_0, end = var_19308_end_0, end_mask = var_19308_end_mask_0, x = q_89_cast_fp16)[name = tensor("op_19308_cast_fp16")]; + tensor var_19312_begin_0 = const()[name = tensor("op_19312_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_19312_end_0 = const()[name = tensor("op_19312_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_19312_end_mask_0 = const()[name = tensor("op_19312_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19312_cast_fp16 = slice_by_index(begin = var_19312_begin_0, end = var_19312_end_0, end_mask = var_19312_end_mask_0, x = q_89_cast_fp16)[name = tensor("op_19312_cast_fp16")]; + tensor var_19316_begin_0 = const()[name = tensor("op_19316_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_19316_end_0 = const()[name = tensor("op_19316_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_19316_end_mask_0 = const()[name = tensor("op_19316_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19316_cast_fp16 = slice_by_index(begin = var_19316_begin_0, end = var_19316_end_0, end_mask = var_19316_end_mask_0, x = q_89_cast_fp16)[name = tensor("op_19316_cast_fp16")]; + tensor var_19320_begin_0 = const()[name = tensor("op_19320_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_19320_end_0 = const()[name = tensor("op_19320_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_19320_end_mask_0 = const()[name = tensor("op_19320_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19320_cast_fp16 = slice_by_index(begin = var_19320_begin_0, end = var_19320_end_0, end_mask = var_19320_end_mask_0, x = q_89_cast_fp16)[name = tensor("op_19320_cast_fp16")]; + tensor var_19324_begin_0 = const()[name = tensor("op_19324_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_19324_end_0 = const()[name = tensor("op_19324_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_19324_end_mask_0 = const()[name = tensor("op_19324_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19324_cast_fp16 = slice_by_index(begin = var_19324_begin_0, end = var_19324_end_0, end_mask = var_19324_end_mask_0, x = q_89_cast_fp16)[name = tensor("op_19324_cast_fp16")]; + tensor var_19328_begin_0 = const()[name = tensor("op_19328_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_19328_end_0 = const()[name = tensor("op_19328_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_19328_end_mask_0 = const()[name = tensor("op_19328_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19328_cast_fp16 = slice_by_index(begin = var_19328_begin_0, end = var_19328_end_0, end_mask = var_19328_end_mask_0, x = q_89_cast_fp16)[name = tensor("op_19328_cast_fp16")]; + tensor var_19332_begin_0 = const()[name = tensor("op_19332_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_19332_end_0 = const()[name = tensor("op_19332_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_19332_end_mask_0 = const()[name = tensor("op_19332_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19332_cast_fp16 = slice_by_index(begin = var_19332_begin_0, end = var_19332_end_0, end_mask = var_19332_end_mask_0, x = q_89_cast_fp16)[name = tensor("op_19332_cast_fp16")]; + tensor k_179_perm_0 = const()[name = tensor("k_179_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_19339_begin_0 = const()[name = tensor("op_19339_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_19339_end_0 = const()[name = tensor("op_19339_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_19339_end_mask_0 = const()[name = tensor("op_19339_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_179_cast_fp16 = transpose(perm = k_179_perm_0, x = k_177_cast_fp16)[name = tensor("transpose_95")]; + tensor var_19339_cast_fp16 = slice_by_index(begin = var_19339_begin_0, end = var_19339_end_0, end_mask = var_19339_end_mask_0, x = k_179_cast_fp16)[name = tensor("op_19339_cast_fp16")]; + tensor var_19343_begin_0 = const()[name = tensor("op_19343_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_19343_end_0 = const()[name = tensor("op_19343_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_19343_end_mask_0 = const()[name = tensor("op_19343_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19343_cast_fp16 = slice_by_index(begin = var_19343_begin_0, end = var_19343_end_0, end_mask = var_19343_end_mask_0, x = k_179_cast_fp16)[name = tensor("op_19343_cast_fp16")]; + tensor var_19347_begin_0 = const()[name = tensor("op_19347_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_19347_end_0 = const()[name = tensor("op_19347_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_19347_end_mask_0 = const()[name = tensor("op_19347_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19347_cast_fp16 = slice_by_index(begin = var_19347_begin_0, end = var_19347_end_0, end_mask = var_19347_end_mask_0, x = k_179_cast_fp16)[name = tensor("op_19347_cast_fp16")]; + tensor var_19351_begin_0 = const()[name = tensor("op_19351_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_19351_end_0 = const()[name = tensor("op_19351_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_19351_end_mask_0 = const()[name = tensor("op_19351_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19351_cast_fp16 = slice_by_index(begin = var_19351_begin_0, end = var_19351_end_0, end_mask = var_19351_end_mask_0, x = k_179_cast_fp16)[name = tensor("op_19351_cast_fp16")]; + tensor var_19355_begin_0 = const()[name = tensor("op_19355_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_19355_end_0 = const()[name = tensor("op_19355_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_19355_end_mask_0 = const()[name = tensor("op_19355_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19355_cast_fp16 = slice_by_index(begin = var_19355_begin_0, end = var_19355_end_0, end_mask = var_19355_end_mask_0, x = k_179_cast_fp16)[name = tensor("op_19355_cast_fp16")]; + tensor var_19359_begin_0 = const()[name = tensor("op_19359_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_19359_end_0 = const()[name = tensor("op_19359_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_19359_end_mask_0 = const()[name = tensor("op_19359_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19359_cast_fp16 = slice_by_index(begin = var_19359_begin_0, end = var_19359_end_0, end_mask = var_19359_end_mask_0, x = k_179_cast_fp16)[name = tensor("op_19359_cast_fp16")]; + tensor var_19363_begin_0 = const()[name = tensor("op_19363_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_19363_end_0 = const()[name = tensor("op_19363_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_19363_end_mask_0 = const()[name = tensor("op_19363_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19363_cast_fp16 = slice_by_index(begin = var_19363_begin_0, end = var_19363_end_0, end_mask = var_19363_end_mask_0, x = k_179_cast_fp16)[name = tensor("op_19363_cast_fp16")]; + tensor var_19367_begin_0 = const()[name = tensor("op_19367_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_19367_end_0 = const()[name = tensor("op_19367_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_19367_end_mask_0 = const()[name = tensor("op_19367_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19367_cast_fp16 = slice_by_index(begin = var_19367_begin_0, end = var_19367_end_0, end_mask = var_19367_end_mask_0, x = k_179_cast_fp16)[name = tensor("op_19367_cast_fp16")]; + tensor var_19371_begin_0 = const()[name = tensor("op_19371_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_19371_end_0 = const()[name = tensor("op_19371_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_19371_end_mask_0 = const()[name = tensor("op_19371_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19371_cast_fp16 = slice_by_index(begin = var_19371_begin_0, end = var_19371_end_0, end_mask = var_19371_end_mask_0, x = k_179_cast_fp16)[name = tensor("op_19371_cast_fp16")]; + tensor var_19375_begin_0 = const()[name = tensor("op_19375_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_19375_end_0 = const()[name = tensor("op_19375_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_19375_end_mask_0 = const()[name = tensor("op_19375_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19375_cast_fp16 = slice_by_index(begin = var_19375_begin_0, end = var_19375_end_0, end_mask = var_19375_end_mask_0, x = k_179_cast_fp16)[name = tensor("op_19375_cast_fp16")]; + tensor var_19379_begin_0 = const()[name = tensor("op_19379_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_19379_end_0 = const()[name = tensor("op_19379_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_19379_end_mask_0 = const()[name = tensor("op_19379_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19379_cast_fp16 = slice_by_index(begin = var_19379_begin_0, end = var_19379_end_0, end_mask = var_19379_end_mask_0, x = k_179_cast_fp16)[name = tensor("op_19379_cast_fp16")]; + tensor var_19383_begin_0 = const()[name = tensor("op_19383_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_19383_end_0 = const()[name = tensor("op_19383_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_19383_end_mask_0 = const()[name = tensor("op_19383_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19383_cast_fp16 = slice_by_index(begin = var_19383_begin_0, end = var_19383_end_0, end_mask = var_19383_end_mask_0, x = k_179_cast_fp16)[name = tensor("op_19383_cast_fp16")]; + tensor var_19387_begin_0 = const()[name = tensor("op_19387_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_19387_end_0 = const()[name = tensor("op_19387_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_19387_end_mask_0 = const()[name = tensor("op_19387_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19387_cast_fp16 = slice_by_index(begin = var_19387_begin_0, end = var_19387_end_0, end_mask = var_19387_end_mask_0, x = k_179_cast_fp16)[name = tensor("op_19387_cast_fp16")]; + tensor var_19391_begin_0 = const()[name = tensor("op_19391_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_19391_end_0 = const()[name = tensor("op_19391_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_19391_end_mask_0 = const()[name = tensor("op_19391_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19391_cast_fp16 = slice_by_index(begin = var_19391_begin_0, end = var_19391_end_0, end_mask = var_19391_end_mask_0, x = k_179_cast_fp16)[name = tensor("op_19391_cast_fp16")]; + tensor var_19395_begin_0 = const()[name = tensor("op_19395_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_19395_end_0 = const()[name = tensor("op_19395_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_19395_end_mask_0 = const()[name = tensor("op_19395_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19395_cast_fp16 = slice_by_index(begin = var_19395_begin_0, end = var_19395_end_0, end_mask = var_19395_end_mask_0, x = k_179_cast_fp16)[name = tensor("op_19395_cast_fp16")]; + tensor var_19399_begin_0 = const()[name = tensor("op_19399_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_19399_end_0 = const()[name = tensor("op_19399_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_19399_end_mask_0 = const()[name = tensor("op_19399_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19399_cast_fp16 = slice_by_index(begin = var_19399_begin_0, end = var_19399_end_0, end_mask = var_19399_end_mask_0, x = k_179_cast_fp16)[name = tensor("op_19399_cast_fp16")]; + tensor var_19403_begin_0 = const()[name = tensor("op_19403_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_19403_end_0 = const()[name = tensor("op_19403_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_19403_end_mask_0 = const()[name = tensor("op_19403_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19403_cast_fp16 = slice_by_index(begin = var_19403_begin_0, end = var_19403_end_0, end_mask = var_19403_end_mask_0, x = k_179_cast_fp16)[name = tensor("op_19403_cast_fp16")]; + tensor var_19407_begin_0 = const()[name = tensor("op_19407_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_19407_end_0 = const()[name = tensor("op_19407_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_19407_end_mask_0 = const()[name = tensor("op_19407_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19407_cast_fp16 = slice_by_index(begin = var_19407_begin_0, end = var_19407_end_0, end_mask = var_19407_end_mask_0, x = k_179_cast_fp16)[name = tensor("op_19407_cast_fp16")]; + tensor var_19411_begin_0 = const()[name = tensor("op_19411_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_19411_end_0 = const()[name = tensor("op_19411_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_19411_end_mask_0 = const()[name = tensor("op_19411_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19411_cast_fp16 = slice_by_index(begin = var_19411_begin_0, end = var_19411_end_0, end_mask = var_19411_end_mask_0, x = k_179_cast_fp16)[name = tensor("op_19411_cast_fp16")]; + tensor var_19415_begin_0 = const()[name = tensor("op_19415_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_19415_end_0 = const()[name = tensor("op_19415_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_19415_end_mask_0 = const()[name = tensor("op_19415_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19415_cast_fp16 = slice_by_index(begin = var_19415_begin_0, end = var_19415_end_0, end_mask = var_19415_end_mask_0, x = k_179_cast_fp16)[name = tensor("op_19415_cast_fp16")]; + tensor var_19417_begin_0 = const()[name = tensor("op_19417_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_19417_end_0 = const()[name = tensor("op_19417_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_19417_end_mask_0 = const()[name = tensor("op_19417_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19417_cast_fp16 = slice_by_index(begin = var_19417_begin_0, end = var_19417_end_0, end_mask = var_19417_end_mask_0, x = v_89_cast_fp16)[name = tensor("op_19417_cast_fp16")]; + tensor var_19421_begin_0 = const()[name = tensor("op_19421_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_19421_end_0 = const()[name = tensor("op_19421_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_19421_end_mask_0 = const()[name = tensor("op_19421_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19421_cast_fp16 = slice_by_index(begin = var_19421_begin_0, end = var_19421_end_0, end_mask = var_19421_end_mask_0, x = v_89_cast_fp16)[name = tensor("op_19421_cast_fp16")]; + tensor var_19425_begin_0 = const()[name = tensor("op_19425_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_19425_end_0 = const()[name = tensor("op_19425_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_19425_end_mask_0 = const()[name = tensor("op_19425_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19425_cast_fp16 = slice_by_index(begin = var_19425_begin_0, end = var_19425_end_0, end_mask = var_19425_end_mask_0, x = v_89_cast_fp16)[name = tensor("op_19425_cast_fp16")]; + tensor var_19429_begin_0 = const()[name = tensor("op_19429_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_19429_end_0 = const()[name = tensor("op_19429_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_19429_end_mask_0 = const()[name = tensor("op_19429_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19429_cast_fp16 = slice_by_index(begin = var_19429_begin_0, end = var_19429_end_0, end_mask = var_19429_end_mask_0, x = v_89_cast_fp16)[name = tensor("op_19429_cast_fp16")]; + tensor var_19433_begin_0 = const()[name = tensor("op_19433_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_19433_end_0 = const()[name = tensor("op_19433_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_19433_end_mask_0 = const()[name = tensor("op_19433_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19433_cast_fp16 = slice_by_index(begin = var_19433_begin_0, end = var_19433_end_0, end_mask = var_19433_end_mask_0, x = v_89_cast_fp16)[name = tensor("op_19433_cast_fp16")]; + tensor var_19437_begin_0 = const()[name = tensor("op_19437_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_19437_end_0 = const()[name = tensor("op_19437_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_19437_end_mask_0 = const()[name = tensor("op_19437_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19437_cast_fp16 = slice_by_index(begin = var_19437_begin_0, end = var_19437_end_0, end_mask = var_19437_end_mask_0, x = v_89_cast_fp16)[name = tensor("op_19437_cast_fp16")]; + tensor var_19441_begin_0 = const()[name = tensor("op_19441_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_19441_end_0 = const()[name = tensor("op_19441_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_19441_end_mask_0 = const()[name = tensor("op_19441_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19441_cast_fp16 = slice_by_index(begin = var_19441_begin_0, end = var_19441_end_0, end_mask = var_19441_end_mask_0, x = v_89_cast_fp16)[name = tensor("op_19441_cast_fp16")]; + tensor var_19445_begin_0 = const()[name = tensor("op_19445_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_19445_end_0 = const()[name = tensor("op_19445_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_19445_end_mask_0 = const()[name = tensor("op_19445_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19445_cast_fp16 = slice_by_index(begin = var_19445_begin_0, end = var_19445_end_0, end_mask = var_19445_end_mask_0, x = v_89_cast_fp16)[name = tensor("op_19445_cast_fp16")]; + tensor var_19449_begin_0 = const()[name = tensor("op_19449_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_19449_end_0 = const()[name = tensor("op_19449_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_19449_end_mask_0 = const()[name = tensor("op_19449_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19449_cast_fp16 = slice_by_index(begin = var_19449_begin_0, end = var_19449_end_0, end_mask = var_19449_end_mask_0, x = v_89_cast_fp16)[name = tensor("op_19449_cast_fp16")]; + tensor var_19453_begin_0 = const()[name = tensor("op_19453_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_19453_end_0 = const()[name = tensor("op_19453_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_19453_end_mask_0 = const()[name = tensor("op_19453_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19453_cast_fp16 = slice_by_index(begin = var_19453_begin_0, end = var_19453_end_0, end_mask = var_19453_end_mask_0, x = v_89_cast_fp16)[name = tensor("op_19453_cast_fp16")]; + tensor var_19457_begin_0 = const()[name = tensor("op_19457_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_19457_end_0 = const()[name = tensor("op_19457_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_19457_end_mask_0 = const()[name = tensor("op_19457_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19457_cast_fp16 = slice_by_index(begin = var_19457_begin_0, end = var_19457_end_0, end_mask = var_19457_end_mask_0, x = v_89_cast_fp16)[name = tensor("op_19457_cast_fp16")]; + tensor var_19461_begin_0 = const()[name = tensor("op_19461_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_19461_end_0 = const()[name = tensor("op_19461_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_19461_end_mask_0 = const()[name = tensor("op_19461_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19461_cast_fp16 = slice_by_index(begin = var_19461_begin_0, end = var_19461_end_0, end_mask = var_19461_end_mask_0, x = v_89_cast_fp16)[name = tensor("op_19461_cast_fp16")]; + tensor var_19465_begin_0 = const()[name = tensor("op_19465_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_19465_end_0 = const()[name = tensor("op_19465_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_19465_end_mask_0 = const()[name = tensor("op_19465_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19465_cast_fp16 = slice_by_index(begin = var_19465_begin_0, end = var_19465_end_0, end_mask = var_19465_end_mask_0, x = v_89_cast_fp16)[name = tensor("op_19465_cast_fp16")]; + tensor var_19469_begin_0 = const()[name = tensor("op_19469_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_19469_end_0 = const()[name = tensor("op_19469_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_19469_end_mask_0 = const()[name = tensor("op_19469_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19469_cast_fp16 = slice_by_index(begin = var_19469_begin_0, end = var_19469_end_0, end_mask = var_19469_end_mask_0, x = v_89_cast_fp16)[name = tensor("op_19469_cast_fp16")]; + tensor var_19473_begin_0 = const()[name = tensor("op_19473_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_19473_end_0 = const()[name = tensor("op_19473_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_19473_end_mask_0 = const()[name = tensor("op_19473_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19473_cast_fp16 = slice_by_index(begin = var_19473_begin_0, end = var_19473_end_0, end_mask = var_19473_end_mask_0, x = v_89_cast_fp16)[name = tensor("op_19473_cast_fp16")]; + tensor var_19477_begin_0 = const()[name = tensor("op_19477_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_19477_end_0 = const()[name = tensor("op_19477_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_19477_end_mask_0 = const()[name = tensor("op_19477_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19477_cast_fp16 = slice_by_index(begin = var_19477_begin_0, end = var_19477_end_0, end_mask = var_19477_end_mask_0, x = v_89_cast_fp16)[name = tensor("op_19477_cast_fp16")]; + tensor var_19481_begin_0 = const()[name = tensor("op_19481_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_19481_end_0 = const()[name = tensor("op_19481_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_19481_end_mask_0 = const()[name = tensor("op_19481_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19481_cast_fp16 = slice_by_index(begin = var_19481_begin_0, end = var_19481_end_0, end_mask = var_19481_end_mask_0, x = v_89_cast_fp16)[name = tensor("op_19481_cast_fp16")]; + tensor var_19485_begin_0 = const()[name = tensor("op_19485_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_19485_end_0 = const()[name = tensor("op_19485_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_19485_end_mask_0 = const()[name = tensor("op_19485_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19485_cast_fp16 = slice_by_index(begin = var_19485_begin_0, end = var_19485_end_0, end_mask = var_19485_end_mask_0, x = v_89_cast_fp16)[name = tensor("op_19485_cast_fp16")]; + tensor var_19489_begin_0 = const()[name = tensor("op_19489_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_19489_end_0 = const()[name = tensor("op_19489_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_19489_end_mask_0 = const()[name = tensor("op_19489_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19489_cast_fp16 = slice_by_index(begin = var_19489_begin_0, end = var_19489_end_0, end_mask = var_19489_end_mask_0, x = v_89_cast_fp16)[name = tensor("op_19489_cast_fp16")]; + tensor var_19493_begin_0 = const()[name = tensor("op_19493_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_19493_end_0 = const()[name = tensor("op_19493_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_19493_end_mask_0 = const()[name = tensor("op_19493_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19493_cast_fp16 = slice_by_index(begin = var_19493_begin_0, end = var_19493_end_0, end_mask = var_19493_end_mask_0, x = v_89_cast_fp16)[name = tensor("op_19493_cast_fp16")]; + tensor var_19497_equation_0 = const()[name = tensor("op_19497_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19497_cast_fp16 = einsum(equation = var_19497_equation_0, values = (var_19339_cast_fp16, var_19256_cast_fp16))[name = tensor("op_19497_cast_fp16")]; + tensor var_19498_to_fp16 = const()[name = tensor("op_19498_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1601_cast_fp16 = mul(x = var_19497_cast_fp16, y = var_19498_to_fp16)[name = tensor("aw_1601_cast_fp16")]; + tensor var_19501_equation_0 = const()[name = tensor("op_19501_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19501_cast_fp16 = einsum(equation = var_19501_equation_0, values = (var_19343_cast_fp16, var_19260_cast_fp16))[name = tensor("op_19501_cast_fp16")]; + tensor var_19502_to_fp16 = const()[name = tensor("op_19502_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1603_cast_fp16 = mul(x = var_19501_cast_fp16, y = var_19502_to_fp16)[name = tensor("aw_1603_cast_fp16")]; + tensor var_19505_equation_0 = const()[name = tensor("op_19505_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19505_cast_fp16 = einsum(equation = var_19505_equation_0, values = (var_19347_cast_fp16, var_19264_cast_fp16))[name = tensor("op_19505_cast_fp16")]; + tensor var_19506_to_fp16 = const()[name = tensor("op_19506_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1605_cast_fp16 = mul(x = var_19505_cast_fp16, y = var_19506_to_fp16)[name = tensor("aw_1605_cast_fp16")]; + tensor var_19509_equation_0 = const()[name = tensor("op_19509_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19509_cast_fp16 = einsum(equation = var_19509_equation_0, values = (var_19351_cast_fp16, var_19268_cast_fp16))[name = tensor("op_19509_cast_fp16")]; + tensor var_19510_to_fp16 = const()[name = tensor("op_19510_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1607_cast_fp16 = mul(x = var_19509_cast_fp16, y = var_19510_to_fp16)[name = tensor("aw_1607_cast_fp16")]; + tensor var_19513_equation_0 = const()[name = tensor("op_19513_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19513_cast_fp16 = einsum(equation = var_19513_equation_0, values = (var_19355_cast_fp16, var_19272_cast_fp16))[name = tensor("op_19513_cast_fp16")]; + tensor var_19514_to_fp16 = const()[name = tensor("op_19514_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1609_cast_fp16 = mul(x = var_19513_cast_fp16, y = var_19514_to_fp16)[name = tensor("aw_1609_cast_fp16")]; + tensor var_19517_equation_0 = const()[name = tensor("op_19517_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19517_cast_fp16 = einsum(equation = var_19517_equation_0, values = (var_19359_cast_fp16, var_19276_cast_fp16))[name = tensor("op_19517_cast_fp16")]; + tensor var_19518_to_fp16 = const()[name = tensor("op_19518_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1611_cast_fp16 = mul(x = var_19517_cast_fp16, y = var_19518_to_fp16)[name = tensor("aw_1611_cast_fp16")]; + tensor var_19521_equation_0 = const()[name = tensor("op_19521_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19521_cast_fp16 = einsum(equation = var_19521_equation_0, values = (var_19363_cast_fp16, var_19280_cast_fp16))[name = tensor("op_19521_cast_fp16")]; + tensor var_19522_to_fp16 = const()[name = tensor("op_19522_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1613_cast_fp16 = mul(x = var_19521_cast_fp16, y = var_19522_to_fp16)[name = tensor("aw_1613_cast_fp16")]; + tensor var_19525_equation_0 = const()[name = tensor("op_19525_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19525_cast_fp16 = einsum(equation = var_19525_equation_0, values = (var_19367_cast_fp16, var_19284_cast_fp16))[name = tensor("op_19525_cast_fp16")]; + tensor var_19526_to_fp16 = const()[name = tensor("op_19526_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1615_cast_fp16 = mul(x = var_19525_cast_fp16, y = var_19526_to_fp16)[name = tensor("aw_1615_cast_fp16")]; + tensor var_19529_equation_0 = const()[name = tensor("op_19529_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19529_cast_fp16 = einsum(equation = var_19529_equation_0, values = (var_19371_cast_fp16, var_19288_cast_fp16))[name = tensor("op_19529_cast_fp16")]; + tensor var_19530_to_fp16 = const()[name = tensor("op_19530_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1617_cast_fp16 = mul(x = var_19529_cast_fp16, y = var_19530_to_fp16)[name = tensor("aw_1617_cast_fp16")]; + tensor var_19533_equation_0 = const()[name = tensor("op_19533_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19533_cast_fp16 = einsum(equation = var_19533_equation_0, values = (var_19375_cast_fp16, var_19292_cast_fp16))[name = tensor("op_19533_cast_fp16")]; + tensor var_19534_to_fp16 = const()[name = tensor("op_19534_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1619_cast_fp16 = mul(x = var_19533_cast_fp16, y = var_19534_to_fp16)[name = tensor("aw_1619_cast_fp16")]; + tensor var_19537_equation_0 = const()[name = tensor("op_19537_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19537_cast_fp16 = einsum(equation = var_19537_equation_0, values = (var_19379_cast_fp16, var_19296_cast_fp16))[name = tensor("op_19537_cast_fp16")]; + tensor var_19538_to_fp16 = const()[name = tensor("op_19538_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1621_cast_fp16 = mul(x = var_19537_cast_fp16, y = var_19538_to_fp16)[name = tensor("aw_1621_cast_fp16")]; + tensor var_19541_equation_0 = const()[name = tensor("op_19541_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19541_cast_fp16 = einsum(equation = var_19541_equation_0, values = (var_19383_cast_fp16, var_19300_cast_fp16))[name = tensor("op_19541_cast_fp16")]; + tensor var_19542_to_fp16 = const()[name = tensor("op_19542_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1623_cast_fp16 = mul(x = var_19541_cast_fp16, y = var_19542_to_fp16)[name = tensor("aw_1623_cast_fp16")]; + tensor var_19545_equation_0 = const()[name = tensor("op_19545_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19545_cast_fp16 = einsum(equation = var_19545_equation_0, values = (var_19387_cast_fp16, var_19304_cast_fp16))[name = tensor("op_19545_cast_fp16")]; + tensor var_19546_to_fp16 = const()[name = tensor("op_19546_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1625_cast_fp16 = mul(x = var_19545_cast_fp16, y = var_19546_to_fp16)[name = tensor("aw_1625_cast_fp16")]; + tensor var_19549_equation_0 = const()[name = tensor("op_19549_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19549_cast_fp16 = einsum(equation = var_19549_equation_0, values = (var_19391_cast_fp16, var_19308_cast_fp16))[name = tensor("op_19549_cast_fp16")]; + tensor var_19550_to_fp16 = const()[name = tensor("op_19550_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1627_cast_fp16 = mul(x = var_19549_cast_fp16, y = var_19550_to_fp16)[name = tensor("aw_1627_cast_fp16")]; + tensor var_19553_equation_0 = const()[name = tensor("op_19553_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19553_cast_fp16 = einsum(equation = var_19553_equation_0, values = (var_19395_cast_fp16, var_19312_cast_fp16))[name = tensor("op_19553_cast_fp16")]; + tensor var_19554_to_fp16 = const()[name = tensor("op_19554_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1629_cast_fp16 = mul(x = var_19553_cast_fp16, y = var_19554_to_fp16)[name = tensor("aw_1629_cast_fp16")]; + tensor var_19557_equation_0 = const()[name = tensor("op_19557_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19557_cast_fp16 = einsum(equation = var_19557_equation_0, values = (var_19399_cast_fp16, var_19316_cast_fp16))[name = tensor("op_19557_cast_fp16")]; + tensor var_19558_to_fp16 = const()[name = tensor("op_19558_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1631_cast_fp16 = mul(x = var_19557_cast_fp16, y = var_19558_to_fp16)[name = tensor("aw_1631_cast_fp16")]; + tensor var_19561_equation_0 = const()[name = tensor("op_19561_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19561_cast_fp16 = einsum(equation = var_19561_equation_0, values = (var_19403_cast_fp16, var_19320_cast_fp16))[name = tensor("op_19561_cast_fp16")]; + tensor var_19562_to_fp16 = const()[name = tensor("op_19562_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1633_cast_fp16 = mul(x = var_19561_cast_fp16, y = var_19562_to_fp16)[name = tensor("aw_1633_cast_fp16")]; + tensor var_19565_equation_0 = const()[name = tensor("op_19565_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19565_cast_fp16 = einsum(equation = var_19565_equation_0, values = (var_19407_cast_fp16, var_19324_cast_fp16))[name = tensor("op_19565_cast_fp16")]; + tensor var_19566_to_fp16 = const()[name = tensor("op_19566_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1635_cast_fp16 = mul(x = var_19565_cast_fp16, y = var_19566_to_fp16)[name = tensor("aw_1635_cast_fp16")]; + tensor var_19569_equation_0 = const()[name = tensor("op_19569_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19569_cast_fp16 = einsum(equation = var_19569_equation_0, values = (var_19411_cast_fp16, var_19328_cast_fp16))[name = tensor("op_19569_cast_fp16")]; + tensor var_19570_to_fp16 = const()[name = tensor("op_19570_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1637_cast_fp16 = mul(x = var_19569_cast_fp16, y = var_19570_to_fp16)[name = tensor("aw_1637_cast_fp16")]; + tensor var_19573_equation_0 = const()[name = tensor("op_19573_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19573_cast_fp16 = einsum(equation = var_19573_equation_0, values = (var_19415_cast_fp16, var_19332_cast_fp16))[name = tensor("op_19573_cast_fp16")]; + tensor var_19574_to_fp16 = const()[name = tensor("op_19574_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1639_cast_fp16 = mul(x = var_19573_cast_fp16, y = var_19574_to_fp16)[name = tensor("aw_1639_cast_fp16")]; + tensor var_19576_cast_fp16 = softmax(axis = var_2624, x = aw_1601_cast_fp16)[name = tensor("op_19576_cast_fp16")]; + tensor var_19577_cast_fp16 = softmax(axis = var_2624, x = aw_1603_cast_fp16)[name = tensor("op_19577_cast_fp16")]; + tensor var_19578_cast_fp16 = softmax(axis = var_2624, x = aw_1605_cast_fp16)[name = tensor("op_19578_cast_fp16")]; + tensor var_19579_cast_fp16 = softmax(axis = var_2624, x = aw_1607_cast_fp16)[name = tensor("op_19579_cast_fp16")]; + tensor var_19580_cast_fp16 = softmax(axis = var_2624, x = aw_1609_cast_fp16)[name = tensor("op_19580_cast_fp16")]; + tensor var_19581_cast_fp16 = softmax(axis = var_2624, x = aw_1611_cast_fp16)[name = tensor("op_19581_cast_fp16")]; + tensor var_19582_cast_fp16 = softmax(axis = var_2624, x = aw_1613_cast_fp16)[name = tensor("op_19582_cast_fp16")]; + tensor var_19583_cast_fp16 = softmax(axis = var_2624, x = aw_1615_cast_fp16)[name = tensor("op_19583_cast_fp16")]; + tensor var_19584_cast_fp16 = softmax(axis = var_2624, x = aw_1617_cast_fp16)[name = tensor("op_19584_cast_fp16")]; + tensor var_19585_cast_fp16 = softmax(axis = var_2624, x = aw_1619_cast_fp16)[name = tensor("op_19585_cast_fp16")]; + tensor var_19586_cast_fp16 = softmax(axis = var_2624, x = aw_1621_cast_fp16)[name = tensor("op_19586_cast_fp16")]; + tensor var_19587_cast_fp16 = softmax(axis = var_2624, x = aw_1623_cast_fp16)[name = tensor("op_19587_cast_fp16")]; + tensor var_19588_cast_fp16 = softmax(axis = var_2624, x = aw_1625_cast_fp16)[name = tensor("op_19588_cast_fp16")]; + tensor var_19589_cast_fp16 = softmax(axis = var_2624, x = aw_1627_cast_fp16)[name = tensor("op_19589_cast_fp16")]; + tensor var_19590_cast_fp16 = softmax(axis = var_2624, x = aw_1629_cast_fp16)[name = tensor("op_19590_cast_fp16")]; + tensor var_19591_cast_fp16 = softmax(axis = var_2624, x = aw_1631_cast_fp16)[name = tensor("op_19591_cast_fp16")]; + tensor var_19592_cast_fp16 = softmax(axis = var_2624, x = aw_1633_cast_fp16)[name = tensor("op_19592_cast_fp16")]; + tensor var_19593_cast_fp16 = softmax(axis = var_2624, x = aw_1635_cast_fp16)[name = tensor("op_19593_cast_fp16")]; + tensor var_19594_cast_fp16 = softmax(axis = var_2624, x = aw_1637_cast_fp16)[name = tensor("op_19594_cast_fp16")]; + tensor var_19595_cast_fp16 = softmax(axis = var_2624, x = aw_1639_cast_fp16)[name = tensor("op_19595_cast_fp16")]; + tensor var_19597_equation_0 = const()[name = tensor("op_19597_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19597_cast_fp16 = einsum(equation = var_19597_equation_0, values = (var_19417_cast_fp16, var_19576_cast_fp16))[name = tensor("op_19597_cast_fp16")]; + tensor var_19599_equation_0 = const()[name = tensor("op_19599_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19599_cast_fp16 = einsum(equation = var_19599_equation_0, values = (var_19421_cast_fp16, var_19577_cast_fp16))[name = tensor("op_19599_cast_fp16")]; + tensor var_19601_equation_0 = const()[name = tensor("op_19601_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19601_cast_fp16 = einsum(equation = var_19601_equation_0, values = (var_19425_cast_fp16, var_19578_cast_fp16))[name = tensor("op_19601_cast_fp16")]; + tensor var_19603_equation_0 = const()[name = tensor("op_19603_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19603_cast_fp16 = einsum(equation = var_19603_equation_0, values = (var_19429_cast_fp16, var_19579_cast_fp16))[name = tensor("op_19603_cast_fp16")]; + tensor var_19605_equation_0 = const()[name = tensor("op_19605_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19605_cast_fp16 = einsum(equation = var_19605_equation_0, values = (var_19433_cast_fp16, var_19580_cast_fp16))[name = tensor("op_19605_cast_fp16")]; + tensor var_19607_equation_0 = const()[name = tensor("op_19607_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19607_cast_fp16 = einsum(equation = var_19607_equation_0, values = (var_19437_cast_fp16, var_19581_cast_fp16))[name = tensor("op_19607_cast_fp16")]; + tensor var_19609_equation_0 = const()[name = tensor("op_19609_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19609_cast_fp16 = einsum(equation = var_19609_equation_0, values = (var_19441_cast_fp16, var_19582_cast_fp16))[name = tensor("op_19609_cast_fp16")]; + tensor var_19611_equation_0 = const()[name = tensor("op_19611_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19611_cast_fp16 = einsum(equation = var_19611_equation_0, values = (var_19445_cast_fp16, var_19583_cast_fp16))[name = tensor("op_19611_cast_fp16")]; + tensor var_19613_equation_0 = const()[name = tensor("op_19613_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19613_cast_fp16 = einsum(equation = var_19613_equation_0, values = (var_19449_cast_fp16, var_19584_cast_fp16))[name = tensor("op_19613_cast_fp16")]; + tensor var_19615_equation_0 = const()[name = tensor("op_19615_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19615_cast_fp16 = einsum(equation = var_19615_equation_0, values = (var_19453_cast_fp16, var_19585_cast_fp16))[name = tensor("op_19615_cast_fp16")]; + tensor var_19617_equation_0 = const()[name = tensor("op_19617_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19617_cast_fp16 = einsum(equation = var_19617_equation_0, values = (var_19457_cast_fp16, var_19586_cast_fp16))[name = tensor("op_19617_cast_fp16")]; + tensor var_19619_equation_0 = const()[name = tensor("op_19619_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19619_cast_fp16 = einsum(equation = var_19619_equation_0, values = (var_19461_cast_fp16, var_19587_cast_fp16))[name = tensor("op_19619_cast_fp16")]; + tensor var_19621_equation_0 = const()[name = tensor("op_19621_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19621_cast_fp16 = einsum(equation = var_19621_equation_0, values = (var_19465_cast_fp16, var_19588_cast_fp16))[name = tensor("op_19621_cast_fp16")]; + tensor var_19623_equation_0 = const()[name = tensor("op_19623_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19623_cast_fp16 = einsum(equation = var_19623_equation_0, values = (var_19469_cast_fp16, var_19589_cast_fp16))[name = tensor("op_19623_cast_fp16")]; + tensor var_19625_equation_0 = const()[name = tensor("op_19625_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19625_cast_fp16 = einsum(equation = var_19625_equation_0, values = (var_19473_cast_fp16, var_19590_cast_fp16))[name = tensor("op_19625_cast_fp16")]; + tensor var_19627_equation_0 = const()[name = tensor("op_19627_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19627_cast_fp16 = einsum(equation = var_19627_equation_0, values = (var_19477_cast_fp16, var_19591_cast_fp16))[name = tensor("op_19627_cast_fp16")]; + tensor var_19629_equation_0 = const()[name = tensor("op_19629_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19629_cast_fp16 = einsum(equation = var_19629_equation_0, values = (var_19481_cast_fp16, var_19592_cast_fp16))[name = tensor("op_19629_cast_fp16")]; + tensor var_19631_equation_0 = const()[name = tensor("op_19631_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19631_cast_fp16 = einsum(equation = var_19631_equation_0, values = (var_19485_cast_fp16, var_19593_cast_fp16))[name = tensor("op_19631_cast_fp16")]; + tensor var_19633_equation_0 = const()[name = tensor("op_19633_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19633_cast_fp16 = einsum(equation = var_19633_equation_0, values = (var_19489_cast_fp16, var_19594_cast_fp16))[name = tensor("op_19633_cast_fp16")]; + tensor var_19635_equation_0 = const()[name = tensor("op_19635_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19635_cast_fp16 = einsum(equation = var_19635_equation_0, values = (var_19493_cast_fp16, var_19595_cast_fp16))[name = tensor("op_19635_cast_fp16")]; + tensor input_293_interleave_0 = const()[name = tensor("input_293_interleave_0"), val = tensor(false)]; + tensor input_293_cast_fp16 = concat(axis = var_2624, interleave = input_293_interleave_0, values = (var_19597_cast_fp16, var_19599_cast_fp16, var_19601_cast_fp16, var_19603_cast_fp16, var_19605_cast_fp16, var_19607_cast_fp16, var_19609_cast_fp16, var_19611_cast_fp16, var_19613_cast_fp16, var_19615_cast_fp16, var_19617_cast_fp16, var_19619_cast_fp16, var_19621_cast_fp16, var_19623_cast_fp16, var_19625_cast_fp16, var_19627_cast_fp16, var_19629_cast_fp16, var_19631_cast_fp16, var_19633_cast_fp16, var_19635_cast_fp16))[name = tensor("input_293_cast_fp16")]; + tensor var_19641 = const()[name = tensor("op_19641"), val = tensor([1, 1])]; + tensor var_19643 = const()[name = tensor("op_19643"), val = tensor([1, 1])]; + tensor var_19645_pad_type_0 = const()[name = tensor("op_19645_pad_type_0"), val = tensor("custom")]; + tensor var_19645_pad_0 = const()[name = tensor("op_19645_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_8_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_8_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1528076992)))]; + tensor down_blocks_2_attentions_1_transformer_blocks_8_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_8_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1531353856)))]; + tensor var_19645_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_8_attn1_to_out_0_bias_to_fp16, dilations = var_19643, groups = var_2624, pad = var_19645_pad_0, pad_type = var_19645_pad_type_0, strides = var_19641, weight = down_blocks_2_attentions_1_transformer_blocks_8_attn1_to_out_0_weight_to_fp16, x = input_293_cast_fp16)[name = tensor("op_19645_cast_fp16")]; + tensor inputs_135_cast_fp16 = add(x = var_19645_cast_fp16, y = inputs_133_cast_fp16)[name = tensor("inputs_135_cast_fp16")]; + tensor hidden_states_187_axes_0 = const()[name = tensor("hidden_states_187_axes_0"), val = tensor([1])]; + tensor hidden_states_187_gamma_0_to_fp16 = const()[name = tensor("hidden_states_187_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1531356480)))]; + tensor hidden_states_187_beta_0_to_fp16 = const()[name = tensor("hidden_states_187_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1531359104)))]; + tensor var_19655_to_fp16 = const()[name = tensor("op_19655_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_187_cast_fp16 = layer_norm(axes = hidden_states_187_axes_0, beta = hidden_states_187_beta_0_to_fp16, epsilon = var_19655_to_fp16, gamma = hidden_states_187_gamma_0_to_fp16, x = inputs_135_cast_fp16)[name = tensor("hidden_states_187_cast_fp16")]; + tensor var_19670 = const()[name = tensor("op_19670"), val = tensor([1, 1])]; + tensor var_19672 = const()[name = tensor("op_19672"), val = tensor([1, 1])]; + tensor q_91_pad_type_0 = const()[name = tensor("q_91_pad_type_0"), val = tensor("custom")]; + tensor q_91_pad_0 = const()[name = tensor("q_91_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_8_attn2_to_q_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_8_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1531361728)))]; + tensor q_91_cast_fp16 = conv(dilations = var_19672, groups = var_2624, pad = q_91_pad_0, pad_type = q_91_pad_type_0, strides = var_19670, weight = down_blocks_2_attentions_1_transformer_blocks_8_attn2_to_q_weight_to_fp16, x = hidden_states_187_cast_fp16)[name = tensor("q_91_cast_fp16")]; + tensor var_19676 = const()[name = tensor("op_19676"), val = tensor([1, 1])]; + tensor var_19678 = const()[name = tensor("op_19678"), val = tensor([1, 1])]; + tensor k_181_pad_type_0 = const()[name = tensor("k_181_pad_type_0"), val = tensor("custom")]; + tensor k_181_pad_0 = const()[name = tensor("k_181_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_8_attn2_to_k_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_8_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1534638592)))]; + tensor k_181_cast_fp16 = conv(dilations = var_19678, groups = var_2624, pad = k_181_pad_0, pad_type = k_181_pad_type_0, strides = var_19676, weight = down_blocks_2_attentions_1_transformer_blocks_8_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_181_cast_fp16")]; + tensor var_19682 = const()[name = tensor("op_19682"), val = tensor([1, 1])]; + tensor var_19684 = const()[name = tensor("op_19684"), val = tensor([1, 1])]; + tensor v_91_pad_type_0 = const()[name = tensor("v_91_pad_type_0"), val = tensor("custom")]; + tensor v_91_pad_0 = const()[name = tensor("v_91_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_8_attn2_to_v_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_8_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1539881536)))]; + tensor v_91_cast_fp16 = conv(dilations = var_19684, groups = var_2624, pad = v_91_pad_0, pad_type = v_91_pad_type_0, strides = var_19682, weight = down_blocks_2_attentions_1_transformer_blocks_8_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_91_cast_fp16")]; + tensor var_19688_begin_0 = const()[name = tensor("op_19688_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_19688_end_0 = const()[name = tensor("op_19688_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_19688_end_mask_0 = const()[name = tensor("op_19688_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19688_cast_fp16 = slice_by_index(begin = var_19688_begin_0, end = var_19688_end_0, end_mask = var_19688_end_mask_0, x = q_91_cast_fp16)[name = tensor("op_19688_cast_fp16")]; + tensor var_19692_begin_0 = const()[name = tensor("op_19692_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_19692_end_0 = const()[name = tensor("op_19692_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_19692_end_mask_0 = const()[name = tensor("op_19692_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19692_cast_fp16 = slice_by_index(begin = var_19692_begin_0, end = var_19692_end_0, end_mask = var_19692_end_mask_0, x = q_91_cast_fp16)[name = tensor("op_19692_cast_fp16")]; + tensor var_19696_begin_0 = const()[name = tensor("op_19696_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_19696_end_0 = const()[name = tensor("op_19696_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_19696_end_mask_0 = const()[name = tensor("op_19696_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19696_cast_fp16 = slice_by_index(begin = var_19696_begin_0, end = var_19696_end_0, end_mask = var_19696_end_mask_0, x = q_91_cast_fp16)[name = tensor("op_19696_cast_fp16")]; + tensor var_19700_begin_0 = const()[name = tensor("op_19700_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_19700_end_0 = const()[name = tensor("op_19700_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_19700_end_mask_0 = const()[name = tensor("op_19700_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19700_cast_fp16 = slice_by_index(begin = var_19700_begin_0, end = var_19700_end_0, end_mask = var_19700_end_mask_0, x = q_91_cast_fp16)[name = tensor("op_19700_cast_fp16")]; + tensor var_19704_begin_0 = const()[name = tensor("op_19704_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_19704_end_0 = const()[name = tensor("op_19704_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_19704_end_mask_0 = const()[name = tensor("op_19704_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19704_cast_fp16 = slice_by_index(begin = var_19704_begin_0, end = var_19704_end_0, end_mask = var_19704_end_mask_0, x = q_91_cast_fp16)[name = tensor("op_19704_cast_fp16")]; + tensor var_19708_begin_0 = const()[name = tensor("op_19708_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_19708_end_0 = const()[name = tensor("op_19708_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_19708_end_mask_0 = const()[name = tensor("op_19708_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19708_cast_fp16 = slice_by_index(begin = var_19708_begin_0, end = var_19708_end_0, end_mask = var_19708_end_mask_0, x = q_91_cast_fp16)[name = tensor("op_19708_cast_fp16")]; + tensor var_19712_begin_0 = const()[name = tensor("op_19712_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_19712_end_0 = const()[name = tensor("op_19712_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_19712_end_mask_0 = const()[name = tensor("op_19712_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19712_cast_fp16 = slice_by_index(begin = var_19712_begin_0, end = var_19712_end_0, end_mask = var_19712_end_mask_0, x = q_91_cast_fp16)[name = tensor("op_19712_cast_fp16")]; + tensor var_19716_begin_0 = const()[name = tensor("op_19716_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_19716_end_0 = const()[name = tensor("op_19716_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_19716_end_mask_0 = const()[name = tensor("op_19716_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19716_cast_fp16 = slice_by_index(begin = var_19716_begin_0, end = var_19716_end_0, end_mask = var_19716_end_mask_0, x = q_91_cast_fp16)[name = tensor("op_19716_cast_fp16")]; + tensor var_19720_begin_0 = const()[name = tensor("op_19720_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_19720_end_0 = const()[name = tensor("op_19720_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_19720_end_mask_0 = const()[name = tensor("op_19720_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19720_cast_fp16 = slice_by_index(begin = var_19720_begin_0, end = var_19720_end_0, end_mask = var_19720_end_mask_0, x = q_91_cast_fp16)[name = tensor("op_19720_cast_fp16")]; + tensor var_19724_begin_0 = const()[name = tensor("op_19724_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_19724_end_0 = const()[name = tensor("op_19724_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_19724_end_mask_0 = const()[name = tensor("op_19724_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19724_cast_fp16 = slice_by_index(begin = var_19724_begin_0, end = var_19724_end_0, end_mask = var_19724_end_mask_0, x = q_91_cast_fp16)[name = tensor("op_19724_cast_fp16")]; + tensor var_19728_begin_0 = const()[name = tensor("op_19728_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_19728_end_0 = const()[name = tensor("op_19728_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_19728_end_mask_0 = const()[name = tensor("op_19728_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19728_cast_fp16 = slice_by_index(begin = var_19728_begin_0, end = var_19728_end_0, end_mask = var_19728_end_mask_0, x = q_91_cast_fp16)[name = tensor("op_19728_cast_fp16")]; + tensor var_19732_begin_0 = const()[name = tensor("op_19732_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_19732_end_0 = const()[name = tensor("op_19732_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_19732_end_mask_0 = const()[name = tensor("op_19732_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19732_cast_fp16 = slice_by_index(begin = var_19732_begin_0, end = var_19732_end_0, end_mask = var_19732_end_mask_0, x = q_91_cast_fp16)[name = tensor("op_19732_cast_fp16")]; + tensor var_19736_begin_0 = const()[name = tensor("op_19736_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_19736_end_0 = const()[name = tensor("op_19736_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_19736_end_mask_0 = const()[name = tensor("op_19736_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19736_cast_fp16 = slice_by_index(begin = var_19736_begin_0, end = var_19736_end_0, end_mask = var_19736_end_mask_0, x = q_91_cast_fp16)[name = tensor("op_19736_cast_fp16")]; + tensor var_19740_begin_0 = const()[name = tensor("op_19740_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_19740_end_0 = const()[name = tensor("op_19740_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_19740_end_mask_0 = const()[name = tensor("op_19740_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19740_cast_fp16 = slice_by_index(begin = var_19740_begin_0, end = var_19740_end_0, end_mask = var_19740_end_mask_0, x = q_91_cast_fp16)[name = tensor("op_19740_cast_fp16")]; + tensor var_19744_begin_0 = const()[name = tensor("op_19744_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_19744_end_0 = const()[name = tensor("op_19744_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_19744_end_mask_0 = const()[name = tensor("op_19744_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19744_cast_fp16 = slice_by_index(begin = var_19744_begin_0, end = var_19744_end_0, end_mask = var_19744_end_mask_0, x = q_91_cast_fp16)[name = tensor("op_19744_cast_fp16")]; + tensor var_19748_begin_0 = const()[name = tensor("op_19748_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_19748_end_0 = const()[name = tensor("op_19748_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_19748_end_mask_0 = const()[name = tensor("op_19748_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19748_cast_fp16 = slice_by_index(begin = var_19748_begin_0, end = var_19748_end_0, end_mask = var_19748_end_mask_0, x = q_91_cast_fp16)[name = tensor("op_19748_cast_fp16")]; + tensor var_19752_begin_0 = const()[name = tensor("op_19752_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_19752_end_0 = const()[name = tensor("op_19752_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_19752_end_mask_0 = const()[name = tensor("op_19752_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19752_cast_fp16 = slice_by_index(begin = var_19752_begin_0, end = var_19752_end_0, end_mask = var_19752_end_mask_0, x = q_91_cast_fp16)[name = tensor("op_19752_cast_fp16")]; + tensor var_19756_begin_0 = const()[name = tensor("op_19756_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_19756_end_0 = const()[name = tensor("op_19756_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_19756_end_mask_0 = const()[name = tensor("op_19756_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19756_cast_fp16 = slice_by_index(begin = var_19756_begin_0, end = var_19756_end_0, end_mask = var_19756_end_mask_0, x = q_91_cast_fp16)[name = tensor("op_19756_cast_fp16")]; + tensor var_19760_begin_0 = const()[name = tensor("op_19760_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_19760_end_0 = const()[name = tensor("op_19760_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_19760_end_mask_0 = const()[name = tensor("op_19760_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19760_cast_fp16 = slice_by_index(begin = var_19760_begin_0, end = var_19760_end_0, end_mask = var_19760_end_mask_0, x = q_91_cast_fp16)[name = tensor("op_19760_cast_fp16")]; + tensor var_19764_begin_0 = const()[name = tensor("op_19764_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_19764_end_0 = const()[name = tensor("op_19764_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_19764_end_mask_0 = const()[name = tensor("op_19764_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19764_cast_fp16 = slice_by_index(begin = var_19764_begin_0, end = var_19764_end_0, end_mask = var_19764_end_mask_0, x = q_91_cast_fp16)[name = tensor("op_19764_cast_fp16")]; + tensor k_183_perm_0 = const()[name = tensor("k_183_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_19771_begin_0 = const()[name = tensor("op_19771_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_19771_end_0 = const()[name = tensor("op_19771_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_19771_end_mask_0 = const()[name = tensor("op_19771_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_183_cast_fp16 = transpose(perm = k_183_perm_0, x = k_181_cast_fp16)[name = tensor("transpose_94")]; + tensor var_19771_cast_fp16 = slice_by_index(begin = var_19771_begin_0, end = var_19771_end_0, end_mask = var_19771_end_mask_0, x = k_183_cast_fp16)[name = tensor("op_19771_cast_fp16")]; + tensor var_19775_begin_0 = const()[name = tensor("op_19775_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_19775_end_0 = const()[name = tensor("op_19775_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_19775_end_mask_0 = const()[name = tensor("op_19775_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19775_cast_fp16 = slice_by_index(begin = var_19775_begin_0, end = var_19775_end_0, end_mask = var_19775_end_mask_0, x = k_183_cast_fp16)[name = tensor("op_19775_cast_fp16")]; + tensor var_19779_begin_0 = const()[name = tensor("op_19779_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_19779_end_0 = const()[name = tensor("op_19779_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_19779_end_mask_0 = const()[name = tensor("op_19779_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19779_cast_fp16 = slice_by_index(begin = var_19779_begin_0, end = var_19779_end_0, end_mask = var_19779_end_mask_0, x = k_183_cast_fp16)[name = tensor("op_19779_cast_fp16")]; + tensor var_19783_begin_0 = const()[name = tensor("op_19783_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_19783_end_0 = const()[name = tensor("op_19783_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_19783_end_mask_0 = const()[name = tensor("op_19783_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19783_cast_fp16 = slice_by_index(begin = var_19783_begin_0, end = var_19783_end_0, end_mask = var_19783_end_mask_0, x = k_183_cast_fp16)[name = tensor("op_19783_cast_fp16")]; + tensor var_19787_begin_0 = const()[name = tensor("op_19787_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_19787_end_0 = const()[name = tensor("op_19787_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_19787_end_mask_0 = const()[name = tensor("op_19787_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19787_cast_fp16 = slice_by_index(begin = var_19787_begin_0, end = var_19787_end_0, end_mask = var_19787_end_mask_0, x = k_183_cast_fp16)[name = tensor("op_19787_cast_fp16")]; + tensor var_19791_begin_0 = const()[name = tensor("op_19791_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_19791_end_0 = const()[name = tensor("op_19791_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_19791_end_mask_0 = const()[name = tensor("op_19791_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19791_cast_fp16 = slice_by_index(begin = var_19791_begin_0, end = var_19791_end_0, end_mask = var_19791_end_mask_0, x = k_183_cast_fp16)[name = tensor("op_19791_cast_fp16")]; + tensor var_19795_begin_0 = const()[name = tensor("op_19795_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_19795_end_0 = const()[name = tensor("op_19795_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_19795_end_mask_0 = const()[name = tensor("op_19795_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19795_cast_fp16 = slice_by_index(begin = var_19795_begin_0, end = var_19795_end_0, end_mask = var_19795_end_mask_0, x = k_183_cast_fp16)[name = tensor("op_19795_cast_fp16")]; + tensor var_19799_begin_0 = const()[name = tensor("op_19799_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_19799_end_0 = const()[name = tensor("op_19799_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_19799_end_mask_0 = const()[name = tensor("op_19799_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19799_cast_fp16 = slice_by_index(begin = var_19799_begin_0, end = var_19799_end_0, end_mask = var_19799_end_mask_0, x = k_183_cast_fp16)[name = tensor("op_19799_cast_fp16")]; + tensor var_19803_begin_0 = const()[name = tensor("op_19803_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_19803_end_0 = const()[name = tensor("op_19803_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_19803_end_mask_0 = const()[name = tensor("op_19803_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19803_cast_fp16 = slice_by_index(begin = var_19803_begin_0, end = var_19803_end_0, end_mask = var_19803_end_mask_0, x = k_183_cast_fp16)[name = tensor("op_19803_cast_fp16")]; + tensor var_19807_begin_0 = const()[name = tensor("op_19807_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_19807_end_0 = const()[name = tensor("op_19807_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_19807_end_mask_0 = const()[name = tensor("op_19807_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19807_cast_fp16 = slice_by_index(begin = var_19807_begin_0, end = var_19807_end_0, end_mask = var_19807_end_mask_0, x = k_183_cast_fp16)[name = tensor("op_19807_cast_fp16")]; + tensor var_19811_begin_0 = const()[name = tensor("op_19811_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_19811_end_0 = const()[name = tensor("op_19811_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_19811_end_mask_0 = const()[name = tensor("op_19811_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19811_cast_fp16 = slice_by_index(begin = var_19811_begin_0, end = var_19811_end_0, end_mask = var_19811_end_mask_0, x = k_183_cast_fp16)[name = tensor("op_19811_cast_fp16")]; + tensor var_19815_begin_0 = const()[name = tensor("op_19815_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_19815_end_0 = const()[name = tensor("op_19815_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_19815_end_mask_0 = const()[name = tensor("op_19815_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19815_cast_fp16 = slice_by_index(begin = var_19815_begin_0, end = var_19815_end_0, end_mask = var_19815_end_mask_0, x = k_183_cast_fp16)[name = tensor("op_19815_cast_fp16")]; + tensor var_19819_begin_0 = const()[name = tensor("op_19819_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_19819_end_0 = const()[name = tensor("op_19819_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_19819_end_mask_0 = const()[name = tensor("op_19819_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19819_cast_fp16 = slice_by_index(begin = var_19819_begin_0, end = var_19819_end_0, end_mask = var_19819_end_mask_0, x = k_183_cast_fp16)[name = tensor("op_19819_cast_fp16")]; + tensor var_19823_begin_0 = const()[name = tensor("op_19823_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_19823_end_0 = const()[name = tensor("op_19823_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_19823_end_mask_0 = const()[name = tensor("op_19823_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19823_cast_fp16 = slice_by_index(begin = var_19823_begin_0, end = var_19823_end_0, end_mask = var_19823_end_mask_0, x = k_183_cast_fp16)[name = tensor("op_19823_cast_fp16")]; + tensor var_19827_begin_0 = const()[name = tensor("op_19827_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_19827_end_0 = const()[name = tensor("op_19827_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_19827_end_mask_0 = const()[name = tensor("op_19827_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19827_cast_fp16 = slice_by_index(begin = var_19827_begin_0, end = var_19827_end_0, end_mask = var_19827_end_mask_0, x = k_183_cast_fp16)[name = tensor("op_19827_cast_fp16")]; + tensor var_19831_begin_0 = const()[name = tensor("op_19831_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_19831_end_0 = const()[name = tensor("op_19831_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_19831_end_mask_0 = const()[name = tensor("op_19831_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19831_cast_fp16 = slice_by_index(begin = var_19831_begin_0, end = var_19831_end_0, end_mask = var_19831_end_mask_0, x = k_183_cast_fp16)[name = tensor("op_19831_cast_fp16")]; + tensor var_19835_begin_0 = const()[name = tensor("op_19835_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_19835_end_0 = const()[name = tensor("op_19835_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_19835_end_mask_0 = const()[name = tensor("op_19835_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19835_cast_fp16 = slice_by_index(begin = var_19835_begin_0, end = var_19835_end_0, end_mask = var_19835_end_mask_0, x = k_183_cast_fp16)[name = tensor("op_19835_cast_fp16")]; + tensor var_19839_begin_0 = const()[name = tensor("op_19839_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_19839_end_0 = const()[name = tensor("op_19839_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_19839_end_mask_0 = const()[name = tensor("op_19839_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19839_cast_fp16 = slice_by_index(begin = var_19839_begin_0, end = var_19839_end_0, end_mask = var_19839_end_mask_0, x = k_183_cast_fp16)[name = tensor("op_19839_cast_fp16")]; + tensor var_19843_begin_0 = const()[name = tensor("op_19843_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_19843_end_0 = const()[name = tensor("op_19843_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_19843_end_mask_0 = const()[name = tensor("op_19843_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19843_cast_fp16 = slice_by_index(begin = var_19843_begin_0, end = var_19843_end_0, end_mask = var_19843_end_mask_0, x = k_183_cast_fp16)[name = tensor("op_19843_cast_fp16")]; + tensor var_19847_begin_0 = const()[name = tensor("op_19847_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_19847_end_0 = const()[name = tensor("op_19847_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_19847_end_mask_0 = const()[name = tensor("op_19847_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19847_cast_fp16 = slice_by_index(begin = var_19847_begin_0, end = var_19847_end_0, end_mask = var_19847_end_mask_0, x = k_183_cast_fp16)[name = tensor("op_19847_cast_fp16")]; + tensor var_19849_begin_0 = const()[name = tensor("op_19849_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_19849_end_0 = const()[name = tensor("op_19849_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_19849_end_mask_0 = const()[name = tensor("op_19849_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19849_cast_fp16 = slice_by_index(begin = var_19849_begin_0, end = var_19849_end_0, end_mask = var_19849_end_mask_0, x = v_91_cast_fp16)[name = tensor("op_19849_cast_fp16")]; + tensor var_19853_begin_0 = const()[name = tensor("op_19853_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_19853_end_0 = const()[name = tensor("op_19853_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_19853_end_mask_0 = const()[name = tensor("op_19853_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19853_cast_fp16 = slice_by_index(begin = var_19853_begin_0, end = var_19853_end_0, end_mask = var_19853_end_mask_0, x = v_91_cast_fp16)[name = tensor("op_19853_cast_fp16")]; + tensor var_19857_begin_0 = const()[name = tensor("op_19857_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_19857_end_0 = const()[name = tensor("op_19857_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_19857_end_mask_0 = const()[name = tensor("op_19857_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19857_cast_fp16 = slice_by_index(begin = var_19857_begin_0, end = var_19857_end_0, end_mask = var_19857_end_mask_0, x = v_91_cast_fp16)[name = tensor("op_19857_cast_fp16")]; + tensor var_19861_begin_0 = const()[name = tensor("op_19861_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_19861_end_0 = const()[name = tensor("op_19861_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_19861_end_mask_0 = const()[name = tensor("op_19861_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19861_cast_fp16 = slice_by_index(begin = var_19861_begin_0, end = var_19861_end_0, end_mask = var_19861_end_mask_0, x = v_91_cast_fp16)[name = tensor("op_19861_cast_fp16")]; + tensor var_19865_begin_0 = const()[name = tensor("op_19865_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_19865_end_0 = const()[name = tensor("op_19865_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_19865_end_mask_0 = const()[name = tensor("op_19865_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19865_cast_fp16 = slice_by_index(begin = var_19865_begin_0, end = var_19865_end_0, end_mask = var_19865_end_mask_0, x = v_91_cast_fp16)[name = tensor("op_19865_cast_fp16")]; + tensor var_19869_begin_0 = const()[name = tensor("op_19869_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_19869_end_0 = const()[name = tensor("op_19869_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_19869_end_mask_0 = const()[name = tensor("op_19869_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19869_cast_fp16 = slice_by_index(begin = var_19869_begin_0, end = var_19869_end_0, end_mask = var_19869_end_mask_0, x = v_91_cast_fp16)[name = tensor("op_19869_cast_fp16")]; + tensor var_19873_begin_0 = const()[name = tensor("op_19873_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_19873_end_0 = const()[name = tensor("op_19873_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_19873_end_mask_0 = const()[name = tensor("op_19873_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19873_cast_fp16 = slice_by_index(begin = var_19873_begin_0, end = var_19873_end_0, end_mask = var_19873_end_mask_0, x = v_91_cast_fp16)[name = tensor("op_19873_cast_fp16")]; + tensor var_19877_begin_0 = const()[name = tensor("op_19877_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_19877_end_0 = const()[name = tensor("op_19877_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_19877_end_mask_0 = const()[name = tensor("op_19877_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19877_cast_fp16 = slice_by_index(begin = var_19877_begin_0, end = var_19877_end_0, end_mask = var_19877_end_mask_0, x = v_91_cast_fp16)[name = tensor("op_19877_cast_fp16")]; + tensor var_19881_begin_0 = const()[name = tensor("op_19881_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_19881_end_0 = const()[name = tensor("op_19881_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_19881_end_mask_0 = const()[name = tensor("op_19881_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19881_cast_fp16 = slice_by_index(begin = var_19881_begin_0, end = var_19881_end_0, end_mask = var_19881_end_mask_0, x = v_91_cast_fp16)[name = tensor("op_19881_cast_fp16")]; + tensor var_19885_begin_0 = const()[name = tensor("op_19885_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_19885_end_0 = const()[name = tensor("op_19885_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_19885_end_mask_0 = const()[name = tensor("op_19885_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19885_cast_fp16 = slice_by_index(begin = var_19885_begin_0, end = var_19885_end_0, end_mask = var_19885_end_mask_0, x = v_91_cast_fp16)[name = tensor("op_19885_cast_fp16")]; + tensor var_19889_begin_0 = const()[name = tensor("op_19889_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_19889_end_0 = const()[name = tensor("op_19889_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_19889_end_mask_0 = const()[name = tensor("op_19889_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19889_cast_fp16 = slice_by_index(begin = var_19889_begin_0, end = var_19889_end_0, end_mask = var_19889_end_mask_0, x = v_91_cast_fp16)[name = tensor("op_19889_cast_fp16")]; + tensor var_19893_begin_0 = const()[name = tensor("op_19893_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_19893_end_0 = const()[name = tensor("op_19893_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_19893_end_mask_0 = const()[name = tensor("op_19893_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19893_cast_fp16 = slice_by_index(begin = var_19893_begin_0, end = var_19893_end_0, end_mask = var_19893_end_mask_0, x = v_91_cast_fp16)[name = tensor("op_19893_cast_fp16")]; + tensor var_19897_begin_0 = const()[name = tensor("op_19897_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_19897_end_0 = const()[name = tensor("op_19897_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_19897_end_mask_0 = const()[name = tensor("op_19897_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19897_cast_fp16 = slice_by_index(begin = var_19897_begin_0, end = var_19897_end_0, end_mask = var_19897_end_mask_0, x = v_91_cast_fp16)[name = tensor("op_19897_cast_fp16")]; + tensor var_19901_begin_0 = const()[name = tensor("op_19901_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_19901_end_0 = const()[name = tensor("op_19901_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_19901_end_mask_0 = const()[name = tensor("op_19901_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19901_cast_fp16 = slice_by_index(begin = var_19901_begin_0, end = var_19901_end_0, end_mask = var_19901_end_mask_0, x = v_91_cast_fp16)[name = tensor("op_19901_cast_fp16")]; + tensor var_19905_begin_0 = const()[name = tensor("op_19905_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_19905_end_0 = const()[name = tensor("op_19905_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_19905_end_mask_0 = const()[name = tensor("op_19905_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19905_cast_fp16 = slice_by_index(begin = var_19905_begin_0, end = var_19905_end_0, end_mask = var_19905_end_mask_0, x = v_91_cast_fp16)[name = tensor("op_19905_cast_fp16")]; + tensor var_19909_begin_0 = const()[name = tensor("op_19909_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_19909_end_0 = const()[name = tensor("op_19909_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_19909_end_mask_0 = const()[name = tensor("op_19909_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19909_cast_fp16 = slice_by_index(begin = var_19909_begin_0, end = var_19909_end_0, end_mask = var_19909_end_mask_0, x = v_91_cast_fp16)[name = tensor("op_19909_cast_fp16")]; + tensor var_19913_begin_0 = const()[name = tensor("op_19913_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_19913_end_0 = const()[name = tensor("op_19913_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_19913_end_mask_0 = const()[name = tensor("op_19913_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19913_cast_fp16 = slice_by_index(begin = var_19913_begin_0, end = var_19913_end_0, end_mask = var_19913_end_mask_0, x = v_91_cast_fp16)[name = tensor("op_19913_cast_fp16")]; + tensor var_19917_begin_0 = const()[name = tensor("op_19917_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_19917_end_0 = const()[name = tensor("op_19917_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_19917_end_mask_0 = const()[name = tensor("op_19917_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19917_cast_fp16 = slice_by_index(begin = var_19917_begin_0, end = var_19917_end_0, end_mask = var_19917_end_mask_0, x = v_91_cast_fp16)[name = tensor("op_19917_cast_fp16")]; + tensor var_19921_begin_0 = const()[name = tensor("op_19921_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_19921_end_0 = const()[name = tensor("op_19921_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_19921_end_mask_0 = const()[name = tensor("op_19921_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19921_cast_fp16 = slice_by_index(begin = var_19921_begin_0, end = var_19921_end_0, end_mask = var_19921_end_mask_0, x = v_91_cast_fp16)[name = tensor("op_19921_cast_fp16")]; + tensor var_19925_begin_0 = const()[name = tensor("op_19925_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_19925_end_0 = const()[name = tensor("op_19925_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_19925_end_mask_0 = const()[name = tensor("op_19925_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19925_cast_fp16 = slice_by_index(begin = var_19925_begin_0, end = var_19925_end_0, end_mask = var_19925_end_mask_0, x = v_91_cast_fp16)[name = tensor("op_19925_cast_fp16")]; + tensor var_19929_equation_0 = const()[name = tensor("op_19929_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19929_cast_fp16 = einsum(equation = var_19929_equation_0, values = (var_19771_cast_fp16, var_19688_cast_fp16))[name = tensor("op_19929_cast_fp16")]; + tensor var_19930_to_fp16 = const()[name = tensor("op_19930_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1641_cast_fp16 = mul(x = var_19929_cast_fp16, y = var_19930_to_fp16)[name = tensor("aw_1641_cast_fp16")]; + tensor var_19933_equation_0 = const()[name = tensor("op_19933_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19933_cast_fp16 = einsum(equation = var_19933_equation_0, values = (var_19775_cast_fp16, var_19692_cast_fp16))[name = tensor("op_19933_cast_fp16")]; + tensor var_19934_to_fp16 = const()[name = tensor("op_19934_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1643_cast_fp16 = mul(x = var_19933_cast_fp16, y = var_19934_to_fp16)[name = tensor("aw_1643_cast_fp16")]; + tensor var_19937_equation_0 = const()[name = tensor("op_19937_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19937_cast_fp16 = einsum(equation = var_19937_equation_0, values = (var_19779_cast_fp16, var_19696_cast_fp16))[name = tensor("op_19937_cast_fp16")]; + tensor var_19938_to_fp16 = const()[name = tensor("op_19938_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1645_cast_fp16 = mul(x = var_19937_cast_fp16, y = var_19938_to_fp16)[name = tensor("aw_1645_cast_fp16")]; + tensor var_19941_equation_0 = const()[name = tensor("op_19941_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19941_cast_fp16 = einsum(equation = var_19941_equation_0, values = (var_19783_cast_fp16, var_19700_cast_fp16))[name = tensor("op_19941_cast_fp16")]; + tensor var_19942_to_fp16 = const()[name = tensor("op_19942_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1647_cast_fp16 = mul(x = var_19941_cast_fp16, y = var_19942_to_fp16)[name = tensor("aw_1647_cast_fp16")]; + tensor var_19945_equation_0 = const()[name = tensor("op_19945_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19945_cast_fp16 = einsum(equation = var_19945_equation_0, values = (var_19787_cast_fp16, var_19704_cast_fp16))[name = tensor("op_19945_cast_fp16")]; + tensor var_19946_to_fp16 = const()[name = tensor("op_19946_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1649_cast_fp16 = mul(x = var_19945_cast_fp16, y = var_19946_to_fp16)[name = tensor("aw_1649_cast_fp16")]; + tensor var_19949_equation_0 = const()[name = tensor("op_19949_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19949_cast_fp16 = einsum(equation = var_19949_equation_0, values = (var_19791_cast_fp16, var_19708_cast_fp16))[name = tensor("op_19949_cast_fp16")]; + tensor var_19950_to_fp16 = const()[name = tensor("op_19950_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1651_cast_fp16 = mul(x = var_19949_cast_fp16, y = var_19950_to_fp16)[name = tensor("aw_1651_cast_fp16")]; + tensor var_19953_equation_0 = const()[name = tensor("op_19953_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19953_cast_fp16 = einsum(equation = var_19953_equation_0, values = (var_19795_cast_fp16, var_19712_cast_fp16))[name = tensor("op_19953_cast_fp16")]; + tensor var_19954_to_fp16 = const()[name = tensor("op_19954_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1653_cast_fp16 = mul(x = var_19953_cast_fp16, y = var_19954_to_fp16)[name = tensor("aw_1653_cast_fp16")]; + tensor var_19957_equation_0 = const()[name = tensor("op_19957_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19957_cast_fp16 = einsum(equation = var_19957_equation_0, values = (var_19799_cast_fp16, var_19716_cast_fp16))[name = tensor("op_19957_cast_fp16")]; + tensor var_19958_to_fp16 = const()[name = tensor("op_19958_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1655_cast_fp16 = mul(x = var_19957_cast_fp16, y = var_19958_to_fp16)[name = tensor("aw_1655_cast_fp16")]; + tensor var_19961_equation_0 = const()[name = tensor("op_19961_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19961_cast_fp16 = einsum(equation = var_19961_equation_0, values = (var_19803_cast_fp16, var_19720_cast_fp16))[name = tensor("op_19961_cast_fp16")]; + tensor var_19962_to_fp16 = const()[name = tensor("op_19962_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1657_cast_fp16 = mul(x = var_19961_cast_fp16, y = var_19962_to_fp16)[name = tensor("aw_1657_cast_fp16")]; + tensor var_19965_equation_0 = const()[name = tensor("op_19965_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19965_cast_fp16 = einsum(equation = var_19965_equation_0, values = (var_19807_cast_fp16, var_19724_cast_fp16))[name = tensor("op_19965_cast_fp16")]; + tensor var_19966_to_fp16 = const()[name = tensor("op_19966_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1659_cast_fp16 = mul(x = var_19965_cast_fp16, y = var_19966_to_fp16)[name = tensor("aw_1659_cast_fp16")]; + tensor var_19969_equation_0 = const()[name = tensor("op_19969_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19969_cast_fp16 = einsum(equation = var_19969_equation_0, values = (var_19811_cast_fp16, var_19728_cast_fp16))[name = tensor("op_19969_cast_fp16")]; + tensor var_19970_to_fp16 = const()[name = tensor("op_19970_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1661_cast_fp16 = mul(x = var_19969_cast_fp16, y = var_19970_to_fp16)[name = tensor("aw_1661_cast_fp16")]; + tensor var_19973_equation_0 = const()[name = tensor("op_19973_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19973_cast_fp16 = einsum(equation = var_19973_equation_0, values = (var_19815_cast_fp16, var_19732_cast_fp16))[name = tensor("op_19973_cast_fp16")]; + tensor var_19974_to_fp16 = const()[name = tensor("op_19974_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1663_cast_fp16 = mul(x = var_19973_cast_fp16, y = var_19974_to_fp16)[name = tensor("aw_1663_cast_fp16")]; + tensor var_19977_equation_0 = const()[name = tensor("op_19977_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19977_cast_fp16 = einsum(equation = var_19977_equation_0, values = (var_19819_cast_fp16, var_19736_cast_fp16))[name = tensor("op_19977_cast_fp16")]; + tensor var_19978_to_fp16 = const()[name = tensor("op_19978_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1665_cast_fp16 = mul(x = var_19977_cast_fp16, y = var_19978_to_fp16)[name = tensor("aw_1665_cast_fp16")]; + tensor var_19981_equation_0 = const()[name = tensor("op_19981_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19981_cast_fp16 = einsum(equation = var_19981_equation_0, values = (var_19823_cast_fp16, var_19740_cast_fp16))[name = tensor("op_19981_cast_fp16")]; + tensor var_19982_to_fp16 = const()[name = tensor("op_19982_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1667_cast_fp16 = mul(x = var_19981_cast_fp16, y = var_19982_to_fp16)[name = tensor("aw_1667_cast_fp16")]; + tensor var_19985_equation_0 = const()[name = tensor("op_19985_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19985_cast_fp16 = einsum(equation = var_19985_equation_0, values = (var_19827_cast_fp16, var_19744_cast_fp16))[name = tensor("op_19985_cast_fp16")]; + tensor var_19986_to_fp16 = const()[name = tensor("op_19986_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1669_cast_fp16 = mul(x = var_19985_cast_fp16, y = var_19986_to_fp16)[name = tensor("aw_1669_cast_fp16")]; + tensor var_19989_equation_0 = const()[name = tensor("op_19989_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19989_cast_fp16 = einsum(equation = var_19989_equation_0, values = (var_19831_cast_fp16, var_19748_cast_fp16))[name = tensor("op_19989_cast_fp16")]; + tensor var_19990_to_fp16 = const()[name = tensor("op_19990_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1671_cast_fp16 = mul(x = var_19989_cast_fp16, y = var_19990_to_fp16)[name = tensor("aw_1671_cast_fp16")]; + tensor var_19993_equation_0 = const()[name = tensor("op_19993_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19993_cast_fp16 = einsum(equation = var_19993_equation_0, values = (var_19835_cast_fp16, var_19752_cast_fp16))[name = tensor("op_19993_cast_fp16")]; + tensor var_19994_to_fp16 = const()[name = tensor("op_19994_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1673_cast_fp16 = mul(x = var_19993_cast_fp16, y = var_19994_to_fp16)[name = tensor("aw_1673_cast_fp16")]; + tensor var_19997_equation_0 = const()[name = tensor("op_19997_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19997_cast_fp16 = einsum(equation = var_19997_equation_0, values = (var_19839_cast_fp16, var_19756_cast_fp16))[name = tensor("op_19997_cast_fp16")]; + tensor var_19998_to_fp16 = const()[name = tensor("op_19998_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1675_cast_fp16 = mul(x = var_19997_cast_fp16, y = var_19998_to_fp16)[name = tensor("aw_1675_cast_fp16")]; + tensor var_20001_equation_0 = const()[name = tensor("op_20001_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20001_cast_fp16 = einsum(equation = var_20001_equation_0, values = (var_19843_cast_fp16, var_19760_cast_fp16))[name = tensor("op_20001_cast_fp16")]; + tensor var_20002_to_fp16 = const()[name = tensor("op_20002_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1677_cast_fp16 = mul(x = var_20001_cast_fp16, y = var_20002_to_fp16)[name = tensor("aw_1677_cast_fp16")]; + tensor var_20005_equation_0 = const()[name = tensor("op_20005_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20005_cast_fp16 = einsum(equation = var_20005_equation_0, values = (var_19847_cast_fp16, var_19764_cast_fp16))[name = tensor("op_20005_cast_fp16")]; + tensor var_20006_to_fp16 = const()[name = tensor("op_20006_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1679_cast_fp16 = mul(x = var_20005_cast_fp16, y = var_20006_to_fp16)[name = tensor("aw_1679_cast_fp16")]; + tensor var_20008_cast_fp16 = softmax(axis = var_2624, x = aw_1641_cast_fp16)[name = tensor("op_20008_cast_fp16")]; + tensor var_20009_cast_fp16 = softmax(axis = var_2624, x = aw_1643_cast_fp16)[name = tensor("op_20009_cast_fp16")]; + tensor var_20010_cast_fp16 = softmax(axis = var_2624, x = aw_1645_cast_fp16)[name = tensor("op_20010_cast_fp16")]; + tensor var_20011_cast_fp16 = softmax(axis = var_2624, x = aw_1647_cast_fp16)[name = tensor("op_20011_cast_fp16")]; + tensor var_20012_cast_fp16 = softmax(axis = var_2624, x = aw_1649_cast_fp16)[name = tensor("op_20012_cast_fp16")]; + tensor var_20013_cast_fp16 = softmax(axis = var_2624, x = aw_1651_cast_fp16)[name = tensor("op_20013_cast_fp16")]; + tensor var_20014_cast_fp16 = softmax(axis = var_2624, x = aw_1653_cast_fp16)[name = tensor("op_20014_cast_fp16")]; + tensor var_20015_cast_fp16 = softmax(axis = var_2624, x = aw_1655_cast_fp16)[name = tensor("op_20015_cast_fp16")]; + tensor var_20016_cast_fp16 = softmax(axis = var_2624, x = aw_1657_cast_fp16)[name = tensor("op_20016_cast_fp16")]; + tensor var_20017_cast_fp16 = softmax(axis = var_2624, x = aw_1659_cast_fp16)[name = tensor("op_20017_cast_fp16")]; + tensor var_20018_cast_fp16 = softmax(axis = var_2624, x = aw_1661_cast_fp16)[name = tensor("op_20018_cast_fp16")]; + tensor var_20019_cast_fp16 = softmax(axis = var_2624, x = aw_1663_cast_fp16)[name = tensor("op_20019_cast_fp16")]; + tensor var_20020_cast_fp16 = softmax(axis = var_2624, x = aw_1665_cast_fp16)[name = tensor("op_20020_cast_fp16")]; + tensor var_20021_cast_fp16 = softmax(axis = var_2624, x = aw_1667_cast_fp16)[name = tensor("op_20021_cast_fp16")]; + tensor var_20022_cast_fp16 = softmax(axis = var_2624, x = aw_1669_cast_fp16)[name = tensor("op_20022_cast_fp16")]; + tensor var_20023_cast_fp16 = softmax(axis = var_2624, x = aw_1671_cast_fp16)[name = tensor("op_20023_cast_fp16")]; + tensor var_20024_cast_fp16 = softmax(axis = var_2624, x = aw_1673_cast_fp16)[name = tensor("op_20024_cast_fp16")]; + tensor var_20025_cast_fp16 = softmax(axis = var_2624, x = aw_1675_cast_fp16)[name = tensor("op_20025_cast_fp16")]; + tensor var_20026_cast_fp16 = softmax(axis = var_2624, x = aw_1677_cast_fp16)[name = tensor("op_20026_cast_fp16")]; + tensor var_20027_cast_fp16 = softmax(axis = var_2624, x = aw_1679_cast_fp16)[name = tensor("op_20027_cast_fp16")]; + tensor var_20029_equation_0 = const()[name = tensor("op_20029_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20029_cast_fp16 = einsum(equation = var_20029_equation_0, values = (var_19849_cast_fp16, var_20008_cast_fp16))[name = tensor("op_20029_cast_fp16")]; + tensor var_20031_equation_0 = const()[name = tensor("op_20031_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20031_cast_fp16 = einsum(equation = var_20031_equation_0, values = (var_19853_cast_fp16, var_20009_cast_fp16))[name = tensor("op_20031_cast_fp16")]; + tensor var_20033_equation_0 = const()[name = tensor("op_20033_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20033_cast_fp16 = einsum(equation = var_20033_equation_0, values = (var_19857_cast_fp16, var_20010_cast_fp16))[name = tensor("op_20033_cast_fp16")]; + tensor var_20035_equation_0 = const()[name = tensor("op_20035_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20035_cast_fp16 = einsum(equation = var_20035_equation_0, values = (var_19861_cast_fp16, var_20011_cast_fp16))[name = tensor("op_20035_cast_fp16")]; + tensor var_20037_equation_0 = const()[name = tensor("op_20037_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20037_cast_fp16 = einsum(equation = var_20037_equation_0, values = (var_19865_cast_fp16, var_20012_cast_fp16))[name = tensor("op_20037_cast_fp16")]; + tensor var_20039_equation_0 = const()[name = tensor("op_20039_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20039_cast_fp16 = einsum(equation = var_20039_equation_0, values = (var_19869_cast_fp16, var_20013_cast_fp16))[name = tensor("op_20039_cast_fp16")]; + tensor var_20041_equation_0 = const()[name = tensor("op_20041_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20041_cast_fp16 = einsum(equation = var_20041_equation_0, values = (var_19873_cast_fp16, var_20014_cast_fp16))[name = tensor("op_20041_cast_fp16")]; + tensor var_20043_equation_0 = const()[name = tensor("op_20043_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20043_cast_fp16 = einsum(equation = var_20043_equation_0, values = (var_19877_cast_fp16, var_20015_cast_fp16))[name = tensor("op_20043_cast_fp16")]; + tensor var_20045_equation_0 = const()[name = tensor("op_20045_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20045_cast_fp16 = einsum(equation = var_20045_equation_0, values = (var_19881_cast_fp16, var_20016_cast_fp16))[name = tensor("op_20045_cast_fp16")]; + tensor var_20047_equation_0 = const()[name = tensor("op_20047_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20047_cast_fp16 = einsum(equation = var_20047_equation_0, values = (var_19885_cast_fp16, var_20017_cast_fp16))[name = tensor("op_20047_cast_fp16")]; + tensor var_20049_equation_0 = const()[name = tensor("op_20049_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20049_cast_fp16 = einsum(equation = var_20049_equation_0, values = (var_19889_cast_fp16, var_20018_cast_fp16))[name = tensor("op_20049_cast_fp16")]; + tensor var_20051_equation_0 = const()[name = tensor("op_20051_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20051_cast_fp16 = einsum(equation = var_20051_equation_0, values = (var_19893_cast_fp16, var_20019_cast_fp16))[name = tensor("op_20051_cast_fp16")]; + tensor var_20053_equation_0 = const()[name = tensor("op_20053_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20053_cast_fp16 = einsum(equation = var_20053_equation_0, values = (var_19897_cast_fp16, var_20020_cast_fp16))[name = tensor("op_20053_cast_fp16")]; + tensor var_20055_equation_0 = const()[name = tensor("op_20055_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20055_cast_fp16 = einsum(equation = var_20055_equation_0, values = (var_19901_cast_fp16, var_20021_cast_fp16))[name = tensor("op_20055_cast_fp16")]; + tensor var_20057_equation_0 = const()[name = tensor("op_20057_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20057_cast_fp16 = einsum(equation = var_20057_equation_0, values = (var_19905_cast_fp16, var_20022_cast_fp16))[name = tensor("op_20057_cast_fp16")]; + tensor var_20059_equation_0 = const()[name = tensor("op_20059_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20059_cast_fp16 = einsum(equation = var_20059_equation_0, values = (var_19909_cast_fp16, var_20023_cast_fp16))[name = tensor("op_20059_cast_fp16")]; + tensor var_20061_equation_0 = const()[name = tensor("op_20061_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20061_cast_fp16 = einsum(equation = var_20061_equation_0, values = (var_19913_cast_fp16, var_20024_cast_fp16))[name = tensor("op_20061_cast_fp16")]; + tensor var_20063_equation_0 = const()[name = tensor("op_20063_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20063_cast_fp16 = einsum(equation = var_20063_equation_0, values = (var_19917_cast_fp16, var_20025_cast_fp16))[name = tensor("op_20063_cast_fp16")]; + tensor var_20065_equation_0 = const()[name = tensor("op_20065_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20065_cast_fp16 = einsum(equation = var_20065_equation_0, values = (var_19921_cast_fp16, var_20026_cast_fp16))[name = tensor("op_20065_cast_fp16")]; + tensor var_20067_equation_0 = const()[name = tensor("op_20067_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20067_cast_fp16 = einsum(equation = var_20067_equation_0, values = (var_19925_cast_fp16, var_20027_cast_fp16))[name = tensor("op_20067_cast_fp16")]; + tensor input_295_interleave_0 = const()[name = tensor("input_295_interleave_0"), val = tensor(false)]; + tensor input_295_cast_fp16 = concat(axis = var_2624, interleave = input_295_interleave_0, values = (var_20029_cast_fp16, var_20031_cast_fp16, var_20033_cast_fp16, var_20035_cast_fp16, var_20037_cast_fp16, var_20039_cast_fp16, var_20041_cast_fp16, var_20043_cast_fp16, var_20045_cast_fp16, var_20047_cast_fp16, var_20049_cast_fp16, var_20051_cast_fp16, var_20053_cast_fp16, var_20055_cast_fp16, var_20057_cast_fp16, var_20059_cast_fp16, var_20061_cast_fp16, var_20063_cast_fp16, var_20065_cast_fp16, var_20067_cast_fp16))[name = tensor("input_295_cast_fp16")]; + tensor var_20073 = const()[name = tensor("op_20073"), val = tensor([1, 1])]; + tensor var_20075 = const()[name = tensor("op_20075"), val = tensor([1, 1])]; + tensor var_20077_pad_type_0 = const()[name = tensor("op_20077_pad_type_0"), val = tensor("custom")]; + tensor var_20077_pad_0 = const()[name = tensor("op_20077_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_8_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_8_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1545124480)))]; + tensor down_blocks_2_attentions_1_transformer_blocks_8_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_8_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1548401344)))]; + tensor var_20077_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_8_attn2_to_out_0_bias_to_fp16, dilations = var_20075, groups = var_2624, pad = var_20077_pad_0, pad_type = var_20077_pad_type_0, strides = var_20073, weight = down_blocks_2_attentions_1_transformer_blocks_8_attn2_to_out_0_weight_to_fp16, x = input_295_cast_fp16)[name = tensor("op_20077_cast_fp16")]; + tensor inputs_137_cast_fp16 = add(x = var_20077_cast_fp16, y = inputs_135_cast_fp16)[name = tensor("inputs_137_cast_fp16")]; + tensor input_297_axes_0 = const()[name = tensor("input_297_axes_0"), val = tensor([1])]; + tensor input_297_gamma_0_to_fp16 = const()[name = tensor("input_297_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1548403968)))]; + tensor input_297_beta_0_to_fp16 = const()[name = tensor("input_297_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1548406592)))]; + tensor var_20087_to_fp16 = const()[name = tensor("op_20087_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_297_cast_fp16 = layer_norm(axes = input_297_axes_0, beta = input_297_beta_0_to_fp16, epsilon = var_20087_to_fp16, gamma = input_297_gamma_0_to_fp16, x = inputs_137_cast_fp16)[name = tensor("input_297_cast_fp16")]; + tensor var_20103 = const()[name = tensor("op_20103"), val = tensor([1, 1])]; + tensor var_20105 = const()[name = tensor("op_20105"), val = tensor([1, 1])]; + tensor var_20107_pad_type_0 = const()[name = tensor("op_20107_pad_type_0"), val = tensor("custom")]; + tensor var_20107_pad_0 = const()[name = tensor("op_20107_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_8_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_8_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1548409216)))]; + tensor down_blocks_2_attentions_1_transformer_blocks_8_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_8_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1574623680)))]; + tensor var_20107_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_8_ff_net_0_proj_bias_to_fp16, dilations = var_20105, groups = var_2624, pad = var_20107_pad_0, pad_type = var_20107_pad_type_0, strides = var_20103, weight = down_blocks_2_attentions_1_transformer_blocks_8_ff_net_0_proj_weight_to_fp16, x = input_297_cast_fp16)[name = tensor("op_20107_cast_fp16")]; + tensor var_20108_split_sizes_0 = const()[name = tensor("op_20108_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_20108_axis_0 = const()[name = tensor("op_20108_axis_0"), val = tensor(1)]; + tensor var_20108_cast_fp16_0, tensor var_20108_cast_fp16_1 = split(axis = var_20108_axis_0, split_sizes = var_20108_split_sizes_0, x = var_20107_cast_fp16)[name = tensor("op_20108_cast_fp16")]; + tensor var_20110_mode_0 = const()[name = tensor("op_20110_mode_0"), val = tensor("EXACT")]; + tensor var_20110_cast_fp16 = gelu(mode = var_20110_mode_0, x = var_20108_cast_fp16_1)[name = tensor("op_20110_cast_fp16")]; + tensor input_299_cast_fp16 = mul(x = var_20108_cast_fp16_0, y = var_20110_cast_fp16)[name = tensor("input_299_cast_fp16")]; + tensor var_20114 = const()[name = tensor("op_20114"), val = tensor([1, 1])]; + tensor var_20116 = const()[name = tensor("op_20116"), val = tensor([1, 1])]; + tensor var_20118_pad_type_0 = const()[name = tensor("op_20118_pad_type_0"), val = tensor("custom")]; + tensor var_20118_pad_0 = const()[name = tensor("op_20118_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_8_ff_net_2_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_8_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1574644224)))]; + tensor down_blocks_2_attentions_1_transformer_blocks_8_ff_net_2_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_8_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1587751488)))]; + tensor var_20118_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_8_ff_net_2_bias_to_fp16, dilations = var_20116, groups = var_2624, pad = var_20118_pad_0, pad_type = var_20118_pad_type_0, strides = var_20114, weight = down_blocks_2_attentions_1_transformer_blocks_8_ff_net_2_weight_to_fp16, x = input_299_cast_fp16)[name = tensor("op_20118_cast_fp16")]; + tensor inputs_139_cast_fp16 = add(x = var_20118_cast_fp16, y = inputs_137_cast_fp16)[name = tensor("inputs_139_cast_fp16")]; + tensor hidden_states_191_axes_0 = const()[name = tensor("hidden_states_191_axes_0"), val = tensor([1])]; + tensor hidden_states_191_gamma_0_to_fp16 = const()[name = tensor("hidden_states_191_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1587754112)))]; + tensor hidden_states_191_beta_0_to_fp16 = const()[name = tensor("hidden_states_191_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1587756736)))]; + tensor var_20134_to_fp16 = const()[name = tensor("op_20134_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_191_cast_fp16 = layer_norm(axes = hidden_states_191_axes_0, beta = hidden_states_191_beta_0_to_fp16, epsilon = var_20134_to_fp16, gamma = hidden_states_191_gamma_0_to_fp16, x = inputs_139_cast_fp16)[name = tensor("hidden_states_191_cast_fp16")]; + tensor var_20149 = const()[name = tensor("op_20149"), val = tensor([1, 1])]; + tensor var_20151 = const()[name = tensor("op_20151"), val = tensor([1, 1])]; + tensor q_93_pad_type_0 = const()[name = tensor("q_93_pad_type_0"), val = tensor("custom")]; + tensor q_93_pad_0 = const()[name = tensor("q_93_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_9_attn1_to_q_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_9_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1587759360)))]; + tensor q_93_cast_fp16 = conv(dilations = var_20151, groups = var_2624, pad = q_93_pad_0, pad_type = q_93_pad_type_0, strides = var_20149, weight = down_blocks_2_attentions_1_transformer_blocks_9_attn1_to_q_weight_to_fp16, x = hidden_states_191_cast_fp16)[name = tensor("q_93_cast_fp16")]; + tensor var_20155 = const()[name = tensor("op_20155"), val = tensor([1, 1])]; + tensor var_20157 = const()[name = tensor("op_20157"), val = tensor([1, 1])]; + tensor k_185_pad_type_0 = const()[name = tensor("k_185_pad_type_0"), val = tensor("custom")]; + tensor k_185_pad_0 = const()[name = tensor("k_185_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_9_attn1_to_k_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_9_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1591036224)))]; + tensor k_185_cast_fp16 = conv(dilations = var_20157, groups = var_2624, pad = k_185_pad_0, pad_type = k_185_pad_type_0, strides = var_20155, weight = down_blocks_2_attentions_1_transformer_blocks_9_attn1_to_k_weight_to_fp16, x = hidden_states_191_cast_fp16)[name = tensor("k_185_cast_fp16")]; + tensor var_20161 = const()[name = tensor("op_20161"), val = tensor([1, 1])]; + tensor var_20163 = const()[name = tensor("op_20163"), val = tensor([1, 1])]; + tensor v_93_pad_type_0 = const()[name = tensor("v_93_pad_type_0"), val = tensor("custom")]; + tensor v_93_pad_0 = const()[name = tensor("v_93_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_9_attn1_to_v_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_9_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1594313088)))]; + tensor v_93_cast_fp16 = conv(dilations = var_20163, groups = var_2624, pad = v_93_pad_0, pad_type = v_93_pad_type_0, strides = var_20161, weight = down_blocks_2_attentions_1_transformer_blocks_9_attn1_to_v_weight_to_fp16, x = hidden_states_191_cast_fp16)[name = tensor("v_93_cast_fp16")]; + tensor var_20167_begin_0 = const()[name = tensor("op_20167_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20167_end_0 = const()[name = tensor("op_20167_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_20167_end_mask_0 = const()[name = tensor("op_20167_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20167_cast_fp16 = slice_by_index(begin = var_20167_begin_0, end = var_20167_end_0, end_mask = var_20167_end_mask_0, x = q_93_cast_fp16)[name = tensor("op_20167_cast_fp16")]; + tensor var_20171_begin_0 = const()[name = tensor("op_20171_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_20171_end_0 = const()[name = tensor("op_20171_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_20171_end_mask_0 = const()[name = tensor("op_20171_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20171_cast_fp16 = slice_by_index(begin = var_20171_begin_0, end = var_20171_end_0, end_mask = var_20171_end_mask_0, x = q_93_cast_fp16)[name = tensor("op_20171_cast_fp16")]; + tensor var_20175_begin_0 = const()[name = tensor("op_20175_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_20175_end_0 = const()[name = tensor("op_20175_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_20175_end_mask_0 = const()[name = tensor("op_20175_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20175_cast_fp16 = slice_by_index(begin = var_20175_begin_0, end = var_20175_end_0, end_mask = var_20175_end_mask_0, x = q_93_cast_fp16)[name = tensor("op_20175_cast_fp16")]; + tensor var_20179_begin_0 = const()[name = tensor("op_20179_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_20179_end_0 = const()[name = tensor("op_20179_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_20179_end_mask_0 = const()[name = tensor("op_20179_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20179_cast_fp16 = slice_by_index(begin = var_20179_begin_0, end = var_20179_end_0, end_mask = var_20179_end_mask_0, x = q_93_cast_fp16)[name = tensor("op_20179_cast_fp16")]; + tensor var_20183_begin_0 = const()[name = tensor("op_20183_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_20183_end_0 = const()[name = tensor("op_20183_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_20183_end_mask_0 = const()[name = tensor("op_20183_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20183_cast_fp16 = slice_by_index(begin = var_20183_begin_0, end = var_20183_end_0, end_mask = var_20183_end_mask_0, x = q_93_cast_fp16)[name = tensor("op_20183_cast_fp16")]; + tensor var_20187_begin_0 = const()[name = tensor("op_20187_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_20187_end_0 = const()[name = tensor("op_20187_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_20187_end_mask_0 = const()[name = tensor("op_20187_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20187_cast_fp16 = slice_by_index(begin = var_20187_begin_0, end = var_20187_end_0, end_mask = var_20187_end_mask_0, x = q_93_cast_fp16)[name = tensor("op_20187_cast_fp16")]; + tensor var_20191_begin_0 = const()[name = tensor("op_20191_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_20191_end_0 = const()[name = tensor("op_20191_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_20191_end_mask_0 = const()[name = tensor("op_20191_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20191_cast_fp16 = slice_by_index(begin = var_20191_begin_0, end = var_20191_end_0, end_mask = var_20191_end_mask_0, x = q_93_cast_fp16)[name = tensor("op_20191_cast_fp16")]; + tensor var_20195_begin_0 = const()[name = tensor("op_20195_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_20195_end_0 = const()[name = tensor("op_20195_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_20195_end_mask_0 = const()[name = tensor("op_20195_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20195_cast_fp16 = slice_by_index(begin = var_20195_begin_0, end = var_20195_end_0, end_mask = var_20195_end_mask_0, x = q_93_cast_fp16)[name = tensor("op_20195_cast_fp16")]; + tensor var_20199_begin_0 = const()[name = tensor("op_20199_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_20199_end_0 = const()[name = tensor("op_20199_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_20199_end_mask_0 = const()[name = tensor("op_20199_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20199_cast_fp16 = slice_by_index(begin = var_20199_begin_0, end = var_20199_end_0, end_mask = var_20199_end_mask_0, x = q_93_cast_fp16)[name = tensor("op_20199_cast_fp16")]; + tensor var_20203_begin_0 = const()[name = tensor("op_20203_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_20203_end_0 = const()[name = tensor("op_20203_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_20203_end_mask_0 = const()[name = tensor("op_20203_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20203_cast_fp16 = slice_by_index(begin = var_20203_begin_0, end = var_20203_end_0, end_mask = var_20203_end_mask_0, x = q_93_cast_fp16)[name = tensor("op_20203_cast_fp16")]; + tensor var_20207_begin_0 = const()[name = tensor("op_20207_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_20207_end_0 = const()[name = tensor("op_20207_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_20207_end_mask_0 = const()[name = tensor("op_20207_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20207_cast_fp16 = slice_by_index(begin = var_20207_begin_0, end = var_20207_end_0, end_mask = var_20207_end_mask_0, x = q_93_cast_fp16)[name = tensor("op_20207_cast_fp16")]; + tensor var_20211_begin_0 = const()[name = tensor("op_20211_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_20211_end_0 = const()[name = tensor("op_20211_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_20211_end_mask_0 = const()[name = tensor("op_20211_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20211_cast_fp16 = slice_by_index(begin = var_20211_begin_0, end = var_20211_end_0, end_mask = var_20211_end_mask_0, x = q_93_cast_fp16)[name = tensor("op_20211_cast_fp16")]; + tensor var_20215_begin_0 = const()[name = tensor("op_20215_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_20215_end_0 = const()[name = tensor("op_20215_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_20215_end_mask_0 = const()[name = tensor("op_20215_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20215_cast_fp16 = slice_by_index(begin = var_20215_begin_0, end = var_20215_end_0, end_mask = var_20215_end_mask_0, x = q_93_cast_fp16)[name = tensor("op_20215_cast_fp16")]; + tensor var_20219_begin_0 = const()[name = tensor("op_20219_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_20219_end_0 = const()[name = tensor("op_20219_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_20219_end_mask_0 = const()[name = tensor("op_20219_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20219_cast_fp16 = slice_by_index(begin = var_20219_begin_0, end = var_20219_end_0, end_mask = var_20219_end_mask_0, x = q_93_cast_fp16)[name = tensor("op_20219_cast_fp16")]; + tensor var_20223_begin_0 = const()[name = tensor("op_20223_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_20223_end_0 = const()[name = tensor("op_20223_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_20223_end_mask_0 = const()[name = tensor("op_20223_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20223_cast_fp16 = slice_by_index(begin = var_20223_begin_0, end = var_20223_end_0, end_mask = var_20223_end_mask_0, x = q_93_cast_fp16)[name = tensor("op_20223_cast_fp16")]; + tensor var_20227_begin_0 = const()[name = tensor("op_20227_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_20227_end_0 = const()[name = tensor("op_20227_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_20227_end_mask_0 = const()[name = tensor("op_20227_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20227_cast_fp16 = slice_by_index(begin = var_20227_begin_0, end = var_20227_end_0, end_mask = var_20227_end_mask_0, x = q_93_cast_fp16)[name = tensor("op_20227_cast_fp16")]; + tensor var_20231_begin_0 = const()[name = tensor("op_20231_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_20231_end_0 = const()[name = tensor("op_20231_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_20231_end_mask_0 = const()[name = tensor("op_20231_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20231_cast_fp16 = slice_by_index(begin = var_20231_begin_0, end = var_20231_end_0, end_mask = var_20231_end_mask_0, x = q_93_cast_fp16)[name = tensor("op_20231_cast_fp16")]; + tensor var_20235_begin_0 = const()[name = tensor("op_20235_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_20235_end_0 = const()[name = tensor("op_20235_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_20235_end_mask_0 = const()[name = tensor("op_20235_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20235_cast_fp16 = slice_by_index(begin = var_20235_begin_0, end = var_20235_end_0, end_mask = var_20235_end_mask_0, x = q_93_cast_fp16)[name = tensor("op_20235_cast_fp16")]; + tensor var_20239_begin_0 = const()[name = tensor("op_20239_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_20239_end_0 = const()[name = tensor("op_20239_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_20239_end_mask_0 = const()[name = tensor("op_20239_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20239_cast_fp16 = slice_by_index(begin = var_20239_begin_0, end = var_20239_end_0, end_mask = var_20239_end_mask_0, x = q_93_cast_fp16)[name = tensor("op_20239_cast_fp16")]; + tensor var_20243_begin_0 = const()[name = tensor("op_20243_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_20243_end_0 = const()[name = tensor("op_20243_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_20243_end_mask_0 = const()[name = tensor("op_20243_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20243_cast_fp16 = slice_by_index(begin = var_20243_begin_0, end = var_20243_end_0, end_mask = var_20243_end_mask_0, x = q_93_cast_fp16)[name = tensor("op_20243_cast_fp16")]; + tensor k_187_perm_0 = const()[name = tensor("k_187_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_20250_begin_0 = const()[name = tensor("op_20250_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20250_end_0 = const()[name = tensor("op_20250_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_20250_end_mask_0 = const()[name = tensor("op_20250_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_187_cast_fp16 = transpose(perm = k_187_perm_0, x = k_185_cast_fp16)[name = tensor("transpose_93")]; + tensor var_20250_cast_fp16 = slice_by_index(begin = var_20250_begin_0, end = var_20250_end_0, end_mask = var_20250_end_mask_0, x = k_187_cast_fp16)[name = tensor("op_20250_cast_fp16")]; + tensor var_20254_begin_0 = const()[name = tensor("op_20254_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_20254_end_0 = const()[name = tensor("op_20254_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_20254_end_mask_0 = const()[name = tensor("op_20254_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20254_cast_fp16 = slice_by_index(begin = var_20254_begin_0, end = var_20254_end_0, end_mask = var_20254_end_mask_0, x = k_187_cast_fp16)[name = tensor("op_20254_cast_fp16")]; + tensor var_20258_begin_0 = const()[name = tensor("op_20258_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_20258_end_0 = const()[name = tensor("op_20258_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_20258_end_mask_0 = const()[name = tensor("op_20258_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20258_cast_fp16 = slice_by_index(begin = var_20258_begin_0, end = var_20258_end_0, end_mask = var_20258_end_mask_0, x = k_187_cast_fp16)[name = tensor("op_20258_cast_fp16")]; + tensor var_20262_begin_0 = const()[name = tensor("op_20262_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_20262_end_0 = const()[name = tensor("op_20262_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_20262_end_mask_0 = const()[name = tensor("op_20262_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20262_cast_fp16 = slice_by_index(begin = var_20262_begin_0, end = var_20262_end_0, end_mask = var_20262_end_mask_0, x = k_187_cast_fp16)[name = tensor("op_20262_cast_fp16")]; + tensor var_20266_begin_0 = const()[name = tensor("op_20266_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_20266_end_0 = const()[name = tensor("op_20266_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_20266_end_mask_0 = const()[name = tensor("op_20266_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20266_cast_fp16 = slice_by_index(begin = var_20266_begin_0, end = var_20266_end_0, end_mask = var_20266_end_mask_0, x = k_187_cast_fp16)[name = tensor("op_20266_cast_fp16")]; + tensor var_20270_begin_0 = const()[name = tensor("op_20270_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_20270_end_0 = const()[name = tensor("op_20270_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_20270_end_mask_0 = const()[name = tensor("op_20270_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20270_cast_fp16 = slice_by_index(begin = var_20270_begin_0, end = var_20270_end_0, end_mask = var_20270_end_mask_0, x = k_187_cast_fp16)[name = tensor("op_20270_cast_fp16")]; + tensor var_20274_begin_0 = const()[name = tensor("op_20274_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_20274_end_0 = const()[name = tensor("op_20274_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_20274_end_mask_0 = const()[name = tensor("op_20274_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20274_cast_fp16 = slice_by_index(begin = var_20274_begin_0, end = var_20274_end_0, end_mask = var_20274_end_mask_0, x = k_187_cast_fp16)[name = tensor("op_20274_cast_fp16")]; + tensor var_20278_begin_0 = const()[name = tensor("op_20278_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_20278_end_0 = const()[name = tensor("op_20278_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_20278_end_mask_0 = const()[name = tensor("op_20278_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20278_cast_fp16 = slice_by_index(begin = var_20278_begin_0, end = var_20278_end_0, end_mask = var_20278_end_mask_0, x = k_187_cast_fp16)[name = tensor("op_20278_cast_fp16")]; + tensor var_20282_begin_0 = const()[name = tensor("op_20282_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_20282_end_0 = const()[name = tensor("op_20282_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_20282_end_mask_0 = const()[name = tensor("op_20282_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20282_cast_fp16 = slice_by_index(begin = var_20282_begin_0, end = var_20282_end_0, end_mask = var_20282_end_mask_0, x = k_187_cast_fp16)[name = tensor("op_20282_cast_fp16")]; + tensor var_20286_begin_0 = const()[name = tensor("op_20286_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_20286_end_0 = const()[name = tensor("op_20286_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_20286_end_mask_0 = const()[name = tensor("op_20286_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20286_cast_fp16 = slice_by_index(begin = var_20286_begin_0, end = var_20286_end_0, end_mask = var_20286_end_mask_0, x = k_187_cast_fp16)[name = tensor("op_20286_cast_fp16")]; + tensor var_20290_begin_0 = const()[name = tensor("op_20290_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_20290_end_0 = const()[name = tensor("op_20290_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_20290_end_mask_0 = const()[name = tensor("op_20290_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20290_cast_fp16 = slice_by_index(begin = var_20290_begin_0, end = var_20290_end_0, end_mask = var_20290_end_mask_0, x = k_187_cast_fp16)[name = tensor("op_20290_cast_fp16")]; + tensor var_20294_begin_0 = const()[name = tensor("op_20294_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_20294_end_0 = const()[name = tensor("op_20294_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_20294_end_mask_0 = const()[name = tensor("op_20294_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20294_cast_fp16 = slice_by_index(begin = var_20294_begin_0, end = var_20294_end_0, end_mask = var_20294_end_mask_0, x = k_187_cast_fp16)[name = tensor("op_20294_cast_fp16")]; + tensor var_20298_begin_0 = const()[name = tensor("op_20298_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_20298_end_0 = const()[name = tensor("op_20298_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_20298_end_mask_0 = const()[name = tensor("op_20298_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20298_cast_fp16 = slice_by_index(begin = var_20298_begin_0, end = var_20298_end_0, end_mask = var_20298_end_mask_0, x = k_187_cast_fp16)[name = tensor("op_20298_cast_fp16")]; + tensor var_20302_begin_0 = const()[name = tensor("op_20302_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_20302_end_0 = const()[name = tensor("op_20302_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_20302_end_mask_0 = const()[name = tensor("op_20302_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20302_cast_fp16 = slice_by_index(begin = var_20302_begin_0, end = var_20302_end_0, end_mask = var_20302_end_mask_0, x = k_187_cast_fp16)[name = tensor("op_20302_cast_fp16")]; + tensor var_20306_begin_0 = const()[name = tensor("op_20306_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_20306_end_0 = const()[name = tensor("op_20306_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_20306_end_mask_0 = const()[name = tensor("op_20306_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20306_cast_fp16 = slice_by_index(begin = var_20306_begin_0, end = var_20306_end_0, end_mask = var_20306_end_mask_0, x = k_187_cast_fp16)[name = tensor("op_20306_cast_fp16")]; + tensor var_20310_begin_0 = const()[name = tensor("op_20310_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_20310_end_0 = const()[name = tensor("op_20310_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_20310_end_mask_0 = const()[name = tensor("op_20310_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20310_cast_fp16 = slice_by_index(begin = var_20310_begin_0, end = var_20310_end_0, end_mask = var_20310_end_mask_0, x = k_187_cast_fp16)[name = tensor("op_20310_cast_fp16")]; + tensor var_20314_begin_0 = const()[name = tensor("op_20314_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_20314_end_0 = const()[name = tensor("op_20314_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_20314_end_mask_0 = const()[name = tensor("op_20314_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20314_cast_fp16 = slice_by_index(begin = var_20314_begin_0, end = var_20314_end_0, end_mask = var_20314_end_mask_0, x = k_187_cast_fp16)[name = tensor("op_20314_cast_fp16")]; + tensor var_20318_begin_0 = const()[name = tensor("op_20318_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_20318_end_0 = const()[name = tensor("op_20318_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_20318_end_mask_0 = const()[name = tensor("op_20318_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20318_cast_fp16 = slice_by_index(begin = var_20318_begin_0, end = var_20318_end_0, end_mask = var_20318_end_mask_0, x = k_187_cast_fp16)[name = tensor("op_20318_cast_fp16")]; + tensor var_20322_begin_0 = const()[name = tensor("op_20322_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_20322_end_0 = const()[name = tensor("op_20322_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_20322_end_mask_0 = const()[name = tensor("op_20322_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20322_cast_fp16 = slice_by_index(begin = var_20322_begin_0, end = var_20322_end_0, end_mask = var_20322_end_mask_0, x = k_187_cast_fp16)[name = tensor("op_20322_cast_fp16")]; + tensor var_20326_begin_0 = const()[name = tensor("op_20326_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_20326_end_0 = const()[name = tensor("op_20326_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_20326_end_mask_0 = const()[name = tensor("op_20326_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20326_cast_fp16 = slice_by_index(begin = var_20326_begin_0, end = var_20326_end_0, end_mask = var_20326_end_mask_0, x = k_187_cast_fp16)[name = tensor("op_20326_cast_fp16")]; + tensor var_20328_begin_0 = const()[name = tensor("op_20328_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20328_end_0 = const()[name = tensor("op_20328_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_20328_end_mask_0 = const()[name = tensor("op_20328_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20328_cast_fp16 = slice_by_index(begin = var_20328_begin_0, end = var_20328_end_0, end_mask = var_20328_end_mask_0, x = v_93_cast_fp16)[name = tensor("op_20328_cast_fp16")]; + tensor var_20332_begin_0 = const()[name = tensor("op_20332_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_20332_end_0 = const()[name = tensor("op_20332_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_20332_end_mask_0 = const()[name = tensor("op_20332_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20332_cast_fp16 = slice_by_index(begin = var_20332_begin_0, end = var_20332_end_0, end_mask = var_20332_end_mask_0, x = v_93_cast_fp16)[name = tensor("op_20332_cast_fp16")]; + tensor var_20336_begin_0 = const()[name = tensor("op_20336_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_20336_end_0 = const()[name = tensor("op_20336_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_20336_end_mask_0 = const()[name = tensor("op_20336_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20336_cast_fp16 = slice_by_index(begin = var_20336_begin_0, end = var_20336_end_0, end_mask = var_20336_end_mask_0, x = v_93_cast_fp16)[name = tensor("op_20336_cast_fp16")]; + tensor var_20340_begin_0 = const()[name = tensor("op_20340_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_20340_end_0 = const()[name = tensor("op_20340_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_20340_end_mask_0 = const()[name = tensor("op_20340_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20340_cast_fp16 = slice_by_index(begin = var_20340_begin_0, end = var_20340_end_0, end_mask = var_20340_end_mask_0, x = v_93_cast_fp16)[name = tensor("op_20340_cast_fp16")]; + tensor var_20344_begin_0 = const()[name = tensor("op_20344_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_20344_end_0 = const()[name = tensor("op_20344_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_20344_end_mask_0 = const()[name = tensor("op_20344_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20344_cast_fp16 = slice_by_index(begin = var_20344_begin_0, end = var_20344_end_0, end_mask = var_20344_end_mask_0, x = v_93_cast_fp16)[name = tensor("op_20344_cast_fp16")]; + tensor var_20348_begin_0 = const()[name = tensor("op_20348_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_20348_end_0 = const()[name = tensor("op_20348_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_20348_end_mask_0 = const()[name = tensor("op_20348_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20348_cast_fp16 = slice_by_index(begin = var_20348_begin_0, end = var_20348_end_0, end_mask = var_20348_end_mask_0, x = v_93_cast_fp16)[name = tensor("op_20348_cast_fp16")]; + tensor var_20352_begin_0 = const()[name = tensor("op_20352_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_20352_end_0 = const()[name = tensor("op_20352_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_20352_end_mask_0 = const()[name = tensor("op_20352_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20352_cast_fp16 = slice_by_index(begin = var_20352_begin_0, end = var_20352_end_0, end_mask = var_20352_end_mask_0, x = v_93_cast_fp16)[name = tensor("op_20352_cast_fp16")]; + tensor var_20356_begin_0 = const()[name = tensor("op_20356_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_20356_end_0 = const()[name = tensor("op_20356_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_20356_end_mask_0 = const()[name = tensor("op_20356_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20356_cast_fp16 = slice_by_index(begin = var_20356_begin_0, end = var_20356_end_0, end_mask = var_20356_end_mask_0, x = v_93_cast_fp16)[name = tensor("op_20356_cast_fp16")]; + tensor var_20360_begin_0 = const()[name = tensor("op_20360_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_20360_end_0 = const()[name = tensor("op_20360_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_20360_end_mask_0 = const()[name = tensor("op_20360_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20360_cast_fp16 = slice_by_index(begin = var_20360_begin_0, end = var_20360_end_0, end_mask = var_20360_end_mask_0, x = v_93_cast_fp16)[name = tensor("op_20360_cast_fp16")]; + tensor var_20364_begin_0 = const()[name = tensor("op_20364_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_20364_end_0 = const()[name = tensor("op_20364_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_20364_end_mask_0 = const()[name = tensor("op_20364_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20364_cast_fp16 = slice_by_index(begin = var_20364_begin_0, end = var_20364_end_0, end_mask = var_20364_end_mask_0, x = v_93_cast_fp16)[name = tensor("op_20364_cast_fp16")]; + tensor var_20368_begin_0 = const()[name = tensor("op_20368_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_20368_end_0 = const()[name = tensor("op_20368_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_20368_end_mask_0 = const()[name = tensor("op_20368_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20368_cast_fp16 = slice_by_index(begin = var_20368_begin_0, end = var_20368_end_0, end_mask = var_20368_end_mask_0, x = v_93_cast_fp16)[name = tensor("op_20368_cast_fp16")]; + tensor var_20372_begin_0 = const()[name = tensor("op_20372_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_20372_end_0 = const()[name = tensor("op_20372_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_20372_end_mask_0 = const()[name = tensor("op_20372_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20372_cast_fp16 = slice_by_index(begin = var_20372_begin_0, end = var_20372_end_0, end_mask = var_20372_end_mask_0, x = v_93_cast_fp16)[name = tensor("op_20372_cast_fp16")]; + tensor var_20376_begin_0 = const()[name = tensor("op_20376_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_20376_end_0 = const()[name = tensor("op_20376_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_20376_end_mask_0 = const()[name = tensor("op_20376_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20376_cast_fp16 = slice_by_index(begin = var_20376_begin_0, end = var_20376_end_0, end_mask = var_20376_end_mask_0, x = v_93_cast_fp16)[name = tensor("op_20376_cast_fp16")]; + tensor var_20380_begin_0 = const()[name = tensor("op_20380_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_20380_end_0 = const()[name = tensor("op_20380_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_20380_end_mask_0 = const()[name = tensor("op_20380_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20380_cast_fp16 = slice_by_index(begin = var_20380_begin_0, end = var_20380_end_0, end_mask = var_20380_end_mask_0, x = v_93_cast_fp16)[name = tensor("op_20380_cast_fp16")]; + tensor var_20384_begin_0 = const()[name = tensor("op_20384_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_20384_end_0 = const()[name = tensor("op_20384_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_20384_end_mask_0 = const()[name = tensor("op_20384_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20384_cast_fp16 = slice_by_index(begin = var_20384_begin_0, end = var_20384_end_0, end_mask = var_20384_end_mask_0, x = v_93_cast_fp16)[name = tensor("op_20384_cast_fp16")]; + tensor var_20388_begin_0 = const()[name = tensor("op_20388_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_20388_end_0 = const()[name = tensor("op_20388_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_20388_end_mask_0 = const()[name = tensor("op_20388_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20388_cast_fp16 = slice_by_index(begin = var_20388_begin_0, end = var_20388_end_0, end_mask = var_20388_end_mask_0, x = v_93_cast_fp16)[name = tensor("op_20388_cast_fp16")]; + tensor var_20392_begin_0 = const()[name = tensor("op_20392_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_20392_end_0 = const()[name = tensor("op_20392_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_20392_end_mask_0 = const()[name = tensor("op_20392_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20392_cast_fp16 = slice_by_index(begin = var_20392_begin_0, end = var_20392_end_0, end_mask = var_20392_end_mask_0, x = v_93_cast_fp16)[name = tensor("op_20392_cast_fp16")]; + tensor var_20396_begin_0 = const()[name = tensor("op_20396_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_20396_end_0 = const()[name = tensor("op_20396_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_20396_end_mask_0 = const()[name = tensor("op_20396_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20396_cast_fp16 = slice_by_index(begin = var_20396_begin_0, end = var_20396_end_0, end_mask = var_20396_end_mask_0, x = v_93_cast_fp16)[name = tensor("op_20396_cast_fp16")]; + tensor var_20400_begin_0 = const()[name = tensor("op_20400_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_20400_end_0 = const()[name = tensor("op_20400_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_20400_end_mask_0 = const()[name = tensor("op_20400_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20400_cast_fp16 = slice_by_index(begin = var_20400_begin_0, end = var_20400_end_0, end_mask = var_20400_end_mask_0, x = v_93_cast_fp16)[name = tensor("op_20400_cast_fp16")]; + tensor var_20404_begin_0 = const()[name = tensor("op_20404_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_20404_end_0 = const()[name = tensor("op_20404_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_20404_end_mask_0 = const()[name = tensor("op_20404_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20404_cast_fp16 = slice_by_index(begin = var_20404_begin_0, end = var_20404_end_0, end_mask = var_20404_end_mask_0, x = v_93_cast_fp16)[name = tensor("op_20404_cast_fp16")]; + tensor var_20408_equation_0 = const()[name = tensor("op_20408_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20408_cast_fp16 = einsum(equation = var_20408_equation_0, values = (var_20250_cast_fp16, var_20167_cast_fp16))[name = tensor("op_20408_cast_fp16")]; + tensor var_20409_to_fp16 = const()[name = tensor("op_20409_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1681_cast_fp16 = mul(x = var_20408_cast_fp16, y = var_20409_to_fp16)[name = tensor("aw_1681_cast_fp16")]; + tensor var_20412_equation_0 = const()[name = tensor("op_20412_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20412_cast_fp16 = einsum(equation = var_20412_equation_0, values = (var_20254_cast_fp16, var_20171_cast_fp16))[name = tensor("op_20412_cast_fp16")]; + tensor var_20413_to_fp16 = const()[name = tensor("op_20413_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1683_cast_fp16 = mul(x = var_20412_cast_fp16, y = var_20413_to_fp16)[name = tensor("aw_1683_cast_fp16")]; + tensor var_20416_equation_0 = const()[name = tensor("op_20416_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20416_cast_fp16 = einsum(equation = var_20416_equation_0, values = (var_20258_cast_fp16, var_20175_cast_fp16))[name = tensor("op_20416_cast_fp16")]; + tensor var_20417_to_fp16 = const()[name = tensor("op_20417_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1685_cast_fp16 = mul(x = var_20416_cast_fp16, y = var_20417_to_fp16)[name = tensor("aw_1685_cast_fp16")]; + tensor var_20420_equation_0 = const()[name = tensor("op_20420_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20420_cast_fp16 = einsum(equation = var_20420_equation_0, values = (var_20262_cast_fp16, var_20179_cast_fp16))[name = tensor("op_20420_cast_fp16")]; + tensor var_20421_to_fp16 = const()[name = tensor("op_20421_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1687_cast_fp16 = mul(x = var_20420_cast_fp16, y = var_20421_to_fp16)[name = tensor("aw_1687_cast_fp16")]; + tensor var_20424_equation_0 = const()[name = tensor("op_20424_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20424_cast_fp16 = einsum(equation = var_20424_equation_0, values = (var_20266_cast_fp16, var_20183_cast_fp16))[name = tensor("op_20424_cast_fp16")]; + tensor var_20425_to_fp16 = const()[name = tensor("op_20425_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1689_cast_fp16 = mul(x = var_20424_cast_fp16, y = var_20425_to_fp16)[name = tensor("aw_1689_cast_fp16")]; + tensor var_20428_equation_0 = const()[name = tensor("op_20428_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20428_cast_fp16 = einsum(equation = var_20428_equation_0, values = (var_20270_cast_fp16, var_20187_cast_fp16))[name = tensor("op_20428_cast_fp16")]; + tensor var_20429_to_fp16 = const()[name = tensor("op_20429_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1691_cast_fp16 = mul(x = var_20428_cast_fp16, y = var_20429_to_fp16)[name = tensor("aw_1691_cast_fp16")]; + tensor var_20432_equation_0 = const()[name = tensor("op_20432_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20432_cast_fp16 = einsum(equation = var_20432_equation_0, values = (var_20274_cast_fp16, var_20191_cast_fp16))[name = tensor("op_20432_cast_fp16")]; + tensor var_20433_to_fp16 = const()[name = tensor("op_20433_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1693_cast_fp16 = mul(x = var_20432_cast_fp16, y = var_20433_to_fp16)[name = tensor("aw_1693_cast_fp16")]; + tensor var_20436_equation_0 = const()[name = tensor("op_20436_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20436_cast_fp16 = einsum(equation = var_20436_equation_0, values = (var_20278_cast_fp16, var_20195_cast_fp16))[name = tensor("op_20436_cast_fp16")]; + tensor var_20437_to_fp16 = const()[name = tensor("op_20437_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1695_cast_fp16 = mul(x = var_20436_cast_fp16, y = var_20437_to_fp16)[name = tensor("aw_1695_cast_fp16")]; + tensor var_20440_equation_0 = const()[name = tensor("op_20440_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20440_cast_fp16 = einsum(equation = var_20440_equation_0, values = (var_20282_cast_fp16, var_20199_cast_fp16))[name = tensor("op_20440_cast_fp16")]; + tensor var_20441_to_fp16 = const()[name = tensor("op_20441_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1697_cast_fp16 = mul(x = var_20440_cast_fp16, y = var_20441_to_fp16)[name = tensor("aw_1697_cast_fp16")]; + tensor var_20444_equation_0 = const()[name = tensor("op_20444_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20444_cast_fp16 = einsum(equation = var_20444_equation_0, values = (var_20286_cast_fp16, var_20203_cast_fp16))[name = tensor("op_20444_cast_fp16")]; + tensor var_20445_to_fp16 = const()[name = tensor("op_20445_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1699_cast_fp16 = mul(x = var_20444_cast_fp16, y = var_20445_to_fp16)[name = tensor("aw_1699_cast_fp16")]; + tensor var_20448_equation_0 = const()[name = tensor("op_20448_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20448_cast_fp16 = einsum(equation = var_20448_equation_0, values = (var_20290_cast_fp16, var_20207_cast_fp16))[name = tensor("op_20448_cast_fp16")]; + tensor var_20449_to_fp16 = const()[name = tensor("op_20449_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1701_cast_fp16 = mul(x = var_20448_cast_fp16, y = var_20449_to_fp16)[name = tensor("aw_1701_cast_fp16")]; + tensor var_20452_equation_0 = const()[name = tensor("op_20452_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20452_cast_fp16 = einsum(equation = var_20452_equation_0, values = (var_20294_cast_fp16, var_20211_cast_fp16))[name = tensor("op_20452_cast_fp16")]; + tensor var_20453_to_fp16 = const()[name = tensor("op_20453_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1703_cast_fp16 = mul(x = var_20452_cast_fp16, y = var_20453_to_fp16)[name = tensor("aw_1703_cast_fp16")]; + tensor var_20456_equation_0 = const()[name = tensor("op_20456_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20456_cast_fp16 = einsum(equation = var_20456_equation_0, values = (var_20298_cast_fp16, var_20215_cast_fp16))[name = tensor("op_20456_cast_fp16")]; + tensor var_20457_to_fp16 = const()[name = tensor("op_20457_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1705_cast_fp16 = mul(x = var_20456_cast_fp16, y = var_20457_to_fp16)[name = tensor("aw_1705_cast_fp16")]; + tensor var_20460_equation_0 = const()[name = tensor("op_20460_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20460_cast_fp16 = einsum(equation = var_20460_equation_0, values = (var_20302_cast_fp16, var_20219_cast_fp16))[name = tensor("op_20460_cast_fp16")]; + tensor var_20461_to_fp16 = const()[name = tensor("op_20461_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1707_cast_fp16 = mul(x = var_20460_cast_fp16, y = var_20461_to_fp16)[name = tensor("aw_1707_cast_fp16")]; + tensor var_20464_equation_0 = const()[name = tensor("op_20464_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20464_cast_fp16 = einsum(equation = var_20464_equation_0, values = (var_20306_cast_fp16, var_20223_cast_fp16))[name = tensor("op_20464_cast_fp16")]; + tensor var_20465_to_fp16 = const()[name = tensor("op_20465_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1709_cast_fp16 = mul(x = var_20464_cast_fp16, y = var_20465_to_fp16)[name = tensor("aw_1709_cast_fp16")]; + tensor var_20468_equation_0 = const()[name = tensor("op_20468_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20468_cast_fp16 = einsum(equation = var_20468_equation_0, values = (var_20310_cast_fp16, var_20227_cast_fp16))[name = tensor("op_20468_cast_fp16")]; + tensor var_20469_to_fp16 = const()[name = tensor("op_20469_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1711_cast_fp16 = mul(x = var_20468_cast_fp16, y = var_20469_to_fp16)[name = tensor("aw_1711_cast_fp16")]; + tensor var_20472_equation_0 = const()[name = tensor("op_20472_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20472_cast_fp16 = einsum(equation = var_20472_equation_0, values = (var_20314_cast_fp16, var_20231_cast_fp16))[name = tensor("op_20472_cast_fp16")]; + tensor var_20473_to_fp16 = const()[name = tensor("op_20473_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1713_cast_fp16 = mul(x = var_20472_cast_fp16, y = var_20473_to_fp16)[name = tensor("aw_1713_cast_fp16")]; + tensor var_20476_equation_0 = const()[name = tensor("op_20476_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20476_cast_fp16 = einsum(equation = var_20476_equation_0, values = (var_20318_cast_fp16, var_20235_cast_fp16))[name = tensor("op_20476_cast_fp16")]; + tensor var_20477_to_fp16 = const()[name = tensor("op_20477_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1715_cast_fp16 = mul(x = var_20476_cast_fp16, y = var_20477_to_fp16)[name = tensor("aw_1715_cast_fp16")]; + tensor var_20480_equation_0 = const()[name = tensor("op_20480_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20480_cast_fp16 = einsum(equation = var_20480_equation_0, values = (var_20322_cast_fp16, var_20239_cast_fp16))[name = tensor("op_20480_cast_fp16")]; + tensor var_20481_to_fp16 = const()[name = tensor("op_20481_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1717_cast_fp16 = mul(x = var_20480_cast_fp16, y = var_20481_to_fp16)[name = tensor("aw_1717_cast_fp16")]; + tensor var_20484_equation_0 = const()[name = tensor("op_20484_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20484_cast_fp16 = einsum(equation = var_20484_equation_0, values = (var_20326_cast_fp16, var_20243_cast_fp16))[name = tensor("op_20484_cast_fp16")]; + tensor var_20485_to_fp16 = const()[name = tensor("op_20485_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1719_cast_fp16 = mul(x = var_20484_cast_fp16, y = var_20485_to_fp16)[name = tensor("aw_1719_cast_fp16")]; + tensor var_20487_cast_fp16 = softmax(axis = var_2624, x = aw_1681_cast_fp16)[name = tensor("op_20487_cast_fp16")]; + tensor var_20488_cast_fp16 = softmax(axis = var_2624, x = aw_1683_cast_fp16)[name = tensor("op_20488_cast_fp16")]; + tensor var_20489_cast_fp16 = softmax(axis = var_2624, x = aw_1685_cast_fp16)[name = tensor("op_20489_cast_fp16")]; + tensor var_20490_cast_fp16 = softmax(axis = var_2624, x = aw_1687_cast_fp16)[name = tensor("op_20490_cast_fp16")]; + tensor var_20491_cast_fp16 = softmax(axis = var_2624, x = aw_1689_cast_fp16)[name = tensor("op_20491_cast_fp16")]; + tensor var_20492_cast_fp16 = softmax(axis = var_2624, x = aw_1691_cast_fp16)[name = tensor("op_20492_cast_fp16")]; + tensor var_20493_cast_fp16 = softmax(axis = var_2624, x = aw_1693_cast_fp16)[name = tensor("op_20493_cast_fp16")]; + tensor var_20494_cast_fp16 = softmax(axis = var_2624, x = aw_1695_cast_fp16)[name = tensor("op_20494_cast_fp16")]; + tensor var_20495_cast_fp16 = softmax(axis = var_2624, x = aw_1697_cast_fp16)[name = tensor("op_20495_cast_fp16")]; + tensor var_20496_cast_fp16 = softmax(axis = var_2624, x = aw_1699_cast_fp16)[name = tensor("op_20496_cast_fp16")]; + tensor var_20497_cast_fp16 = softmax(axis = var_2624, x = aw_1701_cast_fp16)[name = tensor("op_20497_cast_fp16")]; + tensor var_20498_cast_fp16 = softmax(axis = var_2624, x = aw_1703_cast_fp16)[name = tensor("op_20498_cast_fp16")]; + tensor var_20499_cast_fp16 = softmax(axis = var_2624, x = aw_1705_cast_fp16)[name = tensor("op_20499_cast_fp16")]; + tensor var_20500_cast_fp16 = softmax(axis = var_2624, x = aw_1707_cast_fp16)[name = tensor("op_20500_cast_fp16")]; + tensor var_20501_cast_fp16 = softmax(axis = var_2624, x = aw_1709_cast_fp16)[name = tensor("op_20501_cast_fp16")]; + tensor var_20502_cast_fp16 = softmax(axis = var_2624, x = aw_1711_cast_fp16)[name = tensor("op_20502_cast_fp16")]; + tensor var_20503_cast_fp16 = softmax(axis = var_2624, x = aw_1713_cast_fp16)[name = tensor("op_20503_cast_fp16")]; + tensor var_20504_cast_fp16 = softmax(axis = var_2624, x = aw_1715_cast_fp16)[name = tensor("op_20504_cast_fp16")]; + tensor var_20505_cast_fp16 = softmax(axis = var_2624, x = aw_1717_cast_fp16)[name = tensor("op_20505_cast_fp16")]; + tensor var_20506_cast_fp16 = softmax(axis = var_2624, x = aw_1719_cast_fp16)[name = tensor("op_20506_cast_fp16")]; + tensor var_20508_equation_0 = const()[name = tensor("op_20508_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20508_cast_fp16 = einsum(equation = var_20508_equation_0, values = (var_20328_cast_fp16, var_20487_cast_fp16))[name = tensor("op_20508_cast_fp16")]; + tensor var_20510_equation_0 = const()[name = tensor("op_20510_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20510_cast_fp16 = einsum(equation = var_20510_equation_0, values = (var_20332_cast_fp16, var_20488_cast_fp16))[name = tensor("op_20510_cast_fp16")]; + tensor var_20512_equation_0 = const()[name = tensor("op_20512_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20512_cast_fp16 = einsum(equation = var_20512_equation_0, values = (var_20336_cast_fp16, var_20489_cast_fp16))[name = tensor("op_20512_cast_fp16")]; + tensor var_20514_equation_0 = const()[name = tensor("op_20514_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20514_cast_fp16 = einsum(equation = var_20514_equation_0, values = (var_20340_cast_fp16, var_20490_cast_fp16))[name = tensor("op_20514_cast_fp16")]; + tensor var_20516_equation_0 = const()[name = tensor("op_20516_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20516_cast_fp16 = einsum(equation = var_20516_equation_0, values = (var_20344_cast_fp16, var_20491_cast_fp16))[name = tensor("op_20516_cast_fp16")]; + tensor var_20518_equation_0 = const()[name = tensor("op_20518_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20518_cast_fp16 = einsum(equation = var_20518_equation_0, values = (var_20348_cast_fp16, var_20492_cast_fp16))[name = tensor("op_20518_cast_fp16")]; + tensor var_20520_equation_0 = const()[name = tensor("op_20520_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20520_cast_fp16 = einsum(equation = var_20520_equation_0, values = (var_20352_cast_fp16, var_20493_cast_fp16))[name = tensor("op_20520_cast_fp16")]; + tensor var_20522_equation_0 = const()[name = tensor("op_20522_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20522_cast_fp16 = einsum(equation = var_20522_equation_0, values = (var_20356_cast_fp16, var_20494_cast_fp16))[name = tensor("op_20522_cast_fp16")]; + tensor var_20524_equation_0 = const()[name = tensor("op_20524_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20524_cast_fp16 = einsum(equation = var_20524_equation_0, values = (var_20360_cast_fp16, var_20495_cast_fp16))[name = tensor("op_20524_cast_fp16")]; + tensor var_20526_equation_0 = const()[name = tensor("op_20526_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20526_cast_fp16 = einsum(equation = var_20526_equation_0, values = (var_20364_cast_fp16, var_20496_cast_fp16))[name = tensor("op_20526_cast_fp16")]; + tensor var_20528_equation_0 = const()[name = tensor("op_20528_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20528_cast_fp16 = einsum(equation = var_20528_equation_0, values = (var_20368_cast_fp16, var_20497_cast_fp16))[name = tensor("op_20528_cast_fp16")]; + tensor var_20530_equation_0 = const()[name = tensor("op_20530_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20530_cast_fp16 = einsum(equation = var_20530_equation_0, values = (var_20372_cast_fp16, var_20498_cast_fp16))[name = tensor("op_20530_cast_fp16")]; + tensor var_20532_equation_0 = const()[name = tensor("op_20532_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20532_cast_fp16 = einsum(equation = var_20532_equation_0, values = (var_20376_cast_fp16, var_20499_cast_fp16))[name = tensor("op_20532_cast_fp16")]; + tensor var_20534_equation_0 = const()[name = tensor("op_20534_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20534_cast_fp16 = einsum(equation = var_20534_equation_0, values = (var_20380_cast_fp16, var_20500_cast_fp16))[name = tensor("op_20534_cast_fp16")]; + tensor var_20536_equation_0 = const()[name = tensor("op_20536_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20536_cast_fp16 = einsum(equation = var_20536_equation_0, values = (var_20384_cast_fp16, var_20501_cast_fp16))[name = tensor("op_20536_cast_fp16")]; + tensor var_20538_equation_0 = const()[name = tensor("op_20538_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20538_cast_fp16 = einsum(equation = var_20538_equation_0, values = (var_20388_cast_fp16, var_20502_cast_fp16))[name = tensor("op_20538_cast_fp16")]; + tensor var_20540_equation_0 = const()[name = tensor("op_20540_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20540_cast_fp16 = einsum(equation = var_20540_equation_0, values = (var_20392_cast_fp16, var_20503_cast_fp16))[name = tensor("op_20540_cast_fp16")]; + tensor var_20542_equation_0 = const()[name = tensor("op_20542_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20542_cast_fp16 = einsum(equation = var_20542_equation_0, values = (var_20396_cast_fp16, var_20504_cast_fp16))[name = tensor("op_20542_cast_fp16")]; + tensor var_20544_equation_0 = const()[name = tensor("op_20544_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20544_cast_fp16 = einsum(equation = var_20544_equation_0, values = (var_20400_cast_fp16, var_20505_cast_fp16))[name = tensor("op_20544_cast_fp16")]; + tensor var_20546_equation_0 = const()[name = tensor("op_20546_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20546_cast_fp16 = einsum(equation = var_20546_equation_0, values = (var_20404_cast_fp16, var_20506_cast_fp16))[name = tensor("op_20546_cast_fp16")]; + tensor input_301_interleave_0 = const()[name = tensor("input_301_interleave_0"), val = tensor(false)]; + tensor input_301_cast_fp16 = concat(axis = var_2624, interleave = input_301_interleave_0, values = (var_20508_cast_fp16, var_20510_cast_fp16, var_20512_cast_fp16, var_20514_cast_fp16, var_20516_cast_fp16, var_20518_cast_fp16, var_20520_cast_fp16, var_20522_cast_fp16, var_20524_cast_fp16, var_20526_cast_fp16, var_20528_cast_fp16, var_20530_cast_fp16, var_20532_cast_fp16, var_20534_cast_fp16, var_20536_cast_fp16, var_20538_cast_fp16, var_20540_cast_fp16, var_20542_cast_fp16, var_20544_cast_fp16, var_20546_cast_fp16))[name = tensor("input_301_cast_fp16")]; + tensor var_20552 = const()[name = tensor("op_20552"), val = tensor([1, 1])]; + tensor var_20554 = const()[name = tensor("op_20554"), val = tensor([1, 1])]; + tensor var_20556_pad_type_0 = const()[name = tensor("op_20556_pad_type_0"), val = tensor("custom")]; + tensor var_20556_pad_0 = const()[name = tensor("op_20556_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_9_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_9_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1597589952)))]; + tensor down_blocks_2_attentions_1_transformer_blocks_9_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_9_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1600866816)))]; + tensor var_20556_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_9_attn1_to_out_0_bias_to_fp16, dilations = var_20554, groups = var_2624, pad = var_20556_pad_0, pad_type = var_20556_pad_type_0, strides = var_20552, weight = down_blocks_2_attentions_1_transformer_blocks_9_attn1_to_out_0_weight_to_fp16, x = input_301_cast_fp16)[name = tensor("op_20556_cast_fp16")]; + tensor inputs_141_cast_fp16 = add(x = var_20556_cast_fp16, y = inputs_139_cast_fp16)[name = tensor("inputs_141_cast_fp16")]; + tensor hidden_states_193_axes_0 = const()[name = tensor("hidden_states_193_axes_0"), val = tensor([1])]; + tensor hidden_states_193_gamma_0_to_fp16 = const()[name = tensor("hidden_states_193_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1600869440)))]; + tensor hidden_states_193_beta_0_to_fp16 = const()[name = tensor("hidden_states_193_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1600872064)))]; + tensor var_20566_to_fp16 = const()[name = tensor("op_20566_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_193_cast_fp16 = layer_norm(axes = hidden_states_193_axes_0, beta = hidden_states_193_beta_0_to_fp16, epsilon = var_20566_to_fp16, gamma = hidden_states_193_gamma_0_to_fp16, x = inputs_141_cast_fp16)[name = tensor("hidden_states_193_cast_fp16")]; + tensor var_20581 = const()[name = tensor("op_20581"), val = tensor([1, 1])]; + tensor var_20583 = const()[name = tensor("op_20583"), val = tensor([1, 1])]; + tensor q_95_pad_type_0 = const()[name = tensor("q_95_pad_type_0"), val = tensor("custom")]; + tensor q_95_pad_0 = const()[name = tensor("q_95_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_9_attn2_to_q_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_9_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1600874688)))]; + tensor q_95_cast_fp16 = conv(dilations = var_20583, groups = var_2624, pad = q_95_pad_0, pad_type = q_95_pad_type_0, strides = var_20581, weight = down_blocks_2_attentions_1_transformer_blocks_9_attn2_to_q_weight_to_fp16, x = hidden_states_193_cast_fp16)[name = tensor("q_95_cast_fp16")]; + tensor var_20587 = const()[name = tensor("op_20587"), val = tensor([1, 1])]; + tensor var_20589 = const()[name = tensor("op_20589"), val = tensor([1, 1])]; + tensor k_189_pad_type_0 = const()[name = tensor("k_189_pad_type_0"), val = tensor("custom")]; + tensor k_189_pad_0 = const()[name = tensor("k_189_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_9_attn2_to_k_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_9_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1604151552)))]; + tensor k_189_cast_fp16 = conv(dilations = var_20589, groups = var_2624, pad = k_189_pad_0, pad_type = k_189_pad_type_0, strides = var_20587, weight = down_blocks_2_attentions_1_transformer_blocks_9_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_189_cast_fp16")]; + tensor var_20593 = const()[name = tensor("op_20593"), val = tensor([1, 1])]; + tensor var_20595 = const()[name = tensor("op_20595"), val = tensor([1, 1])]; + tensor v_95_pad_type_0 = const()[name = tensor("v_95_pad_type_0"), val = tensor("custom")]; + tensor v_95_pad_0 = const()[name = tensor("v_95_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_9_attn2_to_v_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_9_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1609394496)))]; + tensor v_95_cast_fp16 = conv(dilations = var_20595, groups = var_2624, pad = v_95_pad_0, pad_type = v_95_pad_type_0, strides = var_20593, weight = down_blocks_2_attentions_1_transformer_blocks_9_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_95_cast_fp16")]; + tensor var_20599_begin_0 = const()[name = tensor("op_20599_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20599_end_0 = const()[name = tensor("op_20599_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_20599_end_mask_0 = const()[name = tensor("op_20599_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20599_cast_fp16 = slice_by_index(begin = var_20599_begin_0, end = var_20599_end_0, end_mask = var_20599_end_mask_0, x = q_95_cast_fp16)[name = tensor("op_20599_cast_fp16")]; + tensor var_20603_begin_0 = const()[name = tensor("op_20603_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_20603_end_0 = const()[name = tensor("op_20603_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_20603_end_mask_0 = const()[name = tensor("op_20603_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20603_cast_fp16 = slice_by_index(begin = var_20603_begin_0, end = var_20603_end_0, end_mask = var_20603_end_mask_0, x = q_95_cast_fp16)[name = tensor("op_20603_cast_fp16")]; + tensor var_20607_begin_0 = const()[name = tensor("op_20607_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_20607_end_0 = const()[name = tensor("op_20607_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_20607_end_mask_0 = const()[name = tensor("op_20607_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20607_cast_fp16 = slice_by_index(begin = var_20607_begin_0, end = var_20607_end_0, end_mask = var_20607_end_mask_0, x = q_95_cast_fp16)[name = tensor("op_20607_cast_fp16")]; + tensor var_20611_begin_0 = const()[name = tensor("op_20611_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_20611_end_0 = const()[name = tensor("op_20611_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_20611_end_mask_0 = const()[name = tensor("op_20611_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20611_cast_fp16 = slice_by_index(begin = var_20611_begin_0, end = var_20611_end_0, end_mask = var_20611_end_mask_0, x = q_95_cast_fp16)[name = tensor("op_20611_cast_fp16")]; + tensor var_20615_begin_0 = const()[name = tensor("op_20615_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_20615_end_0 = const()[name = tensor("op_20615_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_20615_end_mask_0 = const()[name = tensor("op_20615_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20615_cast_fp16 = slice_by_index(begin = var_20615_begin_0, end = var_20615_end_0, end_mask = var_20615_end_mask_0, x = q_95_cast_fp16)[name = tensor("op_20615_cast_fp16")]; + tensor var_20619_begin_0 = const()[name = tensor("op_20619_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_20619_end_0 = const()[name = tensor("op_20619_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_20619_end_mask_0 = const()[name = tensor("op_20619_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20619_cast_fp16 = slice_by_index(begin = var_20619_begin_0, end = var_20619_end_0, end_mask = var_20619_end_mask_0, x = q_95_cast_fp16)[name = tensor("op_20619_cast_fp16")]; + tensor var_20623_begin_0 = const()[name = tensor("op_20623_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_20623_end_0 = const()[name = tensor("op_20623_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_20623_end_mask_0 = const()[name = tensor("op_20623_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20623_cast_fp16 = slice_by_index(begin = var_20623_begin_0, end = var_20623_end_0, end_mask = var_20623_end_mask_0, x = q_95_cast_fp16)[name = tensor("op_20623_cast_fp16")]; + tensor var_20627_begin_0 = const()[name = tensor("op_20627_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_20627_end_0 = const()[name = tensor("op_20627_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_20627_end_mask_0 = const()[name = tensor("op_20627_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20627_cast_fp16 = slice_by_index(begin = var_20627_begin_0, end = var_20627_end_0, end_mask = var_20627_end_mask_0, x = q_95_cast_fp16)[name = tensor("op_20627_cast_fp16")]; + tensor var_20631_begin_0 = const()[name = tensor("op_20631_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_20631_end_0 = const()[name = tensor("op_20631_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_20631_end_mask_0 = const()[name = tensor("op_20631_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20631_cast_fp16 = slice_by_index(begin = var_20631_begin_0, end = var_20631_end_0, end_mask = var_20631_end_mask_0, x = q_95_cast_fp16)[name = tensor("op_20631_cast_fp16")]; + tensor var_20635_begin_0 = const()[name = tensor("op_20635_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_20635_end_0 = const()[name = tensor("op_20635_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_20635_end_mask_0 = const()[name = tensor("op_20635_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20635_cast_fp16 = slice_by_index(begin = var_20635_begin_0, end = var_20635_end_0, end_mask = var_20635_end_mask_0, x = q_95_cast_fp16)[name = tensor("op_20635_cast_fp16")]; + tensor var_20639_begin_0 = const()[name = tensor("op_20639_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_20639_end_0 = const()[name = tensor("op_20639_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_20639_end_mask_0 = const()[name = tensor("op_20639_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20639_cast_fp16 = slice_by_index(begin = var_20639_begin_0, end = var_20639_end_0, end_mask = var_20639_end_mask_0, x = q_95_cast_fp16)[name = tensor("op_20639_cast_fp16")]; + tensor var_20643_begin_0 = const()[name = tensor("op_20643_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_20643_end_0 = const()[name = tensor("op_20643_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_20643_end_mask_0 = const()[name = tensor("op_20643_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20643_cast_fp16 = slice_by_index(begin = var_20643_begin_0, end = var_20643_end_0, end_mask = var_20643_end_mask_0, x = q_95_cast_fp16)[name = tensor("op_20643_cast_fp16")]; + tensor var_20647_begin_0 = const()[name = tensor("op_20647_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_20647_end_0 = const()[name = tensor("op_20647_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_20647_end_mask_0 = const()[name = tensor("op_20647_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20647_cast_fp16 = slice_by_index(begin = var_20647_begin_0, end = var_20647_end_0, end_mask = var_20647_end_mask_0, x = q_95_cast_fp16)[name = tensor("op_20647_cast_fp16")]; + tensor var_20651_begin_0 = const()[name = tensor("op_20651_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_20651_end_0 = const()[name = tensor("op_20651_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_20651_end_mask_0 = const()[name = tensor("op_20651_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20651_cast_fp16 = slice_by_index(begin = var_20651_begin_0, end = var_20651_end_0, end_mask = var_20651_end_mask_0, x = q_95_cast_fp16)[name = tensor("op_20651_cast_fp16")]; + tensor var_20655_begin_0 = const()[name = tensor("op_20655_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_20655_end_0 = const()[name = tensor("op_20655_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_20655_end_mask_0 = const()[name = tensor("op_20655_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20655_cast_fp16 = slice_by_index(begin = var_20655_begin_0, end = var_20655_end_0, end_mask = var_20655_end_mask_0, x = q_95_cast_fp16)[name = tensor("op_20655_cast_fp16")]; + tensor var_20659_begin_0 = const()[name = tensor("op_20659_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_20659_end_0 = const()[name = tensor("op_20659_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_20659_end_mask_0 = const()[name = tensor("op_20659_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20659_cast_fp16 = slice_by_index(begin = var_20659_begin_0, end = var_20659_end_0, end_mask = var_20659_end_mask_0, x = q_95_cast_fp16)[name = tensor("op_20659_cast_fp16")]; + tensor var_20663_begin_0 = const()[name = tensor("op_20663_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_20663_end_0 = const()[name = tensor("op_20663_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_20663_end_mask_0 = const()[name = tensor("op_20663_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20663_cast_fp16 = slice_by_index(begin = var_20663_begin_0, end = var_20663_end_0, end_mask = var_20663_end_mask_0, x = q_95_cast_fp16)[name = tensor("op_20663_cast_fp16")]; + tensor var_20667_begin_0 = const()[name = tensor("op_20667_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_20667_end_0 = const()[name = tensor("op_20667_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_20667_end_mask_0 = const()[name = tensor("op_20667_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20667_cast_fp16 = slice_by_index(begin = var_20667_begin_0, end = var_20667_end_0, end_mask = var_20667_end_mask_0, x = q_95_cast_fp16)[name = tensor("op_20667_cast_fp16")]; + tensor var_20671_begin_0 = const()[name = tensor("op_20671_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_20671_end_0 = const()[name = tensor("op_20671_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_20671_end_mask_0 = const()[name = tensor("op_20671_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20671_cast_fp16 = slice_by_index(begin = var_20671_begin_0, end = var_20671_end_0, end_mask = var_20671_end_mask_0, x = q_95_cast_fp16)[name = tensor("op_20671_cast_fp16")]; + tensor var_20675_begin_0 = const()[name = tensor("op_20675_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_20675_end_0 = const()[name = tensor("op_20675_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_20675_end_mask_0 = const()[name = tensor("op_20675_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20675_cast_fp16 = slice_by_index(begin = var_20675_begin_0, end = var_20675_end_0, end_mask = var_20675_end_mask_0, x = q_95_cast_fp16)[name = tensor("op_20675_cast_fp16")]; + tensor k_191_perm_0 = const()[name = tensor("k_191_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_20682_begin_0 = const()[name = tensor("op_20682_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20682_end_0 = const()[name = tensor("op_20682_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_20682_end_mask_0 = const()[name = tensor("op_20682_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_191_cast_fp16 = transpose(perm = k_191_perm_0, x = k_189_cast_fp16)[name = tensor("transpose_92")]; + tensor var_20682_cast_fp16 = slice_by_index(begin = var_20682_begin_0, end = var_20682_end_0, end_mask = var_20682_end_mask_0, x = k_191_cast_fp16)[name = tensor("op_20682_cast_fp16")]; + tensor var_20686_begin_0 = const()[name = tensor("op_20686_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_20686_end_0 = const()[name = tensor("op_20686_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_20686_end_mask_0 = const()[name = tensor("op_20686_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20686_cast_fp16 = slice_by_index(begin = var_20686_begin_0, end = var_20686_end_0, end_mask = var_20686_end_mask_0, x = k_191_cast_fp16)[name = tensor("op_20686_cast_fp16")]; + tensor var_20690_begin_0 = const()[name = tensor("op_20690_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_20690_end_0 = const()[name = tensor("op_20690_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_20690_end_mask_0 = const()[name = tensor("op_20690_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20690_cast_fp16 = slice_by_index(begin = var_20690_begin_0, end = var_20690_end_0, end_mask = var_20690_end_mask_0, x = k_191_cast_fp16)[name = tensor("op_20690_cast_fp16")]; + tensor var_20694_begin_0 = const()[name = tensor("op_20694_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_20694_end_0 = const()[name = tensor("op_20694_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_20694_end_mask_0 = const()[name = tensor("op_20694_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20694_cast_fp16 = slice_by_index(begin = var_20694_begin_0, end = var_20694_end_0, end_mask = var_20694_end_mask_0, x = k_191_cast_fp16)[name = tensor("op_20694_cast_fp16")]; + tensor var_20698_begin_0 = const()[name = tensor("op_20698_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_20698_end_0 = const()[name = tensor("op_20698_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_20698_end_mask_0 = const()[name = tensor("op_20698_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20698_cast_fp16 = slice_by_index(begin = var_20698_begin_0, end = var_20698_end_0, end_mask = var_20698_end_mask_0, x = k_191_cast_fp16)[name = tensor("op_20698_cast_fp16")]; + tensor var_20702_begin_0 = const()[name = tensor("op_20702_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_20702_end_0 = const()[name = tensor("op_20702_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_20702_end_mask_0 = const()[name = tensor("op_20702_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20702_cast_fp16 = slice_by_index(begin = var_20702_begin_0, end = var_20702_end_0, end_mask = var_20702_end_mask_0, x = k_191_cast_fp16)[name = tensor("op_20702_cast_fp16")]; + tensor var_20706_begin_0 = const()[name = tensor("op_20706_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_20706_end_0 = const()[name = tensor("op_20706_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_20706_end_mask_0 = const()[name = tensor("op_20706_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20706_cast_fp16 = slice_by_index(begin = var_20706_begin_0, end = var_20706_end_0, end_mask = var_20706_end_mask_0, x = k_191_cast_fp16)[name = tensor("op_20706_cast_fp16")]; + tensor var_20710_begin_0 = const()[name = tensor("op_20710_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_20710_end_0 = const()[name = tensor("op_20710_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_20710_end_mask_0 = const()[name = tensor("op_20710_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20710_cast_fp16 = slice_by_index(begin = var_20710_begin_0, end = var_20710_end_0, end_mask = var_20710_end_mask_0, x = k_191_cast_fp16)[name = tensor("op_20710_cast_fp16")]; + tensor var_20714_begin_0 = const()[name = tensor("op_20714_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_20714_end_0 = const()[name = tensor("op_20714_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_20714_end_mask_0 = const()[name = tensor("op_20714_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20714_cast_fp16 = slice_by_index(begin = var_20714_begin_0, end = var_20714_end_0, end_mask = var_20714_end_mask_0, x = k_191_cast_fp16)[name = tensor("op_20714_cast_fp16")]; + tensor var_20718_begin_0 = const()[name = tensor("op_20718_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_20718_end_0 = const()[name = tensor("op_20718_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_20718_end_mask_0 = const()[name = tensor("op_20718_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20718_cast_fp16 = slice_by_index(begin = var_20718_begin_0, end = var_20718_end_0, end_mask = var_20718_end_mask_0, x = k_191_cast_fp16)[name = tensor("op_20718_cast_fp16")]; + tensor var_20722_begin_0 = const()[name = tensor("op_20722_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_20722_end_0 = const()[name = tensor("op_20722_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_20722_end_mask_0 = const()[name = tensor("op_20722_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20722_cast_fp16 = slice_by_index(begin = var_20722_begin_0, end = var_20722_end_0, end_mask = var_20722_end_mask_0, x = k_191_cast_fp16)[name = tensor("op_20722_cast_fp16")]; + tensor var_20726_begin_0 = const()[name = tensor("op_20726_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_20726_end_0 = const()[name = tensor("op_20726_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_20726_end_mask_0 = const()[name = tensor("op_20726_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20726_cast_fp16 = slice_by_index(begin = var_20726_begin_0, end = var_20726_end_0, end_mask = var_20726_end_mask_0, x = k_191_cast_fp16)[name = tensor("op_20726_cast_fp16")]; + tensor var_20730_begin_0 = const()[name = tensor("op_20730_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_20730_end_0 = const()[name = tensor("op_20730_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_20730_end_mask_0 = const()[name = tensor("op_20730_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20730_cast_fp16 = slice_by_index(begin = var_20730_begin_0, end = var_20730_end_0, end_mask = var_20730_end_mask_0, x = k_191_cast_fp16)[name = tensor("op_20730_cast_fp16")]; + tensor var_20734_begin_0 = const()[name = tensor("op_20734_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_20734_end_0 = const()[name = tensor("op_20734_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_20734_end_mask_0 = const()[name = tensor("op_20734_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20734_cast_fp16 = slice_by_index(begin = var_20734_begin_0, end = var_20734_end_0, end_mask = var_20734_end_mask_0, x = k_191_cast_fp16)[name = tensor("op_20734_cast_fp16")]; + tensor var_20738_begin_0 = const()[name = tensor("op_20738_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_20738_end_0 = const()[name = tensor("op_20738_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_20738_end_mask_0 = const()[name = tensor("op_20738_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20738_cast_fp16 = slice_by_index(begin = var_20738_begin_0, end = var_20738_end_0, end_mask = var_20738_end_mask_0, x = k_191_cast_fp16)[name = tensor("op_20738_cast_fp16")]; + tensor var_20742_begin_0 = const()[name = tensor("op_20742_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_20742_end_0 = const()[name = tensor("op_20742_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_20742_end_mask_0 = const()[name = tensor("op_20742_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20742_cast_fp16 = slice_by_index(begin = var_20742_begin_0, end = var_20742_end_0, end_mask = var_20742_end_mask_0, x = k_191_cast_fp16)[name = tensor("op_20742_cast_fp16")]; + tensor var_20746_begin_0 = const()[name = tensor("op_20746_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_20746_end_0 = const()[name = tensor("op_20746_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_20746_end_mask_0 = const()[name = tensor("op_20746_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20746_cast_fp16 = slice_by_index(begin = var_20746_begin_0, end = var_20746_end_0, end_mask = var_20746_end_mask_0, x = k_191_cast_fp16)[name = tensor("op_20746_cast_fp16")]; + tensor var_20750_begin_0 = const()[name = tensor("op_20750_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_20750_end_0 = const()[name = tensor("op_20750_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_20750_end_mask_0 = const()[name = tensor("op_20750_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20750_cast_fp16 = slice_by_index(begin = var_20750_begin_0, end = var_20750_end_0, end_mask = var_20750_end_mask_0, x = k_191_cast_fp16)[name = tensor("op_20750_cast_fp16")]; + tensor var_20754_begin_0 = const()[name = tensor("op_20754_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_20754_end_0 = const()[name = tensor("op_20754_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_20754_end_mask_0 = const()[name = tensor("op_20754_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20754_cast_fp16 = slice_by_index(begin = var_20754_begin_0, end = var_20754_end_0, end_mask = var_20754_end_mask_0, x = k_191_cast_fp16)[name = tensor("op_20754_cast_fp16")]; + tensor var_20758_begin_0 = const()[name = tensor("op_20758_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_20758_end_0 = const()[name = tensor("op_20758_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_20758_end_mask_0 = const()[name = tensor("op_20758_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20758_cast_fp16 = slice_by_index(begin = var_20758_begin_0, end = var_20758_end_0, end_mask = var_20758_end_mask_0, x = k_191_cast_fp16)[name = tensor("op_20758_cast_fp16")]; + tensor var_20760_begin_0 = const()[name = tensor("op_20760_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20760_end_0 = const()[name = tensor("op_20760_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_20760_end_mask_0 = const()[name = tensor("op_20760_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20760_cast_fp16 = slice_by_index(begin = var_20760_begin_0, end = var_20760_end_0, end_mask = var_20760_end_mask_0, x = v_95_cast_fp16)[name = tensor("op_20760_cast_fp16")]; + tensor var_20764_begin_0 = const()[name = tensor("op_20764_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_20764_end_0 = const()[name = tensor("op_20764_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_20764_end_mask_0 = const()[name = tensor("op_20764_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20764_cast_fp16 = slice_by_index(begin = var_20764_begin_0, end = var_20764_end_0, end_mask = var_20764_end_mask_0, x = v_95_cast_fp16)[name = tensor("op_20764_cast_fp16")]; + tensor var_20768_begin_0 = const()[name = tensor("op_20768_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_20768_end_0 = const()[name = tensor("op_20768_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_20768_end_mask_0 = const()[name = tensor("op_20768_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20768_cast_fp16 = slice_by_index(begin = var_20768_begin_0, end = var_20768_end_0, end_mask = var_20768_end_mask_0, x = v_95_cast_fp16)[name = tensor("op_20768_cast_fp16")]; + tensor var_20772_begin_0 = const()[name = tensor("op_20772_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_20772_end_0 = const()[name = tensor("op_20772_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_20772_end_mask_0 = const()[name = tensor("op_20772_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20772_cast_fp16 = slice_by_index(begin = var_20772_begin_0, end = var_20772_end_0, end_mask = var_20772_end_mask_0, x = v_95_cast_fp16)[name = tensor("op_20772_cast_fp16")]; + tensor var_20776_begin_0 = const()[name = tensor("op_20776_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_20776_end_0 = const()[name = tensor("op_20776_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_20776_end_mask_0 = const()[name = tensor("op_20776_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20776_cast_fp16 = slice_by_index(begin = var_20776_begin_0, end = var_20776_end_0, end_mask = var_20776_end_mask_0, x = v_95_cast_fp16)[name = tensor("op_20776_cast_fp16")]; + tensor var_20780_begin_0 = const()[name = tensor("op_20780_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_20780_end_0 = const()[name = tensor("op_20780_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_20780_end_mask_0 = const()[name = tensor("op_20780_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20780_cast_fp16 = slice_by_index(begin = var_20780_begin_0, end = var_20780_end_0, end_mask = var_20780_end_mask_0, x = v_95_cast_fp16)[name = tensor("op_20780_cast_fp16")]; + tensor var_20784_begin_0 = const()[name = tensor("op_20784_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_20784_end_0 = const()[name = tensor("op_20784_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_20784_end_mask_0 = const()[name = tensor("op_20784_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20784_cast_fp16 = slice_by_index(begin = var_20784_begin_0, end = var_20784_end_0, end_mask = var_20784_end_mask_0, x = v_95_cast_fp16)[name = tensor("op_20784_cast_fp16")]; + tensor var_20788_begin_0 = const()[name = tensor("op_20788_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_20788_end_0 = const()[name = tensor("op_20788_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_20788_end_mask_0 = const()[name = tensor("op_20788_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20788_cast_fp16 = slice_by_index(begin = var_20788_begin_0, end = var_20788_end_0, end_mask = var_20788_end_mask_0, x = v_95_cast_fp16)[name = tensor("op_20788_cast_fp16")]; + tensor var_20792_begin_0 = const()[name = tensor("op_20792_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_20792_end_0 = const()[name = tensor("op_20792_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_20792_end_mask_0 = const()[name = tensor("op_20792_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20792_cast_fp16 = slice_by_index(begin = var_20792_begin_0, end = var_20792_end_0, end_mask = var_20792_end_mask_0, x = v_95_cast_fp16)[name = tensor("op_20792_cast_fp16")]; + tensor var_20796_begin_0 = const()[name = tensor("op_20796_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_20796_end_0 = const()[name = tensor("op_20796_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_20796_end_mask_0 = const()[name = tensor("op_20796_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20796_cast_fp16 = slice_by_index(begin = var_20796_begin_0, end = var_20796_end_0, end_mask = var_20796_end_mask_0, x = v_95_cast_fp16)[name = tensor("op_20796_cast_fp16")]; + tensor var_20800_begin_0 = const()[name = tensor("op_20800_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_20800_end_0 = const()[name = tensor("op_20800_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_20800_end_mask_0 = const()[name = tensor("op_20800_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20800_cast_fp16 = slice_by_index(begin = var_20800_begin_0, end = var_20800_end_0, end_mask = var_20800_end_mask_0, x = v_95_cast_fp16)[name = tensor("op_20800_cast_fp16")]; + tensor var_20804_begin_0 = const()[name = tensor("op_20804_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_20804_end_0 = const()[name = tensor("op_20804_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_20804_end_mask_0 = const()[name = tensor("op_20804_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20804_cast_fp16 = slice_by_index(begin = var_20804_begin_0, end = var_20804_end_0, end_mask = var_20804_end_mask_0, x = v_95_cast_fp16)[name = tensor("op_20804_cast_fp16")]; + tensor var_20808_begin_0 = const()[name = tensor("op_20808_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_20808_end_0 = const()[name = tensor("op_20808_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_20808_end_mask_0 = const()[name = tensor("op_20808_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20808_cast_fp16 = slice_by_index(begin = var_20808_begin_0, end = var_20808_end_0, end_mask = var_20808_end_mask_0, x = v_95_cast_fp16)[name = tensor("op_20808_cast_fp16")]; + tensor var_20812_begin_0 = const()[name = tensor("op_20812_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_20812_end_0 = const()[name = tensor("op_20812_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_20812_end_mask_0 = const()[name = tensor("op_20812_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20812_cast_fp16 = slice_by_index(begin = var_20812_begin_0, end = var_20812_end_0, end_mask = var_20812_end_mask_0, x = v_95_cast_fp16)[name = tensor("op_20812_cast_fp16")]; + tensor var_20816_begin_0 = const()[name = tensor("op_20816_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_20816_end_0 = const()[name = tensor("op_20816_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_20816_end_mask_0 = const()[name = tensor("op_20816_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20816_cast_fp16 = slice_by_index(begin = var_20816_begin_0, end = var_20816_end_0, end_mask = var_20816_end_mask_0, x = v_95_cast_fp16)[name = tensor("op_20816_cast_fp16")]; + tensor var_20820_begin_0 = const()[name = tensor("op_20820_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_20820_end_0 = const()[name = tensor("op_20820_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_20820_end_mask_0 = const()[name = tensor("op_20820_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20820_cast_fp16 = slice_by_index(begin = var_20820_begin_0, end = var_20820_end_0, end_mask = var_20820_end_mask_0, x = v_95_cast_fp16)[name = tensor("op_20820_cast_fp16")]; + tensor var_20824_begin_0 = const()[name = tensor("op_20824_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_20824_end_0 = const()[name = tensor("op_20824_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_20824_end_mask_0 = const()[name = tensor("op_20824_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20824_cast_fp16 = slice_by_index(begin = var_20824_begin_0, end = var_20824_end_0, end_mask = var_20824_end_mask_0, x = v_95_cast_fp16)[name = tensor("op_20824_cast_fp16")]; + tensor var_20828_begin_0 = const()[name = tensor("op_20828_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_20828_end_0 = const()[name = tensor("op_20828_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_20828_end_mask_0 = const()[name = tensor("op_20828_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20828_cast_fp16 = slice_by_index(begin = var_20828_begin_0, end = var_20828_end_0, end_mask = var_20828_end_mask_0, x = v_95_cast_fp16)[name = tensor("op_20828_cast_fp16")]; + tensor var_20832_begin_0 = const()[name = tensor("op_20832_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_20832_end_0 = const()[name = tensor("op_20832_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_20832_end_mask_0 = const()[name = tensor("op_20832_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20832_cast_fp16 = slice_by_index(begin = var_20832_begin_0, end = var_20832_end_0, end_mask = var_20832_end_mask_0, x = v_95_cast_fp16)[name = tensor("op_20832_cast_fp16")]; + tensor var_20836_begin_0 = const()[name = tensor("op_20836_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_20836_end_0 = const()[name = tensor("op_20836_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_20836_end_mask_0 = const()[name = tensor("op_20836_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20836_cast_fp16 = slice_by_index(begin = var_20836_begin_0, end = var_20836_end_0, end_mask = var_20836_end_mask_0, x = v_95_cast_fp16)[name = tensor("op_20836_cast_fp16")]; + tensor var_20840_equation_0 = const()[name = tensor("op_20840_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20840_cast_fp16 = einsum(equation = var_20840_equation_0, values = (var_20682_cast_fp16, var_20599_cast_fp16))[name = tensor("op_20840_cast_fp16")]; + tensor var_20841_to_fp16 = const()[name = tensor("op_20841_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1721_cast_fp16 = mul(x = var_20840_cast_fp16, y = var_20841_to_fp16)[name = tensor("aw_1721_cast_fp16")]; + tensor var_20844_equation_0 = const()[name = tensor("op_20844_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20844_cast_fp16 = einsum(equation = var_20844_equation_0, values = (var_20686_cast_fp16, var_20603_cast_fp16))[name = tensor("op_20844_cast_fp16")]; + tensor var_20845_to_fp16 = const()[name = tensor("op_20845_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1723_cast_fp16 = mul(x = var_20844_cast_fp16, y = var_20845_to_fp16)[name = tensor("aw_1723_cast_fp16")]; + tensor var_20848_equation_0 = const()[name = tensor("op_20848_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20848_cast_fp16 = einsum(equation = var_20848_equation_0, values = (var_20690_cast_fp16, var_20607_cast_fp16))[name = tensor("op_20848_cast_fp16")]; + tensor var_20849_to_fp16 = const()[name = tensor("op_20849_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1725_cast_fp16 = mul(x = var_20848_cast_fp16, y = var_20849_to_fp16)[name = tensor("aw_1725_cast_fp16")]; + tensor var_20852_equation_0 = const()[name = tensor("op_20852_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20852_cast_fp16 = einsum(equation = var_20852_equation_0, values = (var_20694_cast_fp16, var_20611_cast_fp16))[name = tensor("op_20852_cast_fp16")]; + tensor var_20853_to_fp16 = const()[name = tensor("op_20853_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1727_cast_fp16 = mul(x = var_20852_cast_fp16, y = var_20853_to_fp16)[name = tensor("aw_1727_cast_fp16")]; + tensor var_20856_equation_0 = const()[name = tensor("op_20856_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20856_cast_fp16 = einsum(equation = var_20856_equation_0, values = (var_20698_cast_fp16, var_20615_cast_fp16))[name = tensor("op_20856_cast_fp16")]; + tensor var_20857_to_fp16 = const()[name = tensor("op_20857_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1729_cast_fp16 = mul(x = var_20856_cast_fp16, y = var_20857_to_fp16)[name = tensor("aw_1729_cast_fp16")]; + tensor var_20860_equation_0 = const()[name = tensor("op_20860_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20860_cast_fp16 = einsum(equation = var_20860_equation_0, values = (var_20702_cast_fp16, var_20619_cast_fp16))[name = tensor("op_20860_cast_fp16")]; + tensor var_20861_to_fp16 = const()[name = tensor("op_20861_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1731_cast_fp16 = mul(x = var_20860_cast_fp16, y = var_20861_to_fp16)[name = tensor("aw_1731_cast_fp16")]; + tensor var_20864_equation_0 = const()[name = tensor("op_20864_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20864_cast_fp16 = einsum(equation = var_20864_equation_0, values = (var_20706_cast_fp16, var_20623_cast_fp16))[name = tensor("op_20864_cast_fp16")]; + tensor var_20865_to_fp16 = const()[name = tensor("op_20865_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1733_cast_fp16 = mul(x = var_20864_cast_fp16, y = var_20865_to_fp16)[name = tensor("aw_1733_cast_fp16")]; + tensor var_20868_equation_0 = const()[name = tensor("op_20868_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20868_cast_fp16 = einsum(equation = var_20868_equation_0, values = (var_20710_cast_fp16, var_20627_cast_fp16))[name = tensor("op_20868_cast_fp16")]; + tensor var_20869_to_fp16 = const()[name = tensor("op_20869_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1735_cast_fp16 = mul(x = var_20868_cast_fp16, y = var_20869_to_fp16)[name = tensor("aw_1735_cast_fp16")]; + tensor var_20872_equation_0 = const()[name = tensor("op_20872_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20872_cast_fp16 = einsum(equation = var_20872_equation_0, values = (var_20714_cast_fp16, var_20631_cast_fp16))[name = tensor("op_20872_cast_fp16")]; + tensor var_20873_to_fp16 = const()[name = tensor("op_20873_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1737_cast_fp16 = mul(x = var_20872_cast_fp16, y = var_20873_to_fp16)[name = tensor("aw_1737_cast_fp16")]; + tensor var_20876_equation_0 = const()[name = tensor("op_20876_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20876_cast_fp16 = einsum(equation = var_20876_equation_0, values = (var_20718_cast_fp16, var_20635_cast_fp16))[name = tensor("op_20876_cast_fp16")]; + tensor var_20877_to_fp16 = const()[name = tensor("op_20877_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1739_cast_fp16 = mul(x = var_20876_cast_fp16, y = var_20877_to_fp16)[name = tensor("aw_1739_cast_fp16")]; + tensor var_20880_equation_0 = const()[name = tensor("op_20880_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20880_cast_fp16 = einsum(equation = var_20880_equation_0, values = (var_20722_cast_fp16, var_20639_cast_fp16))[name = tensor("op_20880_cast_fp16")]; + tensor var_20881_to_fp16 = const()[name = tensor("op_20881_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1741_cast_fp16 = mul(x = var_20880_cast_fp16, y = var_20881_to_fp16)[name = tensor("aw_1741_cast_fp16")]; + tensor var_20884_equation_0 = const()[name = tensor("op_20884_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20884_cast_fp16 = einsum(equation = var_20884_equation_0, values = (var_20726_cast_fp16, var_20643_cast_fp16))[name = tensor("op_20884_cast_fp16")]; + tensor var_20885_to_fp16 = const()[name = tensor("op_20885_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1743_cast_fp16 = mul(x = var_20884_cast_fp16, y = var_20885_to_fp16)[name = tensor("aw_1743_cast_fp16")]; + tensor var_20888_equation_0 = const()[name = tensor("op_20888_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20888_cast_fp16 = einsum(equation = var_20888_equation_0, values = (var_20730_cast_fp16, var_20647_cast_fp16))[name = tensor("op_20888_cast_fp16")]; + tensor var_20889_to_fp16 = const()[name = tensor("op_20889_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1745_cast_fp16 = mul(x = var_20888_cast_fp16, y = var_20889_to_fp16)[name = tensor("aw_1745_cast_fp16")]; + tensor var_20892_equation_0 = const()[name = tensor("op_20892_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20892_cast_fp16 = einsum(equation = var_20892_equation_0, values = (var_20734_cast_fp16, var_20651_cast_fp16))[name = tensor("op_20892_cast_fp16")]; + tensor var_20893_to_fp16 = const()[name = tensor("op_20893_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1747_cast_fp16 = mul(x = var_20892_cast_fp16, y = var_20893_to_fp16)[name = tensor("aw_1747_cast_fp16")]; + tensor var_20896_equation_0 = const()[name = tensor("op_20896_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20896_cast_fp16 = einsum(equation = var_20896_equation_0, values = (var_20738_cast_fp16, var_20655_cast_fp16))[name = tensor("op_20896_cast_fp16")]; + tensor var_20897_to_fp16 = const()[name = tensor("op_20897_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1749_cast_fp16 = mul(x = var_20896_cast_fp16, y = var_20897_to_fp16)[name = tensor("aw_1749_cast_fp16")]; + tensor var_20900_equation_0 = const()[name = tensor("op_20900_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20900_cast_fp16 = einsum(equation = var_20900_equation_0, values = (var_20742_cast_fp16, var_20659_cast_fp16))[name = tensor("op_20900_cast_fp16")]; + tensor var_20901_to_fp16 = const()[name = tensor("op_20901_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1751_cast_fp16 = mul(x = var_20900_cast_fp16, y = var_20901_to_fp16)[name = tensor("aw_1751_cast_fp16")]; + tensor var_20904_equation_0 = const()[name = tensor("op_20904_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20904_cast_fp16 = einsum(equation = var_20904_equation_0, values = (var_20746_cast_fp16, var_20663_cast_fp16))[name = tensor("op_20904_cast_fp16")]; + tensor var_20905_to_fp16 = const()[name = tensor("op_20905_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1753_cast_fp16 = mul(x = var_20904_cast_fp16, y = var_20905_to_fp16)[name = tensor("aw_1753_cast_fp16")]; + tensor var_20908_equation_0 = const()[name = tensor("op_20908_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20908_cast_fp16 = einsum(equation = var_20908_equation_0, values = (var_20750_cast_fp16, var_20667_cast_fp16))[name = tensor("op_20908_cast_fp16")]; + tensor var_20909_to_fp16 = const()[name = tensor("op_20909_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1755_cast_fp16 = mul(x = var_20908_cast_fp16, y = var_20909_to_fp16)[name = tensor("aw_1755_cast_fp16")]; + tensor var_20912_equation_0 = const()[name = tensor("op_20912_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20912_cast_fp16 = einsum(equation = var_20912_equation_0, values = (var_20754_cast_fp16, var_20671_cast_fp16))[name = tensor("op_20912_cast_fp16")]; + tensor var_20913_to_fp16 = const()[name = tensor("op_20913_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1757_cast_fp16 = mul(x = var_20912_cast_fp16, y = var_20913_to_fp16)[name = tensor("aw_1757_cast_fp16")]; + tensor var_20916_equation_0 = const()[name = tensor("op_20916_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20916_cast_fp16 = einsum(equation = var_20916_equation_0, values = (var_20758_cast_fp16, var_20675_cast_fp16))[name = tensor("op_20916_cast_fp16")]; + tensor var_20917_to_fp16 = const()[name = tensor("op_20917_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1759_cast_fp16 = mul(x = var_20916_cast_fp16, y = var_20917_to_fp16)[name = tensor("aw_1759_cast_fp16")]; + tensor var_20919_cast_fp16 = softmax(axis = var_2624, x = aw_1721_cast_fp16)[name = tensor("op_20919_cast_fp16")]; + tensor var_20920_cast_fp16 = softmax(axis = var_2624, x = aw_1723_cast_fp16)[name = tensor("op_20920_cast_fp16")]; + tensor var_20921_cast_fp16 = softmax(axis = var_2624, x = aw_1725_cast_fp16)[name = tensor("op_20921_cast_fp16")]; + tensor var_20922_cast_fp16 = softmax(axis = var_2624, x = aw_1727_cast_fp16)[name = tensor("op_20922_cast_fp16")]; + tensor var_20923_cast_fp16 = softmax(axis = var_2624, x = aw_1729_cast_fp16)[name = tensor("op_20923_cast_fp16")]; + tensor var_20924_cast_fp16 = softmax(axis = var_2624, x = aw_1731_cast_fp16)[name = tensor("op_20924_cast_fp16")]; + tensor var_20925_cast_fp16 = softmax(axis = var_2624, x = aw_1733_cast_fp16)[name = tensor("op_20925_cast_fp16")]; + tensor var_20926_cast_fp16 = softmax(axis = var_2624, x = aw_1735_cast_fp16)[name = tensor("op_20926_cast_fp16")]; + tensor var_20927_cast_fp16 = softmax(axis = var_2624, x = aw_1737_cast_fp16)[name = tensor("op_20927_cast_fp16")]; + tensor var_20928_cast_fp16 = softmax(axis = var_2624, x = aw_1739_cast_fp16)[name = tensor("op_20928_cast_fp16")]; + tensor var_20929_cast_fp16 = softmax(axis = var_2624, x = aw_1741_cast_fp16)[name = tensor("op_20929_cast_fp16")]; + tensor var_20930_cast_fp16 = softmax(axis = var_2624, x = aw_1743_cast_fp16)[name = tensor("op_20930_cast_fp16")]; + tensor var_20931_cast_fp16 = softmax(axis = var_2624, x = aw_1745_cast_fp16)[name = tensor("op_20931_cast_fp16")]; + tensor var_20932_cast_fp16 = softmax(axis = var_2624, x = aw_1747_cast_fp16)[name = tensor("op_20932_cast_fp16")]; + tensor var_20933_cast_fp16 = softmax(axis = var_2624, x = aw_1749_cast_fp16)[name = tensor("op_20933_cast_fp16")]; + tensor var_20934_cast_fp16 = softmax(axis = var_2624, x = aw_1751_cast_fp16)[name = tensor("op_20934_cast_fp16")]; + tensor var_20935_cast_fp16 = softmax(axis = var_2624, x = aw_1753_cast_fp16)[name = tensor("op_20935_cast_fp16")]; + tensor var_20936_cast_fp16 = softmax(axis = var_2624, x = aw_1755_cast_fp16)[name = tensor("op_20936_cast_fp16")]; + tensor var_20937_cast_fp16 = softmax(axis = var_2624, x = aw_1757_cast_fp16)[name = tensor("op_20937_cast_fp16")]; + tensor var_20938_cast_fp16 = softmax(axis = var_2624, x = aw_1759_cast_fp16)[name = tensor("op_20938_cast_fp16")]; + tensor var_20940_equation_0 = const()[name = tensor("op_20940_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20940_cast_fp16 = einsum(equation = var_20940_equation_0, values = (var_20760_cast_fp16, var_20919_cast_fp16))[name = tensor("op_20940_cast_fp16")]; + tensor var_20942_equation_0 = const()[name = tensor("op_20942_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20942_cast_fp16 = einsum(equation = var_20942_equation_0, values = (var_20764_cast_fp16, var_20920_cast_fp16))[name = tensor("op_20942_cast_fp16")]; + tensor var_20944_equation_0 = const()[name = tensor("op_20944_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20944_cast_fp16 = einsum(equation = var_20944_equation_0, values = (var_20768_cast_fp16, var_20921_cast_fp16))[name = tensor("op_20944_cast_fp16")]; + tensor var_20946_equation_0 = const()[name = tensor("op_20946_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20946_cast_fp16 = einsum(equation = var_20946_equation_0, values = (var_20772_cast_fp16, var_20922_cast_fp16))[name = tensor("op_20946_cast_fp16")]; + tensor var_20948_equation_0 = const()[name = tensor("op_20948_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20948_cast_fp16 = einsum(equation = var_20948_equation_0, values = (var_20776_cast_fp16, var_20923_cast_fp16))[name = tensor("op_20948_cast_fp16")]; + tensor var_20950_equation_0 = const()[name = tensor("op_20950_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20950_cast_fp16 = einsum(equation = var_20950_equation_0, values = (var_20780_cast_fp16, var_20924_cast_fp16))[name = tensor("op_20950_cast_fp16")]; + tensor var_20952_equation_0 = const()[name = tensor("op_20952_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20952_cast_fp16 = einsum(equation = var_20952_equation_0, values = (var_20784_cast_fp16, var_20925_cast_fp16))[name = tensor("op_20952_cast_fp16")]; + tensor var_20954_equation_0 = const()[name = tensor("op_20954_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20954_cast_fp16 = einsum(equation = var_20954_equation_0, values = (var_20788_cast_fp16, var_20926_cast_fp16))[name = tensor("op_20954_cast_fp16")]; + tensor var_20956_equation_0 = const()[name = tensor("op_20956_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20956_cast_fp16 = einsum(equation = var_20956_equation_0, values = (var_20792_cast_fp16, var_20927_cast_fp16))[name = tensor("op_20956_cast_fp16")]; + tensor var_20958_equation_0 = const()[name = tensor("op_20958_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20958_cast_fp16 = einsum(equation = var_20958_equation_0, values = (var_20796_cast_fp16, var_20928_cast_fp16))[name = tensor("op_20958_cast_fp16")]; + tensor var_20960_equation_0 = const()[name = tensor("op_20960_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20960_cast_fp16 = einsum(equation = var_20960_equation_0, values = (var_20800_cast_fp16, var_20929_cast_fp16))[name = tensor("op_20960_cast_fp16")]; + tensor var_20962_equation_0 = const()[name = tensor("op_20962_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20962_cast_fp16 = einsum(equation = var_20962_equation_0, values = (var_20804_cast_fp16, var_20930_cast_fp16))[name = tensor("op_20962_cast_fp16")]; + tensor var_20964_equation_0 = const()[name = tensor("op_20964_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20964_cast_fp16 = einsum(equation = var_20964_equation_0, values = (var_20808_cast_fp16, var_20931_cast_fp16))[name = tensor("op_20964_cast_fp16")]; + tensor var_20966_equation_0 = const()[name = tensor("op_20966_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20966_cast_fp16 = einsum(equation = var_20966_equation_0, values = (var_20812_cast_fp16, var_20932_cast_fp16))[name = tensor("op_20966_cast_fp16")]; + tensor var_20968_equation_0 = const()[name = tensor("op_20968_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20968_cast_fp16 = einsum(equation = var_20968_equation_0, values = (var_20816_cast_fp16, var_20933_cast_fp16))[name = tensor("op_20968_cast_fp16")]; + tensor var_20970_equation_0 = const()[name = tensor("op_20970_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20970_cast_fp16 = einsum(equation = var_20970_equation_0, values = (var_20820_cast_fp16, var_20934_cast_fp16))[name = tensor("op_20970_cast_fp16")]; + tensor var_20972_equation_0 = const()[name = tensor("op_20972_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20972_cast_fp16 = einsum(equation = var_20972_equation_0, values = (var_20824_cast_fp16, var_20935_cast_fp16))[name = tensor("op_20972_cast_fp16")]; + tensor var_20974_equation_0 = const()[name = tensor("op_20974_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20974_cast_fp16 = einsum(equation = var_20974_equation_0, values = (var_20828_cast_fp16, var_20936_cast_fp16))[name = tensor("op_20974_cast_fp16")]; + tensor var_20976_equation_0 = const()[name = tensor("op_20976_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20976_cast_fp16 = einsum(equation = var_20976_equation_0, values = (var_20832_cast_fp16, var_20937_cast_fp16))[name = tensor("op_20976_cast_fp16")]; + tensor var_20978_equation_0 = const()[name = tensor("op_20978_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_20978_cast_fp16 = einsum(equation = var_20978_equation_0, values = (var_20836_cast_fp16, var_20938_cast_fp16))[name = tensor("op_20978_cast_fp16")]; + tensor input_303_interleave_0 = const()[name = tensor("input_303_interleave_0"), val = tensor(false)]; + tensor input_303_cast_fp16 = concat(axis = var_2624, interleave = input_303_interleave_0, values = (var_20940_cast_fp16, var_20942_cast_fp16, var_20944_cast_fp16, var_20946_cast_fp16, var_20948_cast_fp16, var_20950_cast_fp16, var_20952_cast_fp16, var_20954_cast_fp16, var_20956_cast_fp16, var_20958_cast_fp16, var_20960_cast_fp16, var_20962_cast_fp16, var_20964_cast_fp16, var_20966_cast_fp16, var_20968_cast_fp16, var_20970_cast_fp16, var_20972_cast_fp16, var_20974_cast_fp16, var_20976_cast_fp16, var_20978_cast_fp16))[name = tensor("input_303_cast_fp16")]; + tensor var_20984 = const()[name = tensor("op_20984"), val = tensor([1, 1])]; + tensor var_20986 = const()[name = tensor("op_20986"), val = tensor([1, 1])]; + tensor var_20988_pad_type_0 = const()[name = tensor("op_20988_pad_type_0"), val = tensor("custom")]; + tensor var_20988_pad_0 = const()[name = tensor("op_20988_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_9_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_9_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1614637440)))]; + tensor down_blocks_2_attentions_1_transformer_blocks_9_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_9_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1617914304)))]; + tensor var_20988_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_9_attn2_to_out_0_bias_to_fp16, dilations = var_20986, groups = var_2624, pad = var_20988_pad_0, pad_type = var_20988_pad_type_0, strides = var_20984, weight = down_blocks_2_attentions_1_transformer_blocks_9_attn2_to_out_0_weight_to_fp16, x = input_303_cast_fp16)[name = tensor("op_20988_cast_fp16")]; + tensor inputs_143_cast_fp16 = add(x = var_20988_cast_fp16, y = inputs_141_cast_fp16)[name = tensor("inputs_143_cast_fp16")]; + tensor input_305_axes_0 = const()[name = tensor("input_305_axes_0"), val = tensor([1])]; + tensor input_305_gamma_0_to_fp16 = const()[name = tensor("input_305_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1617916928)))]; + tensor input_305_beta_0_to_fp16 = const()[name = tensor("input_305_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1617919552)))]; + tensor var_20998_to_fp16 = const()[name = tensor("op_20998_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_305_cast_fp16 = layer_norm(axes = input_305_axes_0, beta = input_305_beta_0_to_fp16, epsilon = var_20998_to_fp16, gamma = input_305_gamma_0_to_fp16, x = inputs_143_cast_fp16)[name = tensor("input_305_cast_fp16")]; + tensor var_21014 = const()[name = tensor("op_21014"), val = tensor([1, 1])]; + tensor var_21016 = const()[name = tensor("op_21016"), val = tensor([1, 1])]; + tensor var_21018_pad_type_0 = const()[name = tensor("op_21018_pad_type_0"), val = tensor("custom")]; + tensor var_21018_pad_0 = const()[name = tensor("op_21018_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_9_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_9_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1617922176)))]; + tensor down_blocks_2_attentions_1_transformer_blocks_9_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_9_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1644136640)))]; + tensor var_21018_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_9_ff_net_0_proj_bias_to_fp16, dilations = var_21016, groups = var_2624, pad = var_21018_pad_0, pad_type = var_21018_pad_type_0, strides = var_21014, weight = down_blocks_2_attentions_1_transformer_blocks_9_ff_net_0_proj_weight_to_fp16, x = input_305_cast_fp16)[name = tensor("op_21018_cast_fp16")]; + tensor var_21019_split_sizes_0 = const()[name = tensor("op_21019_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_21019_axis_0 = const()[name = tensor("op_21019_axis_0"), val = tensor(1)]; + tensor var_21019_cast_fp16_0, tensor var_21019_cast_fp16_1 = split(axis = var_21019_axis_0, split_sizes = var_21019_split_sizes_0, x = var_21018_cast_fp16)[name = tensor("op_21019_cast_fp16")]; + tensor var_21021_mode_0 = const()[name = tensor("op_21021_mode_0"), val = tensor("EXACT")]; + tensor var_21021_cast_fp16 = gelu(mode = var_21021_mode_0, x = var_21019_cast_fp16_1)[name = tensor("op_21021_cast_fp16")]; + tensor input_307_cast_fp16 = mul(x = var_21019_cast_fp16_0, y = var_21021_cast_fp16)[name = tensor("input_307_cast_fp16")]; + tensor var_21025 = const()[name = tensor("op_21025"), val = tensor([1, 1])]; + tensor var_21027 = const()[name = tensor("op_21027"), val = tensor([1, 1])]; + tensor var_21029_pad_type_0 = const()[name = tensor("op_21029_pad_type_0"), val = tensor("custom")]; + tensor var_21029_pad_0 = const()[name = tensor("op_21029_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_9_ff_net_2_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_9_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1644157184)))]; + tensor down_blocks_2_attentions_1_transformer_blocks_9_ff_net_2_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_9_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1657264448)))]; + tensor var_21029_cast_fp16 = conv(bias = down_blocks_2_attentions_1_transformer_blocks_9_ff_net_2_bias_to_fp16, dilations = var_21027, groups = var_2624, pad = var_21029_pad_0, pad_type = var_21029_pad_type_0, strides = var_21025, weight = down_blocks_2_attentions_1_transformer_blocks_9_ff_net_2_weight_to_fp16, x = input_307_cast_fp16)[name = tensor("op_21029_cast_fp16")]; + tensor hidden_states_197_cast_fp16 = add(x = var_21029_cast_fp16, y = inputs_143_cast_fp16)[name = tensor("hidden_states_197_cast_fp16")]; + tensor var_21031 = const()[name = tensor("op_21031"), val = tensor([2, 1280, 32, 32])]; + tensor input_309_cast_fp16 = reshape(shape = var_21031, x = hidden_states_197_cast_fp16)[name = tensor("input_309_cast_fp16")]; + tensor var_21035 = const()[name = tensor("op_21035"), val = tensor([1, 1])]; + tensor var_21037 = const()[name = tensor("op_21037"), val = tensor([1, 1])]; + tensor hidden_states_199_pad_type_0 = const()[name = tensor("hidden_states_199_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_199_pad_0 = const()[name = tensor("hidden_states_199_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_proj_out_weight_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_proj_out_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1657267072)))]; + tensor down_blocks_2_attentions_1_proj_out_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_proj_out_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1660543936)))]; + tensor hidden_states_199_cast_fp16 = conv(bias = down_blocks_2_attentions_1_proj_out_bias_to_fp16, dilations = var_21037, groups = var_2624, pad = hidden_states_199_pad_0, pad_type = hidden_states_199_pad_type_0, strides = var_21035, weight = down_blocks_2_attentions_1_proj_out_weight_to_fp16, x = input_309_cast_fp16)[name = tensor("hidden_states_199_cast_fp16")]; + tensor input_311_cast_fp16 = add(x = hidden_states_199_cast_fp16, y = hidden_states_133_cast_fp16)[name = tensor("input_311_cast_fp16")]; + tensor var_21077 = const()[name = tensor("op_21077"), val = tensor(1)]; + tensor reshape_64_shape_0 = const()[name = tensor("reshape_64_shape_0"), val = tensor([2, 32, 40, 32, 32])]; + tensor reshape_64_cast_fp16 = reshape(shape = reshape_64_shape_0, x = input_311_cast_fp16)[name = tensor("reshape_64_cast_fp16")]; + tensor reduce_mean_48_axes_0 = const()[name = tensor("reduce_mean_48_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_48_keep_dims_0 = const()[name = tensor("reduce_mean_48_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_48_cast_fp16 = reduce_mean(axes = reduce_mean_48_axes_0, keep_dims = reduce_mean_48_keep_dims_0, x = reshape_64_cast_fp16)[name = tensor("reduce_mean_48_cast_fp16")]; + tensor sub_32_cast_fp16 = sub(x = reshape_64_cast_fp16, y = reduce_mean_48_cast_fp16)[name = tensor("sub_32_cast_fp16")]; + tensor square_16_cast_fp16 = square(x = sub_32_cast_fp16)[name = tensor("square_16_cast_fp16")]; + tensor reduce_mean_50_axes_0 = const()[name = tensor("reduce_mean_50_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_50_keep_dims_0 = const()[name = tensor("reduce_mean_50_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_50_cast_fp16 = reduce_mean(axes = reduce_mean_50_axes_0, keep_dims = reduce_mean_50_keep_dims_0, x = square_16_cast_fp16)[name = tensor("reduce_mean_50_cast_fp16")]; + tensor add_32_y_0_to_fp16 = const()[name = tensor("add_32_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_32_cast_fp16 = add(x = reduce_mean_50_cast_fp16, y = add_32_y_0_to_fp16)[name = tensor("add_32_cast_fp16")]; + tensor sqrt_16_cast_fp16 = sqrt(x = add_32_cast_fp16)[name = tensor("sqrt_16_cast_fp16")]; + tensor real_div_16_cast_fp16 = real_div(x = sub_32_cast_fp16, y = sqrt_16_cast_fp16)[name = tensor("real_div_16_cast_fp16")]; + tensor reshape_65_shape_0 = const()[name = tensor("reshape_65_shape_0"), val = tensor([2, 1280, 32, 32])]; + tensor reshape_65_cast_fp16 = reshape(shape = reshape_65_shape_0, x = real_div_16_cast_fp16)[name = tensor("reshape_65_cast_fp16")]; + tensor add_33_gamma_0_to_fp16 = const()[name = tensor("add_33_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1660546560)))]; + tensor add_33_beta_0_to_fp16 = const()[name = tensor("add_33_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1660549184)))]; + tensor add_33_epsilon_0_to_fp16 = const()[name = tensor("add_33_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_33_cast_fp16 = batch_norm(beta = add_33_beta_0_to_fp16, epsilon = add_33_epsilon_0_to_fp16, gamma = add_33_gamma_0_to_fp16, mean = add_23_mean_0_to_fp16, variance = add_23_variance_0_to_fp16, x = reshape_65_cast_fp16)[name = tensor("add_33_cast_fp16")]; + tensor input_315_cast_fp16 = silu(x = add_33_cast_fp16)[name = tensor("input_315_cast_fp16")]; + tensor var_21095 = const()[name = tensor("op_21095"), val = tensor([1, 1])]; + tensor var_21097 = const()[name = tensor("op_21097"), val = tensor([1, 1])]; + tensor hidden_states_201_pad_type_0 = const()[name = tensor("hidden_states_201_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_201_pad_0 = const()[name = tensor("hidden_states_201_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor mid_block_resnets_0_conv1_weight_to_fp16 = const()[name = tensor("mid_block_resnets_0_conv1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1660551808)))]; + tensor mid_block_resnets_0_conv1_bias_to_fp16 = const()[name = tensor("mid_block_resnets_0_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1690043072)))]; + tensor hidden_states_201_cast_fp16 = conv(bias = mid_block_resnets_0_conv1_bias_to_fp16, dilations = var_21097, groups = var_21077, pad = hidden_states_201_pad_0, pad_type = hidden_states_201_pad_type_0, strides = var_21095, weight = mid_block_resnets_0_conv1_weight_to_fp16, x = input_315_cast_fp16)[name = tensor("hidden_states_201_cast_fp16")]; + tensor var_21103 = const()[name = tensor("op_21103"), val = tensor([1, 1])]; + tensor var_21105 = const()[name = tensor("op_21105"), val = tensor([1, 1])]; + tensor temb_13_pad_type_0 = const()[name = tensor("temb_13_pad_type_0"), val = tensor("custom")]; + tensor temb_13_pad_0 = const()[name = tensor("temb_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_resnets_0_time_emb_proj_weight_to_fp16 = const()[name = tensor("mid_block_resnets_0_time_emb_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1690045696)))]; + tensor mid_block_resnets_0_time_emb_proj_bias_to_fp16 = const()[name = tensor("mid_block_resnets_0_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1693322560)))]; + tensor temb_13_cast_fp16 = conv(bias = mid_block_resnets_0_time_emb_proj_bias_to_fp16, dilations = var_21105, groups = var_21077, pad = temb_13_pad_0, pad_type = temb_13_pad_type_0, strides = var_21103, weight = mid_block_resnets_0_time_emb_proj_weight_to_fp16, x = input_21_cast_fp16)[name = tensor("temb_13_cast_fp16")]; + tensor input_319_cast_fp16 = add(x = hidden_states_201_cast_fp16, y = temb_13_cast_fp16)[name = tensor("input_319_cast_fp16")]; + tensor reshape_68_shape_0 = const()[name = tensor("reshape_68_shape_0"), val = tensor([2, 32, 40, 32, 32])]; + tensor reshape_68_cast_fp16 = reshape(shape = reshape_68_shape_0, x = input_319_cast_fp16)[name = tensor("reshape_68_cast_fp16")]; + tensor reduce_mean_51_axes_0 = const()[name = tensor("reduce_mean_51_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_51_keep_dims_0 = const()[name = tensor("reduce_mean_51_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_51_cast_fp16 = reduce_mean(axes = reduce_mean_51_axes_0, keep_dims = reduce_mean_51_keep_dims_0, x = reshape_68_cast_fp16)[name = tensor("reduce_mean_51_cast_fp16")]; + tensor sub_34_cast_fp16 = sub(x = reshape_68_cast_fp16, y = reduce_mean_51_cast_fp16)[name = tensor("sub_34_cast_fp16")]; + tensor square_17_cast_fp16 = square(x = sub_34_cast_fp16)[name = tensor("square_17_cast_fp16")]; + tensor reduce_mean_53_axes_0 = const()[name = tensor("reduce_mean_53_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_53_keep_dims_0 = const()[name = tensor("reduce_mean_53_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_53_cast_fp16 = reduce_mean(axes = reduce_mean_53_axes_0, keep_dims = reduce_mean_53_keep_dims_0, x = square_17_cast_fp16)[name = tensor("reduce_mean_53_cast_fp16")]; + tensor add_34_y_0_to_fp16 = const()[name = tensor("add_34_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_34_cast_fp16 = add(x = reduce_mean_53_cast_fp16, y = add_34_y_0_to_fp16)[name = tensor("add_34_cast_fp16")]; + tensor sqrt_17_cast_fp16 = sqrt(x = add_34_cast_fp16)[name = tensor("sqrt_17_cast_fp16")]; + tensor real_div_17_cast_fp16 = real_div(x = sub_34_cast_fp16, y = sqrt_17_cast_fp16)[name = tensor("real_div_17_cast_fp16")]; + tensor reshape_69_shape_0 = const()[name = tensor("reshape_69_shape_0"), val = tensor([2, 1280, 32, 32])]; + tensor reshape_69_cast_fp16 = reshape(shape = reshape_69_shape_0, x = real_div_17_cast_fp16)[name = tensor("reshape_69_cast_fp16")]; + tensor add_35_gamma_0_to_fp16 = const()[name = tensor("add_35_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1693325184)))]; + tensor add_35_beta_0_to_fp16 = const()[name = tensor("add_35_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1693327808)))]; + tensor add_35_epsilon_0_to_fp16 = const()[name = tensor("add_35_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_35_cast_fp16 = batch_norm(beta = add_35_beta_0_to_fp16, epsilon = add_35_epsilon_0_to_fp16, gamma = add_35_gamma_0_to_fp16, mean = add_23_mean_0_to_fp16, variance = add_23_variance_0_to_fp16, x = reshape_69_cast_fp16)[name = tensor("add_35_cast_fp16")]; + tensor input_323_cast_fp16 = silu(x = add_35_cast_fp16)[name = tensor("input_323_cast_fp16")]; + tensor var_21115 = const()[name = tensor("op_21115"), val = tensor([1, 1])]; + tensor var_21117 = const()[name = tensor("op_21117"), val = tensor([1, 1])]; + tensor hidden_states_203_pad_type_0 = const()[name = tensor("hidden_states_203_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_203_pad_0 = const()[name = tensor("hidden_states_203_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor mid_block_resnets_0_conv2_weight_to_fp16 = const()[name = tensor("mid_block_resnets_0_conv2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1693330432)))]; + tensor mid_block_resnets_0_conv2_bias_to_fp16 = const()[name = tensor("mid_block_resnets_0_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1722821696)))]; + tensor hidden_states_203_cast_fp16 = conv(bias = mid_block_resnets_0_conv2_bias_to_fp16, dilations = var_21117, groups = var_21077, pad = hidden_states_203_pad_0, pad_type = hidden_states_203_pad_type_0, strides = var_21115, weight = mid_block_resnets_0_conv2_weight_to_fp16, x = input_323_cast_fp16)[name = tensor("hidden_states_203_cast_fp16")]; + tensor hidden_states_205_cast_fp16 = add(x = input_311_cast_fp16, y = hidden_states_203_cast_fp16)[name = tensor("hidden_states_205_cast_fp16")]; + tensor reshape_72_shape_0 = const()[name = tensor("reshape_72_shape_0"), val = tensor([2, 32, 40, 32, 32])]; + tensor reshape_72_cast_fp16 = reshape(shape = reshape_72_shape_0, x = hidden_states_205_cast_fp16)[name = tensor("reshape_72_cast_fp16")]; + tensor reduce_mean_54_axes_0 = const()[name = tensor("reduce_mean_54_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_54_keep_dims_0 = const()[name = tensor("reduce_mean_54_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_54_cast_fp16 = reduce_mean(axes = reduce_mean_54_axes_0, keep_dims = reduce_mean_54_keep_dims_0, x = reshape_72_cast_fp16)[name = tensor("reduce_mean_54_cast_fp16")]; + tensor sub_36_cast_fp16 = sub(x = reshape_72_cast_fp16, y = reduce_mean_54_cast_fp16)[name = tensor("sub_36_cast_fp16")]; + tensor square_18_cast_fp16 = square(x = sub_36_cast_fp16)[name = tensor("square_18_cast_fp16")]; + tensor reduce_mean_56_axes_0 = const()[name = tensor("reduce_mean_56_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_56_keep_dims_0 = const()[name = tensor("reduce_mean_56_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_56_cast_fp16 = reduce_mean(axes = reduce_mean_56_axes_0, keep_dims = reduce_mean_56_keep_dims_0, x = square_18_cast_fp16)[name = tensor("reduce_mean_56_cast_fp16")]; + tensor add_36_y_0_to_fp16 = const()[name = tensor("add_36_y_0_to_fp16"), val = tensor(0x1.1p-20)]; + tensor add_36_cast_fp16 = add(x = reduce_mean_56_cast_fp16, y = add_36_y_0_to_fp16)[name = tensor("add_36_cast_fp16")]; + tensor sqrt_18_cast_fp16 = sqrt(x = add_36_cast_fp16)[name = tensor("sqrt_18_cast_fp16")]; + tensor real_div_18_cast_fp16 = real_div(x = sub_36_cast_fp16, y = sqrt_18_cast_fp16)[name = tensor("real_div_18_cast_fp16")]; + tensor reshape_73_shape_0 = const()[name = tensor("reshape_73_shape_0"), val = tensor([2, 1280, 32, 32])]; + tensor reshape_73_cast_fp16 = reshape(shape = reshape_73_shape_0, x = real_div_18_cast_fp16)[name = tensor("reshape_73_cast_fp16")]; + tensor add_37_gamma_0_to_fp16 = const()[name = tensor("add_37_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1722824320)))]; + tensor add_37_beta_0_to_fp16 = const()[name = tensor("add_37_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1722826944)))]; + tensor add_37_epsilon_0_to_fp16 = const()[name = tensor("add_37_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_37_cast_fp16 = batch_norm(beta = add_37_beta_0_to_fp16, epsilon = add_37_epsilon_0_to_fp16, gamma = add_37_gamma_0_to_fp16, mean = add_23_mean_0_to_fp16, variance = add_23_variance_0_to_fp16, x = reshape_73_cast_fp16)[name = tensor("add_37_cast_fp16")]; + tensor var_21155 = const()[name = tensor("op_21155"), val = tensor([1, 1])]; + tensor var_21157 = const()[name = tensor("op_21157"), val = tensor([1, 1])]; + tensor hidden_states_207_pad_type_0 = const()[name = tensor("hidden_states_207_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_207_pad_0 = const()[name = tensor("hidden_states_207_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_proj_in_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_proj_in_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1722829568)))]; + tensor mid_block_attentions_0_proj_in_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_proj_in_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1726106432)))]; + tensor hidden_states_207_cast_fp16 = conv(bias = mid_block_attentions_0_proj_in_bias_to_fp16, dilations = var_21157, groups = var_21077, pad = hidden_states_207_pad_0, pad_type = hidden_states_207_pad_type_0, strides = var_21155, weight = mid_block_attentions_0_proj_in_weight_to_fp16, x = add_37_cast_fp16)[name = tensor("hidden_states_207_cast_fp16")]; + tensor var_21162 = const()[name = tensor("op_21162"), val = tensor([2, 1280, 1, 1024])]; + tensor inputs_145_cast_fp16 = reshape(shape = var_21162, x = hidden_states_207_cast_fp16)[name = tensor("inputs_145_cast_fp16")]; + tensor hidden_states_209_axes_0 = const()[name = tensor("hidden_states_209_axes_0"), val = tensor([1])]; + tensor hidden_states_209_gamma_0_to_fp16 = const()[name = tensor("hidden_states_209_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1726109056)))]; + tensor hidden_states_209_beta_0_to_fp16 = const()[name = tensor("hidden_states_209_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1726111680)))]; + tensor var_21178_to_fp16 = const()[name = tensor("op_21178_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_209_cast_fp16 = layer_norm(axes = hidden_states_209_axes_0, beta = hidden_states_209_beta_0_to_fp16, epsilon = var_21178_to_fp16, gamma = hidden_states_209_gamma_0_to_fp16, x = inputs_145_cast_fp16)[name = tensor("hidden_states_209_cast_fp16")]; + tensor var_21193 = const()[name = tensor("op_21193"), val = tensor([1, 1])]; + tensor var_21195 = const()[name = tensor("op_21195"), val = tensor([1, 1])]; + tensor q_97_pad_type_0 = const()[name = tensor("q_97_pad_type_0"), val = tensor("custom")]; + tensor q_97_pad_0 = const()[name = tensor("q_97_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1726114304)))]; + tensor q_97_cast_fp16 = conv(dilations = var_21195, groups = var_21077, pad = q_97_pad_0, pad_type = q_97_pad_type_0, strides = var_21193, weight = mid_block_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16, x = hidden_states_209_cast_fp16)[name = tensor("q_97_cast_fp16")]; + tensor var_21199 = const()[name = tensor("op_21199"), val = tensor([1, 1])]; + tensor var_21201 = const()[name = tensor("op_21201"), val = tensor([1, 1])]; + tensor k_193_pad_type_0 = const()[name = tensor("k_193_pad_type_0"), val = tensor("custom")]; + tensor k_193_pad_0 = const()[name = tensor("k_193_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1729391168)))]; + tensor k_193_cast_fp16 = conv(dilations = var_21201, groups = var_21077, pad = k_193_pad_0, pad_type = k_193_pad_type_0, strides = var_21199, weight = mid_block_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16, x = hidden_states_209_cast_fp16)[name = tensor("k_193_cast_fp16")]; + tensor var_21205 = const()[name = tensor("op_21205"), val = tensor([1, 1])]; + tensor var_21207 = const()[name = tensor("op_21207"), val = tensor([1, 1])]; + tensor v_97_pad_type_0 = const()[name = tensor("v_97_pad_type_0"), val = tensor("custom")]; + tensor v_97_pad_0 = const()[name = tensor("v_97_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1732668032)))]; + tensor v_97_cast_fp16 = conv(dilations = var_21207, groups = var_21077, pad = v_97_pad_0, pad_type = v_97_pad_type_0, strides = var_21205, weight = mid_block_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16, x = hidden_states_209_cast_fp16)[name = tensor("v_97_cast_fp16")]; + tensor var_21211_begin_0 = const()[name = tensor("op_21211_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_21211_end_0 = const()[name = tensor("op_21211_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_21211_end_mask_0 = const()[name = tensor("op_21211_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21211_cast_fp16 = slice_by_index(begin = var_21211_begin_0, end = var_21211_end_0, end_mask = var_21211_end_mask_0, x = q_97_cast_fp16)[name = tensor("op_21211_cast_fp16")]; + tensor var_21215_begin_0 = const()[name = tensor("op_21215_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_21215_end_0 = const()[name = tensor("op_21215_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_21215_end_mask_0 = const()[name = tensor("op_21215_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21215_cast_fp16 = slice_by_index(begin = var_21215_begin_0, end = var_21215_end_0, end_mask = var_21215_end_mask_0, x = q_97_cast_fp16)[name = tensor("op_21215_cast_fp16")]; + tensor var_21219_begin_0 = const()[name = tensor("op_21219_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_21219_end_0 = const()[name = tensor("op_21219_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_21219_end_mask_0 = const()[name = tensor("op_21219_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21219_cast_fp16 = slice_by_index(begin = var_21219_begin_0, end = var_21219_end_0, end_mask = var_21219_end_mask_0, x = q_97_cast_fp16)[name = tensor("op_21219_cast_fp16")]; + tensor var_21223_begin_0 = const()[name = tensor("op_21223_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_21223_end_0 = const()[name = tensor("op_21223_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_21223_end_mask_0 = const()[name = tensor("op_21223_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21223_cast_fp16 = slice_by_index(begin = var_21223_begin_0, end = var_21223_end_0, end_mask = var_21223_end_mask_0, x = q_97_cast_fp16)[name = tensor("op_21223_cast_fp16")]; + tensor var_21227_begin_0 = const()[name = tensor("op_21227_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_21227_end_0 = const()[name = tensor("op_21227_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_21227_end_mask_0 = const()[name = tensor("op_21227_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21227_cast_fp16 = slice_by_index(begin = var_21227_begin_0, end = var_21227_end_0, end_mask = var_21227_end_mask_0, x = q_97_cast_fp16)[name = tensor("op_21227_cast_fp16")]; + tensor var_21231_begin_0 = const()[name = tensor("op_21231_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_21231_end_0 = const()[name = tensor("op_21231_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_21231_end_mask_0 = const()[name = tensor("op_21231_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21231_cast_fp16 = slice_by_index(begin = var_21231_begin_0, end = var_21231_end_0, end_mask = var_21231_end_mask_0, x = q_97_cast_fp16)[name = tensor("op_21231_cast_fp16")]; + tensor var_21235_begin_0 = const()[name = tensor("op_21235_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_21235_end_0 = const()[name = tensor("op_21235_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_21235_end_mask_0 = const()[name = tensor("op_21235_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21235_cast_fp16 = slice_by_index(begin = var_21235_begin_0, end = var_21235_end_0, end_mask = var_21235_end_mask_0, x = q_97_cast_fp16)[name = tensor("op_21235_cast_fp16")]; + tensor var_21239_begin_0 = const()[name = tensor("op_21239_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_21239_end_0 = const()[name = tensor("op_21239_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_21239_end_mask_0 = const()[name = tensor("op_21239_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21239_cast_fp16 = slice_by_index(begin = var_21239_begin_0, end = var_21239_end_0, end_mask = var_21239_end_mask_0, x = q_97_cast_fp16)[name = tensor("op_21239_cast_fp16")]; + tensor var_21243_begin_0 = const()[name = tensor("op_21243_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_21243_end_0 = const()[name = tensor("op_21243_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_21243_end_mask_0 = const()[name = tensor("op_21243_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21243_cast_fp16 = slice_by_index(begin = var_21243_begin_0, end = var_21243_end_0, end_mask = var_21243_end_mask_0, x = q_97_cast_fp16)[name = tensor("op_21243_cast_fp16")]; + tensor var_21247_begin_0 = const()[name = tensor("op_21247_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_21247_end_0 = const()[name = tensor("op_21247_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_21247_end_mask_0 = const()[name = tensor("op_21247_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21247_cast_fp16 = slice_by_index(begin = var_21247_begin_0, end = var_21247_end_0, end_mask = var_21247_end_mask_0, x = q_97_cast_fp16)[name = tensor("op_21247_cast_fp16")]; + tensor var_21251_begin_0 = const()[name = tensor("op_21251_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_21251_end_0 = const()[name = tensor("op_21251_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_21251_end_mask_0 = const()[name = tensor("op_21251_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21251_cast_fp16 = slice_by_index(begin = var_21251_begin_0, end = var_21251_end_0, end_mask = var_21251_end_mask_0, x = q_97_cast_fp16)[name = tensor("op_21251_cast_fp16")]; + tensor var_21255_begin_0 = const()[name = tensor("op_21255_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_21255_end_0 = const()[name = tensor("op_21255_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_21255_end_mask_0 = const()[name = tensor("op_21255_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21255_cast_fp16 = slice_by_index(begin = var_21255_begin_0, end = var_21255_end_0, end_mask = var_21255_end_mask_0, x = q_97_cast_fp16)[name = tensor("op_21255_cast_fp16")]; + tensor var_21259_begin_0 = const()[name = tensor("op_21259_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_21259_end_0 = const()[name = tensor("op_21259_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_21259_end_mask_0 = const()[name = tensor("op_21259_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21259_cast_fp16 = slice_by_index(begin = var_21259_begin_0, end = var_21259_end_0, end_mask = var_21259_end_mask_0, x = q_97_cast_fp16)[name = tensor("op_21259_cast_fp16")]; + tensor var_21263_begin_0 = const()[name = tensor("op_21263_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_21263_end_0 = const()[name = tensor("op_21263_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_21263_end_mask_0 = const()[name = tensor("op_21263_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21263_cast_fp16 = slice_by_index(begin = var_21263_begin_0, end = var_21263_end_0, end_mask = var_21263_end_mask_0, x = q_97_cast_fp16)[name = tensor("op_21263_cast_fp16")]; + tensor var_21267_begin_0 = const()[name = tensor("op_21267_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_21267_end_0 = const()[name = tensor("op_21267_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_21267_end_mask_0 = const()[name = tensor("op_21267_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21267_cast_fp16 = slice_by_index(begin = var_21267_begin_0, end = var_21267_end_0, end_mask = var_21267_end_mask_0, x = q_97_cast_fp16)[name = tensor("op_21267_cast_fp16")]; + tensor var_21271_begin_0 = const()[name = tensor("op_21271_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_21271_end_0 = const()[name = tensor("op_21271_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_21271_end_mask_0 = const()[name = tensor("op_21271_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21271_cast_fp16 = slice_by_index(begin = var_21271_begin_0, end = var_21271_end_0, end_mask = var_21271_end_mask_0, x = q_97_cast_fp16)[name = tensor("op_21271_cast_fp16")]; + tensor var_21275_begin_0 = const()[name = tensor("op_21275_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_21275_end_0 = const()[name = tensor("op_21275_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_21275_end_mask_0 = const()[name = tensor("op_21275_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21275_cast_fp16 = slice_by_index(begin = var_21275_begin_0, end = var_21275_end_0, end_mask = var_21275_end_mask_0, x = q_97_cast_fp16)[name = tensor("op_21275_cast_fp16")]; + tensor var_21279_begin_0 = const()[name = tensor("op_21279_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_21279_end_0 = const()[name = tensor("op_21279_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_21279_end_mask_0 = const()[name = tensor("op_21279_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21279_cast_fp16 = slice_by_index(begin = var_21279_begin_0, end = var_21279_end_0, end_mask = var_21279_end_mask_0, x = q_97_cast_fp16)[name = tensor("op_21279_cast_fp16")]; + tensor var_21283_begin_0 = const()[name = tensor("op_21283_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_21283_end_0 = const()[name = tensor("op_21283_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_21283_end_mask_0 = const()[name = tensor("op_21283_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21283_cast_fp16 = slice_by_index(begin = var_21283_begin_0, end = var_21283_end_0, end_mask = var_21283_end_mask_0, x = q_97_cast_fp16)[name = tensor("op_21283_cast_fp16")]; + tensor var_21287_begin_0 = const()[name = tensor("op_21287_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_21287_end_0 = const()[name = tensor("op_21287_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_21287_end_mask_0 = const()[name = tensor("op_21287_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21287_cast_fp16 = slice_by_index(begin = var_21287_begin_0, end = var_21287_end_0, end_mask = var_21287_end_mask_0, x = q_97_cast_fp16)[name = tensor("op_21287_cast_fp16")]; + tensor k_195_perm_0 = const()[name = tensor("k_195_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_21294_begin_0 = const()[name = tensor("op_21294_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_21294_end_0 = const()[name = tensor("op_21294_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_21294_end_mask_0 = const()[name = tensor("op_21294_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_195_cast_fp16 = transpose(perm = k_195_perm_0, x = k_193_cast_fp16)[name = tensor("transpose_91")]; + tensor var_21294_cast_fp16 = slice_by_index(begin = var_21294_begin_0, end = var_21294_end_0, end_mask = var_21294_end_mask_0, x = k_195_cast_fp16)[name = tensor("op_21294_cast_fp16")]; + tensor var_21298_begin_0 = const()[name = tensor("op_21298_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_21298_end_0 = const()[name = tensor("op_21298_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_21298_end_mask_0 = const()[name = tensor("op_21298_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21298_cast_fp16 = slice_by_index(begin = var_21298_begin_0, end = var_21298_end_0, end_mask = var_21298_end_mask_0, x = k_195_cast_fp16)[name = tensor("op_21298_cast_fp16")]; + tensor var_21302_begin_0 = const()[name = tensor("op_21302_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_21302_end_0 = const()[name = tensor("op_21302_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_21302_end_mask_0 = const()[name = tensor("op_21302_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21302_cast_fp16 = slice_by_index(begin = var_21302_begin_0, end = var_21302_end_0, end_mask = var_21302_end_mask_0, x = k_195_cast_fp16)[name = tensor("op_21302_cast_fp16")]; + tensor var_21306_begin_0 = const()[name = tensor("op_21306_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_21306_end_0 = const()[name = tensor("op_21306_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_21306_end_mask_0 = const()[name = tensor("op_21306_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21306_cast_fp16 = slice_by_index(begin = var_21306_begin_0, end = var_21306_end_0, end_mask = var_21306_end_mask_0, x = k_195_cast_fp16)[name = tensor("op_21306_cast_fp16")]; + tensor var_21310_begin_0 = const()[name = tensor("op_21310_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_21310_end_0 = const()[name = tensor("op_21310_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_21310_end_mask_0 = const()[name = tensor("op_21310_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21310_cast_fp16 = slice_by_index(begin = var_21310_begin_0, end = var_21310_end_0, end_mask = var_21310_end_mask_0, x = k_195_cast_fp16)[name = tensor("op_21310_cast_fp16")]; + tensor var_21314_begin_0 = const()[name = tensor("op_21314_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_21314_end_0 = const()[name = tensor("op_21314_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_21314_end_mask_0 = const()[name = tensor("op_21314_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21314_cast_fp16 = slice_by_index(begin = var_21314_begin_0, end = var_21314_end_0, end_mask = var_21314_end_mask_0, x = k_195_cast_fp16)[name = tensor("op_21314_cast_fp16")]; + tensor var_21318_begin_0 = const()[name = tensor("op_21318_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_21318_end_0 = const()[name = tensor("op_21318_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_21318_end_mask_0 = const()[name = tensor("op_21318_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21318_cast_fp16 = slice_by_index(begin = var_21318_begin_0, end = var_21318_end_0, end_mask = var_21318_end_mask_0, x = k_195_cast_fp16)[name = tensor("op_21318_cast_fp16")]; + tensor var_21322_begin_0 = const()[name = tensor("op_21322_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_21322_end_0 = const()[name = tensor("op_21322_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_21322_end_mask_0 = const()[name = tensor("op_21322_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21322_cast_fp16 = slice_by_index(begin = var_21322_begin_0, end = var_21322_end_0, end_mask = var_21322_end_mask_0, x = k_195_cast_fp16)[name = tensor("op_21322_cast_fp16")]; + tensor var_21326_begin_0 = const()[name = tensor("op_21326_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_21326_end_0 = const()[name = tensor("op_21326_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_21326_end_mask_0 = const()[name = tensor("op_21326_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21326_cast_fp16 = slice_by_index(begin = var_21326_begin_0, end = var_21326_end_0, end_mask = var_21326_end_mask_0, x = k_195_cast_fp16)[name = tensor("op_21326_cast_fp16")]; + tensor var_21330_begin_0 = const()[name = tensor("op_21330_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_21330_end_0 = const()[name = tensor("op_21330_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_21330_end_mask_0 = const()[name = tensor("op_21330_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21330_cast_fp16 = slice_by_index(begin = var_21330_begin_0, end = var_21330_end_0, end_mask = var_21330_end_mask_0, x = k_195_cast_fp16)[name = tensor("op_21330_cast_fp16")]; + tensor var_21334_begin_0 = const()[name = tensor("op_21334_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_21334_end_0 = const()[name = tensor("op_21334_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_21334_end_mask_0 = const()[name = tensor("op_21334_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21334_cast_fp16 = slice_by_index(begin = var_21334_begin_0, end = var_21334_end_0, end_mask = var_21334_end_mask_0, x = k_195_cast_fp16)[name = tensor("op_21334_cast_fp16")]; + tensor var_21338_begin_0 = const()[name = tensor("op_21338_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_21338_end_0 = const()[name = tensor("op_21338_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_21338_end_mask_0 = const()[name = tensor("op_21338_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21338_cast_fp16 = slice_by_index(begin = var_21338_begin_0, end = var_21338_end_0, end_mask = var_21338_end_mask_0, x = k_195_cast_fp16)[name = tensor("op_21338_cast_fp16")]; + tensor var_21342_begin_0 = const()[name = tensor("op_21342_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_21342_end_0 = const()[name = tensor("op_21342_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_21342_end_mask_0 = const()[name = tensor("op_21342_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21342_cast_fp16 = slice_by_index(begin = var_21342_begin_0, end = var_21342_end_0, end_mask = var_21342_end_mask_0, x = k_195_cast_fp16)[name = tensor("op_21342_cast_fp16")]; + tensor var_21346_begin_0 = const()[name = tensor("op_21346_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_21346_end_0 = const()[name = tensor("op_21346_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_21346_end_mask_0 = const()[name = tensor("op_21346_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21346_cast_fp16 = slice_by_index(begin = var_21346_begin_0, end = var_21346_end_0, end_mask = var_21346_end_mask_0, x = k_195_cast_fp16)[name = tensor("op_21346_cast_fp16")]; + tensor var_21350_begin_0 = const()[name = tensor("op_21350_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_21350_end_0 = const()[name = tensor("op_21350_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_21350_end_mask_0 = const()[name = tensor("op_21350_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21350_cast_fp16 = slice_by_index(begin = var_21350_begin_0, end = var_21350_end_0, end_mask = var_21350_end_mask_0, x = k_195_cast_fp16)[name = tensor("op_21350_cast_fp16")]; + tensor var_21354_begin_0 = const()[name = tensor("op_21354_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_21354_end_0 = const()[name = tensor("op_21354_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_21354_end_mask_0 = const()[name = tensor("op_21354_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21354_cast_fp16 = slice_by_index(begin = var_21354_begin_0, end = var_21354_end_0, end_mask = var_21354_end_mask_0, x = k_195_cast_fp16)[name = tensor("op_21354_cast_fp16")]; + tensor var_21358_begin_0 = const()[name = tensor("op_21358_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_21358_end_0 = const()[name = tensor("op_21358_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_21358_end_mask_0 = const()[name = tensor("op_21358_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21358_cast_fp16 = slice_by_index(begin = var_21358_begin_0, end = var_21358_end_0, end_mask = var_21358_end_mask_0, x = k_195_cast_fp16)[name = tensor("op_21358_cast_fp16")]; + tensor var_21362_begin_0 = const()[name = tensor("op_21362_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_21362_end_0 = const()[name = tensor("op_21362_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_21362_end_mask_0 = const()[name = tensor("op_21362_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21362_cast_fp16 = slice_by_index(begin = var_21362_begin_0, end = var_21362_end_0, end_mask = var_21362_end_mask_0, x = k_195_cast_fp16)[name = tensor("op_21362_cast_fp16")]; + tensor var_21366_begin_0 = const()[name = tensor("op_21366_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_21366_end_0 = const()[name = tensor("op_21366_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_21366_end_mask_0 = const()[name = tensor("op_21366_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21366_cast_fp16 = slice_by_index(begin = var_21366_begin_0, end = var_21366_end_0, end_mask = var_21366_end_mask_0, x = k_195_cast_fp16)[name = tensor("op_21366_cast_fp16")]; + tensor var_21370_begin_0 = const()[name = tensor("op_21370_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_21370_end_0 = const()[name = tensor("op_21370_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_21370_end_mask_0 = const()[name = tensor("op_21370_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21370_cast_fp16 = slice_by_index(begin = var_21370_begin_0, end = var_21370_end_0, end_mask = var_21370_end_mask_0, x = k_195_cast_fp16)[name = tensor("op_21370_cast_fp16")]; + tensor var_21372_begin_0 = const()[name = tensor("op_21372_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_21372_end_0 = const()[name = tensor("op_21372_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_21372_end_mask_0 = const()[name = tensor("op_21372_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21372_cast_fp16 = slice_by_index(begin = var_21372_begin_0, end = var_21372_end_0, end_mask = var_21372_end_mask_0, x = v_97_cast_fp16)[name = tensor("op_21372_cast_fp16")]; + tensor var_21376_begin_0 = const()[name = tensor("op_21376_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_21376_end_0 = const()[name = tensor("op_21376_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_21376_end_mask_0 = const()[name = tensor("op_21376_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21376_cast_fp16 = slice_by_index(begin = var_21376_begin_0, end = var_21376_end_0, end_mask = var_21376_end_mask_0, x = v_97_cast_fp16)[name = tensor("op_21376_cast_fp16")]; + tensor var_21380_begin_0 = const()[name = tensor("op_21380_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_21380_end_0 = const()[name = tensor("op_21380_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_21380_end_mask_0 = const()[name = tensor("op_21380_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21380_cast_fp16 = slice_by_index(begin = var_21380_begin_0, end = var_21380_end_0, end_mask = var_21380_end_mask_0, x = v_97_cast_fp16)[name = tensor("op_21380_cast_fp16")]; + tensor var_21384_begin_0 = const()[name = tensor("op_21384_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_21384_end_0 = const()[name = tensor("op_21384_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_21384_end_mask_0 = const()[name = tensor("op_21384_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21384_cast_fp16 = slice_by_index(begin = var_21384_begin_0, end = var_21384_end_0, end_mask = var_21384_end_mask_0, x = v_97_cast_fp16)[name = tensor("op_21384_cast_fp16")]; + tensor var_21388_begin_0 = const()[name = tensor("op_21388_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_21388_end_0 = const()[name = tensor("op_21388_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_21388_end_mask_0 = const()[name = tensor("op_21388_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21388_cast_fp16 = slice_by_index(begin = var_21388_begin_0, end = var_21388_end_0, end_mask = var_21388_end_mask_0, x = v_97_cast_fp16)[name = tensor("op_21388_cast_fp16")]; + tensor var_21392_begin_0 = const()[name = tensor("op_21392_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_21392_end_0 = const()[name = tensor("op_21392_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_21392_end_mask_0 = const()[name = tensor("op_21392_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21392_cast_fp16 = slice_by_index(begin = var_21392_begin_0, end = var_21392_end_0, end_mask = var_21392_end_mask_0, x = v_97_cast_fp16)[name = tensor("op_21392_cast_fp16")]; + tensor var_21396_begin_0 = const()[name = tensor("op_21396_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_21396_end_0 = const()[name = tensor("op_21396_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_21396_end_mask_0 = const()[name = tensor("op_21396_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21396_cast_fp16 = slice_by_index(begin = var_21396_begin_0, end = var_21396_end_0, end_mask = var_21396_end_mask_0, x = v_97_cast_fp16)[name = tensor("op_21396_cast_fp16")]; + tensor var_21400_begin_0 = const()[name = tensor("op_21400_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_21400_end_0 = const()[name = tensor("op_21400_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_21400_end_mask_0 = const()[name = tensor("op_21400_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21400_cast_fp16 = slice_by_index(begin = var_21400_begin_0, end = var_21400_end_0, end_mask = var_21400_end_mask_0, x = v_97_cast_fp16)[name = tensor("op_21400_cast_fp16")]; + tensor var_21404_begin_0 = const()[name = tensor("op_21404_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_21404_end_0 = const()[name = tensor("op_21404_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_21404_end_mask_0 = const()[name = tensor("op_21404_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21404_cast_fp16 = slice_by_index(begin = var_21404_begin_0, end = var_21404_end_0, end_mask = var_21404_end_mask_0, x = v_97_cast_fp16)[name = tensor("op_21404_cast_fp16")]; + tensor var_21408_begin_0 = const()[name = tensor("op_21408_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_21408_end_0 = const()[name = tensor("op_21408_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_21408_end_mask_0 = const()[name = tensor("op_21408_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21408_cast_fp16 = slice_by_index(begin = var_21408_begin_0, end = var_21408_end_0, end_mask = var_21408_end_mask_0, x = v_97_cast_fp16)[name = tensor("op_21408_cast_fp16")]; + tensor var_21412_begin_0 = const()[name = tensor("op_21412_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_21412_end_0 = const()[name = tensor("op_21412_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_21412_end_mask_0 = const()[name = tensor("op_21412_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21412_cast_fp16 = slice_by_index(begin = var_21412_begin_0, end = var_21412_end_0, end_mask = var_21412_end_mask_0, x = v_97_cast_fp16)[name = tensor("op_21412_cast_fp16")]; + tensor var_21416_begin_0 = const()[name = tensor("op_21416_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_21416_end_0 = const()[name = tensor("op_21416_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_21416_end_mask_0 = const()[name = tensor("op_21416_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21416_cast_fp16 = slice_by_index(begin = var_21416_begin_0, end = var_21416_end_0, end_mask = var_21416_end_mask_0, x = v_97_cast_fp16)[name = tensor("op_21416_cast_fp16")]; + tensor var_21420_begin_0 = const()[name = tensor("op_21420_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_21420_end_0 = const()[name = tensor("op_21420_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_21420_end_mask_0 = const()[name = tensor("op_21420_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21420_cast_fp16 = slice_by_index(begin = var_21420_begin_0, end = var_21420_end_0, end_mask = var_21420_end_mask_0, x = v_97_cast_fp16)[name = tensor("op_21420_cast_fp16")]; + tensor var_21424_begin_0 = const()[name = tensor("op_21424_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_21424_end_0 = const()[name = tensor("op_21424_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_21424_end_mask_0 = const()[name = tensor("op_21424_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21424_cast_fp16 = slice_by_index(begin = var_21424_begin_0, end = var_21424_end_0, end_mask = var_21424_end_mask_0, x = v_97_cast_fp16)[name = tensor("op_21424_cast_fp16")]; + tensor var_21428_begin_0 = const()[name = tensor("op_21428_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_21428_end_0 = const()[name = tensor("op_21428_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_21428_end_mask_0 = const()[name = tensor("op_21428_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21428_cast_fp16 = slice_by_index(begin = var_21428_begin_0, end = var_21428_end_0, end_mask = var_21428_end_mask_0, x = v_97_cast_fp16)[name = tensor("op_21428_cast_fp16")]; + tensor var_21432_begin_0 = const()[name = tensor("op_21432_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_21432_end_0 = const()[name = tensor("op_21432_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_21432_end_mask_0 = const()[name = tensor("op_21432_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21432_cast_fp16 = slice_by_index(begin = var_21432_begin_0, end = var_21432_end_0, end_mask = var_21432_end_mask_0, x = v_97_cast_fp16)[name = tensor("op_21432_cast_fp16")]; + tensor var_21436_begin_0 = const()[name = tensor("op_21436_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_21436_end_0 = const()[name = tensor("op_21436_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_21436_end_mask_0 = const()[name = tensor("op_21436_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21436_cast_fp16 = slice_by_index(begin = var_21436_begin_0, end = var_21436_end_0, end_mask = var_21436_end_mask_0, x = v_97_cast_fp16)[name = tensor("op_21436_cast_fp16")]; + tensor var_21440_begin_0 = const()[name = tensor("op_21440_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_21440_end_0 = const()[name = tensor("op_21440_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_21440_end_mask_0 = const()[name = tensor("op_21440_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21440_cast_fp16 = slice_by_index(begin = var_21440_begin_0, end = var_21440_end_0, end_mask = var_21440_end_mask_0, x = v_97_cast_fp16)[name = tensor("op_21440_cast_fp16")]; + tensor var_21444_begin_0 = const()[name = tensor("op_21444_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_21444_end_0 = const()[name = tensor("op_21444_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_21444_end_mask_0 = const()[name = tensor("op_21444_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21444_cast_fp16 = slice_by_index(begin = var_21444_begin_0, end = var_21444_end_0, end_mask = var_21444_end_mask_0, x = v_97_cast_fp16)[name = tensor("op_21444_cast_fp16")]; + tensor var_21448_begin_0 = const()[name = tensor("op_21448_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_21448_end_0 = const()[name = tensor("op_21448_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_21448_end_mask_0 = const()[name = tensor("op_21448_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21448_cast_fp16 = slice_by_index(begin = var_21448_begin_0, end = var_21448_end_0, end_mask = var_21448_end_mask_0, x = v_97_cast_fp16)[name = tensor("op_21448_cast_fp16")]; + tensor var_21452_equation_0 = const()[name = tensor("op_21452_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21452_cast_fp16 = einsum(equation = var_21452_equation_0, values = (var_21294_cast_fp16, var_21211_cast_fp16))[name = tensor("op_21452_cast_fp16")]; + tensor var_21453_to_fp16 = const()[name = tensor("op_21453_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1761_cast_fp16 = mul(x = var_21452_cast_fp16, y = var_21453_to_fp16)[name = tensor("aw_1761_cast_fp16")]; + tensor var_21456_equation_0 = const()[name = tensor("op_21456_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21456_cast_fp16 = einsum(equation = var_21456_equation_0, values = (var_21298_cast_fp16, var_21215_cast_fp16))[name = tensor("op_21456_cast_fp16")]; + tensor var_21457_to_fp16 = const()[name = tensor("op_21457_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1763_cast_fp16 = mul(x = var_21456_cast_fp16, y = var_21457_to_fp16)[name = tensor("aw_1763_cast_fp16")]; + tensor var_21460_equation_0 = const()[name = tensor("op_21460_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21460_cast_fp16 = einsum(equation = var_21460_equation_0, values = (var_21302_cast_fp16, var_21219_cast_fp16))[name = tensor("op_21460_cast_fp16")]; + tensor var_21461_to_fp16 = const()[name = tensor("op_21461_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1765_cast_fp16 = mul(x = var_21460_cast_fp16, y = var_21461_to_fp16)[name = tensor("aw_1765_cast_fp16")]; + tensor var_21464_equation_0 = const()[name = tensor("op_21464_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21464_cast_fp16 = einsum(equation = var_21464_equation_0, values = (var_21306_cast_fp16, var_21223_cast_fp16))[name = tensor("op_21464_cast_fp16")]; + tensor var_21465_to_fp16 = const()[name = tensor("op_21465_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1767_cast_fp16 = mul(x = var_21464_cast_fp16, y = var_21465_to_fp16)[name = tensor("aw_1767_cast_fp16")]; + tensor var_21468_equation_0 = const()[name = tensor("op_21468_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21468_cast_fp16 = einsum(equation = var_21468_equation_0, values = (var_21310_cast_fp16, var_21227_cast_fp16))[name = tensor("op_21468_cast_fp16")]; + tensor var_21469_to_fp16 = const()[name = tensor("op_21469_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1769_cast_fp16 = mul(x = var_21468_cast_fp16, y = var_21469_to_fp16)[name = tensor("aw_1769_cast_fp16")]; + tensor var_21472_equation_0 = const()[name = tensor("op_21472_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21472_cast_fp16 = einsum(equation = var_21472_equation_0, values = (var_21314_cast_fp16, var_21231_cast_fp16))[name = tensor("op_21472_cast_fp16")]; + tensor var_21473_to_fp16 = const()[name = tensor("op_21473_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1771_cast_fp16 = mul(x = var_21472_cast_fp16, y = var_21473_to_fp16)[name = tensor("aw_1771_cast_fp16")]; + tensor var_21476_equation_0 = const()[name = tensor("op_21476_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21476_cast_fp16 = einsum(equation = var_21476_equation_0, values = (var_21318_cast_fp16, var_21235_cast_fp16))[name = tensor("op_21476_cast_fp16")]; + tensor var_21477_to_fp16 = const()[name = tensor("op_21477_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1773_cast_fp16 = mul(x = var_21476_cast_fp16, y = var_21477_to_fp16)[name = tensor("aw_1773_cast_fp16")]; + tensor var_21480_equation_0 = const()[name = tensor("op_21480_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21480_cast_fp16 = einsum(equation = var_21480_equation_0, values = (var_21322_cast_fp16, var_21239_cast_fp16))[name = tensor("op_21480_cast_fp16")]; + tensor var_21481_to_fp16 = const()[name = tensor("op_21481_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1775_cast_fp16 = mul(x = var_21480_cast_fp16, y = var_21481_to_fp16)[name = tensor("aw_1775_cast_fp16")]; + tensor var_21484_equation_0 = const()[name = tensor("op_21484_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21484_cast_fp16 = einsum(equation = var_21484_equation_0, values = (var_21326_cast_fp16, var_21243_cast_fp16))[name = tensor("op_21484_cast_fp16")]; + tensor var_21485_to_fp16 = const()[name = tensor("op_21485_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1777_cast_fp16 = mul(x = var_21484_cast_fp16, y = var_21485_to_fp16)[name = tensor("aw_1777_cast_fp16")]; + tensor var_21488_equation_0 = const()[name = tensor("op_21488_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21488_cast_fp16 = einsum(equation = var_21488_equation_0, values = (var_21330_cast_fp16, var_21247_cast_fp16))[name = tensor("op_21488_cast_fp16")]; + tensor var_21489_to_fp16 = const()[name = tensor("op_21489_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1779_cast_fp16 = mul(x = var_21488_cast_fp16, y = var_21489_to_fp16)[name = tensor("aw_1779_cast_fp16")]; + tensor var_21492_equation_0 = const()[name = tensor("op_21492_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21492_cast_fp16 = einsum(equation = var_21492_equation_0, values = (var_21334_cast_fp16, var_21251_cast_fp16))[name = tensor("op_21492_cast_fp16")]; + tensor var_21493_to_fp16 = const()[name = tensor("op_21493_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1781_cast_fp16 = mul(x = var_21492_cast_fp16, y = var_21493_to_fp16)[name = tensor("aw_1781_cast_fp16")]; + tensor var_21496_equation_0 = const()[name = tensor("op_21496_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21496_cast_fp16 = einsum(equation = var_21496_equation_0, values = (var_21338_cast_fp16, var_21255_cast_fp16))[name = tensor("op_21496_cast_fp16")]; + tensor var_21497_to_fp16 = const()[name = tensor("op_21497_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1783_cast_fp16 = mul(x = var_21496_cast_fp16, y = var_21497_to_fp16)[name = tensor("aw_1783_cast_fp16")]; + tensor var_21500_equation_0 = const()[name = tensor("op_21500_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21500_cast_fp16 = einsum(equation = var_21500_equation_0, values = (var_21342_cast_fp16, var_21259_cast_fp16))[name = tensor("op_21500_cast_fp16")]; + tensor var_21501_to_fp16 = const()[name = tensor("op_21501_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1785_cast_fp16 = mul(x = var_21500_cast_fp16, y = var_21501_to_fp16)[name = tensor("aw_1785_cast_fp16")]; + tensor var_21504_equation_0 = const()[name = tensor("op_21504_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21504_cast_fp16 = einsum(equation = var_21504_equation_0, values = (var_21346_cast_fp16, var_21263_cast_fp16))[name = tensor("op_21504_cast_fp16")]; + tensor var_21505_to_fp16 = const()[name = tensor("op_21505_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1787_cast_fp16 = mul(x = var_21504_cast_fp16, y = var_21505_to_fp16)[name = tensor("aw_1787_cast_fp16")]; + tensor var_21508_equation_0 = const()[name = tensor("op_21508_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21508_cast_fp16 = einsum(equation = var_21508_equation_0, values = (var_21350_cast_fp16, var_21267_cast_fp16))[name = tensor("op_21508_cast_fp16")]; + tensor var_21509_to_fp16 = const()[name = tensor("op_21509_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1789_cast_fp16 = mul(x = var_21508_cast_fp16, y = var_21509_to_fp16)[name = tensor("aw_1789_cast_fp16")]; + tensor var_21512_equation_0 = const()[name = tensor("op_21512_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21512_cast_fp16 = einsum(equation = var_21512_equation_0, values = (var_21354_cast_fp16, var_21271_cast_fp16))[name = tensor("op_21512_cast_fp16")]; + tensor var_21513_to_fp16 = const()[name = tensor("op_21513_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1791_cast_fp16 = mul(x = var_21512_cast_fp16, y = var_21513_to_fp16)[name = tensor("aw_1791_cast_fp16")]; + tensor var_21516_equation_0 = const()[name = tensor("op_21516_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21516_cast_fp16 = einsum(equation = var_21516_equation_0, values = (var_21358_cast_fp16, var_21275_cast_fp16))[name = tensor("op_21516_cast_fp16")]; + tensor var_21517_to_fp16 = const()[name = tensor("op_21517_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1793_cast_fp16 = mul(x = var_21516_cast_fp16, y = var_21517_to_fp16)[name = tensor("aw_1793_cast_fp16")]; + tensor var_21520_equation_0 = const()[name = tensor("op_21520_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21520_cast_fp16 = einsum(equation = var_21520_equation_0, values = (var_21362_cast_fp16, var_21279_cast_fp16))[name = tensor("op_21520_cast_fp16")]; + tensor var_21521_to_fp16 = const()[name = tensor("op_21521_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1795_cast_fp16 = mul(x = var_21520_cast_fp16, y = var_21521_to_fp16)[name = tensor("aw_1795_cast_fp16")]; + tensor var_21524_equation_0 = const()[name = tensor("op_21524_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21524_cast_fp16 = einsum(equation = var_21524_equation_0, values = (var_21366_cast_fp16, var_21283_cast_fp16))[name = tensor("op_21524_cast_fp16")]; + tensor var_21525_to_fp16 = const()[name = tensor("op_21525_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1797_cast_fp16 = mul(x = var_21524_cast_fp16, y = var_21525_to_fp16)[name = tensor("aw_1797_cast_fp16")]; + tensor var_21528_equation_0 = const()[name = tensor("op_21528_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21528_cast_fp16 = einsum(equation = var_21528_equation_0, values = (var_21370_cast_fp16, var_21287_cast_fp16))[name = tensor("op_21528_cast_fp16")]; + tensor var_21529_to_fp16 = const()[name = tensor("op_21529_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1799_cast_fp16 = mul(x = var_21528_cast_fp16, y = var_21529_to_fp16)[name = tensor("aw_1799_cast_fp16")]; + tensor var_21531_cast_fp16 = softmax(axis = var_21077, x = aw_1761_cast_fp16)[name = tensor("op_21531_cast_fp16")]; + tensor var_21532_cast_fp16 = softmax(axis = var_21077, x = aw_1763_cast_fp16)[name = tensor("op_21532_cast_fp16")]; + tensor var_21533_cast_fp16 = softmax(axis = var_21077, x = aw_1765_cast_fp16)[name = tensor("op_21533_cast_fp16")]; + tensor var_21534_cast_fp16 = softmax(axis = var_21077, x = aw_1767_cast_fp16)[name = tensor("op_21534_cast_fp16")]; + tensor var_21535_cast_fp16 = softmax(axis = var_21077, x = aw_1769_cast_fp16)[name = tensor("op_21535_cast_fp16")]; + tensor var_21536_cast_fp16 = softmax(axis = var_21077, x = aw_1771_cast_fp16)[name = tensor("op_21536_cast_fp16")]; + tensor var_21537_cast_fp16 = softmax(axis = var_21077, x = aw_1773_cast_fp16)[name = tensor("op_21537_cast_fp16")]; + tensor var_21538_cast_fp16 = softmax(axis = var_21077, x = aw_1775_cast_fp16)[name = tensor("op_21538_cast_fp16")]; + tensor var_21539_cast_fp16 = softmax(axis = var_21077, x = aw_1777_cast_fp16)[name = tensor("op_21539_cast_fp16")]; + tensor var_21540_cast_fp16 = softmax(axis = var_21077, x = aw_1779_cast_fp16)[name = tensor("op_21540_cast_fp16")]; + tensor var_21541_cast_fp16 = softmax(axis = var_21077, x = aw_1781_cast_fp16)[name = tensor("op_21541_cast_fp16")]; + tensor var_21542_cast_fp16 = softmax(axis = var_21077, x = aw_1783_cast_fp16)[name = tensor("op_21542_cast_fp16")]; + tensor var_21543_cast_fp16 = softmax(axis = var_21077, x = aw_1785_cast_fp16)[name = tensor("op_21543_cast_fp16")]; + tensor var_21544_cast_fp16 = softmax(axis = var_21077, x = aw_1787_cast_fp16)[name = tensor("op_21544_cast_fp16")]; + tensor var_21545_cast_fp16 = softmax(axis = var_21077, x = aw_1789_cast_fp16)[name = tensor("op_21545_cast_fp16")]; + tensor var_21546_cast_fp16 = softmax(axis = var_21077, x = aw_1791_cast_fp16)[name = tensor("op_21546_cast_fp16")]; + tensor var_21547_cast_fp16 = softmax(axis = var_21077, x = aw_1793_cast_fp16)[name = tensor("op_21547_cast_fp16")]; + tensor var_21548_cast_fp16 = softmax(axis = var_21077, x = aw_1795_cast_fp16)[name = tensor("op_21548_cast_fp16")]; + tensor var_21549_cast_fp16 = softmax(axis = var_21077, x = aw_1797_cast_fp16)[name = tensor("op_21549_cast_fp16")]; + tensor var_21550_cast_fp16 = softmax(axis = var_21077, x = aw_1799_cast_fp16)[name = tensor("op_21550_cast_fp16")]; + tensor var_21552_equation_0 = const()[name = tensor("op_21552_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21552_cast_fp16 = einsum(equation = var_21552_equation_0, values = (var_21372_cast_fp16, var_21531_cast_fp16))[name = tensor("op_21552_cast_fp16")]; + tensor var_21554_equation_0 = const()[name = tensor("op_21554_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21554_cast_fp16 = einsum(equation = var_21554_equation_0, values = (var_21376_cast_fp16, var_21532_cast_fp16))[name = tensor("op_21554_cast_fp16")]; + tensor var_21556_equation_0 = const()[name = tensor("op_21556_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21556_cast_fp16 = einsum(equation = var_21556_equation_0, values = (var_21380_cast_fp16, var_21533_cast_fp16))[name = tensor("op_21556_cast_fp16")]; + tensor var_21558_equation_0 = const()[name = tensor("op_21558_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21558_cast_fp16 = einsum(equation = var_21558_equation_0, values = (var_21384_cast_fp16, var_21534_cast_fp16))[name = tensor("op_21558_cast_fp16")]; + tensor var_21560_equation_0 = const()[name = tensor("op_21560_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21560_cast_fp16 = einsum(equation = var_21560_equation_0, values = (var_21388_cast_fp16, var_21535_cast_fp16))[name = tensor("op_21560_cast_fp16")]; + tensor var_21562_equation_0 = const()[name = tensor("op_21562_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21562_cast_fp16 = einsum(equation = var_21562_equation_0, values = (var_21392_cast_fp16, var_21536_cast_fp16))[name = tensor("op_21562_cast_fp16")]; + tensor var_21564_equation_0 = const()[name = tensor("op_21564_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21564_cast_fp16 = einsum(equation = var_21564_equation_0, values = (var_21396_cast_fp16, var_21537_cast_fp16))[name = tensor("op_21564_cast_fp16")]; + tensor var_21566_equation_0 = const()[name = tensor("op_21566_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21566_cast_fp16 = einsum(equation = var_21566_equation_0, values = (var_21400_cast_fp16, var_21538_cast_fp16))[name = tensor("op_21566_cast_fp16")]; + tensor var_21568_equation_0 = const()[name = tensor("op_21568_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21568_cast_fp16 = einsum(equation = var_21568_equation_0, values = (var_21404_cast_fp16, var_21539_cast_fp16))[name = tensor("op_21568_cast_fp16")]; + tensor var_21570_equation_0 = const()[name = tensor("op_21570_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21570_cast_fp16 = einsum(equation = var_21570_equation_0, values = (var_21408_cast_fp16, var_21540_cast_fp16))[name = tensor("op_21570_cast_fp16")]; + tensor var_21572_equation_0 = const()[name = tensor("op_21572_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21572_cast_fp16 = einsum(equation = var_21572_equation_0, values = (var_21412_cast_fp16, var_21541_cast_fp16))[name = tensor("op_21572_cast_fp16")]; + tensor var_21574_equation_0 = const()[name = tensor("op_21574_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21574_cast_fp16 = einsum(equation = var_21574_equation_0, values = (var_21416_cast_fp16, var_21542_cast_fp16))[name = tensor("op_21574_cast_fp16")]; + tensor var_21576_equation_0 = const()[name = tensor("op_21576_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21576_cast_fp16 = einsum(equation = var_21576_equation_0, values = (var_21420_cast_fp16, var_21543_cast_fp16))[name = tensor("op_21576_cast_fp16")]; + tensor var_21578_equation_0 = const()[name = tensor("op_21578_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21578_cast_fp16 = einsum(equation = var_21578_equation_0, values = (var_21424_cast_fp16, var_21544_cast_fp16))[name = tensor("op_21578_cast_fp16")]; + tensor var_21580_equation_0 = const()[name = tensor("op_21580_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21580_cast_fp16 = einsum(equation = var_21580_equation_0, values = (var_21428_cast_fp16, var_21545_cast_fp16))[name = tensor("op_21580_cast_fp16")]; + tensor var_21582_equation_0 = const()[name = tensor("op_21582_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21582_cast_fp16 = einsum(equation = var_21582_equation_0, values = (var_21432_cast_fp16, var_21546_cast_fp16))[name = tensor("op_21582_cast_fp16")]; + tensor var_21584_equation_0 = const()[name = tensor("op_21584_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21584_cast_fp16 = einsum(equation = var_21584_equation_0, values = (var_21436_cast_fp16, var_21547_cast_fp16))[name = tensor("op_21584_cast_fp16")]; + tensor var_21586_equation_0 = const()[name = tensor("op_21586_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21586_cast_fp16 = einsum(equation = var_21586_equation_0, values = (var_21440_cast_fp16, var_21548_cast_fp16))[name = tensor("op_21586_cast_fp16")]; + tensor var_21588_equation_0 = const()[name = tensor("op_21588_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21588_cast_fp16 = einsum(equation = var_21588_equation_0, values = (var_21444_cast_fp16, var_21549_cast_fp16))[name = tensor("op_21588_cast_fp16")]; + tensor var_21590_equation_0 = const()[name = tensor("op_21590_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21590_cast_fp16 = einsum(equation = var_21590_equation_0, values = (var_21448_cast_fp16, var_21550_cast_fp16))[name = tensor("op_21590_cast_fp16")]; + tensor input_327_interleave_0 = const()[name = tensor("input_327_interleave_0"), val = tensor(false)]; + tensor input_327_cast_fp16 = concat(axis = var_21077, interleave = input_327_interleave_0, values = (var_21552_cast_fp16, var_21554_cast_fp16, var_21556_cast_fp16, var_21558_cast_fp16, var_21560_cast_fp16, var_21562_cast_fp16, var_21564_cast_fp16, var_21566_cast_fp16, var_21568_cast_fp16, var_21570_cast_fp16, var_21572_cast_fp16, var_21574_cast_fp16, var_21576_cast_fp16, var_21578_cast_fp16, var_21580_cast_fp16, var_21582_cast_fp16, var_21584_cast_fp16, var_21586_cast_fp16, var_21588_cast_fp16, var_21590_cast_fp16))[name = tensor("input_327_cast_fp16")]; + tensor var_21596 = const()[name = tensor("op_21596"), val = tensor([1, 1])]; + tensor var_21598 = const()[name = tensor("op_21598"), val = tensor([1, 1])]; + tensor var_21600_pad_type_0 = const()[name = tensor("op_21600_pad_type_0"), val = tensor("custom")]; + tensor var_21600_pad_0 = const()[name = tensor("op_21600_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1735944896)))]; + tensor mid_block_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1739221760)))]; + tensor var_21600_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_21598, groups = var_21077, pad = var_21600_pad_0, pad_type = var_21600_pad_type_0, strides = var_21596, weight = mid_block_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16, x = input_327_cast_fp16)[name = tensor("op_21600_cast_fp16")]; + tensor inputs_147_cast_fp16 = add(x = var_21600_cast_fp16, y = inputs_145_cast_fp16)[name = tensor("inputs_147_cast_fp16")]; + tensor hidden_states_211_axes_0 = const()[name = tensor("hidden_states_211_axes_0"), val = tensor([1])]; + tensor hidden_states_211_gamma_0_to_fp16 = const()[name = tensor("hidden_states_211_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1739224384)))]; + tensor hidden_states_211_beta_0_to_fp16 = const()[name = tensor("hidden_states_211_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1739227008)))]; + tensor var_21610_to_fp16 = const()[name = tensor("op_21610_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_211_cast_fp16 = layer_norm(axes = hidden_states_211_axes_0, beta = hidden_states_211_beta_0_to_fp16, epsilon = var_21610_to_fp16, gamma = hidden_states_211_gamma_0_to_fp16, x = inputs_147_cast_fp16)[name = tensor("hidden_states_211_cast_fp16")]; + tensor var_21625 = const()[name = tensor("op_21625"), val = tensor([1, 1])]; + tensor var_21627 = const()[name = tensor("op_21627"), val = tensor([1, 1])]; + tensor q_99_pad_type_0 = const()[name = tensor("q_99_pad_type_0"), val = tensor("custom")]; + tensor q_99_pad_0 = const()[name = tensor("q_99_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1739229632)))]; + tensor q_99_cast_fp16 = conv(dilations = var_21627, groups = var_21077, pad = q_99_pad_0, pad_type = q_99_pad_type_0, strides = var_21625, weight = mid_block_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16, x = hidden_states_211_cast_fp16)[name = tensor("q_99_cast_fp16")]; + tensor var_21631 = const()[name = tensor("op_21631"), val = tensor([1, 1])]; + tensor var_21633 = const()[name = tensor("op_21633"), val = tensor([1, 1])]; + tensor k_197_pad_type_0 = const()[name = tensor("k_197_pad_type_0"), val = tensor("custom")]; + tensor k_197_pad_0 = const()[name = tensor("k_197_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1742506496)))]; + tensor k_197_cast_fp16 = conv(dilations = var_21633, groups = var_21077, pad = k_197_pad_0, pad_type = k_197_pad_type_0, strides = var_21631, weight = mid_block_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_197_cast_fp16")]; + tensor var_21637 = const()[name = tensor("op_21637"), val = tensor([1, 1])]; + tensor var_21639 = const()[name = tensor("op_21639"), val = tensor([1, 1])]; + tensor v_99_pad_type_0 = const()[name = tensor("v_99_pad_type_0"), val = tensor("custom")]; + tensor v_99_pad_0 = const()[name = tensor("v_99_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1747749440)))]; + tensor v_99_cast_fp16 = conv(dilations = var_21639, groups = var_21077, pad = v_99_pad_0, pad_type = v_99_pad_type_0, strides = var_21637, weight = mid_block_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_99_cast_fp16")]; + tensor var_21643_begin_0 = const()[name = tensor("op_21643_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_21643_end_0 = const()[name = tensor("op_21643_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_21643_end_mask_0 = const()[name = tensor("op_21643_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21643_cast_fp16 = slice_by_index(begin = var_21643_begin_0, end = var_21643_end_0, end_mask = var_21643_end_mask_0, x = q_99_cast_fp16)[name = tensor("op_21643_cast_fp16")]; + tensor var_21647_begin_0 = const()[name = tensor("op_21647_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_21647_end_0 = const()[name = tensor("op_21647_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_21647_end_mask_0 = const()[name = tensor("op_21647_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21647_cast_fp16 = slice_by_index(begin = var_21647_begin_0, end = var_21647_end_0, end_mask = var_21647_end_mask_0, x = q_99_cast_fp16)[name = tensor("op_21647_cast_fp16")]; + tensor var_21651_begin_0 = const()[name = tensor("op_21651_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_21651_end_0 = const()[name = tensor("op_21651_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_21651_end_mask_0 = const()[name = tensor("op_21651_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21651_cast_fp16 = slice_by_index(begin = var_21651_begin_0, end = var_21651_end_0, end_mask = var_21651_end_mask_0, x = q_99_cast_fp16)[name = tensor("op_21651_cast_fp16")]; + tensor var_21655_begin_0 = const()[name = tensor("op_21655_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_21655_end_0 = const()[name = tensor("op_21655_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_21655_end_mask_0 = const()[name = tensor("op_21655_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21655_cast_fp16 = slice_by_index(begin = var_21655_begin_0, end = var_21655_end_0, end_mask = var_21655_end_mask_0, x = q_99_cast_fp16)[name = tensor("op_21655_cast_fp16")]; + tensor var_21659_begin_0 = const()[name = tensor("op_21659_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_21659_end_0 = const()[name = tensor("op_21659_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_21659_end_mask_0 = const()[name = tensor("op_21659_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21659_cast_fp16 = slice_by_index(begin = var_21659_begin_0, end = var_21659_end_0, end_mask = var_21659_end_mask_0, x = q_99_cast_fp16)[name = tensor("op_21659_cast_fp16")]; + tensor var_21663_begin_0 = const()[name = tensor("op_21663_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_21663_end_0 = const()[name = tensor("op_21663_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_21663_end_mask_0 = const()[name = tensor("op_21663_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21663_cast_fp16 = slice_by_index(begin = var_21663_begin_0, end = var_21663_end_0, end_mask = var_21663_end_mask_0, x = q_99_cast_fp16)[name = tensor("op_21663_cast_fp16")]; + tensor var_21667_begin_0 = const()[name = tensor("op_21667_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_21667_end_0 = const()[name = tensor("op_21667_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_21667_end_mask_0 = const()[name = tensor("op_21667_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21667_cast_fp16 = slice_by_index(begin = var_21667_begin_0, end = var_21667_end_0, end_mask = var_21667_end_mask_0, x = q_99_cast_fp16)[name = tensor("op_21667_cast_fp16")]; + tensor var_21671_begin_0 = const()[name = tensor("op_21671_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_21671_end_0 = const()[name = tensor("op_21671_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_21671_end_mask_0 = const()[name = tensor("op_21671_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21671_cast_fp16 = slice_by_index(begin = var_21671_begin_0, end = var_21671_end_0, end_mask = var_21671_end_mask_0, x = q_99_cast_fp16)[name = tensor("op_21671_cast_fp16")]; + tensor var_21675_begin_0 = const()[name = tensor("op_21675_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_21675_end_0 = const()[name = tensor("op_21675_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_21675_end_mask_0 = const()[name = tensor("op_21675_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21675_cast_fp16 = slice_by_index(begin = var_21675_begin_0, end = var_21675_end_0, end_mask = var_21675_end_mask_0, x = q_99_cast_fp16)[name = tensor("op_21675_cast_fp16")]; + tensor var_21679_begin_0 = const()[name = tensor("op_21679_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_21679_end_0 = const()[name = tensor("op_21679_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_21679_end_mask_0 = const()[name = tensor("op_21679_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21679_cast_fp16 = slice_by_index(begin = var_21679_begin_0, end = var_21679_end_0, end_mask = var_21679_end_mask_0, x = q_99_cast_fp16)[name = tensor("op_21679_cast_fp16")]; + tensor var_21683_begin_0 = const()[name = tensor("op_21683_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_21683_end_0 = const()[name = tensor("op_21683_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_21683_end_mask_0 = const()[name = tensor("op_21683_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21683_cast_fp16 = slice_by_index(begin = var_21683_begin_0, end = var_21683_end_0, end_mask = var_21683_end_mask_0, x = q_99_cast_fp16)[name = tensor("op_21683_cast_fp16")]; + tensor var_21687_begin_0 = const()[name = tensor("op_21687_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_21687_end_0 = const()[name = tensor("op_21687_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_21687_end_mask_0 = const()[name = tensor("op_21687_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21687_cast_fp16 = slice_by_index(begin = var_21687_begin_0, end = var_21687_end_0, end_mask = var_21687_end_mask_0, x = q_99_cast_fp16)[name = tensor("op_21687_cast_fp16")]; + tensor var_21691_begin_0 = const()[name = tensor("op_21691_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_21691_end_0 = const()[name = tensor("op_21691_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_21691_end_mask_0 = const()[name = tensor("op_21691_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21691_cast_fp16 = slice_by_index(begin = var_21691_begin_0, end = var_21691_end_0, end_mask = var_21691_end_mask_0, x = q_99_cast_fp16)[name = tensor("op_21691_cast_fp16")]; + tensor var_21695_begin_0 = const()[name = tensor("op_21695_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_21695_end_0 = const()[name = tensor("op_21695_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_21695_end_mask_0 = const()[name = tensor("op_21695_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21695_cast_fp16 = slice_by_index(begin = var_21695_begin_0, end = var_21695_end_0, end_mask = var_21695_end_mask_0, x = q_99_cast_fp16)[name = tensor("op_21695_cast_fp16")]; + tensor var_21699_begin_0 = const()[name = tensor("op_21699_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_21699_end_0 = const()[name = tensor("op_21699_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_21699_end_mask_0 = const()[name = tensor("op_21699_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21699_cast_fp16 = slice_by_index(begin = var_21699_begin_0, end = var_21699_end_0, end_mask = var_21699_end_mask_0, x = q_99_cast_fp16)[name = tensor("op_21699_cast_fp16")]; + tensor var_21703_begin_0 = const()[name = tensor("op_21703_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_21703_end_0 = const()[name = tensor("op_21703_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_21703_end_mask_0 = const()[name = tensor("op_21703_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21703_cast_fp16 = slice_by_index(begin = var_21703_begin_0, end = var_21703_end_0, end_mask = var_21703_end_mask_0, x = q_99_cast_fp16)[name = tensor("op_21703_cast_fp16")]; + tensor var_21707_begin_0 = const()[name = tensor("op_21707_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_21707_end_0 = const()[name = tensor("op_21707_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_21707_end_mask_0 = const()[name = tensor("op_21707_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21707_cast_fp16 = slice_by_index(begin = var_21707_begin_0, end = var_21707_end_0, end_mask = var_21707_end_mask_0, x = q_99_cast_fp16)[name = tensor("op_21707_cast_fp16")]; + tensor var_21711_begin_0 = const()[name = tensor("op_21711_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_21711_end_0 = const()[name = tensor("op_21711_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_21711_end_mask_0 = const()[name = tensor("op_21711_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21711_cast_fp16 = slice_by_index(begin = var_21711_begin_0, end = var_21711_end_0, end_mask = var_21711_end_mask_0, x = q_99_cast_fp16)[name = tensor("op_21711_cast_fp16")]; + tensor var_21715_begin_0 = const()[name = tensor("op_21715_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_21715_end_0 = const()[name = tensor("op_21715_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_21715_end_mask_0 = const()[name = tensor("op_21715_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21715_cast_fp16 = slice_by_index(begin = var_21715_begin_0, end = var_21715_end_0, end_mask = var_21715_end_mask_0, x = q_99_cast_fp16)[name = tensor("op_21715_cast_fp16")]; + tensor var_21719_begin_0 = const()[name = tensor("op_21719_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_21719_end_0 = const()[name = tensor("op_21719_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_21719_end_mask_0 = const()[name = tensor("op_21719_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21719_cast_fp16 = slice_by_index(begin = var_21719_begin_0, end = var_21719_end_0, end_mask = var_21719_end_mask_0, x = q_99_cast_fp16)[name = tensor("op_21719_cast_fp16")]; + tensor k_199_perm_0 = const()[name = tensor("k_199_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_21726_begin_0 = const()[name = tensor("op_21726_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_21726_end_0 = const()[name = tensor("op_21726_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_21726_end_mask_0 = const()[name = tensor("op_21726_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_199_cast_fp16 = transpose(perm = k_199_perm_0, x = k_197_cast_fp16)[name = tensor("transpose_90")]; + tensor var_21726_cast_fp16 = slice_by_index(begin = var_21726_begin_0, end = var_21726_end_0, end_mask = var_21726_end_mask_0, x = k_199_cast_fp16)[name = tensor("op_21726_cast_fp16")]; + tensor var_21730_begin_0 = const()[name = tensor("op_21730_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_21730_end_0 = const()[name = tensor("op_21730_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_21730_end_mask_0 = const()[name = tensor("op_21730_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21730_cast_fp16 = slice_by_index(begin = var_21730_begin_0, end = var_21730_end_0, end_mask = var_21730_end_mask_0, x = k_199_cast_fp16)[name = tensor("op_21730_cast_fp16")]; + tensor var_21734_begin_0 = const()[name = tensor("op_21734_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_21734_end_0 = const()[name = tensor("op_21734_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_21734_end_mask_0 = const()[name = tensor("op_21734_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21734_cast_fp16 = slice_by_index(begin = var_21734_begin_0, end = var_21734_end_0, end_mask = var_21734_end_mask_0, x = k_199_cast_fp16)[name = tensor("op_21734_cast_fp16")]; + tensor var_21738_begin_0 = const()[name = tensor("op_21738_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_21738_end_0 = const()[name = tensor("op_21738_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_21738_end_mask_0 = const()[name = tensor("op_21738_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21738_cast_fp16 = slice_by_index(begin = var_21738_begin_0, end = var_21738_end_0, end_mask = var_21738_end_mask_0, x = k_199_cast_fp16)[name = tensor("op_21738_cast_fp16")]; + tensor var_21742_begin_0 = const()[name = tensor("op_21742_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_21742_end_0 = const()[name = tensor("op_21742_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_21742_end_mask_0 = const()[name = tensor("op_21742_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21742_cast_fp16 = slice_by_index(begin = var_21742_begin_0, end = var_21742_end_0, end_mask = var_21742_end_mask_0, x = k_199_cast_fp16)[name = tensor("op_21742_cast_fp16")]; + tensor var_21746_begin_0 = const()[name = tensor("op_21746_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_21746_end_0 = const()[name = tensor("op_21746_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_21746_end_mask_0 = const()[name = tensor("op_21746_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21746_cast_fp16 = slice_by_index(begin = var_21746_begin_0, end = var_21746_end_0, end_mask = var_21746_end_mask_0, x = k_199_cast_fp16)[name = tensor("op_21746_cast_fp16")]; + tensor var_21750_begin_0 = const()[name = tensor("op_21750_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_21750_end_0 = const()[name = tensor("op_21750_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_21750_end_mask_0 = const()[name = tensor("op_21750_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21750_cast_fp16 = slice_by_index(begin = var_21750_begin_0, end = var_21750_end_0, end_mask = var_21750_end_mask_0, x = k_199_cast_fp16)[name = tensor("op_21750_cast_fp16")]; + tensor var_21754_begin_0 = const()[name = tensor("op_21754_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_21754_end_0 = const()[name = tensor("op_21754_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_21754_end_mask_0 = const()[name = tensor("op_21754_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21754_cast_fp16 = slice_by_index(begin = var_21754_begin_0, end = var_21754_end_0, end_mask = var_21754_end_mask_0, x = k_199_cast_fp16)[name = tensor("op_21754_cast_fp16")]; + tensor var_21758_begin_0 = const()[name = tensor("op_21758_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_21758_end_0 = const()[name = tensor("op_21758_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_21758_end_mask_0 = const()[name = tensor("op_21758_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21758_cast_fp16 = slice_by_index(begin = var_21758_begin_0, end = var_21758_end_0, end_mask = var_21758_end_mask_0, x = k_199_cast_fp16)[name = tensor("op_21758_cast_fp16")]; + tensor var_21762_begin_0 = const()[name = tensor("op_21762_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_21762_end_0 = const()[name = tensor("op_21762_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_21762_end_mask_0 = const()[name = tensor("op_21762_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21762_cast_fp16 = slice_by_index(begin = var_21762_begin_0, end = var_21762_end_0, end_mask = var_21762_end_mask_0, x = k_199_cast_fp16)[name = tensor("op_21762_cast_fp16")]; + tensor var_21766_begin_0 = const()[name = tensor("op_21766_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_21766_end_0 = const()[name = tensor("op_21766_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_21766_end_mask_0 = const()[name = tensor("op_21766_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21766_cast_fp16 = slice_by_index(begin = var_21766_begin_0, end = var_21766_end_0, end_mask = var_21766_end_mask_0, x = k_199_cast_fp16)[name = tensor("op_21766_cast_fp16")]; + tensor var_21770_begin_0 = const()[name = tensor("op_21770_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_21770_end_0 = const()[name = tensor("op_21770_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_21770_end_mask_0 = const()[name = tensor("op_21770_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21770_cast_fp16 = slice_by_index(begin = var_21770_begin_0, end = var_21770_end_0, end_mask = var_21770_end_mask_0, x = k_199_cast_fp16)[name = tensor("op_21770_cast_fp16")]; + tensor var_21774_begin_0 = const()[name = tensor("op_21774_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_21774_end_0 = const()[name = tensor("op_21774_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_21774_end_mask_0 = const()[name = tensor("op_21774_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21774_cast_fp16 = slice_by_index(begin = var_21774_begin_0, end = var_21774_end_0, end_mask = var_21774_end_mask_0, x = k_199_cast_fp16)[name = tensor("op_21774_cast_fp16")]; + tensor var_21778_begin_0 = const()[name = tensor("op_21778_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_21778_end_0 = const()[name = tensor("op_21778_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_21778_end_mask_0 = const()[name = tensor("op_21778_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21778_cast_fp16 = slice_by_index(begin = var_21778_begin_0, end = var_21778_end_0, end_mask = var_21778_end_mask_0, x = k_199_cast_fp16)[name = tensor("op_21778_cast_fp16")]; + tensor var_21782_begin_0 = const()[name = tensor("op_21782_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_21782_end_0 = const()[name = tensor("op_21782_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_21782_end_mask_0 = const()[name = tensor("op_21782_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21782_cast_fp16 = slice_by_index(begin = var_21782_begin_0, end = var_21782_end_0, end_mask = var_21782_end_mask_0, x = k_199_cast_fp16)[name = tensor("op_21782_cast_fp16")]; + tensor var_21786_begin_0 = const()[name = tensor("op_21786_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_21786_end_0 = const()[name = tensor("op_21786_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_21786_end_mask_0 = const()[name = tensor("op_21786_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21786_cast_fp16 = slice_by_index(begin = var_21786_begin_0, end = var_21786_end_0, end_mask = var_21786_end_mask_0, x = k_199_cast_fp16)[name = tensor("op_21786_cast_fp16")]; + tensor var_21790_begin_0 = const()[name = tensor("op_21790_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_21790_end_0 = const()[name = tensor("op_21790_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_21790_end_mask_0 = const()[name = tensor("op_21790_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21790_cast_fp16 = slice_by_index(begin = var_21790_begin_0, end = var_21790_end_0, end_mask = var_21790_end_mask_0, x = k_199_cast_fp16)[name = tensor("op_21790_cast_fp16")]; + tensor var_21794_begin_0 = const()[name = tensor("op_21794_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_21794_end_0 = const()[name = tensor("op_21794_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_21794_end_mask_0 = const()[name = tensor("op_21794_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21794_cast_fp16 = slice_by_index(begin = var_21794_begin_0, end = var_21794_end_0, end_mask = var_21794_end_mask_0, x = k_199_cast_fp16)[name = tensor("op_21794_cast_fp16")]; + tensor var_21798_begin_0 = const()[name = tensor("op_21798_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_21798_end_0 = const()[name = tensor("op_21798_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_21798_end_mask_0 = const()[name = tensor("op_21798_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21798_cast_fp16 = slice_by_index(begin = var_21798_begin_0, end = var_21798_end_0, end_mask = var_21798_end_mask_0, x = k_199_cast_fp16)[name = tensor("op_21798_cast_fp16")]; + tensor var_21802_begin_0 = const()[name = tensor("op_21802_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_21802_end_0 = const()[name = tensor("op_21802_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_21802_end_mask_0 = const()[name = tensor("op_21802_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21802_cast_fp16 = slice_by_index(begin = var_21802_begin_0, end = var_21802_end_0, end_mask = var_21802_end_mask_0, x = k_199_cast_fp16)[name = tensor("op_21802_cast_fp16")]; + tensor var_21804_begin_0 = const()[name = tensor("op_21804_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_21804_end_0 = const()[name = tensor("op_21804_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_21804_end_mask_0 = const()[name = tensor("op_21804_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21804_cast_fp16 = slice_by_index(begin = var_21804_begin_0, end = var_21804_end_0, end_mask = var_21804_end_mask_0, x = v_99_cast_fp16)[name = tensor("op_21804_cast_fp16")]; + tensor var_21808_begin_0 = const()[name = tensor("op_21808_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_21808_end_0 = const()[name = tensor("op_21808_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_21808_end_mask_0 = const()[name = tensor("op_21808_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21808_cast_fp16 = slice_by_index(begin = var_21808_begin_0, end = var_21808_end_0, end_mask = var_21808_end_mask_0, x = v_99_cast_fp16)[name = tensor("op_21808_cast_fp16")]; + tensor var_21812_begin_0 = const()[name = tensor("op_21812_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_21812_end_0 = const()[name = tensor("op_21812_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_21812_end_mask_0 = const()[name = tensor("op_21812_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21812_cast_fp16 = slice_by_index(begin = var_21812_begin_0, end = var_21812_end_0, end_mask = var_21812_end_mask_0, x = v_99_cast_fp16)[name = tensor("op_21812_cast_fp16")]; + tensor var_21816_begin_0 = const()[name = tensor("op_21816_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_21816_end_0 = const()[name = tensor("op_21816_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_21816_end_mask_0 = const()[name = tensor("op_21816_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21816_cast_fp16 = slice_by_index(begin = var_21816_begin_0, end = var_21816_end_0, end_mask = var_21816_end_mask_0, x = v_99_cast_fp16)[name = tensor("op_21816_cast_fp16")]; + tensor var_21820_begin_0 = const()[name = tensor("op_21820_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_21820_end_0 = const()[name = tensor("op_21820_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_21820_end_mask_0 = const()[name = tensor("op_21820_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21820_cast_fp16 = slice_by_index(begin = var_21820_begin_0, end = var_21820_end_0, end_mask = var_21820_end_mask_0, x = v_99_cast_fp16)[name = tensor("op_21820_cast_fp16")]; + tensor var_21824_begin_0 = const()[name = tensor("op_21824_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_21824_end_0 = const()[name = tensor("op_21824_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_21824_end_mask_0 = const()[name = tensor("op_21824_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21824_cast_fp16 = slice_by_index(begin = var_21824_begin_0, end = var_21824_end_0, end_mask = var_21824_end_mask_0, x = v_99_cast_fp16)[name = tensor("op_21824_cast_fp16")]; + tensor var_21828_begin_0 = const()[name = tensor("op_21828_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_21828_end_0 = const()[name = tensor("op_21828_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_21828_end_mask_0 = const()[name = tensor("op_21828_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21828_cast_fp16 = slice_by_index(begin = var_21828_begin_0, end = var_21828_end_0, end_mask = var_21828_end_mask_0, x = v_99_cast_fp16)[name = tensor("op_21828_cast_fp16")]; + tensor var_21832_begin_0 = const()[name = tensor("op_21832_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_21832_end_0 = const()[name = tensor("op_21832_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_21832_end_mask_0 = const()[name = tensor("op_21832_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21832_cast_fp16 = slice_by_index(begin = var_21832_begin_0, end = var_21832_end_0, end_mask = var_21832_end_mask_0, x = v_99_cast_fp16)[name = tensor("op_21832_cast_fp16")]; + tensor var_21836_begin_0 = const()[name = tensor("op_21836_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_21836_end_0 = const()[name = tensor("op_21836_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_21836_end_mask_0 = const()[name = tensor("op_21836_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21836_cast_fp16 = slice_by_index(begin = var_21836_begin_0, end = var_21836_end_0, end_mask = var_21836_end_mask_0, x = v_99_cast_fp16)[name = tensor("op_21836_cast_fp16")]; + tensor var_21840_begin_0 = const()[name = tensor("op_21840_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_21840_end_0 = const()[name = tensor("op_21840_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_21840_end_mask_0 = const()[name = tensor("op_21840_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21840_cast_fp16 = slice_by_index(begin = var_21840_begin_0, end = var_21840_end_0, end_mask = var_21840_end_mask_0, x = v_99_cast_fp16)[name = tensor("op_21840_cast_fp16")]; + tensor var_21844_begin_0 = const()[name = tensor("op_21844_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_21844_end_0 = const()[name = tensor("op_21844_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_21844_end_mask_0 = const()[name = tensor("op_21844_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21844_cast_fp16 = slice_by_index(begin = var_21844_begin_0, end = var_21844_end_0, end_mask = var_21844_end_mask_0, x = v_99_cast_fp16)[name = tensor("op_21844_cast_fp16")]; + tensor var_21848_begin_0 = const()[name = tensor("op_21848_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_21848_end_0 = const()[name = tensor("op_21848_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_21848_end_mask_0 = const()[name = tensor("op_21848_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21848_cast_fp16 = slice_by_index(begin = var_21848_begin_0, end = var_21848_end_0, end_mask = var_21848_end_mask_0, x = v_99_cast_fp16)[name = tensor("op_21848_cast_fp16")]; + tensor var_21852_begin_0 = const()[name = tensor("op_21852_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_21852_end_0 = const()[name = tensor("op_21852_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_21852_end_mask_0 = const()[name = tensor("op_21852_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21852_cast_fp16 = slice_by_index(begin = var_21852_begin_0, end = var_21852_end_0, end_mask = var_21852_end_mask_0, x = v_99_cast_fp16)[name = tensor("op_21852_cast_fp16")]; + tensor var_21856_begin_0 = const()[name = tensor("op_21856_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_21856_end_0 = const()[name = tensor("op_21856_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_21856_end_mask_0 = const()[name = tensor("op_21856_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21856_cast_fp16 = slice_by_index(begin = var_21856_begin_0, end = var_21856_end_0, end_mask = var_21856_end_mask_0, x = v_99_cast_fp16)[name = tensor("op_21856_cast_fp16")]; + tensor var_21860_begin_0 = const()[name = tensor("op_21860_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_21860_end_0 = const()[name = tensor("op_21860_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_21860_end_mask_0 = const()[name = tensor("op_21860_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21860_cast_fp16 = slice_by_index(begin = var_21860_begin_0, end = var_21860_end_0, end_mask = var_21860_end_mask_0, x = v_99_cast_fp16)[name = tensor("op_21860_cast_fp16")]; + tensor var_21864_begin_0 = const()[name = tensor("op_21864_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_21864_end_0 = const()[name = tensor("op_21864_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_21864_end_mask_0 = const()[name = tensor("op_21864_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21864_cast_fp16 = slice_by_index(begin = var_21864_begin_0, end = var_21864_end_0, end_mask = var_21864_end_mask_0, x = v_99_cast_fp16)[name = tensor("op_21864_cast_fp16")]; + tensor var_21868_begin_0 = const()[name = tensor("op_21868_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_21868_end_0 = const()[name = tensor("op_21868_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_21868_end_mask_0 = const()[name = tensor("op_21868_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21868_cast_fp16 = slice_by_index(begin = var_21868_begin_0, end = var_21868_end_0, end_mask = var_21868_end_mask_0, x = v_99_cast_fp16)[name = tensor("op_21868_cast_fp16")]; + tensor var_21872_begin_0 = const()[name = tensor("op_21872_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_21872_end_0 = const()[name = tensor("op_21872_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_21872_end_mask_0 = const()[name = tensor("op_21872_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21872_cast_fp16 = slice_by_index(begin = var_21872_begin_0, end = var_21872_end_0, end_mask = var_21872_end_mask_0, x = v_99_cast_fp16)[name = tensor("op_21872_cast_fp16")]; + tensor var_21876_begin_0 = const()[name = tensor("op_21876_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_21876_end_0 = const()[name = tensor("op_21876_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_21876_end_mask_0 = const()[name = tensor("op_21876_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21876_cast_fp16 = slice_by_index(begin = var_21876_begin_0, end = var_21876_end_0, end_mask = var_21876_end_mask_0, x = v_99_cast_fp16)[name = tensor("op_21876_cast_fp16")]; + tensor var_21880_begin_0 = const()[name = tensor("op_21880_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_21880_end_0 = const()[name = tensor("op_21880_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_21880_end_mask_0 = const()[name = tensor("op_21880_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21880_cast_fp16 = slice_by_index(begin = var_21880_begin_0, end = var_21880_end_0, end_mask = var_21880_end_mask_0, x = v_99_cast_fp16)[name = tensor("op_21880_cast_fp16")]; + tensor var_21884_equation_0 = const()[name = tensor("op_21884_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21884_cast_fp16 = einsum(equation = var_21884_equation_0, values = (var_21726_cast_fp16, var_21643_cast_fp16))[name = tensor("op_21884_cast_fp16")]; + tensor var_21885_to_fp16 = const()[name = tensor("op_21885_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1801_cast_fp16 = mul(x = var_21884_cast_fp16, y = var_21885_to_fp16)[name = tensor("aw_1801_cast_fp16")]; + tensor var_21888_equation_0 = const()[name = tensor("op_21888_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21888_cast_fp16 = einsum(equation = var_21888_equation_0, values = (var_21730_cast_fp16, var_21647_cast_fp16))[name = tensor("op_21888_cast_fp16")]; + tensor var_21889_to_fp16 = const()[name = tensor("op_21889_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1803_cast_fp16 = mul(x = var_21888_cast_fp16, y = var_21889_to_fp16)[name = tensor("aw_1803_cast_fp16")]; + tensor var_21892_equation_0 = const()[name = tensor("op_21892_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21892_cast_fp16 = einsum(equation = var_21892_equation_0, values = (var_21734_cast_fp16, var_21651_cast_fp16))[name = tensor("op_21892_cast_fp16")]; + tensor var_21893_to_fp16 = const()[name = tensor("op_21893_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1805_cast_fp16 = mul(x = var_21892_cast_fp16, y = var_21893_to_fp16)[name = tensor("aw_1805_cast_fp16")]; + tensor var_21896_equation_0 = const()[name = tensor("op_21896_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21896_cast_fp16 = einsum(equation = var_21896_equation_0, values = (var_21738_cast_fp16, var_21655_cast_fp16))[name = tensor("op_21896_cast_fp16")]; + tensor var_21897_to_fp16 = const()[name = tensor("op_21897_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1807_cast_fp16 = mul(x = var_21896_cast_fp16, y = var_21897_to_fp16)[name = tensor("aw_1807_cast_fp16")]; + tensor var_21900_equation_0 = const()[name = tensor("op_21900_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21900_cast_fp16 = einsum(equation = var_21900_equation_0, values = (var_21742_cast_fp16, var_21659_cast_fp16))[name = tensor("op_21900_cast_fp16")]; + tensor var_21901_to_fp16 = const()[name = tensor("op_21901_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1809_cast_fp16 = mul(x = var_21900_cast_fp16, y = var_21901_to_fp16)[name = tensor("aw_1809_cast_fp16")]; + tensor var_21904_equation_0 = const()[name = tensor("op_21904_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21904_cast_fp16 = einsum(equation = var_21904_equation_0, values = (var_21746_cast_fp16, var_21663_cast_fp16))[name = tensor("op_21904_cast_fp16")]; + tensor var_21905_to_fp16 = const()[name = tensor("op_21905_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1811_cast_fp16 = mul(x = var_21904_cast_fp16, y = var_21905_to_fp16)[name = tensor("aw_1811_cast_fp16")]; + tensor var_21908_equation_0 = const()[name = tensor("op_21908_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21908_cast_fp16 = einsum(equation = var_21908_equation_0, values = (var_21750_cast_fp16, var_21667_cast_fp16))[name = tensor("op_21908_cast_fp16")]; + tensor var_21909_to_fp16 = const()[name = tensor("op_21909_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1813_cast_fp16 = mul(x = var_21908_cast_fp16, y = var_21909_to_fp16)[name = tensor("aw_1813_cast_fp16")]; + tensor var_21912_equation_0 = const()[name = tensor("op_21912_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21912_cast_fp16 = einsum(equation = var_21912_equation_0, values = (var_21754_cast_fp16, var_21671_cast_fp16))[name = tensor("op_21912_cast_fp16")]; + tensor var_21913_to_fp16 = const()[name = tensor("op_21913_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1815_cast_fp16 = mul(x = var_21912_cast_fp16, y = var_21913_to_fp16)[name = tensor("aw_1815_cast_fp16")]; + tensor var_21916_equation_0 = const()[name = tensor("op_21916_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21916_cast_fp16 = einsum(equation = var_21916_equation_0, values = (var_21758_cast_fp16, var_21675_cast_fp16))[name = tensor("op_21916_cast_fp16")]; + tensor var_21917_to_fp16 = const()[name = tensor("op_21917_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1817_cast_fp16 = mul(x = var_21916_cast_fp16, y = var_21917_to_fp16)[name = tensor("aw_1817_cast_fp16")]; + tensor var_21920_equation_0 = const()[name = tensor("op_21920_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21920_cast_fp16 = einsum(equation = var_21920_equation_0, values = (var_21762_cast_fp16, var_21679_cast_fp16))[name = tensor("op_21920_cast_fp16")]; + tensor var_21921_to_fp16 = const()[name = tensor("op_21921_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1819_cast_fp16 = mul(x = var_21920_cast_fp16, y = var_21921_to_fp16)[name = tensor("aw_1819_cast_fp16")]; + tensor var_21924_equation_0 = const()[name = tensor("op_21924_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21924_cast_fp16 = einsum(equation = var_21924_equation_0, values = (var_21766_cast_fp16, var_21683_cast_fp16))[name = tensor("op_21924_cast_fp16")]; + tensor var_21925_to_fp16 = const()[name = tensor("op_21925_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1821_cast_fp16 = mul(x = var_21924_cast_fp16, y = var_21925_to_fp16)[name = tensor("aw_1821_cast_fp16")]; + tensor var_21928_equation_0 = const()[name = tensor("op_21928_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21928_cast_fp16 = einsum(equation = var_21928_equation_0, values = (var_21770_cast_fp16, var_21687_cast_fp16))[name = tensor("op_21928_cast_fp16")]; + tensor var_21929_to_fp16 = const()[name = tensor("op_21929_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1823_cast_fp16 = mul(x = var_21928_cast_fp16, y = var_21929_to_fp16)[name = tensor("aw_1823_cast_fp16")]; + tensor var_21932_equation_0 = const()[name = tensor("op_21932_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21932_cast_fp16 = einsum(equation = var_21932_equation_0, values = (var_21774_cast_fp16, var_21691_cast_fp16))[name = tensor("op_21932_cast_fp16")]; + tensor var_21933_to_fp16 = const()[name = tensor("op_21933_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1825_cast_fp16 = mul(x = var_21932_cast_fp16, y = var_21933_to_fp16)[name = tensor("aw_1825_cast_fp16")]; + tensor var_21936_equation_0 = const()[name = tensor("op_21936_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21936_cast_fp16 = einsum(equation = var_21936_equation_0, values = (var_21778_cast_fp16, var_21695_cast_fp16))[name = tensor("op_21936_cast_fp16")]; + tensor var_21937_to_fp16 = const()[name = tensor("op_21937_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1827_cast_fp16 = mul(x = var_21936_cast_fp16, y = var_21937_to_fp16)[name = tensor("aw_1827_cast_fp16")]; + tensor var_21940_equation_0 = const()[name = tensor("op_21940_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21940_cast_fp16 = einsum(equation = var_21940_equation_0, values = (var_21782_cast_fp16, var_21699_cast_fp16))[name = tensor("op_21940_cast_fp16")]; + tensor var_21941_to_fp16 = const()[name = tensor("op_21941_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1829_cast_fp16 = mul(x = var_21940_cast_fp16, y = var_21941_to_fp16)[name = tensor("aw_1829_cast_fp16")]; + tensor var_21944_equation_0 = const()[name = tensor("op_21944_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21944_cast_fp16 = einsum(equation = var_21944_equation_0, values = (var_21786_cast_fp16, var_21703_cast_fp16))[name = tensor("op_21944_cast_fp16")]; + tensor var_21945_to_fp16 = const()[name = tensor("op_21945_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1831_cast_fp16 = mul(x = var_21944_cast_fp16, y = var_21945_to_fp16)[name = tensor("aw_1831_cast_fp16")]; + tensor var_21948_equation_0 = const()[name = tensor("op_21948_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21948_cast_fp16 = einsum(equation = var_21948_equation_0, values = (var_21790_cast_fp16, var_21707_cast_fp16))[name = tensor("op_21948_cast_fp16")]; + tensor var_21949_to_fp16 = const()[name = tensor("op_21949_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1833_cast_fp16 = mul(x = var_21948_cast_fp16, y = var_21949_to_fp16)[name = tensor("aw_1833_cast_fp16")]; + tensor var_21952_equation_0 = const()[name = tensor("op_21952_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21952_cast_fp16 = einsum(equation = var_21952_equation_0, values = (var_21794_cast_fp16, var_21711_cast_fp16))[name = tensor("op_21952_cast_fp16")]; + tensor var_21953_to_fp16 = const()[name = tensor("op_21953_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1835_cast_fp16 = mul(x = var_21952_cast_fp16, y = var_21953_to_fp16)[name = tensor("aw_1835_cast_fp16")]; + tensor var_21956_equation_0 = const()[name = tensor("op_21956_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21956_cast_fp16 = einsum(equation = var_21956_equation_0, values = (var_21798_cast_fp16, var_21715_cast_fp16))[name = tensor("op_21956_cast_fp16")]; + tensor var_21957_to_fp16 = const()[name = tensor("op_21957_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1837_cast_fp16 = mul(x = var_21956_cast_fp16, y = var_21957_to_fp16)[name = tensor("aw_1837_cast_fp16")]; + tensor var_21960_equation_0 = const()[name = tensor("op_21960_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21960_cast_fp16 = einsum(equation = var_21960_equation_0, values = (var_21802_cast_fp16, var_21719_cast_fp16))[name = tensor("op_21960_cast_fp16")]; + tensor var_21961_to_fp16 = const()[name = tensor("op_21961_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1839_cast_fp16 = mul(x = var_21960_cast_fp16, y = var_21961_to_fp16)[name = tensor("aw_1839_cast_fp16")]; + tensor var_21963_cast_fp16 = softmax(axis = var_21077, x = aw_1801_cast_fp16)[name = tensor("op_21963_cast_fp16")]; + tensor var_21964_cast_fp16 = softmax(axis = var_21077, x = aw_1803_cast_fp16)[name = tensor("op_21964_cast_fp16")]; + tensor var_21965_cast_fp16 = softmax(axis = var_21077, x = aw_1805_cast_fp16)[name = tensor("op_21965_cast_fp16")]; + tensor var_21966_cast_fp16 = softmax(axis = var_21077, x = aw_1807_cast_fp16)[name = tensor("op_21966_cast_fp16")]; + tensor var_21967_cast_fp16 = softmax(axis = var_21077, x = aw_1809_cast_fp16)[name = tensor("op_21967_cast_fp16")]; + tensor var_21968_cast_fp16 = softmax(axis = var_21077, x = aw_1811_cast_fp16)[name = tensor("op_21968_cast_fp16")]; + tensor var_21969_cast_fp16 = softmax(axis = var_21077, x = aw_1813_cast_fp16)[name = tensor("op_21969_cast_fp16")]; + tensor var_21970_cast_fp16 = softmax(axis = var_21077, x = aw_1815_cast_fp16)[name = tensor("op_21970_cast_fp16")]; + tensor var_21971_cast_fp16 = softmax(axis = var_21077, x = aw_1817_cast_fp16)[name = tensor("op_21971_cast_fp16")]; + tensor var_21972_cast_fp16 = softmax(axis = var_21077, x = aw_1819_cast_fp16)[name = tensor("op_21972_cast_fp16")]; + tensor var_21973_cast_fp16 = softmax(axis = var_21077, x = aw_1821_cast_fp16)[name = tensor("op_21973_cast_fp16")]; + tensor var_21974_cast_fp16 = softmax(axis = var_21077, x = aw_1823_cast_fp16)[name = tensor("op_21974_cast_fp16")]; + tensor var_21975_cast_fp16 = softmax(axis = var_21077, x = aw_1825_cast_fp16)[name = tensor("op_21975_cast_fp16")]; + tensor var_21976_cast_fp16 = softmax(axis = var_21077, x = aw_1827_cast_fp16)[name = tensor("op_21976_cast_fp16")]; + tensor var_21977_cast_fp16 = softmax(axis = var_21077, x = aw_1829_cast_fp16)[name = tensor("op_21977_cast_fp16")]; + tensor var_21978_cast_fp16 = softmax(axis = var_21077, x = aw_1831_cast_fp16)[name = tensor("op_21978_cast_fp16")]; + tensor var_21979_cast_fp16 = softmax(axis = var_21077, x = aw_1833_cast_fp16)[name = tensor("op_21979_cast_fp16")]; + tensor var_21980_cast_fp16 = softmax(axis = var_21077, x = aw_1835_cast_fp16)[name = tensor("op_21980_cast_fp16")]; + tensor var_21981_cast_fp16 = softmax(axis = var_21077, x = aw_1837_cast_fp16)[name = tensor("op_21981_cast_fp16")]; + tensor var_21982_cast_fp16 = softmax(axis = var_21077, x = aw_1839_cast_fp16)[name = tensor("op_21982_cast_fp16")]; + tensor var_21984_equation_0 = const()[name = tensor("op_21984_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21984_cast_fp16 = einsum(equation = var_21984_equation_0, values = (var_21804_cast_fp16, var_21963_cast_fp16))[name = tensor("op_21984_cast_fp16")]; + tensor var_21986_equation_0 = const()[name = tensor("op_21986_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21986_cast_fp16 = einsum(equation = var_21986_equation_0, values = (var_21808_cast_fp16, var_21964_cast_fp16))[name = tensor("op_21986_cast_fp16")]; + tensor var_21988_equation_0 = const()[name = tensor("op_21988_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21988_cast_fp16 = einsum(equation = var_21988_equation_0, values = (var_21812_cast_fp16, var_21965_cast_fp16))[name = tensor("op_21988_cast_fp16")]; + tensor var_21990_equation_0 = const()[name = tensor("op_21990_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21990_cast_fp16 = einsum(equation = var_21990_equation_0, values = (var_21816_cast_fp16, var_21966_cast_fp16))[name = tensor("op_21990_cast_fp16")]; + tensor var_21992_equation_0 = const()[name = tensor("op_21992_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21992_cast_fp16 = einsum(equation = var_21992_equation_0, values = (var_21820_cast_fp16, var_21967_cast_fp16))[name = tensor("op_21992_cast_fp16")]; + tensor var_21994_equation_0 = const()[name = tensor("op_21994_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21994_cast_fp16 = einsum(equation = var_21994_equation_0, values = (var_21824_cast_fp16, var_21968_cast_fp16))[name = tensor("op_21994_cast_fp16")]; + tensor var_21996_equation_0 = const()[name = tensor("op_21996_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21996_cast_fp16 = einsum(equation = var_21996_equation_0, values = (var_21828_cast_fp16, var_21969_cast_fp16))[name = tensor("op_21996_cast_fp16")]; + tensor var_21998_equation_0 = const()[name = tensor("op_21998_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21998_cast_fp16 = einsum(equation = var_21998_equation_0, values = (var_21832_cast_fp16, var_21970_cast_fp16))[name = tensor("op_21998_cast_fp16")]; + tensor var_22000_equation_0 = const()[name = tensor("op_22000_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22000_cast_fp16 = einsum(equation = var_22000_equation_0, values = (var_21836_cast_fp16, var_21971_cast_fp16))[name = tensor("op_22000_cast_fp16")]; + tensor var_22002_equation_0 = const()[name = tensor("op_22002_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22002_cast_fp16 = einsum(equation = var_22002_equation_0, values = (var_21840_cast_fp16, var_21972_cast_fp16))[name = tensor("op_22002_cast_fp16")]; + tensor var_22004_equation_0 = const()[name = tensor("op_22004_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22004_cast_fp16 = einsum(equation = var_22004_equation_0, values = (var_21844_cast_fp16, var_21973_cast_fp16))[name = tensor("op_22004_cast_fp16")]; + tensor var_22006_equation_0 = const()[name = tensor("op_22006_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22006_cast_fp16 = einsum(equation = var_22006_equation_0, values = (var_21848_cast_fp16, var_21974_cast_fp16))[name = tensor("op_22006_cast_fp16")]; + tensor var_22008_equation_0 = const()[name = tensor("op_22008_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22008_cast_fp16 = einsum(equation = var_22008_equation_0, values = (var_21852_cast_fp16, var_21975_cast_fp16))[name = tensor("op_22008_cast_fp16")]; + tensor var_22010_equation_0 = const()[name = tensor("op_22010_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22010_cast_fp16 = einsum(equation = var_22010_equation_0, values = (var_21856_cast_fp16, var_21976_cast_fp16))[name = tensor("op_22010_cast_fp16")]; + tensor var_22012_equation_0 = const()[name = tensor("op_22012_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22012_cast_fp16 = einsum(equation = var_22012_equation_0, values = (var_21860_cast_fp16, var_21977_cast_fp16))[name = tensor("op_22012_cast_fp16")]; + tensor var_22014_equation_0 = const()[name = tensor("op_22014_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22014_cast_fp16 = einsum(equation = var_22014_equation_0, values = (var_21864_cast_fp16, var_21978_cast_fp16))[name = tensor("op_22014_cast_fp16")]; + tensor var_22016_equation_0 = const()[name = tensor("op_22016_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22016_cast_fp16 = einsum(equation = var_22016_equation_0, values = (var_21868_cast_fp16, var_21979_cast_fp16))[name = tensor("op_22016_cast_fp16")]; + tensor var_22018_equation_0 = const()[name = tensor("op_22018_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22018_cast_fp16 = einsum(equation = var_22018_equation_0, values = (var_21872_cast_fp16, var_21980_cast_fp16))[name = tensor("op_22018_cast_fp16")]; + tensor var_22020_equation_0 = const()[name = tensor("op_22020_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22020_cast_fp16 = einsum(equation = var_22020_equation_0, values = (var_21876_cast_fp16, var_21981_cast_fp16))[name = tensor("op_22020_cast_fp16")]; + tensor var_22022_equation_0 = const()[name = tensor("op_22022_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22022_cast_fp16 = einsum(equation = var_22022_equation_0, values = (var_21880_cast_fp16, var_21982_cast_fp16))[name = tensor("op_22022_cast_fp16")]; + tensor input_329_interleave_0 = const()[name = tensor("input_329_interleave_0"), val = tensor(false)]; + tensor input_329_cast_fp16 = concat(axis = var_21077, interleave = input_329_interleave_0, values = (var_21984_cast_fp16, var_21986_cast_fp16, var_21988_cast_fp16, var_21990_cast_fp16, var_21992_cast_fp16, var_21994_cast_fp16, var_21996_cast_fp16, var_21998_cast_fp16, var_22000_cast_fp16, var_22002_cast_fp16, var_22004_cast_fp16, var_22006_cast_fp16, var_22008_cast_fp16, var_22010_cast_fp16, var_22012_cast_fp16, var_22014_cast_fp16, var_22016_cast_fp16, var_22018_cast_fp16, var_22020_cast_fp16, var_22022_cast_fp16))[name = tensor("input_329_cast_fp16")]; + tensor var_22028 = const()[name = tensor("op_22028"), val = tensor([1, 1])]; + tensor var_22030 = const()[name = tensor("op_22030"), val = tensor([1, 1])]; + tensor var_22032_pad_type_0 = const()[name = tensor("op_22032_pad_type_0"), val = tensor("custom")]; + tensor var_22032_pad_0 = const()[name = tensor("op_22032_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1752992384)))]; + tensor mid_block_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1756269248)))]; + tensor var_22032_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_22030, groups = var_21077, pad = var_22032_pad_0, pad_type = var_22032_pad_type_0, strides = var_22028, weight = mid_block_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16, x = input_329_cast_fp16)[name = tensor("op_22032_cast_fp16")]; + tensor inputs_149_cast_fp16 = add(x = var_22032_cast_fp16, y = inputs_147_cast_fp16)[name = tensor("inputs_149_cast_fp16")]; + tensor input_331_axes_0 = const()[name = tensor("input_331_axes_0"), val = tensor([1])]; + tensor input_331_gamma_0_to_fp16 = const()[name = tensor("input_331_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1756271872)))]; + tensor input_331_beta_0_to_fp16 = const()[name = tensor("input_331_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1756274496)))]; + tensor var_22042_to_fp16 = const()[name = tensor("op_22042_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_331_cast_fp16 = layer_norm(axes = input_331_axes_0, beta = input_331_beta_0_to_fp16, epsilon = var_22042_to_fp16, gamma = input_331_gamma_0_to_fp16, x = inputs_149_cast_fp16)[name = tensor("input_331_cast_fp16")]; + tensor var_22058 = const()[name = tensor("op_22058"), val = tensor([1, 1])]; + tensor var_22060 = const()[name = tensor("op_22060"), val = tensor([1, 1])]; + tensor var_22062_pad_type_0 = const()[name = tensor("op_22062_pad_type_0"), val = tensor("custom")]; + tensor var_22062_pad_0 = const()[name = tensor("op_22062_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1756277120)))]; + tensor mid_block_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1782491584)))]; + tensor var_22062_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_22060, groups = var_21077, pad = var_22062_pad_0, pad_type = var_22062_pad_type_0, strides = var_22058, weight = mid_block_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16, x = input_331_cast_fp16)[name = tensor("op_22062_cast_fp16")]; + tensor var_22063_split_sizes_0 = const()[name = tensor("op_22063_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_22063_axis_0 = const()[name = tensor("op_22063_axis_0"), val = tensor(1)]; + tensor var_22063_cast_fp16_0, tensor var_22063_cast_fp16_1 = split(axis = var_22063_axis_0, split_sizes = var_22063_split_sizes_0, x = var_22062_cast_fp16)[name = tensor("op_22063_cast_fp16")]; + tensor var_22065_mode_0 = const()[name = tensor("op_22065_mode_0"), val = tensor("EXACT")]; + tensor var_22065_cast_fp16 = gelu(mode = var_22065_mode_0, x = var_22063_cast_fp16_1)[name = tensor("op_22065_cast_fp16")]; + tensor input_333_cast_fp16 = mul(x = var_22063_cast_fp16_0, y = var_22065_cast_fp16)[name = tensor("input_333_cast_fp16")]; + tensor var_22069 = const()[name = tensor("op_22069"), val = tensor([1, 1])]; + tensor var_22071 = const()[name = tensor("op_22071"), val = tensor([1, 1])]; + tensor var_22073_pad_type_0 = const()[name = tensor("op_22073_pad_type_0"), val = tensor("custom")]; + tensor var_22073_pad_0 = const()[name = tensor("op_22073_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1782512128)))]; + tensor mid_block_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1795619392)))]; + tensor var_22073_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_22071, groups = var_21077, pad = var_22073_pad_0, pad_type = var_22073_pad_type_0, strides = var_22069, weight = mid_block_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16, x = input_333_cast_fp16)[name = tensor("op_22073_cast_fp16")]; + tensor inputs_151_cast_fp16 = add(x = var_22073_cast_fp16, y = inputs_149_cast_fp16)[name = tensor("inputs_151_cast_fp16")]; + tensor hidden_states_215_axes_0 = const()[name = tensor("hidden_states_215_axes_0"), val = tensor([1])]; + tensor hidden_states_215_gamma_0_to_fp16 = const()[name = tensor("hidden_states_215_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1795622016)))]; + tensor hidden_states_215_beta_0_to_fp16 = const()[name = tensor("hidden_states_215_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1795624640)))]; + tensor var_22089_to_fp16 = const()[name = tensor("op_22089_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_215_cast_fp16 = layer_norm(axes = hidden_states_215_axes_0, beta = hidden_states_215_beta_0_to_fp16, epsilon = var_22089_to_fp16, gamma = hidden_states_215_gamma_0_to_fp16, x = inputs_151_cast_fp16)[name = tensor("hidden_states_215_cast_fp16")]; + tensor var_22104 = const()[name = tensor("op_22104"), val = tensor([1, 1])]; + tensor var_22106 = const()[name = tensor("op_22106"), val = tensor([1, 1])]; + tensor q_101_pad_type_0 = const()[name = tensor("q_101_pad_type_0"), val = tensor("custom")]; + tensor q_101_pad_0 = const()[name = tensor("q_101_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_1_attn1_to_q_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_1_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1795627264)))]; + tensor q_101_cast_fp16 = conv(dilations = var_22106, groups = var_21077, pad = q_101_pad_0, pad_type = q_101_pad_type_0, strides = var_22104, weight = mid_block_attentions_0_transformer_blocks_1_attn1_to_q_weight_to_fp16, x = hidden_states_215_cast_fp16)[name = tensor("q_101_cast_fp16")]; + tensor var_22110 = const()[name = tensor("op_22110"), val = tensor([1, 1])]; + tensor var_22112 = const()[name = tensor("op_22112"), val = tensor([1, 1])]; + tensor k_201_pad_type_0 = const()[name = tensor("k_201_pad_type_0"), val = tensor("custom")]; + tensor k_201_pad_0 = const()[name = tensor("k_201_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_1_attn1_to_k_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_1_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1798904128)))]; + tensor k_201_cast_fp16 = conv(dilations = var_22112, groups = var_21077, pad = k_201_pad_0, pad_type = k_201_pad_type_0, strides = var_22110, weight = mid_block_attentions_0_transformer_blocks_1_attn1_to_k_weight_to_fp16, x = hidden_states_215_cast_fp16)[name = tensor("k_201_cast_fp16")]; + tensor var_22116 = const()[name = tensor("op_22116"), val = tensor([1, 1])]; + tensor var_22118 = const()[name = tensor("op_22118"), val = tensor([1, 1])]; + tensor v_101_pad_type_0 = const()[name = tensor("v_101_pad_type_0"), val = tensor("custom")]; + tensor v_101_pad_0 = const()[name = tensor("v_101_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_1_attn1_to_v_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_1_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1802180992)))]; + tensor v_101_cast_fp16 = conv(dilations = var_22118, groups = var_21077, pad = v_101_pad_0, pad_type = v_101_pad_type_0, strides = var_22116, weight = mid_block_attentions_0_transformer_blocks_1_attn1_to_v_weight_to_fp16, x = hidden_states_215_cast_fp16)[name = tensor("v_101_cast_fp16")]; + tensor var_22122_begin_0 = const()[name = tensor("op_22122_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_22122_end_0 = const()[name = tensor("op_22122_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_22122_end_mask_0 = const()[name = tensor("op_22122_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22122_cast_fp16 = slice_by_index(begin = var_22122_begin_0, end = var_22122_end_0, end_mask = var_22122_end_mask_0, x = q_101_cast_fp16)[name = tensor("op_22122_cast_fp16")]; + tensor var_22126_begin_0 = const()[name = tensor("op_22126_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_22126_end_0 = const()[name = tensor("op_22126_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_22126_end_mask_0 = const()[name = tensor("op_22126_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22126_cast_fp16 = slice_by_index(begin = var_22126_begin_0, end = var_22126_end_0, end_mask = var_22126_end_mask_0, x = q_101_cast_fp16)[name = tensor("op_22126_cast_fp16")]; + tensor var_22130_begin_0 = const()[name = tensor("op_22130_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_22130_end_0 = const()[name = tensor("op_22130_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_22130_end_mask_0 = const()[name = tensor("op_22130_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22130_cast_fp16 = slice_by_index(begin = var_22130_begin_0, end = var_22130_end_0, end_mask = var_22130_end_mask_0, x = q_101_cast_fp16)[name = tensor("op_22130_cast_fp16")]; + tensor var_22134_begin_0 = const()[name = tensor("op_22134_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_22134_end_0 = const()[name = tensor("op_22134_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_22134_end_mask_0 = const()[name = tensor("op_22134_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22134_cast_fp16 = slice_by_index(begin = var_22134_begin_0, end = var_22134_end_0, end_mask = var_22134_end_mask_0, x = q_101_cast_fp16)[name = tensor("op_22134_cast_fp16")]; + tensor var_22138_begin_0 = const()[name = tensor("op_22138_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_22138_end_0 = const()[name = tensor("op_22138_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_22138_end_mask_0 = const()[name = tensor("op_22138_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22138_cast_fp16 = slice_by_index(begin = var_22138_begin_0, end = var_22138_end_0, end_mask = var_22138_end_mask_0, x = q_101_cast_fp16)[name = tensor("op_22138_cast_fp16")]; + tensor var_22142_begin_0 = const()[name = tensor("op_22142_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_22142_end_0 = const()[name = tensor("op_22142_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_22142_end_mask_0 = const()[name = tensor("op_22142_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22142_cast_fp16 = slice_by_index(begin = var_22142_begin_0, end = var_22142_end_0, end_mask = var_22142_end_mask_0, x = q_101_cast_fp16)[name = tensor("op_22142_cast_fp16")]; + tensor var_22146_begin_0 = const()[name = tensor("op_22146_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_22146_end_0 = const()[name = tensor("op_22146_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_22146_end_mask_0 = const()[name = tensor("op_22146_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22146_cast_fp16 = slice_by_index(begin = var_22146_begin_0, end = var_22146_end_0, end_mask = var_22146_end_mask_0, x = q_101_cast_fp16)[name = tensor("op_22146_cast_fp16")]; + tensor var_22150_begin_0 = const()[name = tensor("op_22150_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_22150_end_0 = const()[name = tensor("op_22150_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_22150_end_mask_0 = const()[name = tensor("op_22150_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22150_cast_fp16 = slice_by_index(begin = var_22150_begin_0, end = var_22150_end_0, end_mask = var_22150_end_mask_0, x = q_101_cast_fp16)[name = tensor("op_22150_cast_fp16")]; + tensor var_22154_begin_0 = const()[name = tensor("op_22154_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_22154_end_0 = const()[name = tensor("op_22154_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_22154_end_mask_0 = const()[name = tensor("op_22154_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22154_cast_fp16 = slice_by_index(begin = var_22154_begin_0, end = var_22154_end_0, end_mask = var_22154_end_mask_0, x = q_101_cast_fp16)[name = tensor("op_22154_cast_fp16")]; + tensor var_22158_begin_0 = const()[name = tensor("op_22158_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_22158_end_0 = const()[name = tensor("op_22158_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_22158_end_mask_0 = const()[name = tensor("op_22158_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22158_cast_fp16 = slice_by_index(begin = var_22158_begin_0, end = var_22158_end_0, end_mask = var_22158_end_mask_0, x = q_101_cast_fp16)[name = tensor("op_22158_cast_fp16")]; + tensor var_22162_begin_0 = const()[name = tensor("op_22162_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_22162_end_0 = const()[name = tensor("op_22162_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_22162_end_mask_0 = const()[name = tensor("op_22162_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22162_cast_fp16 = slice_by_index(begin = var_22162_begin_0, end = var_22162_end_0, end_mask = var_22162_end_mask_0, x = q_101_cast_fp16)[name = tensor("op_22162_cast_fp16")]; + tensor var_22166_begin_0 = const()[name = tensor("op_22166_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_22166_end_0 = const()[name = tensor("op_22166_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_22166_end_mask_0 = const()[name = tensor("op_22166_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22166_cast_fp16 = slice_by_index(begin = var_22166_begin_0, end = var_22166_end_0, end_mask = var_22166_end_mask_0, x = q_101_cast_fp16)[name = tensor("op_22166_cast_fp16")]; + tensor var_22170_begin_0 = const()[name = tensor("op_22170_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_22170_end_0 = const()[name = tensor("op_22170_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_22170_end_mask_0 = const()[name = tensor("op_22170_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22170_cast_fp16 = slice_by_index(begin = var_22170_begin_0, end = var_22170_end_0, end_mask = var_22170_end_mask_0, x = q_101_cast_fp16)[name = tensor("op_22170_cast_fp16")]; + tensor var_22174_begin_0 = const()[name = tensor("op_22174_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_22174_end_0 = const()[name = tensor("op_22174_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_22174_end_mask_0 = const()[name = tensor("op_22174_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22174_cast_fp16 = slice_by_index(begin = var_22174_begin_0, end = var_22174_end_0, end_mask = var_22174_end_mask_0, x = q_101_cast_fp16)[name = tensor("op_22174_cast_fp16")]; + tensor var_22178_begin_0 = const()[name = tensor("op_22178_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_22178_end_0 = const()[name = tensor("op_22178_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_22178_end_mask_0 = const()[name = tensor("op_22178_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22178_cast_fp16 = slice_by_index(begin = var_22178_begin_0, end = var_22178_end_0, end_mask = var_22178_end_mask_0, x = q_101_cast_fp16)[name = tensor("op_22178_cast_fp16")]; + tensor var_22182_begin_0 = const()[name = tensor("op_22182_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_22182_end_0 = const()[name = tensor("op_22182_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_22182_end_mask_0 = const()[name = tensor("op_22182_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22182_cast_fp16 = slice_by_index(begin = var_22182_begin_0, end = var_22182_end_0, end_mask = var_22182_end_mask_0, x = q_101_cast_fp16)[name = tensor("op_22182_cast_fp16")]; + tensor var_22186_begin_0 = const()[name = tensor("op_22186_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_22186_end_0 = const()[name = tensor("op_22186_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_22186_end_mask_0 = const()[name = tensor("op_22186_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22186_cast_fp16 = slice_by_index(begin = var_22186_begin_0, end = var_22186_end_0, end_mask = var_22186_end_mask_0, x = q_101_cast_fp16)[name = tensor("op_22186_cast_fp16")]; + tensor var_22190_begin_0 = const()[name = tensor("op_22190_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_22190_end_0 = const()[name = tensor("op_22190_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_22190_end_mask_0 = const()[name = tensor("op_22190_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22190_cast_fp16 = slice_by_index(begin = var_22190_begin_0, end = var_22190_end_0, end_mask = var_22190_end_mask_0, x = q_101_cast_fp16)[name = tensor("op_22190_cast_fp16")]; + tensor var_22194_begin_0 = const()[name = tensor("op_22194_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_22194_end_0 = const()[name = tensor("op_22194_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_22194_end_mask_0 = const()[name = tensor("op_22194_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22194_cast_fp16 = slice_by_index(begin = var_22194_begin_0, end = var_22194_end_0, end_mask = var_22194_end_mask_0, x = q_101_cast_fp16)[name = tensor("op_22194_cast_fp16")]; + tensor var_22198_begin_0 = const()[name = tensor("op_22198_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_22198_end_0 = const()[name = tensor("op_22198_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_22198_end_mask_0 = const()[name = tensor("op_22198_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22198_cast_fp16 = slice_by_index(begin = var_22198_begin_0, end = var_22198_end_0, end_mask = var_22198_end_mask_0, x = q_101_cast_fp16)[name = tensor("op_22198_cast_fp16")]; + tensor k_203_perm_0 = const()[name = tensor("k_203_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_22205_begin_0 = const()[name = tensor("op_22205_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_22205_end_0 = const()[name = tensor("op_22205_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_22205_end_mask_0 = const()[name = tensor("op_22205_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_203_cast_fp16 = transpose(perm = k_203_perm_0, x = k_201_cast_fp16)[name = tensor("transpose_89")]; + tensor var_22205_cast_fp16 = slice_by_index(begin = var_22205_begin_0, end = var_22205_end_0, end_mask = var_22205_end_mask_0, x = k_203_cast_fp16)[name = tensor("op_22205_cast_fp16")]; + tensor var_22209_begin_0 = const()[name = tensor("op_22209_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_22209_end_0 = const()[name = tensor("op_22209_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_22209_end_mask_0 = const()[name = tensor("op_22209_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22209_cast_fp16 = slice_by_index(begin = var_22209_begin_0, end = var_22209_end_0, end_mask = var_22209_end_mask_0, x = k_203_cast_fp16)[name = tensor("op_22209_cast_fp16")]; + tensor var_22213_begin_0 = const()[name = tensor("op_22213_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_22213_end_0 = const()[name = tensor("op_22213_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_22213_end_mask_0 = const()[name = tensor("op_22213_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22213_cast_fp16 = slice_by_index(begin = var_22213_begin_0, end = var_22213_end_0, end_mask = var_22213_end_mask_0, x = k_203_cast_fp16)[name = tensor("op_22213_cast_fp16")]; + tensor var_22217_begin_0 = const()[name = tensor("op_22217_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_22217_end_0 = const()[name = tensor("op_22217_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_22217_end_mask_0 = const()[name = tensor("op_22217_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22217_cast_fp16 = slice_by_index(begin = var_22217_begin_0, end = var_22217_end_0, end_mask = var_22217_end_mask_0, x = k_203_cast_fp16)[name = tensor("op_22217_cast_fp16")]; + tensor var_22221_begin_0 = const()[name = tensor("op_22221_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_22221_end_0 = const()[name = tensor("op_22221_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_22221_end_mask_0 = const()[name = tensor("op_22221_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22221_cast_fp16 = slice_by_index(begin = var_22221_begin_0, end = var_22221_end_0, end_mask = var_22221_end_mask_0, x = k_203_cast_fp16)[name = tensor("op_22221_cast_fp16")]; + tensor var_22225_begin_0 = const()[name = tensor("op_22225_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_22225_end_0 = const()[name = tensor("op_22225_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_22225_end_mask_0 = const()[name = tensor("op_22225_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22225_cast_fp16 = slice_by_index(begin = var_22225_begin_0, end = var_22225_end_0, end_mask = var_22225_end_mask_0, x = k_203_cast_fp16)[name = tensor("op_22225_cast_fp16")]; + tensor var_22229_begin_0 = const()[name = tensor("op_22229_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_22229_end_0 = const()[name = tensor("op_22229_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_22229_end_mask_0 = const()[name = tensor("op_22229_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22229_cast_fp16 = slice_by_index(begin = var_22229_begin_0, end = var_22229_end_0, end_mask = var_22229_end_mask_0, x = k_203_cast_fp16)[name = tensor("op_22229_cast_fp16")]; + tensor var_22233_begin_0 = const()[name = tensor("op_22233_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_22233_end_0 = const()[name = tensor("op_22233_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_22233_end_mask_0 = const()[name = tensor("op_22233_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22233_cast_fp16 = slice_by_index(begin = var_22233_begin_0, end = var_22233_end_0, end_mask = var_22233_end_mask_0, x = k_203_cast_fp16)[name = tensor("op_22233_cast_fp16")]; + tensor var_22237_begin_0 = const()[name = tensor("op_22237_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_22237_end_0 = const()[name = tensor("op_22237_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_22237_end_mask_0 = const()[name = tensor("op_22237_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22237_cast_fp16 = slice_by_index(begin = var_22237_begin_0, end = var_22237_end_0, end_mask = var_22237_end_mask_0, x = k_203_cast_fp16)[name = tensor("op_22237_cast_fp16")]; + tensor var_22241_begin_0 = const()[name = tensor("op_22241_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_22241_end_0 = const()[name = tensor("op_22241_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_22241_end_mask_0 = const()[name = tensor("op_22241_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22241_cast_fp16 = slice_by_index(begin = var_22241_begin_0, end = var_22241_end_0, end_mask = var_22241_end_mask_0, x = k_203_cast_fp16)[name = tensor("op_22241_cast_fp16")]; + tensor var_22245_begin_0 = const()[name = tensor("op_22245_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_22245_end_0 = const()[name = tensor("op_22245_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_22245_end_mask_0 = const()[name = tensor("op_22245_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22245_cast_fp16 = slice_by_index(begin = var_22245_begin_0, end = var_22245_end_0, end_mask = var_22245_end_mask_0, x = k_203_cast_fp16)[name = tensor("op_22245_cast_fp16")]; + tensor var_22249_begin_0 = const()[name = tensor("op_22249_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_22249_end_0 = const()[name = tensor("op_22249_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_22249_end_mask_0 = const()[name = tensor("op_22249_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22249_cast_fp16 = slice_by_index(begin = var_22249_begin_0, end = var_22249_end_0, end_mask = var_22249_end_mask_0, x = k_203_cast_fp16)[name = tensor("op_22249_cast_fp16")]; + tensor var_22253_begin_0 = const()[name = tensor("op_22253_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_22253_end_0 = const()[name = tensor("op_22253_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_22253_end_mask_0 = const()[name = tensor("op_22253_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22253_cast_fp16 = slice_by_index(begin = var_22253_begin_0, end = var_22253_end_0, end_mask = var_22253_end_mask_0, x = k_203_cast_fp16)[name = tensor("op_22253_cast_fp16")]; + tensor var_22257_begin_0 = const()[name = tensor("op_22257_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_22257_end_0 = const()[name = tensor("op_22257_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_22257_end_mask_0 = const()[name = tensor("op_22257_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22257_cast_fp16 = slice_by_index(begin = var_22257_begin_0, end = var_22257_end_0, end_mask = var_22257_end_mask_0, x = k_203_cast_fp16)[name = tensor("op_22257_cast_fp16")]; + tensor var_22261_begin_0 = const()[name = tensor("op_22261_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_22261_end_0 = const()[name = tensor("op_22261_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_22261_end_mask_0 = const()[name = tensor("op_22261_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22261_cast_fp16 = slice_by_index(begin = var_22261_begin_0, end = var_22261_end_0, end_mask = var_22261_end_mask_0, x = k_203_cast_fp16)[name = tensor("op_22261_cast_fp16")]; + tensor var_22265_begin_0 = const()[name = tensor("op_22265_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_22265_end_0 = const()[name = tensor("op_22265_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_22265_end_mask_0 = const()[name = tensor("op_22265_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22265_cast_fp16 = slice_by_index(begin = var_22265_begin_0, end = var_22265_end_0, end_mask = var_22265_end_mask_0, x = k_203_cast_fp16)[name = tensor("op_22265_cast_fp16")]; + tensor var_22269_begin_0 = const()[name = tensor("op_22269_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_22269_end_0 = const()[name = tensor("op_22269_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_22269_end_mask_0 = const()[name = tensor("op_22269_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22269_cast_fp16 = slice_by_index(begin = var_22269_begin_0, end = var_22269_end_0, end_mask = var_22269_end_mask_0, x = k_203_cast_fp16)[name = tensor("op_22269_cast_fp16")]; + tensor var_22273_begin_0 = const()[name = tensor("op_22273_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_22273_end_0 = const()[name = tensor("op_22273_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_22273_end_mask_0 = const()[name = tensor("op_22273_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22273_cast_fp16 = slice_by_index(begin = var_22273_begin_0, end = var_22273_end_0, end_mask = var_22273_end_mask_0, x = k_203_cast_fp16)[name = tensor("op_22273_cast_fp16")]; + tensor var_22277_begin_0 = const()[name = tensor("op_22277_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_22277_end_0 = const()[name = tensor("op_22277_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_22277_end_mask_0 = const()[name = tensor("op_22277_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22277_cast_fp16 = slice_by_index(begin = var_22277_begin_0, end = var_22277_end_0, end_mask = var_22277_end_mask_0, x = k_203_cast_fp16)[name = tensor("op_22277_cast_fp16")]; + tensor var_22281_begin_0 = const()[name = tensor("op_22281_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_22281_end_0 = const()[name = tensor("op_22281_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_22281_end_mask_0 = const()[name = tensor("op_22281_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22281_cast_fp16 = slice_by_index(begin = var_22281_begin_0, end = var_22281_end_0, end_mask = var_22281_end_mask_0, x = k_203_cast_fp16)[name = tensor("op_22281_cast_fp16")]; + tensor var_22283_begin_0 = const()[name = tensor("op_22283_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_22283_end_0 = const()[name = tensor("op_22283_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_22283_end_mask_0 = const()[name = tensor("op_22283_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22283_cast_fp16 = slice_by_index(begin = var_22283_begin_0, end = var_22283_end_0, end_mask = var_22283_end_mask_0, x = v_101_cast_fp16)[name = tensor("op_22283_cast_fp16")]; + tensor var_22287_begin_0 = const()[name = tensor("op_22287_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_22287_end_0 = const()[name = tensor("op_22287_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_22287_end_mask_0 = const()[name = tensor("op_22287_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22287_cast_fp16 = slice_by_index(begin = var_22287_begin_0, end = var_22287_end_0, end_mask = var_22287_end_mask_0, x = v_101_cast_fp16)[name = tensor("op_22287_cast_fp16")]; + tensor var_22291_begin_0 = const()[name = tensor("op_22291_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_22291_end_0 = const()[name = tensor("op_22291_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_22291_end_mask_0 = const()[name = tensor("op_22291_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22291_cast_fp16 = slice_by_index(begin = var_22291_begin_0, end = var_22291_end_0, end_mask = var_22291_end_mask_0, x = v_101_cast_fp16)[name = tensor("op_22291_cast_fp16")]; + tensor var_22295_begin_0 = const()[name = tensor("op_22295_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_22295_end_0 = const()[name = tensor("op_22295_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_22295_end_mask_0 = const()[name = tensor("op_22295_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22295_cast_fp16 = slice_by_index(begin = var_22295_begin_0, end = var_22295_end_0, end_mask = var_22295_end_mask_0, x = v_101_cast_fp16)[name = tensor("op_22295_cast_fp16")]; + tensor var_22299_begin_0 = const()[name = tensor("op_22299_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_22299_end_0 = const()[name = tensor("op_22299_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_22299_end_mask_0 = const()[name = tensor("op_22299_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22299_cast_fp16 = slice_by_index(begin = var_22299_begin_0, end = var_22299_end_0, end_mask = var_22299_end_mask_0, x = v_101_cast_fp16)[name = tensor("op_22299_cast_fp16")]; + tensor var_22303_begin_0 = const()[name = tensor("op_22303_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_22303_end_0 = const()[name = tensor("op_22303_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_22303_end_mask_0 = const()[name = tensor("op_22303_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22303_cast_fp16 = slice_by_index(begin = var_22303_begin_0, end = var_22303_end_0, end_mask = var_22303_end_mask_0, x = v_101_cast_fp16)[name = tensor("op_22303_cast_fp16")]; + tensor var_22307_begin_0 = const()[name = tensor("op_22307_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_22307_end_0 = const()[name = tensor("op_22307_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_22307_end_mask_0 = const()[name = tensor("op_22307_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22307_cast_fp16 = slice_by_index(begin = var_22307_begin_0, end = var_22307_end_0, end_mask = var_22307_end_mask_0, x = v_101_cast_fp16)[name = tensor("op_22307_cast_fp16")]; + tensor var_22311_begin_0 = const()[name = tensor("op_22311_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_22311_end_0 = const()[name = tensor("op_22311_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_22311_end_mask_0 = const()[name = tensor("op_22311_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22311_cast_fp16 = slice_by_index(begin = var_22311_begin_0, end = var_22311_end_0, end_mask = var_22311_end_mask_0, x = v_101_cast_fp16)[name = tensor("op_22311_cast_fp16")]; + tensor var_22315_begin_0 = const()[name = tensor("op_22315_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_22315_end_0 = const()[name = tensor("op_22315_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_22315_end_mask_0 = const()[name = tensor("op_22315_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22315_cast_fp16 = slice_by_index(begin = var_22315_begin_0, end = var_22315_end_0, end_mask = var_22315_end_mask_0, x = v_101_cast_fp16)[name = tensor("op_22315_cast_fp16")]; + tensor var_22319_begin_0 = const()[name = tensor("op_22319_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_22319_end_0 = const()[name = tensor("op_22319_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_22319_end_mask_0 = const()[name = tensor("op_22319_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22319_cast_fp16 = slice_by_index(begin = var_22319_begin_0, end = var_22319_end_0, end_mask = var_22319_end_mask_0, x = v_101_cast_fp16)[name = tensor("op_22319_cast_fp16")]; + tensor var_22323_begin_0 = const()[name = tensor("op_22323_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_22323_end_0 = const()[name = tensor("op_22323_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_22323_end_mask_0 = const()[name = tensor("op_22323_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22323_cast_fp16 = slice_by_index(begin = var_22323_begin_0, end = var_22323_end_0, end_mask = var_22323_end_mask_0, x = v_101_cast_fp16)[name = tensor("op_22323_cast_fp16")]; + tensor var_22327_begin_0 = const()[name = tensor("op_22327_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_22327_end_0 = const()[name = tensor("op_22327_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_22327_end_mask_0 = const()[name = tensor("op_22327_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22327_cast_fp16 = slice_by_index(begin = var_22327_begin_0, end = var_22327_end_0, end_mask = var_22327_end_mask_0, x = v_101_cast_fp16)[name = tensor("op_22327_cast_fp16")]; + tensor var_22331_begin_0 = const()[name = tensor("op_22331_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_22331_end_0 = const()[name = tensor("op_22331_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_22331_end_mask_0 = const()[name = tensor("op_22331_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22331_cast_fp16 = slice_by_index(begin = var_22331_begin_0, end = var_22331_end_0, end_mask = var_22331_end_mask_0, x = v_101_cast_fp16)[name = tensor("op_22331_cast_fp16")]; + tensor var_22335_begin_0 = const()[name = tensor("op_22335_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_22335_end_0 = const()[name = tensor("op_22335_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_22335_end_mask_0 = const()[name = tensor("op_22335_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22335_cast_fp16 = slice_by_index(begin = var_22335_begin_0, end = var_22335_end_0, end_mask = var_22335_end_mask_0, x = v_101_cast_fp16)[name = tensor("op_22335_cast_fp16")]; + tensor var_22339_begin_0 = const()[name = tensor("op_22339_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_22339_end_0 = const()[name = tensor("op_22339_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_22339_end_mask_0 = const()[name = tensor("op_22339_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22339_cast_fp16 = slice_by_index(begin = var_22339_begin_0, end = var_22339_end_0, end_mask = var_22339_end_mask_0, x = v_101_cast_fp16)[name = tensor("op_22339_cast_fp16")]; + tensor var_22343_begin_0 = const()[name = tensor("op_22343_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_22343_end_0 = const()[name = tensor("op_22343_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_22343_end_mask_0 = const()[name = tensor("op_22343_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22343_cast_fp16 = slice_by_index(begin = var_22343_begin_0, end = var_22343_end_0, end_mask = var_22343_end_mask_0, x = v_101_cast_fp16)[name = tensor("op_22343_cast_fp16")]; + tensor var_22347_begin_0 = const()[name = tensor("op_22347_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_22347_end_0 = const()[name = tensor("op_22347_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_22347_end_mask_0 = const()[name = tensor("op_22347_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22347_cast_fp16 = slice_by_index(begin = var_22347_begin_0, end = var_22347_end_0, end_mask = var_22347_end_mask_0, x = v_101_cast_fp16)[name = tensor("op_22347_cast_fp16")]; + tensor var_22351_begin_0 = const()[name = tensor("op_22351_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_22351_end_0 = const()[name = tensor("op_22351_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_22351_end_mask_0 = const()[name = tensor("op_22351_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22351_cast_fp16 = slice_by_index(begin = var_22351_begin_0, end = var_22351_end_0, end_mask = var_22351_end_mask_0, x = v_101_cast_fp16)[name = tensor("op_22351_cast_fp16")]; + tensor var_22355_begin_0 = const()[name = tensor("op_22355_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_22355_end_0 = const()[name = tensor("op_22355_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_22355_end_mask_0 = const()[name = tensor("op_22355_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22355_cast_fp16 = slice_by_index(begin = var_22355_begin_0, end = var_22355_end_0, end_mask = var_22355_end_mask_0, x = v_101_cast_fp16)[name = tensor("op_22355_cast_fp16")]; + tensor var_22359_begin_0 = const()[name = tensor("op_22359_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_22359_end_0 = const()[name = tensor("op_22359_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_22359_end_mask_0 = const()[name = tensor("op_22359_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22359_cast_fp16 = slice_by_index(begin = var_22359_begin_0, end = var_22359_end_0, end_mask = var_22359_end_mask_0, x = v_101_cast_fp16)[name = tensor("op_22359_cast_fp16")]; + tensor var_22363_equation_0 = const()[name = tensor("op_22363_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22363_cast_fp16 = einsum(equation = var_22363_equation_0, values = (var_22205_cast_fp16, var_22122_cast_fp16))[name = tensor("op_22363_cast_fp16")]; + tensor var_22364_to_fp16 = const()[name = tensor("op_22364_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1841_cast_fp16 = mul(x = var_22363_cast_fp16, y = var_22364_to_fp16)[name = tensor("aw_1841_cast_fp16")]; + tensor var_22367_equation_0 = const()[name = tensor("op_22367_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22367_cast_fp16 = einsum(equation = var_22367_equation_0, values = (var_22209_cast_fp16, var_22126_cast_fp16))[name = tensor("op_22367_cast_fp16")]; + tensor var_22368_to_fp16 = const()[name = tensor("op_22368_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1843_cast_fp16 = mul(x = var_22367_cast_fp16, y = var_22368_to_fp16)[name = tensor("aw_1843_cast_fp16")]; + tensor var_22371_equation_0 = const()[name = tensor("op_22371_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22371_cast_fp16 = einsum(equation = var_22371_equation_0, values = (var_22213_cast_fp16, var_22130_cast_fp16))[name = tensor("op_22371_cast_fp16")]; + tensor var_22372_to_fp16 = const()[name = tensor("op_22372_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1845_cast_fp16 = mul(x = var_22371_cast_fp16, y = var_22372_to_fp16)[name = tensor("aw_1845_cast_fp16")]; + tensor var_22375_equation_0 = const()[name = tensor("op_22375_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22375_cast_fp16 = einsum(equation = var_22375_equation_0, values = (var_22217_cast_fp16, var_22134_cast_fp16))[name = tensor("op_22375_cast_fp16")]; + tensor var_22376_to_fp16 = const()[name = tensor("op_22376_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1847_cast_fp16 = mul(x = var_22375_cast_fp16, y = var_22376_to_fp16)[name = tensor("aw_1847_cast_fp16")]; + tensor var_22379_equation_0 = const()[name = tensor("op_22379_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22379_cast_fp16 = einsum(equation = var_22379_equation_0, values = (var_22221_cast_fp16, var_22138_cast_fp16))[name = tensor("op_22379_cast_fp16")]; + tensor var_22380_to_fp16 = const()[name = tensor("op_22380_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1849_cast_fp16 = mul(x = var_22379_cast_fp16, y = var_22380_to_fp16)[name = tensor("aw_1849_cast_fp16")]; + tensor var_22383_equation_0 = const()[name = tensor("op_22383_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22383_cast_fp16 = einsum(equation = var_22383_equation_0, values = (var_22225_cast_fp16, var_22142_cast_fp16))[name = tensor("op_22383_cast_fp16")]; + tensor var_22384_to_fp16 = const()[name = tensor("op_22384_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1851_cast_fp16 = mul(x = var_22383_cast_fp16, y = var_22384_to_fp16)[name = tensor("aw_1851_cast_fp16")]; + tensor var_22387_equation_0 = const()[name = tensor("op_22387_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22387_cast_fp16 = einsum(equation = var_22387_equation_0, values = (var_22229_cast_fp16, var_22146_cast_fp16))[name = tensor("op_22387_cast_fp16")]; + tensor var_22388_to_fp16 = const()[name = tensor("op_22388_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1853_cast_fp16 = mul(x = var_22387_cast_fp16, y = var_22388_to_fp16)[name = tensor("aw_1853_cast_fp16")]; + tensor var_22391_equation_0 = const()[name = tensor("op_22391_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22391_cast_fp16 = einsum(equation = var_22391_equation_0, values = (var_22233_cast_fp16, var_22150_cast_fp16))[name = tensor("op_22391_cast_fp16")]; + tensor var_22392_to_fp16 = const()[name = tensor("op_22392_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1855_cast_fp16 = mul(x = var_22391_cast_fp16, y = var_22392_to_fp16)[name = tensor("aw_1855_cast_fp16")]; + tensor var_22395_equation_0 = const()[name = tensor("op_22395_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22395_cast_fp16 = einsum(equation = var_22395_equation_0, values = (var_22237_cast_fp16, var_22154_cast_fp16))[name = tensor("op_22395_cast_fp16")]; + tensor var_22396_to_fp16 = const()[name = tensor("op_22396_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1857_cast_fp16 = mul(x = var_22395_cast_fp16, y = var_22396_to_fp16)[name = tensor("aw_1857_cast_fp16")]; + tensor var_22399_equation_0 = const()[name = tensor("op_22399_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22399_cast_fp16 = einsum(equation = var_22399_equation_0, values = (var_22241_cast_fp16, var_22158_cast_fp16))[name = tensor("op_22399_cast_fp16")]; + tensor var_22400_to_fp16 = const()[name = tensor("op_22400_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1859_cast_fp16 = mul(x = var_22399_cast_fp16, y = var_22400_to_fp16)[name = tensor("aw_1859_cast_fp16")]; + tensor var_22403_equation_0 = const()[name = tensor("op_22403_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22403_cast_fp16 = einsum(equation = var_22403_equation_0, values = (var_22245_cast_fp16, var_22162_cast_fp16))[name = tensor("op_22403_cast_fp16")]; + tensor var_22404_to_fp16 = const()[name = tensor("op_22404_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1861_cast_fp16 = mul(x = var_22403_cast_fp16, y = var_22404_to_fp16)[name = tensor("aw_1861_cast_fp16")]; + tensor var_22407_equation_0 = const()[name = tensor("op_22407_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22407_cast_fp16 = einsum(equation = var_22407_equation_0, values = (var_22249_cast_fp16, var_22166_cast_fp16))[name = tensor("op_22407_cast_fp16")]; + tensor var_22408_to_fp16 = const()[name = tensor("op_22408_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1863_cast_fp16 = mul(x = var_22407_cast_fp16, y = var_22408_to_fp16)[name = tensor("aw_1863_cast_fp16")]; + tensor var_22411_equation_0 = const()[name = tensor("op_22411_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22411_cast_fp16 = einsum(equation = var_22411_equation_0, values = (var_22253_cast_fp16, var_22170_cast_fp16))[name = tensor("op_22411_cast_fp16")]; + tensor var_22412_to_fp16 = const()[name = tensor("op_22412_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1865_cast_fp16 = mul(x = var_22411_cast_fp16, y = var_22412_to_fp16)[name = tensor("aw_1865_cast_fp16")]; + tensor var_22415_equation_0 = const()[name = tensor("op_22415_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22415_cast_fp16 = einsum(equation = var_22415_equation_0, values = (var_22257_cast_fp16, var_22174_cast_fp16))[name = tensor("op_22415_cast_fp16")]; + tensor var_22416_to_fp16 = const()[name = tensor("op_22416_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1867_cast_fp16 = mul(x = var_22415_cast_fp16, y = var_22416_to_fp16)[name = tensor("aw_1867_cast_fp16")]; + tensor var_22419_equation_0 = const()[name = tensor("op_22419_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22419_cast_fp16 = einsum(equation = var_22419_equation_0, values = (var_22261_cast_fp16, var_22178_cast_fp16))[name = tensor("op_22419_cast_fp16")]; + tensor var_22420_to_fp16 = const()[name = tensor("op_22420_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1869_cast_fp16 = mul(x = var_22419_cast_fp16, y = var_22420_to_fp16)[name = tensor("aw_1869_cast_fp16")]; + tensor var_22423_equation_0 = const()[name = tensor("op_22423_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22423_cast_fp16 = einsum(equation = var_22423_equation_0, values = (var_22265_cast_fp16, var_22182_cast_fp16))[name = tensor("op_22423_cast_fp16")]; + tensor var_22424_to_fp16 = const()[name = tensor("op_22424_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1871_cast_fp16 = mul(x = var_22423_cast_fp16, y = var_22424_to_fp16)[name = tensor("aw_1871_cast_fp16")]; + tensor var_22427_equation_0 = const()[name = tensor("op_22427_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22427_cast_fp16 = einsum(equation = var_22427_equation_0, values = (var_22269_cast_fp16, var_22186_cast_fp16))[name = tensor("op_22427_cast_fp16")]; + tensor var_22428_to_fp16 = const()[name = tensor("op_22428_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1873_cast_fp16 = mul(x = var_22427_cast_fp16, y = var_22428_to_fp16)[name = tensor("aw_1873_cast_fp16")]; + tensor var_22431_equation_0 = const()[name = tensor("op_22431_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22431_cast_fp16 = einsum(equation = var_22431_equation_0, values = (var_22273_cast_fp16, var_22190_cast_fp16))[name = tensor("op_22431_cast_fp16")]; + tensor var_22432_to_fp16 = const()[name = tensor("op_22432_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1875_cast_fp16 = mul(x = var_22431_cast_fp16, y = var_22432_to_fp16)[name = tensor("aw_1875_cast_fp16")]; + tensor var_22435_equation_0 = const()[name = tensor("op_22435_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22435_cast_fp16 = einsum(equation = var_22435_equation_0, values = (var_22277_cast_fp16, var_22194_cast_fp16))[name = tensor("op_22435_cast_fp16")]; + tensor var_22436_to_fp16 = const()[name = tensor("op_22436_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1877_cast_fp16 = mul(x = var_22435_cast_fp16, y = var_22436_to_fp16)[name = tensor("aw_1877_cast_fp16")]; + tensor var_22439_equation_0 = const()[name = tensor("op_22439_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22439_cast_fp16 = einsum(equation = var_22439_equation_0, values = (var_22281_cast_fp16, var_22198_cast_fp16))[name = tensor("op_22439_cast_fp16")]; + tensor var_22440_to_fp16 = const()[name = tensor("op_22440_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1879_cast_fp16 = mul(x = var_22439_cast_fp16, y = var_22440_to_fp16)[name = tensor("aw_1879_cast_fp16")]; + tensor var_22442_cast_fp16 = softmax(axis = var_21077, x = aw_1841_cast_fp16)[name = tensor("op_22442_cast_fp16")]; + tensor var_22443_cast_fp16 = softmax(axis = var_21077, x = aw_1843_cast_fp16)[name = tensor("op_22443_cast_fp16")]; + tensor var_22444_cast_fp16 = softmax(axis = var_21077, x = aw_1845_cast_fp16)[name = tensor("op_22444_cast_fp16")]; + tensor var_22445_cast_fp16 = softmax(axis = var_21077, x = aw_1847_cast_fp16)[name = tensor("op_22445_cast_fp16")]; + tensor var_22446_cast_fp16 = softmax(axis = var_21077, x = aw_1849_cast_fp16)[name = tensor("op_22446_cast_fp16")]; + tensor var_22447_cast_fp16 = softmax(axis = var_21077, x = aw_1851_cast_fp16)[name = tensor("op_22447_cast_fp16")]; + tensor var_22448_cast_fp16 = softmax(axis = var_21077, x = aw_1853_cast_fp16)[name = tensor("op_22448_cast_fp16")]; + tensor var_22449_cast_fp16 = softmax(axis = var_21077, x = aw_1855_cast_fp16)[name = tensor("op_22449_cast_fp16")]; + tensor var_22450_cast_fp16 = softmax(axis = var_21077, x = aw_1857_cast_fp16)[name = tensor("op_22450_cast_fp16")]; + tensor var_22451_cast_fp16 = softmax(axis = var_21077, x = aw_1859_cast_fp16)[name = tensor("op_22451_cast_fp16")]; + tensor var_22452_cast_fp16 = softmax(axis = var_21077, x = aw_1861_cast_fp16)[name = tensor("op_22452_cast_fp16")]; + tensor var_22453_cast_fp16 = softmax(axis = var_21077, x = aw_1863_cast_fp16)[name = tensor("op_22453_cast_fp16")]; + tensor var_22454_cast_fp16 = softmax(axis = var_21077, x = aw_1865_cast_fp16)[name = tensor("op_22454_cast_fp16")]; + tensor var_22455_cast_fp16 = softmax(axis = var_21077, x = aw_1867_cast_fp16)[name = tensor("op_22455_cast_fp16")]; + tensor var_22456_cast_fp16 = softmax(axis = var_21077, x = aw_1869_cast_fp16)[name = tensor("op_22456_cast_fp16")]; + tensor var_22457_cast_fp16 = softmax(axis = var_21077, x = aw_1871_cast_fp16)[name = tensor("op_22457_cast_fp16")]; + tensor var_22458_cast_fp16 = softmax(axis = var_21077, x = aw_1873_cast_fp16)[name = tensor("op_22458_cast_fp16")]; + tensor var_22459_cast_fp16 = softmax(axis = var_21077, x = aw_1875_cast_fp16)[name = tensor("op_22459_cast_fp16")]; + tensor var_22460_cast_fp16 = softmax(axis = var_21077, x = aw_1877_cast_fp16)[name = tensor("op_22460_cast_fp16")]; + tensor var_22461_cast_fp16 = softmax(axis = var_21077, x = aw_1879_cast_fp16)[name = tensor("op_22461_cast_fp16")]; + tensor var_22463_equation_0 = const()[name = tensor("op_22463_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22463_cast_fp16 = einsum(equation = var_22463_equation_0, values = (var_22283_cast_fp16, var_22442_cast_fp16))[name = tensor("op_22463_cast_fp16")]; + tensor var_22465_equation_0 = const()[name = tensor("op_22465_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22465_cast_fp16 = einsum(equation = var_22465_equation_0, values = (var_22287_cast_fp16, var_22443_cast_fp16))[name = tensor("op_22465_cast_fp16")]; + tensor var_22467_equation_0 = const()[name = tensor("op_22467_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22467_cast_fp16 = einsum(equation = var_22467_equation_0, values = (var_22291_cast_fp16, var_22444_cast_fp16))[name = tensor("op_22467_cast_fp16")]; + tensor var_22469_equation_0 = const()[name = tensor("op_22469_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22469_cast_fp16 = einsum(equation = var_22469_equation_0, values = (var_22295_cast_fp16, var_22445_cast_fp16))[name = tensor("op_22469_cast_fp16")]; + tensor var_22471_equation_0 = const()[name = tensor("op_22471_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22471_cast_fp16 = einsum(equation = var_22471_equation_0, values = (var_22299_cast_fp16, var_22446_cast_fp16))[name = tensor("op_22471_cast_fp16")]; + tensor var_22473_equation_0 = const()[name = tensor("op_22473_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22473_cast_fp16 = einsum(equation = var_22473_equation_0, values = (var_22303_cast_fp16, var_22447_cast_fp16))[name = tensor("op_22473_cast_fp16")]; + tensor var_22475_equation_0 = const()[name = tensor("op_22475_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22475_cast_fp16 = einsum(equation = var_22475_equation_0, values = (var_22307_cast_fp16, var_22448_cast_fp16))[name = tensor("op_22475_cast_fp16")]; + tensor var_22477_equation_0 = const()[name = tensor("op_22477_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22477_cast_fp16 = einsum(equation = var_22477_equation_0, values = (var_22311_cast_fp16, var_22449_cast_fp16))[name = tensor("op_22477_cast_fp16")]; + tensor var_22479_equation_0 = const()[name = tensor("op_22479_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22479_cast_fp16 = einsum(equation = var_22479_equation_0, values = (var_22315_cast_fp16, var_22450_cast_fp16))[name = tensor("op_22479_cast_fp16")]; + tensor var_22481_equation_0 = const()[name = tensor("op_22481_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22481_cast_fp16 = einsum(equation = var_22481_equation_0, values = (var_22319_cast_fp16, var_22451_cast_fp16))[name = tensor("op_22481_cast_fp16")]; + tensor var_22483_equation_0 = const()[name = tensor("op_22483_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22483_cast_fp16 = einsum(equation = var_22483_equation_0, values = (var_22323_cast_fp16, var_22452_cast_fp16))[name = tensor("op_22483_cast_fp16")]; + tensor var_22485_equation_0 = const()[name = tensor("op_22485_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22485_cast_fp16 = einsum(equation = var_22485_equation_0, values = (var_22327_cast_fp16, var_22453_cast_fp16))[name = tensor("op_22485_cast_fp16")]; + tensor var_22487_equation_0 = const()[name = tensor("op_22487_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22487_cast_fp16 = einsum(equation = var_22487_equation_0, values = (var_22331_cast_fp16, var_22454_cast_fp16))[name = tensor("op_22487_cast_fp16")]; + tensor var_22489_equation_0 = const()[name = tensor("op_22489_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22489_cast_fp16 = einsum(equation = var_22489_equation_0, values = (var_22335_cast_fp16, var_22455_cast_fp16))[name = tensor("op_22489_cast_fp16")]; + tensor var_22491_equation_0 = const()[name = tensor("op_22491_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22491_cast_fp16 = einsum(equation = var_22491_equation_0, values = (var_22339_cast_fp16, var_22456_cast_fp16))[name = tensor("op_22491_cast_fp16")]; + tensor var_22493_equation_0 = const()[name = tensor("op_22493_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22493_cast_fp16 = einsum(equation = var_22493_equation_0, values = (var_22343_cast_fp16, var_22457_cast_fp16))[name = tensor("op_22493_cast_fp16")]; + tensor var_22495_equation_0 = const()[name = tensor("op_22495_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22495_cast_fp16 = einsum(equation = var_22495_equation_0, values = (var_22347_cast_fp16, var_22458_cast_fp16))[name = tensor("op_22495_cast_fp16")]; + tensor var_22497_equation_0 = const()[name = tensor("op_22497_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22497_cast_fp16 = einsum(equation = var_22497_equation_0, values = (var_22351_cast_fp16, var_22459_cast_fp16))[name = tensor("op_22497_cast_fp16")]; + tensor var_22499_equation_0 = const()[name = tensor("op_22499_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22499_cast_fp16 = einsum(equation = var_22499_equation_0, values = (var_22355_cast_fp16, var_22460_cast_fp16))[name = tensor("op_22499_cast_fp16")]; + tensor var_22501_equation_0 = const()[name = tensor("op_22501_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22501_cast_fp16 = einsum(equation = var_22501_equation_0, values = (var_22359_cast_fp16, var_22461_cast_fp16))[name = tensor("op_22501_cast_fp16")]; + tensor input_335_interleave_0 = const()[name = tensor("input_335_interleave_0"), val = tensor(false)]; + tensor input_335_cast_fp16 = concat(axis = var_21077, interleave = input_335_interleave_0, values = (var_22463_cast_fp16, var_22465_cast_fp16, var_22467_cast_fp16, var_22469_cast_fp16, var_22471_cast_fp16, var_22473_cast_fp16, var_22475_cast_fp16, var_22477_cast_fp16, var_22479_cast_fp16, var_22481_cast_fp16, var_22483_cast_fp16, var_22485_cast_fp16, var_22487_cast_fp16, var_22489_cast_fp16, var_22491_cast_fp16, var_22493_cast_fp16, var_22495_cast_fp16, var_22497_cast_fp16, var_22499_cast_fp16, var_22501_cast_fp16))[name = tensor("input_335_cast_fp16")]; + tensor var_22507 = const()[name = tensor("op_22507"), val = tensor([1, 1])]; + tensor var_22509 = const()[name = tensor("op_22509"), val = tensor([1, 1])]; + tensor var_22511_pad_type_0 = const()[name = tensor("op_22511_pad_type_0"), val = tensor("custom")]; + tensor var_22511_pad_0 = const()[name = tensor("op_22511_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_1_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_1_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1805457856)))]; + tensor mid_block_attentions_0_transformer_blocks_1_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_1_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1808734720)))]; + tensor var_22511_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_1_attn1_to_out_0_bias_to_fp16, dilations = var_22509, groups = var_21077, pad = var_22511_pad_0, pad_type = var_22511_pad_type_0, strides = var_22507, weight = mid_block_attentions_0_transformer_blocks_1_attn1_to_out_0_weight_to_fp16, x = input_335_cast_fp16)[name = tensor("op_22511_cast_fp16")]; + tensor inputs_153_cast_fp16 = add(x = var_22511_cast_fp16, y = inputs_151_cast_fp16)[name = tensor("inputs_153_cast_fp16")]; + tensor hidden_states_217_axes_0 = const()[name = tensor("hidden_states_217_axes_0"), val = tensor([1])]; + tensor hidden_states_217_gamma_0_to_fp16 = const()[name = tensor("hidden_states_217_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1808737344)))]; + tensor hidden_states_217_beta_0_to_fp16 = const()[name = tensor("hidden_states_217_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1808739968)))]; + tensor var_22521_to_fp16 = const()[name = tensor("op_22521_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_217_cast_fp16 = layer_norm(axes = hidden_states_217_axes_0, beta = hidden_states_217_beta_0_to_fp16, epsilon = var_22521_to_fp16, gamma = hidden_states_217_gamma_0_to_fp16, x = inputs_153_cast_fp16)[name = tensor("hidden_states_217_cast_fp16")]; + tensor var_22536 = const()[name = tensor("op_22536"), val = tensor([1, 1])]; + tensor var_22538 = const()[name = tensor("op_22538"), val = tensor([1, 1])]; + tensor q_103_pad_type_0 = const()[name = tensor("q_103_pad_type_0"), val = tensor("custom")]; + tensor q_103_pad_0 = const()[name = tensor("q_103_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_1_attn2_to_q_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_1_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1808742592)))]; + tensor q_103_cast_fp16 = conv(dilations = var_22538, groups = var_21077, pad = q_103_pad_0, pad_type = q_103_pad_type_0, strides = var_22536, weight = mid_block_attentions_0_transformer_blocks_1_attn2_to_q_weight_to_fp16, x = hidden_states_217_cast_fp16)[name = tensor("q_103_cast_fp16")]; + tensor var_22542 = const()[name = tensor("op_22542"), val = tensor([1, 1])]; + tensor var_22544 = const()[name = tensor("op_22544"), val = tensor([1, 1])]; + tensor k_205_pad_type_0 = const()[name = tensor("k_205_pad_type_0"), val = tensor("custom")]; + tensor k_205_pad_0 = const()[name = tensor("k_205_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_1_attn2_to_k_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_1_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1812019456)))]; + tensor k_205_cast_fp16 = conv(dilations = var_22544, groups = var_21077, pad = k_205_pad_0, pad_type = k_205_pad_type_0, strides = var_22542, weight = mid_block_attentions_0_transformer_blocks_1_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_205_cast_fp16")]; + tensor var_22548 = const()[name = tensor("op_22548"), val = tensor([1, 1])]; + tensor var_22550 = const()[name = tensor("op_22550"), val = tensor([1, 1])]; + tensor v_103_pad_type_0 = const()[name = tensor("v_103_pad_type_0"), val = tensor("custom")]; + tensor v_103_pad_0 = const()[name = tensor("v_103_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_1_attn2_to_v_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_1_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1817262400)))]; + tensor v_103_cast_fp16 = conv(dilations = var_22550, groups = var_21077, pad = v_103_pad_0, pad_type = v_103_pad_type_0, strides = var_22548, weight = mid_block_attentions_0_transformer_blocks_1_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_103_cast_fp16")]; + tensor var_22554_begin_0 = const()[name = tensor("op_22554_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_22554_end_0 = const()[name = tensor("op_22554_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_22554_end_mask_0 = const()[name = tensor("op_22554_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22554_cast_fp16 = slice_by_index(begin = var_22554_begin_0, end = var_22554_end_0, end_mask = var_22554_end_mask_0, x = q_103_cast_fp16)[name = tensor("op_22554_cast_fp16")]; + tensor var_22558_begin_0 = const()[name = tensor("op_22558_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_22558_end_0 = const()[name = tensor("op_22558_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_22558_end_mask_0 = const()[name = tensor("op_22558_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22558_cast_fp16 = slice_by_index(begin = var_22558_begin_0, end = var_22558_end_0, end_mask = var_22558_end_mask_0, x = q_103_cast_fp16)[name = tensor("op_22558_cast_fp16")]; + tensor var_22562_begin_0 = const()[name = tensor("op_22562_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_22562_end_0 = const()[name = tensor("op_22562_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_22562_end_mask_0 = const()[name = tensor("op_22562_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22562_cast_fp16 = slice_by_index(begin = var_22562_begin_0, end = var_22562_end_0, end_mask = var_22562_end_mask_0, x = q_103_cast_fp16)[name = tensor("op_22562_cast_fp16")]; + tensor var_22566_begin_0 = const()[name = tensor("op_22566_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_22566_end_0 = const()[name = tensor("op_22566_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_22566_end_mask_0 = const()[name = tensor("op_22566_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22566_cast_fp16 = slice_by_index(begin = var_22566_begin_0, end = var_22566_end_0, end_mask = var_22566_end_mask_0, x = q_103_cast_fp16)[name = tensor("op_22566_cast_fp16")]; + tensor var_22570_begin_0 = const()[name = tensor("op_22570_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_22570_end_0 = const()[name = tensor("op_22570_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_22570_end_mask_0 = const()[name = tensor("op_22570_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22570_cast_fp16 = slice_by_index(begin = var_22570_begin_0, end = var_22570_end_0, end_mask = var_22570_end_mask_0, x = q_103_cast_fp16)[name = tensor("op_22570_cast_fp16")]; + tensor var_22574_begin_0 = const()[name = tensor("op_22574_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_22574_end_0 = const()[name = tensor("op_22574_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_22574_end_mask_0 = const()[name = tensor("op_22574_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22574_cast_fp16 = slice_by_index(begin = var_22574_begin_0, end = var_22574_end_0, end_mask = var_22574_end_mask_0, x = q_103_cast_fp16)[name = tensor("op_22574_cast_fp16")]; + tensor var_22578_begin_0 = const()[name = tensor("op_22578_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_22578_end_0 = const()[name = tensor("op_22578_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_22578_end_mask_0 = const()[name = tensor("op_22578_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22578_cast_fp16 = slice_by_index(begin = var_22578_begin_0, end = var_22578_end_0, end_mask = var_22578_end_mask_0, x = q_103_cast_fp16)[name = tensor("op_22578_cast_fp16")]; + tensor var_22582_begin_0 = const()[name = tensor("op_22582_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_22582_end_0 = const()[name = tensor("op_22582_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_22582_end_mask_0 = const()[name = tensor("op_22582_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22582_cast_fp16 = slice_by_index(begin = var_22582_begin_0, end = var_22582_end_0, end_mask = var_22582_end_mask_0, x = q_103_cast_fp16)[name = tensor("op_22582_cast_fp16")]; + tensor var_22586_begin_0 = const()[name = tensor("op_22586_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_22586_end_0 = const()[name = tensor("op_22586_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_22586_end_mask_0 = const()[name = tensor("op_22586_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22586_cast_fp16 = slice_by_index(begin = var_22586_begin_0, end = var_22586_end_0, end_mask = var_22586_end_mask_0, x = q_103_cast_fp16)[name = tensor("op_22586_cast_fp16")]; + tensor var_22590_begin_0 = const()[name = tensor("op_22590_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_22590_end_0 = const()[name = tensor("op_22590_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_22590_end_mask_0 = const()[name = tensor("op_22590_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22590_cast_fp16 = slice_by_index(begin = var_22590_begin_0, end = var_22590_end_0, end_mask = var_22590_end_mask_0, x = q_103_cast_fp16)[name = tensor("op_22590_cast_fp16")]; + tensor var_22594_begin_0 = const()[name = tensor("op_22594_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_22594_end_0 = const()[name = tensor("op_22594_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_22594_end_mask_0 = const()[name = tensor("op_22594_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22594_cast_fp16 = slice_by_index(begin = var_22594_begin_0, end = var_22594_end_0, end_mask = var_22594_end_mask_0, x = q_103_cast_fp16)[name = tensor("op_22594_cast_fp16")]; + tensor var_22598_begin_0 = const()[name = tensor("op_22598_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_22598_end_0 = const()[name = tensor("op_22598_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_22598_end_mask_0 = const()[name = tensor("op_22598_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22598_cast_fp16 = slice_by_index(begin = var_22598_begin_0, end = var_22598_end_0, end_mask = var_22598_end_mask_0, x = q_103_cast_fp16)[name = tensor("op_22598_cast_fp16")]; + tensor var_22602_begin_0 = const()[name = tensor("op_22602_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_22602_end_0 = const()[name = tensor("op_22602_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_22602_end_mask_0 = const()[name = tensor("op_22602_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22602_cast_fp16 = slice_by_index(begin = var_22602_begin_0, end = var_22602_end_0, end_mask = var_22602_end_mask_0, x = q_103_cast_fp16)[name = tensor("op_22602_cast_fp16")]; + tensor var_22606_begin_0 = const()[name = tensor("op_22606_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_22606_end_0 = const()[name = tensor("op_22606_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_22606_end_mask_0 = const()[name = tensor("op_22606_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22606_cast_fp16 = slice_by_index(begin = var_22606_begin_0, end = var_22606_end_0, end_mask = var_22606_end_mask_0, x = q_103_cast_fp16)[name = tensor("op_22606_cast_fp16")]; + tensor var_22610_begin_0 = const()[name = tensor("op_22610_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_22610_end_0 = const()[name = tensor("op_22610_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_22610_end_mask_0 = const()[name = tensor("op_22610_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22610_cast_fp16 = slice_by_index(begin = var_22610_begin_0, end = var_22610_end_0, end_mask = var_22610_end_mask_0, x = q_103_cast_fp16)[name = tensor("op_22610_cast_fp16")]; + tensor var_22614_begin_0 = const()[name = tensor("op_22614_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_22614_end_0 = const()[name = tensor("op_22614_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_22614_end_mask_0 = const()[name = tensor("op_22614_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22614_cast_fp16 = slice_by_index(begin = var_22614_begin_0, end = var_22614_end_0, end_mask = var_22614_end_mask_0, x = q_103_cast_fp16)[name = tensor("op_22614_cast_fp16")]; + tensor var_22618_begin_0 = const()[name = tensor("op_22618_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_22618_end_0 = const()[name = tensor("op_22618_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_22618_end_mask_0 = const()[name = tensor("op_22618_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22618_cast_fp16 = slice_by_index(begin = var_22618_begin_0, end = var_22618_end_0, end_mask = var_22618_end_mask_0, x = q_103_cast_fp16)[name = tensor("op_22618_cast_fp16")]; + tensor var_22622_begin_0 = const()[name = tensor("op_22622_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_22622_end_0 = const()[name = tensor("op_22622_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_22622_end_mask_0 = const()[name = tensor("op_22622_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22622_cast_fp16 = slice_by_index(begin = var_22622_begin_0, end = var_22622_end_0, end_mask = var_22622_end_mask_0, x = q_103_cast_fp16)[name = tensor("op_22622_cast_fp16")]; + tensor var_22626_begin_0 = const()[name = tensor("op_22626_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_22626_end_0 = const()[name = tensor("op_22626_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_22626_end_mask_0 = const()[name = tensor("op_22626_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22626_cast_fp16 = slice_by_index(begin = var_22626_begin_0, end = var_22626_end_0, end_mask = var_22626_end_mask_0, x = q_103_cast_fp16)[name = tensor("op_22626_cast_fp16")]; + tensor var_22630_begin_0 = const()[name = tensor("op_22630_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_22630_end_0 = const()[name = tensor("op_22630_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_22630_end_mask_0 = const()[name = tensor("op_22630_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22630_cast_fp16 = slice_by_index(begin = var_22630_begin_0, end = var_22630_end_0, end_mask = var_22630_end_mask_0, x = q_103_cast_fp16)[name = tensor("op_22630_cast_fp16")]; + tensor k_207_perm_0 = const()[name = tensor("k_207_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_22637_begin_0 = const()[name = tensor("op_22637_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_22637_end_0 = const()[name = tensor("op_22637_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_22637_end_mask_0 = const()[name = tensor("op_22637_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_207_cast_fp16 = transpose(perm = k_207_perm_0, x = k_205_cast_fp16)[name = tensor("transpose_88")]; + tensor var_22637_cast_fp16 = slice_by_index(begin = var_22637_begin_0, end = var_22637_end_0, end_mask = var_22637_end_mask_0, x = k_207_cast_fp16)[name = tensor("op_22637_cast_fp16")]; + tensor var_22641_begin_0 = const()[name = tensor("op_22641_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_22641_end_0 = const()[name = tensor("op_22641_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_22641_end_mask_0 = const()[name = tensor("op_22641_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22641_cast_fp16 = slice_by_index(begin = var_22641_begin_0, end = var_22641_end_0, end_mask = var_22641_end_mask_0, x = k_207_cast_fp16)[name = tensor("op_22641_cast_fp16")]; + tensor var_22645_begin_0 = const()[name = tensor("op_22645_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_22645_end_0 = const()[name = tensor("op_22645_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_22645_end_mask_0 = const()[name = tensor("op_22645_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22645_cast_fp16 = slice_by_index(begin = var_22645_begin_0, end = var_22645_end_0, end_mask = var_22645_end_mask_0, x = k_207_cast_fp16)[name = tensor("op_22645_cast_fp16")]; + tensor var_22649_begin_0 = const()[name = tensor("op_22649_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_22649_end_0 = const()[name = tensor("op_22649_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_22649_end_mask_0 = const()[name = tensor("op_22649_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22649_cast_fp16 = slice_by_index(begin = var_22649_begin_0, end = var_22649_end_0, end_mask = var_22649_end_mask_0, x = k_207_cast_fp16)[name = tensor("op_22649_cast_fp16")]; + tensor var_22653_begin_0 = const()[name = tensor("op_22653_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_22653_end_0 = const()[name = tensor("op_22653_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_22653_end_mask_0 = const()[name = tensor("op_22653_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22653_cast_fp16 = slice_by_index(begin = var_22653_begin_0, end = var_22653_end_0, end_mask = var_22653_end_mask_0, x = k_207_cast_fp16)[name = tensor("op_22653_cast_fp16")]; + tensor var_22657_begin_0 = const()[name = tensor("op_22657_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_22657_end_0 = const()[name = tensor("op_22657_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_22657_end_mask_0 = const()[name = tensor("op_22657_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22657_cast_fp16 = slice_by_index(begin = var_22657_begin_0, end = var_22657_end_0, end_mask = var_22657_end_mask_0, x = k_207_cast_fp16)[name = tensor("op_22657_cast_fp16")]; + tensor var_22661_begin_0 = const()[name = tensor("op_22661_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_22661_end_0 = const()[name = tensor("op_22661_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_22661_end_mask_0 = const()[name = tensor("op_22661_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22661_cast_fp16 = slice_by_index(begin = var_22661_begin_0, end = var_22661_end_0, end_mask = var_22661_end_mask_0, x = k_207_cast_fp16)[name = tensor("op_22661_cast_fp16")]; + tensor var_22665_begin_0 = const()[name = tensor("op_22665_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_22665_end_0 = const()[name = tensor("op_22665_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_22665_end_mask_0 = const()[name = tensor("op_22665_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22665_cast_fp16 = slice_by_index(begin = var_22665_begin_0, end = var_22665_end_0, end_mask = var_22665_end_mask_0, x = k_207_cast_fp16)[name = tensor("op_22665_cast_fp16")]; + tensor var_22669_begin_0 = const()[name = tensor("op_22669_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_22669_end_0 = const()[name = tensor("op_22669_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_22669_end_mask_0 = const()[name = tensor("op_22669_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22669_cast_fp16 = slice_by_index(begin = var_22669_begin_0, end = var_22669_end_0, end_mask = var_22669_end_mask_0, x = k_207_cast_fp16)[name = tensor("op_22669_cast_fp16")]; + tensor var_22673_begin_0 = const()[name = tensor("op_22673_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_22673_end_0 = const()[name = tensor("op_22673_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_22673_end_mask_0 = const()[name = tensor("op_22673_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22673_cast_fp16 = slice_by_index(begin = var_22673_begin_0, end = var_22673_end_0, end_mask = var_22673_end_mask_0, x = k_207_cast_fp16)[name = tensor("op_22673_cast_fp16")]; + tensor var_22677_begin_0 = const()[name = tensor("op_22677_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_22677_end_0 = const()[name = tensor("op_22677_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_22677_end_mask_0 = const()[name = tensor("op_22677_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22677_cast_fp16 = slice_by_index(begin = var_22677_begin_0, end = var_22677_end_0, end_mask = var_22677_end_mask_0, x = k_207_cast_fp16)[name = tensor("op_22677_cast_fp16")]; + tensor var_22681_begin_0 = const()[name = tensor("op_22681_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_22681_end_0 = const()[name = tensor("op_22681_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_22681_end_mask_0 = const()[name = tensor("op_22681_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22681_cast_fp16 = slice_by_index(begin = var_22681_begin_0, end = var_22681_end_0, end_mask = var_22681_end_mask_0, x = k_207_cast_fp16)[name = tensor("op_22681_cast_fp16")]; + tensor var_22685_begin_0 = const()[name = tensor("op_22685_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_22685_end_0 = const()[name = tensor("op_22685_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_22685_end_mask_0 = const()[name = tensor("op_22685_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22685_cast_fp16 = slice_by_index(begin = var_22685_begin_0, end = var_22685_end_0, end_mask = var_22685_end_mask_0, x = k_207_cast_fp16)[name = tensor("op_22685_cast_fp16")]; + tensor var_22689_begin_0 = const()[name = tensor("op_22689_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_22689_end_0 = const()[name = tensor("op_22689_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_22689_end_mask_0 = const()[name = tensor("op_22689_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22689_cast_fp16 = slice_by_index(begin = var_22689_begin_0, end = var_22689_end_0, end_mask = var_22689_end_mask_0, x = k_207_cast_fp16)[name = tensor("op_22689_cast_fp16")]; + tensor var_22693_begin_0 = const()[name = tensor("op_22693_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_22693_end_0 = const()[name = tensor("op_22693_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_22693_end_mask_0 = const()[name = tensor("op_22693_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22693_cast_fp16 = slice_by_index(begin = var_22693_begin_0, end = var_22693_end_0, end_mask = var_22693_end_mask_0, x = k_207_cast_fp16)[name = tensor("op_22693_cast_fp16")]; + tensor var_22697_begin_0 = const()[name = tensor("op_22697_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_22697_end_0 = const()[name = tensor("op_22697_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_22697_end_mask_0 = const()[name = tensor("op_22697_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22697_cast_fp16 = slice_by_index(begin = var_22697_begin_0, end = var_22697_end_0, end_mask = var_22697_end_mask_0, x = k_207_cast_fp16)[name = tensor("op_22697_cast_fp16")]; + tensor var_22701_begin_0 = const()[name = tensor("op_22701_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_22701_end_0 = const()[name = tensor("op_22701_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_22701_end_mask_0 = const()[name = tensor("op_22701_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22701_cast_fp16 = slice_by_index(begin = var_22701_begin_0, end = var_22701_end_0, end_mask = var_22701_end_mask_0, x = k_207_cast_fp16)[name = tensor("op_22701_cast_fp16")]; + tensor var_22705_begin_0 = const()[name = tensor("op_22705_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_22705_end_0 = const()[name = tensor("op_22705_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_22705_end_mask_0 = const()[name = tensor("op_22705_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22705_cast_fp16 = slice_by_index(begin = var_22705_begin_0, end = var_22705_end_0, end_mask = var_22705_end_mask_0, x = k_207_cast_fp16)[name = tensor("op_22705_cast_fp16")]; + tensor var_22709_begin_0 = const()[name = tensor("op_22709_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_22709_end_0 = const()[name = tensor("op_22709_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_22709_end_mask_0 = const()[name = tensor("op_22709_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22709_cast_fp16 = slice_by_index(begin = var_22709_begin_0, end = var_22709_end_0, end_mask = var_22709_end_mask_0, x = k_207_cast_fp16)[name = tensor("op_22709_cast_fp16")]; + tensor var_22713_begin_0 = const()[name = tensor("op_22713_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_22713_end_0 = const()[name = tensor("op_22713_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_22713_end_mask_0 = const()[name = tensor("op_22713_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22713_cast_fp16 = slice_by_index(begin = var_22713_begin_0, end = var_22713_end_0, end_mask = var_22713_end_mask_0, x = k_207_cast_fp16)[name = tensor("op_22713_cast_fp16")]; + tensor var_22715_begin_0 = const()[name = tensor("op_22715_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_22715_end_0 = const()[name = tensor("op_22715_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_22715_end_mask_0 = const()[name = tensor("op_22715_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22715_cast_fp16 = slice_by_index(begin = var_22715_begin_0, end = var_22715_end_0, end_mask = var_22715_end_mask_0, x = v_103_cast_fp16)[name = tensor("op_22715_cast_fp16")]; + tensor var_22719_begin_0 = const()[name = tensor("op_22719_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_22719_end_0 = const()[name = tensor("op_22719_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_22719_end_mask_0 = const()[name = tensor("op_22719_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22719_cast_fp16 = slice_by_index(begin = var_22719_begin_0, end = var_22719_end_0, end_mask = var_22719_end_mask_0, x = v_103_cast_fp16)[name = tensor("op_22719_cast_fp16")]; + tensor var_22723_begin_0 = const()[name = tensor("op_22723_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_22723_end_0 = const()[name = tensor("op_22723_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_22723_end_mask_0 = const()[name = tensor("op_22723_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22723_cast_fp16 = slice_by_index(begin = var_22723_begin_0, end = var_22723_end_0, end_mask = var_22723_end_mask_0, x = v_103_cast_fp16)[name = tensor("op_22723_cast_fp16")]; + tensor var_22727_begin_0 = const()[name = tensor("op_22727_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_22727_end_0 = const()[name = tensor("op_22727_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_22727_end_mask_0 = const()[name = tensor("op_22727_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22727_cast_fp16 = slice_by_index(begin = var_22727_begin_0, end = var_22727_end_0, end_mask = var_22727_end_mask_0, x = v_103_cast_fp16)[name = tensor("op_22727_cast_fp16")]; + tensor var_22731_begin_0 = const()[name = tensor("op_22731_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_22731_end_0 = const()[name = tensor("op_22731_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_22731_end_mask_0 = const()[name = tensor("op_22731_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22731_cast_fp16 = slice_by_index(begin = var_22731_begin_0, end = var_22731_end_0, end_mask = var_22731_end_mask_0, x = v_103_cast_fp16)[name = tensor("op_22731_cast_fp16")]; + tensor var_22735_begin_0 = const()[name = tensor("op_22735_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_22735_end_0 = const()[name = tensor("op_22735_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_22735_end_mask_0 = const()[name = tensor("op_22735_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22735_cast_fp16 = slice_by_index(begin = var_22735_begin_0, end = var_22735_end_0, end_mask = var_22735_end_mask_0, x = v_103_cast_fp16)[name = tensor("op_22735_cast_fp16")]; + tensor var_22739_begin_0 = const()[name = tensor("op_22739_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_22739_end_0 = const()[name = tensor("op_22739_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_22739_end_mask_0 = const()[name = tensor("op_22739_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22739_cast_fp16 = slice_by_index(begin = var_22739_begin_0, end = var_22739_end_0, end_mask = var_22739_end_mask_0, x = v_103_cast_fp16)[name = tensor("op_22739_cast_fp16")]; + tensor var_22743_begin_0 = const()[name = tensor("op_22743_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_22743_end_0 = const()[name = tensor("op_22743_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_22743_end_mask_0 = const()[name = tensor("op_22743_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22743_cast_fp16 = slice_by_index(begin = var_22743_begin_0, end = var_22743_end_0, end_mask = var_22743_end_mask_0, x = v_103_cast_fp16)[name = tensor("op_22743_cast_fp16")]; + tensor var_22747_begin_0 = const()[name = tensor("op_22747_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_22747_end_0 = const()[name = tensor("op_22747_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_22747_end_mask_0 = const()[name = tensor("op_22747_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22747_cast_fp16 = slice_by_index(begin = var_22747_begin_0, end = var_22747_end_0, end_mask = var_22747_end_mask_0, x = v_103_cast_fp16)[name = tensor("op_22747_cast_fp16")]; + tensor var_22751_begin_0 = const()[name = tensor("op_22751_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_22751_end_0 = const()[name = tensor("op_22751_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_22751_end_mask_0 = const()[name = tensor("op_22751_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22751_cast_fp16 = slice_by_index(begin = var_22751_begin_0, end = var_22751_end_0, end_mask = var_22751_end_mask_0, x = v_103_cast_fp16)[name = tensor("op_22751_cast_fp16")]; + tensor var_22755_begin_0 = const()[name = tensor("op_22755_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_22755_end_0 = const()[name = tensor("op_22755_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_22755_end_mask_0 = const()[name = tensor("op_22755_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22755_cast_fp16 = slice_by_index(begin = var_22755_begin_0, end = var_22755_end_0, end_mask = var_22755_end_mask_0, x = v_103_cast_fp16)[name = tensor("op_22755_cast_fp16")]; + tensor var_22759_begin_0 = const()[name = tensor("op_22759_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_22759_end_0 = const()[name = tensor("op_22759_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_22759_end_mask_0 = const()[name = tensor("op_22759_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22759_cast_fp16 = slice_by_index(begin = var_22759_begin_0, end = var_22759_end_0, end_mask = var_22759_end_mask_0, x = v_103_cast_fp16)[name = tensor("op_22759_cast_fp16")]; + tensor var_22763_begin_0 = const()[name = tensor("op_22763_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_22763_end_0 = const()[name = tensor("op_22763_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_22763_end_mask_0 = const()[name = tensor("op_22763_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22763_cast_fp16 = slice_by_index(begin = var_22763_begin_0, end = var_22763_end_0, end_mask = var_22763_end_mask_0, x = v_103_cast_fp16)[name = tensor("op_22763_cast_fp16")]; + tensor var_22767_begin_0 = const()[name = tensor("op_22767_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_22767_end_0 = const()[name = tensor("op_22767_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_22767_end_mask_0 = const()[name = tensor("op_22767_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22767_cast_fp16 = slice_by_index(begin = var_22767_begin_0, end = var_22767_end_0, end_mask = var_22767_end_mask_0, x = v_103_cast_fp16)[name = tensor("op_22767_cast_fp16")]; + tensor var_22771_begin_0 = const()[name = tensor("op_22771_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_22771_end_0 = const()[name = tensor("op_22771_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_22771_end_mask_0 = const()[name = tensor("op_22771_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22771_cast_fp16 = slice_by_index(begin = var_22771_begin_0, end = var_22771_end_0, end_mask = var_22771_end_mask_0, x = v_103_cast_fp16)[name = tensor("op_22771_cast_fp16")]; + tensor var_22775_begin_0 = const()[name = tensor("op_22775_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_22775_end_0 = const()[name = tensor("op_22775_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_22775_end_mask_0 = const()[name = tensor("op_22775_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22775_cast_fp16 = slice_by_index(begin = var_22775_begin_0, end = var_22775_end_0, end_mask = var_22775_end_mask_0, x = v_103_cast_fp16)[name = tensor("op_22775_cast_fp16")]; + tensor var_22779_begin_0 = const()[name = tensor("op_22779_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_22779_end_0 = const()[name = tensor("op_22779_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_22779_end_mask_0 = const()[name = tensor("op_22779_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22779_cast_fp16 = slice_by_index(begin = var_22779_begin_0, end = var_22779_end_0, end_mask = var_22779_end_mask_0, x = v_103_cast_fp16)[name = tensor("op_22779_cast_fp16")]; + tensor var_22783_begin_0 = const()[name = tensor("op_22783_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_22783_end_0 = const()[name = tensor("op_22783_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_22783_end_mask_0 = const()[name = tensor("op_22783_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22783_cast_fp16 = slice_by_index(begin = var_22783_begin_0, end = var_22783_end_0, end_mask = var_22783_end_mask_0, x = v_103_cast_fp16)[name = tensor("op_22783_cast_fp16")]; + tensor var_22787_begin_0 = const()[name = tensor("op_22787_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_22787_end_0 = const()[name = tensor("op_22787_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_22787_end_mask_0 = const()[name = tensor("op_22787_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22787_cast_fp16 = slice_by_index(begin = var_22787_begin_0, end = var_22787_end_0, end_mask = var_22787_end_mask_0, x = v_103_cast_fp16)[name = tensor("op_22787_cast_fp16")]; + tensor var_22791_begin_0 = const()[name = tensor("op_22791_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_22791_end_0 = const()[name = tensor("op_22791_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_22791_end_mask_0 = const()[name = tensor("op_22791_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22791_cast_fp16 = slice_by_index(begin = var_22791_begin_0, end = var_22791_end_0, end_mask = var_22791_end_mask_0, x = v_103_cast_fp16)[name = tensor("op_22791_cast_fp16")]; + tensor var_22795_equation_0 = const()[name = tensor("op_22795_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22795_cast_fp16 = einsum(equation = var_22795_equation_0, values = (var_22637_cast_fp16, var_22554_cast_fp16))[name = tensor("op_22795_cast_fp16")]; + tensor var_22796_to_fp16 = const()[name = tensor("op_22796_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1881_cast_fp16 = mul(x = var_22795_cast_fp16, y = var_22796_to_fp16)[name = tensor("aw_1881_cast_fp16")]; + tensor var_22799_equation_0 = const()[name = tensor("op_22799_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22799_cast_fp16 = einsum(equation = var_22799_equation_0, values = (var_22641_cast_fp16, var_22558_cast_fp16))[name = tensor("op_22799_cast_fp16")]; + tensor var_22800_to_fp16 = const()[name = tensor("op_22800_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1883_cast_fp16 = mul(x = var_22799_cast_fp16, y = var_22800_to_fp16)[name = tensor("aw_1883_cast_fp16")]; + tensor var_22803_equation_0 = const()[name = tensor("op_22803_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22803_cast_fp16 = einsum(equation = var_22803_equation_0, values = (var_22645_cast_fp16, var_22562_cast_fp16))[name = tensor("op_22803_cast_fp16")]; + tensor var_22804_to_fp16 = const()[name = tensor("op_22804_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1885_cast_fp16 = mul(x = var_22803_cast_fp16, y = var_22804_to_fp16)[name = tensor("aw_1885_cast_fp16")]; + tensor var_22807_equation_0 = const()[name = tensor("op_22807_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22807_cast_fp16 = einsum(equation = var_22807_equation_0, values = (var_22649_cast_fp16, var_22566_cast_fp16))[name = tensor("op_22807_cast_fp16")]; + tensor var_22808_to_fp16 = const()[name = tensor("op_22808_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1887_cast_fp16 = mul(x = var_22807_cast_fp16, y = var_22808_to_fp16)[name = tensor("aw_1887_cast_fp16")]; + tensor var_22811_equation_0 = const()[name = tensor("op_22811_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22811_cast_fp16 = einsum(equation = var_22811_equation_0, values = (var_22653_cast_fp16, var_22570_cast_fp16))[name = tensor("op_22811_cast_fp16")]; + tensor var_22812_to_fp16 = const()[name = tensor("op_22812_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1889_cast_fp16 = mul(x = var_22811_cast_fp16, y = var_22812_to_fp16)[name = tensor("aw_1889_cast_fp16")]; + tensor var_22815_equation_0 = const()[name = tensor("op_22815_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22815_cast_fp16 = einsum(equation = var_22815_equation_0, values = (var_22657_cast_fp16, var_22574_cast_fp16))[name = tensor("op_22815_cast_fp16")]; + tensor var_22816_to_fp16 = const()[name = tensor("op_22816_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1891_cast_fp16 = mul(x = var_22815_cast_fp16, y = var_22816_to_fp16)[name = tensor("aw_1891_cast_fp16")]; + tensor var_22819_equation_0 = const()[name = tensor("op_22819_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22819_cast_fp16 = einsum(equation = var_22819_equation_0, values = (var_22661_cast_fp16, var_22578_cast_fp16))[name = tensor("op_22819_cast_fp16")]; + tensor var_22820_to_fp16 = const()[name = tensor("op_22820_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1893_cast_fp16 = mul(x = var_22819_cast_fp16, y = var_22820_to_fp16)[name = tensor("aw_1893_cast_fp16")]; + tensor var_22823_equation_0 = const()[name = tensor("op_22823_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22823_cast_fp16 = einsum(equation = var_22823_equation_0, values = (var_22665_cast_fp16, var_22582_cast_fp16))[name = tensor("op_22823_cast_fp16")]; + tensor var_22824_to_fp16 = const()[name = tensor("op_22824_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1895_cast_fp16 = mul(x = var_22823_cast_fp16, y = var_22824_to_fp16)[name = tensor("aw_1895_cast_fp16")]; + tensor var_22827_equation_0 = const()[name = tensor("op_22827_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22827_cast_fp16 = einsum(equation = var_22827_equation_0, values = (var_22669_cast_fp16, var_22586_cast_fp16))[name = tensor("op_22827_cast_fp16")]; + tensor var_22828_to_fp16 = const()[name = tensor("op_22828_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1897_cast_fp16 = mul(x = var_22827_cast_fp16, y = var_22828_to_fp16)[name = tensor("aw_1897_cast_fp16")]; + tensor var_22831_equation_0 = const()[name = tensor("op_22831_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22831_cast_fp16 = einsum(equation = var_22831_equation_0, values = (var_22673_cast_fp16, var_22590_cast_fp16))[name = tensor("op_22831_cast_fp16")]; + tensor var_22832_to_fp16 = const()[name = tensor("op_22832_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1899_cast_fp16 = mul(x = var_22831_cast_fp16, y = var_22832_to_fp16)[name = tensor("aw_1899_cast_fp16")]; + tensor var_22835_equation_0 = const()[name = tensor("op_22835_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22835_cast_fp16 = einsum(equation = var_22835_equation_0, values = (var_22677_cast_fp16, var_22594_cast_fp16))[name = tensor("op_22835_cast_fp16")]; + tensor var_22836_to_fp16 = const()[name = tensor("op_22836_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1901_cast_fp16 = mul(x = var_22835_cast_fp16, y = var_22836_to_fp16)[name = tensor("aw_1901_cast_fp16")]; + tensor var_22839_equation_0 = const()[name = tensor("op_22839_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22839_cast_fp16 = einsum(equation = var_22839_equation_0, values = (var_22681_cast_fp16, var_22598_cast_fp16))[name = tensor("op_22839_cast_fp16")]; + tensor var_22840_to_fp16 = const()[name = tensor("op_22840_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1903_cast_fp16 = mul(x = var_22839_cast_fp16, y = var_22840_to_fp16)[name = tensor("aw_1903_cast_fp16")]; + tensor var_22843_equation_0 = const()[name = tensor("op_22843_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22843_cast_fp16 = einsum(equation = var_22843_equation_0, values = (var_22685_cast_fp16, var_22602_cast_fp16))[name = tensor("op_22843_cast_fp16")]; + tensor var_22844_to_fp16 = const()[name = tensor("op_22844_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1905_cast_fp16 = mul(x = var_22843_cast_fp16, y = var_22844_to_fp16)[name = tensor("aw_1905_cast_fp16")]; + tensor var_22847_equation_0 = const()[name = tensor("op_22847_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22847_cast_fp16 = einsum(equation = var_22847_equation_0, values = (var_22689_cast_fp16, var_22606_cast_fp16))[name = tensor("op_22847_cast_fp16")]; + tensor var_22848_to_fp16 = const()[name = tensor("op_22848_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1907_cast_fp16 = mul(x = var_22847_cast_fp16, y = var_22848_to_fp16)[name = tensor("aw_1907_cast_fp16")]; + tensor var_22851_equation_0 = const()[name = tensor("op_22851_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22851_cast_fp16 = einsum(equation = var_22851_equation_0, values = (var_22693_cast_fp16, var_22610_cast_fp16))[name = tensor("op_22851_cast_fp16")]; + tensor var_22852_to_fp16 = const()[name = tensor("op_22852_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1909_cast_fp16 = mul(x = var_22851_cast_fp16, y = var_22852_to_fp16)[name = tensor("aw_1909_cast_fp16")]; + tensor var_22855_equation_0 = const()[name = tensor("op_22855_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22855_cast_fp16 = einsum(equation = var_22855_equation_0, values = (var_22697_cast_fp16, var_22614_cast_fp16))[name = tensor("op_22855_cast_fp16")]; + tensor var_22856_to_fp16 = const()[name = tensor("op_22856_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1911_cast_fp16 = mul(x = var_22855_cast_fp16, y = var_22856_to_fp16)[name = tensor("aw_1911_cast_fp16")]; + tensor var_22859_equation_0 = const()[name = tensor("op_22859_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22859_cast_fp16 = einsum(equation = var_22859_equation_0, values = (var_22701_cast_fp16, var_22618_cast_fp16))[name = tensor("op_22859_cast_fp16")]; + tensor var_22860_to_fp16 = const()[name = tensor("op_22860_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1913_cast_fp16 = mul(x = var_22859_cast_fp16, y = var_22860_to_fp16)[name = tensor("aw_1913_cast_fp16")]; + tensor var_22863_equation_0 = const()[name = tensor("op_22863_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22863_cast_fp16 = einsum(equation = var_22863_equation_0, values = (var_22705_cast_fp16, var_22622_cast_fp16))[name = tensor("op_22863_cast_fp16")]; + tensor var_22864_to_fp16 = const()[name = tensor("op_22864_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1915_cast_fp16 = mul(x = var_22863_cast_fp16, y = var_22864_to_fp16)[name = tensor("aw_1915_cast_fp16")]; + tensor var_22867_equation_0 = const()[name = tensor("op_22867_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22867_cast_fp16 = einsum(equation = var_22867_equation_0, values = (var_22709_cast_fp16, var_22626_cast_fp16))[name = tensor("op_22867_cast_fp16")]; + tensor var_22868_to_fp16 = const()[name = tensor("op_22868_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1917_cast_fp16 = mul(x = var_22867_cast_fp16, y = var_22868_to_fp16)[name = tensor("aw_1917_cast_fp16")]; + tensor var_22871_equation_0 = const()[name = tensor("op_22871_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22871_cast_fp16 = einsum(equation = var_22871_equation_0, values = (var_22713_cast_fp16, var_22630_cast_fp16))[name = tensor("op_22871_cast_fp16")]; + tensor var_22872_to_fp16 = const()[name = tensor("op_22872_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1919_cast_fp16 = mul(x = var_22871_cast_fp16, y = var_22872_to_fp16)[name = tensor("aw_1919_cast_fp16")]; + tensor var_22874_cast_fp16 = softmax(axis = var_21077, x = aw_1881_cast_fp16)[name = tensor("op_22874_cast_fp16")]; + tensor var_22875_cast_fp16 = softmax(axis = var_21077, x = aw_1883_cast_fp16)[name = tensor("op_22875_cast_fp16")]; + tensor var_22876_cast_fp16 = softmax(axis = var_21077, x = aw_1885_cast_fp16)[name = tensor("op_22876_cast_fp16")]; + tensor var_22877_cast_fp16 = softmax(axis = var_21077, x = aw_1887_cast_fp16)[name = tensor("op_22877_cast_fp16")]; + tensor var_22878_cast_fp16 = softmax(axis = var_21077, x = aw_1889_cast_fp16)[name = tensor("op_22878_cast_fp16")]; + tensor var_22879_cast_fp16 = softmax(axis = var_21077, x = aw_1891_cast_fp16)[name = tensor("op_22879_cast_fp16")]; + tensor var_22880_cast_fp16 = softmax(axis = var_21077, x = aw_1893_cast_fp16)[name = tensor("op_22880_cast_fp16")]; + tensor var_22881_cast_fp16 = softmax(axis = var_21077, x = aw_1895_cast_fp16)[name = tensor("op_22881_cast_fp16")]; + tensor var_22882_cast_fp16 = softmax(axis = var_21077, x = aw_1897_cast_fp16)[name = tensor("op_22882_cast_fp16")]; + tensor var_22883_cast_fp16 = softmax(axis = var_21077, x = aw_1899_cast_fp16)[name = tensor("op_22883_cast_fp16")]; + tensor var_22884_cast_fp16 = softmax(axis = var_21077, x = aw_1901_cast_fp16)[name = tensor("op_22884_cast_fp16")]; + tensor var_22885_cast_fp16 = softmax(axis = var_21077, x = aw_1903_cast_fp16)[name = tensor("op_22885_cast_fp16")]; + tensor var_22886_cast_fp16 = softmax(axis = var_21077, x = aw_1905_cast_fp16)[name = tensor("op_22886_cast_fp16")]; + tensor var_22887_cast_fp16 = softmax(axis = var_21077, x = aw_1907_cast_fp16)[name = tensor("op_22887_cast_fp16")]; + tensor var_22888_cast_fp16 = softmax(axis = var_21077, x = aw_1909_cast_fp16)[name = tensor("op_22888_cast_fp16")]; + tensor var_22889_cast_fp16 = softmax(axis = var_21077, x = aw_1911_cast_fp16)[name = tensor("op_22889_cast_fp16")]; + tensor var_22890_cast_fp16 = softmax(axis = var_21077, x = aw_1913_cast_fp16)[name = tensor("op_22890_cast_fp16")]; + tensor var_22891_cast_fp16 = softmax(axis = var_21077, x = aw_1915_cast_fp16)[name = tensor("op_22891_cast_fp16")]; + tensor var_22892_cast_fp16 = softmax(axis = var_21077, x = aw_1917_cast_fp16)[name = tensor("op_22892_cast_fp16")]; + tensor var_22893_cast_fp16 = softmax(axis = var_21077, x = aw_1919_cast_fp16)[name = tensor("op_22893_cast_fp16")]; + tensor var_22895_equation_0 = const()[name = tensor("op_22895_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22895_cast_fp16 = einsum(equation = var_22895_equation_0, values = (var_22715_cast_fp16, var_22874_cast_fp16))[name = tensor("op_22895_cast_fp16")]; + tensor var_22897_equation_0 = const()[name = tensor("op_22897_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22897_cast_fp16 = einsum(equation = var_22897_equation_0, values = (var_22719_cast_fp16, var_22875_cast_fp16))[name = tensor("op_22897_cast_fp16")]; + tensor var_22899_equation_0 = const()[name = tensor("op_22899_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22899_cast_fp16 = einsum(equation = var_22899_equation_0, values = (var_22723_cast_fp16, var_22876_cast_fp16))[name = tensor("op_22899_cast_fp16")]; + tensor var_22901_equation_0 = const()[name = tensor("op_22901_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22901_cast_fp16 = einsum(equation = var_22901_equation_0, values = (var_22727_cast_fp16, var_22877_cast_fp16))[name = tensor("op_22901_cast_fp16")]; + tensor var_22903_equation_0 = const()[name = tensor("op_22903_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22903_cast_fp16 = einsum(equation = var_22903_equation_0, values = (var_22731_cast_fp16, var_22878_cast_fp16))[name = tensor("op_22903_cast_fp16")]; + tensor var_22905_equation_0 = const()[name = tensor("op_22905_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22905_cast_fp16 = einsum(equation = var_22905_equation_0, values = (var_22735_cast_fp16, var_22879_cast_fp16))[name = tensor("op_22905_cast_fp16")]; + tensor var_22907_equation_0 = const()[name = tensor("op_22907_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22907_cast_fp16 = einsum(equation = var_22907_equation_0, values = (var_22739_cast_fp16, var_22880_cast_fp16))[name = tensor("op_22907_cast_fp16")]; + tensor var_22909_equation_0 = const()[name = tensor("op_22909_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22909_cast_fp16 = einsum(equation = var_22909_equation_0, values = (var_22743_cast_fp16, var_22881_cast_fp16))[name = tensor("op_22909_cast_fp16")]; + tensor var_22911_equation_0 = const()[name = tensor("op_22911_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22911_cast_fp16 = einsum(equation = var_22911_equation_0, values = (var_22747_cast_fp16, var_22882_cast_fp16))[name = tensor("op_22911_cast_fp16")]; + tensor var_22913_equation_0 = const()[name = tensor("op_22913_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22913_cast_fp16 = einsum(equation = var_22913_equation_0, values = (var_22751_cast_fp16, var_22883_cast_fp16))[name = tensor("op_22913_cast_fp16")]; + tensor var_22915_equation_0 = const()[name = tensor("op_22915_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22915_cast_fp16 = einsum(equation = var_22915_equation_0, values = (var_22755_cast_fp16, var_22884_cast_fp16))[name = tensor("op_22915_cast_fp16")]; + tensor var_22917_equation_0 = const()[name = tensor("op_22917_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22917_cast_fp16 = einsum(equation = var_22917_equation_0, values = (var_22759_cast_fp16, var_22885_cast_fp16))[name = tensor("op_22917_cast_fp16")]; + tensor var_22919_equation_0 = const()[name = tensor("op_22919_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22919_cast_fp16 = einsum(equation = var_22919_equation_0, values = (var_22763_cast_fp16, var_22886_cast_fp16))[name = tensor("op_22919_cast_fp16")]; + tensor var_22921_equation_0 = const()[name = tensor("op_22921_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22921_cast_fp16 = einsum(equation = var_22921_equation_0, values = (var_22767_cast_fp16, var_22887_cast_fp16))[name = tensor("op_22921_cast_fp16")]; + tensor var_22923_equation_0 = const()[name = tensor("op_22923_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22923_cast_fp16 = einsum(equation = var_22923_equation_0, values = (var_22771_cast_fp16, var_22888_cast_fp16))[name = tensor("op_22923_cast_fp16")]; + tensor var_22925_equation_0 = const()[name = tensor("op_22925_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22925_cast_fp16 = einsum(equation = var_22925_equation_0, values = (var_22775_cast_fp16, var_22889_cast_fp16))[name = tensor("op_22925_cast_fp16")]; + tensor var_22927_equation_0 = const()[name = tensor("op_22927_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22927_cast_fp16 = einsum(equation = var_22927_equation_0, values = (var_22779_cast_fp16, var_22890_cast_fp16))[name = tensor("op_22927_cast_fp16")]; + tensor var_22929_equation_0 = const()[name = tensor("op_22929_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22929_cast_fp16 = einsum(equation = var_22929_equation_0, values = (var_22783_cast_fp16, var_22891_cast_fp16))[name = tensor("op_22929_cast_fp16")]; + tensor var_22931_equation_0 = const()[name = tensor("op_22931_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22931_cast_fp16 = einsum(equation = var_22931_equation_0, values = (var_22787_cast_fp16, var_22892_cast_fp16))[name = tensor("op_22931_cast_fp16")]; + tensor var_22933_equation_0 = const()[name = tensor("op_22933_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22933_cast_fp16 = einsum(equation = var_22933_equation_0, values = (var_22791_cast_fp16, var_22893_cast_fp16))[name = tensor("op_22933_cast_fp16")]; + tensor input_337_interleave_0 = const()[name = tensor("input_337_interleave_0"), val = tensor(false)]; + tensor input_337_cast_fp16 = concat(axis = var_21077, interleave = input_337_interleave_0, values = (var_22895_cast_fp16, var_22897_cast_fp16, var_22899_cast_fp16, var_22901_cast_fp16, var_22903_cast_fp16, var_22905_cast_fp16, var_22907_cast_fp16, var_22909_cast_fp16, var_22911_cast_fp16, var_22913_cast_fp16, var_22915_cast_fp16, var_22917_cast_fp16, var_22919_cast_fp16, var_22921_cast_fp16, var_22923_cast_fp16, var_22925_cast_fp16, var_22927_cast_fp16, var_22929_cast_fp16, var_22931_cast_fp16, var_22933_cast_fp16))[name = tensor("input_337_cast_fp16")]; + tensor var_22939 = const()[name = tensor("op_22939"), val = tensor([1, 1])]; + tensor var_22941 = const()[name = tensor("op_22941"), val = tensor([1, 1])]; + tensor var_22943_pad_type_0 = const()[name = tensor("op_22943_pad_type_0"), val = tensor("custom")]; + tensor var_22943_pad_0 = const()[name = tensor("op_22943_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_1_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_1_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1822505344)))]; + tensor mid_block_attentions_0_transformer_blocks_1_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_1_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1825782208)))]; + tensor var_22943_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_1_attn2_to_out_0_bias_to_fp16, dilations = var_22941, groups = var_21077, pad = var_22943_pad_0, pad_type = var_22943_pad_type_0, strides = var_22939, weight = mid_block_attentions_0_transformer_blocks_1_attn2_to_out_0_weight_to_fp16, x = input_337_cast_fp16)[name = tensor("op_22943_cast_fp16")]; + tensor inputs_155_cast_fp16 = add(x = var_22943_cast_fp16, y = inputs_153_cast_fp16)[name = tensor("inputs_155_cast_fp16")]; + tensor input_339_axes_0 = const()[name = tensor("input_339_axes_0"), val = tensor([1])]; + tensor input_339_gamma_0_to_fp16 = const()[name = tensor("input_339_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1825784832)))]; + tensor input_339_beta_0_to_fp16 = const()[name = tensor("input_339_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1825787456)))]; + tensor var_22953_to_fp16 = const()[name = tensor("op_22953_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_339_cast_fp16 = layer_norm(axes = input_339_axes_0, beta = input_339_beta_0_to_fp16, epsilon = var_22953_to_fp16, gamma = input_339_gamma_0_to_fp16, x = inputs_155_cast_fp16)[name = tensor("input_339_cast_fp16")]; + tensor var_22969 = const()[name = tensor("op_22969"), val = tensor([1, 1])]; + tensor var_22971 = const()[name = tensor("op_22971"), val = tensor([1, 1])]; + tensor var_22973_pad_type_0 = const()[name = tensor("op_22973_pad_type_0"), val = tensor("custom")]; + tensor var_22973_pad_0 = const()[name = tensor("op_22973_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_1_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_1_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1825790080)))]; + tensor mid_block_attentions_0_transformer_blocks_1_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_1_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1852004544)))]; + tensor var_22973_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_1_ff_net_0_proj_bias_to_fp16, dilations = var_22971, groups = var_21077, pad = var_22973_pad_0, pad_type = var_22973_pad_type_0, strides = var_22969, weight = mid_block_attentions_0_transformer_blocks_1_ff_net_0_proj_weight_to_fp16, x = input_339_cast_fp16)[name = tensor("op_22973_cast_fp16")]; + tensor var_22974_split_sizes_0 = const()[name = tensor("op_22974_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_22974_axis_0 = const()[name = tensor("op_22974_axis_0"), val = tensor(1)]; + tensor var_22974_cast_fp16_0, tensor var_22974_cast_fp16_1 = split(axis = var_22974_axis_0, split_sizes = var_22974_split_sizes_0, x = var_22973_cast_fp16)[name = tensor("op_22974_cast_fp16")]; + tensor var_22976_mode_0 = const()[name = tensor("op_22976_mode_0"), val = tensor("EXACT")]; + tensor var_22976_cast_fp16 = gelu(mode = var_22976_mode_0, x = var_22974_cast_fp16_1)[name = tensor("op_22976_cast_fp16")]; + tensor input_341_cast_fp16 = mul(x = var_22974_cast_fp16_0, y = var_22976_cast_fp16)[name = tensor("input_341_cast_fp16")]; + tensor var_22980 = const()[name = tensor("op_22980"), val = tensor([1, 1])]; + tensor var_22982 = const()[name = tensor("op_22982"), val = tensor([1, 1])]; + tensor var_22984_pad_type_0 = const()[name = tensor("op_22984_pad_type_0"), val = tensor("custom")]; + tensor var_22984_pad_0 = const()[name = tensor("op_22984_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_1_ff_net_2_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_1_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1852025088)))]; + tensor mid_block_attentions_0_transformer_blocks_1_ff_net_2_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_1_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1865132352)))]; + tensor var_22984_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_1_ff_net_2_bias_to_fp16, dilations = var_22982, groups = var_21077, pad = var_22984_pad_0, pad_type = var_22984_pad_type_0, strides = var_22980, weight = mid_block_attentions_0_transformer_blocks_1_ff_net_2_weight_to_fp16, x = input_341_cast_fp16)[name = tensor("op_22984_cast_fp16")]; + tensor inputs_157_cast_fp16 = add(x = var_22984_cast_fp16, y = inputs_155_cast_fp16)[name = tensor("inputs_157_cast_fp16")]; + tensor hidden_states_221_axes_0 = const()[name = tensor("hidden_states_221_axes_0"), val = tensor([1])]; + tensor hidden_states_221_gamma_0_to_fp16 = const()[name = tensor("hidden_states_221_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1865134976)))]; + tensor hidden_states_221_beta_0_to_fp16 = const()[name = tensor("hidden_states_221_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1865137600)))]; + tensor var_23000_to_fp16 = const()[name = tensor("op_23000_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_221_cast_fp16 = layer_norm(axes = hidden_states_221_axes_0, beta = hidden_states_221_beta_0_to_fp16, epsilon = var_23000_to_fp16, gamma = hidden_states_221_gamma_0_to_fp16, x = inputs_157_cast_fp16)[name = tensor("hidden_states_221_cast_fp16")]; + tensor var_23015 = const()[name = tensor("op_23015"), val = tensor([1, 1])]; + tensor var_23017 = const()[name = tensor("op_23017"), val = tensor([1, 1])]; + tensor q_105_pad_type_0 = const()[name = tensor("q_105_pad_type_0"), val = tensor("custom")]; + tensor q_105_pad_0 = const()[name = tensor("q_105_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_2_attn1_to_q_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_2_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1865140224)))]; + tensor q_105_cast_fp16 = conv(dilations = var_23017, groups = var_21077, pad = q_105_pad_0, pad_type = q_105_pad_type_0, strides = var_23015, weight = mid_block_attentions_0_transformer_blocks_2_attn1_to_q_weight_to_fp16, x = hidden_states_221_cast_fp16)[name = tensor("q_105_cast_fp16")]; + tensor var_23021 = const()[name = tensor("op_23021"), val = tensor([1, 1])]; + tensor var_23023 = const()[name = tensor("op_23023"), val = tensor([1, 1])]; + tensor k_209_pad_type_0 = const()[name = tensor("k_209_pad_type_0"), val = tensor("custom")]; + tensor k_209_pad_0 = const()[name = tensor("k_209_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_2_attn1_to_k_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_2_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1868417088)))]; + tensor k_209_cast_fp16 = conv(dilations = var_23023, groups = var_21077, pad = k_209_pad_0, pad_type = k_209_pad_type_0, strides = var_23021, weight = mid_block_attentions_0_transformer_blocks_2_attn1_to_k_weight_to_fp16, x = hidden_states_221_cast_fp16)[name = tensor("k_209_cast_fp16")]; + tensor var_23027 = const()[name = tensor("op_23027"), val = tensor([1, 1])]; + tensor var_23029 = const()[name = tensor("op_23029"), val = tensor([1, 1])]; + tensor v_105_pad_type_0 = const()[name = tensor("v_105_pad_type_0"), val = tensor("custom")]; + tensor v_105_pad_0 = const()[name = tensor("v_105_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_2_attn1_to_v_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_2_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1871693952)))]; + tensor v_105_cast_fp16 = conv(dilations = var_23029, groups = var_21077, pad = v_105_pad_0, pad_type = v_105_pad_type_0, strides = var_23027, weight = mid_block_attentions_0_transformer_blocks_2_attn1_to_v_weight_to_fp16, x = hidden_states_221_cast_fp16)[name = tensor("v_105_cast_fp16")]; + tensor var_23033_begin_0 = const()[name = tensor("op_23033_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23033_end_0 = const()[name = tensor("op_23033_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_23033_end_mask_0 = const()[name = tensor("op_23033_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23033_cast_fp16 = slice_by_index(begin = var_23033_begin_0, end = var_23033_end_0, end_mask = var_23033_end_mask_0, x = q_105_cast_fp16)[name = tensor("op_23033_cast_fp16")]; + tensor var_23037_begin_0 = const()[name = tensor("op_23037_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_23037_end_0 = const()[name = tensor("op_23037_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_23037_end_mask_0 = const()[name = tensor("op_23037_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23037_cast_fp16 = slice_by_index(begin = var_23037_begin_0, end = var_23037_end_0, end_mask = var_23037_end_mask_0, x = q_105_cast_fp16)[name = tensor("op_23037_cast_fp16")]; + tensor var_23041_begin_0 = const()[name = tensor("op_23041_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_23041_end_0 = const()[name = tensor("op_23041_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_23041_end_mask_0 = const()[name = tensor("op_23041_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23041_cast_fp16 = slice_by_index(begin = var_23041_begin_0, end = var_23041_end_0, end_mask = var_23041_end_mask_0, x = q_105_cast_fp16)[name = tensor("op_23041_cast_fp16")]; + tensor var_23045_begin_0 = const()[name = tensor("op_23045_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_23045_end_0 = const()[name = tensor("op_23045_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_23045_end_mask_0 = const()[name = tensor("op_23045_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23045_cast_fp16 = slice_by_index(begin = var_23045_begin_0, end = var_23045_end_0, end_mask = var_23045_end_mask_0, x = q_105_cast_fp16)[name = tensor("op_23045_cast_fp16")]; + tensor var_23049_begin_0 = const()[name = tensor("op_23049_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_23049_end_0 = const()[name = tensor("op_23049_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_23049_end_mask_0 = const()[name = tensor("op_23049_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23049_cast_fp16 = slice_by_index(begin = var_23049_begin_0, end = var_23049_end_0, end_mask = var_23049_end_mask_0, x = q_105_cast_fp16)[name = tensor("op_23049_cast_fp16")]; + tensor var_23053_begin_0 = const()[name = tensor("op_23053_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_23053_end_0 = const()[name = tensor("op_23053_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_23053_end_mask_0 = const()[name = tensor("op_23053_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23053_cast_fp16 = slice_by_index(begin = var_23053_begin_0, end = var_23053_end_0, end_mask = var_23053_end_mask_0, x = q_105_cast_fp16)[name = tensor("op_23053_cast_fp16")]; + tensor var_23057_begin_0 = const()[name = tensor("op_23057_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_23057_end_0 = const()[name = tensor("op_23057_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_23057_end_mask_0 = const()[name = tensor("op_23057_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23057_cast_fp16 = slice_by_index(begin = var_23057_begin_0, end = var_23057_end_0, end_mask = var_23057_end_mask_0, x = q_105_cast_fp16)[name = tensor("op_23057_cast_fp16")]; + tensor var_23061_begin_0 = const()[name = tensor("op_23061_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_23061_end_0 = const()[name = tensor("op_23061_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_23061_end_mask_0 = const()[name = tensor("op_23061_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23061_cast_fp16 = slice_by_index(begin = var_23061_begin_0, end = var_23061_end_0, end_mask = var_23061_end_mask_0, x = q_105_cast_fp16)[name = tensor("op_23061_cast_fp16")]; + tensor var_23065_begin_0 = const()[name = tensor("op_23065_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_23065_end_0 = const()[name = tensor("op_23065_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_23065_end_mask_0 = const()[name = tensor("op_23065_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23065_cast_fp16 = slice_by_index(begin = var_23065_begin_0, end = var_23065_end_0, end_mask = var_23065_end_mask_0, x = q_105_cast_fp16)[name = tensor("op_23065_cast_fp16")]; + tensor var_23069_begin_0 = const()[name = tensor("op_23069_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_23069_end_0 = const()[name = tensor("op_23069_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_23069_end_mask_0 = const()[name = tensor("op_23069_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23069_cast_fp16 = slice_by_index(begin = var_23069_begin_0, end = var_23069_end_0, end_mask = var_23069_end_mask_0, x = q_105_cast_fp16)[name = tensor("op_23069_cast_fp16")]; + tensor var_23073_begin_0 = const()[name = tensor("op_23073_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_23073_end_0 = const()[name = tensor("op_23073_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_23073_end_mask_0 = const()[name = tensor("op_23073_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23073_cast_fp16 = slice_by_index(begin = var_23073_begin_0, end = var_23073_end_0, end_mask = var_23073_end_mask_0, x = q_105_cast_fp16)[name = tensor("op_23073_cast_fp16")]; + tensor var_23077_begin_0 = const()[name = tensor("op_23077_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_23077_end_0 = const()[name = tensor("op_23077_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_23077_end_mask_0 = const()[name = tensor("op_23077_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23077_cast_fp16 = slice_by_index(begin = var_23077_begin_0, end = var_23077_end_0, end_mask = var_23077_end_mask_0, x = q_105_cast_fp16)[name = tensor("op_23077_cast_fp16")]; + tensor var_23081_begin_0 = const()[name = tensor("op_23081_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_23081_end_0 = const()[name = tensor("op_23081_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_23081_end_mask_0 = const()[name = tensor("op_23081_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23081_cast_fp16 = slice_by_index(begin = var_23081_begin_0, end = var_23081_end_0, end_mask = var_23081_end_mask_0, x = q_105_cast_fp16)[name = tensor("op_23081_cast_fp16")]; + tensor var_23085_begin_0 = const()[name = tensor("op_23085_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_23085_end_0 = const()[name = tensor("op_23085_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_23085_end_mask_0 = const()[name = tensor("op_23085_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23085_cast_fp16 = slice_by_index(begin = var_23085_begin_0, end = var_23085_end_0, end_mask = var_23085_end_mask_0, x = q_105_cast_fp16)[name = tensor("op_23085_cast_fp16")]; + tensor var_23089_begin_0 = const()[name = tensor("op_23089_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_23089_end_0 = const()[name = tensor("op_23089_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_23089_end_mask_0 = const()[name = tensor("op_23089_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23089_cast_fp16 = slice_by_index(begin = var_23089_begin_0, end = var_23089_end_0, end_mask = var_23089_end_mask_0, x = q_105_cast_fp16)[name = tensor("op_23089_cast_fp16")]; + tensor var_23093_begin_0 = const()[name = tensor("op_23093_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_23093_end_0 = const()[name = tensor("op_23093_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_23093_end_mask_0 = const()[name = tensor("op_23093_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23093_cast_fp16 = slice_by_index(begin = var_23093_begin_0, end = var_23093_end_0, end_mask = var_23093_end_mask_0, x = q_105_cast_fp16)[name = tensor("op_23093_cast_fp16")]; + tensor var_23097_begin_0 = const()[name = tensor("op_23097_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_23097_end_0 = const()[name = tensor("op_23097_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_23097_end_mask_0 = const()[name = tensor("op_23097_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23097_cast_fp16 = slice_by_index(begin = var_23097_begin_0, end = var_23097_end_0, end_mask = var_23097_end_mask_0, x = q_105_cast_fp16)[name = tensor("op_23097_cast_fp16")]; + tensor var_23101_begin_0 = const()[name = tensor("op_23101_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_23101_end_0 = const()[name = tensor("op_23101_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_23101_end_mask_0 = const()[name = tensor("op_23101_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23101_cast_fp16 = slice_by_index(begin = var_23101_begin_0, end = var_23101_end_0, end_mask = var_23101_end_mask_0, x = q_105_cast_fp16)[name = tensor("op_23101_cast_fp16")]; + tensor var_23105_begin_0 = const()[name = tensor("op_23105_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_23105_end_0 = const()[name = tensor("op_23105_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_23105_end_mask_0 = const()[name = tensor("op_23105_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23105_cast_fp16 = slice_by_index(begin = var_23105_begin_0, end = var_23105_end_0, end_mask = var_23105_end_mask_0, x = q_105_cast_fp16)[name = tensor("op_23105_cast_fp16")]; + tensor var_23109_begin_0 = const()[name = tensor("op_23109_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_23109_end_0 = const()[name = tensor("op_23109_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_23109_end_mask_0 = const()[name = tensor("op_23109_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23109_cast_fp16 = slice_by_index(begin = var_23109_begin_0, end = var_23109_end_0, end_mask = var_23109_end_mask_0, x = q_105_cast_fp16)[name = tensor("op_23109_cast_fp16")]; + tensor k_211_perm_0 = const()[name = tensor("k_211_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_23116_begin_0 = const()[name = tensor("op_23116_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23116_end_0 = const()[name = tensor("op_23116_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_23116_end_mask_0 = const()[name = tensor("op_23116_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_211_cast_fp16 = transpose(perm = k_211_perm_0, x = k_209_cast_fp16)[name = tensor("transpose_87")]; + tensor var_23116_cast_fp16 = slice_by_index(begin = var_23116_begin_0, end = var_23116_end_0, end_mask = var_23116_end_mask_0, x = k_211_cast_fp16)[name = tensor("op_23116_cast_fp16")]; + tensor var_23120_begin_0 = const()[name = tensor("op_23120_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_23120_end_0 = const()[name = tensor("op_23120_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_23120_end_mask_0 = const()[name = tensor("op_23120_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23120_cast_fp16 = slice_by_index(begin = var_23120_begin_0, end = var_23120_end_0, end_mask = var_23120_end_mask_0, x = k_211_cast_fp16)[name = tensor("op_23120_cast_fp16")]; + tensor var_23124_begin_0 = const()[name = tensor("op_23124_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_23124_end_0 = const()[name = tensor("op_23124_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_23124_end_mask_0 = const()[name = tensor("op_23124_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23124_cast_fp16 = slice_by_index(begin = var_23124_begin_0, end = var_23124_end_0, end_mask = var_23124_end_mask_0, x = k_211_cast_fp16)[name = tensor("op_23124_cast_fp16")]; + tensor var_23128_begin_0 = const()[name = tensor("op_23128_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_23128_end_0 = const()[name = tensor("op_23128_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_23128_end_mask_0 = const()[name = tensor("op_23128_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23128_cast_fp16 = slice_by_index(begin = var_23128_begin_0, end = var_23128_end_0, end_mask = var_23128_end_mask_0, x = k_211_cast_fp16)[name = tensor("op_23128_cast_fp16")]; + tensor var_23132_begin_0 = const()[name = tensor("op_23132_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_23132_end_0 = const()[name = tensor("op_23132_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_23132_end_mask_0 = const()[name = tensor("op_23132_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23132_cast_fp16 = slice_by_index(begin = var_23132_begin_0, end = var_23132_end_0, end_mask = var_23132_end_mask_0, x = k_211_cast_fp16)[name = tensor("op_23132_cast_fp16")]; + tensor var_23136_begin_0 = const()[name = tensor("op_23136_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_23136_end_0 = const()[name = tensor("op_23136_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_23136_end_mask_0 = const()[name = tensor("op_23136_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23136_cast_fp16 = slice_by_index(begin = var_23136_begin_0, end = var_23136_end_0, end_mask = var_23136_end_mask_0, x = k_211_cast_fp16)[name = tensor("op_23136_cast_fp16")]; + tensor var_23140_begin_0 = const()[name = tensor("op_23140_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_23140_end_0 = const()[name = tensor("op_23140_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_23140_end_mask_0 = const()[name = tensor("op_23140_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23140_cast_fp16 = slice_by_index(begin = var_23140_begin_0, end = var_23140_end_0, end_mask = var_23140_end_mask_0, x = k_211_cast_fp16)[name = tensor("op_23140_cast_fp16")]; + tensor var_23144_begin_0 = const()[name = tensor("op_23144_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_23144_end_0 = const()[name = tensor("op_23144_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_23144_end_mask_0 = const()[name = tensor("op_23144_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23144_cast_fp16 = slice_by_index(begin = var_23144_begin_0, end = var_23144_end_0, end_mask = var_23144_end_mask_0, x = k_211_cast_fp16)[name = tensor("op_23144_cast_fp16")]; + tensor var_23148_begin_0 = const()[name = tensor("op_23148_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_23148_end_0 = const()[name = tensor("op_23148_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_23148_end_mask_0 = const()[name = tensor("op_23148_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23148_cast_fp16 = slice_by_index(begin = var_23148_begin_0, end = var_23148_end_0, end_mask = var_23148_end_mask_0, x = k_211_cast_fp16)[name = tensor("op_23148_cast_fp16")]; + tensor var_23152_begin_0 = const()[name = tensor("op_23152_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_23152_end_0 = const()[name = tensor("op_23152_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_23152_end_mask_0 = const()[name = tensor("op_23152_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23152_cast_fp16 = slice_by_index(begin = var_23152_begin_0, end = var_23152_end_0, end_mask = var_23152_end_mask_0, x = k_211_cast_fp16)[name = tensor("op_23152_cast_fp16")]; + tensor var_23156_begin_0 = const()[name = tensor("op_23156_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_23156_end_0 = const()[name = tensor("op_23156_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_23156_end_mask_0 = const()[name = tensor("op_23156_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23156_cast_fp16 = slice_by_index(begin = var_23156_begin_0, end = var_23156_end_0, end_mask = var_23156_end_mask_0, x = k_211_cast_fp16)[name = tensor("op_23156_cast_fp16")]; + tensor var_23160_begin_0 = const()[name = tensor("op_23160_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_23160_end_0 = const()[name = tensor("op_23160_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_23160_end_mask_0 = const()[name = tensor("op_23160_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23160_cast_fp16 = slice_by_index(begin = var_23160_begin_0, end = var_23160_end_0, end_mask = var_23160_end_mask_0, x = k_211_cast_fp16)[name = tensor("op_23160_cast_fp16")]; + tensor var_23164_begin_0 = const()[name = tensor("op_23164_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_23164_end_0 = const()[name = tensor("op_23164_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_23164_end_mask_0 = const()[name = tensor("op_23164_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23164_cast_fp16 = slice_by_index(begin = var_23164_begin_0, end = var_23164_end_0, end_mask = var_23164_end_mask_0, x = k_211_cast_fp16)[name = tensor("op_23164_cast_fp16")]; + tensor var_23168_begin_0 = const()[name = tensor("op_23168_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_23168_end_0 = const()[name = tensor("op_23168_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_23168_end_mask_0 = const()[name = tensor("op_23168_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23168_cast_fp16 = slice_by_index(begin = var_23168_begin_0, end = var_23168_end_0, end_mask = var_23168_end_mask_0, x = k_211_cast_fp16)[name = tensor("op_23168_cast_fp16")]; + tensor var_23172_begin_0 = const()[name = tensor("op_23172_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_23172_end_0 = const()[name = tensor("op_23172_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_23172_end_mask_0 = const()[name = tensor("op_23172_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23172_cast_fp16 = slice_by_index(begin = var_23172_begin_0, end = var_23172_end_0, end_mask = var_23172_end_mask_0, x = k_211_cast_fp16)[name = tensor("op_23172_cast_fp16")]; + tensor var_23176_begin_0 = const()[name = tensor("op_23176_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_23176_end_0 = const()[name = tensor("op_23176_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_23176_end_mask_0 = const()[name = tensor("op_23176_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23176_cast_fp16 = slice_by_index(begin = var_23176_begin_0, end = var_23176_end_0, end_mask = var_23176_end_mask_0, x = k_211_cast_fp16)[name = tensor("op_23176_cast_fp16")]; + tensor var_23180_begin_0 = const()[name = tensor("op_23180_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_23180_end_0 = const()[name = tensor("op_23180_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_23180_end_mask_0 = const()[name = tensor("op_23180_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23180_cast_fp16 = slice_by_index(begin = var_23180_begin_0, end = var_23180_end_0, end_mask = var_23180_end_mask_0, x = k_211_cast_fp16)[name = tensor("op_23180_cast_fp16")]; + tensor var_23184_begin_0 = const()[name = tensor("op_23184_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_23184_end_0 = const()[name = tensor("op_23184_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_23184_end_mask_0 = const()[name = tensor("op_23184_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23184_cast_fp16 = slice_by_index(begin = var_23184_begin_0, end = var_23184_end_0, end_mask = var_23184_end_mask_0, x = k_211_cast_fp16)[name = tensor("op_23184_cast_fp16")]; + tensor var_23188_begin_0 = const()[name = tensor("op_23188_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_23188_end_0 = const()[name = tensor("op_23188_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_23188_end_mask_0 = const()[name = tensor("op_23188_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23188_cast_fp16 = slice_by_index(begin = var_23188_begin_0, end = var_23188_end_0, end_mask = var_23188_end_mask_0, x = k_211_cast_fp16)[name = tensor("op_23188_cast_fp16")]; + tensor var_23192_begin_0 = const()[name = tensor("op_23192_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_23192_end_0 = const()[name = tensor("op_23192_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_23192_end_mask_0 = const()[name = tensor("op_23192_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23192_cast_fp16 = slice_by_index(begin = var_23192_begin_0, end = var_23192_end_0, end_mask = var_23192_end_mask_0, x = k_211_cast_fp16)[name = tensor("op_23192_cast_fp16")]; + tensor var_23194_begin_0 = const()[name = tensor("op_23194_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23194_end_0 = const()[name = tensor("op_23194_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_23194_end_mask_0 = const()[name = tensor("op_23194_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23194_cast_fp16 = slice_by_index(begin = var_23194_begin_0, end = var_23194_end_0, end_mask = var_23194_end_mask_0, x = v_105_cast_fp16)[name = tensor("op_23194_cast_fp16")]; + tensor var_23198_begin_0 = const()[name = tensor("op_23198_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_23198_end_0 = const()[name = tensor("op_23198_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_23198_end_mask_0 = const()[name = tensor("op_23198_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23198_cast_fp16 = slice_by_index(begin = var_23198_begin_0, end = var_23198_end_0, end_mask = var_23198_end_mask_0, x = v_105_cast_fp16)[name = tensor("op_23198_cast_fp16")]; + tensor var_23202_begin_0 = const()[name = tensor("op_23202_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_23202_end_0 = const()[name = tensor("op_23202_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_23202_end_mask_0 = const()[name = tensor("op_23202_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23202_cast_fp16 = slice_by_index(begin = var_23202_begin_0, end = var_23202_end_0, end_mask = var_23202_end_mask_0, x = v_105_cast_fp16)[name = tensor("op_23202_cast_fp16")]; + tensor var_23206_begin_0 = const()[name = tensor("op_23206_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_23206_end_0 = const()[name = tensor("op_23206_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_23206_end_mask_0 = const()[name = tensor("op_23206_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23206_cast_fp16 = slice_by_index(begin = var_23206_begin_0, end = var_23206_end_0, end_mask = var_23206_end_mask_0, x = v_105_cast_fp16)[name = tensor("op_23206_cast_fp16")]; + tensor var_23210_begin_0 = const()[name = tensor("op_23210_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_23210_end_0 = const()[name = tensor("op_23210_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_23210_end_mask_0 = const()[name = tensor("op_23210_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23210_cast_fp16 = slice_by_index(begin = var_23210_begin_0, end = var_23210_end_0, end_mask = var_23210_end_mask_0, x = v_105_cast_fp16)[name = tensor("op_23210_cast_fp16")]; + tensor var_23214_begin_0 = const()[name = tensor("op_23214_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_23214_end_0 = const()[name = tensor("op_23214_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_23214_end_mask_0 = const()[name = tensor("op_23214_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23214_cast_fp16 = slice_by_index(begin = var_23214_begin_0, end = var_23214_end_0, end_mask = var_23214_end_mask_0, x = v_105_cast_fp16)[name = tensor("op_23214_cast_fp16")]; + tensor var_23218_begin_0 = const()[name = tensor("op_23218_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_23218_end_0 = const()[name = tensor("op_23218_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_23218_end_mask_0 = const()[name = tensor("op_23218_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23218_cast_fp16 = slice_by_index(begin = var_23218_begin_0, end = var_23218_end_0, end_mask = var_23218_end_mask_0, x = v_105_cast_fp16)[name = tensor("op_23218_cast_fp16")]; + tensor var_23222_begin_0 = const()[name = tensor("op_23222_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_23222_end_0 = const()[name = tensor("op_23222_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_23222_end_mask_0 = const()[name = tensor("op_23222_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23222_cast_fp16 = slice_by_index(begin = var_23222_begin_0, end = var_23222_end_0, end_mask = var_23222_end_mask_0, x = v_105_cast_fp16)[name = tensor("op_23222_cast_fp16")]; + tensor var_23226_begin_0 = const()[name = tensor("op_23226_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_23226_end_0 = const()[name = tensor("op_23226_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_23226_end_mask_0 = const()[name = tensor("op_23226_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23226_cast_fp16 = slice_by_index(begin = var_23226_begin_0, end = var_23226_end_0, end_mask = var_23226_end_mask_0, x = v_105_cast_fp16)[name = tensor("op_23226_cast_fp16")]; + tensor var_23230_begin_0 = const()[name = tensor("op_23230_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_23230_end_0 = const()[name = tensor("op_23230_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_23230_end_mask_0 = const()[name = tensor("op_23230_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23230_cast_fp16 = slice_by_index(begin = var_23230_begin_0, end = var_23230_end_0, end_mask = var_23230_end_mask_0, x = v_105_cast_fp16)[name = tensor("op_23230_cast_fp16")]; + tensor var_23234_begin_0 = const()[name = tensor("op_23234_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_23234_end_0 = const()[name = tensor("op_23234_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_23234_end_mask_0 = const()[name = tensor("op_23234_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23234_cast_fp16 = slice_by_index(begin = var_23234_begin_0, end = var_23234_end_0, end_mask = var_23234_end_mask_0, x = v_105_cast_fp16)[name = tensor("op_23234_cast_fp16")]; + tensor var_23238_begin_0 = const()[name = tensor("op_23238_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_23238_end_0 = const()[name = tensor("op_23238_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_23238_end_mask_0 = const()[name = tensor("op_23238_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23238_cast_fp16 = slice_by_index(begin = var_23238_begin_0, end = var_23238_end_0, end_mask = var_23238_end_mask_0, x = v_105_cast_fp16)[name = tensor("op_23238_cast_fp16")]; + tensor var_23242_begin_0 = const()[name = tensor("op_23242_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_23242_end_0 = const()[name = tensor("op_23242_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_23242_end_mask_0 = const()[name = tensor("op_23242_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23242_cast_fp16 = slice_by_index(begin = var_23242_begin_0, end = var_23242_end_0, end_mask = var_23242_end_mask_0, x = v_105_cast_fp16)[name = tensor("op_23242_cast_fp16")]; + tensor var_23246_begin_0 = const()[name = tensor("op_23246_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_23246_end_0 = const()[name = tensor("op_23246_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_23246_end_mask_0 = const()[name = tensor("op_23246_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23246_cast_fp16 = slice_by_index(begin = var_23246_begin_0, end = var_23246_end_0, end_mask = var_23246_end_mask_0, x = v_105_cast_fp16)[name = tensor("op_23246_cast_fp16")]; + tensor var_23250_begin_0 = const()[name = tensor("op_23250_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_23250_end_0 = const()[name = tensor("op_23250_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_23250_end_mask_0 = const()[name = tensor("op_23250_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23250_cast_fp16 = slice_by_index(begin = var_23250_begin_0, end = var_23250_end_0, end_mask = var_23250_end_mask_0, x = v_105_cast_fp16)[name = tensor("op_23250_cast_fp16")]; + tensor var_23254_begin_0 = const()[name = tensor("op_23254_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_23254_end_0 = const()[name = tensor("op_23254_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_23254_end_mask_0 = const()[name = tensor("op_23254_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23254_cast_fp16 = slice_by_index(begin = var_23254_begin_0, end = var_23254_end_0, end_mask = var_23254_end_mask_0, x = v_105_cast_fp16)[name = tensor("op_23254_cast_fp16")]; + tensor var_23258_begin_0 = const()[name = tensor("op_23258_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_23258_end_0 = const()[name = tensor("op_23258_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_23258_end_mask_0 = const()[name = tensor("op_23258_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23258_cast_fp16 = slice_by_index(begin = var_23258_begin_0, end = var_23258_end_0, end_mask = var_23258_end_mask_0, x = v_105_cast_fp16)[name = tensor("op_23258_cast_fp16")]; + tensor var_23262_begin_0 = const()[name = tensor("op_23262_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_23262_end_0 = const()[name = tensor("op_23262_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_23262_end_mask_0 = const()[name = tensor("op_23262_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23262_cast_fp16 = slice_by_index(begin = var_23262_begin_0, end = var_23262_end_0, end_mask = var_23262_end_mask_0, x = v_105_cast_fp16)[name = tensor("op_23262_cast_fp16")]; + tensor var_23266_begin_0 = const()[name = tensor("op_23266_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_23266_end_0 = const()[name = tensor("op_23266_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_23266_end_mask_0 = const()[name = tensor("op_23266_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23266_cast_fp16 = slice_by_index(begin = var_23266_begin_0, end = var_23266_end_0, end_mask = var_23266_end_mask_0, x = v_105_cast_fp16)[name = tensor("op_23266_cast_fp16")]; + tensor var_23270_begin_0 = const()[name = tensor("op_23270_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_23270_end_0 = const()[name = tensor("op_23270_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_23270_end_mask_0 = const()[name = tensor("op_23270_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23270_cast_fp16 = slice_by_index(begin = var_23270_begin_0, end = var_23270_end_0, end_mask = var_23270_end_mask_0, x = v_105_cast_fp16)[name = tensor("op_23270_cast_fp16")]; + tensor var_23274_equation_0 = const()[name = tensor("op_23274_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23274_cast_fp16 = einsum(equation = var_23274_equation_0, values = (var_23116_cast_fp16, var_23033_cast_fp16))[name = tensor("op_23274_cast_fp16")]; + tensor var_23275_to_fp16 = const()[name = tensor("op_23275_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1921_cast_fp16 = mul(x = var_23274_cast_fp16, y = var_23275_to_fp16)[name = tensor("aw_1921_cast_fp16")]; + tensor var_23278_equation_0 = const()[name = tensor("op_23278_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23278_cast_fp16 = einsum(equation = var_23278_equation_0, values = (var_23120_cast_fp16, var_23037_cast_fp16))[name = tensor("op_23278_cast_fp16")]; + tensor var_23279_to_fp16 = const()[name = tensor("op_23279_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1923_cast_fp16 = mul(x = var_23278_cast_fp16, y = var_23279_to_fp16)[name = tensor("aw_1923_cast_fp16")]; + tensor var_23282_equation_0 = const()[name = tensor("op_23282_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23282_cast_fp16 = einsum(equation = var_23282_equation_0, values = (var_23124_cast_fp16, var_23041_cast_fp16))[name = tensor("op_23282_cast_fp16")]; + tensor var_23283_to_fp16 = const()[name = tensor("op_23283_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1925_cast_fp16 = mul(x = var_23282_cast_fp16, y = var_23283_to_fp16)[name = tensor("aw_1925_cast_fp16")]; + tensor var_23286_equation_0 = const()[name = tensor("op_23286_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23286_cast_fp16 = einsum(equation = var_23286_equation_0, values = (var_23128_cast_fp16, var_23045_cast_fp16))[name = tensor("op_23286_cast_fp16")]; + tensor var_23287_to_fp16 = const()[name = tensor("op_23287_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1927_cast_fp16 = mul(x = var_23286_cast_fp16, y = var_23287_to_fp16)[name = tensor("aw_1927_cast_fp16")]; + tensor var_23290_equation_0 = const()[name = tensor("op_23290_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23290_cast_fp16 = einsum(equation = var_23290_equation_0, values = (var_23132_cast_fp16, var_23049_cast_fp16))[name = tensor("op_23290_cast_fp16")]; + tensor var_23291_to_fp16 = const()[name = tensor("op_23291_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1929_cast_fp16 = mul(x = var_23290_cast_fp16, y = var_23291_to_fp16)[name = tensor("aw_1929_cast_fp16")]; + tensor var_23294_equation_0 = const()[name = tensor("op_23294_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23294_cast_fp16 = einsum(equation = var_23294_equation_0, values = (var_23136_cast_fp16, var_23053_cast_fp16))[name = tensor("op_23294_cast_fp16")]; + tensor var_23295_to_fp16 = const()[name = tensor("op_23295_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1931_cast_fp16 = mul(x = var_23294_cast_fp16, y = var_23295_to_fp16)[name = tensor("aw_1931_cast_fp16")]; + tensor var_23298_equation_0 = const()[name = tensor("op_23298_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23298_cast_fp16 = einsum(equation = var_23298_equation_0, values = (var_23140_cast_fp16, var_23057_cast_fp16))[name = tensor("op_23298_cast_fp16")]; + tensor var_23299_to_fp16 = const()[name = tensor("op_23299_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1933_cast_fp16 = mul(x = var_23298_cast_fp16, y = var_23299_to_fp16)[name = tensor("aw_1933_cast_fp16")]; + tensor var_23302_equation_0 = const()[name = tensor("op_23302_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23302_cast_fp16 = einsum(equation = var_23302_equation_0, values = (var_23144_cast_fp16, var_23061_cast_fp16))[name = tensor("op_23302_cast_fp16")]; + tensor var_23303_to_fp16 = const()[name = tensor("op_23303_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1935_cast_fp16 = mul(x = var_23302_cast_fp16, y = var_23303_to_fp16)[name = tensor("aw_1935_cast_fp16")]; + tensor var_23306_equation_0 = const()[name = tensor("op_23306_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23306_cast_fp16 = einsum(equation = var_23306_equation_0, values = (var_23148_cast_fp16, var_23065_cast_fp16))[name = tensor("op_23306_cast_fp16")]; + tensor var_23307_to_fp16 = const()[name = tensor("op_23307_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1937_cast_fp16 = mul(x = var_23306_cast_fp16, y = var_23307_to_fp16)[name = tensor("aw_1937_cast_fp16")]; + tensor var_23310_equation_0 = const()[name = tensor("op_23310_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23310_cast_fp16 = einsum(equation = var_23310_equation_0, values = (var_23152_cast_fp16, var_23069_cast_fp16))[name = tensor("op_23310_cast_fp16")]; + tensor var_23311_to_fp16 = const()[name = tensor("op_23311_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1939_cast_fp16 = mul(x = var_23310_cast_fp16, y = var_23311_to_fp16)[name = tensor("aw_1939_cast_fp16")]; + tensor var_23314_equation_0 = const()[name = tensor("op_23314_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23314_cast_fp16 = einsum(equation = var_23314_equation_0, values = (var_23156_cast_fp16, var_23073_cast_fp16))[name = tensor("op_23314_cast_fp16")]; + tensor var_23315_to_fp16 = const()[name = tensor("op_23315_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1941_cast_fp16 = mul(x = var_23314_cast_fp16, y = var_23315_to_fp16)[name = tensor("aw_1941_cast_fp16")]; + tensor var_23318_equation_0 = const()[name = tensor("op_23318_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23318_cast_fp16 = einsum(equation = var_23318_equation_0, values = (var_23160_cast_fp16, var_23077_cast_fp16))[name = tensor("op_23318_cast_fp16")]; + tensor var_23319_to_fp16 = const()[name = tensor("op_23319_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1943_cast_fp16 = mul(x = var_23318_cast_fp16, y = var_23319_to_fp16)[name = tensor("aw_1943_cast_fp16")]; + tensor var_23322_equation_0 = const()[name = tensor("op_23322_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23322_cast_fp16 = einsum(equation = var_23322_equation_0, values = (var_23164_cast_fp16, var_23081_cast_fp16))[name = tensor("op_23322_cast_fp16")]; + tensor var_23323_to_fp16 = const()[name = tensor("op_23323_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1945_cast_fp16 = mul(x = var_23322_cast_fp16, y = var_23323_to_fp16)[name = tensor("aw_1945_cast_fp16")]; + tensor var_23326_equation_0 = const()[name = tensor("op_23326_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23326_cast_fp16 = einsum(equation = var_23326_equation_0, values = (var_23168_cast_fp16, var_23085_cast_fp16))[name = tensor("op_23326_cast_fp16")]; + tensor var_23327_to_fp16 = const()[name = tensor("op_23327_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1947_cast_fp16 = mul(x = var_23326_cast_fp16, y = var_23327_to_fp16)[name = tensor("aw_1947_cast_fp16")]; + tensor var_23330_equation_0 = const()[name = tensor("op_23330_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23330_cast_fp16 = einsum(equation = var_23330_equation_0, values = (var_23172_cast_fp16, var_23089_cast_fp16))[name = tensor("op_23330_cast_fp16")]; + tensor var_23331_to_fp16 = const()[name = tensor("op_23331_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1949_cast_fp16 = mul(x = var_23330_cast_fp16, y = var_23331_to_fp16)[name = tensor("aw_1949_cast_fp16")]; + tensor var_23334_equation_0 = const()[name = tensor("op_23334_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23334_cast_fp16 = einsum(equation = var_23334_equation_0, values = (var_23176_cast_fp16, var_23093_cast_fp16))[name = tensor("op_23334_cast_fp16")]; + tensor var_23335_to_fp16 = const()[name = tensor("op_23335_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1951_cast_fp16 = mul(x = var_23334_cast_fp16, y = var_23335_to_fp16)[name = tensor("aw_1951_cast_fp16")]; + tensor var_23338_equation_0 = const()[name = tensor("op_23338_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23338_cast_fp16 = einsum(equation = var_23338_equation_0, values = (var_23180_cast_fp16, var_23097_cast_fp16))[name = tensor("op_23338_cast_fp16")]; + tensor var_23339_to_fp16 = const()[name = tensor("op_23339_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1953_cast_fp16 = mul(x = var_23338_cast_fp16, y = var_23339_to_fp16)[name = tensor("aw_1953_cast_fp16")]; + tensor var_23342_equation_0 = const()[name = tensor("op_23342_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23342_cast_fp16 = einsum(equation = var_23342_equation_0, values = (var_23184_cast_fp16, var_23101_cast_fp16))[name = tensor("op_23342_cast_fp16")]; + tensor var_23343_to_fp16 = const()[name = tensor("op_23343_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1955_cast_fp16 = mul(x = var_23342_cast_fp16, y = var_23343_to_fp16)[name = tensor("aw_1955_cast_fp16")]; + tensor var_23346_equation_0 = const()[name = tensor("op_23346_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23346_cast_fp16 = einsum(equation = var_23346_equation_0, values = (var_23188_cast_fp16, var_23105_cast_fp16))[name = tensor("op_23346_cast_fp16")]; + tensor var_23347_to_fp16 = const()[name = tensor("op_23347_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1957_cast_fp16 = mul(x = var_23346_cast_fp16, y = var_23347_to_fp16)[name = tensor("aw_1957_cast_fp16")]; + tensor var_23350_equation_0 = const()[name = tensor("op_23350_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23350_cast_fp16 = einsum(equation = var_23350_equation_0, values = (var_23192_cast_fp16, var_23109_cast_fp16))[name = tensor("op_23350_cast_fp16")]; + tensor var_23351_to_fp16 = const()[name = tensor("op_23351_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1959_cast_fp16 = mul(x = var_23350_cast_fp16, y = var_23351_to_fp16)[name = tensor("aw_1959_cast_fp16")]; + tensor var_23353_cast_fp16 = softmax(axis = var_21077, x = aw_1921_cast_fp16)[name = tensor("op_23353_cast_fp16")]; + tensor var_23354_cast_fp16 = softmax(axis = var_21077, x = aw_1923_cast_fp16)[name = tensor("op_23354_cast_fp16")]; + tensor var_23355_cast_fp16 = softmax(axis = var_21077, x = aw_1925_cast_fp16)[name = tensor("op_23355_cast_fp16")]; + tensor var_23356_cast_fp16 = softmax(axis = var_21077, x = aw_1927_cast_fp16)[name = tensor("op_23356_cast_fp16")]; + tensor var_23357_cast_fp16 = softmax(axis = var_21077, x = aw_1929_cast_fp16)[name = tensor("op_23357_cast_fp16")]; + tensor var_23358_cast_fp16 = softmax(axis = var_21077, x = aw_1931_cast_fp16)[name = tensor("op_23358_cast_fp16")]; + tensor var_23359_cast_fp16 = softmax(axis = var_21077, x = aw_1933_cast_fp16)[name = tensor("op_23359_cast_fp16")]; + tensor var_23360_cast_fp16 = softmax(axis = var_21077, x = aw_1935_cast_fp16)[name = tensor("op_23360_cast_fp16")]; + tensor var_23361_cast_fp16 = softmax(axis = var_21077, x = aw_1937_cast_fp16)[name = tensor("op_23361_cast_fp16")]; + tensor var_23362_cast_fp16 = softmax(axis = var_21077, x = aw_1939_cast_fp16)[name = tensor("op_23362_cast_fp16")]; + tensor var_23363_cast_fp16 = softmax(axis = var_21077, x = aw_1941_cast_fp16)[name = tensor("op_23363_cast_fp16")]; + tensor var_23364_cast_fp16 = softmax(axis = var_21077, x = aw_1943_cast_fp16)[name = tensor("op_23364_cast_fp16")]; + tensor var_23365_cast_fp16 = softmax(axis = var_21077, x = aw_1945_cast_fp16)[name = tensor("op_23365_cast_fp16")]; + tensor var_23366_cast_fp16 = softmax(axis = var_21077, x = aw_1947_cast_fp16)[name = tensor("op_23366_cast_fp16")]; + tensor var_23367_cast_fp16 = softmax(axis = var_21077, x = aw_1949_cast_fp16)[name = tensor("op_23367_cast_fp16")]; + tensor var_23368_cast_fp16 = softmax(axis = var_21077, x = aw_1951_cast_fp16)[name = tensor("op_23368_cast_fp16")]; + tensor var_23369_cast_fp16 = softmax(axis = var_21077, x = aw_1953_cast_fp16)[name = tensor("op_23369_cast_fp16")]; + tensor var_23370_cast_fp16 = softmax(axis = var_21077, x = aw_1955_cast_fp16)[name = tensor("op_23370_cast_fp16")]; + tensor var_23371_cast_fp16 = softmax(axis = var_21077, x = aw_1957_cast_fp16)[name = tensor("op_23371_cast_fp16")]; + tensor var_23372_cast_fp16 = softmax(axis = var_21077, x = aw_1959_cast_fp16)[name = tensor("op_23372_cast_fp16")]; + tensor var_23374_equation_0 = const()[name = tensor("op_23374_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23374_cast_fp16 = einsum(equation = var_23374_equation_0, values = (var_23194_cast_fp16, var_23353_cast_fp16))[name = tensor("op_23374_cast_fp16")]; + tensor var_23376_equation_0 = const()[name = tensor("op_23376_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23376_cast_fp16 = einsum(equation = var_23376_equation_0, values = (var_23198_cast_fp16, var_23354_cast_fp16))[name = tensor("op_23376_cast_fp16")]; + tensor var_23378_equation_0 = const()[name = tensor("op_23378_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23378_cast_fp16 = einsum(equation = var_23378_equation_0, values = (var_23202_cast_fp16, var_23355_cast_fp16))[name = tensor("op_23378_cast_fp16")]; + tensor var_23380_equation_0 = const()[name = tensor("op_23380_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23380_cast_fp16 = einsum(equation = var_23380_equation_0, values = (var_23206_cast_fp16, var_23356_cast_fp16))[name = tensor("op_23380_cast_fp16")]; + tensor var_23382_equation_0 = const()[name = tensor("op_23382_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23382_cast_fp16 = einsum(equation = var_23382_equation_0, values = (var_23210_cast_fp16, var_23357_cast_fp16))[name = tensor("op_23382_cast_fp16")]; + tensor var_23384_equation_0 = const()[name = tensor("op_23384_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23384_cast_fp16 = einsum(equation = var_23384_equation_0, values = (var_23214_cast_fp16, var_23358_cast_fp16))[name = tensor("op_23384_cast_fp16")]; + tensor var_23386_equation_0 = const()[name = tensor("op_23386_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23386_cast_fp16 = einsum(equation = var_23386_equation_0, values = (var_23218_cast_fp16, var_23359_cast_fp16))[name = tensor("op_23386_cast_fp16")]; + tensor var_23388_equation_0 = const()[name = tensor("op_23388_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23388_cast_fp16 = einsum(equation = var_23388_equation_0, values = (var_23222_cast_fp16, var_23360_cast_fp16))[name = tensor("op_23388_cast_fp16")]; + tensor var_23390_equation_0 = const()[name = tensor("op_23390_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23390_cast_fp16 = einsum(equation = var_23390_equation_0, values = (var_23226_cast_fp16, var_23361_cast_fp16))[name = tensor("op_23390_cast_fp16")]; + tensor var_23392_equation_0 = const()[name = tensor("op_23392_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23392_cast_fp16 = einsum(equation = var_23392_equation_0, values = (var_23230_cast_fp16, var_23362_cast_fp16))[name = tensor("op_23392_cast_fp16")]; + tensor var_23394_equation_0 = const()[name = tensor("op_23394_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23394_cast_fp16 = einsum(equation = var_23394_equation_0, values = (var_23234_cast_fp16, var_23363_cast_fp16))[name = tensor("op_23394_cast_fp16")]; + tensor var_23396_equation_0 = const()[name = tensor("op_23396_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23396_cast_fp16 = einsum(equation = var_23396_equation_0, values = (var_23238_cast_fp16, var_23364_cast_fp16))[name = tensor("op_23396_cast_fp16")]; + tensor var_23398_equation_0 = const()[name = tensor("op_23398_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23398_cast_fp16 = einsum(equation = var_23398_equation_0, values = (var_23242_cast_fp16, var_23365_cast_fp16))[name = tensor("op_23398_cast_fp16")]; + tensor var_23400_equation_0 = const()[name = tensor("op_23400_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23400_cast_fp16 = einsum(equation = var_23400_equation_0, values = (var_23246_cast_fp16, var_23366_cast_fp16))[name = tensor("op_23400_cast_fp16")]; + tensor var_23402_equation_0 = const()[name = tensor("op_23402_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23402_cast_fp16 = einsum(equation = var_23402_equation_0, values = (var_23250_cast_fp16, var_23367_cast_fp16))[name = tensor("op_23402_cast_fp16")]; + tensor var_23404_equation_0 = const()[name = tensor("op_23404_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23404_cast_fp16 = einsum(equation = var_23404_equation_0, values = (var_23254_cast_fp16, var_23368_cast_fp16))[name = tensor("op_23404_cast_fp16")]; + tensor var_23406_equation_0 = const()[name = tensor("op_23406_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23406_cast_fp16 = einsum(equation = var_23406_equation_0, values = (var_23258_cast_fp16, var_23369_cast_fp16))[name = tensor("op_23406_cast_fp16")]; + tensor var_23408_equation_0 = const()[name = tensor("op_23408_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23408_cast_fp16 = einsum(equation = var_23408_equation_0, values = (var_23262_cast_fp16, var_23370_cast_fp16))[name = tensor("op_23408_cast_fp16")]; + tensor var_23410_equation_0 = const()[name = tensor("op_23410_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23410_cast_fp16 = einsum(equation = var_23410_equation_0, values = (var_23266_cast_fp16, var_23371_cast_fp16))[name = tensor("op_23410_cast_fp16")]; + tensor var_23412_equation_0 = const()[name = tensor("op_23412_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23412_cast_fp16 = einsum(equation = var_23412_equation_0, values = (var_23270_cast_fp16, var_23372_cast_fp16))[name = tensor("op_23412_cast_fp16")]; + tensor input_343_interleave_0 = const()[name = tensor("input_343_interleave_0"), val = tensor(false)]; + tensor input_343_cast_fp16 = concat(axis = var_21077, interleave = input_343_interleave_0, values = (var_23374_cast_fp16, var_23376_cast_fp16, var_23378_cast_fp16, var_23380_cast_fp16, var_23382_cast_fp16, var_23384_cast_fp16, var_23386_cast_fp16, var_23388_cast_fp16, var_23390_cast_fp16, var_23392_cast_fp16, var_23394_cast_fp16, var_23396_cast_fp16, var_23398_cast_fp16, var_23400_cast_fp16, var_23402_cast_fp16, var_23404_cast_fp16, var_23406_cast_fp16, var_23408_cast_fp16, var_23410_cast_fp16, var_23412_cast_fp16))[name = tensor("input_343_cast_fp16")]; + tensor var_23418 = const()[name = tensor("op_23418"), val = tensor([1, 1])]; + tensor var_23420 = const()[name = tensor("op_23420"), val = tensor([1, 1])]; + tensor var_23422_pad_type_0 = const()[name = tensor("op_23422_pad_type_0"), val = tensor("custom")]; + tensor var_23422_pad_0 = const()[name = tensor("op_23422_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_2_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_2_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1874970816)))]; + tensor mid_block_attentions_0_transformer_blocks_2_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_2_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1878247680)))]; + tensor var_23422_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_2_attn1_to_out_0_bias_to_fp16, dilations = var_23420, groups = var_21077, pad = var_23422_pad_0, pad_type = var_23422_pad_type_0, strides = var_23418, weight = mid_block_attentions_0_transformer_blocks_2_attn1_to_out_0_weight_to_fp16, x = input_343_cast_fp16)[name = tensor("op_23422_cast_fp16")]; + tensor inputs_159_cast_fp16 = add(x = var_23422_cast_fp16, y = inputs_157_cast_fp16)[name = tensor("inputs_159_cast_fp16")]; + tensor hidden_states_223_axes_0 = const()[name = tensor("hidden_states_223_axes_0"), val = tensor([1])]; + tensor hidden_states_223_gamma_0_to_fp16 = const()[name = tensor("hidden_states_223_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1878250304)))]; + tensor hidden_states_223_beta_0_to_fp16 = const()[name = tensor("hidden_states_223_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1878252928)))]; + tensor var_23432_to_fp16 = const()[name = tensor("op_23432_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_223_cast_fp16 = layer_norm(axes = hidden_states_223_axes_0, beta = hidden_states_223_beta_0_to_fp16, epsilon = var_23432_to_fp16, gamma = hidden_states_223_gamma_0_to_fp16, x = inputs_159_cast_fp16)[name = tensor("hidden_states_223_cast_fp16")]; + tensor var_23447 = const()[name = tensor("op_23447"), val = tensor([1, 1])]; + tensor var_23449 = const()[name = tensor("op_23449"), val = tensor([1, 1])]; + tensor q_107_pad_type_0 = const()[name = tensor("q_107_pad_type_0"), val = tensor("custom")]; + tensor q_107_pad_0 = const()[name = tensor("q_107_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_2_attn2_to_q_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_2_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1878255552)))]; + tensor q_107_cast_fp16 = conv(dilations = var_23449, groups = var_21077, pad = q_107_pad_0, pad_type = q_107_pad_type_0, strides = var_23447, weight = mid_block_attentions_0_transformer_blocks_2_attn2_to_q_weight_to_fp16, x = hidden_states_223_cast_fp16)[name = tensor("q_107_cast_fp16")]; + tensor var_23453 = const()[name = tensor("op_23453"), val = tensor([1, 1])]; + tensor var_23455 = const()[name = tensor("op_23455"), val = tensor([1, 1])]; + tensor k_213_pad_type_0 = const()[name = tensor("k_213_pad_type_0"), val = tensor("custom")]; + tensor k_213_pad_0 = const()[name = tensor("k_213_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_2_attn2_to_k_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_2_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1881532416)))]; + tensor k_213_cast_fp16 = conv(dilations = var_23455, groups = var_21077, pad = k_213_pad_0, pad_type = k_213_pad_type_0, strides = var_23453, weight = mid_block_attentions_0_transformer_blocks_2_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_213_cast_fp16")]; + tensor var_23459 = const()[name = tensor("op_23459"), val = tensor([1, 1])]; + tensor var_23461 = const()[name = tensor("op_23461"), val = tensor([1, 1])]; + tensor v_107_pad_type_0 = const()[name = tensor("v_107_pad_type_0"), val = tensor("custom")]; + tensor v_107_pad_0 = const()[name = tensor("v_107_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_2_attn2_to_v_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_2_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1886775360)))]; + tensor v_107_cast_fp16 = conv(dilations = var_23461, groups = var_21077, pad = v_107_pad_0, pad_type = v_107_pad_type_0, strides = var_23459, weight = mid_block_attentions_0_transformer_blocks_2_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_107_cast_fp16")]; + tensor var_23465_begin_0 = const()[name = tensor("op_23465_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23465_end_0 = const()[name = tensor("op_23465_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_23465_end_mask_0 = const()[name = tensor("op_23465_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23465_cast_fp16 = slice_by_index(begin = var_23465_begin_0, end = var_23465_end_0, end_mask = var_23465_end_mask_0, x = q_107_cast_fp16)[name = tensor("op_23465_cast_fp16")]; + tensor var_23469_begin_0 = const()[name = tensor("op_23469_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_23469_end_0 = const()[name = tensor("op_23469_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_23469_end_mask_0 = const()[name = tensor("op_23469_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23469_cast_fp16 = slice_by_index(begin = var_23469_begin_0, end = var_23469_end_0, end_mask = var_23469_end_mask_0, x = q_107_cast_fp16)[name = tensor("op_23469_cast_fp16")]; + tensor var_23473_begin_0 = const()[name = tensor("op_23473_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_23473_end_0 = const()[name = tensor("op_23473_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_23473_end_mask_0 = const()[name = tensor("op_23473_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23473_cast_fp16 = slice_by_index(begin = var_23473_begin_0, end = var_23473_end_0, end_mask = var_23473_end_mask_0, x = q_107_cast_fp16)[name = tensor("op_23473_cast_fp16")]; + tensor var_23477_begin_0 = const()[name = tensor("op_23477_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_23477_end_0 = const()[name = tensor("op_23477_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_23477_end_mask_0 = const()[name = tensor("op_23477_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23477_cast_fp16 = slice_by_index(begin = var_23477_begin_0, end = var_23477_end_0, end_mask = var_23477_end_mask_0, x = q_107_cast_fp16)[name = tensor("op_23477_cast_fp16")]; + tensor var_23481_begin_0 = const()[name = tensor("op_23481_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_23481_end_0 = const()[name = tensor("op_23481_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_23481_end_mask_0 = const()[name = tensor("op_23481_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23481_cast_fp16 = slice_by_index(begin = var_23481_begin_0, end = var_23481_end_0, end_mask = var_23481_end_mask_0, x = q_107_cast_fp16)[name = tensor("op_23481_cast_fp16")]; + tensor var_23485_begin_0 = const()[name = tensor("op_23485_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_23485_end_0 = const()[name = tensor("op_23485_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_23485_end_mask_0 = const()[name = tensor("op_23485_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23485_cast_fp16 = slice_by_index(begin = var_23485_begin_0, end = var_23485_end_0, end_mask = var_23485_end_mask_0, x = q_107_cast_fp16)[name = tensor("op_23485_cast_fp16")]; + tensor var_23489_begin_0 = const()[name = tensor("op_23489_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_23489_end_0 = const()[name = tensor("op_23489_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_23489_end_mask_0 = const()[name = tensor("op_23489_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23489_cast_fp16 = slice_by_index(begin = var_23489_begin_0, end = var_23489_end_0, end_mask = var_23489_end_mask_0, x = q_107_cast_fp16)[name = tensor("op_23489_cast_fp16")]; + tensor var_23493_begin_0 = const()[name = tensor("op_23493_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_23493_end_0 = const()[name = tensor("op_23493_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_23493_end_mask_0 = const()[name = tensor("op_23493_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23493_cast_fp16 = slice_by_index(begin = var_23493_begin_0, end = var_23493_end_0, end_mask = var_23493_end_mask_0, x = q_107_cast_fp16)[name = tensor("op_23493_cast_fp16")]; + tensor var_23497_begin_0 = const()[name = tensor("op_23497_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_23497_end_0 = const()[name = tensor("op_23497_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_23497_end_mask_0 = const()[name = tensor("op_23497_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23497_cast_fp16 = slice_by_index(begin = var_23497_begin_0, end = var_23497_end_0, end_mask = var_23497_end_mask_0, x = q_107_cast_fp16)[name = tensor("op_23497_cast_fp16")]; + tensor var_23501_begin_0 = const()[name = tensor("op_23501_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_23501_end_0 = const()[name = tensor("op_23501_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_23501_end_mask_0 = const()[name = tensor("op_23501_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23501_cast_fp16 = slice_by_index(begin = var_23501_begin_0, end = var_23501_end_0, end_mask = var_23501_end_mask_0, x = q_107_cast_fp16)[name = tensor("op_23501_cast_fp16")]; + tensor var_23505_begin_0 = const()[name = tensor("op_23505_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_23505_end_0 = const()[name = tensor("op_23505_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_23505_end_mask_0 = const()[name = tensor("op_23505_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23505_cast_fp16 = slice_by_index(begin = var_23505_begin_0, end = var_23505_end_0, end_mask = var_23505_end_mask_0, x = q_107_cast_fp16)[name = tensor("op_23505_cast_fp16")]; + tensor var_23509_begin_0 = const()[name = tensor("op_23509_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_23509_end_0 = const()[name = tensor("op_23509_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_23509_end_mask_0 = const()[name = tensor("op_23509_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23509_cast_fp16 = slice_by_index(begin = var_23509_begin_0, end = var_23509_end_0, end_mask = var_23509_end_mask_0, x = q_107_cast_fp16)[name = tensor("op_23509_cast_fp16")]; + tensor var_23513_begin_0 = const()[name = tensor("op_23513_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_23513_end_0 = const()[name = tensor("op_23513_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_23513_end_mask_0 = const()[name = tensor("op_23513_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23513_cast_fp16 = slice_by_index(begin = var_23513_begin_0, end = var_23513_end_0, end_mask = var_23513_end_mask_0, x = q_107_cast_fp16)[name = tensor("op_23513_cast_fp16")]; + tensor var_23517_begin_0 = const()[name = tensor("op_23517_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_23517_end_0 = const()[name = tensor("op_23517_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_23517_end_mask_0 = const()[name = tensor("op_23517_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23517_cast_fp16 = slice_by_index(begin = var_23517_begin_0, end = var_23517_end_0, end_mask = var_23517_end_mask_0, x = q_107_cast_fp16)[name = tensor("op_23517_cast_fp16")]; + tensor var_23521_begin_0 = const()[name = tensor("op_23521_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_23521_end_0 = const()[name = tensor("op_23521_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_23521_end_mask_0 = const()[name = tensor("op_23521_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23521_cast_fp16 = slice_by_index(begin = var_23521_begin_0, end = var_23521_end_0, end_mask = var_23521_end_mask_0, x = q_107_cast_fp16)[name = tensor("op_23521_cast_fp16")]; + tensor var_23525_begin_0 = const()[name = tensor("op_23525_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_23525_end_0 = const()[name = tensor("op_23525_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_23525_end_mask_0 = const()[name = tensor("op_23525_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23525_cast_fp16 = slice_by_index(begin = var_23525_begin_0, end = var_23525_end_0, end_mask = var_23525_end_mask_0, x = q_107_cast_fp16)[name = tensor("op_23525_cast_fp16")]; + tensor var_23529_begin_0 = const()[name = tensor("op_23529_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_23529_end_0 = const()[name = tensor("op_23529_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_23529_end_mask_0 = const()[name = tensor("op_23529_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23529_cast_fp16 = slice_by_index(begin = var_23529_begin_0, end = var_23529_end_0, end_mask = var_23529_end_mask_0, x = q_107_cast_fp16)[name = tensor("op_23529_cast_fp16")]; + tensor var_23533_begin_0 = const()[name = tensor("op_23533_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_23533_end_0 = const()[name = tensor("op_23533_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_23533_end_mask_0 = const()[name = tensor("op_23533_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23533_cast_fp16 = slice_by_index(begin = var_23533_begin_0, end = var_23533_end_0, end_mask = var_23533_end_mask_0, x = q_107_cast_fp16)[name = tensor("op_23533_cast_fp16")]; + tensor var_23537_begin_0 = const()[name = tensor("op_23537_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_23537_end_0 = const()[name = tensor("op_23537_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_23537_end_mask_0 = const()[name = tensor("op_23537_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23537_cast_fp16 = slice_by_index(begin = var_23537_begin_0, end = var_23537_end_0, end_mask = var_23537_end_mask_0, x = q_107_cast_fp16)[name = tensor("op_23537_cast_fp16")]; + tensor var_23541_begin_0 = const()[name = tensor("op_23541_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_23541_end_0 = const()[name = tensor("op_23541_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_23541_end_mask_0 = const()[name = tensor("op_23541_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23541_cast_fp16 = slice_by_index(begin = var_23541_begin_0, end = var_23541_end_0, end_mask = var_23541_end_mask_0, x = q_107_cast_fp16)[name = tensor("op_23541_cast_fp16")]; + tensor k_215_perm_0 = const()[name = tensor("k_215_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_23548_begin_0 = const()[name = tensor("op_23548_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23548_end_0 = const()[name = tensor("op_23548_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_23548_end_mask_0 = const()[name = tensor("op_23548_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_215_cast_fp16 = transpose(perm = k_215_perm_0, x = k_213_cast_fp16)[name = tensor("transpose_86")]; + tensor var_23548_cast_fp16 = slice_by_index(begin = var_23548_begin_0, end = var_23548_end_0, end_mask = var_23548_end_mask_0, x = k_215_cast_fp16)[name = tensor("op_23548_cast_fp16")]; + tensor var_23552_begin_0 = const()[name = tensor("op_23552_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_23552_end_0 = const()[name = tensor("op_23552_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_23552_end_mask_0 = const()[name = tensor("op_23552_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23552_cast_fp16 = slice_by_index(begin = var_23552_begin_0, end = var_23552_end_0, end_mask = var_23552_end_mask_0, x = k_215_cast_fp16)[name = tensor("op_23552_cast_fp16")]; + tensor var_23556_begin_0 = const()[name = tensor("op_23556_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_23556_end_0 = const()[name = tensor("op_23556_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_23556_end_mask_0 = const()[name = tensor("op_23556_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23556_cast_fp16 = slice_by_index(begin = var_23556_begin_0, end = var_23556_end_0, end_mask = var_23556_end_mask_0, x = k_215_cast_fp16)[name = tensor("op_23556_cast_fp16")]; + tensor var_23560_begin_0 = const()[name = tensor("op_23560_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_23560_end_0 = const()[name = tensor("op_23560_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_23560_end_mask_0 = const()[name = tensor("op_23560_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23560_cast_fp16 = slice_by_index(begin = var_23560_begin_0, end = var_23560_end_0, end_mask = var_23560_end_mask_0, x = k_215_cast_fp16)[name = tensor("op_23560_cast_fp16")]; + tensor var_23564_begin_0 = const()[name = tensor("op_23564_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_23564_end_0 = const()[name = tensor("op_23564_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_23564_end_mask_0 = const()[name = tensor("op_23564_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23564_cast_fp16 = slice_by_index(begin = var_23564_begin_0, end = var_23564_end_0, end_mask = var_23564_end_mask_0, x = k_215_cast_fp16)[name = tensor("op_23564_cast_fp16")]; + tensor var_23568_begin_0 = const()[name = tensor("op_23568_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_23568_end_0 = const()[name = tensor("op_23568_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_23568_end_mask_0 = const()[name = tensor("op_23568_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23568_cast_fp16 = slice_by_index(begin = var_23568_begin_0, end = var_23568_end_0, end_mask = var_23568_end_mask_0, x = k_215_cast_fp16)[name = tensor("op_23568_cast_fp16")]; + tensor var_23572_begin_0 = const()[name = tensor("op_23572_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_23572_end_0 = const()[name = tensor("op_23572_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_23572_end_mask_0 = const()[name = tensor("op_23572_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23572_cast_fp16 = slice_by_index(begin = var_23572_begin_0, end = var_23572_end_0, end_mask = var_23572_end_mask_0, x = k_215_cast_fp16)[name = tensor("op_23572_cast_fp16")]; + tensor var_23576_begin_0 = const()[name = tensor("op_23576_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_23576_end_0 = const()[name = tensor("op_23576_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_23576_end_mask_0 = const()[name = tensor("op_23576_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23576_cast_fp16 = slice_by_index(begin = var_23576_begin_0, end = var_23576_end_0, end_mask = var_23576_end_mask_0, x = k_215_cast_fp16)[name = tensor("op_23576_cast_fp16")]; + tensor var_23580_begin_0 = const()[name = tensor("op_23580_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_23580_end_0 = const()[name = tensor("op_23580_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_23580_end_mask_0 = const()[name = tensor("op_23580_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23580_cast_fp16 = slice_by_index(begin = var_23580_begin_0, end = var_23580_end_0, end_mask = var_23580_end_mask_0, x = k_215_cast_fp16)[name = tensor("op_23580_cast_fp16")]; + tensor var_23584_begin_0 = const()[name = tensor("op_23584_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_23584_end_0 = const()[name = tensor("op_23584_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_23584_end_mask_0 = const()[name = tensor("op_23584_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23584_cast_fp16 = slice_by_index(begin = var_23584_begin_0, end = var_23584_end_0, end_mask = var_23584_end_mask_0, x = k_215_cast_fp16)[name = tensor("op_23584_cast_fp16")]; + tensor var_23588_begin_0 = const()[name = tensor("op_23588_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_23588_end_0 = const()[name = tensor("op_23588_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_23588_end_mask_0 = const()[name = tensor("op_23588_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23588_cast_fp16 = slice_by_index(begin = var_23588_begin_0, end = var_23588_end_0, end_mask = var_23588_end_mask_0, x = k_215_cast_fp16)[name = tensor("op_23588_cast_fp16")]; + tensor var_23592_begin_0 = const()[name = tensor("op_23592_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_23592_end_0 = const()[name = tensor("op_23592_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_23592_end_mask_0 = const()[name = tensor("op_23592_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23592_cast_fp16 = slice_by_index(begin = var_23592_begin_0, end = var_23592_end_0, end_mask = var_23592_end_mask_0, x = k_215_cast_fp16)[name = tensor("op_23592_cast_fp16")]; + tensor var_23596_begin_0 = const()[name = tensor("op_23596_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_23596_end_0 = const()[name = tensor("op_23596_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_23596_end_mask_0 = const()[name = tensor("op_23596_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23596_cast_fp16 = slice_by_index(begin = var_23596_begin_0, end = var_23596_end_0, end_mask = var_23596_end_mask_0, x = k_215_cast_fp16)[name = tensor("op_23596_cast_fp16")]; + tensor var_23600_begin_0 = const()[name = tensor("op_23600_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_23600_end_0 = const()[name = tensor("op_23600_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_23600_end_mask_0 = const()[name = tensor("op_23600_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23600_cast_fp16 = slice_by_index(begin = var_23600_begin_0, end = var_23600_end_0, end_mask = var_23600_end_mask_0, x = k_215_cast_fp16)[name = tensor("op_23600_cast_fp16")]; + tensor var_23604_begin_0 = const()[name = tensor("op_23604_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_23604_end_0 = const()[name = tensor("op_23604_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_23604_end_mask_0 = const()[name = tensor("op_23604_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23604_cast_fp16 = slice_by_index(begin = var_23604_begin_0, end = var_23604_end_0, end_mask = var_23604_end_mask_0, x = k_215_cast_fp16)[name = tensor("op_23604_cast_fp16")]; + tensor var_23608_begin_0 = const()[name = tensor("op_23608_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_23608_end_0 = const()[name = tensor("op_23608_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_23608_end_mask_0 = const()[name = tensor("op_23608_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23608_cast_fp16 = slice_by_index(begin = var_23608_begin_0, end = var_23608_end_0, end_mask = var_23608_end_mask_0, x = k_215_cast_fp16)[name = tensor("op_23608_cast_fp16")]; + tensor var_23612_begin_0 = const()[name = tensor("op_23612_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_23612_end_0 = const()[name = tensor("op_23612_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_23612_end_mask_0 = const()[name = tensor("op_23612_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23612_cast_fp16 = slice_by_index(begin = var_23612_begin_0, end = var_23612_end_0, end_mask = var_23612_end_mask_0, x = k_215_cast_fp16)[name = tensor("op_23612_cast_fp16")]; + tensor var_23616_begin_0 = const()[name = tensor("op_23616_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_23616_end_0 = const()[name = tensor("op_23616_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_23616_end_mask_0 = const()[name = tensor("op_23616_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23616_cast_fp16 = slice_by_index(begin = var_23616_begin_0, end = var_23616_end_0, end_mask = var_23616_end_mask_0, x = k_215_cast_fp16)[name = tensor("op_23616_cast_fp16")]; + tensor var_23620_begin_0 = const()[name = tensor("op_23620_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_23620_end_0 = const()[name = tensor("op_23620_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_23620_end_mask_0 = const()[name = tensor("op_23620_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23620_cast_fp16 = slice_by_index(begin = var_23620_begin_0, end = var_23620_end_0, end_mask = var_23620_end_mask_0, x = k_215_cast_fp16)[name = tensor("op_23620_cast_fp16")]; + tensor var_23624_begin_0 = const()[name = tensor("op_23624_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_23624_end_0 = const()[name = tensor("op_23624_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_23624_end_mask_0 = const()[name = tensor("op_23624_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23624_cast_fp16 = slice_by_index(begin = var_23624_begin_0, end = var_23624_end_0, end_mask = var_23624_end_mask_0, x = k_215_cast_fp16)[name = tensor("op_23624_cast_fp16")]; + tensor var_23626_begin_0 = const()[name = tensor("op_23626_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23626_end_0 = const()[name = tensor("op_23626_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_23626_end_mask_0 = const()[name = tensor("op_23626_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23626_cast_fp16 = slice_by_index(begin = var_23626_begin_0, end = var_23626_end_0, end_mask = var_23626_end_mask_0, x = v_107_cast_fp16)[name = tensor("op_23626_cast_fp16")]; + tensor var_23630_begin_0 = const()[name = tensor("op_23630_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_23630_end_0 = const()[name = tensor("op_23630_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_23630_end_mask_0 = const()[name = tensor("op_23630_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23630_cast_fp16 = slice_by_index(begin = var_23630_begin_0, end = var_23630_end_0, end_mask = var_23630_end_mask_0, x = v_107_cast_fp16)[name = tensor("op_23630_cast_fp16")]; + tensor var_23634_begin_0 = const()[name = tensor("op_23634_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_23634_end_0 = const()[name = tensor("op_23634_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_23634_end_mask_0 = const()[name = tensor("op_23634_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23634_cast_fp16 = slice_by_index(begin = var_23634_begin_0, end = var_23634_end_0, end_mask = var_23634_end_mask_0, x = v_107_cast_fp16)[name = tensor("op_23634_cast_fp16")]; + tensor var_23638_begin_0 = const()[name = tensor("op_23638_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_23638_end_0 = const()[name = tensor("op_23638_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_23638_end_mask_0 = const()[name = tensor("op_23638_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23638_cast_fp16 = slice_by_index(begin = var_23638_begin_0, end = var_23638_end_0, end_mask = var_23638_end_mask_0, x = v_107_cast_fp16)[name = tensor("op_23638_cast_fp16")]; + tensor var_23642_begin_0 = const()[name = tensor("op_23642_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_23642_end_0 = const()[name = tensor("op_23642_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_23642_end_mask_0 = const()[name = tensor("op_23642_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23642_cast_fp16 = slice_by_index(begin = var_23642_begin_0, end = var_23642_end_0, end_mask = var_23642_end_mask_0, x = v_107_cast_fp16)[name = tensor("op_23642_cast_fp16")]; + tensor var_23646_begin_0 = const()[name = tensor("op_23646_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_23646_end_0 = const()[name = tensor("op_23646_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_23646_end_mask_0 = const()[name = tensor("op_23646_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23646_cast_fp16 = slice_by_index(begin = var_23646_begin_0, end = var_23646_end_0, end_mask = var_23646_end_mask_0, x = v_107_cast_fp16)[name = tensor("op_23646_cast_fp16")]; + tensor var_23650_begin_0 = const()[name = tensor("op_23650_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_23650_end_0 = const()[name = tensor("op_23650_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_23650_end_mask_0 = const()[name = tensor("op_23650_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23650_cast_fp16 = slice_by_index(begin = var_23650_begin_0, end = var_23650_end_0, end_mask = var_23650_end_mask_0, x = v_107_cast_fp16)[name = tensor("op_23650_cast_fp16")]; + tensor var_23654_begin_0 = const()[name = tensor("op_23654_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_23654_end_0 = const()[name = tensor("op_23654_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_23654_end_mask_0 = const()[name = tensor("op_23654_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23654_cast_fp16 = slice_by_index(begin = var_23654_begin_0, end = var_23654_end_0, end_mask = var_23654_end_mask_0, x = v_107_cast_fp16)[name = tensor("op_23654_cast_fp16")]; + tensor var_23658_begin_0 = const()[name = tensor("op_23658_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_23658_end_0 = const()[name = tensor("op_23658_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_23658_end_mask_0 = const()[name = tensor("op_23658_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23658_cast_fp16 = slice_by_index(begin = var_23658_begin_0, end = var_23658_end_0, end_mask = var_23658_end_mask_0, x = v_107_cast_fp16)[name = tensor("op_23658_cast_fp16")]; + tensor var_23662_begin_0 = const()[name = tensor("op_23662_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_23662_end_0 = const()[name = tensor("op_23662_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_23662_end_mask_0 = const()[name = tensor("op_23662_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23662_cast_fp16 = slice_by_index(begin = var_23662_begin_0, end = var_23662_end_0, end_mask = var_23662_end_mask_0, x = v_107_cast_fp16)[name = tensor("op_23662_cast_fp16")]; + tensor var_23666_begin_0 = const()[name = tensor("op_23666_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_23666_end_0 = const()[name = tensor("op_23666_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_23666_end_mask_0 = const()[name = tensor("op_23666_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23666_cast_fp16 = slice_by_index(begin = var_23666_begin_0, end = var_23666_end_0, end_mask = var_23666_end_mask_0, x = v_107_cast_fp16)[name = tensor("op_23666_cast_fp16")]; + tensor var_23670_begin_0 = const()[name = tensor("op_23670_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_23670_end_0 = const()[name = tensor("op_23670_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_23670_end_mask_0 = const()[name = tensor("op_23670_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23670_cast_fp16 = slice_by_index(begin = var_23670_begin_0, end = var_23670_end_0, end_mask = var_23670_end_mask_0, x = v_107_cast_fp16)[name = tensor("op_23670_cast_fp16")]; + tensor var_23674_begin_0 = const()[name = tensor("op_23674_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_23674_end_0 = const()[name = tensor("op_23674_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_23674_end_mask_0 = const()[name = tensor("op_23674_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23674_cast_fp16 = slice_by_index(begin = var_23674_begin_0, end = var_23674_end_0, end_mask = var_23674_end_mask_0, x = v_107_cast_fp16)[name = tensor("op_23674_cast_fp16")]; + tensor var_23678_begin_0 = const()[name = tensor("op_23678_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_23678_end_0 = const()[name = tensor("op_23678_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_23678_end_mask_0 = const()[name = tensor("op_23678_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23678_cast_fp16 = slice_by_index(begin = var_23678_begin_0, end = var_23678_end_0, end_mask = var_23678_end_mask_0, x = v_107_cast_fp16)[name = tensor("op_23678_cast_fp16")]; + tensor var_23682_begin_0 = const()[name = tensor("op_23682_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_23682_end_0 = const()[name = tensor("op_23682_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_23682_end_mask_0 = const()[name = tensor("op_23682_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23682_cast_fp16 = slice_by_index(begin = var_23682_begin_0, end = var_23682_end_0, end_mask = var_23682_end_mask_0, x = v_107_cast_fp16)[name = tensor("op_23682_cast_fp16")]; + tensor var_23686_begin_0 = const()[name = tensor("op_23686_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_23686_end_0 = const()[name = tensor("op_23686_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_23686_end_mask_0 = const()[name = tensor("op_23686_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23686_cast_fp16 = slice_by_index(begin = var_23686_begin_0, end = var_23686_end_0, end_mask = var_23686_end_mask_0, x = v_107_cast_fp16)[name = tensor("op_23686_cast_fp16")]; + tensor var_23690_begin_0 = const()[name = tensor("op_23690_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_23690_end_0 = const()[name = tensor("op_23690_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_23690_end_mask_0 = const()[name = tensor("op_23690_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23690_cast_fp16 = slice_by_index(begin = var_23690_begin_0, end = var_23690_end_0, end_mask = var_23690_end_mask_0, x = v_107_cast_fp16)[name = tensor("op_23690_cast_fp16")]; + tensor var_23694_begin_0 = const()[name = tensor("op_23694_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_23694_end_0 = const()[name = tensor("op_23694_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_23694_end_mask_0 = const()[name = tensor("op_23694_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23694_cast_fp16 = slice_by_index(begin = var_23694_begin_0, end = var_23694_end_0, end_mask = var_23694_end_mask_0, x = v_107_cast_fp16)[name = tensor("op_23694_cast_fp16")]; + tensor var_23698_begin_0 = const()[name = tensor("op_23698_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_23698_end_0 = const()[name = tensor("op_23698_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_23698_end_mask_0 = const()[name = tensor("op_23698_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23698_cast_fp16 = slice_by_index(begin = var_23698_begin_0, end = var_23698_end_0, end_mask = var_23698_end_mask_0, x = v_107_cast_fp16)[name = tensor("op_23698_cast_fp16")]; + tensor var_23702_begin_0 = const()[name = tensor("op_23702_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_23702_end_0 = const()[name = tensor("op_23702_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_23702_end_mask_0 = const()[name = tensor("op_23702_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23702_cast_fp16 = slice_by_index(begin = var_23702_begin_0, end = var_23702_end_0, end_mask = var_23702_end_mask_0, x = v_107_cast_fp16)[name = tensor("op_23702_cast_fp16")]; + tensor var_23706_equation_0 = const()[name = tensor("op_23706_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23706_cast_fp16 = einsum(equation = var_23706_equation_0, values = (var_23548_cast_fp16, var_23465_cast_fp16))[name = tensor("op_23706_cast_fp16")]; + tensor var_23707_to_fp16 = const()[name = tensor("op_23707_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1961_cast_fp16 = mul(x = var_23706_cast_fp16, y = var_23707_to_fp16)[name = tensor("aw_1961_cast_fp16")]; + tensor var_23710_equation_0 = const()[name = tensor("op_23710_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23710_cast_fp16 = einsum(equation = var_23710_equation_0, values = (var_23552_cast_fp16, var_23469_cast_fp16))[name = tensor("op_23710_cast_fp16")]; + tensor var_23711_to_fp16 = const()[name = tensor("op_23711_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1963_cast_fp16 = mul(x = var_23710_cast_fp16, y = var_23711_to_fp16)[name = tensor("aw_1963_cast_fp16")]; + tensor var_23714_equation_0 = const()[name = tensor("op_23714_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23714_cast_fp16 = einsum(equation = var_23714_equation_0, values = (var_23556_cast_fp16, var_23473_cast_fp16))[name = tensor("op_23714_cast_fp16")]; + tensor var_23715_to_fp16 = const()[name = tensor("op_23715_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1965_cast_fp16 = mul(x = var_23714_cast_fp16, y = var_23715_to_fp16)[name = tensor("aw_1965_cast_fp16")]; + tensor var_23718_equation_0 = const()[name = tensor("op_23718_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23718_cast_fp16 = einsum(equation = var_23718_equation_0, values = (var_23560_cast_fp16, var_23477_cast_fp16))[name = tensor("op_23718_cast_fp16")]; + tensor var_23719_to_fp16 = const()[name = tensor("op_23719_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1967_cast_fp16 = mul(x = var_23718_cast_fp16, y = var_23719_to_fp16)[name = tensor("aw_1967_cast_fp16")]; + tensor var_23722_equation_0 = const()[name = tensor("op_23722_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23722_cast_fp16 = einsum(equation = var_23722_equation_0, values = (var_23564_cast_fp16, var_23481_cast_fp16))[name = tensor("op_23722_cast_fp16")]; + tensor var_23723_to_fp16 = const()[name = tensor("op_23723_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1969_cast_fp16 = mul(x = var_23722_cast_fp16, y = var_23723_to_fp16)[name = tensor("aw_1969_cast_fp16")]; + tensor var_23726_equation_0 = const()[name = tensor("op_23726_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23726_cast_fp16 = einsum(equation = var_23726_equation_0, values = (var_23568_cast_fp16, var_23485_cast_fp16))[name = tensor("op_23726_cast_fp16")]; + tensor var_23727_to_fp16 = const()[name = tensor("op_23727_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1971_cast_fp16 = mul(x = var_23726_cast_fp16, y = var_23727_to_fp16)[name = tensor("aw_1971_cast_fp16")]; + tensor var_23730_equation_0 = const()[name = tensor("op_23730_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23730_cast_fp16 = einsum(equation = var_23730_equation_0, values = (var_23572_cast_fp16, var_23489_cast_fp16))[name = tensor("op_23730_cast_fp16")]; + tensor var_23731_to_fp16 = const()[name = tensor("op_23731_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1973_cast_fp16 = mul(x = var_23730_cast_fp16, y = var_23731_to_fp16)[name = tensor("aw_1973_cast_fp16")]; + tensor var_23734_equation_0 = const()[name = tensor("op_23734_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23734_cast_fp16 = einsum(equation = var_23734_equation_0, values = (var_23576_cast_fp16, var_23493_cast_fp16))[name = tensor("op_23734_cast_fp16")]; + tensor var_23735_to_fp16 = const()[name = tensor("op_23735_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1975_cast_fp16 = mul(x = var_23734_cast_fp16, y = var_23735_to_fp16)[name = tensor("aw_1975_cast_fp16")]; + tensor var_23738_equation_0 = const()[name = tensor("op_23738_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23738_cast_fp16 = einsum(equation = var_23738_equation_0, values = (var_23580_cast_fp16, var_23497_cast_fp16))[name = tensor("op_23738_cast_fp16")]; + tensor var_23739_to_fp16 = const()[name = tensor("op_23739_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1977_cast_fp16 = mul(x = var_23738_cast_fp16, y = var_23739_to_fp16)[name = tensor("aw_1977_cast_fp16")]; + tensor var_23742_equation_0 = const()[name = tensor("op_23742_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23742_cast_fp16 = einsum(equation = var_23742_equation_0, values = (var_23584_cast_fp16, var_23501_cast_fp16))[name = tensor("op_23742_cast_fp16")]; + tensor var_23743_to_fp16 = const()[name = tensor("op_23743_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1979_cast_fp16 = mul(x = var_23742_cast_fp16, y = var_23743_to_fp16)[name = tensor("aw_1979_cast_fp16")]; + tensor var_23746_equation_0 = const()[name = tensor("op_23746_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23746_cast_fp16 = einsum(equation = var_23746_equation_0, values = (var_23588_cast_fp16, var_23505_cast_fp16))[name = tensor("op_23746_cast_fp16")]; + tensor var_23747_to_fp16 = const()[name = tensor("op_23747_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1981_cast_fp16 = mul(x = var_23746_cast_fp16, y = var_23747_to_fp16)[name = tensor("aw_1981_cast_fp16")]; + tensor var_23750_equation_0 = const()[name = tensor("op_23750_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23750_cast_fp16 = einsum(equation = var_23750_equation_0, values = (var_23592_cast_fp16, var_23509_cast_fp16))[name = tensor("op_23750_cast_fp16")]; + tensor var_23751_to_fp16 = const()[name = tensor("op_23751_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1983_cast_fp16 = mul(x = var_23750_cast_fp16, y = var_23751_to_fp16)[name = tensor("aw_1983_cast_fp16")]; + tensor var_23754_equation_0 = const()[name = tensor("op_23754_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23754_cast_fp16 = einsum(equation = var_23754_equation_0, values = (var_23596_cast_fp16, var_23513_cast_fp16))[name = tensor("op_23754_cast_fp16")]; + tensor var_23755_to_fp16 = const()[name = tensor("op_23755_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1985_cast_fp16 = mul(x = var_23754_cast_fp16, y = var_23755_to_fp16)[name = tensor("aw_1985_cast_fp16")]; + tensor var_23758_equation_0 = const()[name = tensor("op_23758_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23758_cast_fp16 = einsum(equation = var_23758_equation_0, values = (var_23600_cast_fp16, var_23517_cast_fp16))[name = tensor("op_23758_cast_fp16")]; + tensor var_23759_to_fp16 = const()[name = tensor("op_23759_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1987_cast_fp16 = mul(x = var_23758_cast_fp16, y = var_23759_to_fp16)[name = tensor("aw_1987_cast_fp16")]; + tensor var_23762_equation_0 = const()[name = tensor("op_23762_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23762_cast_fp16 = einsum(equation = var_23762_equation_0, values = (var_23604_cast_fp16, var_23521_cast_fp16))[name = tensor("op_23762_cast_fp16")]; + tensor var_23763_to_fp16 = const()[name = tensor("op_23763_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1989_cast_fp16 = mul(x = var_23762_cast_fp16, y = var_23763_to_fp16)[name = tensor("aw_1989_cast_fp16")]; + tensor var_23766_equation_0 = const()[name = tensor("op_23766_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23766_cast_fp16 = einsum(equation = var_23766_equation_0, values = (var_23608_cast_fp16, var_23525_cast_fp16))[name = tensor("op_23766_cast_fp16")]; + tensor var_23767_to_fp16 = const()[name = tensor("op_23767_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1991_cast_fp16 = mul(x = var_23766_cast_fp16, y = var_23767_to_fp16)[name = tensor("aw_1991_cast_fp16")]; + tensor var_23770_equation_0 = const()[name = tensor("op_23770_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23770_cast_fp16 = einsum(equation = var_23770_equation_0, values = (var_23612_cast_fp16, var_23529_cast_fp16))[name = tensor("op_23770_cast_fp16")]; + tensor var_23771_to_fp16 = const()[name = tensor("op_23771_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1993_cast_fp16 = mul(x = var_23770_cast_fp16, y = var_23771_to_fp16)[name = tensor("aw_1993_cast_fp16")]; + tensor var_23774_equation_0 = const()[name = tensor("op_23774_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23774_cast_fp16 = einsum(equation = var_23774_equation_0, values = (var_23616_cast_fp16, var_23533_cast_fp16))[name = tensor("op_23774_cast_fp16")]; + tensor var_23775_to_fp16 = const()[name = tensor("op_23775_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1995_cast_fp16 = mul(x = var_23774_cast_fp16, y = var_23775_to_fp16)[name = tensor("aw_1995_cast_fp16")]; + tensor var_23778_equation_0 = const()[name = tensor("op_23778_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23778_cast_fp16 = einsum(equation = var_23778_equation_0, values = (var_23620_cast_fp16, var_23537_cast_fp16))[name = tensor("op_23778_cast_fp16")]; + tensor var_23779_to_fp16 = const()[name = tensor("op_23779_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1997_cast_fp16 = mul(x = var_23778_cast_fp16, y = var_23779_to_fp16)[name = tensor("aw_1997_cast_fp16")]; + tensor var_23782_equation_0 = const()[name = tensor("op_23782_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23782_cast_fp16 = einsum(equation = var_23782_equation_0, values = (var_23624_cast_fp16, var_23541_cast_fp16))[name = tensor("op_23782_cast_fp16")]; + tensor var_23783_to_fp16 = const()[name = tensor("op_23783_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1999_cast_fp16 = mul(x = var_23782_cast_fp16, y = var_23783_to_fp16)[name = tensor("aw_1999_cast_fp16")]; + tensor var_23785_cast_fp16 = softmax(axis = var_21077, x = aw_1961_cast_fp16)[name = tensor("op_23785_cast_fp16")]; + tensor var_23786_cast_fp16 = softmax(axis = var_21077, x = aw_1963_cast_fp16)[name = tensor("op_23786_cast_fp16")]; + tensor var_23787_cast_fp16 = softmax(axis = var_21077, x = aw_1965_cast_fp16)[name = tensor("op_23787_cast_fp16")]; + tensor var_23788_cast_fp16 = softmax(axis = var_21077, x = aw_1967_cast_fp16)[name = tensor("op_23788_cast_fp16")]; + tensor var_23789_cast_fp16 = softmax(axis = var_21077, x = aw_1969_cast_fp16)[name = tensor("op_23789_cast_fp16")]; + tensor var_23790_cast_fp16 = softmax(axis = var_21077, x = aw_1971_cast_fp16)[name = tensor("op_23790_cast_fp16")]; + tensor var_23791_cast_fp16 = softmax(axis = var_21077, x = aw_1973_cast_fp16)[name = tensor("op_23791_cast_fp16")]; + tensor var_23792_cast_fp16 = softmax(axis = var_21077, x = aw_1975_cast_fp16)[name = tensor("op_23792_cast_fp16")]; + tensor var_23793_cast_fp16 = softmax(axis = var_21077, x = aw_1977_cast_fp16)[name = tensor("op_23793_cast_fp16")]; + tensor var_23794_cast_fp16 = softmax(axis = var_21077, x = aw_1979_cast_fp16)[name = tensor("op_23794_cast_fp16")]; + tensor var_23795_cast_fp16 = softmax(axis = var_21077, x = aw_1981_cast_fp16)[name = tensor("op_23795_cast_fp16")]; + tensor var_23796_cast_fp16 = softmax(axis = var_21077, x = aw_1983_cast_fp16)[name = tensor("op_23796_cast_fp16")]; + tensor var_23797_cast_fp16 = softmax(axis = var_21077, x = aw_1985_cast_fp16)[name = tensor("op_23797_cast_fp16")]; + tensor var_23798_cast_fp16 = softmax(axis = var_21077, x = aw_1987_cast_fp16)[name = tensor("op_23798_cast_fp16")]; + tensor var_23799_cast_fp16 = softmax(axis = var_21077, x = aw_1989_cast_fp16)[name = tensor("op_23799_cast_fp16")]; + tensor var_23800_cast_fp16 = softmax(axis = var_21077, x = aw_1991_cast_fp16)[name = tensor("op_23800_cast_fp16")]; + tensor var_23801_cast_fp16 = softmax(axis = var_21077, x = aw_1993_cast_fp16)[name = tensor("op_23801_cast_fp16")]; + tensor var_23802_cast_fp16 = softmax(axis = var_21077, x = aw_1995_cast_fp16)[name = tensor("op_23802_cast_fp16")]; + tensor var_23803_cast_fp16 = softmax(axis = var_21077, x = aw_1997_cast_fp16)[name = tensor("op_23803_cast_fp16")]; + tensor var_23804_cast_fp16 = softmax(axis = var_21077, x = aw_1999_cast_fp16)[name = tensor("op_23804_cast_fp16")]; + tensor var_23806_equation_0 = const()[name = tensor("op_23806_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23806_cast_fp16 = einsum(equation = var_23806_equation_0, values = (var_23626_cast_fp16, var_23785_cast_fp16))[name = tensor("op_23806_cast_fp16")]; + tensor var_23808_equation_0 = const()[name = tensor("op_23808_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23808_cast_fp16 = einsum(equation = var_23808_equation_0, values = (var_23630_cast_fp16, var_23786_cast_fp16))[name = tensor("op_23808_cast_fp16")]; + tensor var_23810_equation_0 = const()[name = tensor("op_23810_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23810_cast_fp16 = einsum(equation = var_23810_equation_0, values = (var_23634_cast_fp16, var_23787_cast_fp16))[name = tensor("op_23810_cast_fp16")]; + tensor var_23812_equation_0 = const()[name = tensor("op_23812_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23812_cast_fp16 = einsum(equation = var_23812_equation_0, values = (var_23638_cast_fp16, var_23788_cast_fp16))[name = tensor("op_23812_cast_fp16")]; + tensor var_23814_equation_0 = const()[name = tensor("op_23814_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23814_cast_fp16 = einsum(equation = var_23814_equation_0, values = (var_23642_cast_fp16, var_23789_cast_fp16))[name = tensor("op_23814_cast_fp16")]; + tensor var_23816_equation_0 = const()[name = tensor("op_23816_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23816_cast_fp16 = einsum(equation = var_23816_equation_0, values = (var_23646_cast_fp16, var_23790_cast_fp16))[name = tensor("op_23816_cast_fp16")]; + tensor var_23818_equation_0 = const()[name = tensor("op_23818_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23818_cast_fp16 = einsum(equation = var_23818_equation_0, values = (var_23650_cast_fp16, var_23791_cast_fp16))[name = tensor("op_23818_cast_fp16")]; + tensor var_23820_equation_0 = const()[name = tensor("op_23820_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23820_cast_fp16 = einsum(equation = var_23820_equation_0, values = (var_23654_cast_fp16, var_23792_cast_fp16))[name = tensor("op_23820_cast_fp16")]; + tensor var_23822_equation_0 = const()[name = tensor("op_23822_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23822_cast_fp16 = einsum(equation = var_23822_equation_0, values = (var_23658_cast_fp16, var_23793_cast_fp16))[name = tensor("op_23822_cast_fp16")]; + tensor var_23824_equation_0 = const()[name = tensor("op_23824_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23824_cast_fp16 = einsum(equation = var_23824_equation_0, values = (var_23662_cast_fp16, var_23794_cast_fp16))[name = tensor("op_23824_cast_fp16")]; + tensor var_23826_equation_0 = const()[name = tensor("op_23826_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23826_cast_fp16 = einsum(equation = var_23826_equation_0, values = (var_23666_cast_fp16, var_23795_cast_fp16))[name = tensor("op_23826_cast_fp16")]; + tensor var_23828_equation_0 = const()[name = tensor("op_23828_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23828_cast_fp16 = einsum(equation = var_23828_equation_0, values = (var_23670_cast_fp16, var_23796_cast_fp16))[name = tensor("op_23828_cast_fp16")]; + tensor var_23830_equation_0 = const()[name = tensor("op_23830_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23830_cast_fp16 = einsum(equation = var_23830_equation_0, values = (var_23674_cast_fp16, var_23797_cast_fp16))[name = tensor("op_23830_cast_fp16")]; + tensor var_23832_equation_0 = const()[name = tensor("op_23832_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23832_cast_fp16 = einsum(equation = var_23832_equation_0, values = (var_23678_cast_fp16, var_23798_cast_fp16))[name = tensor("op_23832_cast_fp16")]; + tensor var_23834_equation_0 = const()[name = tensor("op_23834_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23834_cast_fp16 = einsum(equation = var_23834_equation_0, values = (var_23682_cast_fp16, var_23799_cast_fp16))[name = tensor("op_23834_cast_fp16")]; + tensor var_23836_equation_0 = const()[name = tensor("op_23836_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23836_cast_fp16 = einsum(equation = var_23836_equation_0, values = (var_23686_cast_fp16, var_23800_cast_fp16))[name = tensor("op_23836_cast_fp16")]; + tensor var_23838_equation_0 = const()[name = tensor("op_23838_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23838_cast_fp16 = einsum(equation = var_23838_equation_0, values = (var_23690_cast_fp16, var_23801_cast_fp16))[name = tensor("op_23838_cast_fp16")]; + tensor var_23840_equation_0 = const()[name = tensor("op_23840_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23840_cast_fp16 = einsum(equation = var_23840_equation_0, values = (var_23694_cast_fp16, var_23802_cast_fp16))[name = tensor("op_23840_cast_fp16")]; + tensor var_23842_equation_0 = const()[name = tensor("op_23842_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23842_cast_fp16 = einsum(equation = var_23842_equation_0, values = (var_23698_cast_fp16, var_23803_cast_fp16))[name = tensor("op_23842_cast_fp16")]; + tensor var_23844_equation_0 = const()[name = tensor("op_23844_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_23844_cast_fp16 = einsum(equation = var_23844_equation_0, values = (var_23702_cast_fp16, var_23804_cast_fp16))[name = tensor("op_23844_cast_fp16")]; + tensor input_345_interleave_0 = const()[name = tensor("input_345_interleave_0"), val = tensor(false)]; + tensor input_345_cast_fp16 = concat(axis = var_21077, interleave = input_345_interleave_0, values = (var_23806_cast_fp16, var_23808_cast_fp16, var_23810_cast_fp16, var_23812_cast_fp16, var_23814_cast_fp16, var_23816_cast_fp16, var_23818_cast_fp16, var_23820_cast_fp16, var_23822_cast_fp16, var_23824_cast_fp16, var_23826_cast_fp16, var_23828_cast_fp16, var_23830_cast_fp16, var_23832_cast_fp16, var_23834_cast_fp16, var_23836_cast_fp16, var_23838_cast_fp16, var_23840_cast_fp16, var_23842_cast_fp16, var_23844_cast_fp16))[name = tensor("input_345_cast_fp16")]; + tensor var_23850 = const()[name = tensor("op_23850"), val = tensor([1, 1])]; + tensor var_23852 = const()[name = tensor("op_23852"), val = tensor([1, 1])]; + tensor var_23854_pad_type_0 = const()[name = tensor("op_23854_pad_type_0"), val = tensor("custom")]; + tensor var_23854_pad_0 = const()[name = tensor("op_23854_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_2_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_2_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1892018304)))]; + tensor mid_block_attentions_0_transformer_blocks_2_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_2_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1895295168)))]; + tensor var_23854_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_2_attn2_to_out_0_bias_to_fp16, dilations = var_23852, groups = var_21077, pad = var_23854_pad_0, pad_type = var_23854_pad_type_0, strides = var_23850, weight = mid_block_attentions_0_transformer_blocks_2_attn2_to_out_0_weight_to_fp16, x = input_345_cast_fp16)[name = tensor("op_23854_cast_fp16")]; + tensor inputs_161_cast_fp16 = add(x = var_23854_cast_fp16, y = inputs_159_cast_fp16)[name = tensor("inputs_161_cast_fp16")]; + tensor input_347_axes_0 = const()[name = tensor("input_347_axes_0"), val = tensor([1])]; + tensor input_347_gamma_0_to_fp16 = const()[name = tensor("input_347_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1895297792)))]; + tensor input_347_beta_0_to_fp16 = const()[name = tensor("input_347_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1895300416)))]; + tensor var_23864_to_fp16 = const()[name = tensor("op_23864_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_347_cast_fp16 = layer_norm(axes = input_347_axes_0, beta = input_347_beta_0_to_fp16, epsilon = var_23864_to_fp16, gamma = input_347_gamma_0_to_fp16, x = inputs_161_cast_fp16)[name = tensor("input_347_cast_fp16")]; + tensor var_23880 = const()[name = tensor("op_23880"), val = tensor([1, 1])]; + tensor var_23882 = const()[name = tensor("op_23882"), val = tensor([1, 1])]; + tensor var_23884_pad_type_0 = const()[name = tensor("op_23884_pad_type_0"), val = tensor("custom")]; + tensor var_23884_pad_0 = const()[name = tensor("op_23884_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_2_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_2_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1895303040)))]; + tensor mid_block_attentions_0_transformer_blocks_2_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_2_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1921517504)))]; + tensor var_23884_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_2_ff_net_0_proj_bias_to_fp16, dilations = var_23882, groups = var_21077, pad = var_23884_pad_0, pad_type = var_23884_pad_type_0, strides = var_23880, weight = mid_block_attentions_0_transformer_blocks_2_ff_net_0_proj_weight_to_fp16, x = input_347_cast_fp16)[name = tensor("op_23884_cast_fp16")]; + tensor var_23885_split_sizes_0 = const()[name = tensor("op_23885_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_23885_axis_0 = const()[name = tensor("op_23885_axis_0"), val = tensor(1)]; + tensor var_23885_cast_fp16_0, tensor var_23885_cast_fp16_1 = split(axis = var_23885_axis_0, split_sizes = var_23885_split_sizes_0, x = var_23884_cast_fp16)[name = tensor("op_23885_cast_fp16")]; + tensor var_23887_mode_0 = const()[name = tensor("op_23887_mode_0"), val = tensor("EXACT")]; + tensor var_23887_cast_fp16 = gelu(mode = var_23887_mode_0, x = var_23885_cast_fp16_1)[name = tensor("op_23887_cast_fp16")]; + tensor input_349_cast_fp16 = mul(x = var_23885_cast_fp16_0, y = var_23887_cast_fp16)[name = tensor("input_349_cast_fp16")]; + tensor var_23891 = const()[name = tensor("op_23891"), val = tensor([1, 1])]; + tensor var_23893 = const()[name = tensor("op_23893"), val = tensor([1, 1])]; + tensor var_23895_pad_type_0 = const()[name = tensor("op_23895_pad_type_0"), val = tensor("custom")]; + tensor var_23895_pad_0 = const()[name = tensor("op_23895_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_2_ff_net_2_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_2_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1921538048)))]; + tensor mid_block_attentions_0_transformer_blocks_2_ff_net_2_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_2_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1934645312)))]; + tensor var_23895_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_2_ff_net_2_bias_to_fp16, dilations = var_23893, groups = var_21077, pad = var_23895_pad_0, pad_type = var_23895_pad_type_0, strides = var_23891, weight = mid_block_attentions_0_transformer_blocks_2_ff_net_2_weight_to_fp16, x = input_349_cast_fp16)[name = tensor("op_23895_cast_fp16")]; + tensor inputs_163_cast_fp16 = add(x = var_23895_cast_fp16, y = inputs_161_cast_fp16)[name = tensor("inputs_163_cast_fp16")]; + tensor hidden_states_227_axes_0 = const()[name = tensor("hidden_states_227_axes_0"), val = tensor([1])]; + tensor hidden_states_227_gamma_0_to_fp16 = const()[name = tensor("hidden_states_227_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1934647936)))]; + tensor hidden_states_227_beta_0_to_fp16 = const()[name = tensor("hidden_states_227_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1934650560)))]; + tensor var_23911_to_fp16 = const()[name = tensor("op_23911_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_227_cast_fp16 = layer_norm(axes = hidden_states_227_axes_0, beta = hidden_states_227_beta_0_to_fp16, epsilon = var_23911_to_fp16, gamma = hidden_states_227_gamma_0_to_fp16, x = inputs_163_cast_fp16)[name = tensor("hidden_states_227_cast_fp16")]; + tensor var_23926 = const()[name = tensor("op_23926"), val = tensor([1, 1])]; + tensor var_23928 = const()[name = tensor("op_23928"), val = tensor([1, 1])]; + tensor q_109_pad_type_0 = const()[name = tensor("q_109_pad_type_0"), val = tensor("custom")]; + tensor q_109_pad_0 = const()[name = tensor("q_109_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_3_attn1_to_q_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_3_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1934653184)))]; + tensor q_109_cast_fp16 = conv(dilations = var_23928, groups = var_21077, pad = q_109_pad_0, pad_type = q_109_pad_type_0, strides = var_23926, weight = mid_block_attentions_0_transformer_blocks_3_attn1_to_q_weight_to_fp16, x = hidden_states_227_cast_fp16)[name = tensor("q_109_cast_fp16")]; + tensor var_23932 = const()[name = tensor("op_23932"), val = tensor([1, 1])]; + tensor var_23934 = const()[name = tensor("op_23934"), val = tensor([1, 1])]; + tensor k_217_pad_type_0 = const()[name = tensor("k_217_pad_type_0"), val = tensor("custom")]; + tensor k_217_pad_0 = const()[name = tensor("k_217_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_3_attn1_to_k_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_3_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1937930048)))]; + tensor k_217_cast_fp16 = conv(dilations = var_23934, groups = var_21077, pad = k_217_pad_0, pad_type = k_217_pad_type_0, strides = var_23932, weight = mid_block_attentions_0_transformer_blocks_3_attn1_to_k_weight_to_fp16, x = hidden_states_227_cast_fp16)[name = tensor("k_217_cast_fp16")]; + tensor var_23938 = const()[name = tensor("op_23938"), val = tensor([1, 1])]; + tensor var_23940 = const()[name = tensor("op_23940"), val = tensor([1, 1])]; + tensor v_109_pad_type_0 = const()[name = tensor("v_109_pad_type_0"), val = tensor("custom")]; + tensor v_109_pad_0 = const()[name = tensor("v_109_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_3_attn1_to_v_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_3_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1941206912)))]; + tensor v_109_cast_fp16 = conv(dilations = var_23940, groups = var_21077, pad = v_109_pad_0, pad_type = v_109_pad_type_0, strides = var_23938, weight = mid_block_attentions_0_transformer_blocks_3_attn1_to_v_weight_to_fp16, x = hidden_states_227_cast_fp16)[name = tensor("v_109_cast_fp16")]; + tensor var_23944_begin_0 = const()[name = tensor("op_23944_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23944_end_0 = const()[name = tensor("op_23944_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_23944_end_mask_0 = const()[name = tensor("op_23944_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23944_cast_fp16 = slice_by_index(begin = var_23944_begin_0, end = var_23944_end_0, end_mask = var_23944_end_mask_0, x = q_109_cast_fp16)[name = tensor("op_23944_cast_fp16")]; + tensor var_23948_begin_0 = const()[name = tensor("op_23948_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_23948_end_0 = const()[name = tensor("op_23948_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_23948_end_mask_0 = const()[name = tensor("op_23948_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23948_cast_fp16 = slice_by_index(begin = var_23948_begin_0, end = var_23948_end_0, end_mask = var_23948_end_mask_0, x = q_109_cast_fp16)[name = tensor("op_23948_cast_fp16")]; + tensor var_23952_begin_0 = const()[name = tensor("op_23952_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_23952_end_0 = const()[name = tensor("op_23952_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_23952_end_mask_0 = const()[name = tensor("op_23952_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23952_cast_fp16 = slice_by_index(begin = var_23952_begin_0, end = var_23952_end_0, end_mask = var_23952_end_mask_0, x = q_109_cast_fp16)[name = tensor("op_23952_cast_fp16")]; + tensor var_23956_begin_0 = const()[name = tensor("op_23956_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_23956_end_0 = const()[name = tensor("op_23956_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_23956_end_mask_0 = const()[name = tensor("op_23956_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23956_cast_fp16 = slice_by_index(begin = var_23956_begin_0, end = var_23956_end_0, end_mask = var_23956_end_mask_0, x = q_109_cast_fp16)[name = tensor("op_23956_cast_fp16")]; + tensor var_23960_begin_0 = const()[name = tensor("op_23960_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_23960_end_0 = const()[name = tensor("op_23960_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_23960_end_mask_0 = const()[name = tensor("op_23960_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23960_cast_fp16 = slice_by_index(begin = var_23960_begin_0, end = var_23960_end_0, end_mask = var_23960_end_mask_0, x = q_109_cast_fp16)[name = tensor("op_23960_cast_fp16")]; + tensor var_23964_begin_0 = const()[name = tensor("op_23964_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_23964_end_0 = const()[name = tensor("op_23964_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_23964_end_mask_0 = const()[name = tensor("op_23964_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23964_cast_fp16 = slice_by_index(begin = var_23964_begin_0, end = var_23964_end_0, end_mask = var_23964_end_mask_0, x = q_109_cast_fp16)[name = tensor("op_23964_cast_fp16")]; + tensor var_23968_begin_0 = const()[name = tensor("op_23968_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_23968_end_0 = const()[name = tensor("op_23968_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_23968_end_mask_0 = const()[name = tensor("op_23968_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23968_cast_fp16 = slice_by_index(begin = var_23968_begin_0, end = var_23968_end_0, end_mask = var_23968_end_mask_0, x = q_109_cast_fp16)[name = tensor("op_23968_cast_fp16")]; + tensor var_23972_begin_0 = const()[name = tensor("op_23972_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_23972_end_0 = const()[name = tensor("op_23972_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_23972_end_mask_0 = const()[name = tensor("op_23972_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23972_cast_fp16 = slice_by_index(begin = var_23972_begin_0, end = var_23972_end_0, end_mask = var_23972_end_mask_0, x = q_109_cast_fp16)[name = tensor("op_23972_cast_fp16")]; + tensor var_23976_begin_0 = const()[name = tensor("op_23976_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_23976_end_0 = const()[name = tensor("op_23976_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_23976_end_mask_0 = const()[name = tensor("op_23976_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23976_cast_fp16 = slice_by_index(begin = var_23976_begin_0, end = var_23976_end_0, end_mask = var_23976_end_mask_0, x = q_109_cast_fp16)[name = tensor("op_23976_cast_fp16")]; + tensor var_23980_begin_0 = const()[name = tensor("op_23980_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_23980_end_0 = const()[name = tensor("op_23980_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_23980_end_mask_0 = const()[name = tensor("op_23980_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23980_cast_fp16 = slice_by_index(begin = var_23980_begin_0, end = var_23980_end_0, end_mask = var_23980_end_mask_0, x = q_109_cast_fp16)[name = tensor("op_23980_cast_fp16")]; + tensor var_23984_begin_0 = const()[name = tensor("op_23984_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_23984_end_0 = const()[name = tensor("op_23984_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_23984_end_mask_0 = const()[name = tensor("op_23984_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23984_cast_fp16 = slice_by_index(begin = var_23984_begin_0, end = var_23984_end_0, end_mask = var_23984_end_mask_0, x = q_109_cast_fp16)[name = tensor("op_23984_cast_fp16")]; + tensor var_23988_begin_0 = const()[name = tensor("op_23988_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_23988_end_0 = const()[name = tensor("op_23988_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_23988_end_mask_0 = const()[name = tensor("op_23988_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23988_cast_fp16 = slice_by_index(begin = var_23988_begin_0, end = var_23988_end_0, end_mask = var_23988_end_mask_0, x = q_109_cast_fp16)[name = tensor("op_23988_cast_fp16")]; + tensor var_23992_begin_0 = const()[name = tensor("op_23992_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_23992_end_0 = const()[name = tensor("op_23992_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_23992_end_mask_0 = const()[name = tensor("op_23992_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23992_cast_fp16 = slice_by_index(begin = var_23992_begin_0, end = var_23992_end_0, end_mask = var_23992_end_mask_0, x = q_109_cast_fp16)[name = tensor("op_23992_cast_fp16")]; + tensor var_23996_begin_0 = const()[name = tensor("op_23996_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_23996_end_0 = const()[name = tensor("op_23996_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_23996_end_mask_0 = const()[name = tensor("op_23996_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23996_cast_fp16 = slice_by_index(begin = var_23996_begin_0, end = var_23996_end_0, end_mask = var_23996_end_mask_0, x = q_109_cast_fp16)[name = tensor("op_23996_cast_fp16")]; + tensor var_24000_begin_0 = const()[name = tensor("op_24000_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_24000_end_0 = const()[name = tensor("op_24000_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_24000_end_mask_0 = const()[name = tensor("op_24000_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24000_cast_fp16 = slice_by_index(begin = var_24000_begin_0, end = var_24000_end_0, end_mask = var_24000_end_mask_0, x = q_109_cast_fp16)[name = tensor("op_24000_cast_fp16")]; + tensor var_24004_begin_0 = const()[name = tensor("op_24004_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_24004_end_0 = const()[name = tensor("op_24004_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_24004_end_mask_0 = const()[name = tensor("op_24004_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24004_cast_fp16 = slice_by_index(begin = var_24004_begin_0, end = var_24004_end_0, end_mask = var_24004_end_mask_0, x = q_109_cast_fp16)[name = tensor("op_24004_cast_fp16")]; + tensor var_24008_begin_0 = const()[name = tensor("op_24008_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_24008_end_0 = const()[name = tensor("op_24008_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_24008_end_mask_0 = const()[name = tensor("op_24008_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24008_cast_fp16 = slice_by_index(begin = var_24008_begin_0, end = var_24008_end_0, end_mask = var_24008_end_mask_0, x = q_109_cast_fp16)[name = tensor("op_24008_cast_fp16")]; + tensor var_24012_begin_0 = const()[name = tensor("op_24012_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_24012_end_0 = const()[name = tensor("op_24012_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_24012_end_mask_0 = const()[name = tensor("op_24012_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24012_cast_fp16 = slice_by_index(begin = var_24012_begin_0, end = var_24012_end_0, end_mask = var_24012_end_mask_0, x = q_109_cast_fp16)[name = tensor("op_24012_cast_fp16")]; + tensor var_24016_begin_0 = const()[name = tensor("op_24016_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_24016_end_0 = const()[name = tensor("op_24016_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_24016_end_mask_0 = const()[name = tensor("op_24016_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24016_cast_fp16 = slice_by_index(begin = var_24016_begin_0, end = var_24016_end_0, end_mask = var_24016_end_mask_0, x = q_109_cast_fp16)[name = tensor("op_24016_cast_fp16")]; + tensor var_24020_begin_0 = const()[name = tensor("op_24020_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_24020_end_0 = const()[name = tensor("op_24020_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_24020_end_mask_0 = const()[name = tensor("op_24020_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24020_cast_fp16 = slice_by_index(begin = var_24020_begin_0, end = var_24020_end_0, end_mask = var_24020_end_mask_0, x = q_109_cast_fp16)[name = tensor("op_24020_cast_fp16")]; + tensor k_219_perm_0 = const()[name = tensor("k_219_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_24027_begin_0 = const()[name = tensor("op_24027_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_24027_end_0 = const()[name = tensor("op_24027_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_24027_end_mask_0 = const()[name = tensor("op_24027_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_219_cast_fp16 = transpose(perm = k_219_perm_0, x = k_217_cast_fp16)[name = tensor("transpose_85")]; + tensor var_24027_cast_fp16 = slice_by_index(begin = var_24027_begin_0, end = var_24027_end_0, end_mask = var_24027_end_mask_0, x = k_219_cast_fp16)[name = tensor("op_24027_cast_fp16")]; + tensor var_24031_begin_0 = const()[name = tensor("op_24031_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_24031_end_0 = const()[name = tensor("op_24031_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_24031_end_mask_0 = const()[name = tensor("op_24031_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24031_cast_fp16 = slice_by_index(begin = var_24031_begin_0, end = var_24031_end_0, end_mask = var_24031_end_mask_0, x = k_219_cast_fp16)[name = tensor("op_24031_cast_fp16")]; + tensor var_24035_begin_0 = const()[name = tensor("op_24035_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_24035_end_0 = const()[name = tensor("op_24035_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_24035_end_mask_0 = const()[name = tensor("op_24035_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24035_cast_fp16 = slice_by_index(begin = var_24035_begin_0, end = var_24035_end_0, end_mask = var_24035_end_mask_0, x = k_219_cast_fp16)[name = tensor("op_24035_cast_fp16")]; + tensor var_24039_begin_0 = const()[name = tensor("op_24039_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_24039_end_0 = const()[name = tensor("op_24039_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_24039_end_mask_0 = const()[name = tensor("op_24039_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24039_cast_fp16 = slice_by_index(begin = var_24039_begin_0, end = var_24039_end_0, end_mask = var_24039_end_mask_0, x = k_219_cast_fp16)[name = tensor("op_24039_cast_fp16")]; + tensor var_24043_begin_0 = const()[name = tensor("op_24043_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_24043_end_0 = const()[name = tensor("op_24043_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_24043_end_mask_0 = const()[name = tensor("op_24043_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24043_cast_fp16 = slice_by_index(begin = var_24043_begin_0, end = var_24043_end_0, end_mask = var_24043_end_mask_0, x = k_219_cast_fp16)[name = tensor("op_24043_cast_fp16")]; + tensor var_24047_begin_0 = const()[name = tensor("op_24047_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_24047_end_0 = const()[name = tensor("op_24047_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_24047_end_mask_0 = const()[name = tensor("op_24047_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24047_cast_fp16 = slice_by_index(begin = var_24047_begin_0, end = var_24047_end_0, end_mask = var_24047_end_mask_0, x = k_219_cast_fp16)[name = tensor("op_24047_cast_fp16")]; + tensor var_24051_begin_0 = const()[name = tensor("op_24051_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_24051_end_0 = const()[name = tensor("op_24051_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_24051_end_mask_0 = const()[name = tensor("op_24051_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24051_cast_fp16 = slice_by_index(begin = var_24051_begin_0, end = var_24051_end_0, end_mask = var_24051_end_mask_0, x = k_219_cast_fp16)[name = tensor("op_24051_cast_fp16")]; + tensor var_24055_begin_0 = const()[name = tensor("op_24055_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_24055_end_0 = const()[name = tensor("op_24055_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_24055_end_mask_0 = const()[name = tensor("op_24055_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24055_cast_fp16 = slice_by_index(begin = var_24055_begin_0, end = var_24055_end_0, end_mask = var_24055_end_mask_0, x = k_219_cast_fp16)[name = tensor("op_24055_cast_fp16")]; + tensor var_24059_begin_0 = const()[name = tensor("op_24059_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_24059_end_0 = const()[name = tensor("op_24059_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_24059_end_mask_0 = const()[name = tensor("op_24059_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24059_cast_fp16 = slice_by_index(begin = var_24059_begin_0, end = var_24059_end_0, end_mask = var_24059_end_mask_0, x = k_219_cast_fp16)[name = tensor("op_24059_cast_fp16")]; + tensor var_24063_begin_0 = const()[name = tensor("op_24063_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_24063_end_0 = const()[name = tensor("op_24063_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_24063_end_mask_0 = const()[name = tensor("op_24063_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24063_cast_fp16 = slice_by_index(begin = var_24063_begin_0, end = var_24063_end_0, end_mask = var_24063_end_mask_0, x = k_219_cast_fp16)[name = tensor("op_24063_cast_fp16")]; + tensor var_24067_begin_0 = const()[name = tensor("op_24067_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_24067_end_0 = const()[name = tensor("op_24067_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_24067_end_mask_0 = const()[name = tensor("op_24067_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24067_cast_fp16 = slice_by_index(begin = var_24067_begin_0, end = var_24067_end_0, end_mask = var_24067_end_mask_0, x = k_219_cast_fp16)[name = tensor("op_24067_cast_fp16")]; + tensor var_24071_begin_0 = const()[name = tensor("op_24071_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_24071_end_0 = const()[name = tensor("op_24071_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_24071_end_mask_0 = const()[name = tensor("op_24071_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24071_cast_fp16 = slice_by_index(begin = var_24071_begin_0, end = var_24071_end_0, end_mask = var_24071_end_mask_0, x = k_219_cast_fp16)[name = tensor("op_24071_cast_fp16")]; + tensor var_24075_begin_0 = const()[name = tensor("op_24075_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_24075_end_0 = const()[name = tensor("op_24075_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_24075_end_mask_0 = const()[name = tensor("op_24075_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24075_cast_fp16 = slice_by_index(begin = var_24075_begin_0, end = var_24075_end_0, end_mask = var_24075_end_mask_0, x = k_219_cast_fp16)[name = tensor("op_24075_cast_fp16")]; + tensor var_24079_begin_0 = const()[name = tensor("op_24079_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_24079_end_0 = const()[name = tensor("op_24079_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_24079_end_mask_0 = const()[name = tensor("op_24079_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24079_cast_fp16 = slice_by_index(begin = var_24079_begin_0, end = var_24079_end_0, end_mask = var_24079_end_mask_0, x = k_219_cast_fp16)[name = tensor("op_24079_cast_fp16")]; + tensor var_24083_begin_0 = const()[name = tensor("op_24083_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_24083_end_0 = const()[name = tensor("op_24083_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_24083_end_mask_0 = const()[name = tensor("op_24083_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24083_cast_fp16 = slice_by_index(begin = var_24083_begin_0, end = var_24083_end_0, end_mask = var_24083_end_mask_0, x = k_219_cast_fp16)[name = tensor("op_24083_cast_fp16")]; + tensor var_24087_begin_0 = const()[name = tensor("op_24087_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_24087_end_0 = const()[name = tensor("op_24087_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_24087_end_mask_0 = const()[name = tensor("op_24087_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24087_cast_fp16 = slice_by_index(begin = var_24087_begin_0, end = var_24087_end_0, end_mask = var_24087_end_mask_0, x = k_219_cast_fp16)[name = tensor("op_24087_cast_fp16")]; + tensor var_24091_begin_0 = const()[name = tensor("op_24091_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_24091_end_0 = const()[name = tensor("op_24091_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_24091_end_mask_0 = const()[name = tensor("op_24091_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24091_cast_fp16 = slice_by_index(begin = var_24091_begin_0, end = var_24091_end_0, end_mask = var_24091_end_mask_0, x = k_219_cast_fp16)[name = tensor("op_24091_cast_fp16")]; + tensor var_24095_begin_0 = const()[name = tensor("op_24095_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_24095_end_0 = const()[name = tensor("op_24095_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_24095_end_mask_0 = const()[name = tensor("op_24095_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24095_cast_fp16 = slice_by_index(begin = var_24095_begin_0, end = var_24095_end_0, end_mask = var_24095_end_mask_0, x = k_219_cast_fp16)[name = tensor("op_24095_cast_fp16")]; + tensor var_24099_begin_0 = const()[name = tensor("op_24099_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_24099_end_0 = const()[name = tensor("op_24099_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_24099_end_mask_0 = const()[name = tensor("op_24099_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24099_cast_fp16 = slice_by_index(begin = var_24099_begin_0, end = var_24099_end_0, end_mask = var_24099_end_mask_0, x = k_219_cast_fp16)[name = tensor("op_24099_cast_fp16")]; + tensor var_24103_begin_0 = const()[name = tensor("op_24103_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_24103_end_0 = const()[name = tensor("op_24103_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_24103_end_mask_0 = const()[name = tensor("op_24103_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24103_cast_fp16 = slice_by_index(begin = var_24103_begin_0, end = var_24103_end_0, end_mask = var_24103_end_mask_0, x = k_219_cast_fp16)[name = tensor("op_24103_cast_fp16")]; + tensor var_24105_begin_0 = const()[name = tensor("op_24105_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_24105_end_0 = const()[name = tensor("op_24105_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_24105_end_mask_0 = const()[name = tensor("op_24105_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24105_cast_fp16 = slice_by_index(begin = var_24105_begin_0, end = var_24105_end_0, end_mask = var_24105_end_mask_0, x = v_109_cast_fp16)[name = tensor("op_24105_cast_fp16")]; + tensor var_24109_begin_0 = const()[name = tensor("op_24109_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_24109_end_0 = const()[name = tensor("op_24109_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_24109_end_mask_0 = const()[name = tensor("op_24109_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24109_cast_fp16 = slice_by_index(begin = var_24109_begin_0, end = var_24109_end_0, end_mask = var_24109_end_mask_0, x = v_109_cast_fp16)[name = tensor("op_24109_cast_fp16")]; + tensor var_24113_begin_0 = const()[name = tensor("op_24113_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_24113_end_0 = const()[name = tensor("op_24113_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_24113_end_mask_0 = const()[name = tensor("op_24113_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24113_cast_fp16 = slice_by_index(begin = var_24113_begin_0, end = var_24113_end_0, end_mask = var_24113_end_mask_0, x = v_109_cast_fp16)[name = tensor("op_24113_cast_fp16")]; + tensor var_24117_begin_0 = const()[name = tensor("op_24117_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_24117_end_0 = const()[name = tensor("op_24117_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_24117_end_mask_0 = const()[name = tensor("op_24117_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24117_cast_fp16 = slice_by_index(begin = var_24117_begin_0, end = var_24117_end_0, end_mask = var_24117_end_mask_0, x = v_109_cast_fp16)[name = tensor("op_24117_cast_fp16")]; + tensor var_24121_begin_0 = const()[name = tensor("op_24121_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_24121_end_0 = const()[name = tensor("op_24121_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_24121_end_mask_0 = const()[name = tensor("op_24121_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24121_cast_fp16 = slice_by_index(begin = var_24121_begin_0, end = var_24121_end_0, end_mask = var_24121_end_mask_0, x = v_109_cast_fp16)[name = tensor("op_24121_cast_fp16")]; + tensor var_24125_begin_0 = const()[name = tensor("op_24125_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_24125_end_0 = const()[name = tensor("op_24125_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_24125_end_mask_0 = const()[name = tensor("op_24125_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24125_cast_fp16 = slice_by_index(begin = var_24125_begin_0, end = var_24125_end_0, end_mask = var_24125_end_mask_0, x = v_109_cast_fp16)[name = tensor("op_24125_cast_fp16")]; + tensor var_24129_begin_0 = const()[name = tensor("op_24129_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_24129_end_0 = const()[name = tensor("op_24129_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_24129_end_mask_0 = const()[name = tensor("op_24129_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24129_cast_fp16 = slice_by_index(begin = var_24129_begin_0, end = var_24129_end_0, end_mask = var_24129_end_mask_0, x = v_109_cast_fp16)[name = tensor("op_24129_cast_fp16")]; + tensor var_24133_begin_0 = const()[name = tensor("op_24133_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_24133_end_0 = const()[name = tensor("op_24133_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_24133_end_mask_0 = const()[name = tensor("op_24133_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24133_cast_fp16 = slice_by_index(begin = var_24133_begin_0, end = var_24133_end_0, end_mask = var_24133_end_mask_0, x = v_109_cast_fp16)[name = tensor("op_24133_cast_fp16")]; + tensor var_24137_begin_0 = const()[name = tensor("op_24137_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_24137_end_0 = const()[name = tensor("op_24137_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_24137_end_mask_0 = const()[name = tensor("op_24137_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24137_cast_fp16 = slice_by_index(begin = var_24137_begin_0, end = var_24137_end_0, end_mask = var_24137_end_mask_0, x = v_109_cast_fp16)[name = tensor("op_24137_cast_fp16")]; + tensor var_24141_begin_0 = const()[name = tensor("op_24141_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_24141_end_0 = const()[name = tensor("op_24141_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_24141_end_mask_0 = const()[name = tensor("op_24141_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24141_cast_fp16 = slice_by_index(begin = var_24141_begin_0, end = var_24141_end_0, end_mask = var_24141_end_mask_0, x = v_109_cast_fp16)[name = tensor("op_24141_cast_fp16")]; + tensor var_24145_begin_0 = const()[name = tensor("op_24145_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_24145_end_0 = const()[name = tensor("op_24145_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_24145_end_mask_0 = const()[name = tensor("op_24145_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24145_cast_fp16 = slice_by_index(begin = var_24145_begin_0, end = var_24145_end_0, end_mask = var_24145_end_mask_0, x = v_109_cast_fp16)[name = tensor("op_24145_cast_fp16")]; + tensor var_24149_begin_0 = const()[name = tensor("op_24149_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_24149_end_0 = const()[name = tensor("op_24149_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_24149_end_mask_0 = const()[name = tensor("op_24149_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24149_cast_fp16 = slice_by_index(begin = var_24149_begin_0, end = var_24149_end_0, end_mask = var_24149_end_mask_0, x = v_109_cast_fp16)[name = tensor("op_24149_cast_fp16")]; + tensor var_24153_begin_0 = const()[name = tensor("op_24153_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_24153_end_0 = const()[name = tensor("op_24153_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_24153_end_mask_0 = const()[name = tensor("op_24153_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24153_cast_fp16 = slice_by_index(begin = var_24153_begin_0, end = var_24153_end_0, end_mask = var_24153_end_mask_0, x = v_109_cast_fp16)[name = tensor("op_24153_cast_fp16")]; + tensor var_24157_begin_0 = const()[name = tensor("op_24157_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_24157_end_0 = const()[name = tensor("op_24157_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_24157_end_mask_0 = const()[name = tensor("op_24157_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24157_cast_fp16 = slice_by_index(begin = var_24157_begin_0, end = var_24157_end_0, end_mask = var_24157_end_mask_0, x = v_109_cast_fp16)[name = tensor("op_24157_cast_fp16")]; + tensor var_24161_begin_0 = const()[name = tensor("op_24161_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_24161_end_0 = const()[name = tensor("op_24161_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_24161_end_mask_0 = const()[name = tensor("op_24161_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24161_cast_fp16 = slice_by_index(begin = var_24161_begin_0, end = var_24161_end_0, end_mask = var_24161_end_mask_0, x = v_109_cast_fp16)[name = tensor("op_24161_cast_fp16")]; + tensor var_24165_begin_0 = const()[name = tensor("op_24165_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_24165_end_0 = const()[name = tensor("op_24165_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_24165_end_mask_0 = const()[name = tensor("op_24165_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24165_cast_fp16 = slice_by_index(begin = var_24165_begin_0, end = var_24165_end_0, end_mask = var_24165_end_mask_0, x = v_109_cast_fp16)[name = tensor("op_24165_cast_fp16")]; + tensor var_24169_begin_0 = const()[name = tensor("op_24169_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_24169_end_0 = const()[name = tensor("op_24169_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_24169_end_mask_0 = const()[name = tensor("op_24169_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24169_cast_fp16 = slice_by_index(begin = var_24169_begin_0, end = var_24169_end_0, end_mask = var_24169_end_mask_0, x = v_109_cast_fp16)[name = tensor("op_24169_cast_fp16")]; + tensor var_24173_begin_0 = const()[name = tensor("op_24173_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_24173_end_0 = const()[name = tensor("op_24173_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_24173_end_mask_0 = const()[name = tensor("op_24173_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24173_cast_fp16 = slice_by_index(begin = var_24173_begin_0, end = var_24173_end_0, end_mask = var_24173_end_mask_0, x = v_109_cast_fp16)[name = tensor("op_24173_cast_fp16")]; + tensor var_24177_begin_0 = const()[name = tensor("op_24177_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_24177_end_0 = const()[name = tensor("op_24177_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_24177_end_mask_0 = const()[name = tensor("op_24177_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24177_cast_fp16 = slice_by_index(begin = var_24177_begin_0, end = var_24177_end_0, end_mask = var_24177_end_mask_0, x = v_109_cast_fp16)[name = tensor("op_24177_cast_fp16")]; + tensor var_24181_begin_0 = const()[name = tensor("op_24181_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_24181_end_0 = const()[name = tensor("op_24181_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_24181_end_mask_0 = const()[name = tensor("op_24181_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24181_cast_fp16 = slice_by_index(begin = var_24181_begin_0, end = var_24181_end_0, end_mask = var_24181_end_mask_0, x = v_109_cast_fp16)[name = tensor("op_24181_cast_fp16")]; + tensor var_24185_equation_0 = const()[name = tensor("op_24185_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24185_cast_fp16 = einsum(equation = var_24185_equation_0, values = (var_24027_cast_fp16, var_23944_cast_fp16))[name = tensor("op_24185_cast_fp16")]; + tensor var_24186_to_fp16 = const()[name = tensor("op_24186_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2001_cast_fp16 = mul(x = var_24185_cast_fp16, y = var_24186_to_fp16)[name = tensor("aw_2001_cast_fp16")]; + tensor var_24189_equation_0 = const()[name = tensor("op_24189_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24189_cast_fp16 = einsum(equation = var_24189_equation_0, values = (var_24031_cast_fp16, var_23948_cast_fp16))[name = tensor("op_24189_cast_fp16")]; + tensor var_24190_to_fp16 = const()[name = tensor("op_24190_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2003_cast_fp16 = mul(x = var_24189_cast_fp16, y = var_24190_to_fp16)[name = tensor("aw_2003_cast_fp16")]; + tensor var_24193_equation_0 = const()[name = tensor("op_24193_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24193_cast_fp16 = einsum(equation = var_24193_equation_0, values = (var_24035_cast_fp16, var_23952_cast_fp16))[name = tensor("op_24193_cast_fp16")]; + tensor var_24194_to_fp16 = const()[name = tensor("op_24194_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2005_cast_fp16 = mul(x = var_24193_cast_fp16, y = var_24194_to_fp16)[name = tensor("aw_2005_cast_fp16")]; + tensor var_24197_equation_0 = const()[name = tensor("op_24197_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24197_cast_fp16 = einsum(equation = var_24197_equation_0, values = (var_24039_cast_fp16, var_23956_cast_fp16))[name = tensor("op_24197_cast_fp16")]; + tensor var_24198_to_fp16 = const()[name = tensor("op_24198_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2007_cast_fp16 = mul(x = var_24197_cast_fp16, y = var_24198_to_fp16)[name = tensor("aw_2007_cast_fp16")]; + tensor var_24201_equation_0 = const()[name = tensor("op_24201_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24201_cast_fp16 = einsum(equation = var_24201_equation_0, values = (var_24043_cast_fp16, var_23960_cast_fp16))[name = tensor("op_24201_cast_fp16")]; + tensor var_24202_to_fp16 = const()[name = tensor("op_24202_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2009_cast_fp16 = mul(x = var_24201_cast_fp16, y = var_24202_to_fp16)[name = tensor("aw_2009_cast_fp16")]; + tensor var_24205_equation_0 = const()[name = tensor("op_24205_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24205_cast_fp16 = einsum(equation = var_24205_equation_0, values = (var_24047_cast_fp16, var_23964_cast_fp16))[name = tensor("op_24205_cast_fp16")]; + tensor var_24206_to_fp16 = const()[name = tensor("op_24206_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2011_cast_fp16 = mul(x = var_24205_cast_fp16, y = var_24206_to_fp16)[name = tensor("aw_2011_cast_fp16")]; + tensor var_24209_equation_0 = const()[name = tensor("op_24209_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24209_cast_fp16 = einsum(equation = var_24209_equation_0, values = (var_24051_cast_fp16, var_23968_cast_fp16))[name = tensor("op_24209_cast_fp16")]; + tensor var_24210_to_fp16 = const()[name = tensor("op_24210_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2013_cast_fp16 = mul(x = var_24209_cast_fp16, y = var_24210_to_fp16)[name = tensor("aw_2013_cast_fp16")]; + tensor var_24213_equation_0 = const()[name = tensor("op_24213_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24213_cast_fp16 = einsum(equation = var_24213_equation_0, values = (var_24055_cast_fp16, var_23972_cast_fp16))[name = tensor("op_24213_cast_fp16")]; + tensor var_24214_to_fp16 = const()[name = tensor("op_24214_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2015_cast_fp16 = mul(x = var_24213_cast_fp16, y = var_24214_to_fp16)[name = tensor("aw_2015_cast_fp16")]; + tensor var_24217_equation_0 = const()[name = tensor("op_24217_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24217_cast_fp16 = einsum(equation = var_24217_equation_0, values = (var_24059_cast_fp16, var_23976_cast_fp16))[name = tensor("op_24217_cast_fp16")]; + tensor var_24218_to_fp16 = const()[name = tensor("op_24218_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2017_cast_fp16 = mul(x = var_24217_cast_fp16, y = var_24218_to_fp16)[name = tensor("aw_2017_cast_fp16")]; + tensor var_24221_equation_0 = const()[name = tensor("op_24221_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24221_cast_fp16 = einsum(equation = var_24221_equation_0, values = (var_24063_cast_fp16, var_23980_cast_fp16))[name = tensor("op_24221_cast_fp16")]; + tensor var_24222_to_fp16 = const()[name = tensor("op_24222_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2019_cast_fp16 = mul(x = var_24221_cast_fp16, y = var_24222_to_fp16)[name = tensor("aw_2019_cast_fp16")]; + tensor var_24225_equation_0 = const()[name = tensor("op_24225_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24225_cast_fp16 = einsum(equation = var_24225_equation_0, values = (var_24067_cast_fp16, var_23984_cast_fp16))[name = tensor("op_24225_cast_fp16")]; + tensor var_24226_to_fp16 = const()[name = tensor("op_24226_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2021_cast_fp16 = mul(x = var_24225_cast_fp16, y = var_24226_to_fp16)[name = tensor("aw_2021_cast_fp16")]; + tensor var_24229_equation_0 = const()[name = tensor("op_24229_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24229_cast_fp16 = einsum(equation = var_24229_equation_0, values = (var_24071_cast_fp16, var_23988_cast_fp16))[name = tensor("op_24229_cast_fp16")]; + tensor var_24230_to_fp16 = const()[name = tensor("op_24230_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2023_cast_fp16 = mul(x = var_24229_cast_fp16, y = var_24230_to_fp16)[name = tensor("aw_2023_cast_fp16")]; + tensor var_24233_equation_0 = const()[name = tensor("op_24233_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24233_cast_fp16 = einsum(equation = var_24233_equation_0, values = (var_24075_cast_fp16, var_23992_cast_fp16))[name = tensor("op_24233_cast_fp16")]; + tensor var_24234_to_fp16 = const()[name = tensor("op_24234_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2025_cast_fp16 = mul(x = var_24233_cast_fp16, y = var_24234_to_fp16)[name = tensor("aw_2025_cast_fp16")]; + tensor var_24237_equation_0 = const()[name = tensor("op_24237_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24237_cast_fp16 = einsum(equation = var_24237_equation_0, values = (var_24079_cast_fp16, var_23996_cast_fp16))[name = tensor("op_24237_cast_fp16")]; + tensor var_24238_to_fp16 = const()[name = tensor("op_24238_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2027_cast_fp16 = mul(x = var_24237_cast_fp16, y = var_24238_to_fp16)[name = tensor("aw_2027_cast_fp16")]; + tensor var_24241_equation_0 = const()[name = tensor("op_24241_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24241_cast_fp16 = einsum(equation = var_24241_equation_0, values = (var_24083_cast_fp16, var_24000_cast_fp16))[name = tensor("op_24241_cast_fp16")]; + tensor var_24242_to_fp16 = const()[name = tensor("op_24242_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2029_cast_fp16 = mul(x = var_24241_cast_fp16, y = var_24242_to_fp16)[name = tensor("aw_2029_cast_fp16")]; + tensor var_24245_equation_0 = const()[name = tensor("op_24245_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24245_cast_fp16 = einsum(equation = var_24245_equation_0, values = (var_24087_cast_fp16, var_24004_cast_fp16))[name = tensor("op_24245_cast_fp16")]; + tensor var_24246_to_fp16 = const()[name = tensor("op_24246_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2031_cast_fp16 = mul(x = var_24245_cast_fp16, y = var_24246_to_fp16)[name = tensor("aw_2031_cast_fp16")]; + tensor var_24249_equation_0 = const()[name = tensor("op_24249_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24249_cast_fp16 = einsum(equation = var_24249_equation_0, values = (var_24091_cast_fp16, var_24008_cast_fp16))[name = tensor("op_24249_cast_fp16")]; + tensor var_24250_to_fp16 = const()[name = tensor("op_24250_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2033_cast_fp16 = mul(x = var_24249_cast_fp16, y = var_24250_to_fp16)[name = tensor("aw_2033_cast_fp16")]; + tensor var_24253_equation_0 = const()[name = tensor("op_24253_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24253_cast_fp16 = einsum(equation = var_24253_equation_0, values = (var_24095_cast_fp16, var_24012_cast_fp16))[name = tensor("op_24253_cast_fp16")]; + tensor var_24254_to_fp16 = const()[name = tensor("op_24254_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2035_cast_fp16 = mul(x = var_24253_cast_fp16, y = var_24254_to_fp16)[name = tensor("aw_2035_cast_fp16")]; + tensor var_24257_equation_0 = const()[name = tensor("op_24257_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24257_cast_fp16 = einsum(equation = var_24257_equation_0, values = (var_24099_cast_fp16, var_24016_cast_fp16))[name = tensor("op_24257_cast_fp16")]; + tensor var_24258_to_fp16 = const()[name = tensor("op_24258_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2037_cast_fp16 = mul(x = var_24257_cast_fp16, y = var_24258_to_fp16)[name = tensor("aw_2037_cast_fp16")]; + tensor var_24261_equation_0 = const()[name = tensor("op_24261_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24261_cast_fp16 = einsum(equation = var_24261_equation_0, values = (var_24103_cast_fp16, var_24020_cast_fp16))[name = tensor("op_24261_cast_fp16")]; + tensor var_24262_to_fp16 = const()[name = tensor("op_24262_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2039_cast_fp16 = mul(x = var_24261_cast_fp16, y = var_24262_to_fp16)[name = tensor("aw_2039_cast_fp16")]; + tensor var_24264_cast_fp16 = softmax(axis = var_21077, x = aw_2001_cast_fp16)[name = tensor("op_24264_cast_fp16")]; + tensor var_24265_cast_fp16 = softmax(axis = var_21077, x = aw_2003_cast_fp16)[name = tensor("op_24265_cast_fp16")]; + tensor var_24266_cast_fp16 = softmax(axis = var_21077, x = aw_2005_cast_fp16)[name = tensor("op_24266_cast_fp16")]; + tensor var_24267_cast_fp16 = softmax(axis = var_21077, x = aw_2007_cast_fp16)[name = tensor("op_24267_cast_fp16")]; + tensor var_24268_cast_fp16 = softmax(axis = var_21077, x = aw_2009_cast_fp16)[name = tensor("op_24268_cast_fp16")]; + tensor var_24269_cast_fp16 = softmax(axis = var_21077, x = aw_2011_cast_fp16)[name = tensor("op_24269_cast_fp16")]; + tensor var_24270_cast_fp16 = softmax(axis = var_21077, x = aw_2013_cast_fp16)[name = tensor("op_24270_cast_fp16")]; + tensor var_24271_cast_fp16 = softmax(axis = var_21077, x = aw_2015_cast_fp16)[name = tensor("op_24271_cast_fp16")]; + tensor var_24272_cast_fp16 = softmax(axis = var_21077, x = aw_2017_cast_fp16)[name = tensor("op_24272_cast_fp16")]; + tensor var_24273_cast_fp16 = softmax(axis = var_21077, x = aw_2019_cast_fp16)[name = tensor("op_24273_cast_fp16")]; + tensor var_24274_cast_fp16 = softmax(axis = var_21077, x = aw_2021_cast_fp16)[name = tensor("op_24274_cast_fp16")]; + tensor var_24275_cast_fp16 = softmax(axis = var_21077, x = aw_2023_cast_fp16)[name = tensor("op_24275_cast_fp16")]; + tensor var_24276_cast_fp16 = softmax(axis = var_21077, x = aw_2025_cast_fp16)[name = tensor("op_24276_cast_fp16")]; + tensor var_24277_cast_fp16 = softmax(axis = var_21077, x = aw_2027_cast_fp16)[name = tensor("op_24277_cast_fp16")]; + tensor var_24278_cast_fp16 = softmax(axis = var_21077, x = aw_2029_cast_fp16)[name = tensor("op_24278_cast_fp16")]; + tensor var_24279_cast_fp16 = softmax(axis = var_21077, x = aw_2031_cast_fp16)[name = tensor("op_24279_cast_fp16")]; + tensor var_24280_cast_fp16 = softmax(axis = var_21077, x = aw_2033_cast_fp16)[name = tensor("op_24280_cast_fp16")]; + tensor var_24281_cast_fp16 = softmax(axis = var_21077, x = aw_2035_cast_fp16)[name = tensor("op_24281_cast_fp16")]; + tensor var_24282_cast_fp16 = softmax(axis = var_21077, x = aw_2037_cast_fp16)[name = tensor("op_24282_cast_fp16")]; + tensor var_24283_cast_fp16 = softmax(axis = var_21077, x = aw_2039_cast_fp16)[name = tensor("op_24283_cast_fp16")]; + tensor var_24285_equation_0 = const()[name = tensor("op_24285_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24285_cast_fp16 = einsum(equation = var_24285_equation_0, values = (var_24105_cast_fp16, var_24264_cast_fp16))[name = tensor("op_24285_cast_fp16")]; + tensor var_24287_equation_0 = const()[name = tensor("op_24287_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24287_cast_fp16 = einsum(equation = var_24287_equation_0, values = (var_24109_cast_fp16, var_24265_cast_fp16))[name = tensor("op_24287_cast_fp16")]; + tensor var_24289_equation_0 = const()[name = tensor("op_24289_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24289_cast_fp16 = einsum(equation = var_24289_equation_0, values = (var_24113_cast_fp16, var_24266_cast_fp16))[name = tensor("op_24289_cast_fp16")]; + tensor var_24291_equation_0 = const()[name = tensor("op_24291_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24291_cast_fp16 = einsum(equation = var_24291_equation_0, values = (var_24117_cast_fp16, var_24267_cast_fp16))[name = tensor("op_24291_cast_fp16")]; + tensor var_24293_equation_0 = const()[name = tensor("op_24293_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24293_cast_fp16 = einsum(equation = var_24293_equation_0, values = (var_24121_cast_fp16, var_24268_cast_fp16))[name = tensor("op_24293_cast_fp16")]; + tensor var_24295_equation_0 = const()[name = tensor("op_24295_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24295_cast_fp16 = einsum(equation = var_24295_equation_0, values = (var_24125_cast_fp16, var_24269_cast_fp16))[name = tensor("op_24295_cast_fp16")]; + tensor var_24297_equation_0 = const()[name = tensor("op_24297_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24297_cast_fp16 = einsum(equation = var_24297_equation_0, values = (var_24129_cast_fp16, var_24270_cast_fp16))[name = tensor("op_24297_cast_fp16")]; + tensor var_24299_equation_0 = const()[name = tensor("op_24299_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24299_cast_fp16 = einsum(equation = var_24299_equation_0, values = (var_24133_cast_fp16, var_24271_cast_fp16))[name = tensor("op_24299_cast_fp16")]; + tensor var_24301_equation_0 = const()[name = tensor("op_24301_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24301_cast_fp16 = einsum(equation = var_24301_equation_0, values = (var_24137_cast_fp16, var_24272_cast_fp16))[name = tensor("op_24301_cast_fp16")]; + tensor var_24303_equation_0 = const()[name = tensor("op_24303_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24303_cast_fp16 = einsum(equation = var_24303_equation_0, values = (var_24141_cast_fp16, var_24273_cast_fp16))[name = tensor("op_24303_cast_fp16")]; + tensor var_24305_equation_0 = const()[name = tensor("op_24305_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24305_cast_fp16 = einsum(equation = var_24305_equation_0, values = (var_24145_cast_fp16, var_24274_cast_fp16))[name = tensor("op_24305_cast_fp16")]; + tensor var_24307_equation_0 = const()[name = tensor("op_24307_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24307_cast_fp16 = einsum(equation = var_24307_equation_0, values = (var_24149_cast_fp16, var_24275_cast_fp16))[name = tensor("op_24307_cast_fp16")]; + tensor var_24309_equation_0 = const()[name = tensor("op_24309_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24309_cast_fp16 = einsum(equation = var_24309_equation_0, values = (var_24153_cast_fp16, var_24276_cast_fp16))[name = tensor("op_24309_cast_fp16")]; + tensor var_24311_equation_0 = const()[name = tensor("op_24311_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24311_cast_fp16 = einsum(equation = var_24311_equation_0, values = (var_24157_cast_fp16, var_24277_cast_fp16))[name = tensor("op_24311_cast_fp16")]; + tensor var_24313_equation_0 = const()[name = tensor("op_24313_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24313_cast_fp16 = einsum(equation = var_24313_equation_0, values = (var_24161_cast_fp16, var_24278_cast_fp16))[name = tensor("op_24313_cast_fp16")]; + tensor var_24315_equation_0 = const()[name = tensor("op_24315_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24315_cast_fp16 = einsum(equation = var_24315_equation_0, values = (var_24165_cast_fp16, var_24279_cast_fp16))[name = tensor("op_24315_cast_fp16")]; + tensor var_24317_equation_0 = const()[name = tensor("op_24317_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24317_cast_fp16 = einsum(equation = var_24317_equation_0, values = (var_24169_cast_fp16, var_24280_cast_fp16))[name = tensor("op_24317_cast_fp16")]; + tensor var_24319_equation_0 = const()[name = tensor("op_24319_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24319_cast_fp16 = einsum(equation = var_24319_equation_0, values = (var_24173_cast_fp16, var_24281_cast_fp16))[name = tensor("op_24319_cast_fp16")]; + tensor var_24321_equation_0 = const()[name = tensor("op_24321_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24321_cast_fp16 = einsum(equation = var_24321_equation_0, values = (var_24177_cast_fp16, var_24282_cast_fp16))[name = tensor("op_24321_cast_fp16")]; + tensor var_24323_equation_0 = const()[name = tensor("op_24323_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24323_cast_fp16 = einsum(equation = var_24323_equation_0, values = (var_24181_cast_fp16, var_24283_cast_fp16))[name = tensor("op_24323_cast_fp16")]; + tensor input_351_interleave_0 = const()[name = tensor("input_351_interleave_0"), val = tensor(false)]; + tensor input_351_cast_fp16 = concat(axis = var_21077, interleave = input_351_interleave_0, values = (var_24285_cast_fp16, var_24287_cast_fp16, var_24289_cast_fp16, var_24291_cast_fp16, var_24293_cast_fp16, var_24295_cast_fp16, var_24297_cast_fp16, var_24299_cast_fp16, var_24301_cast_fp16, var_24303_cast_fp16, var_24305_cast_fp16, var_24307_cast_fp16, var_24309_cast_fp16, var_24311_cast_fp16, var_24313_cast_fp16, var_24315_cast_fp16, var_24317_cast_fp16, var_24319_cast_fp16, var_24321_cast_fp16, var_24323_cast_fp16))[name = tensor("input_351_cast_fp16")]; + tensor var_24329 = const()[name = tensor("op_24329"), val = tensor([1, 1])]; + tensor var_24331 = const()[name = tensor("op_24331"), val = tensor([1, 1])]; + tensor var_24333_pad_type_0 = const()[name = tensor("op_24333_pad_type_0"), val = tensor("custom")]; + tensor var_24333_pad_0 = const()[name = tensor("op_24333_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_3_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_3_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1944483776)))]; + tensor mid_block_attentions_0_transformer_blocks_3_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_3_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1947760640)))]; + tensor var_24333_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_3_attn1_to_out_0_bias_to_fp16, dilations = var_24331, groups = var_21077, pad = var_24333_pad_0, pad_type = var_24333_pad_type_0, strides = var_24329, weight = mid_block_attentions_0_transformer_blocks_3_attn1_to_out_0_weight_to_fp16, x = input_351_cast_fp16)[name = tensor("op_24333_cast_fp16")]; + tensor inputs_165_cast_fp16 = add(x = var_24333_cast_fp16, y = inputs_163_cast_fp16)[name = tensor("inputs_165_cast_fp16")]; + tensor hidden_states_229_axes_0 = const()[name = tensor("hidden_states_229_axes_0"), val = tensor([1])]; + tensor hidden_states_229_gamma_0_to_fp16 = const()[name = tensor("hidden_states_229_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1947763264)))]; + tensor hidden_states_229_beta_0_to_fp16 = const()[name = tensor("hidden_states_229_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1947765888)))]; + tensor var_24343_to_fp16 = const()[name = tensor("op_24343_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_229_cast_fp16 = layer_norm(axes = hidden_states_229_axes_0, beta = hidden_states_229_beta_0_to_fp16, epsilon = var_24343_to_fp16, gamma = hidden_states_229_gamma_0_to_fp16, x = inputs_165_cast_fp16)[name = tensor("hidden_states_229_cast_fp16")]; + tensor var_24358 = const()[name = tensor("op_24358"), val = tensor([1, 1])]; + tensor var_24360 = const()[name = tensor("op_24360"), val = tensor([1, 1])]; + tensor q_111_pad_type_0 = const()[name = tensor("q_111_pad_type_0"), val = tensor("custom")]; + tensor q_111_pad_0 = const()[name = tensor("q_111_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_3_attn2_to_q_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_3_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1947768512)))]; + tensor q_111_cast_fp16 = conv(dilations = var_24360, groups = var_21077, pad = q_111_pad_0, pad_type = q_111_pad_type_0, strides = var_24358, weight = mid_block_attentions_0_transformer_blocks_3_attn2_to_q_weight_to_fp16, x = hidden_states_229_cast_fp16)[name = tensor("q_111_cast_fp16")]; + tensor var_24364 = const()[name = tensor("op_24364"), val = tensor([1, 1])]; + tensor var_24366 = const()[name = tensor("op_24366"), val = tensor([1, 1])]; + tensor k_221_pad_type_0 = const()[name = tensor("k_221_pad_type_0"), val = tensor("custom")]; + tensor k_221_pad_0 = const()[name = tensor("k_221_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_3_attn2_to_k_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_3_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1951045376)))]; + tensor k_221_cast_fp16 = conv(dilations = var_24366, groups = var_21077, pad = k_221_pad_0, pad_type = k_221_pad_type_0, strides = var_24364, weight = mid_block_attentions_0_transformer_blocks_3_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_221_cast_fp16")]; + tensor var_24370 = const()[name = tensor("op_24370"), val = tensor([1, 1])]; + tensor var_24372 = const()[name = tensor("op_24372"), val = tensor([1, 1])]; + tensor v_111_pad_type_0 = const()[name = tensor("v_111_pad_type_0"), val = tensor("custom")]; + tensor v_111_pad_0 = const()[name = tensor("v_111_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_3_attn2_to_v_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_3_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1956288320)))]; + tensor v_111_cast_fp16 = conv(dilations = var_24372, groups = var_21077, pad = v_111_pad_0, pad_type = v_111_pad_type_0, strides = var_24370, weight = mid_block_attentions_0_transformer_blocks_3_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_111_cast_fp16")]; + tensor var_24376_begin_0 = const()[name = tensor("op_24376_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_24376_end_0 = const()[name = tensor("op_24376_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_24376_end_mask_0 = const()[name = tensor("op_24376_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24376_cast_fp16 = slice_by_index(begin = var_24376_begin_0, end = var_24376_end_0, end_mask = var_24376_end_mask_0, x = q_111_cast_fp16)[name = tensor("op_24376_cast_fp16")]; + tensor var_24380_begin_0 = const()[name = tensor("op_24380_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_24380_end_0 = const()[name = tensor("op_24380_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_24380_end_mask_0 = const()[name = tensor("op_24380_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24380_cast_fp16 = slice_by_index(begin = var_24380_begin_0, end = var_24380_end_0, end_mask = var_24380_end_mask_0, x = q_111_cast_fp16)[name = tensor("op_24380_cast_fp16")]; + tensor var_24384_begin_0 = const()[name = tensor("op_24384_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_24384_end_0 = const()[name = tensor("op_24384_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_24384_end_mask_0 = const()[name = tensor("op_24384_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24384_cast_fp16 = slice_by_index(begin = var_24384_begin_0, end = var_24384_end_0, end_mask = var_24384_end_mask_0, x = q_111_cast_fp16)[name = tensor("op_24384_cast_fp16")]; + tensor var_24388_begin_0 = const()[name = tensor("op_24388_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_24388_end_0 = const()[name = tensor("op_24388_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_24388_end_mask_0 = const()[name = tensor("op_24388_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24388_cast_fp16 = slice_by_index(begin = var_24388_begin_0, end = var_24388_end_0, end_mask = var_24388_end_mask_0, x = q_111_cast_fp16)[name = tensor("op_24388_cast_fp16")]; + tensor var_24392_begin_0 = const()[name = tensor("op_24392_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_24392_end_0 = const()[name = tensor("op_24392_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_24392_end_mask_0 = const()[name = tensor("op_24392_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24392_cast_fp16 = slice_by_index(begin = var_24392_begin_0, end = var_24392_end_0, end_mask = var_24392_end_mask_0, x = q_111_cast_fp16)[name = tensor("op_24392_cast_fp16")]; + tensor var_24396_begin_0 = const()[name = tensor("op_24396_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_24396_end_0 = const()[name = tensor("op_24396_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_24396_end_mask_0 = const()[name = tensor("op_24396_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24396_cast_fp16 = slice_by_index(begin = var_24396_begin_0, end = var_24396_end_0, end_mask = var_24396_end_mask_0, x = q_111_cast_fp16)[name = tensor("op_24396_cast_fp16")]; + tensor var_24400_begin_0 = const()[name = tensor("op_24400_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_24400_end_0 = const()[name = tensor("op_24400_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_24400_end_mask_0 = const()[name = tensor("op_24400_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24400_cast_fp16 = slice_by_index(begin = var_24400_begin_0, end = var_24400_end_0, end_mask = var_24400_end_mask_0, x = q_111_cast_fp16)[name = tensor("op_24400_cast_fp16")]; + tensor var_24404_begin_0 = const()[name = tensor("op_24404_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_24404_end_0 = const()[name = tensor("op_24404_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_24404_end_mask_0 = const()[name = tensor("op_24404_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24404_cast_fp16 = slice_by_index(begin = var_24404_begin_0, end = var_24404_end_0, end_mask = var_24404_end_mask_0, x = q_111_cast_fp16)[name = tensor("op_24404_cast_fp16")]; + tensor var_24408_begin_0 = const()[name = tensor("op_24408_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_24408_end_0 = const()[name = tensor("op_24408_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_24408_end_mask_0 = const()[name = tensor("op_24408_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24408_cast_fp16 = slice_by_index(begin = var_24408_begin_0, end = var_24408_end_0, end_mask = var_24408_end_mask_0, x = q_111_cast_fp16)[name = tensor("op_24408_cast_fp16")]; + tensor var_24412_begin_0 = const()[name = tensor("op_24412_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_24412_end_0 = const()[name = tensor("op_24412_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_24412_end_mask_0 = const()[name = tensor("op_24412_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24412_cast_fp16 = slice_by_index(begin = var_24412_begin_0, end = var_24412_end_0, end_mask = var_24412_end_mask_0, x = q_111_cast_fp16)[name = tensor("op_24412_cast_fp16")]; + tensor var_24416_begin_0 = const()[name = tensor("op_24416_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_24416_end_0 = const()[name = tensor("op_24416_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_24416_end_mask_0 = const()[name = tensor("op_24416_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24416_cast_fp16 = slice_by_index(begin = var_24416_begin_0, end = var_24416_end_0, end_mask = var_24416_end_mask_0, x = q_111_cast_fp16)[name = tensor("op_24416_cast_fp16")]; + tensor var_24420_begin_0 = const()[name = tensor("op_24420_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_24420_end_0 = const()[name = tensor("op_24420_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_24420_end_mask_0 = const()[name = tensor("op_24420_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24420_cast_fp16 = slice_by_index(begin = var_24420_begin_0, end = var_24420_end_0, end_mask = var_24420_end_mask_0, x = q_111_cast_fp16)[name = tensor("op_24420_cast_fp16")]; + tensor var_24424_begin_0 = const()[name = tensor("op_24424_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_24424_end_0 = const()[name = tensor("op_24424_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_24424_end_mask_0 = const()[name = tensor("op_24424_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24424_cast_fp16 = slice_by_index(begin = var_24424_begin_0, end = var_24424_end_0, end_mask = var_24424_end_mask_0, x = q_111_cast_fp16)[name = tensor("op_24424_cast_fp16")]; + tensor var_24428_begin_0 = const()[name = tensor("op_24428_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_24428_end_0 = const()[name = tensor("op_24428_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_24428_end_mask_0 = const()[name = tensor("op_24428_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24428_cast_fp16 = slice_by_index(begin = var_24428_begin_0, end = var_24428_end_0, end_mask = var_24428_end_mask_0, x = q_111_cast_fp16)[name = tensor("op_24428_cast_fp16")]; + tensor var_24432_begin_0 = const()[name = tensor("op_24432_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_24432_end_0 = const()[name = tensor("op_24432_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_24432_end_mask_0 = const()[name = tensor("op_24432_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24432_cast_fp16 = slice_by_index(begin = var_24432_begin_0, end = var_24432_end_0, end_mask = var_24432_end_mask_0, x = q_111_cast_fp16)[name = tensor("op_24432_cast_fp16")]; + tensor var_24436_begin_0 = const()[name = tensor("op_24436_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_24436_end_0 = const()[name = tensor("op_24436_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_24436_end_mask_0 = const()[name = tensor("op_24436_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24436_cast_fp16 = slice_by_index(begin = var_24436_begin_0, end = var_24436_end_0, end_mask = var_24436_end_mask_0, x = q_111_cast_fp16)[name = tensor("op_24436_cast_fp16")]; + tensor var_24440_begin_0 = const()[name = tensor("op_24440_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_24440_end_0 = const()[name = tensor("op_24440_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_24440_end_mask_0 = const()[name = tensor("op_24440_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24440_cast_fp16 = slice_by_index(begin = var_24440_begin_0, end = var_24440_end_0, end_mask = var_24440_end_mask_0, x = q_111_cast_fp16)[name = tensor("op_24440_cast_fp16")]; + tensor var_24444_begin_0 = const()[name = tensor("op_24444_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_24444_end_0 = const()[name = tensor("op_24444_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_24444_end_mask_0 = const()[name = tensor("op_24444_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24444_cast_fp16 = slice_by_index(begin = var_24444_begin_0, end = var_24444_end_0, end_mask = var_24444_end_mask_0, x = q_111_cast_fp16)[name = tensor("op_24444_cast_fp16")]; + tensor var_24448_begin_0 = const()[name = tensor("op_24448_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_24448_end_0 = const()[name = tensor("op_24448_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_24448_end_mask_0 = const()[name = tensor("op_24448_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24448_cast_fp16 = slice_by_index(begin = var_24448_begin_0, end = var_24448_end_0, end_mask = var_24448_end_mask_0, x = q_111_cast_fp16)[name = tensor("op_24448_cast_fp16")]; + tensor var_24452_begin_0 = const()[name = tensor("op_24452_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_24452_end_0 = const()[name = tensor("op_24452_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_24452_end_mask_0 = const()[name = tensor("op_24452_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24452_cast_fp16 = slice_by_index(begin = var_24452_begin_0, end = var_24452_end_0, end_mask = var_24452_end_mask_0, x = q_111_cast_fp16)[name = tensor("op_24452_cast_fp16")]; + tensor k_223_perm_0 = const()[name = tensor("k_223_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_24459_begin_0 = const()[name = tensor("op_24459_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_24459_end_0 = const()[name = tensor("op_24459_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_24459_end_mask_0 = const()[name = tensor("op_24459_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_223_cast_fp16 = transpose(perm = k_223_perm_0, x = k_221_cast_fp16)[name = tensor("transpose_84")]; + tensor var_24459_cast_fp16 = slice_by_index(begin = var_24459_begin_0, end = var_24459_end_0, end_mask = var_24459_end_mask_0, x = k_223_cast_fp16)[name = tensor("op_24459_cast_fp16")]; + tensor var_24463_begin_0 = const()[name = tensor("op_24463_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_24463_end_0 = const()[name = tensor("op_24463_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_24463_end_mask_0 = const()[name = tensor("op_24463_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24463_cast_fp16 = slice_by_index(begin = var_24463_begin_0, end = var_24463_end_0, end_mask = var_24463_end_mask_0, x = k_223_cast_fp16)[name = tensor("op_24463_cast_fp16")]; + tensor var_24467_begin_0 = const()[name = tensor("op_24467_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_24467_end_0 = const()[name = tensor("op_24467_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_24467_end_mask_0 = const()[name = tensor("op_24467_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24467_cast_fp16 = slice_by_index(begin = var_24467_begin_0, end = var_24467_end_0, end_mask = var_24467_end_mask_0, x = k_223_cast_fp16)[name = tensor("op_24467_cast_fp16")]; + tensor var_24471_begin_0 = const()[name = tensor("op_24471_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_24471_end_0 = const()[name = tensor("op_24471_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_24471_end_mask_0 = const()[name = tensor("op_24471_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24471_cast_fp16 = slice_by_index(begin = var_24471_begin_0, end = var_24471_end_0, end_mask = var_24471_end_mask_0, x = k_223_cast_fp16)[name = tensor("op_24471_cast_fp16")]; + tensor var_24475_begin_0 = const()[name = tensor("op_24475_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_24475_end_0 = const()[name = tensor("op_24475_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_24475_end_mask_0 = const()[name = tensor("op_24475_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24475_cast_fp16 = slice_by_index(begin = var_24475_begin_0, end = var_24475_end_0, end_mask = var_24475_end_mask_0, x = k_223_cast_fp16)[name = tensor("op_24475_cast_fp16")]; + tensor var_24479_begin_0 = const()[name = tensor("op_24479_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_24479_end_0 = const()[name = tensor("op_24479_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_24479_end_mask_0 = const()[name = tensor("op_24479_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24479_cast_fp16 = slice_by_index(begin = var_24479_begin_0, end = var_24479_end_0, end_mask = var_24479_end_mask_0, x = k_223_cast_fp16)[name = tensor("op_24479_cast_fp16")]; + tensor var_24483_begin_0 = const()[name = tensor("op_24483_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_24483_end_0 = const()[name = tensor("op_24483_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_24483_end_mask_0 = const()[name = tensor("op_24483_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24483_cast_fp16 = slice_by_index(begin = var_24483_begin_0, end = var_24483_end_0, end_mask = var_24483_end_mask_0, x = k_223_cast_fp16)[name = tensor("op_24483_cast_fp16")]; + tensor var_24487_begin_0 = const()[name = tensor("op_24487_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_24487_end_0 = const()[name = tensor("op_24487_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_24487_end_mask_0 = const()[name = tensor("op_24487_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24487_cast_fp16 = slice_by_index(begin = var_24487_begin_0, end = var_24487_end_0, end_mask = var_24487_end_mask_0, x = k_223_cast_fp16)[name = tensor("op_24487_cast_fp16")]; + tensor var_24491_begin_0 = const()[name = tensor("op_24491_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_24491_end_0 = const()[name = tensor("op_24491_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_24491_end_mask_0 = const()[name = tensor("op_24491_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24491_cast_fp16 = slice_by_index(begin = var_24491_begin_0, end = var_24491_end_0, end_mask = var_24491_end_mask_0, x = k_223_cast_fp16)[name = tensor("op_24491_cast_fp16")]; + tensor var_24495_begin_0 = const()[name = tensor("op_24495_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_24495_end_0 = const()[name = tensor("op_24495_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_24495_end_mask_0 = const()[name = tensor("op_24495_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24495_cast_fp16 = slice_by_index(begin = var_24495_begin_0, end = var_24495_end_0, end_mask = var_24495_end_mask_0, x = k_223_cast_fp16)[name = tensor("op_24495_cast_fp16")]; + tensor var_24499_begin_0 = const()[name = tensor("op_24499_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_24499_end_0 = const()[name = tensor("op_24499_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_24499_end_mask_0 = const()[name = tensor("op_24499_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24499_cast_fp16 = slice_by_index(begin = var_24499_begin_0, end = var_24499_end_0, end_mask = var_24499_end_mask_0, x = k_223_cast_fp16)[name = tensor("op_24499_cast_fp16")]; + tensor var_24503_begin_0 = const()[name = tensor("op_24503_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_24503_end_0 = const()[name = tensor("op_24503_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_24503_end_mask_0 = const()[name = tensor("op_24503_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24503_cast_fp16 = slice_by_index(begin = var_24503_begin_0, end = var_24503_end_0, end_mask = var_24503_end_mask_0, x = k_223_cast_fp16)[name = tensor("op_24503_cast_fp16")]; + tensor var_24507_begin_0 = const()[name = tensor("op_24507_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_24507_end_0 = const()[name = tensor("op_24507_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_24507_end_mask_0 = const()[name = tensor("op_24507_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24507_cast_fp16 = slice_by_index(begin = var_24507_begin_0, end = var_24507_end_0, end_mask = var_24507_end_mask_0, x = k_223_cast_fp16)[name = tensor("op_24507_cast_fp16")]; + tensor var_24511_begin_0 = const()[name = tensor("op_24511_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_24511_end_0 = const()[name = tensor("op_24511_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_24511_end_mask_0 = const()[name = tensor("op_24511_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24511_cast_fp16 = slice_by_index(begin = var_24511_begin_0, end = var_24511_end_0, end_mask = var_24511_end_mask_0, x = k_223_cast_fp16)[name = tensor("op_24511_cast_fp16")]; + tensor var_24515_begin_0 = const()[name = tensor("op_24515_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_24515_end_0 = const()[name = tensor("op_24515_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_24515_end_mask_0 = const()[name = tensor("op_24515_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24515_cast_fp16 = slice_by_index(begin = var_24515_begin_0, end = var_24515_end_0, end_mask = var_24515_end_mask_0, x = k_223_cast_fp16)[name = tensor("op_24515_cast_fp16")]; + tensor var_24519_begin_0 = const()[name = tensor("op_24519_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_24519_end_0 = const()[name = tensor("op_24519_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_24519_end_mask_0 = const()[name = tensor("op_24519_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24519_cast_fp16 = slice_by_index(begin = var_24519_begin_0, end = var_24519_end_0, end_mask = var_24519_end_mask_0, x = k_223_cast_fp16)[name = tensor("op_24519_cast_fp16")]; + tensor var_24523_begin_0 = const()[name = tensor("op_24523_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_24523_end_0 = const()[name = tensor("op_24523_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_24523_end_mask_0 = const()[name = tensor("op_24523_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24523_cast_fp16 = slice_by_index(begin = var_24523_begin_0, end = var_24523_end_0, end_mask = var_24523_end_mask_0, x = k_223_cast_fp16)[name = tensor("op_24523_cast_fp16")]; + tensor var_24527_begin_0 = const()[name = tensor("op_24527_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_24527_end_0 = const()[name = tensor("op_24527_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_24527_end_mask_0 = const()[name = tensor("op_24527_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24527_cast_fp16 = slice_by_index(begin = var_24527_begin_0, end = var_24527_end_0, end_mask = var_24527_end_mask_0, x = k_223_cast_fp16)[name = tensor("op_24527_cast_fp16")]; + tensor var_24531_begin_0 = const()[name = tensor("op_24531_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_24531_end_0 = const()[name = tensor("op_24531_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_24531_end_mask_0 = const()[name = tensor("op_24531_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24531_cast_fp16 = slice_by_index(begin = var_24531_begin_0, end = var_24531_end_0, end_mask = var_24531_end_mask_0, x = k_223_cast_fp16)[name = tensor("op_24531_cast_fp16")]; + tensor var_24535_begin_0 = const()[name = tensor("op_24535_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_24535_end_0 = const()[name = tensor("op_24535_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_24535_end_mask_0 = const()[name = tensor("op_24535_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24535_cast_fp16 = slice_by_index(begin = var_24535_begin_0, end = var_24535_end_0, end_mask = var_24535_end_mask_0, x = k_223_cast_fp16)[name = tensor("op_24535_cast_fp16")]; + tensor var_24537_begin_0 = const()[name = tensor("op_24537_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_24537_end_0 = const()[name = tensor("op_24537_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_24537_end_mask_0 = const()[name = tensor("op_24537_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24537_cast_fp16 = slice_by_index(begin = var_24537_begin_0, end = var_24537_end_0, end_mask = var_24537_end_mask_0, x = v_111_cast_fp16)[name = tensor("op_24537_cast_fp16")]; + tensor var_24541_begin_0 = const()[name = tensor("op_24541_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_24541_end_0 = const()[name = tensor("op_24541_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_24541_end_mask_0 = const()[name = tensor("op_24541_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24541_cast_fp16 = slice_by_index(begin = var_24541_begin_0, end = var_24541_end_0, end_mask = var_24541_end_mask_0, x = v_111_cast_fp16)[name = tensor("op_24541_cast_fp16")]; + tensor var_24545_begin_0 = const()[name = tensor("op_24545_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_24545_end_0 = const()[name = tensor("op_24545_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_24545_end_mask_0 = const()[name = tensor("op_24545_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24545_cast_fp16 = slice_by_index(begin = var_24545_begin_0, end = var_24545_end_0, end_mask = var_24545_end_mask_0, x = v_111_cast_fp16)[name = tensor("op_24545_cast_fp16")]; + tensor var_24549_begin_0 = const()[name = tensor("op_24549_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_24549_end_0 = const()[name = tensor("op_24549_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_24549_end_mask_0 = const()[name = tensor("op_24549_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24549_cast_fp16 = slice_by_index(begin = var_24549_begin_0, end = var_24549_end_0, end_mask = var_24549_end_mask_0, x = v_111_cast_fp16)[name = tensor("op_24549_cast_fp16")]; + tensor var_24553_begin_0 = const()[name = tensor("op_24553_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_24553_end_0 = const()[name = tensor("op_24553_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_24553_end_mask_0 = const()[name = tensor("op_24553_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24553_cast_fp16 = slice_by_index(begin = var_24553_begin_0, end = var_24553_end_0, end_mask = var_24553_end_mask_0, x = v_111_cast_fp16)[name = tensor("op_24553_cast_fp16")]; + tensor var_24557_begin_0 = const()[name = tensor("op_24557_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_24557_end_0 = const()[name = tensor("op_24557_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_24557_end_mask_0 = const()[name = tensor("op_24557_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24557_cast_fp16 = slice_by_index(begin = var_24557_begin_0, end = var_24557_end_0, end_mask = var_24557_end_mask_0, x = v_111_cast_fp16)[name = tensor("op_24557_cast_fp16")]; + tensor var_24561_begin_0 = const()[name = tensor("op_24561_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_24561_end_0 = const()[name = tensor("op_24561_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_24561_end_mask_0 = const()[name = tensor("op_24561_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24561_cast_fp16 = slice_by_index(begin = var_24561_begin_0, end = var_24561_end_0, end_mask = var_24561_end_mask_0, x = v_111_cast_fp16)[name = tensor("op_24561_cast_fp16")]; + tensor var_24565_begin_0 = const()[name = tensor("op_24565_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_24565_end_0 = const()[name = tensor("op_24565_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_24565_end_mask_0 = const()[name = tensor("op_24565_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24565_cast_fp16 = slice_by_index(begin = var_24565_begin_0, end = var_24565_end_0, end_mask = var_24565_end_mask_0, x = v_111_cast_fp16)[name = tensor("op_24565_cast_fp16")]; + tensor var_24569_begin_0 = const()[name = tensor("op_24569_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_24569_end_0 = const()[name = tensor("op_24569_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_24569_end_mask_0 = const()[name = tensor("op_24569_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24569_cast_fp16 = slice_by_index(begin = var_24569_begin_0, end = var_24569_end_0, end_mask = var_24569_end_mask_0, x = v_111_cast_fp16)[name = tensor("op_24569_cast_fp16")]; + tensor var_24573_begin_0 = const()[name = tensor("op_24573_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_24573_end_0 = const()[name = tensor("op_24573_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_24573_end_mask_0 = const()[name = tensor("op_24573_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24573_cast_fp16 = slice_by_index(begin = var_24573_begin_0, end = var_24573_end_0, end_mask = var_24573_end_mask_0, x = v_111_cast_fp16)[name = tensor("op_24573_cast_fp16")]; + tensor var_24577_begin_0 = const()[name = tensor("op_24577_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_24577_end_0 = const()[name = tensor("op_24577_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_24577_end_mask_0 = const()[name = tensor("op_24577_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24577_cast_fp16 = slice_by_index(begin = var_24577_begin_0, end = var_24577_end_0, end_mask = var_24577_end_mask_0, x = v_111_cast_fp16)[name = tensor("op_24577_cast_fp16")]; + tensor var_24581_begin_0 = const()[name = tensor("op_24581_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_24581_end_0 = const()[name = tensor("op_24581_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_24581_end_mask_0 = const()[name = tensor("op_24581_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24581_cast_fp16 = slice_by_index(begin = var_24581_begin_0, end = var_24581_end_0, end_mask = var_24581_end_mask_0, x = v_111_cast_fp16)[name = tensor("op_24581_cast_fp16")]; + tensor var_24585_begin_0 = const()[name = tensor("op_24585_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_24585_end_0 = const()[name = tensor("op_24585_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_24585_end_mask_0 = const()[name = tensor("op_24585_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24585_cast_fp16 = slice_by_index(begin = var_24585_begin_0, end = var_24585_end_0, end_mask = var_24585_end_mask_0, x = v_111_cast_fp16)[name = tensor("op_24585_cast_fp16")]; + tensor var_24589_begin_0 = const()[name = tensor("op_24589_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_24589_end_0 = const()[name = tensor("op_24589_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_24589_end_mask_0 = const()[name = tensor("op_24589_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24589_cast_fp16 = slice_by_index(begin = var_24589_begin_0, end = var_24589_end_0, end_mask = var_24589_end_mask_0, x = v_111_cast_fp16)[name = tensor("op_24589_cast_fp16")]; + tensor var_24593_begin_0 = const()[name = tensor("op_24593_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_24593_end_0 = const()[name = tensor("op_24593_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_24593_end_mask_0 = const()[name = tensor("op_24593_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24593_cast_fp16 = slice_by_index(begin = var_24593_begin_0, end = var_24593_end_0, end_mask = var_24593_end_mask_0, x = v_111_cast_fp16)[name = tensor("op_24593_cast_fp16")]; + tensor var_24597_begin_0 = const()[name = tensor("op_24597_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_24597_end_0 = const()[name = tensor("op_24597_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_24597_end_mask_0 = const()[name = tensor("op_24597_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24597_cast_fp16 = slice_by_index(begin = var_24597_begin_0, end = var_24597_end_0, end_mask = var_24597_end_mask_0, x = v_111_cast_fp16)[name = tensor("op_24597_cast_fp16")]; + tensor var_24601_begin_0 = const()[name = tensor("op_24601_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_24601_end_0 = const()[name = tensor("op_24601_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_24601_end_mask_0 = const()[name = tensor("op_24601_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24601_cast_fp16 = slice_by_index(begin = var_24601_begin_0, end = var_24601_end_0, end_mask = var_24601_end_mask_0, x = v_111_cast_fp16)[name = tensor("op_24601_cast_fp16")]; + tensor var_24605_begin_0 = const()[name = tensor("op_24605_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_24605_end_0 = const()[name = tensor("op_24605_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_24605_end_mask_0 = const()[name = tensor("op_24605_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24605_cast_fp16 = slice_by_index(begin = var_24605_begin_0, end = var_24605_end_0, end_mask = var_24605_end_mask_0, x = v_111_cast_fp16)[name = tensor("op_24605_cast_fp16")]; + tensor var_24609_begin_0 = const()[name = tensor("op_24609_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_24609_end_0 = const()[name = tensor("op_24609_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_24609_end_mask_0 = const()[name = tensor("op_24609_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24609_cast_fp16 = slice_by_index(begin = var_24609_begin_0, end = var_24609_end_0, end_mask = var_24609_end_mask_0, x = v_111_cast_fp16)[name = tensor("op_24609_cast_fp16")]; + tensor var_24613_begin_0 = const()[name = tensor("op_24613_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_24613_end_0 = const()[name = tensor("op_24613_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_24613_end_mask_0 = const()[name = tensor("op_24613_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24613_cast_fp16 = slice_by_index(begin = var_24613_begin_0, end = var_24613_end_0, end_mask = var_24613_end_mask_0, x = v_111_cast_fp16)[name = tensor("op_24613_cast_fp16")]; + tensor var_24617_equation_0 = const()[name = tensor("op_24617_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24617_cast_fp16 = einsum(equation = var_24617_equation_0, values = (var_24459_cast_fp16, var_24376_cast_fp16))[name = tensor("op_24617_cast_fp16")]; + tensor var_24618_to_fp16 = const()[name = tensor("op_24618_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2041_cast_fp16 = mul(x = var_24617_cast_fp16, y = var_24618_to_fp16)[name = tensor("aw_2041_cast_fp16")]; + tensor var_24621_equation_0 = const()[name = tensor("op_24621_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24621_cast_fp16 = einsum(equation = var_24621_equation_0, values = (var_24463_cast_fp16, var_24380_cast_fp16))[name = tensor("op_24621_cast_fp16")]; + tensor var_24622_to_fp16 = const()[name = tensor("op_24622_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2043_cast_fp16 = mul(x = var_24621_cast_fp16, y = var_24622_to_fp16)[name = tensor("aw_2043_cast_fp16")]; + tensor var_24625_equation_0 = const()[name = tensor("op_24625_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24625_cast_fp16 = einsum(equation = var_24625_equation_0, values = (var_24467_cast_fp16, var_24384_cast_fp16))[name = tensor("op_24625_cast_fp16")]; + tensor var_24626_to_fp16 = const()[name = tensor("op_24626_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2045_cast_fp16 = mul(x = var_24625_cast_fp16, y = var_24626_to_fp16)[name = tensor("aw_2045_cast_fp16")]; + tensor var_24629_equation_0 = const()[name = tensor("op_24629_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24629_cast_fp16 = einsum(equation = var_24629_equation_0, values = (var_24471_cast_fp16, var_24388_cast_fp16))[name = tensor("op_24629_cast_fp16")]; + tensor var_24630_to_fp16 = const()[name = tensor("op_24630_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2047_cast_fp16 = mul(x = var_24629_cast_fp16, y = var_24630_to_fp16)[name = tensor("aw_2047_cast_fp16")]; + tensor var_24633_equation_0 = const()[name = tensor("op_24633_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24633_cast_fp16 = einsum(equation = var_24633_equation_0, values = (var_24475_cast_fp16, var_24392_cast_fp16))[name = tensor("op_24633_cast_fp16")]; + tensor var_24634_to_fp16 = const()[name = tensor("op_24634_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2049_cast_fp16 = mul(x = var_24633_cast_fp16, y = var_24634_to_fp16)[name = tensor("aw_2049_cast_fp16")]; + tensor var_24637_equation_0 = const()[name = tensor("op_24637_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24637_cast_fp16 = einsum(equation = var_24637_equation_0, values = (var_24479_cast_fp16, var_24396_cast_fp16))[name = tensor("op_24637_cast_fp16")]; + tensor var_24638_to_fp16 = const()[name = tensor("op_24638_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2051_cast_fp16 = mul(x = var_24637_cast_fp16, y = var_24638_to_fp16)[name = tensor("aw_2051_cast_fp16")]; + tensor var_24641_equation_0 = const()[name = tensor("op_24641_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24641_cast_fp16 = einsum(equation = var_24641_equation_0, values = (var_24483_cast_fp16, var_24400_cast_fp16))[name = tensor("op_24641_cast_fp16")]; + tensor var_24642_to_fp16 = const()[name = tensor("op_24642_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2053_cast_fp16 = mul(x = var_24641_cast_fp16, y = var_24642_to_fp16)[name = tensor("aw_2053_cast_fp16")]; + tensor var_24645_equation_0 = const()[name = tensor("op_24645_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24645_cast_fp16 = einsum(equation = var_24645_equation_0, values = (var_24487_cast_fp16, var_24404_cast_fp16))[name = tensor("op_24645_cast_fp16")]; + tensor var_24646_to_fp16 = const()[name = tensor("op_24646_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2055_cast_fp16 = mul(x = var_24645_cast_fp16, y = var_24646_to_fp16)[name = tensor("aw_2055_cast_fp16")]; + tensor var_24649_equation_0 = const()[name = tensor("op_24649_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24649_cast_fp16 = einsum(equation = var_24649_equation_0, values = (var_24491_cast_fp16, var_24408_cast_fp16))[name = tensor("op_24649_cast_fp16")]; + tensor var_24650_to_fp16 = const()[name = tensor("op_24650_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2057_cast_fp16 = mul(x = var_24649_cast_fp16, y = var_24650_to_fp16)[name = tensor("aw_2057_cast_fp16")]; + tensor var_24653_equation_0 = const()[name = tensor("op_24653_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24653_cast_fp16 = einsum(equation = var_24653_equation_0, values = (var_24495_cast_fp16, var_24412_cast_fp16))[name = tensor("op_24653_cast_fp16")]; + tensor var_24654_to_fp16 = const()[name = tensor("op_24654_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2059_cast_fp16 = mul(x = var_24653_cast_fp16, y = var_24654_to_fp16)[name = tensor("aw_2059_cast_fp16")]; + tensor var_24657_equation_0 = const()[name = tensor("op_24657_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24657_cast_fp16 = einsum(equation = var_24657_equation_0, values = (var_24499_cast_fp16, var_24416_cast_fp16))[name = tensor("op_24657_cast_fp16")]; + tensor var_24658_to_fp16 = const()[name = tensor("op_24658_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2061_cast_fp16 = mul(x = var_24657_cast_fp16, y = var_24658_to_fp16)[name = tensor("aw_2061_cast_fp16")]; + tensor var_24661_equation_0 = const()[name = tensor("op_24661_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24661_cast_fp16 = einsum(equation = var_24661_equation_0, values = (var_24503_cast_fp16, var_24420_cast_fp16))[name = tensor("op_24661_cast_fp16")]; + tensor var_24662_to_fp16 = const()[name = tensor("op_24662_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2063_cast_fp16 = mul(x = var_24661_cast_fp16, y = var_24662_to_fp16)[name = tensor("aw_2063_cast_fp16")]; + tensor var_24665_equation_0 = const()[name = tensor("op_24665_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24665_cast_fp16 = einsum(equation = var_24665_equation_0, values = (var_24507_cast_fp16, var_24424_cast_fp16))[name = tensor("op_24665_cast_fp16")]; + tensor var_24666_to_fp16 = const()[name = tensor("op_24666_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2065_cast_fp16 = mul(x = var_24665_cast_fp16, y = var_24666_to_fp16)[name = tensor("aw_2065_cast_fp16")]; + tensor var_24669_equation_0 = const()[name = tensor("op_24669_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24669_cast_fp16 = einsum(equation = var_24669_equation_0, values = (var_24511_cast_fp16, var_24428_cast_fp16))[name = tensor("op_24669_cast_fp16")]; + tensor var_24670_to_fp16 = const()[name = tensor("op_24670_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2067_cast_fp16 = mul(x = var_24669_cast_fp16, y = var_24670_to_fp16)[name = tensor("aw_2067_cast_fp16")]; + tensor var_24673_equation_0 = const()[name = tensor("op_24673_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24673_cast_fp16 = einsum(equation = var_24673_equation_0, values = (var_24515_cast_fp16, var_24432_cast_fp16))[name = tensor("op_24673_cast_fp16")]; + tensor var_24674_to_fp16 = const()[name = tensor("op_24674_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2069_cast_fp16 = mul(x = var_24673_cast_fp16, y = var_24674_to_fp16)[name = tensor("aw_2069_cast_fp16")]; + tensor var_24677_equation_0 = const()[name = tensor("op_24677_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24677_cast_fp16 = einsum(equation = var_24677_equation_0, values = (var_24519_cast_fp16, var_24436_cast_fp16))[name = tensor("op_24677_cast_fp16")]; + tensor var_24678_to_fp16 = const()[name = tensor("op_24678_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2071_cast_fp16 = mul(x = var_24677_cast_fp16, y = var_24678_to_fp16)[name = tensor("aw_2071_cast_fp16")]; + tensor var_24681_equation_0 = const()[name = tensor("op_24681_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24681_cast_fp16 = einsum(equation = var_24681_equation_0, values = (var_24523_cast_fp16, var_24440_cast_fp16))[name = tensor("op_24681_cast_fp16")]; + tensor var_24682_to_fp16 = const()[name = tensor("op_24682_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2073_cast_fp16 = mul(x = var_24681_cast_fp16, y = var_24682_to_fp16)[name = tensor("aw_2073_cast_fp16")]; + tensor var_24685_equation_0 = const()[name = tensor("op_24685_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24685_cast_fp16 = einsum(equation = var_24685_equation_0, values = (var_24527_cast_fp16, var_24444_cast_fp16))[name = tensor("op_24685_cast_fp16")]; + tensor var_24686_to_fp16 = const()[name = tensor("op_24686_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2075_cast_fp16 = mul(x = var_24685_cast_fp16, y = var_24686_to_fp16)[name = tensor("aw_2075_cast_fp16")]; + tensor var_24689_equation_0 = const()[name = tensor("op_24689_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24689_cast_fp16 = einsum(equation = var_24689_equation_0, values = (var_24531_cast_fp16, var_24448_cast_fp16))[name = tensor("op_24689_cast_fp16")]; + tensor var_24690_to_fp16 = const()[name = tensor("op_24690_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2077_cast_fp16 = mul(x = var_24689_cast_fp16, y = var_24690_to_fp16)[name = tensor("aw_2077_cast_fp16")]; + tensor var_24693_equation_0 = const()[name = tensor("op_24693_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24693_cast_fp16 = einsum(equation = var_24693_equation_0, values = (var_24535_cast_fp16, var_24452_cast_fp16))[name = tensor("op_24693_cast_fp16")]; + tensor var_24694_to_fp16 = const()[name = tensor("op_24694_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2079_cast_fp16 = mul(x = var_24693_cast_fp16, y = var_24694_to_fp16)[name = tensor("aw_2079_cast_fp16")]; + tensor var_24696_cast_fp16 = softmax(axis = var_21077, x = aw_2041_cast_fp16)[name = tensor("op_24696_cast_fp16")]; + tensor var_24697_cast_fp16 = softmax(axis = var_21077, x = aw_2043_cast_fp16)[name = tensor("op_24697_cast_fp16")]; + tensor var_24698_cast_fp16 = softmax(axis = var_21077, x = aw_2045_cast_fp16)[name = tensor("op_24698_cast_fp16")]; + tensor var_24699_cast_fp16 = softmax(axis = var_21077, x = aw_2047_cast_fp16)[name = tensor("op_24699_cast_fp16")]; + tensor var_24700_cast_fp16 = softmax(axis = var_21077, x = aw_2049_cast_fp16)[name = tensor("op_24700_cast_fp16")]; + tensor var_24701_cast_fp16 = softmax(axis = var_21077, x = aw_2051_cast_fp16)[name = tensor("op_24701_cast_fp16")]; + tensor var_24702_cast_fp16 = softmax(axis = var_21077, x = aw_2053_cast_fp16)[name = tensor("op_24702_cast_fp16")]; + tensor var_24703_cast_fp16 = softmax(axis = var_21077, x = aw_2055_cast_fp16)[name = tensor("op_24703_cast_fp16")]; + tensor var_24704_cast_fp16 = softmax(axis = var_21077, x = aw_2057_cast_fp16)[name = tensor("op_24704_cast_fp16")]; + tensor var_24705_cast_fp16 = softmax(axis = var_21077, x = aw_2059_cast_fp16)[name = tensor("op_24705_cast_fp16")]; + tensor var_24706_cast_fp16 = softmax(axis = var_21077, x = aw_2061_cast_fp16)[name = tensor("op_24706_cast_fp16")]; + tensor var_24707_cast_fp16 = softmax(axis = var_21077, x = aw_2063_cast_fp16)[name = tensor("op_24707_cast_fp16")]; + tensor var_24708_cast_fp16 = softmax(axis = var_21077, x = aw_2065_cast_fp16)[name = tensor("op_24708_cast_fp16")]; + tensor var_24709_cast_fp16 = softmax(axis = var_21077, x = aw_2067_cast_fp16)[name = tensor("op_24709_cast_fp16")]; + tensor var_24710_cast_fp16 = softmax(axis = var_21077, x = aw_2069_cast_fp16)[name = tensor("op_24710_cast_fp16")]; + tensor var_24711_cast_fp16 = softmax(axis = var_21077, x = aw_2071_cast_fp16)[name = tensor("op_24711_cast_fp16")]; + tensor var_24712_cast_fp16 = softmax(axis = var_21077, x = aw_2073_cast_fp16)[name = tensor("op_24712_cast_fp16")]; + tensor var_24713_cast_fp16 = softmax(axis = var_21077, x = aw_2075_cast_fp16)[name = tensor("op_24713_cast_fp16")]; + tensor var_24714_cast_fp16 = softmax(axis = var_21077, x = aw_2077_cast_fp16)[name = tensor("op_24714_cast_fp16")]; + tensor var_24715_cast_fp16 = softmax(axis = var_21077, x = aw_2079_cast_fp16)[name = tensor("op_24715_cast_fp16")]; + tensor var_24717_equation_0 = const()[name = tensor("op_24717_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24717_cast_fp16 = einsum(equation = var_24717_equation_0, values = (var_24537_cast_fp16, var_24696_cast_fp16))[name = tensor("op_24717_cast_fp16")]; + tensor var_24719_equation_0 = const()[name = tensor("op_24719_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24719_cast_fp16 = einsum(equation = var_24719_equation_0, values = (var_24541_cast_fp16, var_24697_cast_fp16))[name = tensor("op_24719_cast_fp16")]; + tensor var_24721_equation_0 = const()[name = tensor("op_24721_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24721_cast_fp16 = einsum(equation = var_24721_equation_0, values = (var_24545_cast_fp16, var_24698_cast_fp16))[name = tensor("op_24721_cast_fp16")]; + tensor var_24723_equation_0 = const()[name = tensor("op_24723_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24723_cast_fp16 = einsum(equation = var_24723_equation_0, values = (var_24549_cast_fp16, var_24699_cast_fp16))[name = tensor("op_24723_cast_fp16")]; + tensor var_24725_equation_0 = const()[name = tensor("op_24725_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24725_cast_fp16 = einsum(equation = var_24725_equation_0, values = (var_24553_cast_fp16, var_24700_cast_fp16))[name = tensor("op_24725_cast_fp16")]; + tensor var_24727_equation_0 = const()[name = tensor("op_24727_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24727_cast_fp16 = einsum(equation = var_24727_equation_0, values = (var_24557_cast_fp16, var_24701_cast_fp16))[name = tensor("op_24727_cast_fp16")]; + tensor var_24729_equation_0 = const()[name = tensor("op_24729_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24729_cast_fp16 = einsum(equation = var_24729_equation_0, values = (var_24561_cast_fp16, var_24702_cast_fp16))[name = tensor("op_24729_cast_fp16")]; + tensor var_24731_equation_0 = const()[name = tensor("op_24731_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24731_cast_fp16 = einsum(equation = var_24731_equation_0, values = (var_24565_cast_fp16, var_24703_cast_fp16))[name = tensor("op_24731_cast_fp16")]; + tensor var_24733_equation_0 = const()[name = tensor("op_24733_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24733_cast_fp16 = einsum(equation = var_24733_equation_0, values = (var_24569_cast_fp16, var_24704_cast_fp16))[name = tensor("op_24733_cast_fp16")]; + tensor var_24735_equation_0 = const()[name = tensor("op_24735_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24735_cast_fp16 = einsum(equation = var_24735_equation_0, values = (var_24573_cast_fp16, var_24705_cast_fp16))[name = tensor("op_24735_cast_fp16")]; + tensor var_24737_equation_0 = const()[name = tensor("op_24737_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24737_cast_fp16 = einsum(equation = var_24737_equation_0, values = (var_24577_cast_fp16, var_24706_cast_fp16))[name = tensor("op_24737_cast_fp16")]; + tensor var_24739_equation_0 = const()[name = tensor("op_24739_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24739_cast_fp16 = einsum(equation = var_24739_equation_0, values = (var_24581_cast_fp16, var_24707_cast_fp16))[name = tensor("op_24739_cast_fp16")]; + tensor var_24741_equation_0 = const()[name = tensor("op_24741_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24741_cast_fp16 = einsum(equation = var_24741_equation_0, values = (var_24585_cast_fp16, var_24708_cast_fp16))[name = tensor("op_24741_cast_fp16")]; + tensor var_24743_equation_0 = const()[name = tensor("op_24743_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24743_cast_fp16 = einsum(equation = var_24743_equation_0, values = (var_24589_cast_fp16, var_24709_cast_fp16))[name = tensor("op_24743_cast_fp16")]; + tensor var_24745_equation_0 = const()[name = tensor("op_24745_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24745_cast_fp16 = einsum(equation = var_24745_equation_0, values = (var_24593_cast_fp16, var_24710_cast_fp16))[name = tensor("op_24745_cast_fp16")]; + tensor var_24747_equation_0 = const()[name = tensor("op_24747_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24747_cast_fp16 = einsum(equation = var_24747_equation_0, values = (var_24597_cast_fp16, var_24711_cast_fp16))[name = tensor("op_24747_cast_fp16")]; + tensor var_24749_equation_0 = const()[name = tensor("op_24749_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24749_cast_fp16 = einsum(equation = var_24749_equation_0, values = (var_24601_cast_fp16, var_24712_cast_fp16))[name = tensor("op_24749_cast_fp16")]; + tensor var_24751_equation_0 = const()[name = tensor("op_24751_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24751_cast_fp16 = einsum(equation = var_24751_equation_0, values = (var_24605_cast_fp16, var_24713_cast_fp16))[name = tensor("op_24751_cast_fp16")]; + tensor var_24753_equation_0 = const()[name = tensor("op_24753_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24753_cast_fp16 = einsum(equation = var_24753_equation_0, values = (var_24609_cast_fp16, var_24714_cast_fp16))[name = tensor("op_24753_cast_fp16")]; + tensor var_24755_equation_0 = const()[name = tensor("op_24755_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24755_cast_fp16 = einsum(equation = var_24755_equation_0, values = (var_24613_cast_fp16, var_24715_cast_fp16))[name = tensor("op_24755_cast_fp16")]; + tensor input_353_interleave_0 = const()[name = tensor("input_353_interleave_0"), val = tensor(false)]; + tensor input_353_cast_fp16 = concat(axis = var_21077, interleave = input_353_interleave_0, values = (var_24717_cast_fp16, var_24719_cast_fp16, var_24721_cast_fp16, var_24723_cast_fp16, var_24725_cast_fp16, var_24727_cast_fp16, var_24729_cast_fp16, var_24731_cast_fp16, var_24733_cast_fp16, var_24735_cast_fp16, var_24737_cast_fp16, var_24739_cast_fp16, var_24741_cast_fp16, var_24743_cast_fp16, var_24745_cast_fp16, var_24747_cast_fp16, var_24749_cast_fp16, var_24751_cast_fp16, var_24753_cast_fp16, var_24755_cast_fp16))[name = tensor("input_353_cast_fp16")]; + tensor var_24761 = const()[name = tensor("op_24761"), val = tensor([1, 1])]; + tensor var_24763 = const()[name = tensor("op_24763"), val = tensor([1, 1])]; + tensor var_24765_pad_type_0 = const()[name = tensor("op_24765_pad_type_0"), val = tensor("custom")]; + tensor var_24765_pad_0 = const()[name = tensor("op_24765_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_3_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_3_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1961531264)))]; + tensor mid_block_attentions_0_transformer_blocks_3_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_3_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1964808128)))]; + tensor var_24765_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_3_attn2_to_out_0_bias_to_fp16, dilations = var_24763, groups = var_21077, pad = var_24765_pad_0, pad_type = var_24765_pad_type_0, strides = var_24761, weight = mid_block_attentions_0_transformer_blocks_3_attn2_to_out_0_weight_to_fp16, x = input_353_cast_fp16)[name = tensor("op_24765_cast_fp16")]; + tensor inputs_167_cast_fp16 = add(x = var_24765_cast_fp16, y = inputs_165_cast_fp16)[name = tensor("inputs_167_cast_fp16")]; + tensor input_355_axes_0 = const()[name = tensor("input_355_axes_0"), val = tensor([1])]; + tensor input_355_gamma_0_to_fp16 = const()[name = tensor("input_355_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1964810752)))]; + tensor input_355_beta_0_to_fp16 = const()[name = tensor("input_355_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1964813376)))]; + tensor var_24775_to_fp16 = const()[name = tensor("op_24775_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_355_cast_fp16 = layer_norm(axes = input_355_axes_0, beta = input_355_beta_0_to_fp16, epsilon = var_24775_to_fp16, gamma = input_355_gamma_0_to_fp16, x = inputs_167_cast_fp16)[name = tensor("input_355_cast_fp16")]; + tensor var_24791 = const()[name = tensor("op_24791"), val = tensor([1, 1])]; + tensor var_24793 = const()[name = tensor("op_24793"), val = tensor([1, 1])]; + tensor var_24795_pad_type_0 = const()[name = tensor("op_24795_pad_type_0"), val = tensor("custom")]; + tensor var_24795_pad_0 = const()[name = tensor("op_24795_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_3_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_3_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1964816000)))]; + tensor mid_block_attentions_0_transformer_blocks_3_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_3_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1991030464)))]; + tensor var_24795_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_3_ff_net_0_proj_bias_to_fp16, dilations = var_24793, groups = var_21077, pad = var_24795_pad_0, pad_type = var_24795_pad_type_0, strides = var_24791, weight = mid_block_attentions_0_transformer_blocks_3_ff_net_0_proj_weight_to_fp16, x = input_355_cast_fp16)[name = tensor("op_24795_cast_fp16")]; + tensor var_24796_split_sizes_0 = const()[name = tensor("op_24796_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_24796_axis_0 = const()[name = tensor("op_24796_axis_0"), val = tensor(1)]; + tensor var_24796_cast_fp16_0, tensor var_24796_cast_fp16_1 = split(axis = var_24796_axis_0, split_sizes = var_24796_split_sizes_0, x = var_24795_cast_fp16)[name = tensor("op_24796_cast_fp16")]; + tensor var_24798_mode_0 = const()[name = tensor("op_24798_mode_0"), val = tensor("EXACT")]; + tensor var_24798_cast_fp16 = gelu(mode = var_24798_mode_0, x = var_24796_cast_fp16_1)[name = tensor("op_24798_cast_fp16")]; + tensor input_357_cast_fp16 = mul(x = var_24796_cast_fp16_0, y = var_24798_cast_fp16)[name = tensor("input_357_cast_fp16")]; + tensor var_24802 = const()[name = tensor("op_24802"), val = tensor([1, 1])]; + tensor var_24804 = const()[name = tensor("op_24804"), val = tensor([1, 1])]; + tensor var_24806_pad_type_0 = const()[name = tensor("op_24806_pad_type_0"), val = tensor("custom")]; + tensor var_24806_pad_0 = const()[name = tensor("op_24806_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_3_ff_net_2_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_3_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1991051008)))]; + tensor mid_block_attentions_0_transformer_blocks_3_ff_net_2_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_3_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2004158272)))]; + tensor var_24806_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_3_ff_net_2_bias_to_fp16, dilations = var_24804, groups = var_21077, pad = var_24806_pad_0, pad_type = var_24806_pad_type_0, strides = var_24802, weight = mid_block_attentions_0_transformer_blocks_3_ff_net_2_weight_to_fp16, x = input_357_cast_fp16)[name = tensor("op_24806_cast_fp16")]; + tensor inputs_169_cast_fp16 = add(x = var_24806_cast_fp16, y = inputs_167_cast_fp16)[name = tensor("inputs_169_cast_fp16")]; + tensor hidden_states_233_axes_0 = const()[name = tensor("hidden_states_233_axes_0"), val = tensor([1])]; + tensor hidden_states_233_gamma_0_to_fp16 = const()[name = tensor("hidden_states_233_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2004160896)))]; + tensor hidden_states_233_beta_0_to_fp16 = const()[name = tensor("hidden_states_233_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2004163520)))]; + tensor var_24822_to_fp16 = const()[name = tensor("op_24822_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_233_cast_fp16 = layer_norm(axes = hidden_states_233_axes_0, beta = hidden_states_233_beta_0_to_fp16, epsilon = var_24822_to_fp16, gamma = hidden_states_233_gamma_0_to_fp16, x = inputs_169_cast_fp16)[name = tensor("hidden_states_233_cast_fp16")]; + tensor var_24837 = const()[name = tensor("op_24837"), val = tensor([1, 1])]; + tensor var_24839 = const()[name = tensor("op_24839"), val = tensor([1, 1])]; + tensor q_113_pad_type_0 = const()[name = tensor("q_113_pad_type_0"), val = tensor("custom")]; + tensor q_113_pad_0 = const()[name = tensor("q_113_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_4_attn1_to_q_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_4_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2004166144)))]; + tensor q_113_cast_fp16 = conv(dilations = var_24839, groups = var_21077, pad = q_113_pad_0, pad_type = q_113_pad_type_0, strides = var_24837, weight = mid_block_attentions_0_transformer_blocks_4_attn1_to_q_weight_to_fp16, x = hidden_states_233_cast_fp16)[name = tensor("q_113_cast_fp16")]; + tensor var_24843 = const()[name = tensor("op_24843"), val = tensor([1, 1])]; + tensor var_24845 = const()[name = tensor("op_24845"), val = tensor([1, 1])]; + tensor k_225_pad_type_0 = const()[name = tensor("k_225_pad_type_0"), val = tensor("custom")]; + tensor k_225_pad_0 = const()[name = tensor("k_225_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_4_attn1_to_k_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_4_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2007443008)))]; + tensor k_225_cast_fp16 = conv(dilations = var_24845, groups = var_21077, pad = k_225_pad_0, pad_type = k_225_pad_type_0, strides = var_24843, weight = mid_block_attentions_0_transformer_blocks_4_attn1_to_k_weight_to_fp16, x = hidden_states_233_cast_fp16)[name = tensor("k_225_cast_fp16")]; + tensor var_24849 = const()[name = tensor("op_24849"), val = tensor([1, 1])]; + tensor var_24851 = const()[name = tensor("op_24851"), val = tensor([1, 1])]; + tensor v_113_pad_type_0 = const()[name = tensor("v_113_pad_type_0"), val = tensor("custom")]; + tensor v_113_pad_0 = const()[name = tensor("v_113_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_4_attn1_to_v_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_4_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2010719872)))]; + tensor v_113_cast_fp16 = conv(dilations = var_24851, groups = var_21077, pad = v_113_pad_0, pad_type = v_113_pad_type_0, strides = var_24849, weight = mid_block_attentions_0_transformer_blocks_4_attn1_to_v_weight_to_fp16, x = hidden_states_233_cast_fp16)[name = tensor("v_113_cast_fp16")]; + tensor var_24855_begin_0 = const()[name = tensor("op_24855_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_24855_end_0 = const()[name = tensor("op_24855_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_24855_end_mask_0 = const()[name = tensor("op_24855_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24855_cast_fp16 = slice_by_index(begin = var_24855_begin_0, end = var_24855_end_0, end_mask = var_24855_end_mask_0, x = q_113_cast_fp16)[name = tensor("op_24855_cast_fp16")]; + tensor var_24859_begin_0 = const()[name = tensor("op_24859_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_24859_end_0 = const()[name = tensor("op_24859_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_24859_end_mask_0 = const()[name = tensor("op_24859_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24859_cast_fp16 = slice_by_index(begin = var_24859_begin_0, end = var_24859_end_0, end_mask = var_24859_end_mask_0, x = q_113_cast_fp16)[name = tensor("op_24859_cast_fp16")]; + tensor var_24863_begin_0 = const()[name = tensor("op_24863_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_24863_end_0 = const()[name = tensor("op_24863_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_24863_end_mask_0 = const()[name = tensor("op_24863_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24863_cast_fp16 = slice_by_index(begin = var_24863_begin_0, end = var_24863_end_0, end_mask = var_24863_end_mask_0, x = q_113_cast_fp16)[name = tensor("op_24863_cast_fp16")]; + tensor var_24867_begin_0 = const()[name = tensor("op_24867_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_24867_end_0 = const()[name = tensor("op_24867_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_24867_end_mask_0 = const()[name = tensor("op_24867_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24867_cast_fp16 = slice_by_index(begin = var_24867_begin_0, end = var_24867_end_0, end_mask = var_24867_end_mask_0, x = q_113_cast_fp16)[name = tensor("op_24867_cast_fp16")]; + tensor var_24871_begin_0 = const()[name = tensor("op_24871_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_24871_end_0 = const()[name = tensor("op_24871_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_24871_end_mask_0 = const()[name = tensor("op_24871_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24871_cast_fp16 = slice_by_index(begin = var_24871_begin_0, end = var_24871_end_0, end_mask = var_24871_end_mask_0, x = q_113_cast_fp16)[name = tensor("op_24871_cast_fp16")]; + tensor var_24875_begin_0 = const()[name = tensor("op_24875_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_24875_end_0 = const()[name = tensor("op_24875_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_24875_end_mask_0 = const()[name = tensor("op_24875_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24875_cast_fp16 = slice_by_index(begin = var_24875_begin_0, end = var_24875_end_0, end_mask = var_24875_end_mask_0, x = q_113_cast_fp16)[name = tensor("op_24875_cast_fp16")]; + tensor var_24879_begin_0 = const()[name = tensor("op_24879_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_24879_end_0 = const()[name = tensor("op_24879_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_24879_end_mask_0 = const()[name = tensor("op_24879_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24879_cast_fp16 = slice_by_index(begin = var_24879_begin_0, end = var_24879_end_0, end_mask = var_24879_end_mask_0, x = q_113_cast_fp16)[name = tensor("op_24879_cast_fp16")]; + tensor var_24883_begin_0 = const()[name = tensor("op_24883_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_24883_end_0 = const()[name = tensor("op_24883_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_24883_end_mask_0 = const()[name = tensor("op_24883_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24883_cast_fp16 = slice_by_index(begin = var_24883_begin_0, end = var_24883_end_0, end_mask = var_24883_end_mask_0, x = q_113_cast_fp16)[name = tensor("op_24883_cast_fp16")]; + tensor var_24887_begin_0 = const()[name = tensor("op_24887_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_24887_end_0 = const()[name = tensor("op_24887_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_24887_end_mask_0 = const()[name = tensor("op_24887_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24887_cast_fp16 = slice_by_index(begin = var_24887_begin_0, end = var_24887_end_0, end_mask = var_24887_end_mask_0, x = q_113_cast_fp16)[name = tensor("op_24887_cast_fp16")]; + tensor var_24891_begin_0 = const()[name = tensor("op_24891_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_24891_end_0 = const()[name = tensor("op_24891_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_24891_end_mask_0 = const()[name = tensor("op_24891_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24891_cast_fp16 = slice_by_index(begin = var_24891_begin_0, end = var_24891_end_0, end_mask = var_24891_end_mask_0, x = q_113_cast_fp16)[name = tensor("op_24891_cast_fp16")]; + tensor var_24895_begin_0 = const()[name = tensor("op_24895_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_24895_end_0 = const()[name = tensor("op_24895_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_24895_end_mask_0 = const()[name = tensor("op_24895_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24895_cast_fp16 = slice_by_index(begin = var_24895_begin_0, end = var_24895_end_0, end_mask = var_24895_end_mask_0, x = q_113_cast_fp16)[name = tensor("op_24895_cast_fp16")]; + tensor var_24899_begin_0 = const()[name = tensor("op_24899_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_24899_end_0 = const()[name = tensor("op_24899_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_24899_end_mask_0 = const()[name = tensor("op_24899_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24899_cast_fp16 = slice_by_index(begin = var_24899_begin_0, end = var_24899_end_0, end_mask = var_24899_end_mask_0, x = q_113_cast_fp16)[name = tensor("op_24899_cast_fp16")]; + tensor var_24903_begin_0 = const()[name = tensor("op_24903_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_24903_end_0 = const()[name = tensor("op_24903_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_24903_end_mask_0 = const()[name = tensor("op_24903_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24903_cast_fp16 = slice_by_index(begin = var_24903_begin_0, end = var_24903_end_0, end_mask = var_24903_end_mask_0, x = q_113_cast_fp16)[name = tensor("op_24903_cast_fp16")]; + tensor var_24907_begin_0 = const()[name = tensor("op_24907_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_24907_end_0 = const()[name = tensor("op_24907_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_24907_end_mask_0 = const()[name = tensor("op_24907_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24907_cast_fp16 = slice_by_index(begin = var_24907_begin_0, end = var_24907_end_0, end_mask = var_24907_end_mask_0, x = q_113_cast_fp16)[name = tensor("op_24907_cast_fp16")]; + tensor var_24911_begin_0 = const()[name = tensor("op_24911_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_24911_end_0 = const()[name = tensor("op_24911_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_24911_end_mask_0 = const()[name = tensor("op_24911_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24911_cast_fp16 = slice_by_index(begin = var_24911_begin_0, end = var_24911_end_0, end_mask = var_24911_end_mask_0, x = q_113_cast_fp16)[name = tensor("op_24911_cast_fp16")]; + tensor var_24915_begin_0 = const()[name = tensor("op_24915_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_24915_end_0 = const()[name = tensor("op_24915_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_24915_end_mask_0 = const()[name = tensor("op_24915_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24915_cast_fp16 = slice_by_index(begin = var_24915_begin_0, end = var_24915_end_0, end_mask = var_24915_end_mask_0, x = q_113_cast_fp16)[name = tensor("op_24915_cast_fp16")]; + tensor var_24919_begin_0 = const()[name = tensor("op_24919_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_24919_end_0 = const()[name = tensor("op_24919_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_24919_end_mask_0 = const()[name = tensor("op_24919_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24919_cast_fp16 = slice_by_index(begin = var_24919_begin_0, end = var_24919_end_0, end_mask = var_24919_end_mask_0, x = q_113_cast_fp16)[name = tensor("op_24919_cast_fp16")]; + tensor var_24923_begin_0 = const()[name = tensor("op_24923_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_24923_end_0 = const()[name = tensor("op_24923_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_24923_end_mask_0 = const()[name = tensor("op_24923_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24923_cast_fp16 = slice_by_index(begin = var_24923_begin_0, end = var_24923_end_0, end_mask = var_24923_end_mask_0, x = q_113_cast_fp16)[name = tensor("op_24923_cast_fp16")]; + tensor var_24927_begin_0 = const()[name = tensor("op_24927_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_24927_end_0 = const()[name = tensor("op_24927_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_24927_end_mask_0 = const()[name = tensor("op_24927_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24927_cast_fp16 = slice_by_index(begin = var_24927_begin_0, end = var_24927_end_0, end_mask = var_24927_end_mask_0, x = q_113_cast_fp16)[name = tensor("op_24927_cast_fp16")]; + tensor var_24931_begin_0 = const()[name = tensor("op_24931_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_24931_end_0 = const()[name = tensor("op_24931_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_24931_end_mask_0 = const()[name = tensor("op_24931_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24931_cast_fp16 = slice_by_index(begin = var_24931_begin_0, end = var_24931_end_0, end_mask = var_24931_end_mask_0, x = q_113_cast_fp16)[name = tensor("op_24931_cast_fp16")]; + tensor k_227_perm_0 = const()[name = tensor("k_227_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_24938_begin_0 = const()[name = tensor("op_24938_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_24938_end_0 = const()[name = tensor("op_24938_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_24938_end_mask_0 = const()[name = tensor("op_24938_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_227_cast_fp16 = transpose(perm = k_227_perm_0, x = k_225_cast_fp16)[name = tensor("transpose_83")]; + tensor var_24938_cast_fp16 = slice_by_index(begin = var_24938_begin_0, end = var_24938_end_0, end_mask = var_24938_end_mask_0, x = k_227_cast_fp16)[name = tensor("op_24938_cast_fp16")]; + tensor var_24942_begin_0 = const()[name = tensor("op_24942_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_24942_end_0 = const()[name = tensor("op_24942_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_24942_end_mask_0 = const()[name = tensor("op_24942_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24942_cast_fp16 = slice_by_index(begin = var_24942_begin_0, end = var_24942_end_0, end_mask = var_24942_end_mask_0, x = k_227_cast_fp16)[name = tensor("op_24942_cast_fp16")]; + tensor var_24946_begin_0 = const()[name = tensor("op_24946_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_24946_end_0 = const()[name = tensor("op_24946_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_24946_end_mask_0 = const()[name = tensor("op_24946_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24946_cast_fp16 = slice_by_index(begin = var_24946_begin_0, end = var_24946_end_0, end_mask = var_24946_end_mask_0, x = k_227_cast_fp16)[name = tensor("op_24946_cast_fp16")]; + tensor var_24950_begin_0 = const()[name = tensor("op_24950_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_24950_end_0 = const()[name = tensor("op_24950_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_24950_end_mask_0 = const()[name = tensor("op_24950_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24950_cast_fp16 = slice_by_index(begin = var_24950_begin_0, end = var_24950_end_0, end_mask = var_24950_end_mask_0, x = k_227_cast_fp16)[name = tensor("op_24950_cast_fp16")]; + tensor var_24954_begin_0 = const()[name = tensor("op_24954_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_24954_end_0 = const()[name = tensor("op_24954_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_24954_end_mask_0 = const()[name = tensor("op_24954_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24954_cast_fp16 = slice_by_index(begin = var_24954_begin_0, end = var_24954_end_0, end_mask = var_24954_end_mask_0, x = k_227_cast_fp16)[name = tensor("op_24954_cast_fp16")]; + tensor var_24958_begin_0 = const()[name = tensor("op_24958_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_24958_end_0 = const()[name = tensor("op_24958_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_24958_end_mask_0 = const()[name = tensor("op_24958_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24958_cast_fp16 = slice_by_index(begin = var_24958_begin_0, end = var_24958_end_0, end_mask = var_24958_end_mask_0, x = k_227_cast_fp16)[name = tensor("op_24958_cast_fp16")]; + tensor var_24962_begin_0 = const()[name = tensor("op_24962_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_24962_end_0 = const()[name = tensor("op_24962_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_24962_end_mask_0 = const()[name = tensor("op_24962_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24962_cast_fp16 = slice_by_index(begin = var_24962_begin_0, end = var_24962_end_0, end_mask = var_24962_end_mask_0, x = k_227_cast_fp16)[name = tensor("op_24962_cast_fp16")]; + tensor var_24966_begin_0 = const()[name = tensor("op_24966_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_24966_end_0 = const()[name = tensor("op_24966_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_24966_end_mask_0 = const()[name = tensor("op_24966_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24966_cast_fp16 = slice_by_index(begin = var_24966_begin_0, end = var_24966_end_0, end_mask = var_24966_end_mask_0, x = k_227_cast_fp16)[name = tensor("op_24966_cast_fp16")]; + tensor var_24970_begin_0 = const()[name = tensor("op_24970_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_24970_end_0 = const()[name = tensor("op_24970_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_24970_end_mask_0 = const()[name = tensor("op_24970_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24970_cast_fp16 = slice_by_index(begin = var_24970_begin_0, end = var_24970_end_0, end_mask = var_24970_end_mask_0, x = k_227_cast_fp16)[name = tensor("op_24970_cast_fp16")]; + tensor var_24974_begin_0 = const()[name = tensor("op_24974_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_24974_end_0 = const()[name = tensor("op_24974_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_24974_end_mask_0 = const()[name = tensor("op_24974_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24974_cast_fp16 = slice_by_index(begin = var_24974_begin_0, end = var_24974_end_0, end_mask = var_24974_end_mask_0, x = k_227_cast_fp16)[name = tensor("op_24974_cast_fp16")]; + tensor var_24978_begin_0 = const()[name = tensor("op_24978_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_24978_end_0 = const()[name = tensor("op_24978_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_24978_end_mask_0 = const()[name = tensor("op_24978_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24978_cast_fp16 = slice_by_index(begin = var_24978_begin_0, end = var_24978_end_0, end_mask = var_24978_end_mask_0, x = k_227_cast_fp16)[name = tensor("op_24978_cast_fp16")]; + tensor var_24982_begin_0 = const()[name = tensor("op_24982_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_24982_end_0 = const()[name = tensor("op_24982_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_24982_end_mask_0 = const()[name = tensor("op_24982_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24982_cast_fp16 = slice_by_index(begin = var_24982_begin_0, end = var_24982_end_0, end_mask = var_24982_end_mask_0, x = k_227_cast_fp16)[name = tensor("op_24982_cast_fp16")]; + tensor var_24986_begin_0 = const()[name = tensor("op_24986_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_24986_end_0 = const()[name = tensor("op_24986_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_24986_end_mask_0 = const()[name = tensor("op_24986_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24986_cast_fp16 = slice_by_index(begin = var_24986_begin_0, end = var_24986_end_0, end_mask = var_24986_end_mask_0, x = k_227_cast_fp16)[name = tensor("op_24986_cast_fp16")]; + tensor var_24990_begin_0 = const()[name = tensor("op_24990_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_24990_end_0 = const()[name = tensor("op_24990_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_24990_end_mask_0 = const()[name = tensor("op_24990_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24990_cast_fp16 = slice_by_index(begin = var_24990_begin_0, end = var_24990_end_0, end_mask = var_24990_end_mask_0, x = k_227_cast_fp16)[name = tensor("op_24990_cast_fp16")]; + tensor var_24994_begin_0 = const()[name = tensor("op_24994_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_24994_end_0 = const()[name = tensor("op_24994_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_24994_end_mask_0 = const()[name = tensor("op_24994_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24994_cast_fp16 = slice_by_index(begin = var_24994_begin_0, end = var_24994_end_0, end_mask = var_24994_end_mask_0, x = k_227_cast_fp16)[name = tensor("op_24994_cast_fp16")]; + tensor var_24998_begin_0 = const()[name = tensor("op_24998_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_24998_end_0 = const()[name = tensor("op_24998_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_24998_end_mask_0 = const()[name = tensor("op_24998_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24998_cast_fp16 = slice_by_index(begin = var_24998_begin_0, end = var_24998_end_0, end_mask = var_24998_end_mask_0, x = k_227_cast_fp16)[name = tensor("op_24998_cast_fp16")]; + tensor var_25002_begin_0 = const()[name = tensor("op_25002_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_25002_end_0 = const()[name = tensor("op_25002_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_25002_end_mask_0 = const()[name = tensor("op_25002_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25002_cast_fp16 = slice_by_index(begin = var_25002_begin_0, end = var_25002_end_0, end_mask = var_25002_end_mask_0, x = k_227_cast_fp16)[name = tensor("op_25002_cast_fp16")]; + tensor var_25006_begin_0 = const()[name = tensor("op_25006_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_25006_end_0 = const()[name = tensor("op_25006_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_25006_end_mask_0 = const()[name = tensor("op_25006_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25006_cast_fp16 = slice_by_index(begin = var_25006_begin_0, end = var_25006_end_0, end_mask = var_25006_end_mask_0, x = k_227_cast_fp16)[name = tensor("op_25006_cast_fp16")]; + tensor var_25010_begin_0 = const()[name = tensor("op_25010_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_25010_end_0 = const()[name = tensor("op_25010_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_25010_end_mask_0 = const()[name = tensor("op_25010_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25010_cast_fp16 = slice_by_index(begin = var_25010_begin_0, end = var_25010_end_0, end_mask = var_25010_end_mask_0, x = k_227_cast_fp16)[name = tensor("op_25010_cast_fp16")]; + tensor var_25014_begin_0 = const()[name = tensor("op_25014_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_25014_end_0 = const()[name = tensor("op_25014_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_25014_end_mask_0 = const()[name = tensor("op_25014_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25014_cast_fp16 = slice_by_index(begin = var_25014_begin_0, end = var_25014_end_0, end_mask = var_25014_end_mask_0, x = k_227_cast_fp16)[name = tensor("op_25014_cast_fp16")]; + tensor var_25016_begin_0 = const()[name = tensor("op_25016_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_25016_end_0 = const()[name = tensor("op_25016_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_25016_end_mask_0 = const()[name = tensor("op_25016_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25016_cast_fp16 = slice_by_index(begin = var_25016_begin_0, end = var_25016_end_0, end_mask = var_25016_end_mask_0, x = v_113_cast_fp16)[name = tensor("op_25016_cast_fp16")]; + tensor var_25020_begin_0 = const()[name = tensor("op_25020_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_25020_end_0 = const()[name = tensor("op_25020_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_25020_end_mask_0 = const()[name = tensor("op_25020_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25020_cast_fp16 = slice_by_index(begin = var_25020_begin_0, end = var_25020_end_0, end_mask = var_25020_end_mask_0, x = v_113_cast_fp16)[name = tensor("op_25020_cast_fp16")]; + tensor var_25024_begin_0 = const()[name = tensor("op_25024_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_25024_end_0 = const()[name = tensor("op_25024_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_25024_end_mask_0 = const()[name = tensor("op_25024_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25024_cast_fp16 = slice_by_index(begin = var_25024_begin_0, end = var_25024_end_0, end_mask = var_25024_end_mask_0, x = v_113_cast_fp16)[name = tensor("op_25024_cast_fp16")]; + tensor var_25028_begin_0 = const()[name = tensor("op_25028_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_25028_end_0 = const()[name = tensor("op_25028_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_25028_end_mask_0 = const()[name = tensor("op_25028_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25028_cast_fp16 = slice_by_index(begin = var_25028_begin_0, end = var_25028_end_0, end_mask = var_25028_end_mask_0, x = v_113_cast_fp16)[name = tensor("op_25028_cast_fp16")]; + tensor var_25032_begin_0 = const()[name = tensor("op_25032_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_25032_end_0 = const()[name = tensor("op_25032_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_25032_end_mask_0 = const()[name = tensor("op_25032_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25032_cast_fp16 = slice_by_index(begin = var_25032_begin_0, end = var_25032_end_0, end_mask = var_25032_end_mask_0, x = v_113_cast_fp16)[name = tensor("op_25032_cast_fp16")]; + tensor var_25036_begin_0 = const()[name = tensor("op_25036_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_25036_end_0 = const()[name = tensor("op_25036_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_25036_end_mask_0 = const()[name = tensor("op_25036_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25036_cast_fp16 = slice_by_index(begin = var_25036_begin_0, end = var_25036_end_0, end_mask = var_25036_end_mask_0, x = v_113_cast_fp16)[name = tensor("op_25036_cast_fp16")]; + tensor var_25040_begin_0 = const()[name = tensor("op_25040_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_25040_end_0 = const()[name = tensor("op_25040_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_25040_end_mask_0 = const()[name = tensor("op_25040_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25040_cast_fp16 = slice_by_index(begin = var_25040_begin_0, end = var_25040_end_0, end_mask = var_25040_end_mask_0, x = v_113_cast_fp16)[name = tensor("op_25040_cast_fp16")]; + tensor var_25044_begin_0 = const()[name = tensor("op_25044_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_25044_end_0 = const()[name = tensor("op_25044_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_25044_end_mask_0 = const()[name = tensor("op_25044_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25044_cast_fp16 = slice_by_index(begin = var_25044_begin_0, end = var_25044_end_0, end_mask = var_25044_end_mask_0, x = v_113_cast_fp16)[name = tensor("op_25044_cast_fp16")]; + tensor var_25048_begin_0 = const()[name = tensor("op_25048_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_25048_end_0 = const()[name = tensor("op_25048_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_25048_end_mask_0 = const()[name = tensor("op_25048_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25048_cast_fp16 = slice_by_index(begin = var_25048_begin_0, end = var_25048_end_0, end_mask = var_25048_end_mask_0, x = v_113_cast_fp16)[name = tensor("op_25048_cast_fp16")]; + tensor var_25052_begin_0 = const()[name = tensor("op_25052_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_25052_end_0 = const()[name = tensor("op_25052_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_25052_end_mask_0 = const()[name = tensor("op_25052_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25052_cast_fp16 = slice_by_index(begin = var_25052_begin_0, end = var_25052_end_0, end_mask = var_25052_end_mask_0, x = v_113_cast_fp16)[name = tensor("op_25052_cast_fp16")]; + tensor var_25056_begin_0 = const()[name = tensor("op_25056_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_25056_end_0 = const()[name = tensor("op_25056_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_25056_end_mask_0 = const()[name = tensor("op_25056_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25056_cast_fp16 = slice_by_index(begin = var_25056_begin_0, end = var_25056_end_0, end_mask = var_25056_end_mask_0, x = v_113_cast_fp16)[name = tensor("op_25056_cast_fp16")]; + tensor var_25060_begin_0 = const()[name = tensor("op_25060_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_25060_end_0 = const()[name = tensor("op_25060_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_25060_end_mask_0 = const()[name = tensor("op_25060_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25060_cast_fp16 = slice_by_index(begin = var_25060_begin_0, end = var_25060_end_0, end_mask = var_25060_end_mask_0, x = v_113_cast_fp16)[name = tensor("op_25060_cast_fp16")]; + tensor var_25064_begin_0 = const()[name = tensor("op_25064_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_25064_end_0 = const()[name = tensor("op_25064_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_25064_end_mask_0 = const()[name = tensor("op_25064_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25064_cast_fp16 = slice_by_index(begin = var_25064_begin_0, end = var_25064_end_0, end_mask = var_25064_end_mask_0, x = v_113_cast_fp16)[name = tensor("op_25064_cast_fp16")]; + tensor var_25068_begin_0 = const()[name = tensor("op_25068_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_25068_end_0 = const()[name = tensor("op_25068_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_25068_end_mask_0 = const()[name = tensor("op_25068_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25068_cast_fp16 = slice_by_index(begin = var_25068_begin_0, end = var_25068_end_0, end_mask = var_25068_end_mask_0, x = v_113_cast_fp16)[name = tensor("op_25068_cast_fp16")]; + tensor var_25072_begin_0 = const()[name = tensor("op_25072_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_25072_end_0 = const()[name = tensor("op_25072_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_25072_end_mask_0 = const()[name = tensor("op_25072_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25072_cast_fp16 = slice_by_index(begin = var_25072_begin_0, end = var_25072_end_0, end_mask = var_25072_end_mask_0, x = v_113_cast_fp16)[name = tensor("op_25072_cast_fp16")]; + tensor var_25076_begin_0 = const()[name = tensor("op_25076_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_25076_end_0 = const()[name = tensor("op_25076_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_25076_end_mask_0 = const()[name = tensor("op_25076_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25076_cast_fp16 = slice_by_index(begin = var_25076_begin_0, end = var_25076_end_0, end_mask = var_25076_end_mask_0, x = v_113_cast_fp16)[name = tensor("op_25076_cast_fp16")]; + tensor var_25080_begin_0 = const()[name = tensor("op_25080_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_25080_end_0 = const()[name = tensor("op_25080_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_25080_end_mask_0 = const()[name = tensor("op_25080_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25080_cast_fp16 = slice_by_index(begin = var_25080_begin_0, end = var_25080_end_0, end_mask = var_25080_end_mask_0, x = v_113_cast_fp16)[name = tensor("op_25080_cast_fp16")]; + tensor var_25084_begin_0 = const()[name = tensor("op_25084_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_25084_end_0 = const()[name = tensor("op_25084_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_25084_end_mask_0 = const()[name = tensor("op_25084_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25084_cast_fp16 = slice_by_index(begin = var_25084_begin_0, end = var_25084_end_0, end_mask = var_25084_end_mask_0, x = v_113_cast_fp16)[name = tensor("op_25084_cast_fp16")]; + tensor var_25088_begin_0 = const()[name = tensor("op_25088_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_25088_end_0 = const()[name = tensor("op_25088_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_25088_end_mask_0 = const()[name = tensor("op_25088_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25088_cast_fp16 = slice_by_index(begin = var_25088_begin_0, end = var_25088_end_0, end_mask = var_25088_end_mask_0, x = v_113_cast_fp16)[name = tensor("op_25088_cast_fp16")]; + tensor var_25092_begin_0 = const()[name = tensor("op_25092_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_25092_end_0 = const()[name = tensor("op_25092_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_25092_end_mask_0 = const()[name = tensor("op_25092_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25092_cast_fp16 = slice_by_index(begin = var_25092_begin_0, end = var_25092_end_0, end_mask = var_25092_end_mask_0, x = v_113_cast_fp16)[name = tensor("op_25092_cast_fp16")]; + tensor var_25096_equation_0 = const()[name = tensor("op_25096_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25096_cast_fp16 = einsum(equation = var_25096_equation_0, values = (var_24938_cast_fp16, var_24855_cast_fp16))[name = tensor("op_25096_cast_fp16")]; + tensor var_25097_to_fp16 = const()[name = tensor("op_25097_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2081_cast_fp16 = mul(x = var_25096_cast_fp16, y = var_25097_to_fp16)[name = tensor("aw_2081_cast_fp16")]; + tensor var_25100_equation_0 = const()[name = tensor("op_25100_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25100_cast_fp16 = einsum(equation = var_25100_equation_0, values = (var_24942_cast_fp16, var_24859_cast_fp16))[name = tensor("op_25100_cast_fp16")]; + tensor var_25101_to_fp16 = const()[name = tensor("op_25101_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2083_cast_fp16 = mul(x = var_25100_cast_fp16, y = var_25101_to_fp16)[name = tensor("aw_2083_cast_fp16")]; + tensor var_25104_equation_0 = const()[name = tensor("op_25104_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25104_cast_fp16 = einsum(equation = var_25104_equation_0, values = (var_24946_cast_fp16, var_24863_cast_fp16))[name = tensor("op_25104_cast_fp16")]; + tensor var_25105_to_fp16 = const()[name = tensor("op_25105_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2085_cast_fp16 = mul(x = var_25104_cast_fp16, y = var_25105_to_fp16)[name = tensor("aw_2085_cast_fp16")]; + tensor var_25108_equation_0 = const()[name = tensor("op_25108_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25108_cast_fp16 = einsum(equation = var_25108_equation_0, values = (var_24950_cast_fp16, var_24867_cast_fp16))[name = tensor("op_25108_cast_fp16")]; + tensor var_25109_to_fp16 = const()[name = tensor("op_25109_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2087_cast_fp16 = mul(x = var_25108_cast_fp16, y = var_25109_to_fp16)[name = tensor("aw_2087_cast_fp16")]; + tensor var_25112_equation_0 = const()[name = tensor("op_25112_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25112_cast_fp16 = einsum(equation = var_25112_equation_0, values = (var_24954_cast_fp16, var_24871_cast_fp16))[name = tensor("op_25112_cast_fp16")]; + tensor var_25113_to_fp16 = const()[name = tensor("op_25113_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2089_cast_fp16 = mul(x = var_25112_cast_fp16, y = var_25113_to_fp16)[name = tensor("aw_2089_cast_fp16")]; + tensor var_25116_equation_0 = const()[name = tensor("op_25116_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25116_cast_fp16 = einsum(equation = var_25116_equation_0, values = (var_24958_cast_fp16, var_24875_cast_fp16))[name = tensor("op_25116_cast_fp16")]; + tensor var_25117_to_fp16 = const()[name = tensor("op_25117_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2091_cast_fp16 = mul(x = var_25116_cast_fp16, y = var_25117_to_fp16)[name = tensor("aw_2091_cast_fp16")]; + tensor var_25120_equation_0 = const()[name = tensor("op_25120_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25120_cast_fp16 = einsum(equation = var_25120_equation_0, values = (var_24962_cast_fp16, var_24879_cast_fp16))[name = tensor("op_25120_cast_fp16")]; + tensor var_25121_to_fp16 = const()[name = tensor("op_25121_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2093_cast_fp16 = mul(x = var_25120_cast_fp16, y = var_25121_to_fp16)[name = tensor("aw_2093_cast_fp16")]; + tensor var_25124_equation_0 = const()[name = tensor("op_25124_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25124_cast_fp16 = einsum(equation = var_25124_equation_0, values = (var_24966_cast_fp16, var_24883_cast_fp16))[name = tensor("op_25124_cast_fp16")]; + tensor var_25125_to_fp16 = const()[name = tensor("op_25125_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2095_cast_fp16 = mul(x = var_25124_cast_fp16, y = var_25125_to_fp16)[name = tensor("aw_2095_cast_fp16")]; + tensor var_25128_equation_0 = const()[name = tensor("op_25128_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25128_cast_fp16 = einsum(equation = var_25128_equation_0, values = (var_24970_cast_fp16, var_24887_cast_fp16))[name = tensor("op_25128_cast_fp16")]; + tensor var_25129_to_fp16 = const()[name = tensor("op_25129_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2097_cast_fp16 = mul(x = var_25128_cast_fp16, y = var_25129_to_fp16)[name = tensor("aw_2097_cast_fp16")]; + tensor var_25132_equation_0 = const()[name = tensor("op_25132_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25132_cast_fp16 = einsum(equation = var_25132_equation_0, values = (var_24974_cast_fp16, var_24891_cast_fp16))[name = tensor("op_25132_cast_fp16")]; + tensor var_25133_to_fp16 = const()[name = tensor("op_25133_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2099_cast_fp16 = mul(x = var_25132_cast_fp16, y = var_25133_to_fp16)[name = tensor("aw_2099_cast_fp16")]; + tensor var_25136_equation_0 = const()[name = tensor("op_25136_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25136_cast_fp16 = einsum(equation = var_25136_equation_0, values = (var_24978_cast_fp16, var_24895_cast_fp16))[name = tensor("op_25136_cast_fp16")]; + tensor var_25137_to_fp16 = const()[name = tensor("op_25137_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2101_cast_fp16 = mul(x = var_25136_cast_fp16, y = var_25137_to_fp16)[name = tensor("aw_2101_cast_fp16")]; + tensor var_25140_equation_0 = const()[name = tensor("op_25140_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25140_cast_fp16 = einsum(equation = var_25140_equation_0, values = (var_24982_cast_fp16, var_24899_cast_fp16))[name = tensor("op_25140_cast_fp16")]; + tensor var_25141_to_fp16 = const()[name = tensor("op_25141_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2103_cast_fp16 = mul(x = var_25140_cast_fp16, y = var_25141_to_fp16)[name = tensor("aw_2103_cast_fp16")]; + tensor var_25144_equation_0 = const()[name = tensor("op_25144_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25144_cast_fp16 = einsum(equation = var_25144_equation_0, values = (var_24986_cast_fp16, var_24903_cast_fp16))[name = tensor("op_25144_cast_fp16")]; + tensor var_25145_to_fp16 = const()[name = tensor("op_25145_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2105_cast_fp16 = mul(x = var_25144_cast_fp16, y = var_25145_to_fp16)[name = tensor("aw_2105_cast_fp16")]; + tensor var_25148_equation_0 = const()[name = tensor("op_25148_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25148_cast_fp16 = einsum(equation = var_25148_equation_0, values = (var_24990_cast_fp16, var_24907_cast_fp16))[name = tensor("op_25148_cast_fp16")]; + tensor var_25149_to_fp16 = const()[name = tensor("op_25149_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2107_cast_fp16 = mul(x = var_25148_cast_fp16, y = var_25149_to_fp16)[name = tensor("aw_2107_cast_fp16")]; + tensor var_25152_equation_0 = const()[name = tensor("op_25152_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25152_cast_fp16 = einsum(equation = var_25152_equation_0, values = (var_24994_cast_fp16, var_24911_cast_fp16))[name = tensor("op_25152_cast_fp16")]; + tensor var_25153_to_fp16 = const()[name = tensor("op_25153_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2109_cast_fp16 = mul(x = var_25152_cast_fp16, y = var_25153_to_fp16)[name = tensor("aw_2109_cast_fp16")]; + tensor var_25156_equation_0 = const()[name = tensor("op_25156_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25156_cast_fp16 = einsum(equation = var_25156_equation_0, values = (var_24998_cast_fp16, var_24915_cast_fp16))[name = tensor("op_25156_cast_fp16")]; + tensor var_25157_to_fp16 = const()[name = tensor("op_25157_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2111_cast_fp16 = mul(x = var_25156_cast_fp16, y = var_25157_to_fp16)[name = tensor("aw_2111_cast_fp16")]; + tensor var_25160_equation_0 = const()[name = tensor("op_25160_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25160_cast_fp16 = einsum(equation = var_25160_equation_0, values = (var_25002_cast_fp16, var_24919_cast_fp16))[name = tensor("op_25160_cast_fp16")]; + tensor var_25161_to_fp16 = const()[name = tensor("op_25161_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2113_cast_fp16 = mul(x = var_25160_cast_fp16, y = var_25161_to_fp16)[name = tensor("aw_2113_cast_fp16")]; + tensor var_25164_equation_0 = const()[name = tensor("op_25164_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25164_cast_fp16 = einsum(equation = var_25164_equation_0, values = (var_25006_cast_fp16, var_24923_cast_fp16))[name = tensor("op_25164_cast_fp16")]; + tensor var_25165_to_fp16 = const()[name = tensor("op_25165_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2115_cast_fp16 = mul(x = var_25164_cast_fp16, y = var_25165_to_fp16)[name = tensor("aw_2115_cast_fp16")]; + tensor var_25168_equation_0 = const()[name = tensor("op_25168_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25168_cast_fp16 = einsum(equation = var_25168_equation_0, values = (var_25010_cast_fp16, var_24927_cast_fp16))[name = tensor("op_25168_cast_fp16")]; + tensor var_25169_to_fp16 = const()[name = tensor("op_25169_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2117_cast_fp16 = mul(x = var_25168_cast_fp16, y = var_25169_to_fp16)[name = tensor("aw_2117_cast_fp16")]; + tensor var_25172_equation_0 = const()[name = tensor("op_25172_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25172_cast_fp16 = einsum(equation = var_25172_equation_0, values = (var_25014_cast_fp16, var_24931_cast_fp16))[name = tensor("op_25172_cast_fp16")]; + tensor var_25173_to_fp16 = const()[name = tensor("op_25173_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2119_cast_fp16 = mul(x = var_25172_cast_fp16, y = var_25173_to_fp16)[name = tensor("aw_2119_cast_fp16")]; + tensor var_25175_cast_fp16 = softmax(axis = var_21077, x = aw_2081_cast_fp16)[name = tensor("op_25175_cast_fp16")]; + tensor var_25176_cast_fp16 = softmax(axis = var_21077, x = aw_2083_cast_fp16)[name = tensor("op_25176_cast_fp16")]; + tensor var_25177_cast_fp16 = softmax(axis = var_21077, x = aw_2085_cast_fp16)[name = tensor("op_25177_cast_fp16")]; + tensor var_25178_cast_fp16 = softmax(axis = var_21077, x = aw_2087_cast_fp16)[name = tensor("op_25178_cast_fp16")]; + tensor var_25179_cast_fp16 = softmax(axis = var_21077, x = aw_2089_cast_fp16)[name = tensor("op_25179_cast_fp16")]; + tensor var_25180_cast_fp16 = softmax(axis = var_21077, x = aw_2091_cast_fp16)[name = tensor("op_25180_cast_fp16")]; + tensor var_25181_cast_fp16 = softmax(axis = var_21077, x = aw_2093_cast_fp16)[name = tensor("op_25181_cast_fp16")]; + tensor var_25182_cast_fp16 = softmax(axis = var_21077, x = aw_2095_cast_fp16)[name = tensor("op_25182_cast_fp16")]; + tensor var_25183_cast_fp16 = softmax(axis = var_21077, x = aw_2097_cast_fp16)[name = tensor("op_25183_cast_fp16")]; + tensor var_25184_cast_fp16 = softmax(axis = var_21077, x = aw_2099_cast_fp16)[name = tensor("op_25184_cast_fp16")]; + tensor var_25185_cast_fp16 = softmax(axis = var_21077, x = aw_2101_cast_fp16)[name = tensor("op_25185_cast_fp16")]; + tensor var_25186_cast_fp16 = softmax(axis = var_21077, x = aw_2103_cast_fp16)[name = tensor("op_25186_cast_fp16")]; + tensor var_25187_cast_fp16 = softmax(axis = var_21077, x = aw_2105_cast_fp16)[name = tensor("op_25187_cast_fp16")]; + tensor var_25188_cast_fp16 = softmax(axis = var_21077, x = aw_2107_cast_fp16)[name = tensor("op_25188_cast_fp16")]; + tensor var_25189_cast_fp16 = softmax(axis = var_21077, x = aw_2109_cast_fp16)[name = tensor("op_25189_cast_fp16")]; + tensor var_25190_cast_fp16 = softmax(axis = var_21077, x = aw_2111_cast_fp16)[name = tensor("op_25190_cast_fp16")]; + tensor var_25191_cast_fp16 = softmax(axis = var_21077, x = aw_2113_cast_fp16)[name = tensor("op_25191_cast_fp16")]; + tensor var_25192_cast_fp16 = softmax(axis = var_21077, x = aw_2115_cast_fp16)[name = tensor("op_25192_cast_fp16")]; + tensor var_25193_cast_fp16 = softmax(axis = var_21077, x = aw_2117_cast_fp16)[name = tensor("op_25193_cast_fp16")]; + tensor var_25194_cast_fp16 = softmax(axis = var_21077, x = aw_2119_cast_fp16)[name = tensor("op_25194_cast_fp16")]; + tensor var_25196_equation_0 = const()[name = tensor("op_25196_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25196_cast_fp16 = einsum(equation = var_25196_equation_0, values = (var_25016_cast_fp16, var_25175_cast_fp16))[name = tensor("op_25196_cast_fp16")]; + tensor var_25198_equation_0 = const()[name = tensor("op_25198_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25198_cast_fp16 = einsum(equation = var_25198_equation_0, values = (var_25020_cast_fp16, var_25176_cast_fp16))[name = tensor("op_25198_cast_fp16")]; + tensor var_25200_equation_0 = const()[name = tensor("op_25200_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25200_cast_fp16 = einsum(equation = var_25200_equation_0, values = (var_25024_cast_fp16, var_25177_cast_fp16))[name = tensor("op_25200_cast_fp16")]; + tensor var_25202_equation_0 = const()[name = tensor("op_25202_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25202_cast_fp16 = einsum(equation = var_25202_equation_0, values = (var_25028_cast_fp16, var_25178_cast_fp16))[name = tensor("op_25202_cast_fp16")]; + tensor var_25204_equation_0 = const()[name = tensor("op_25204_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25204_cast_fp16 = einsum(equation = var_25204_equation_0, values = (var_25032_cast_fp16, var_25179_cast_fp16))[name = tensor("op_25204_cast_fp16")]; + tensor var_25206_equation_0 = const()[name = tensor("op_25206_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25206_cast_fp16 = einsum(equation = var_25206_equation_0, values = (var_25036_cast_fp16, var_25180_cast_fp16))[name = tensor("op_25206_cast_fp16")]; + tensor var_25208_equation_0 = const()[name = tensor("op_25208_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25208_cast_fp16 = einsum(equation = var_25208_equation_0, values = (var_25040_cast_fp16, var_25181_cast_fp16))[name = tensor("op_25208_cast_fp16")]; + tensor var_25210_equation_0 = const()[name = tensor("op_25210_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25210_cast_fp16 = einsum(equation = var_25210_equation_0, values = (var_25044_cast_fp16, var_25182_cast_fp16))[name = tensor("op_25210_cast_fp16")]; + tensor var_25212_equation_0 = const()[name = tensor("op_25212_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25212_cast_fp16 = einsum(equation = var_25212_equation_0, values = (var_25048_cast_fp16, var_25183_cast_fp16))[name = tensor("op_25212_cast_fp16")]; + tensor var_25214_equation_0 = const()[name = tensor("op_25214_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25214_cast_fp16 = einsum(equation = var_25214_equation_0, values = (var_25052_cast_fp16, var_25184_cast_fp16))[name = tensor("op_25214_cast_fp16")]; + tensor var_25216_equation_0 = const()[name = tensor("op_25216_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25216_cast_fp16 = einsum(equation = var_25216_equation_0, values = (var_25056_cast_fp16, var_25185_cast_fp16))[name = tensor("op_25216_cast_fp16")]; + tensor var_25218_equation_0 = const()[name = tensor("op_25218_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25218_cast_fp16 = einsum(equation = var_25218_equation_0, values = (var_25060_cast_fp16, var_25186_cast_fp16))[name = tensor("op_25218_cast_fp16")]; + tensor var_25220_equation_0 = const()[name = tensor("op_25220_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25220_cast_fp16 = einsum(equation = var_25220_equation_0, values = (var_25064_cast_fp16, var_25187_cast_fp16))[name = tensor("op_25220_cast_fp16")]; + tensor var_25222_equation_0 = const()[name = tensor("op_25222_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25222_cast_fp16 = einsum(equation = var_25222_equation_0, values = (var_25068_cast_fp16, var_25188_cast_fp16))[name = tensor("op_25222_cast_fp16")]; + tensor var_25224_equation_0 = const()[name = tensor("op_25224_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25224_cast_fp16 = einsum(equation = var_25224_equation_0, values = (var_25072_cast_fp16, var_25189_cast_fp16))[name = tensor("op_25224_cast_fp16")]; + tensor var_25226_equation_0 = const()[name = tensor("op_25226_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25226_cast_fp16 = einsum(equation = var_25226_equation_0, values = (var_25076_cast_fp16, var_25190_cast_fp16))[name = tensor("op_25226_cast_fp16")]; + tensor var_25228_equation_0 = const()[name = tensor("op_25228_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25228_cast_fp16 = einsum(equation = var_25228_equation_0, values = (var_25080_cast_fp16, var_25191_cast_fp16))[name = tensor("op_25228_cast_fp16")]; + tensor var_25230_equation_0 = const()[name = tensor("op_25230_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25230_cast_fp16 = einsum(equation = var_25230_equation_0, values = (var_25084_cast_fp16, var_25192_cast_fp16))[name = tensor("op_25230_cast_fp16")]; + tensor var_25232_equation_0 = const()[name = tensor("op_25232_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25232_cast_fp16 = einsum(equation = var_25232_equation_0, values = (var_25088_cast_fp16, var_25193_cast_fp16))[name = tensor("op_25232_cast_fp16")]; + tensor var_25234_equation_0 = const()[name = tensor("op_25234_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25234_cast_fp16 = einsum(equation = var_25234_equation_0, values = (var_25092_cast_fp16, var_25194_cast_fp16))[name = tensor("op_25234_cast_fp16")]; + tensor input_359_interleave_0 = const()[name = tensor("input_359_interleave_0"), val = tensor(false)]; + tensor input_359_cast_fp16 = concat(axis = var_21077, interleave = input_359_interleave_0, values = (var_25196_cast_fp16, var_25198_cast_fp16, var_25200_cast_fp16, var_25202_cast_fp16, var_25204_cast_fp16, var_25206_cast_fp16, var_25208_cast_fp16, var_25210_cast_fp16, var_25212_cast_fp16, var_25214_cast_fp16, var_25216_cast_fp16, var_25218_cast_fp16, var_25220_cast_fp16, var_25222_cast_fp16, var_25224_cast_fp16, var_25226_cast_fp16, var_25228_cast_fp16, var_25230_cast_fp16, var_25232_cast_fp16, var_25234_cast_fp16))[name = tensor("input_359_cast_fp16")]; + tensor var_25240 = const()[name = tensor("op_25240"), val = tensor([1, 1])]; + tensor var_25242 = const()[name = tensor("op_25242"), val = tensor([1, 1])]; + tensor var_25244_pad_type_0 = const()[name = tensor("op_25244_pad_type_0"), val = tensor("custom")]; + tensor var_25244_pad_0 = const()[name = tensor("op_25244_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_4_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_4_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2013996736)))]; + tensor mid_block_attentions_0_transformer_blocks_4_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_4_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2017273600)))]; + tensor var_25244_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_4_attn1_to_out_0_bias_to_fp16, dilations = var_25242, groups = var_21077, pad = var_25244_pad_0, pad_type = var_25244_pad_type_0, strides = var_25240, weight = mid_block_attentions_0_transformer_blocks_4_attn1_to_out_0_weight_to_fp16, x = input_359_cast_fp16)[name = tensor("op_25244_cast_fp16")]; + tensor inputs_171_cast_fp16 = add(x = var_25244_cast_fp16, y = inputs_169_cast_fp16)[name = tensor("inputs_171_cast_fp16")]; + tensor hidden_states_235_axes_0 = const()[name = tensor("hidden_states_235_axes_0"), val = tensor([1])]; + tensor hidden_states_235_gamma_0_to_fp16 = const()[name = tensor("hidden_states_235_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2017276224)))]; + tensor hidden_states_235_beta_0_to_fp16 = const()[name = tensor("hidden_states_235_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2017278848)))]; + tensor var_25254_to_fp16 = const()[name = tensor("op_25254_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_235_cast_fp16 = layer_norm(axes = hidden_states_235_axes_0, beta = hidden_states_235_beta_0_to_fp16, epsilon = var_25254_to_fp16, gamma = hidden_states_235_gamma_0_to_fp16, x = inputs_171_cast_fp16)[name = tensor("hidden_states_235_cast_fp16")]; + tensor var_25269 = const()[name = tensor("op_25269"), val = tensor([1, 1])]; + tensor var_25271 = const()[name = tensor("op_25271"), val = tensor([1, 1])]; + tensor q_115_pad_type_0 = const()[name = tensor("q_115_pad_type_0"), val = tensor("custom")]; + tensor q_115_pad_0 = const()[name = tensor("q_115_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_4_attn2_to_q_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_4_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2017281472)))]; + tensor q_115_cast_fp16 = conv(dilations = var_25271, groups = var_21077, pad = q_115_pad_0, pad_type = q_115_pad_type_0, strides = var_25269, weight = mid_block_attentions_0_transformer_blocks_4_attn2_to_q_weight_to_fp16, x = hidden_states_235_cast_fp16)[name = tensor("q_115_cast_fp16")]; + tensor var_25275 = const()[name = tensor("op_25275"), val = tensor([1, 1])]; + tensor var_25277 = const()[name = tensor("op_25277"), val = tensor([1, 1])]; + tensor k_229_pad_type_0 = const()[name = tensor("k_229_pad_type_0"), val = tensor("custom")]; + tensor k_229_pad_0 = const()[name = tensor("k_229_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_4_attn2_to_k_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_4_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2020558336)))]; + tensor k_229_cast_fp16 = conv(dilations = var_25277, groups = var_21077, pad = k_229_pad_0, pad_type = k_229_pad_type_0, strides = var_25275, weight = mid_block_attentions_0_transformer_blocks_4_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_229_cast_fp16")]; + tensor var_25281 = const()[name = tensor("op_25281"), val = tensor([1, 1])]; + tensor var_25283 = const()[name = tensor("op_25283"), val = tensor([1, 1])]; + tensor v_115_pad_type_0 = const()[name = tensor("v_115_pad_type_0"), val = tensor("custom")]; + tensor v_115_pad_0 = const()[name = tensor("v_115_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_4_attn2_to_v_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_4_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2025801280)))]; + tensor v_115_cast_fp16 = conv(dilations = var_25283, groups = var_21077, pad = v_115_pad_0, pad_type = v_115_pad_type_0, strides = var_25281, weight = mid_block_attentions_0_transformer_blocks_4_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_115_cast_fp16")]; + tensor var_25287_begin_0 = const()[name = tensor("op_25287_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_25287_end_0 = const()[name = tensor("op_25287_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_25287_end_mask_0 = const()[name = tensor("op_25287_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25287_cast_fp16 = slice_by_index(begin = var_25287_begin_0, end = var_25287_end_0, end_mask = var_25287_end_mask_0, x = q_115_cast_fp16)[name = tensor("op_25287_cast_fp16")]; + tensor var_25291_begin_0 = const()[name = tensor("op_25291_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_25291_end_0 = const()[name = tensor("op_25291_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_25291_end_mask_0 = const()[name = tensor("op_25291_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25291_cast_fp16 = slice_by_index(begin = var_25291_begin_0, end = var_25291_end_0, end_mask = var_25291_end_mask_0, x = q_115_cast_fp16)[name = tensor("op_25291_cast_fp16")]; + tensor var_25295_begin_0 = const()[name = tensor("op_25295_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_25295_end_0 = const()[name = tensor("op_25295_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_25295_end_mask_0 = const()[name = tensor("op_25295_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25295_cast_fp16 = slice_by_index(begin = var_25295_begin_0, end = var_25295_end_0, end_mask = var_25295_end_mask_0, x = q_115_cast_fp16)[name = tensor("op_25295_cast_fp16")]; + tensor var_25299_begin_0 = const()[name = tensor("op_25299_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_25299_end_0 = const()[name = tensor("op_25299_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_25299_end_mask_0 = const()[name = tensor("op_25299_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25299_cast_fp16 = slice_by_index(begin = var_25299_begin_0, end = var_25299_end_0, end_mask = var_25299_end_mask_0, x = q_115_cast_fp16)[name = tensor("op_25299_cast_fp16")]; + tensor var_25303_begin_0 = const()[name = tensor("op_25303_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_25303_end_0 = const()[name = tensor("op_25303_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_25303_end_mask_0 = const()[name = tensor("op_25303_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25303_cast_fp16 = slice_by_index(begin = var_25303_begin_0, end = var_25303_end_0, end_mask = var_25303_end_mask_0, x = q_115_cast_fp16)[name = tensor("op_25303_cast_fp16")]; + tensor var_25307_begin_0 = const()[name = tensor("op_25307_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_25307_end_0 = const()[name = tensor("op_25307_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_25307_end_mask_0 = const()[name = tensor("op_25307_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25307_cast_fp16 = slice_by_index(begin = var_25307_begin_0, end = var_25307_end_0, end_mask = var_25307_end_mask_0, x = q_115_cast_fp16)[name = tensor("op_25307_cast_fp16")]; + tensor var_25311_begin_0 = const()[name = tensor("op_25311_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_25311_end_0 = const()[name = tensor("op_25311_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_25311_end_mask_0 = const()[name = tensor("op_25311_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25311_cast_fp16 = slice_by_index(begin = var_25311_begin_0, end = var_25311_end_0, end_mask = var_25311_end_mask_0, x = q_115_cast_fp16)[name = tensor("op_25311_cast_fp16")]; + tensor var_25315_begin_0 = const()[name = tensor("op_25315_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_25315_end_0 = const()[name = tensor("op_25315_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_25315_end_mask_0 = const()[name = tensor("op_25315_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25315_cast_fp16 = slice_by_index(begin = var_25315_begin_0, end = var_25315_end_0, end_mask = var_25315_end_mask_0, x = q_115_cast_fp16)[name = tensor("op_25315_cast_fp16")]; + tensor var_25319_begin_0 = const()[name = tensor("op_25319_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_25319_end_0 = const()[name = tensor("op_25319_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_25319_end_mask_0 = const()[name = tensor("op_25319_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25319_cast_fp16 = slice_by_index(begin = var_25319_begin_0, end = var_25319_end_0, end_mask = var_25319_end_mask_0, x = q_115_cast_fp16)[name = tensor("op_25319_cast_fp16")]; + tensor var_25323_begin_0 = const()[name = tensor("op_25323_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_25323_end_0 = const()[name = tensor("op_25323_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_25323_end_mask_0 = const()[name = tensor("op_25323_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25323_cast_fp16 = slice_by_index(begin = var_25323_begin_0, end = var_25323_end_0, end_mask = var_25323_end_mask_0, x = q_115_cast_fp16)[name = tensor("op_25323_cast_fp16")]; + tensor var_25327_begin_0 = const()[name = tensor("op_25327_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_25327_end_0 = const()[name = tensor("op_25327_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_25327_end_mask_0 = const()[name = tensor("op_25327_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25327_cast_fp16 = slice_by_index(begin = var_25327_begin_0, end = var_25327_end_0, end_mask = var_25327_end_mask_0, x = q_115_cast_fp16)[name = tensor("op_25327_cast_fp16")]; + tensor var_25331_begin_0 = const()[name = tensor("op_25331_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_25331_end_0 = const()[name = tensor("op_25331_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_25331_end_mask_0 = const()[name = tensor("op_25331_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25331_cast_fp16 = slice_by_index(begin = var_25331_begin_0, end = var_25331_end_0, end_mask = var_25331_end_mask_0, x = q_115_cast_fp16)[name = tensor("op_25331_cast_fp16")]; + tensor var_25335_begin_0 = const()[name = tensor("op_25335_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_25335_end_0 = const()[name = tensor("op_25335_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_25335_end_mask_0 = const()[name = tensor("op_25335_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25335_cast_fp16 = slice_by_index(begin = var_25335_begin_0, end = var_25335_end_0, end_mask = var_25335_end_mask_0, x = q_115_cast_fp16)[name = tensor("op_25335_cast_fp16")]; + tensor var_25339_begin_0 = const()[name = tensor("op_25339_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_25339_end_0 = const()[name = tensor("op_25339_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_25339_end_mask_0 = const()[name = tensor("op_25339_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25339_cast_fp16 = slice_by_index(begin = var_25339_begin_0, end = var_25339_end_0, end_mask = var_25339_end_mask_0, x = q_115_cast_fp16)[name = tensor("op_25339_cast_fp16")]; + tensor var_25343_begin_0 = const()[name = tensor("op_25343_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_25343_end_0 = const()[name = tensor("op_25343_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_25343_end_mask_0 = const()[name = tensor("op_25343_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25343_cast_fp16 = slice_by_index(begin = var_25343_begin_0, end = var_25343_end_0, end_mask = var_25343_end_mask_0, x = q_115_cast_fp16)[name = tensor("op_25343_cast_fp16")]; + tensor var_25347_begin_0 = const()[name = tensor("op_25347_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_25347_end_0 = const()[name = tensor("op_25347_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_25347_end_mask_0 = const()[name = tensor("op_25347_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25347_cast_fp16 = slice_by_index(begin = var_25347_begin_0, end = var_25347_end_0, end_mask = var_25347_end_mask_0, x = q_115_cast_fp16)[name = tensor("op_25347_cast_fp16")]; + tensor var_25351_begin_0 = const()[name = tensor("op_25351_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_25351_end_0 = const()[name = tensor("op_25351_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_25351_end_mask_0 = const()[name = tensor("op_25351_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25351_cast_fp16 = slice_by_index(begin = var_25351_begin_0, end = var_25351_end_0, end_mask = var_25351_end_mask_0, x = q_115_cast_fp16)[name = tensor("op_25351_cast_fp16")]; + tensor var_25355_begin_0 = const()[name = tensor("op_25355_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_25355_end_0 = const()[name = tensor("op_25355_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_25355_end_mask_0 = const()[name = tensor("op_25355_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25355_cast_fp16 = slice_by_index(begin = var_25355_begin_0, end = var_25355_end_0, end_mask = var_25355_end_mask_0, x = q_115_cast_fp16)[name = tensor("op_25355_cast_fp16")]; + tensor var_25359_begin_0 = const()[name = tensor("op_25359_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_25359_end_0 = const()[name = tensor("op_25359_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_25359_end_mask_0 = const()[name = tensor("op_25359_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25359_cast_fp16 = slice_by_index(begin = var_25359_begin_0, end = var_25359_end_0, end_mask = var_25359_end_mask_0, x = q_115_cast_fp16)[name = tensor("op_25359_cast_fp16")]; + tensor var_25363_begin_0 = const()[name = tensor("op_25363_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_25363_end_0 = const()[name = tensor("op_25363_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_25363_end_mask_0 = const()[name = tensor("op_25363_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25363_cast_fp16 = slice_by_index(begin = var_25363_begin_0, end = var_25363_end_0, end_mask = var_25363_end_mask_0, x = q_115_cast_fp16)[name = tensor("op_25363_cast_fp16")]; + tensor k_231_perm_0 = const()[name = tensor("k_231_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_25370_begin_0 = const()[name = tensor("op_25370_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_25370_end_0 = const()[name = tensor("op_25370_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_25370_end_mask_0 = const()[name = tensor("op_25370_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_231_cast_fp16 = transpose(perm = k_231_perm_0, x = k_229_cast_fp16)[name = tensor("transpose_82")]; + tensor var_25370_cast_fp16 = slice_by_index(begin = var_25370_begin_0, end = var_25370_end_0, end_mask = var_25370_end_mask_0, x = k_231_cast_fp16)[name = tensor("op_25370_cast_fp16")]; + tensor var_25374_begin_0 = const()[name = tensor("op_25374_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_25374_end_0 = const()[name = tensor("op_25374_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_25374_end_mask_0 = const()[name = tensor("op_25374_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25374_cast_fp16 = slice_by_index(begin = var_25374_begin_0, end = var_25374_end_0, end_mask = var_25374_end_mask_0, x = k_231_cast_fp16)[name = tensor("op_25374_cast_fp16")]; + tensor var_25378_begin_0 = const()[name = tensor("op_25378_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_25378_end_0 = const()[name = tensor("op_25378_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_25378_end_mask_0 = const()[name = tensor("op_25378_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25378_cast_fp16 = slice_by_index(begin = var_25378_begin_0, end = var_25378_end_0, end_mask = var_25378_end_mask_0, x = k_231_cast_fp16)[name = tensor("op_25378_cast_fp16")]; + tensor var_25382_begin_0 = const()[name = tensor("op_25382_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_25382_end_0 = const()[name = tensor("op_25382_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_25382_end_mask_0 = const()[name = tensor("op_25382_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25382_cast_fp16 = slice_by_index(begin = var_25382_begin_0, end = var_25382_end_0, end_mask = var_25382_end_mask_0, x = k_231_cast_fp16)[name = tensor("op_25382_cast_fp16")]; + tensor var_25386_begin_0 = const()[name = tensor("op_25386_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_25386_end_0 = const()[name = tensor("op_25386_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_25386_end_mask_0 = const()[name = tensor("op_25386_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25386_cast_fp16 = slice_by_index(begin = var_25386_begin_0, end = var_25386_end_0, end_mask = var_25386_end_mask_0, x = k_231_cast_fp16)[name = tensor("op_25386_cast_fp16")]; + tensor var_25390_begin_0 = const()[name = tensor("op_25390_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_25390_end_0 = const()[name = tensor("op_25390_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_25390_end_mask_0 = const()[name = tensor("op_25390_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25390_cast_fp16 = slice_by_index(begin = var_25390_begin_0, end = var_25390_end_0, end_mask = var_25390_end_mask_0, x = k_231_cast_fp16)[name = tensor("op_25390_cast_fp16")]; + tensor var_25394_begin_0 = const()[name = tensor("op_25394_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_25394_end_0 = const()[name = tensor("op_25394_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_25394_end_mask_0 = const()[name = tensor("op_25394_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25394_cast_fp16 = slice_by_index(begin = var_25394_begin_0, end = var_25394_end_0, end_mask = var_25394_end_mask_0, x = k_231_cast_fp16)[name = tensor("op_25394_cast_fp16")]; + tensor var_25398_begin_0 = const()[name = tensor("op_25398_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_25398_end_0 = const()[name = tensor("op_25398_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_25398_end_mask_0 = const()[name = tensor("op_25398_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25398_cast_fp16 = slice_by_index(begin = var_25398_begin_0, end = var_25398_end_0, end_mask = var_25398_end_mask_0, x = k_231_cast_fp16)[name = tensor("op_25398_cast_fp16")]; + tensor var_25402_begin_0 = const()[name = tensor("op_25402_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_25402_end_0 = const()[name = tensor("op_25402_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_25402_end_mask_0 = const()[name = tensor("op_25402_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25402_cast_fp16 = slice_by_index(begin = var_25402_begin_0, end = var_25402_end_0, end_mask = var_25402_end_mask_0, x = k_231_cast_fp16)[name = tensor("op_25402_cast_fp16")]; + tensor var_25406_begin_0 = const()[name = tensor("op_25406_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_25406_end_0 = const()[name = tensor("op_25406_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_25406_end_mask_0 = const()[name = tensor("op_25406_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25406_cast_fp16 = slice_by_index(begin = var_25406_begin_0, end = var_25406_end_0, end_mask = var_25406_end_mask_0, x = k_231_cast_fp16)[name = tensor("op_25406_cast_fp16")]; + tensor var_25410_begin_0 = const()[name = tensor("op_25410_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_25410_end_0 = const()[name = tensor("op_25410_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_25410_end_mask_0 = const()[name = tensor("op_25410_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25410_cast_fp16 = slice_by_index(begin = var_25410_begin_0, end = var_25410_end_0, end_mask = var_25410_end_mask_0, x = k_231_cast_fp16)[name = tensor("op_25410_cast_fp16")]; + tensor var_25414_begin_0 = const()[name = tensor("op_25414_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_25414_end_0 = const()[name = tensor("op_25414_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_25414_end_mask_0 = const()[name = tensor("op_25414_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25414_cast_fp16 = slice_by_index(begin = var_25414_begin_0, end = var_25414_end_0, end_mask = var_25414_end_mask_0, x = k_231_cast_fp16)[name = tensor("op_25414_cast_fp16")]; + tensor var_25418_begin_0 = const()[name = tensor("op_25418_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_25418_end_0 = const()[name = tensor("op_25418_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_25418_end_mask_0 = const()[name = tensor("op_25418_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25418_cast_fp16 = slice_by_index(begin = var_25418_begin_0, end = var_25418_end_0, end_mask = var_25418_end_mask_0, x = k_231_cast_fp16)[name = tensor("op_25418_cast_fp16")]; + tensor var_25422_begin_0 = const()[name = tensor("op_25422_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_25422_end_0 = const()[name = tensor("op_25422_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_25422_end_mask_0 = const()[name = tensor("op_25422_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25422_cast_fp16 = slice_by_index(begin = var_25422_begin_0, end = var_25422_end_0, end_mask = var_25422_end_mask_0, x = k_231_cast_fp16)[name = tensor("op_25422_cast_fp16")]; + tensor var_25426_begin_0 = const()[name = tensor("op_25426_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_25426_end_0 = const()[name = tensor("op_25426_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_25426_end_mask_0 = const()[name = tensor("op_25426_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25426_cast_fp16 = slice_by_index(begin = var_25426_begin_0, end = var_25426_end_0, end_mask = var_25426_end_mask_0, x = k_231_cast_fp16)[name = tensor("op_25426_cast_fp16")]; + tensor var_25430_begin_0 = const()[name = tensor("op_25430_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_25430_end_0 = const()[name = tensor("op_25430_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_25430_end_mask_0 = const()[name = tensor("op_25430_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25430_cast_fp16 = slice_by_index(begin = var_25430_begin_0, end = var_25430_end_0, end_mask = var_25430_end_mask_0, x = k_231_cast_fp16)[name = tensor("op_25430_cast_fp16")]; + tensor var_25434_begin_0 = const()[name = tensor("op_25434_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_25434_end_0 = const()[name = tensor("op_25434_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_25434_end_mask_0 = const()[name = tensor("op_25434_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25434_cast_fp16 = slice_by_index(begin = var_25434_begin_0, end = var_25434_end_0, end_mask = var_25434_end_mask_0, x = k_231_cast_fp16)[name = tensor("op_25434_cast_fp16")]; + tensor var_25438_begin_0 = const()[name = tensor("op_25438_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_25438_end_0 = const()[name = tensor("op_25438_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_25438_end_mask_0 = const()[name = tensor("op_25438_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25438_cast_fp16 = slice_by_index(begin = var_25438_begin_0, end = var_25438_end_0, end_mask = var_25438_end_mask_0, x = k_231_cast_fp16)[name = tensor("op_25438_cast_fp16")]; + tensor var_25442_begin_0 = const()[name = tensor("op_25442_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_25442_end_0 = const()[name = tensor("op_25442_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_25442_end_mask_0 = const()[name = tensor("op_25442_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25442_cast_fp16 = slice_by_index(begin = var_25442_begin_0, end = var_25442_end_0, end_mask = var_25442_end_mask_0, x = k_231_cast_fp16)[name = tensor("op_25442_cast_fp16")]; + tensor var_25446_begin_0 = const()[name = tensor("op_25446_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_25446_end_0 = const()[name = tensor("op_25446_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_25446_end_mask_0 = const()[name = tensor("op_25446_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25446_cast_fp16 = slice_by_index(begin = var_25446_begin_0, end = var_25446_end_0, end_mask = var_25446_end_mask_0, x = k_231_cast_fp16)[name = tensor("op_25446_cast_fp16")]; + tensor var_25448_begin_0 = const()[name = tensor("op_25448_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_25448_end_0 = const()[name = tensor("op_25448_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_25448_end_mask_0 = const()[name = tensor("op_25448_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25448_cast_fp16 = slice_by_index(begin = var_25448_begin_0, end = var_25448_end_0, end_mask = var_25448_end_mask_0, x = v_115_cast_fp16)[name = tensor("op_25448_cast_fp16")]; + tensor var_25452_begin_0 = const()[name = tensor("op_25452_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_25452_end_0 = const()[name = tensor("op_25452_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_25452_end_mask_0 = const()[name = tensor("op_25452_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25452_cast_fp16 = slice_by_index(begin = var_25452_begin_0, end = var_25452_end_0, end_mask = var_25452_end_mask_0, x = v_115_cast_fp16)[name = tensor("op_25452_cast_fp16")]; + tensor var_25456_begin_0 = const()[name = tensor("op_25456_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_25456_end_0 = const()[name = tensor("op_25456_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_25456_end_mask_0 = const()[name = tensor("op_25456_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25456_cast_fp16 = slice_by_index(begin = var_25456_begin_0, end = var_25456_end_0, end_mask = var_25456_end_mask_0, x = v_115_cast_fp16)[name = tensor("op_25456_cast_fp16")]; + tensor var_25460_begin_0 = const()[name = tensor("op_25460_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_25460_end_0 = const()[name = tensor("op_25460_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_25460_end_mask_0 = const()[name = tensor("op_25460_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25460_cast_fp16 = slice_by_index(begin = var_25460_begin_0, end = var_25460_end_0, end_mask = var_25460_end_mask_0, x = v_115_cast_fp16)[name = tensor("op_25460_cast_fp16")]; + tensor var_25464_begin_0 = const()[name = tensor("op_25464_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_25464_end_0 = const()[name = tensor("op_25464_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_25464_end_mask_0 = const()[name = tensor("op_25464_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25464_cast_fp16 = slice_by_index(begin = var_25464_begin_0, end = var_25464_end_0, end_mask = var_25464_end_mask_0, x = v_115_cast_fp16)[name = tensor("op_25464_cast_fp16")]; + tensor var_25468_begin_0 = const()[name = tensor("op_25468_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_25468_end_0 = const()[name = tensor("op_25468_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_25468_end_mask_0 = const()[name = tensor("op_25468_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25468_cast_fp16 = slice_by_index(begin = var_25468_begin_0, end = var_25468_end_0, end_mask = var_25468_end_mask_0, x = v_115_cast_fp16)[name = tensor("op_25468_cast_fp16")]; + tensor var_25472_begin_0 = const()[name = tensor("op_25472_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_25472_end_0 = const()[name = tensor("op_25472_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_25472_end_mask_0 = const()[name = tensor("op_25472_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25472_cast_fp16 = slice_by_index(begin = var_25472_begin_0, end = var_25472_end_0, end_mask = var_25472_end_mask_0, x = v_115_cast_fp16)[name = tensor("op_25472_cast_fp16")]; + tensor var_25476_begin_0 = const()[name = tensor("op_25476_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_25476_end_0 = const()[name = tensor("op_25476_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_25476_end_mask_0 = const()[name = tensor("op_25476_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25476_cast_fp16 = slice_by_index(begin = var_25476_begin_0, end = var_25476_end_0, end_mask = var_25476_end_mask_0, x = v_115_cast_fp16)[name = tensor("op_25476_cast_fp16")]; + tensor var_25480_begin_0 = const()[name = tensor("op_25480_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_25480_end_0 = const()[name = tensor("op_25480_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_25480_end_mask_0 = const()[name = tensor("op_25480_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25480_cast_fp16 = slice_by_index(begin = var_25480_begin_0, end = var_25480_end_0, end_mask = var_25480_end_mask_0, x = v_115_cast_fp16)[name = tensor("op_25480_cast_fp16")]; + tensor var_25484_begin_0 = const()[name = tensor("op_25484_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_25484_end_0 = const()[name = tensor("op_25484_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_25484_end_mask_0 = const()[name = tensor("op_25484_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25484_cast_fp16 = slice_by_index(begin = var_25484_begin_0, end = var_25484_end_0, end_mask = var_25484_end_mask_0, x = v_115_cast_fp16)[name = tensor("op_25484_cast_fp16")]; + tensor var_25488_begin_0 = const()[name = tensor("op_25488_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_25488_end_0 = const()[name = tensor("op_25488_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_25488_end_mask_0 = const()[name = tensor("op_25488_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25488_cast_fp16 = slice_by_index(begin = var_25488_begin_0, end = var_25488_end_0, end_mask = var_25488_end_mask_0, x = v_115_cast_fp16)[name = tensor("op_25488_cast_fp16")]; + tensor var_25492_begin_0 = const()[name = tensor("op_25492_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_25492_end_0 = const()[name = tensor("op_25492_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_25492_end_mask_0 = const()[name = tensor("op_25492_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25492_cast_fp16 = slice_by_index(begin = var_25492_begin_0, end = var_25492_end_0, end_mask = var_25492_end_mask_0, x = v_115_cast_fp16)[name = tensor("op_25492_cast_fp16")]; + tensor var_25496_begin_0 = const()[name = tensor("op_25496_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_25496_end_0 = const()[name = tensor("op_25496_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_25496_end_mask_0 = const()[name = tensor("op_25496_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25496_cast_fp16 = slice_by_index(begin = var_25496_begin_0, end = var_25496_end_0, end_mask = var_25496_end_mask_0, x = v_115_cast_fp16)[name = tensor("op_25496_cast_fp16")]; + tensor var_25500_begin_0 = const()[name = tensor("op_25500_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_25500_end_0 = const()[name = tensor("op_25500_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_25500_end_mask_0 = const()[name = tensor("op_25500_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25500_cast_fp16 = slice_by_index(begin = var_25500_begin_0, end = var_25500_end_0, end_mask = var_25500_end_mask_0, x = v_115_cast_fp16)[name = tensor("op_25500_cast_fp16")]; + tensor var_25504_begin_0 = const()[name = tensor("op_25504_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_25504_end_0 = const()[name = tensor("op_25504_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_25504_end_mask_0 = const()[name = tensor("op_25504_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25504_cast_fp16 = slice_by_index(begin = var_25504_begin_0, end = var_25504_end_0, end_mask = var_25504_end_mask_0, x = v_115_cast_fp16)[name = tensor("op_25504_cast_fp16")]; + tensor var_25508_begin_0 = const()[name = tensor("op_25508_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_25508_end_0 = const()[name = tensor("op_25508_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_25508_end_mask_0 = const()[name = tensor("op_25508_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25508_cast_fp16 = slice_by_index(begin = var_25508_begin_0, end = var_25508_end_0, end_mask = var_25508_end_mask_0, x = v_115_cast_fp16)[name = tensor("op_25508_cast_fp16")]; + tensor var_25512_begin_0 = const()[name = tensor("op_25512_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_25512_end_0 = const()[name = tensor("op_25512_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_25512_end_mask_0 = const()[name = tensor("op_25512_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25512_cast_fp16 = slice_by_index(begin = var_25512_begin_0, end = var_25512_end_0, end_mask = var_25512_end_mask_0, x = v_115_cast_fp16)[name = tensor("op_25512_cast_fp16")]; + tensor var_25516_begin_0 = const()[name = tensor("op_25516_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_25516_end_0 = const()[name = tensor("op_25516_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_25516_end_mask_0 = const()[name = tensor("op_25516_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25516_cast_fp16 = slice_by_index(begin = var_25516_begin_0, end = var_25516_end_0, end_mask = var_25516_end_mask_0, x = v_115_cast_fp16)[name = tensor("op_25516_cast_fp16")]; + tensor var_25520_begin_0 = const()[name = tensor("op_25520_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_25520_end_0 = const()[name = tensor("op_25520_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_25520_end_mask_0 = const()[name = tensor("op_25520_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25520_cast_fp16 = slice_by_index(begin = var_25520_begin_0, end = var_25520_end_0, end_mask = var_25520_end_mask_0, x = v_115_cast_fp16)[name = tensor("op_25520_cast_fp16")]; + tensor var_25524_begin_0 = const()[name = tensor("op_25524_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_25524_end_0 = const()[name = tensor("op_25524_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_25524_end_mask_0 = const()[name = tensor("op_25524_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25524_cast_fp16 = slice_by_index(begin = var_25524_begin_0, end = var_25524_end_0, end_mask = var_25524_end_mask_0, x = v_115_cast_fp16)[name = tensor("op_25524_cast_fp16")]; + tensor var_25528_equation_0 = const()[name = tensor("op_25528_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25528_cast_fp16 = einsum(equation = var_25528_equation_0, values = (var_25370_cast_fp16, var_25287_cast_fp16))[name = tensor("op_25528_cast_fp16")]; + tensor var_25529_to_fp16 = const()[name = tensor("op_25529_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2121_cast_fp16 = mul(x = var_25528_cast_fp16, y = var_25529_to_fp16)[name = tensor("aw_2121_cast_fp16")]; + tensor var_25532_equation_0 = const()[name = tensor("op_25532_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25532_cast_fp16 = einsum(equation = var_25532_equation_0, values = (var_25374_cast_fp16, var_25291_cast_fp16))[name = tensor("op_25532_cast_fp16")]; + tensor var_25533_to_fp16 = const()[name = tensor("op_25533_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2123_cast_fp16 = mul(x = var_25532_cast_fp16, y = var_25533_to_fp16)[name = tensor("aw_2123_cast_fp16")]; + tensor var_25536_equation_0 = const()[name = tensor("op_25536_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25536_cast_fp16 = einsum(equation = var_25536_equation_0, values = (var_25378_cast_fp16, var_25295_cast_fp16))[name = tensor("op_25536_cast_fp16")]; + tensor var_25537_to_fp16 = const()[name = tensor("op_25537_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2125_cast_fp16 = mul(x = var_25536_cast_fp16, y = var_25537_to_fp16)[name = tensor("aw_2125_cast_fp16")]; + tensor var_25540_equation_0 = const()[name = tensor("op_25540_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25540_cast_fp16 = einsum(equation = var_25540_equation_0, values = (var_25382_cast_fp16, var_25299_cast_fp16))[name = tensor("op_25540_cast_fp16")]; + tensor var_25541_to_fp16 = const()[name = tensor("op_25541_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2127_cast_fp16 = mul(x = var_25540_cast_fp16, y = var_25541_to_fp16)[name = tensor("aw_2127_cast_fp16")]; + tensor var_25544_equation_0 = const()[name = tensor("op_25544_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25544_cast_fp16 = einsum(equation = var_25544_equation_0, values = (var_25386_cast_fp16, var_25303_cast_fp16))[name = tensor("op_25544_cast_fp16")]; + tensor var_25545_to_fp16 = const()[name = tensor("op_25545_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2129_cast_fp16 = mul(x = var_25544_cast_fp16, y = var_25545_to_fp16)[name = tensor("aw_2129_cast_fp16")]; + tensor var_25548_equation_0 = const()[name = tensor("op_25548_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25548_cast_fp16 = einsum(equation = var_25548_equation_0, values = (var_25390_cast_fp16, var_25307_cast_fp16))[name = tensor("op_25548_cast_fp16")]; + tensor var_25549_to_fp16 = const()[name = tensor("op_25549_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2131_cast_fp16 = mul(x = var_25548_cast_fp16, y = var_25549_to_fp16)[name = tensor("aw_2131_cast_fp16")]; + tensor var_25552_equation_0 = const()[name = tensor("op_25552_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25552_cast_fp16 = einsum(equation = var_25552_equation_0, values = (var_25394_cast_fp16, var_25311_cast_fp16))[name = tensor("op_25552_cast_fp16")]; + tensor var_25553_to_fp16 = const()[name = tensor("op_25553_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2133_cast_fp16 = mul(x = var_25552_cast_fp16, y = var_25553_to_fp16)[name = tensor("aw_2133_cast_fp16")]; + tensor var_25556_equation_0 = const()[name = tensor("op_25556_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25556_cast_fp16 = einsum(equation = var_25556_equation_0, values = (var_25398_cast_fp16, var_25315_cast_fp16))[name = tensor("op_25556_cast_fp16")]; + tensor var_25557_to_fp16 = const()[name = tensor("op_25557_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2135_cast_fp16 = mul(x = var_25556_cast_fp16, y = var_25557_to_fp16)[name = tensor("aw_2135_cast_fp16")]; + tensor var_25560_equation_0 = const()[name = tensor("op_25560_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25560_cast_fp16 = einsum(equation = var_25560_equation_0, values = (var_25402_cast_fp16, var_25319_cast_fp16))[name = tensor("op_25560_cast_fp16")]; + tensor var_25561_to_fp16 = const()[name = tensor("op_25561_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2137_cast_fp16 = mul(x = var_25560_cast_fp16, y = var_25561_to_fp16)[name = tensor("aw_2137_cast_fp16")]; + tensor var_25564_equation_0 = const()[name = tensor("op_25564_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25564_cast_fp16 = einsum(equation = var_25564_equation_0, values = (var_25406_cast_fp16, var_25323_cast_fp16))[name = tensor("op_25564_cast_fp16")]; + tensor var_25565_to_fp16 = const()[name = tensor("op_25565_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2139_cast_fp16 = mul(x = var_25564_cast_fp16, y = var_25565_to_fp16)[name = tensor("aw_2139_cast_fp16")]; + tensor var_25568_equation_0 = const()[name = tensor("op_25568_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25568_cast_fp16 = einsum(equation = var_25568_equation_0, values = (var_25410_cast_fp16, var_25327_cast_fp16))[name = tensor("op_25568_cast_fp16")]; + tensor var_25569_to_fp16 = const()[name = tensor("op_25569_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2141_cast_fp16 = mul(x = var_25568_cast_fp16, y = var_25569_to_fp16)[name = tensor("aw_2141_cast_fp16")]; + tensor var_25572_equation_0 = const()[name = tensor("op_25572_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25572_cast_fp16 = einsum(equation = var_25572_equation_0, values = (var_25414_cast_fp16, var_25331_cast_fp16))[name = tensor("op_25572_cast_fp16")]; + tensor var_25573_to_fp16 = const()[name = tensor("op_25573_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2143_cast_fp16 = mul(x = var_25572_cast_fp16, y = var_25573_to_fp16)[name = tensor("aw_2143_cast_fp16")]; + tensor var_25576_equation_0 = const()[name = tensor("op_25576_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25576_cast_fp16 = einsum(equation = var_25576_equation_0, values = (var_25418_cast_fp16, var_25335_cast_fp16))[name = tensor("op_25576_cast_fp16")]; + tensor var_25577_to_fp16 = const()[name = tensor("op_25577_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2145_cast_fp16 = mul(x = var_25576_cast_fp16, y = var_25577_to_fp16)[name = tensor("aw_2145_cast_fp16")]; + tensor var_25580_equation_0 = const()[name = tensor("op_25580_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25580_cast_fp16 = einsum(equation = var_25580_equation_0, values = (var_25422_cast_fp16, var_25339_cast_fp16))[name = tensor("op_25580_cast_fp16")]; + tensor var_25581_to_fp16 = const()[name = tensor("op_25581_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2147_cast_fp16 = mul(x = var_25580_cast_fp16, y = var_25581_to_fp16)[name = tensor("aw_2147_cast_fp16")]; + tensor var_25584_equation_0 = const()[name = tensor("op_25584_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25584_cast_fp16 = einsum(equation = var_25584_equation_0, values = (var_25426_cast_fp16, var_25343_cast_fp16))[name = tensor("op_25584_cast_fp16")]; + tensor var_25585_to_fp16 = const()[name = tensor("op_25585_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2149_cast_fp16 = mul(x = var_25584_cast_fp16, y = var_25585_to_fp16)[name = tensor("aw_2149_cast_fp16")]; + tensor var_25588_equation_0 = const()[name = tensor("op_25588_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25588_cast_fp16 = einsum(equation = var_25588_equation_0, values = (var_25430_cast_fp16, var_25347_cast_fp16))[name = tensor("op_25588_cast_fp16")]; + tensor var_25589_to_fp16 = const()[name = tensor("op_25589_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2151_cast_fp16 = mul(x = var_25588_cast_fp16, y = var_25589_to_fp16)[name = tensor("aw_2151_cast_fp16")]; + tensor var_25592_equation_0 = const()[name = tensor("op_25592_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25592_cast_fp16 = einsum(equation = var_25592_equation_0, values = (var_25434_cast_fp16, var_25351_cast_fp16))[name = tensor("op_25592_cast_fp16")]; + tensor var_25593_to_fp16 = const()[name = tensor("op_25593_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2153_cast_fp16 = mul(x = var_25592_cast_fp16, y = var_25593_to_fp16)[name = tensor("aw_2153_cast_fp16")]; + tensor var_25596_equation_0 = const()[name = tensor("op_25596_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25596_cast_fp16 = einsum(equation = var_25596_equation_0, values = (var_25438_cast_fp16, var_25355_cast_fp16))[name = tensor("op_25596_cast_fp16")]; + tensor var_25597_to_fp16 = const()[name = tensor("op_25597_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2155_cast_fp16 = mul(x = var_25596_cast_fp16, y = var_25597_to_fp16)[name = tensor("aw_2155_cast_fp16")]; + tensor var_25600_equation_0 = const()[name = tensor("op_25600_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25600_cast_fp16 = einsum(equation = var_25600_equation_0, values = (var_25442_cast_fp16, var_25359_cast_fp16))[name = tensor("op_25600_cast_fp16")]; + tensor var_25601_to_fp16 = const()[name = tensor("op_25601_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2157_cast_fp16 = mul(x = var_25600_cast_fp16, y = var_25601_to_fp16)[name = tensor("aw_2157_cast_fp16")]; + tensor var_25604_equation_0 = const()[name = tensor("op_25604_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25604_cast_fp16 = einsum(equation = var_25604_equation_0, values = (var_25446_cast_fp16, var_25363_cast_fp16))[name = tensor("op_25604_cast_fp16")]; + tensor var_25605_to_fp16 = const()[name = tensor("op_25605_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2159_cast_fp16 = mul(x = var_25604_cast_fp16, y = var_25605_to_fp16)[name = tensor("aw_2159_cast_fp16")]; + tensor var_25607_cast_fp16 = softmax(axis = var_21077, x = aw_2121_cast_fp16)[name = tensor("op_25607_cast_fp16")]; + tensor var_25608_cast_fp16 = softmax(axis = var_21077, x = aw_2123_cast_fp16)[name = tensor("op_25608_cast_fp16")]; + tensor var_25609_cast_fp16 = softmax(axis = var_21077, x = aw_2125_cast_fp16)[name = tensor("op_25609_cast_fp16")]; + tensor var_25610_cast_fp16 = softmax(axis = var_21077, x = aw_2127_cast_fp16)[name = tensor("op_25610_cast_fp16")]; + tensor var_25611_cast_fp16 = softmax(axis = var_21077, x = aw_2129_cast_fp16)[name = tensor("op_25611_cast_fp16")]; + tensor var_25612_cast_fp16 = softmax(axis = var_21077, x = aw_2131_cast_fp16)[name = tensor("op_25612_cast_fp16")]; + tensor var_25613_cast_fp16 = softmax(axis = var_21077, x = aw_2133_cast_fp16)[name = tensor("op_25613_cast_fp16")]; + tensor var_25614_cast_fp16 = softmax(axis = var_21077, x = aw_2135_cast_fp16)[name = tensor("op_25614_cast_fp16")]; + tensor var_25615_cast_fp16 = softmax(axis = var_21077, x = aw_2137_cast_fp16)[name = tensor("op_25615_cast_fp16")]; + tensor var_25616_cast_fp16 = softmax(axis = var_21077, x = aw_2139_cast_fp16)[name = tensor("op_25616_cast_fp16")]; + tensor var_25617_cast_fp16 = softmax(axis = var_21077, x = aw_2141_cast_fp16)[name = tensor("op_25617_cast_fp16")]; + tensor var_25618_cast_fp16 = softmax(axis = var_21077, x = aw_2143_cast_fp16)[name = tensor("op_25618_cast_fp16")]; + tensor var_25619_cast_fp16 = softmax(axis = var_21077, x = aw_2145_cast_fp16)[name = tensor("op_25619_cast_fp16")]; + tensor var_25620_cast_fp16 = softmax(axis = var_21077, x = aw_2147_cast_fp16)[name = tensor("op_25620_cast_fp16")]; + tensor var_25621_cast_fp16 = softmax(axis = var_21077, x = aw_2149_cast_fp16)[name = tensor("op_25621_cast_fp16")]; + tensor var_25622_cast_fp16 = softmax(axis = var_21077, x = aw_2151_cast_fp16)[name = tensor("op_25622_cast_fp16")]; + tensor var_25623_cast_fp16 = softmax(axis = var_21077, x = aw_2153_cast_fp16)[name = tensor("op_25623_cast_fp16")]; + tensor var_25624_cast_fp16 = softmax(axis = var_21077, x = aw_2155_cast_fp16)[name = tensor("op_25624_cast_fp16")]; + tensor var_25625_cast_fp16 = softmax(axis = var_21077, x = aw_2157_cast_fp16)[name = tensor("op_25625_cast_fp16")]; + tensor var_25626_cast_fp16 = softmax(axis = var_21077, x = aw_2159_cast_fp16)[name = tensor("op_25626_cast_fp16")]; + tensor var_25628_equation_0 = const()[name = tensor("op_25628_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25628_cast_fp16 = einsum(equation = var_25628_equation_0, values = (var_25448_cast_fp16, var_25607_cast_fp16))[name = tensor("op_25628_cast_fp16")]; + tensor var_25630_equation_0 = const()[name = tensor("op_25630_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25630_cast_fp16 = einsum(equation = var_25630_equation_0, values = (var_25452_cast_fp16, var_25608_cast_fp16))[name = tensor("op_25630_cast_fp16")]; + tensor var_25632_equation_0 = const()[name = tensor("op_25632_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25632_cast_fp16 = einsum(equation = var_25632_equation_0, values = (var_25456_cast_fp16, var_25609_cast_fp16))[name = tensor("op_25632_cast_fp16")]; + tensor var_25634_equation_0 = const()[name = tensor("op_25634_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25634_cast_fp16 = einsum(equation = var_25634_equation_0, values = (var_25460_cast_fp16, var_25610_cast_fp16))[name = tensor("op_25634_cast_fp16")]; + tensor var_25636_equation_0 = const()[name = tensor("op_25636_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25636_cast_fp16 = einsum(equation = var_25636_equation_0, values = (var_25464_cast_fp16, var_25611_cast_fp16))[name = tensor("op_25636_cast_fp16")]; + tensor var_25638_equation_0 = const()[name = tensor("op_25638_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25638_cast_fp16 = einsum(equation = var_25638_equation_0, values = (var_25468_cast_fp16, var_25612_cast_fp16))[name = tensor("op_25638_cast_fp16")]; + tensor var_25640_equation_0 = const()[name = tensor("op_25640_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25640_cast_fp16 = einsum(equation = var_25640_equation_0, values = (var_25472_cast_fp16, var_25613_cast_fp16))[name = tensor("op_25640_cast_fp16")]; + tensor var_25642_equation_0 = const()[name = tensor("op_25642_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25642_cast_fp16 = einsum(equation = var_25642_equation_0, values = (var_25476_cast_fp16, var_25614_cast_fp16))[name = tensor("op_25642_cast_fp16")]; + tensor var_25644_equation_0 = const()[name = tensor("op_25644_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25644_cast_fp16 = einsum(equation = var_25644_equation_0, values = (var_25480_cast_fp16, var_25615_cast_fp16))[name = tensor("op_25644_cast_fp16")]; + tensor var_25646_equation_0 = const()[name = tensor("op_25646_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25646_cast_fp16 = einsum(equation = var_25646_equation_0, values = (var_25484_cast_fp16, var_25616_cast_fp16))[name = tensor("op_25646_cast_fp16")]; + tensor var_25648_equation_0 = const()[name = tensor("op_25648_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25648_cast_fp16 = einsum(equation = var_25648_equation_0, values = (var_25488_cast_fp16, var_25617_cast_fp16))[name = tensor("op_25648_cast_fp16")]; + tensor var_25650_equation_0 = const()[name = tensor("op_25650_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25650_cast_fp16 = einsum(equation = var_25650_equation_0, values = (var_25492_cast_fp16, var_25618_cast_fp16))[name = tensor("op_25650_cast_fp16")]; + tensor var_25652_equation_0 = const()[name = tensor("op_25652_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25652_cast_fp16 = einsum(equation = var_25652_equation_0, values = (var_25496_cast_fp16, var_25619_cast_fp16))[name = tensor("op_25652_cast_fp16")]; + tensor var_25654_equation_0 = const()[name = tensor("op_25654_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25654_cast_fp16 = einsum(equation = var_25654_equation_0, values = (var_25500_cast_fp16, var_25620_cast_fp16))[name = tensor("op_25654_cast_fp16")]; + tensor var_25656_equation_0 = const()[name = tensor("op_25656_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25656_cast_fp16 = einsum(equation = var_25656_equation_0, values = (var_25504_cast_fp16, var_25621_cast_fp16))[name = tensor("op_25656_cast_fp16")]; + tensor var_25658_equation_0 = const()[name = tensor("op_25658_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25658_cast_fp16 = einsum(equation = var_25658_equation_0, values = (var_25508_cast_fp16, var_25622_cast_fp16))[name = tensor("op_25658_cast_fp16")]; + tensor var_25660_equation_0 = const()[name = tensor("op_25660_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25660_cast_fp16 = einsum(equation = var_25660_equation_0, values = (var_25512_cast_fp16, var_25623_cast_fp16))[name = tensor("op_25660_cast_fp16")]; + tensor var_25662_equation_0 = const()[name = tensor("op_25662_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25662_cast_fp16 = einsum(equation = var_25662_equation_0, values = (var_25516_cast_fp16, var_25624_cast_fp16))[name = tensor("op_25662_cast_fp16")]; + tensor var_25664_equation_0 = const()[name = tensor("op_25664_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25664_cast_fp16 = einsum(equation = var_25664_equation_0, values = (var_25520_cast_fp16, var_25625_cast_fp16))[name = tensor("op_25664_cast_fp16")]; + tensor var_25666_equation_0 = const()[name = tensor("op_25666_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_25666_cast_fp16 = einsum(equation = var_25666_equation_0, values = (var_25524_cast_fp16, var_25626_cast_fp16))[name = tensor("op_25666_cast_fp16")]; + tensor input_361_interleave_0 = const()[name = tensor("input_361_interleave_0"), val = tensor(false)]; + tensor input_361_cast_fp16 = concat(axis = var_21077, interleave = input_361_interleave_0, values = (var_25628_cast_fp16, var_25630_cast_fp16, var_25632_cast_fp16, var_25634_cast_fp16, var_25636_cast_fp16, var_25638_cast_fp16, var_25640_cast_fp16, var_25642_cast_fp16, var_25644_cast_fp16, var_25646_cast_fp16, var_25648_cast_fp16, var_25650_cast_fp16, var_25652_cast_fp16, var_25654_cast_fp16, var_25656_cast_fp16, var_25658_cast_fp16, var_25660_cast_fp16, var_25662_cast_fp16, var_25664_cast_fp16, var_25666_cast_fp16))[name = tensor("input_361_cast_fp16")]; + tensor var_25672 = const()[name = tensor("op_25672"), val = tensor([1, 1])]; + tensor var_25674 = const()[name = tensor("op_25674"), val = tensor([1, 1])]; + tensor var_25676_pad_type_0 = const()[name = tensor("op_25676_pad_type_0"), val = tensor("custom")]; + tensor var_25676_pad_0 = const()[name = tensor("op_25676_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_4_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_4_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2031044224)))]; + tensor mid_block_attentions_0_transformer_blocks_4_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_4_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2034321088)))]; + tensor var_25676_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_4_attn2_to_out_0_bias_to_fp16, dilations = var_25674, groups = var_21077, pad = var_25676_pad_0, pad_type = var_25676_pad_type_0, strides = var_25672, weight = mid_block_attentions_0_transformer_blocks_4_attn2_to_out_0_weight_to_fp16, x = input_361_cast_fp16)[name = tensor("op_25676_cast_fp16")]; + tensor inputs_173_cast_fp16 = add(x = var_25676_cast_fp16, y = inputs_171_cast_fp16)[name = tensor("inputs_173_cast_fp16")]; + tensor input_363_axes_0 = const()[name = tensor("input_363_axes_0"), val = tensor([1])]; + tensor input_363_gamma_0_to_fp16 = const()[name = tensor("input_363_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2034323712)))]; + tensor input_363_beta_0_to_fp16 = const()[name = tensor("input_363_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2034326336)))]; + tensor var_25686_to_fp16 = const()[name = tensor("op_25686_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_363_cast_fp16 = layer_norm(axes = input_363_axes_0, beta = input_363_beta_0_to_fp16, epsilon = var_25686_to_fp16, gamma = input_363_gamma_0_to_fp16, x = inputs_173_cast_fp16)[name = tensor("input_363_cast_fp16")]; + tensor var_25702 = const()[name = tensor("op_25702"), val = tensor([1, 1])]; + tensor var_25704 = const()[name = tensor("op_25704"), val = tensor([1, 1])]; + tensor var_25706_pad_type_0 = const()[name = tensor("op_25706_pad_type_0"), val = tensor("custom")]; + tensor var_25706_pad_0 = const()[name = tensor("op_25706_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_4_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_4_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2034328960)))]; + tensor mid_block_attentions_0_transformer_blocks_4_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_4_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2060543424)))]; + tensor var_25706_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_4_ff_net_0_proj_bias_to_fp16, dilations = var_25704, groups = var_21077, pad = var_25706_pad_0, pad_type = var_25706_pad_type_0, strides = var_25702, weight = mid_block_attentions_0_transformer_blocks_4_ff_net_0_proj_weight_to_fp16, x = input_363_cast_fp16)[name = tensor("op_25706_cast_fp16")]; + tensor var_25707_split_sizes_0 = const()[name = tensor("op_25707_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_25707_axis_0 = const()[name = tensor("op_25707_axis_0"), val = tensor(1)]; + tensor var_25707_cast_fp16_0, tensor var_25707_cast_fp16_1 = split(axis = var_25707_axis_0, split_sizes = var_25707_split_sizes_0, x = var_25706_cast_fp16)[name = tensor("op_25707_cast_fp16")]; + tensor var_25709_mode_0 = const()[name = tensor("op_25709_mode_0"), val = tensor("EXACT")]; + tensor var_25709_cast_fp16 = gelu(mode = var_25709_mode_0, x = var_25707_cast_fp16_1)[name = tensor("op_25709_cast_fp16")]; + tensor input_365_cast_fp16 = mul(x = var_25707_cast_fp16_0, y = var_25709_cast_fp16)[name = tensor("input_365_cast_fp16")]; + tensor var_25713 = const()[name = tensor("op_25713"), val = tensor([1, 1])]; + tensor var_25715 = const()[name = tensor("op_25715"), val = tensor([1, 1])]; + tensor var_25717_pad_type_0 = const()[name = tensor("op_25717_pad_type_0"), val = tensor("custom")]; + tensor var_25717_pad_0 = const()[name = tensor("op_25717_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_4_ff_net_2_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_4_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2060563968)))]; + tensor mid_block_attentions_0_transformer_blocks_4_ff_net_2_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_4_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2073671232)))]; + tensor var_25717_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_4_ff_net_2_bias_to_fp16, dilations = var_25715, groups = var_21077, pad = var_25717_pad_0, pad_type = var_25717_pad_type_0, strides = var_25713, weight = mid_block_attentions_0_transformer_blocks_4_ff_net_2_weight_to_fp16, x = input_365_cast_fp16)[name = tensor("op_25717_cast_fp16")]; + tensor inputs_175_cast_fp16 = add(x = var_25717_cast_fp16, y = inputs_173_cast_fp16)[name = tensor("inputs_175_cast_fp16")]; + tensor hidden_states_239_axes_0 = const()[name = tensor("hidden_states_239_axes_0"), val = tensor([1])]; + tensor hidden_states_239_gamma_0_to_fp16 = const()[name = tensor("hidden_states_239_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2073673856)))]; + tensor hidden_states_239_beta_0_to_fp16 = const()[name = tensor("hidden_states_239_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2073676480)))]; + tensor var_25733_to_fp16 = const()[name = tensor("op_25733_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_239_cast_fp16 = layer_norm(axes = hidden_states_239_axes_0, beta = hidden_states_239_beta_0_to_fp16, epsilon = var_25733_to_fp16, gamma = hidden_states_239_gamma_0_to_fp16, x = inputs_175_cast_fp16)[name = tensor("hidden_states_239_cast_fp16")]; + tensor var_25748 = const()[name = tensor("op_25748"), val = tensor([1, 1])]; + tensor var_25750 = const()[name = tensor("op_25750"), val = tensor([1, 1])]; + tensor q_117_pad_type_0 = const()[name = tensor("q_117_pad_type_0"), val = tensor("custom")]; + tensor q_117_pad_0 = const()[name = tensor("q_117_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_5_attn1_to_q_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_5_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2073679104)))]; + tensor q_117_cast_fp16 = conv(dilations = var_25750, groups = var_21077, pad = q_117_pad_0, pad_type = q_117_pad_type_0, strides = var_25748, weight = mid_block_attentions_0_transformer_blocks_5_attn1_to_q_weight_to_fp16, x = hidden_states_239_cast_fp16)[name = tensor("q_117_cast_fp16")]; + tensor var_25754 = const()[name = tensor("op_25754"), val = tensor([1, 1])]; + tensor var_25756 = const()[name = tensor("op_25756"), val = tensor([1, 1])]; + tensor k_233_pad_type_0 = const()[name = tensor("k_233_pad_type_0"), val = tensor("custom")]; + tensor k_233_pad_0 = const()[name = tensor("k_233_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_5_attn1_to_k_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_5_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2076955968)))]; + tensor k_233_cast_fp16 = conv(dilations = var_25756, groups = var_21077, pad = k_233_pad_0, pad_type = k_233_pad_type_0, strides = var_25754, weight = mid_block_attentions_0_transformer_blocks_5_attn1_to_k_weight_to_fp16, x = hidden_states_239_cast_fp16)[name = tensor("k_233_cast_fp16")]; + tensor var_25760 = const()[name = tensor("op_25760"), val = tensor([1, 1])]; + tensor var_25762 = const()[name = tensor("op_25762"), val = tensor([1, 1])]; + tensor v_117_pad_type_0 = const()[name = tensor("v_117_pad_type_0"), val = tensor("custom")]; + tensor v_117_pad_0 = const()[name = tensor("v_117_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_5_attn1_to_v_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_5_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2080232832)))]; + tensor v_117_cast_fp16 = conv(dilations = var_25762, groups = var_21077, pad = v_117_pad_0, pad_type = v_117_pad_type_0, strides = var_25760, weight = mid_block_attentions_0_transformer_blocks_5_attn1_to_v_weight_to_fp16, x = hidden_states_239_cast_fp16)[name = tensor("v_117_cast_fp16")]; + tensor var_25766_begin_0 = const()[name = tensor("op_25766_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_25766_end_0 = const()[name = tensor("op_25766_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_25766_end_mask_0 = const()[name = tensor("op_25766_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25766_cast_fp16 = slice_by_index(begin = var_25766_begin_0, end = var_25766_end_0, end_mask = var_25766_end_mask_0, x = q_117_cast_fp16)[name = tensor("op_25766_cast_fp16")]; + tensor var_25770_begin_0 = const()[name = tensor("op_25770_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_25770_end_0 = const()[name = tensor("op_25770_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_25770_end_mask_0 = const()[name = tensor("op_25770_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25770_cast_fp16 = slice_by_index(begin = var_25770_begin_0, end = var_25770_end_0, end_mask = var_25770_end_mask_0, x = q_117_cast_fp16)[name = tensor("op_25770_cast_fp16")]; + tensor var_25774_begin_0 = const()[name = tensor("op_25774_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_25774_end_0 = const()[name = tensor("op_25774_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_25774_end_mask_0 = const()[name = tensor("op_25774_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25774_cast_fp16 = slice_by_index(begin = var_25774_begin_0, end = var_25774_end_0, end_mask = var_25774_end_mask_0, x = q_117_cast_fp16)[name = tensor("op_25774_cast_fp16")]; + tensor var_25778_begin_0 = const()[name = tensor("op_25778_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_25778_end_0 = const()[name = tensor("op_25778_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_25778_end_mask_0 = const()[name = tensor("op_25778_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25778_cast_fp16 = slice_by_index(begin = var_25778_begin_0, end = var_25778_end_0, end_mask = var_25778_end_mask_0, x = q_117_cast_fp16)[name = tensor("op_25778_cast_fp16")]; + tensor var_25782_begin_0 = const()[name = tensor("op_25782_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_25782_end_0 = const()[name = tensor("op_25782_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_25782_end_mask_0 = const()[name = tensor("op_25782_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25782_cast_fp16 = slice_by_index(begin = var_25782_begin_0, end = var_25782_end_0, end_mask = var_25782_end_mask_0, x = q_117_cast_fp16)[name = tensor("op_25782_cast_fp16")]; + tensor var_25786_begin_0 = const()[name = tensor("op_25786_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_25786_end_0 = const()[name = tensor("op_25786_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_25786_end_mask_0 = const()[name = tensor("op_25786_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25786_cast_fp16 = slice_by_index(begin = var_25786_begin_0, end = var_25786_end_0, end_mask = var_25786_end_mask_0, x = q_117_cast_fp16)[name = tensor("op_25786_cast_fp16")]; + tensor var_25790_begin_0 = const()[name = tensor("op_25790_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_25790_end_0 = const()[name = tensor("op_25790_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_25790_end_mask_0 = const()[name = tensor("op_25790_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25790_cast_fp16 = slice_by_index(begin = var_25790_begin_0, end = var_25790_end_0, end_mask = var_25790_end_mask_0, x = q_117_cast_fp16)[name = tensor("op_25790_cast_fp16")]; + tensor var_25794_begin_0 = const()[name = tensor("op_25794_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_25794_end_0 = const()[name = tensor("op_25794_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_25794_end_mask_0 = const()[name = tensor("op_25794_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25794_cast_fp16 = slice_by_index(begin = var_25794_begin_0, end = var_25794_end_0, end_mask = var_25794_end_mask_0, x = q_117_cast_fp16)[name = tensor("op_25794_cast_fp16")]; + tensor var_25798_begin_0 = const()[name = tensor("op_25798_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_25798_end_0 = const()[name = tensor("op_25798_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_25798_end_mask_0 = const()[name = tensor("op_25798_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25798_cast_fp16 = slice_by_index(begin = var_25798_begin_0, end = var_25798_end_0, end_mask = var_25798_end_mask_0, x = q_117_cast_fp16)[name = tensor("op_25798_cast_fp16")]; + tensor var_25802_begin_0 = const()[name = tensor("op_25802_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_25802_end_0 = const()[name = tensor("op_25802_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_25802_end_mask_0 = const()[name = tensor("op_25802_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25802_cast_fp16 = slice_by_index(begin = var_25802_begin_0, end = var_25802_end_0, end_mask = var_25802_end_mask_0, x = q_117_cast_fp16)[name = tensor("op_25802_cast_fp16")]; + tensor var_25806_begin_0 = const()[name = tensor("op_25806_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_25806_end_0 = const()[name = tensor("op_25806_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_25806_end_mask_0 = const()[name = tensor("op_25806_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25806_cast_fp16 = slice_by_index(begin = var_25806_begin_0, end = var_25806_end_0, end_mask = var_25806_end_mask_0, x = q_117_cast_fp16)[name = tensor("op_25806_cast_fp16")]; + tensor var_25810_begin_0 = const()[name = tensor("op_25810_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_25810_end_0 = const()[name = tensor("op_25810_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_25810_end_mask_0 = const()[name = tensor("op_25810_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25810_cast_fp16 = slice_by_index(begin = var_25810_begin_0, end = var_25810_end_0, end_mask = var_25810_end_mask_0, x = q_117_cast_fp16)[name = tensor("op_25810_cast_fp16")]; + tensor var_25814_begin_0 = const()[name = tensor("op_25814_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_25814_end_0 = const()[name = tensor("op_25814_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_25814_end_mask_0 = const()[name = tensor("op_25814_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25814_cast_fp16 = slice_by_index(begin = var_25814_begin_0, end = var_25814_end_0, end_mask = var_25814_end_mask_0, x = q_117_cast_fp16)[name = tensor("op_25814_cast_fp16")]; + tensor var_25818_begin_0 = const()[name = tensor("op_25818_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_25818_end_0 = const()[name = tensor("op_25818_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_25818_end_mask_0 = const()[name = tensor("op_25818_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25818_cast_fp16 = slice_by_index(begin = var_25818_begin_0, end = var_25818_end_0, end_mask = var_25818_end_mask_0, x = q_117_cast_fp16)[name = tensor("op_25818_cast_fp16")]; + tensor var_25822_begin_0 = const()[name = tensor("op_25822_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_25822_end_0 = const()[name = tensor("op_25822_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_25822_end_mask_0 = const()[name = tensor("op_25822_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25822_cast_fp16 = slice_by_index(begin = var_25822_begin_0, end = var_25822_end_0, end_mask = var_25822_end_mask_0, x = q_117_cast_fp16)[name = tensor("op_25822_cast_fp16")]; + tensor var_25826_begin_0 = const()[name = tensor("op_25826_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_25826_end_0 = const()[name = tensor("op_25826_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_25826_end_mask_0 = const()[name = tensor("op_25826_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25826_cast_fp16 = slice_by_index(begin = var_25826_begin_0, end = var_25826_end_0, end_mask = var_25826_end_mask_0, x = q_117_cast_fp16)[name = tensor("op_25826_cast_fp16")]; + tensor var_25830_begin_0 = const()[name = tensor("op_25830_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_25830_end_0 = const()[name = tensor("op_25830_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_25830_end_mask_0 = const()[name = tensor("op_25830_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25830_cast_fp16 = slice_by_index(begin = var_25830_begin_0, end = var_25830_end_0, end_mask = var_25830_end_mask_0, x = q_117_cast_fp16)[name = tensor("op_25830_cast_fp16")]; + tensor var_25834_begin_0 = const()[name = tensor("op_25834_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_25834_end_0 = const()[name = tensor("op_25834_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_25834_end_mask_0 = const()[name = tensor("op_25834_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25834_cast_fp16 = slice_by_index(begin = var_25834_begin_0, end = var_25834_end_0, end_mask = var_25834_end_mask_0, x = q_117_cast_fp16)[name = tensor("op_25834_cast_fp16")]; + tensor var_25838_begin_0 = const()[name = tensor("op_25838_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_25838_end_0 = const()[name = tensor("op_25838_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_25838_end_mask_0 = const()[name = tensor("op_25838_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25838_cast_fp16 = slice_by_index(begin = var_25838_begin_0, end = var_25838_end_0, end_mask = var_25838_end_mask_0, x = q_117_cast_fp16)[name = tensor("op_25838_cast_fp16")]; + tensor var_25842_begin_0 = const()[name = tensor("op_25842_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_25842_end_0 = const()[name = tensor("op_25842_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_25842_end_mask_0 = const()[name = tensor("op_25842_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25842_cast_fp16 = slice_by_index(begin = var_25842_begin_0, end = var_25842_end_0, end_mask = var_25842_end_mask_0, x = q_117_cast_fp16)[name = tensor("op_25842_cast_fp16")]; + tensor k_235_perm_0 = const()[name = tensor("k_235_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_25849_begin_0 = const()[name = tensor("op_25849_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_25849_end_0 = const()[name = tensor("op_25849_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_25849_end_mask_0 = const()[name = tensor("op_25849_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_235_cast_fp16 = transpose(perm = k_235_perm_0, x = k_233_cast_fp16)[name = tensor("transpose_81")]; + tensor var_25849_cast_fp16 = slice_by_index(begin = var_25849_begin_0, end = var_25849_end_0, end_mask = var_25849_end_mask_0, x = k_235_cast_fp16)[name = tensor("op_25849_cast_fp16")]; + tensor var_25853_begin_0 = const()[name = tensor("op_25853_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_25853_end_0 = const()[name = tensor("op_25853_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_25853_end_mask_0 = const()[name = tensor("op_25853_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25853_cast_fp16 = slice_by_index(begin = var_25853_begin_0, end = var_25853_end_0, end_mask = var_25853_end_mask_0, x = k_235_cast_fp16)[name = tensor("op_25853_cast_fp16")]; + tensor var_25857_begin_0 = const()[name = tensor("op_25857_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_25857_end_0 = const()[name = tensor("op_25857_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_25857_end_mask_0 = const()[name = tensor("op_25857_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25857_cast_fp16 = slice_by_index(begin = var_25857_begin_0, end = var_25857_end_0, end_mask = var_25857_end_mask_0, x = k_235_cast_fp16)[name = tensor("op_25857_cast_fp16")]; + tensor var_25861_begin_0 = const()[name = tensor("op_25861_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_25861_end_0 = const()[name = tensor("op_25861_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_25861_end_mask_0 = const()[name = tensor("op_25861_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25861_cast_fp16 = slice_by_index(begin = var_25861_begin_0, end = var_25861_end_0, end_mask = var_25861_end_mask_0, x = k_235_cast_fp16)[name = tensor("op_25861_cast_fp16")]; + tensor var_25865_begin_0 = const()[name = tensor("op_25865_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_25865_end_0 = const()[name = tensor("op_25865_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_25865_end_mask_0 = const()[name = tensor("op_25865_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25865_cast_fp16 = slice_by_index(begin = var_25865_begin_0, end = var_25865_end_0, end_mask = var_25865_end_mask_0, x = k_235_cast_fp16)[name = tensor("op_25865_cast_fp16")]; + tensor var_25869_begin_0 = const()[name = tensor("op_25869_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_25869_end_0 = const()[name = tensor("op_25869_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_25869_end_mask_0 = const()[name = tensor("op_25869_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25869_cast_fp16 = slice_by_index(begin = var_25869_begin_0, end = var_25869_end_0, end_mask = var_25869_end_mask_0, x = k_235_cast_fp16)[name = tensor("op_25869_cast_fp16")]; + tensor var_25873_begin_0 = const()[name = tensor("op_25873_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_25873_end_0 = const()[name = tensor("op_25873_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_25873_end_mask_0 = const()[name = tensor("op_25873_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25873_cast_fp16 = slice_by_index(begin = var_25873_begin_0, end = var_25873_end_0, end_mask = var_25873_end_mask_0, x = k_235_cast_fp16)[name = tensor("op_25873_cast_fp16")]; + tensor var_25877_begin_0 = const()[name = tensor("op_25877_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_25877_end_0 = const()[name = tensor("op_25877_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_25877_end_mask_0 = const()[name = tensor("op_25877_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25877_cast_fp16 = slice_by_index(begin = var_25877_begin_0, end = var_25877_end_0, end_mask = var_25877_end_mask_0, x = k_235_cast_fp16)[name = tensor("op_25877_cast_fp16")]; + tensor var_25881_begin_0 = const()[name = tensor("op_25881_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_25881_end_0 = const()[name = tensor("op_25881_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_25881_end_mask_0 = const()[name = tensor("op_25881_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25881_cast_fp16 = slice_by_index(begin = var_25881_begin_0, end = var_25881_end_0, end_mask = var_25881_end_mask_0, x = k_235_cast_fp16)[name = tensor("op_25881_cast_fp16")]; + tensor var_25885_begin_0 = const()[name = tensor("op_25885_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_25885_end_0 = const()[name = tensor("op_25885_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_25885_end_mask_0 = const()[name = tensor("op_25885_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25885_cast_fp16 = slice_by_index(begin = var_25885_begin_0, end = var_25885_end_0, end_mask = var_25885_end_mask_0, x = k_235_cast_fp16)[name = tensor("op_25885_cast_fp16")]; + tensor var_25889_begin_0 = const()[name = tensor("op_25889_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_25889_end_0 = const()[name = tensor("op_25889_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_25889_end_mask_0 = const()[name = tensor("op_25889_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25889_cast_fp16 = slice_by_index(begin = var_25889_begin_0, end = var_25889_end_0, end_mask = var_25889_end_mask_0, x = k_235_cast_fp16)[name = tensor("op_25889_cast_fp16")]; + tensor var_25893_begin_0 = const()[name = tensor("op_25893_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_25893_end_0 = const()[name = tensor("op_25893_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_25893_end_mask_0 = const()[name = tensor("op_25893_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25893_cast_fp16 = slice_by_index(begin = var_25893_begin_0, end = var_25893_end_0, end_mask = var_25893_end_mask_0, x = k_235_cast_fp16)[name = tensor("op_25893_cast_fp16")]; + tensor var_25897_begin_0 = const()[name = tensor("op_25897_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_25897_end_0 = const()[name = tensor("op_25897_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_25897_end_mask_0 = const()[name = tensor("op_25897_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25897_cast_fp16 = slice_by_index(begin = var_25897_begin_0, end = var_25897_end_0, end_mask = var_25897_end_mask_0, x = k_235_cast_fp16)[name = tensor("op_25897_cast_fp16")]; + tensor var_25901_begin_0 = const()[name = tensor("op_25901_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_25901_end_0 = const()[name = tensor("op_25901_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_25901_end_mask_0 = const()[name = tensor("op_25901_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25901_cast_fp16 = slice_by_index(begin = var_25901_begin_0, end = var_25901_end_0, end_mask = var_25901_end_mask_0, x = k_235_cast_fp16)[name = tensor("op_25901_cast_fp16")]; + tensor var_25905_begin_0 = const()[name = tensor("op_25905_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_25905_end_0 = const()[name = tensor("op_25905_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_25905_end_mask_0 = const()[name = tensor("op_25905_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25905_cast_fp16 = slice_by_index(begin = var_25905_begin_0, end = var_25905_end_0, end_mask = var_25905_end_mask_0, x = k_235_cast_fp16)[name = tensor("op_25905_cast_fp16")]; + tensor var_25909_begin_0 = const()[name = tensor("op_25909_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_25909_end_0 = const()[name = tensor("op_25909_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_25909_end_mask_0 = const()[name = tensor("op_25909_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25909_cast_fp16 = slice_by_index(begin = var_25909_begin_0, end = var_25909_end_0, end_mask = var_25909_end_mask_0, x = k_235_cast_fp16)[name = tensor("op_25909_cast_fp16")]; + tensor var_25913_begin_0 = const()[name = tensor("op_25913_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_25913_end_0 = const()[name = tensor("op_25913_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_25913_end_mask_0 = const()[name = tensor("op_25913_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25913_cast_fp16 = slice_by_index(begin = var_25913_begin_0, end = var_25913_end_0, end_mask = var_25913_end_mask_0, x = k_235_cast_fp16)[name = tensor("op_25913_cast_fp16")]; + tensor var_25917_begin_0 = const()[name = tensor("op_25917_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_25917_end_0 = const()[name = tensor("op_25917_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_25917_end_mask_0 = const()[name = tensor("op_25917_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25917_cast_fp16 = slice_by_index(begin = var_25917_begin_0, end = var_25917_end_0, end_mask = var_25917_end_mask_0, x = k_235_cast_fp16)[name = tensor("op_25917_cast_fp16")]; + tensor var_25921_begin_0 = const()[name = tensor("op_25921_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_25921_end_0 = const()[name = tensor("op_25921_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_25921_end_mask_0 = const()[name = tensor("op_25921_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25921_cast_fp16 = slice_by_index(begin = var_25921_begin_0, end = var_25921_end_0, end_mask = var_25921_end_mask_0, x = k_235_cast_fp16)[name = tensor("op_25921_cast_fp16")]; + tensor var_25925_begin_0 = const()[name = tensor("op_25925_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_25925_end_0 = const()[name = tensor("op_25925_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_25925_end_mask_0 = const()[name = tensor("op_25925_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25925_cast_fp16 = slice_by_index(begin = var_25925_begin_0, end = var_25925_end_0, end_mask = var_25925_end_mask_0, x = k_235_cast_fp16)[name = tensor("op_25925_cast_fp16")]; + tensor var_25927_begin_0 = const()[name = tensor("op_25927_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_25927_end_0 = const()[name = tensor("op_25927_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_25927_end_mask_0 = const()[name = tensor("op_25927_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25927_cast_fp16 = slice_by_index(begin = var_25927_begin_0, end = var_25927_end_0, end_mask = var_25927_end_mask_0, x = v_117_cast_fp16)[name = tensor("op_25927_cast_fp16")]; + tensor var_25931_begin_0 = const()[name = tensor("op_25931_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_25931_end_0 = const()[name = tensor("op_25931_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_25931_end_mask_0 = const()[name = tensor("op_25931_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25931_cast_fp16 = slice_by_index(begin = var_25931_begin_0, end = var_25931_end_0, end_mask = var_25931_end_mask_0, x = v_117_cast_fp16)[name = tensor("op_25931_cast_fp16")]; + tensor var_25935_begin_0 = const()[name = tensor("op_25935_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_25935_end_0 = const()[name = tensor("op_25935_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_25935_end_mask_0 = const()[name = tensor("op_25935_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25935_cast_fp16 = slice_by_index(begin = var_25935_begin_0, end = var_25935_end_0, end_mask = var_25935_end_mask_0, x = v_117_cast_fp16)[name = tensor("op_25935_cast_fp16")]; + tensor var_25939_begin_0 = const()[name = tensor("op_25939_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_25939_end_0 = const()[name = tensor("op_25939_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_25939_end_mask_0 = const()[name = tensor("op_25939_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25939_cast_fp16 = slice_by_index(begin = var_25939_begin_0, end = var_25939_end_0, end_mask = var_25939_end_mask_0, x = v_117_cast_fp16)[name = tensor("op_25939_cast_fp16")]; + tensor var_25943_begin_0 = const()[name = tensor("op_25943_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_25943_end_0 = const()[name = tensor("op_25943_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_25943_end_mask_0 = const()[name = tensor("op_25943_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25943_cast_fp16 = slice_by_index(begin = var_25943_begin_0, end = var_25943_end_0, end_mask = var_25943_end_mask_0, x = v_117_cast_fp16)[name = tensor("op_25943_cast_fp16")]; + tensor var_25947_begin_0 = const()[name = tensor("op_25947_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_25947_end_0 = const()[name = tensor("op_25947_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_25947_end_mask_0 = const()[name = tensor("op_25947_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25947_cast_fp16 = slice_by_index(begin = var_25947_begin_0, end = var_25947_end_0, end_mask = var_25947_end_mask_0, x = v_117_cast_fp16)[name = tensor("op_25947_cast_fp16")]; + tensor var_25951_begin_0 = const()[name = tensor("op_25951_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_25951_end_0 = const()[name = tensor("op_25951_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_25951_end_mask_0 = const()[name = tensor("op_25951_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25951_cast_fp16 = slice_by_index(begin = var_25951_begin_0, end = var_25951_end_0, end_mask = var_25951_end_mask_0, x = v_117_cast_fp16)[name = tensor("op_25951_cast_fp16")]; + tensor var_25955_begin_0 = const()[name = tensor("op_25955_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_25955_end_0 = const()[name = tensor("op_25955_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_25955_end_mask_0 = const()[name = tensor("op_25955_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25955_cast_fp16 = slice_by_index(begin = var_25955_begin_0, end = var_25955_end_0, end_mask = var_25955_end_mask_0, x = v_117_cast_fp16)[name = tensor("op_25955_cast_fp16")]; + tensor var_25959_begin_0 = const()[name = tensor("op_25959_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_25959_end_0 = const()[name = tensor("op_25959_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_25959_end_mask_0 = const()[name = tensor("op_25959_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25959_cast_fp16 = slice_by_index(begin = var_25959_begin_0, end = var_25959_end_0, end_mask = var_25959_end_mask_0, x = v_117_cast_fp16)[name = tensor("op_25959_cast_fp16")]; + tensor var_25963_begin_0 = const()[name = tensor("op_25963_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_25963_end_0 = const()[name = tensor("op_25963_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_25963_end_mask_0 = const()[name = tensor("op_25963_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25963_cast_fp16 = slice_by_index(begin = var_25963_begin_0, end = var_25963_end_0, end_mask = var_25963_end_mask_0, x = v_117_cast_fp16)[name = tensor("op_25963_cast_fp16")]; + tensor var_25967_begin_0 = const()[name = tensor("op_25967_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_25967_end_0 = const()[name = tensor("op_25967_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_25967_end_mask_0 = const()[name = tensor("op_25967_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25967_cast_fp16 = slice_by_index(begin = var_25967_begin_0, end = var_25967_end_0, end_mask = var_25967_end_mask_0, x = v_117_cast_fp16)[name = tensor("op_25967_cast_fp16")]; + tensor var_25971_begin_0 = const()[name = tensor("op_25971_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_25971_end_0 = const()[name = tensor("op_25971_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_25971_end_mask_0 = const()[name = tensor("op_25971_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25971_cast_fp16 = slice_by_index(begin = var_25971_begin_0, end = var_25971_end_0, end_mask = var_25971_end_mask_0, x = v_117_cast_fp16)[name = tensor("op_25971_cast_fp16")]; + tensor var_25975_begin_0 = const()[name = tensor("op_25975_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_25975_end_0 = const()[name = tensor("op_25975_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_25975_end_mask_0 = const()[name = tensor("op_25975_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25975_cast_fp16 = slice_by_index(begin = var_25975_begin_0, end = var_25975_end_0, end_mask = var_25975_end_mask_0, x = v_117_cast_fp16)[name = tensor("op_25975_cast_fp16")]; + tensor var_25979_begin_0 = const()[name = tensor("op_25979_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_25979_end_0 = const()[name = tensor("op_25979_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_25979_end_mask_0 = const()[name = tensor("op_25979_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25979_cast_fp16 = slice_by_index(begin = var_25979_begin_0, end = var_25979_end_0, end_mask = var_25979_end_mask_0, x = v_117_cast_fp16)[name = tensor("op_25979_cast_fp16")]; + tensor var_25983_begin_0 = const()[name = tensor("op_25983_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_25983_end_0 = const()[name = tensor("op_25983_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_25983_end_mask_0 = const()[name = tensor("op_25983_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25983_cast_fp16 = slice_by_index(begin = var_25983_begin_0, end = var_25983_end_0, end_mask = var_25983_end_mask_0, x = v_117_cast_fp16)[name = tensor("op_25983_cast_fp16")]; + tensor var_25987_begin_0 = const()[name = tensor("op_25987_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_25987_end_0 = const()[name = tensor("op_25987_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_25987_end_mask_0 = const()[name = tensor("op_25987_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25987_cast_fp16 = slice_by_index(begin = var_25987_begin_0, end = var_25987_end_0, end_mask = var_25987_end_mask_0, x = v_117_cast_fp16)[name = tensor("op_25987_cast_fp16")]; + tensor var_25991_begin_0 = const()[name = tensor("op_25991_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_25991_end_0 = const()[name = tensor("op_25991_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_25991_end_mask_0 = const()[name = tensor("op_25991_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25991_cast_fp16 = slice_by_index(begin = var_25991_begin_0, end = var_25991_end_0, end_mask = var_25991_end_mask_0, x = v_117_cast_fp16)[name = tensor("op_25991_cast_fp16")]; + tensor var_25995_begin_0 = const()[name = tensor("op_25995_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_25995_end_0 = const()[name = tensor("op_25995_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_25995_end_mask_0 = const()[name = tensor("op_25995_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25995_cast_fp16 = slice_by_index(begin = var_25995_begin_0, end = var_25995_end_0, end_mask = var_25995_end_mask_0, x = v_117_cast_fp16)[name = tensor("op_25995_cast_fp16")]; + tensor var_25999_begin_0 = const()[name = tensor("op_25999_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_25999_end_0 = const()[name = tensor("op_25999_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_25999_end_mask_0 = const()[name = tensor("op_25999_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25999_cast_fp16 = slice_by_index(begin = var_25999_begin_0, end = var_25999_end_0, end_mask = var_25999_end_mask_0, x = v_117_cast_fp16)[name = tensor("op_25999_cast_fp16")]; + tensor var_26003_begin_0 = const()[name = tensor("op_26003_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_26003_end_0 = const()[name = tensor("op_26003_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_26003_end_mask_0 = const()[name = tensor("op_26003_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26003_cast_fp16 = slice_by_index(begin = var_26003_begin_0, end = var_26003_end_0, end_mask = var_26003_end_mask_0, x = v_117_cast_fp16)[name = tensor("op_26003_cast_fp16")]; + tensor var_26007_equation_0 = const()[name = tensor("op_26007_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26007_cast_fp16 = einsum(equation = var_26007_equation_0, values = (var_25849_cast_fp16, var_25766_cast_fp16))[name = tensor("op_26007_cast_fp16")]; + tensor var_26008_to_fp16 = const()[name = tensor("op_26008_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2161_cast_fp16 = mul(x = var_26007_cast_fp16, y = var_26008_to_fp16)[name = tensor("aw_2161_cast_fp16")]; + tensor var_26011_equation_0 = const()[name = tensor("op_26011_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26011_cast_fp16 = einsum(equation = var_26011_equation_0, values = (var_25853_cast_fp16, var_25770_cast_fp16))[name = tensor("op_26011_cast_fp16")]; + tensor var_26012_to_fp16 = const()[name = tensor("op_26012_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2163_cast_fp16 = mul(x = var_26011_cast_fp16, y = var_26012_to_fp16)[name = tensor("aw_2163_cast_fp16")]; + tensor var_26015_equation_0 = const()[name = tensor("op_26015_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26015_cast_fp16 = einsum(equation = var_26015_equation_0, values = (var_25857_cast_fp16, var_25774_cast_fp16))[name = tensor("op_26015_cast_fp16")]; + tensor var_26016_to_fp16 = const()[name = tensor("op_26016_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2165_cast_fp16 = mul(x = var_26015_cast_fp16, y = var_26016_to_fp16)[name = tensor("aw_2165_cast_fp16")]; + tensor var_26019_equation_0 = const()[name = tensor("op_26019_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26019_cast_fp16 = einsum(equation = var_26019_equation_0, values = (var_25861_cast_fp16, var_25778_cast_fp16))[name = tensor("op_26019_cast_fp16")]; + tensor var_26020_to_fp16 = const()[name = tensor("op_26020_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2167_cast_fp16 = mul(x = var_26019_cast_fp16, y = var_26020_to_fp16)[name = tensor("aw_2167_cast_fp16")]; + tensor var_26023_equation_0 = const()[name = tensor("op_26023_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26023_cast_fp16 = einsum(equation = var_26023_equation_0, values = (var_25865_cast_fp16, var_25782_cast_fp16))[name = tensor("op_26023_cast_fp16")]; + tensor var_26024_to_fp16 = const()[name = tensor("op_26024_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2169_cast_fp16 = mul(x = var_26023_cast_fp16, y = var_26024_to_fp16)[name = tensor("aw_2169_cast_fp16")]; + tensor var_26027_equation_0 = const()[name = tensor("op_26027_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26027_cast_fp16 = einsum(equation = var_26027_equation_0, values = (var_25869_cast_fp16, var_25786_cast_fp16))[name = tensor("op_26027_cast_fp16")]; + tensor var_26028_to_fp16 = const()[name = tensor("op_26028_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2171_cast_fp16 = mul(x = var_26027_cast_fp16, y = var_26028_to_fp16)[name = tensor("aw_2171_cast_fp16")]; + tensor var_26031_equation_0 = const()[name = tensor("op_26031_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26031_cast_fp16 = einsum(equation = var_26031_equation_0, values = (var_25873_cast_fp16, var_25790_cast_fp16))[name = tensor("op_26031_cast_fp16")]; + tensor var_26032_to_fp16 = const()[name = tensor("op_26032_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2173_cast_fp16 = mul(x = var_26031_cast_fp16, y = var_26032_to_fp16)[name = tensor("aw_2173_cast_fp16")]; + tensor var_26035_equation_0 = const()[name = tensor("op_26035_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26035_cast_fp16 = einsum(equation = var_26035_equation_0, values = (var_25877_cast_fp16, var_25794_cast_fp16))[name = tensor("op_26035_cast_fp16")]; + tensor var_26036_to_fp16 = const()[name = tensor("op_26036_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2175_cast_fp16 = mul(x = var_26035_cast_fp16, y = var_26036_to_fp16)[name = tensor("aw_2175_cast_fp16")]; + tensor var_26039_equation_0 = const()[name = tensor("op_26039_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26039_cast_fp16 = einsum(equation = var_26039_equation_0, values = (var_25881_cast_fp16, var_25798_cast_fp16))[name = tensor("op_26039_cast_fp16")]; + tensor var_26040_to_fp16 = const()[name = tensor("op_26040_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2177_cast_fp16 = mul(x = var_26039_cast_fp16, y = var_26040_to_fp16)[name = tensor("aw_2177_cast_fp16")]; + tensor var_26043_equation_0 = const()[name = tensor("op_26043_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26043_cast_fp16 = einsum(equation = var_26043_equation_0, values = (var_25885_cast_fp16, var_25802_cast_fp16))[name = tensor("op_26043_cast_fp16")]; + tensor var_26044_to_fp16 = const()[name = tensor("op_26044_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2179_cast_fp16 = mul(x = var_26043_cast_fp16, y = var_26044_to_fp16)[name = tensor("aw_2179_cast_fp16")]; + tensor var_26047_equation_0 = const()[name = tensor("op_26047_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26047_cast_fp16 = einsum(equation = var_26047_equation_0, values = (var_25889_cast_fp16, var_25806_cast_fp16))[name = tensor("op_26047_cast_fp16")]; + tensor var_26048_to_fp16 = const()[name = tensor("op_26048_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2181_cast_fp16 = mul(x = var_26047_cast_fp16, y = var_26048_to_fp16)[name = tensor("aw_2181_cast_fp16")]; + tensor var_26051_equation_0 = const()[name = tensor("op_26051_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26051_cast_fp16 = einsum(equation = var_26051_equation_0, values = (var_25893_cast_fp16, var_25810_cast_fp16))[name = tensor("op_26051_cast_fp16")]; + tensor var_26052_to_fp16 = const()[name = tensor("op_26052_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2183_cast_fp16 = mul(x = var_26051_cast_fp16, y = var_26052_to_fp16)[name = tensor("aw_2183_cast_fp16")]; + tensor var_26055_equation_0 = const()[name = tensor("op_26055_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26055_cast_fp16 = einsum(equation = var_26055_equation_0, values = (var_25897_cast_fp16, var_25814_cast_fp16))[name = tensor("op_26055_cast_fp16")]; + tensor var_26056_to_fp16 = const()[name = tensor("op_26056_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2185_cast_fp16 = mul(x = var_26055_cast_fp16, y = var_26056_to_fp16)[name = tensor("aw_2185_cast_fp16")]; + tensor var_26059_equation_0 = const()[name = tensor("op_26059_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26059_cast_fp16 = einsum(equation = var_26059_equation_0, values = (var_25901_cast_fp16, var_25818_cast_fp16))[name = tensor("op_26059_cast_fp16")]; + tensor var_26060_to_fp16 = const()[name = tensor("op_26060_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2187_cast_fp16 = mul(x = var_26059_cast_fp16, y = var_26060_to_fp16)[name = tensor("aw_2187_cast_fp16")]; + tensor var_26063_equation_0 = const()[name = tensor("op_26063_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26063_cast_fp16 = einsum(equation = var_26063_equation_0, values = (var_25905_cast_fp16, var_25822_cast_fp16))[name = tensor("op_26063_cast_fp16")]; + tensor var_26064_to_fp16 = const()[name = tensor("op_26064_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2189_cast_fp16 = mul(x = var_26063_cast_fp16, y = var_26064_to_fp16)[name = tensor("aw_2189_cast_fp16")]; + tensor var_26067_equation_0 = const()[name = tensor("op_26067_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26067_cast_fp16 = einsum(equation = var_26067_equation_0, values = (var_25909_cast_fp16, var_25826_cast_fp16))[name = tensor("op_26067_cast_fp16")]; + tensor var_26068_to_fp16 = const()[name = tensor("op_26068_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2191_cast_fp16 = mul(x = var_26067_cast_fp16, y = var_26068_to_fp16)[name = tensor("aw_2191_cast_fp16")]; + tensor var_26071_equation_0 = const()[name = tensor("op_26071_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26071_cast_fp16 = einsum(equation = var_26071_equation_0, values = (var_25913_cast_fp16, var_25830_cast_fp16))[name = tensor("op_26071_cast_fp16")]; + tensor var_26072_to_fp16 = const()[name = tensor("op_26072_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2193_cast_fp16 = mul(x = var_26071_cast_fp16, y = var_26072_to_fp16)[name = tensor("aw_2193_cast_fp16")]; + tensor var_26075_equation_0 = const()[name = tensor("op_26075_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26075_cast_fp16 = einsum(equation = var_26075_equation_0, values = (var_25917_cast_fp16, var_25834_cast_fp16))[name = tensor("op_26075_cast_fp16")]; + tensor var_26076_to_fp16 = const()[name = tensor("op_26076_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2195_cast_fp16 = mul(x = var_26075_cast_fp16, y = var_26076_to_fp16)[name = tensor("aw_2195_cast_fp16")]; + tensor var_26079_equation_0 = const()[name = tensor("op_26079_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26079_cast_fp16 = einsum(equation = var_26079_equation_0, values = (var_25921_cast_fp16, var_25838_cast_fp16))[name = tensor("op_26079_cast_fp16")]; + tensor var_26080_to_fp16 = const()[name = tensor("op_26080_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2197_cast_fp16 = mul(x = var_26079_cast_fp16, y = var_26080_to_fp16)[name = tensor("aw_2197_cast_fp16")]; + tensor var_26083_equation_0 = const()[name = tensor("op_26083_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26083_cast_fp16 = einsum(equation = var_26083_equation_0, values = (var_25925_cast_fp16, var_25842_cast_fp16))[name = tensor("op_26083_cast_fp16")]; + tensor var_26084_to_fp16 = const()[name = tensor("op_26084_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2199_cast_fp16 = mul(x = var_26083_cast_fp16, y = var_26084_to_fp16)[name = tensor("aw_2199_cast_fp16")]; + tensor var_26086_cast_fp16 = softmax(axis = var_21077, x = aw_2161_cast_fp16)[name = tensor("op_26086_cast_fp16")]; + tensor var_26087_cast_fp16 = softmax(axis = var_21077, x = aw_2163_cast_fp16)[name = tensor("op_26087_cast_fp16")]; + tensor var_26088_cast_fp16 = softmax(axis = var_21077, x = aw_2165_cast_fp16)[name = tensor("op_26088_cast_fp16")]; + tensor var_26089_cast_fp16 = softmax(axis = var_21077, x = aw_2167_cast_fp16)[name = tensor("op_26089_cast_fp16")]; + tensor var_26090_cast_fp16 = softmax(axis = var_21077, x = aw_2169_cast_fp16)[name = tensor("op_26090_cast_fp16")]; + tensor var_26091_cast_fp16 = softmax(axis = var_21077, x = aw_2171_cast_fp16)[name = tensor("op_26091_cast_fp16")]; + tensor var_26092_cast_fp16 = softmax(axis = var_21077, x = aw_2173_cast_fp16)[name = tensor("op_26092_cast_fp16")]; + tensor var_26093_cast_fp16 = softmax(axis = var_21077, x = aw_2175_cast_fp16)[name = tensor("op_26093_cast_fp16")]; + tensor var_26094_cast_fp16 = softmax(axis = var_21077, x = aw_2177_cast_fp16)[name = tensor("op_26094_cast_fp16")]; + tensor var_26095_cast_fp16 = softmax(axis = var_21077, x = aw_2179_cast_fp16)[name = tensor("op_26095_cast_fp16")]; + tensor var_26096_cast_fp16 = softmax(axis = var_21077, x = aw_2181_cast_fp16)[name = tensor("op_26096_cast_fp16")]; + tensor var_26097_cast_fp16 = softmax(axis = var_21077, x = aw_2183_cast_fp16)[name = tensor("op_26097_cast_fp16")]; + tensor var_26098_cast_fp16 = softmax(axis = var_21077, x = aw_2185_cast_fp16)[name = tensor("op_26098_cast_fp16")]; + tensor var_26099_cast_fp16 = softmax(axis = var_21077, x = aw_2187_cast_fp16)[name = tensor("op_26099_cast_fp16")]; + tensor var_26100_cast_fp16 = softmax(axis = var_21077, x = aw_2189_cast_fp16)[name = tensor("op_26100_cast_fp16")]; + tensor var_26101_cast_fp16 = softmax(axis = var_21077, x = aw_2191_cast_fp16)[name = tensor("op_26101_cast_fp16")]; + tensor var_26102_cast_fp16 = softmax(axis = var_21077, x = aw_2193_cast_fp16)[name = tensor("op_26102_cast_fp16")]; + tensor var_26103_cast_fp16 = softmax(axis = var_21077, x = aw_2195_cast_fp16)[name = tensor("op_26103_cast_fp16")]; + tensor var_26104_cast_fp16 = softmax(axis = var_21077, x = aw_2197_cast_fp16)[name = tensor("op_26104_cast_fp16")]; + tensor var_26105_cast_fp16 = softmax(axis = var_21077, x = aw_2199_cast_fp16)[name = tensor("op_26105_cast_fp16")]; + tensor var_26107_equation_0 = const()[name = tensor("op_26107_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26107_cast_fp16 = einsum(equation = var_26107_equation_0, values = (var_25927_cast_fp16, var_26086_cast_fp16))[name = tensor("op_26107_cast_fp16")]; + tensor var_26109_equation_0 = const()[name = tensor("op_26109_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26109_cast_fp16 = einsum(equation = var_26109_equation_0, values = (var_25931_cast_fp16, var_26087_cast_fp16))[name = tensor("op_26109_cast_fp16")]; + tensor var_26111_equation_0 = const()[name = tensor("op_26111_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26111_cast_fp16 = einsum(equation = var_26111_equation_0, values = (var_25935_cast_fp16, var_26088_cast_fp16))[name = tensor("op_26111_cast_fp16")]; + tensor var_26113_equation_0 = const()[name = tensor("op_26113_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26113_cast_fp16 = einsum(equation = var_26113_equation_0, values = (var_25939_cast_fp16, var_26089_cast_fp16))[name = tensor("op_26113_cast_fp16")]; + tensor var_26115_equation_0 = const()[name = tensor("op_26115_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26115_cast_fp16 = einsum(equation = var_26115_equation_0, values = (var_25943_cast_fp16, var_26090_cast_fp16))[name = tensor("op_26115_cast_fp16")]; + tensor var_26117_equation_0 = const()[name = tensor("op_26117_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26117_cast_fp16 = einsum(equation = var_26117_equation_0, values = (var_25947_cast_fp16, var_26091_cast_fp16))[name = tensor("op_26117_cast_fp16")]; + tensor var_26119_equation_0 = const()[name = tensor("op_26119_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26119_cast_fp16 = einsum(equation = var_26119_equation_0, values = (var_25951_cast_fp16, var_26092_cast_fp16))[name = tensor("op_26119_cast_fp16")]; + tensor var_26121_equation_0 = const()[name = tensor("op_26121_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26121_cast_fp16 = einsum(equation = var_26121_equation_0, values = (var_25955_cast_fp16, var_26093_cast_fp16))[name = tensor("op_26121_cast_fp16")]; + tensor var_26123_equation_0 = const()[name = tensor("op_26123_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26123_cast_fp16 = einsum(equation = var_26123_equation_0, values = (var_25959_cast_fp16, var_26094_cast_fp16))[name = tensor("op_26123_cast_fp16")]; + tensor var_26125_equation_0 = const()[name = tensor("op_26125_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26125_cast_fp16 = einsum(equation = var_26125_equation_0, values = (var_25963_cast_fp16, var_26095_cast_fp16))[name = tensor("op_26125_cast_fp16")]; + tensor var_26127_equation_0 = const()[name = tensor("op_26127_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26127_cast_fp16 = einsum(equation = var_26127_equation_0, values = (var_25967_cast_fp16, var_26096_cast_fp16))[name = tensor("op_26127_cast_fp16")]; + tensor var_26129_equation_0 = const()[name = tensor("op_26129_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26129_cast_fp16 = einsum(equation = var_26129_equation_0, values = (var_25971_cast_fp16, var_26097_cast_fp16))[name = tensor("op_26129_cast_fp16")]; + tensor var_26131_equation_0 = const()[name = tensor("op_26131_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26131_cast_fp16 = einsum(equation = var_26131_equation_0, values = (var_25975_cast_fp16, var_26098_cast_fp16))[name = tensor("op_26131_cast_fp16")]; + tensor var_26133_equation_0 = const()[name = tensor("op_26133_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26133_cast_fp16 = einsum(equation = var_26133_equation_0, values = (var_25979_cast_fp16, var_26099_cast_fp16))[name = tensor("op_26133_cast_fp16")]; + tensor var_26135_equation_0 = const()[name = tensor("op_26135_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26135_cast_fp16 = einsum(equation = var_26135_equation_0, values = (var_25983_cast_fp16, var_26100_cast_fp16))[name = tensor("op_26135_cast_fp16")]; + tensor var_26137_equation_0 = const()[name = tensor("op_26137_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26137_cast_fp16 = einsum(equation = var_26137_equation_0, values = (var_25987_cast_fp16, var_26101_cast_fp16))[name = tensor("op_26137_cast_fp16")]; + tensor var_26139_equation_0 = const()[name = tensor("op_26139_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26139_cast_fp16 = einsum(equation = var_26139_equation_0, values = (var_25991_cast_fp16, var_26102_cast_fp16))[name = tensor("op_26139_cast_fp16")]; + tensor var_26141_equation_0 = const()[name = tensor("op_26141_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26141_cast_fp16 = einsum(equation = var_26141_equation_0, values = (var_25995_cast_fp16, var_26103_cast_fp16))[name = tensor("op_26141_cast_fp16")]; + tensor var_26143_equation_0 = const()[name = tensor("op_26143_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26143_cast_fp16 = einsum(equation = var_26143_equation_0, values = (var_25999_cast_fp16, var_26104_cast_fp16))[name = tensor("op_26143_cast_fp16")]; + tensor var_26145_equation_0 = const()[name = tensor("op_26145_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26145_cast_fp16 = einsum(equation = var_26145_equation_0, values = (var_26003_cast_fp16, var_26105_cast_fp16))[name = tensor("op_26145_cast_fp16")]; + tensor input_367_interleave_0 = const()[name = tensor("input_367_interleave_0"), val = tensor(false)]; + tensor input_367_cast_fp16 = concat(axis = var_21077, interleave = input_367_interleave_0, values = (var_26107_cast_fp16, var_26109_cast_fp16, var_26111_cast_fp16, var_26113_cast_fp16, var_26115_cast_fp16, var_26117_cast_fp16, var_26119_cast_fp16, var_26121_cast_fp16, var_26123_cast_fp16, var_26125_cast_fp16, var_26127_cast_fp16, var_26129_cast_fp16, var_26131_cast_fp16, var_26133_cast_fp16, var_26135_cast_fp16, var_26137_cast_fp16, var_26139_cast_fp16, var_26141_cast_fp16, var_26143_cast_fp16, var_26145_cast_fp16))[name = tensor("input_367_cast_fp16")]; + tensor var_26151 = const()[name = tensor("op_26151"), val = tensor([1, 1])]; + tensor var_26153 = const()[name = tensor("op_26153"), val = tensor([1, 1])]; + tensor var_26155_pad_type_0 = const()[name = tensor("op_26155_pad_type_0"), val = tensor("custom")]; + tensor var_26155_pad_0 = const()[name = tensor("op_26155_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_5_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_5_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2083509696)))]; + tensor mid_block_attentions_0_transformer_blocks_5_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_5_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2086786560)))]; + tensor var_26155_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_5_attn1_to_out_0_bias_to_fp16, dilations = var_26153, groups = var_21077, pad = var_26155_pad_0, pad_type = var_26155_pad_type_0, strides = var_26151, weight = mid_block_attentions_0_transformer_blocks_5_attn1_to_out_0_weight_to_fp16, x = input_367_cast_fp16)[name = tensor("op_26155_cast_fp16")]; + tensor inputs_177_cast_fp16 = add(x = var_26155_cast_fp16, y = inputs_175_cast_fp16)[name = tensor("inputs_177_cast_fp16")]; + tensor hidden_states_241_axes_0 = const()[name = tensor("hidden_states_241_axes_0"), val = tensor([1])]; + tensor hidden_states_241_gamma_0_to_fp16 = const()[name = tensor("hidden_states_241_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2086789184)))]; + tensor hidden_states_241_beta_0_to_fp16 = const()[name = tensor("hidden_states_241_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2086791808)))]; + tensor var_26165_to_fp16 = const()[name = tensor("op_26165_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_241_cast_fp16 = layer_norm(axes = hidden_states_241_axes_0, beta = hidden_states_241_beta_0_to_fp16, epsilon = var_26165_to_fp16, gamma = hidden_states_241_gamma_0_to_fp16, x = inputs_177_cast_fp16)[name = tensor("hidden_states_241_cast_fp16")]; + tensor var_26180 = const()[name = tensor("op_26180"), val = tensor([1, 1])]; + tensor var_26182 = const()[name = tensor("op_26182"), val = tensor([1, 1])]; + tensor q_119_pad_type_0 = const()[name = tensor("q_119_pad_type_0"), val = tensor("custom")]; + tensor q_119_pad_0 = const()[name = tensor("q_119_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_5_attn2_to_q_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_5_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2086794432)))]; + tensor q_119_cast_fp16 = conv(dilations = var_26182, groups = var_21077, pad = q_119_pad_0, pad_type = q_119_pad_type_0, strides = var_26180, weight = mid_block_attentions_0_transformer_blocks_5_attn2_to_q_weight_to_fp16, x = hidden_states_241_cast_fp16)[name = tensor("q_119_cast_fp16")]; + tensor var_26186 = const()[name = tensor("op_26186"), val = tensor([1, 1])]; + tensor var_26188 = const()[name = tensor("op_26188"), val = tensor([1, 1])]; + tensor k_237_pad_type_0 = const()[name = tensor("k_237_pad_type_0"), val = tensor("custom")]; + tensor k_237_pad_0 = const()[name = tensor("k_237_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_5_attn2_to_k_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_5_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2090071296)))]; + tensor k_237_cast_fp16 = conv(dilations = var_26188, groups = var_21077, pad = k_237_pad_0, pad_type = k_237_pad_type_0, strides = var_26186, weight = mid_block_attentions_0_transformer_blocks_5_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_237_cast_fp16")]; + tensor var_26192 = const()[name = tensor("op_26192"), val = tensor([1, 1])]; + tensor var_26194 = const()[name = tensor("op_26194"), val = tensor([1, 1])]; + tensor v_119_pad_type_0 = const()[name = tensor("v_119_pad_type_0"), val = tensor("custom")]; + tensor v_119_pad_0 = const()[name = tensor("v_119_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_5_attn2_to_v_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_5_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2095314240)))]; + tensor v_119_cast_fp16 = conv(dilations = var_26194, groups = var_21077, pad = v_119_pad_0, pad_type = v_119_pad_type_0, strides = var_26192, weight = mid_block_attentions_0_transformer_blocks_5_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_119_cast_fp16")]; + tensor var_26198_begin_0 = const()[name = tensor("op_26198_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26198_end_0 = const()[name = tensor("op_26198_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_26198_end_mask_0 = const()[name = tensor("op_26198_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26198_cast_fp16 = slice_by_index(begin = var_26198_begin_0, end = var_26198_end_0, end_mask = var_26198_end_mask_0, x = q_119_cast_fp16)[name = tensor("op_26198_cast_fp16")]; + tensor var_26202_begin_0 = const()[name = tensor("op_26202_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_26202_end_0 = const()[name = tensor("op_26202_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_26202_end_mask_0 = const()[name = tensor("op_26202_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26202_cast_fp16 = slice_by_index(begin = var_26202_begin_0, end = var_26202_end_0, end_mask = var_26202_end_mask_0, x = q_119_cast_fp16)[name = tensor("op_26202_cast_fp16")]; + tensor var_26206_begin_0 = const()[name = tensor("op_26206_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_26206_end_0 = const()[name = tensor("op_26206_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_26206_end_mask_0 = const()[name = tensor("op_26206_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26206_cast_fp16 = slice_by_index(begin = var_26206_begin_0, end = var_26206_end_0, end_mask = var_26206_end_mask_0, x = q_119_cast_fp16)[name = tensor("op_26206_cast_fp16")]; + tensor var_26210_begin_0 = const()[name = tensor("op_26210_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_26210_end_0 = const()[name = tensor("op_26210_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_26210_end_mask_0 = const()[name = tensor("op_26210_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26210_cast_fp16 = slice_by_index(begin = var_26210_begin_0, end = var_26210_end_0, end_mask = var_26210_end_mask_0, x = q_119_cast_fp16)[name = tensor("op_26210_cast_fp16")]; + tensor var_26214_begin_0 = const()[name = tensor("op_26214_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_26214_end_0 = const()[name = tensor("op_26214_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_26214_end_mask_0 = const()[name = tensor("op_26214_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26214_cast_fp16 = slice_by_index(begin = var_26214_begin_0, end = var_26214_end_0, end_mask = var_26214_end_mask_0, x = q_119_cast_fp16)[name = tensor("op_26214_cast_fp16")]; + tensor var_26218_begin_0 = const()[name = tensor("op_26218_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_26218_end_0 = const()[name = tensor("op_26218_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_26218_end_mask_0 = const()[name = tensor("op_26218_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26218_cast_fp16 = slice_by_index(begin = var_26218_begin_0, end = var_26218_end_0, end_mask = var_26218_end_mask_0, x = q_119_cast_fp16)[name = tensor("op_26218_cast_fp16")]; + tensor var_26222_begin_0 = const()[name = tensor("op_26222_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_26222_end_0 = const()[name = tensor("op_26222_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_26222_end_mask_0 = const()[name = tensor("op_26222_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26222_cast_fp16 = slice_by_index(begin = var_26222_begin_0, end = var_26222_end_0, end_mask = var_26222_end_mask_0, x = q_119_cast_fp16)[name = tensor("op_26222_cast_fp16")]; + tensor var_26226_begin_0 = const()[name = tensor("op_26226_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_26226_end_0 = const()[name = tensor("op_26226_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_26226_end_mask_0 = const()[name = tensor("op_26226_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26226_cast_fp16 = slice_by_index(begin = var_26226_begin_0, end = var_26226_end_0, end_mask = var_26226_end_mask_0, x = q_119_cast_fp16)[name = tensor("op_26226_cast_fp16")]; + tensor var_26230_begin_0 = const()[name = tensor("op_26230_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_26230_end_0 = const()[name = tensor("op_26230_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_26230_end_mask_0 = const()[name = tensor("op_26230_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26230_cast_fp16 = slice_by_index(begin = var_26230_begin_0, end = var_26230_end_0, end_mask = var_26230_end_mask_0, x = q_119_cast_fp16)[name = tensor("op_26230_cast_fp16")]; + tensor var_26234_begin_0 = const()[name = tensor("op_26234_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_26234_end_0 = const()[name = tensor("op_26234_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_26234_end_mask_0 = const()[name = tensor("op_26234_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26234_cast_fp16 = slice_by_index(begin = var_26234_begin_0, end = var_26234_end_0, end_mask = var_26234_end_mask_0, x = q_119_cast_fp16)[name = tensor("op_26234_cast_fp16")]; + tensor var_26238_begin_0 = const()[name = tensor("op_26238_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_26238_end_0 = const()[name = tensor("op_26238_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_26238_end_mask_0 = const()[name = tensor("op_26238_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26238_cast_fp16 = slice_by_index(begin = var_26238_begin_0, end = var_26238_end_0, end_mask = var_26238_end_mask_0, x = q_119_cast_fp16)[name = tensor("op_26238_cast_fp16")]; + tensor var_26242_begin_0 = const()[name = tensor("op_26242_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_26242_end_0 = const()[name = tensor("op_26242_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_26242_end_mask_0 = const()[name = tensor("op_26242_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26242_cast_fp16 = slice_by_index(begin = var_26242_begin_0, end = var_26242_end_0, end_mask = var_26242_end_mask_0, x = q_119_cast_fp16)[name = tensor("op_26242_cast_fp16")]; + tensor var_26246_begin_0 = const()[name = tensor("op_26246_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_26246_end_0 = const()[name = tensor("op_26246_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_26246_end_mask_0 = const()[name = tensor("op_26246_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26246_cast_fp16 = slice_by_index(begin = var_26246_begin_0, end = var_26246_end_0, end_mask = var_26246_end_mask_0, x = q_119_cast_fp16)[name = tensor("op_26246_cast_fp16")]; + tensor var_26250_begin_0 = const()[name = tensor("op_26250_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_26250_end_0 = const()[name = tensor("op_26250_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_26250_end_mask_0 = const()[name = tensor("op_26250_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26250_cast_fp16 = slice_by_index(begin = var_26250_begin_0, end = var_26250_end_0, end_mask = var_26250_end_mask_0, x = q_119_cast_fp16)[name = tensor("op_26250_cast_fp16")]; + tensor var_26254_begin_0 = const()[name = tensor("op_26254_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_26254_end_0 = const()[name = tensor("op_26254_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_26254_end_mask_0 = const()[name = tensor("op_26254_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26254_cast_fp16 = slice_by_index(begin = var_26254_begin_0, end = var_26254_end_0, end_mask = var_26254_end_mask_0, x = q_119_cast_fp16)[name = tensor("op_26254_cast_fp16")]; + tensor var_26258_begin_0 = const()[name = tensor("op_26258_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_26258_end_0 = const()[name = tensor("op_26258_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_26258_end_mask_0 = const()[name = tensor("op_26258_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26258_cast_fp16 = slice_by_index(begin = var_26258_begin_0, end = var_26258_end_0, end_mask = var_26258_end_mask_0, x = q_119_cast_fp16)[name = tensor("op_26258_cast_fp16")]; + tensor var_26262_begin_0 = const()[name = tensor("op_26262_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_26262_end_0 = const()[name = tensor("op_26262_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_26262_end_mask_0 = const()[name = tensor("op_26262_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26262_cast_fp16 = slice_by_index(begin = var_26262_begin_0, end = var_26262_end_0, end_mask = var_26262_end_mask_0, x = q_119_cast_fp16)[name = tensor("op_26262_cast_fp16")]; + tensor var_26266_begin_0 = const()[name = tensor("op_26266_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_26266_end_0 = const()[name = tensor("op_26266_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_26266_end_mask_0 = const()[name = tensor("op_26266_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26266_cast_fp16 = slice_by_index(begin = var_26266_begin_0, end = var_26266_end_0, end_mask = var_26266_end_mask_0, x = q_119_cast_fp16)[name = tensor("op_26266_cast_fp16")]; + tensor var_26270_begin_0 = const()[name = tensor("op_26270_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_26270_end_0 = const()[name = tensor("op_26270_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_26270_end_mask_0 = const()[name = tensor("op_26270_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26270_cast_fp16 = slice_by_index(begin = var_26270_begin_0, end = var_26270_end_0, end_mask = var_26270_end_mask_0, x = q_119_cast_fp16)[name = tensor("op_26270_cast_fp16")]; + tensor var_26274_begin_0 = const()[name = tensor("op_26274_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_26274_end_0 = const()[name = tensor("op_26274_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_26274_end_mask_0 = const()[name = tensor("op_26274_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26274_cast_fp16 = slice_by_index(begin = var_26274_begin_0, end = var_26274_end_0, end_mask = var_26274_end_mask_0, x = q_119_cast_fp16)[name = tensor("op_26274_cast_fp16")]; + tensor k_239_perm_0 = const()[name = tensor("k_239_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_26281_begin_0 = const()[name = tensor("op_26281_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26281_end_0 = const()[name = tensor("op_26281_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_26281_end_mask_0 = const()[name = tensor("op_26281_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_239_cast_fp16 = transpose(perm = k_239_perm_0, x = k_237_cast_fp16)[name = tensor("transpose_80")]; + tensor var_26281_cast_fp16 = slice_by_index(begin = var_26281_begin_0, end = var_26281_end_0, end_mask = var_26281_end_mask_0, x = k_239_cast_fp16)[name = tensor("op_26281_cast_fp16")]; + tensor var_26285_begin_0 = const()[name = tensor("op_26285_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_26285_end_0 = const()[name = tensor("op_26285_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_26285_end_mask_0 = const()[name = tensor("op_26285_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26285_cast_fp16 = slice_by_index(begin = var_26285_begin_0, end = var_26285_end_0, end_mask = var_26285_end_mask_0, x = k_239_cast_fp16)[name = tensor("op_26285_cast_fp16")]; + tensor var_26289_begin_0 = const()[name = tensor("op_26289_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_26289_end_0 = const()[name = tensor("op_26289_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_26289_end_mask_0 = const()[name = tensor("op_26289_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26289_cast_fp16 = slice_by_index(begin = var_26289_begin_0, end = var_26289_end_0, end_mask = var_26289_end_mask_0, x = k_239_cast_fp16)[name = tensor("op_26289_cast_fp16")]; + tensor var_26293_begin_0 = const()[name = tensor("op_26293_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_26293_end_0 = const()[name = tensor("op_26293_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_26293_end_mask_0 = const()[name = tensor("op_26293_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26293_cast_fp16 = slice_by_index(begin = var_26293_begin_0, end = var_26293_end_0, end_mask = var_26293_end_mask_0, x = k_239_cast_fp16)[name = tensor("op_26293_cast_fp16")]; + tensor var_26297_begin_0 = const()[name = tensor("op_26297_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_26297_end_0 = const()[name = tensor("op_26297_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_26297_end_mask_0 = const()[name = tensor("op_26297_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26297_cast_fp16 = slice_by_index(begin = var_26297_begin_0, end = var_26297_end_0, end_mask = var_26297_end_mask_0, x = k_239_cast_fp16)[name = tensor("op_26297_cast_fp16")]; + tensor var_26301_begin_0 = const()[name = tensor("op_26301_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_26301_end_0 = const()[name = tensor("op_26301_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_26301_end_mask_0 = const()[name = tensor("op_26301_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26301_cast_fp16 = slice_by_index(begin = var_26301_begin_0, end = var_26301_end_0, end_mask = var_26301_end_mask_0, x = k_239_cast_fp16)[name = tensor("op_26301_cast_fp16")]; + tensor var_26305_begin_0 = const()[name = tensor("op_26305_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_26305_end_0 = const()[name = tensor("op_26305_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_26305_end_mask_0 = const()[name = tensor("op_26305_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26305_cast_fp16 = slice_by_index(begin = var_26305_begin_0, end = var_26305_end_0, end_mask = var_26305_end_mask_0, x = k_239_cast_fp16)[name = tensor("op_26305_cast_fp16")]; + tensor var_26309_begin_0 = const()[name = tensor("op_26309_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_26309_end_0 = const()[name = tensor("op_26309_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_26309_end_mask_0 = const()[name = tensor("op_26309_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26309_cast_fp16 = slice_by_index(begin = var_26309_begin_0, end = var_26309_end_0, end_mask = var_26309_end_mask_0, x = k_239_cast_fp16)[name = tensor("op_26309_cast_fp16")]; + tensor var_26313_begin_0 = const()[name = tensor("op_26313_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_26313_end_0 = const()[name = tensor("op_26313_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_26313_end_mask_0 = const()[name = tensor("op_26313_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26313_cast_fp16 = slice_by_index(begin = var_26313_begin_0, end = var_26313_end_0, end_mask = var_26313_end_mask_0, x = k_239_cast_fp16)[name = tensor("op_26313_cast_fp16")]; + tensor var_26317_begin_0 = const()[name = tensor("op_26317_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_26317_end_0 = const()[name = tensor("op_26317_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_26317_end_mask_0 = const()[name = tensor("op_26317_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26317_cast_fp16 = slice_by_index(begin = var_26317_begin_0, end = var_26317_end_0, end_mask = var_26317_end_mask_0, x = k_239_cast_fp16)[name = tensor("op_26317_cast_fp16")]; + tensor var_26321_begin_0 = const()[name = tensor("op_26321_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_26321_end_0 = const()[name = tensor("op_26321_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_26321_end_mask_0 = const()[name = tensor("op_26321_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26321_cast_fp16 = slice_by_index(begin = var_26321_begin_0, end = var_26321_end_0, end_mask = var_26321_end_mask_0, x = k_239_cast_fp16)[name = tensor("op_26321_cast_fp16")]; + tensor var_26325_begin_0 = const()[name = tensor("op_26325_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_26325_end_0 = const()[name = tensor("op_26325_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_26325_end_mask_0 = const()[name = tensor("op_26325_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26325_cast_fp16 = slice_by_index(begin = var_26325_begin_0, end = var_26325_end_0, end_mask = var_26325_end_mask_0, x = k_239_cast_fp16)[name = tensor("op_26325_cast_fp16")]; + tensor var_26329_begin_0 = const()[name = tensor("op_26329_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_26329_end_0 = const()[name = tensor("op_26329_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_26329_end_mask_0 = const()[name = tensor("op_26329_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26329_cast_fp16 = slice_by_index(begin = var_26329_begin_0, end = var_26329_end_0, end_mask = var_26329_end_mask_0, x = k_239_cast_fp16)[name = tensor("op_26329_cast_fp16")]; + tensor var_26333_begin_0 = const()[name = tensor("op_26333_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_26333_end_0 = const()[name = tensor("op_26333_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_26333_end_mask_0 = const()[name = tensor("op_26333_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26333_cast_fp16 = slice_by_index(begin = var_26333_begin_0, end = var_26333_end_0, end_mask = var_26333_end_mask_0, x = k_239_cast_fp16)[name = tensor("op_26333_cast_fp16")]; + tensor var_26337_begin_0 = const()[name = tensor("op_26337_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_26337_end_0 = const()[name = tensor("op_26337_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_26337_end_mask_0 = const()[name = tensor("op_26337_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26337_cast_fp16 = slice_by_index(begin = var_26337_begin_0, end = var_26337_end_0, end_mask = var_26337_end_mask_0, x = k_239_cast_fp16)[name = tensor("op_26337_cast_fp16")]; + tensor var_26341_begin_0 = const()[name = tensor("op_26341_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_26341_end_0 = const()[name = tensor("op_26341_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_26341_end_mask_0 = const()[name = tensor("op_26341_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26341_cast_fp16 = slice_by_index(begin = var_26341_begin_0, end = var_26341_end_0, end_mask = var_26341_end_mask_0, x = k_239_cast_fp16)[name = tensor("op_26341_cast_fp16")]; + tensor var_26345_begin_0 = const()[name = tensor("op_26345_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_26345_end_0 = const()[name = tensor("op_26345_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_26345_end_mask_0 = const()[name = tensor("op_26345_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26345_cast_fp16 = slice_by_index(begin = var_26345_begin_0, end = var_26345_end_0, end_mask = var_26345_end_mask_0, x = k_239_cast_fp16)[name = tensor("op_26345_cast_fp16")]; + tensor var_26349_begin_0 = const()[name = tensor("op_26349_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_26349_end_0 = const()[name = tensor("op_26349_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_26349_end_mask_0 = const()[name = tensor("op_26349_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26349_cast_fp16 = slice_by_index(begin = var_26349_begin_0, end = var_26349_end_0, end_mask = var_26349_end_mask_0, x = k_239_cast_fp16)[name = tensor("op_26349_cast_fp16")]; + tensor var_26353_begin_0 = const()[name = tensor("op_26353_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_26353_end_0 = const()[name = tensor("op_26353_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_26353_end_mask_0 = const()[name = tensor("op_26353_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26353_cast_fp16 = slice_by_index(begin = var_26353_begin_0, end = var_26353_end_0, end_mask = var_26353_end_mask_0, x = k_239_cast_fp16)[name = tensor("op_26353_cast_fp16")]; + tensor var_26357_begin_0 = const()[name = tensor("op_26357_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_26357_end_0 = const()[name = tensor("op_26357_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_26357_end_mask_0 = const()[name = tensor("op_26357_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26357_cast_fp16 = slice_by_index(begin = var_26357_begin_0, end = var_26357_end_0, end_mask = var_26357_end_mask_0, x = k_239_cast_fp16)[name = tensor("op_26357_cast_fp16")]; + tensor var_26359_begin_0 = const()[name = tensor("op_26359_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26359_end_0 = const()[name = tensor("op_26359_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_26359_end_mask_0 = const()[name = tensor("op_26359_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26359_cast_fp16 = slice_by_index(begin = var_26359_begin_0, end = var_26359_end_0, end_mask = var_26359_end_mask_0, x = v_119_cast_fp16)[name = tensor("op_26359_cast_fp16")]; + tensor var_26363_begin_0 = const()[name = tensor("op_26363_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_26363_end_0 = const()[name = tensor("op_26363_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_26363_end_mask_0 = const()[name = tensor("op_26363_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26363_cast_fp16 = slice_by_index(begin = var_26363_begin_0, end = var_26363_end_0, end_mask = var_26363_end_mask_0, x = v_119_cast_fp16)[name = tensor("op_26363_cast_fp16")]; + tensor var_26367_begin_0 = const()[name = tensor("op_26367_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_26367_end_0 = const()[name = tensor("op_26367_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_26367_end_mask_0 = const()[name = tensor("op_26367_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26367_cast_fp16 = slice_by_index(begin = var_26367_begin_0, end = var_26367_end_0, end_mask = var_26367_end_mask_0, x = v_119_cast_fp16)[name = tensor("op_26367_cast_fp16")]; + tensor var_26371_begin_0 = const()[name = tensor("op_26371_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_26371_end_0 = const()[name = tensor("op_26371_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_26371_end_mask_0 = const()[name = tensor("op_26371_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26371_cast_fp16 = slice_by_index(begin = var_26371_begin_0, end = var_26371_end_0, end_mask = var_26371_end_mask_0, x = v_119_cast_fp16)[name = tensor("op_26371_cast_fp16")]; + tensor var_26375_begin_0 = const()[name = tensor("op_26375_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_26375_end_0 = const()[name = tensor("op_26375_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_26375_end_mask_0 = const()[name = tensor("op_26375_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26375_cast_fp16 = slice_by_index(begin = var_26375_begin_0, end = var_26375_end_0, end_mask = var_26375_end_mask_0, x = v_119_cast_fp16)[name = tensor("op_26375_cast_fp16")]; + tensor var_26379_begin_0 = const()[name = tensor("op_26379_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_26379_end_0 = const()[name = tensor("op_26379_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_26379_end_mask_0 = const()[name = tensor("op_26379_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26379_cast_fp16 = slice_by_index(begin = var_26379_begin_0, end = var_26379_end_0, end_mask = var_26379_end_mask_0, x = v_119_cast_fp16)[name = tensor("op_26379_cast_fp16")]; + tensor var_26383_begin_0 = const()[name = tensor("op_26383_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_26383_end_0 = const()[name = tensor("op_26383_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_26383_end_mask_0 = const()[name = tensor("op_26383_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26383_cast_fp16 = slice_by_index(begin = var_26383_begin_0, end = var_26383_end_0, end_mask = var_26383_end_mask_0, x = v_119_cast_fp16)[name = tensor("op_26383_cast_fp16")]; + tensor var_26387_begin_0 = const()[name = tensor("op_26387_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_26387_end_0 = const()[name = tensor("op_26387_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_26387_end_mask_0 = const()[name = tensor("op_26387_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26387_cast_fp16 = slice_by_index(begin = var_26387_begin_0, end = var_26387_end_0, end_mask = var_26387_end_mask_0, x = v_119_cast_fp16)[name = tensor("op_26387_cast_fp16")]; + tensor var_26391_begin_0 = const()[name = tensor("op_26391_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_26391_end_0 = const()[name = tensor("op_26391_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_26391_end_mask_0 = const()[name = tensor("op_26391_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26391_cast_fp16 = slice_by_index(begin = var_26391_begin_0, end = var_26391_end_0, end_mask = var_26391_end_mask_0, x = v_119_cast_fp16)[name = tensor("op_26391_cast_fp16")]; + tensor var_26395_begin_0 = const()[name = tensor("op_26395_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_26395_end_0 = const()[name = tensor("op_26395_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_26395_end_mask_0 = const()[name = tensor("op_26395_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26395_cast_fp16 = slice_by_index(begin = var_26395_begin_0, end = var_26395_end_0, end_mask = var_26395_end_mask_0, x = v_119_cast_fp16)[name = tensor("op_26395_cast_fp16")]; + tensor var_26399_begin_0 = const()[name = tensor("op_26399_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_26399_end_0 = const()[name = tensor("op_26399_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_26399_end_mask_0 = const()[name = tensor("op_26399_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26399_cast_fp16 = slice_by_index(begin = var_26399_begin_0, end = var_26399_end_0, end_mask = var_26399_end_mask_0, x = v_119_cast_fp16)[name = tensor("op_26399_cast_fp16")]; + tensor var_26403_begin_0 = const()[name = tensor("op_26403_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_26403_end_0 = const()[name = tensor("op_26403_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_26403_end_mask_0 = const()[name = tensor("op_26403_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26403_cast_fp16 = slice_by_index(begin = var_26403_begin_0, end = var_26403_end_0, end_mask = var_26403_end_mask_0, x = v_119_cast_fp16)[name = tensor("op_26403_cast_fp16")]; + tensor var_26407_begin_0 = const()[name = tensor("op_26407_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_26407_end_0 = const()[name = tensor("op_26407_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_26407_end_mask_0 = const()[name = tensor("op_26407_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26407_cast_fp16 = slice_by_index(begin = var_26407_begin_0, end = var_26407_end_0, end_mask = var_26407_end_mask_0, x = v_119_cast_fp16)[name = tensor("op_26407_cast_fp16")]; + tensor var_26411_begin_0 = const()[name = tensor("op_26411_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_26411_end_0 = const()[name = tensor("op_26411_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_26411_end_mask_0 = const()[name = tensor("op_26411_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26411_cast_fp16 = slice_by_index(begin = var_26411_begin_0, end = var_26411_end_0, end_mask = var_26411_end_mask_0, x = v_119_cast_fp16)[name = tensor("op_26411_cast_fp16")]; + tensor var_26415_begin_0 = const()[name = tensor("op_26415_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_26415_end_0 = const()[name = tensor("op_26415_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_26415_end_mask_0 = const()[name = tensor("op_26415_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26415_cast_fp16 = slice_by_index(begin = var_26415_begin_0, end = var_26415_end_0, end_mask = var_26415_end_mask_0, x = v_119_cast_fp16)[name = tensor("op_26415_cast_fp16")]; + tensor var_26419_begin_0 = const()[name = tensor("op_26419_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_26419_end_0 = const()[name = tensor("op_26419_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_26419_end_mask_0 = const()[name = tensor("op_26419_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26419_cast_fp16 = slice_by_index(begin = var_26419_begin_0, end = var_26419_end_0, end_mask = var_26419_end_mask_0, x = v_119_cast_fp16)[name = tensor("op_26419_cast_fp16")]; + tensor var_26423_begin_0 = const()[name = tensor("op_26423_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_26423_end_0 = const()[name = tensor("op_26423_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_26423_end_mask_0 = const()[name = tensor("op_26423_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26423_cast_fp16 = slice_by_index(begin = var_26423_begin_0, end = var_26423_end_0, end_mask = var_26423_end_mask_0, x = v_119_cast_fp16)[name = tensor("op_26423_cast_fp16")]; + tensor var_26427_begin_0 = const()[name = tensor("op_26427_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_26427_end_0 = const()[name = tensor("op_26427_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_26427_end_mask_0 = const()[name = tensor("op_26427_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26427_cast_fp16 = slice_by_index(begin = var_26427_begin_0, end = var_26427_end_0, end_mask = var_26427_end_mask_0, x = v_119_cast_fp16)[name = tensor("op_26427_cast_fp16")]; + tensor var_26431_begin_0 = const()[name = tensor("op_26431_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_26431_end_0 = const()[name = tensor("op_26431_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_26431_end_mask_0 = const()[name = tensor("op_26431_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26431_cast_fp16 = slice_by_index(begin = var_26431_begin_0, end = var_26431_end_0, end_mask = var_26431_end_mask_0, x = v_119_cast_fp16)[name = tensor("op_26431_cast_fp16")]; + tensor var_26435_begin_0 = const()[name = tensor("op_26435_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_26435_end_0 = const()[name = tensor("op_26435_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_26435_end_mask_0 = const()[name = tensor("op_26435_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26435_cast_fp16 = slice_by_index(begin = var_26435_begin_0, end = var_26435_end_0, end_mask = var_26435_end_mask_0, x = v_119_cast_fp16)[name = tensor("op_26435_cast_fp16")]; + tensor var_26439_equation_0 = const()[name = tensor("op_26439_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26439_cast_fp16 = einsum(equation = var_26439_equation_0, values = (var_26281_cast_fp16, var_26198_cast_fp16))[name = tensor("op_26439_cast_fp16")]; + tensor var_26440_to_fp16 = const()[name = tensor("op_26440_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2201_cast_fp16 = mul(x = var_26439_cast_fp16, y = var_26440_to_fp16)[name = tensor("aw_2201_cast_fp16")]; + tensor var_26443_equation_0 = const()[name = tensor("op_26443_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26443_cast_fp16 = einsum(equation = var_26443_equation_0, values = (var_26285_cast_fp16, var_26202_cast_fp16))[name = tensor("op_26443_cast_fp16")]; + tensor var_26444_to_fp16 = const()[name = tensor("op_26444_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2203_cast_fp16 = mul(x = var_26443_cast_fp16, y = var_26444_to_fp16)[name = tensor("aw_2203_cast_fp16")]; + tensor var_26447_equation_0 = const()[name = tensor("op_26447_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26447_cast_fp16 = einsum(equation = var_26447_equation_0, values = (var_26289_cast_fp16, var_26206_cast_fp16))[name = tensor("op_26447_cast_fp16")]; + tensor var_26448_to_fp16 = const()[name = tensor("op_26448_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2205_cast_fp16 = mul(x = var_26447_cast_fp16, y = var_26448_to_fp16)[name = tensor("aw_2205_cast_fp16")]; + tensor var_26451_equation_0 = const()[name = tensor("op_26451_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26451_cast_fp16 = einsum(equation = var_26451_equation_0, values = (var_26293_cast_fp16, var_26210_cast_fp16))[name = tensor("op_26451_cast_fp16")]; + tensor var_26452_to_fp16 = const()[name = tensor("op_26452_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2207_cast_fp16 = mul(x = var_26451_cast_fp16, y = var_26452_to_fp16)[name = tensor("aw_2207_cast_fp16")]; + tensor var_26455_equation_0 = const()[name = tensor("op_26455_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26455_cast_fp16 = einsum(equation = var_26455_equation_0, values = (var_26297_cast_fp16, var_26214_cast_fp16))[name = tensor("op_26455_cast_fp16")]; + tensor var_26456_to_fp16 = const()[name = tensor("op_26456_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2209_cast_fp16 = mul(x = var_26455_cast_fp16, y = var_26456_to_fp16)[name = tensor("aw_2209_cast_fp16")]; + tensor var_26459_equation_0 = const()[name = tensor("op_26459_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26459_cast_fp16 = einsum(equation = var_26459_equation_0, values = (var_26301_cast_fp16, var_26218_cast_fp16))[name = tensor("op_26459_cast_fp16")]; + tensor var_26460_to_fp16 = const()[name = tensor("op_26460_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2211_cast_fp16 = mul(x = var_26459_cast_fp16, y = var_26460_to_fp16)[name = tensor("aw_2211_cast_fp16")]; + tensor var_26463_equation_0 = const()[name = tensor("op_26463_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26463_cast_fp16 = einsum(equation = var_26463_equation_0, values = (var_26305_cast_fp16, var_26222_cast_fp16))[name = tensor("op_26463_cast_fp16")]; + tensor var_26464_to_fp16 = const()[name = tensor("op_26464_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2213_cast_fp16 = mul(x = var_26463_cast_fp16, y = var_26464_to_fp16)[name = tensor("aw_2213_cast_fp16")]; + tensor var_26467_equation_0 = const()[name = tensor("op_26467_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26467_cast_fp16 = einsum(equation = var_26467_equation_0, values = (var_26309_cast_fp16, var_26226_cast_fp16))[name = tensor("op_26467_cast_fp16")]; + tensor var_26468_to_fp16 = const()[name = tensor("op_26468_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2215_cast_fp16 = mul(x = var_26467_cast_fp16, y = var_26468_to_fp16)[name = tensor("aw_2215_cast_fp16")]; + tensor var_26471_equation_0 = const()[name = tensor("op_26471_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26471_cast_fp16 = einsum(equation = var_26471_equation_0, values = (var_26313_cast_fp16, var_26230_cast_fp16))[name = tensor("op_26471_cast_fp16")]; + tensor var_26472_to_fp16 = const()[name = tensor("op_26472_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2217_cast_fp16 = mul(x = var_26471_cast_fp16, y = var_26472_to_fp16)[name = tensor("aw_2217_cast_fp16")]; + tensor var_26475_equation_0 = const()[name = tensor("op_26475_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26475_cast_fp16 = einsum(equation = var_26475_equation_0, values = (var_26317_cast_fp16, var_26234_cast_fp16))[name = tensor("op_26475_cast_fp16")]; + tensor var_26476_to_fp16 = const()[name = tensor("op_26476_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2219_cast_fp16 = mul(x = var_26475_cast_fp16, y = var_26476_to_fp16)[name = tensor("aw_2219_cast_fp16")]; + tensor var_26479_equation_0 = const()[name = tensor("op_26479_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26479_cast_fp16 = einsum(equation = var_26479_equation_0, values = (var_26321_cast_fp16, var_26238_cast_fp16))[name = tensor("op_26479_cast_fp16")]; + tensor var_26480_to_fp16 = const()[name = tensor("op_26480_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2221_cast_fp16 = mul(x = var_26479_cast_fp16, y = var_26480_to_fp16)[name = tensor("aw_2221_cast_fp16")]; + tensor var_26483_equation_0 = const()[name = tensor("op_26483_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26483_cast_fp16 = einsum(equation = var_26483_equation_0, values = (var_26325_cast_fp16, var_26242_cast_fp16))[name = tensor("op_26483_cast_fp16")]; + tensor var_26484_to_fp16 = const()[name = tensor("op_26484_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2223_cast_fp16 = mul(x = var_26483_cast_fp16, y = var_26484_to_fp16)[name = tensor("aw_2223_cast_fp16")]; + tensor var_26487_equation_0 = const()[name = tensor("op_26487_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26487_cast_fp16 = einsum(equation = var_26487_equation_0, values = (var_26329_cast_fp16, var_26246_cast_fp16))[name = tensor("op_26487_cast_fp16")]; + tensor var_26488_to_fp16 = const()[name = tensor("op_26488_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2225_cast_fp16 = mul(x = var_26487_cast_fp16, y = var_26488_to_fp16)[name = tensor("aw_2225_cast_fp16")]; + tensor var_26491_equation_0 = const()[name = tensor("op_26491_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26491_cast_fp16 = einsum(equation = var_26491_equation_0, values = (var_26333_cast_fp16, var_26250_cast_fp16))[name = tensor("op_26491_cast_fp16")]; + tensor var_26492_to_fp16 = const()[name = tensor("op_26492_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2227_cast_fp16 = mul(x = var_26491_cast_fp16, y = var_26492_to_fp16)[name = tensor("aw_2227_cast_fp16")]; + tensor var_26495_equation_0 = const()[name = tensor("op_26495_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26495_cast_fp16 = einsum(equation = var_26495_equation_0, values = (var_26337_cast_fp16, var_26254_cast_fp16))[name = tensor("op_26495_cast_fp16")]; + tensor var_26496_to_fp16 = const()[name = tensor("op_26496_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2229_cast_fp16 = mul(x = var_26495_cast_fp16, y = var_26496_to_fp16)[name = tensor("aw_2229_cast_fp16")]; + tensor var_26499_equation_0 = const()[name = tensor("op_26499_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26499_cast_fp16 = einsum(equation = var_26499_equation_0, values = (var_26341_cast_fp16, var_26258_cast_fp16))[name = tensor("op_26499_cast_fp16")]; + tensor var_26500_to_fp16 = const()[name = tensor("op_26500_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2231_cast_fp16 = mul(x = var_26499_cast_fp16, y = var_26500_to_fp16)[name = tensor("aw_2231_cast_fp16")]; + tensor var_26503_equation_0 = const()[name = tensor("op_26503_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26503_cast_fp16 = einsum(equation = var_26503_equation_0, values = (var_26345_cast_fp16, var_26262_cast_fp16))[name = tensor("op_26503_cast_fp16")]; + tensor var_26504_to_fp16 = const()[name = tensor("op_26504_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2233_cast_fp16 = mul(x = var_26503_cast_fp16, y = var_26504_to_fp16)[name = tensor("aw_2233_cast_fp16")]; + tensor var_26507_equation_0 = const()[name = tensor("op_26507_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26507_cast_fp16 = einsum(equation = var_26507_equation_0, values = (var_26349_cast_fp16, var_26266_cast_fp16))[name = tensor("op_26507_cast_fp16")]; + tensor var_26508_to_fp16 = const()[name = tensor("op_26508_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2235_cast_fp16 = mul(x = var_26507_cast_fp16, y = var_26508_to_fp16)[name = tensor("aw_2235_cast_fp16")]; + tensor var_26511_equation_0 = const()[name = tensor("op_26511_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26511_cast_fp16 = einsum(equation = var_26511_equation_0, values = (var_26353_cast_fp16, var_26270_cast_fp16))[name = tensor("op_26511_cast_fp16")]; + tensor var_26512_to_fp16 = const()[name = tensor("op_26512_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2237_cast_fp16 = mul(x = var_26511_cast_fp16, y = var_26512_to_fp16)[name = tensor("aw_2237_cast_fp16")]; + tensor var_26515_equation_0 = const()[name = tensor("op_26515_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26515_cast_fp16 = einsum(equation = var_26515_equation_0, values = (var_26357_cast_fp16, var_26274_cast_fp16))[name = tensor("op_26515_cast_fp16")]; + tensor var_26516_to_fp16 = const()[name = tensor("op_26516_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2239_cast_fp16 = mul(x = var_26515_cast_fp16, y = var_26516_to_fp16)[name = tensor("aw_2239_cast_fp16")]; + tensor var_26518_cast_fp16 = softmax(axis = var_21077, x = aw_2201_cast_fp16)[name = tensor("op_26518_cast_fp16")]; + tensor var_26519_cast_fp16 = softmax(axis = var_21077, x = aw_2203_cast_fp16)[name = tensor("op_26519_cast_fp16")]; + tensor var_26520_cast_fp16 = softmax(axis = var_21077, x = aw_2205_cast_fp16)[name = tensor("op_26520_cast_fp16")]; + tensor var_26521_cast_fp16 = softmax(axis = var_21077, x = aw_2207_cast_fp16)[name = tensor("op_26521_cast_fp16")]; + tensor var_26522_cast_fp16 = softmax(axis = var_21077, x = aw_2209_cast_fp16)[name = tensor("op_26522_cast_fp16")]; + tensor var_26523_cast_fp16 = softmax(axis = var_21077, x = aw_2211_cast_fp16)[name = tensor("op_26523_cast_fp16")]; + tensor var_26524_cast_fp16 = softmax(axis = var_21077, x = aw_2213_cast_fp16)[name = tensor("op_26524_cast_fp16")]; + tensor var_26525_cast_fp16 = softmax(axis = var_21077, x = aw_2215_cast_fp16)[name = tensor("op_26525_cast_fp16")]; + tensor var_26526_cast_fp16 = softmax(axis = var_21077, x = aw_2217_cast_fp16)[name = tensor("op_26526_cast_fp16")]; + tensor var_26527_cast_fp16 = softmax(axis = var_21077, x = aw_2219_cast_fp16)[name = tensor("op_26527_cast_fp16")]; + tensor var_26528_cast_fp16 = softmax(axis = var_21077, x = aw_2221_cast_fp16)[name = tensor("op_26528_cast_fp16")]; + tensor var_26529_cast_fp16 = softmax(axis = var_21077, x = aw_2223_cast_fp16)[name = tensor("op_26529_cast_fp16")]; + tensor var_26530_cast_fp16 = softmax(axis = var_21077, x = aw_2225_cast_fp16)[name = tensor("op_26530_cast_fp16")]; + tensor var_26531_cast_fp16 = softmax(axis = var_21077, x = aw_2227_cast_fp16)[name = tensor("op_26531_cast_fp16")]; + tensor var_26532_cast_fp16 = softmax(axis = var_21077, x = aw_2229_cast_fp16)[name = tensor("op_26532_cast_fp16")]; + tensor var_26533_cast_fp16 = softmax(axis = var_21077, x = aw_2231_cast_fp16)[name = tensor("op_26533_cast_fp16")]; + tensor var_26534_cast_fp16 = softmax(axis = var_21077, x = aw_2233_cast_fp16)[name = tensor("op_26534_cast_fp16")]; + tensor var_26535_cast_fp16 = softmax(axis = var_21077, x = aw_2235_cast_fp16)[name = tensor("op_26535_cast_fp16")]; + tensor var_26536_cast_fp16 = softmax(axis = var_21077, x = aw_2237_cast_fp16)[name = tensor("op_26536_cast_fp16")]; + tensor var_26537_cast_fp16 = softmax(axis = var_21077, x = aw_2239_cast_fp16)[name = tensor("op_26537_cast_fp16")]; + tensor var_26539_equation_0 = const()[name = tensor("op_26539_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26539_cast_fp16 = einsum(equation = var_26539_equation_0, values = (var_26359_cast_fp16, var_26518_cast_fp16))[name = tensor("op_26539_cast_fp16")]; + tensor var_26541_equation_0 = const()[name = tensor("op_26541_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26541_cast_fp16 = einsum(equation = var_26541_equation_0, values = (var_26363_cast_fp16, var_26519_cast_fp16))[name = tensor("op_26541_cast_fp16")]; + tensor var_26543_equation_0 = const()[name = tensor("op_26543_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26543_cast_fp16 = einsum(equation = var_26543_equation_0, values = (var_26367_cast_fp16, var_26520_cast_fp16))[name = tensor("op_26543_cast_fp16")]; + tensor var_26545_equation_0 = const()[name = tensor("op_26545_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26545_cast_fp16 = einsum(equation = var_26545_equation_0, values = (var_26371_cast_fp16, var_26521_cast_fp16))[name = tensor("op_26545_cast_fp16")]; + tensor var_26547_equation_0 = const()[name = tensor("op_26547_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26547_cast_fp16 = einsum(equation = var_26547_equation_0, values = (var_26375_cast_fp16, var_26522_cast_fp16))[name = tensor("op_26547_cast_fp16")]; + tensor var_26549_equation_0 = const()[name = tensor("op_26549_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26549_cast_fp16 = einsum(equation = var_26549_equation_0, values = (var_26379_cast_fp16, var_26523_cast_fp16))[name = tensor("op_26549_cast_fp16")]; + tensor var_26551_equation_0 = const()[name = tensor("op_26551_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26551_cast_fp16 = einsum(equation = var_26551_equation_0, values = (var_26383_cast_fp16, var_26524_cast_fp16))[name = tensor("op_26551_cast_fp16")]; + tensor var_26553_equation_0 = const()[name = tensor("op_26553_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26553_cast_fp16 = einsum(equation = var_26553_equation_0, values = (var_26387_cast_fp16, var_26525_cast_fp16))[name = tensor("op_26553_cast_fp16")]; + tensor var_26555_equation_0 = const()[name = tensor("op_26555_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26555_cast_fp16 = einsum(equation = var_26555_equation_0, values = (var_26391_cast_fp16, var_26526_cast_fp16))[name = tensor("op_26555_cast_fp16")]; + tensor var_26557_equation_0 = const()[name = tensor("op_26557_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26557_cast_fp16 = einsum(equation = var_26557_equation_0, values = (var_26395_cast_fp16, var_26527_cast_fp16))[name = tensor("op_26557_cast_fp16")]; + tensor var_26559_equation_0 = const()[name = tensor("op_26559_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26559_cast_fp16 = einsum(equation = var_26559_equation_0, values = (var_26399_cast_fp16, var_26528_cast_fp16))[name = tensor("op_26559_cast_fp16")]; + tensor var_26561_equation_0 = const()[name = tensor("op_26561_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26561_cast_fp16 = einsum(equation = var_26561_equation_0, values = (var_26403_cast_fp16, var_26529_cast_fp16))[name = tensor("op_26561_cast_fp16")]; + tensor var_26563_equation_0 = const()[name = tensor("op_26563_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26563_cast_fp16 = einsum(equation = var_26563_equation_0, values = (var_26407_cast_fp16, var_26530_cast_fp16))[name = tensor("op_26563_cast_fp16")]; + tensor var_26565_equation_0 = const()[name = tensor("op_26565_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26565_cast_fp16 = einsum(equation = var_26565_equation_0, values = (var_26411_cast_fp16, var_26531_cast_fp16))[name = tensor("op_26565_cast_fp16")]; + tensor var_26567_equation_0 = const()[name = tensor("op_26567_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26567_cast_fp16 = einsum(equation = var_26567_equation_0, values = (var_26415_cast_fp16, var_26532_cast_fp16))[name = tensor("op_26567_cast_fp16")]; + tensor var_26569_equation_0 = const()[name = tensor("op_26569_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26569_cast_fp16 = einsum(equation = var_26569_equation_0, values = (var_26419_cast_fp16, var_26533_cast_fp16))[name = tensor("op_26569_cast_fp16")]; + tensor var_26571_equation_0 = const()[name = tensor("op_26571_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26571_cast_fp16 = einsum(equation = var_26571_equation_0, values = (var_26423_cast_fp16, var_26534_cast_fp16))[name = tensor("op_26571_cast_fp16")]; + tensor var_26573_equation_0 = const()[name = tensor("op_26573_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26573_cast_fp16 = einsum(equation = var_26573_equation_0, values = (var_26427_cast_fp16, var_26535_cast_fp16))[name = tensor("op_26573_cast_fp16")]; + tensor var_26575_equation_0 = const()[name = tensor("op_26575_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26575_cast_fp16 = einsum(equation = var_26575_equation_0, values = (var_26431_cast_fp16, var_26536_cast_fp16))[name = tensor("op_26575_cast_fp16")]; + tensor var_26577_equation_0 = const()[name = tensor("op_26577_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26577_cast_fp16 = einsum(equation = var_26577_equation_0, values = (var_26435_cast_fp16, var_26537_cast_fp16))[name = tensor("op_26577_cast_fp16")]; + tensor input_369_interleave_0 = const()[name = tensor("input_369_interleave_0"), val = tensor(false)]; + tensor input_369_cast_fp16 = concat(axis = var_21077, interleave = input_369_interleave_0, values = (var_26539_cast_fp16, var_26541_cast_fp16, var_26543_cast_fp16, var_26545_cast_fp16, var_26547_cast_fp16, var_26549_cast_fp16, var_26551_cast_fp16, var_26553_cast_fp16, var_26555_cast_fp16, var_26557_cast_fp16, var_26559_cast_fp16, var_26561_cast_fp16, var_26563_cast_fp16, var_26565_cast_fp16, var_26567_cast_fp16, var_26569_cast_fp16, var_26571_cast_fp16, var_26573_cast_fp16, var_26575_cast_fp16, var_26577_cast_fp16))[name = tensor("input_369_cast_fp16")]; + tensor var_26583 = const()[name = tensor("op_26583"), val = tensor([1, 1])]; + tensor var_26585 = const()[name = tensor("op_26585"), val = tensor([1, 1])]; + tensor var_26587_pad_type_0 = const()[name = tensor("op_26587_pad_type_0"), val = tensor("custom")]; + tensor var_26587_pad_0 = const()[name = tensor("op_26587_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_5_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_5_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2100557184)))]; + tensor mid_block_attentions_0_transformer_blocks_5_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_5_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2103834048)))]; + tensor var_26587_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_5_attn2_to_out_0_bias_to_fp16, dilations = var_26585, groups = var_21077, pad = var_26587_pad_0, pad_type = var_26587_pad_type_0, strides = var_26583, weight = mid_block_attentions_0_transformer_blocks_5_attn2_to_out_0_weight_to_fp16, x = input_369_cast_fp16)[name = tensor("op_26587_cast_fp16")]; + tensor inputs_179_cast_fp16 = add(x = var_26587_cast_fp16, y = inputs_177_cast_fp16)[name = tensor("inputs_179_cast_fp16")]; + tensor input_371_axes_0 = const()[name = tensor("input_371_axes_0"), val = tensor([1])]; + tensor input_371_gamma_0_to_fp16 = const()[name = tensor("input_371_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2103836672)))]; + tensor input_371_beta_0_to_fp16 = const()[name = tensor("input_371_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2103839296)))]; + tensor var_26597_to_fp16 = const()[name = tensor("op_26597_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_371_cast_fp16 = layer_norm(axes = input_371_axes_0, beta = input_371_beta_0_to_fp16, epsilon = var_26597_to_fp16, gamma = input_371_gamma_0_to_fp16, x = inputs_179_cast_fp16)[name = tensor("input_371_cast_fp16")]; + tensor var_26613 = const()[name = tensor("op_26613"), val = tensor([1, 1])]; + tensor var_26615 = const()[name = tensor("op_26615"), val = tensor([1, 1])]; + tensor var_26617_pad_type_0 = const()[name = tensor("op_26617_pad_type_0"), val = tensor("custom")]; + tensor var_26617_pad_0 = const()[name = tensor("op_26617_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_5_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_5_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2103841920)))]; + tensor mid_block_attentions_0_transformer_blocks_5_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_5_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2130056384)))]; + tensor var_26617_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_5_ff_net_0_proj_bias_to_fp16, dilations = var_26615, groups = var_21077, pad = var_26617_pad_0, pad_type = var_26617_pad_type_0, strides = var_26613, weight = mid_block_attentions_0_transformer_blocks_5_ff_net_0_proj_weight_to_fp16, x = input_371_cast_fp16)[name = tensor("op_26617_cast_fp16")]; + tensor var_26618_split_sizes_0 = const()[name = tensor("op_26618_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_26618_axis_0 = const()[name = tensor("op_26618_axis_0"), val = tensor(1)]; + tensor var_26618_cast_fp16_0, tensor var_26618_cast_fp16_1 = split(axis = var_26618_axis_0, split_sizes = var_26618_split_sizes_0, x = var_26617_cast_fp16)[name = tensor("op_26618_cast_fp16")]; + tensor var_26620_mode_0 = const()[name = tensor("op_26620_mode_0"), val = tensor("EXACT")]; + tensor var_26620_cast_fp16 = gelu(mode = var_26620_mode_0, x = var_26618_cast_fp16_1)[name = tensor("op_26620_cast_fp16")]; + tensor input_373_cast_fp16 = mul(x = var_26618_cast_fp16_0, y = var_26620_cast_fp16)[name = tensor("input_373_cast_fp16")]; + tensor var_26624 = const()[name = tensor("op_26624"), val = tensor([1, 1])]; + tensor var_26626 = const()[name = tensor("op_26626"), val = tensor([1, 1])]; + tensor var_26628_pad_type_0 = const()[name = tensor("op_26628_pad_type_0"), val = tensor("custom")]; + tensor var_26628_pad_0 = const()[name = tensor("op_26628_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_5_ff_net_2_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_5_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2130076928)))]; + tensor mid_block_attentions_0_transformer_blocks_5_ff_net_2_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_5_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2143184192)))]; + tensor var_26628_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_5_ff_net_2_bias_to_fp16, dilations = var_26626, groups = var_21077, pad = var_26628_pad_0, pad_type = var_26628_pad_type_0, strides = var_26624, weight = mid_block_attentions_0_transformer_blocks_5_ff_net_2_weight_to_fp16, x = input_373_cast_fp16)[name = tensor("op_26628_cast_fp16")]; + tensor inputs_181_cast_fp16 = add(x = var_26628_cast_fp16, y = inputs_179_cast_fp16)[name = tensor("inputs_181_cast_fp16")]; + tensor hidden_states_245_axes_0 = const()[name = tensor("hidden_states_245_axes_0"), val = tensor([1])]; + tensor hidden_states_245_gamma_0_to_fp16 = const()[name = tensor("hidden_states_245_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2143186816)))]; + tensor hidden_states_245_beta_0_to_fp16 = const()[name = tensor("hidden_states_245_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2143189440)))]; + tensor var_26644_to_fp16 = const()[name = tensor("op_26644_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_245_cast_fp16 = layer_norm(axes = hidden_states_245_axes_0, beta = hidden_states_245_beta_0_to_fp16, epsilon = var_26644_to_fp16, gamma = hidden_states_245_gamma_0_to_fp16, x = inputs_181_cast_fp16)[name = tensor("hidden_states_245_cast_fp16")]; + tensor var_26659 = const()[name = tensor("op_26659"), val = tensor([1, 1])]; + tensor var_26661 = const()[name = tensor("op_26661"), val = tensor([1, 1])]; + tensor q_121_pad_type_0 = const()[name = tensor("q_121_pad_type_0"), val = tensor("custom")]; + tensor q_121_pad_0 = const()[name = tensor("q_121_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_6_attn1_to_q_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_6_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2143192064)))]; + tensor q_121_cast_fp16 = conv(dilations = var_26661, groups = var_21077, pad = q_121_pad_0, pad_type = q_121_pad_type_0, strides = var_26659, weight = mid_block_attentions_0_transformer_blocks_6_attn1_to_q_weight_to_fp16, x = hidden_states_245_cast_fp16)[name = tensor("q_121_cast_fp16")]; + tensor var_26665 = const()[name = tensor("op_26665"), val = tensor([1, 1])]; + tensor var_26667 = const()[name = tensor("op_26667"), val = tensor([1, 1])]; + tensor k_241_pad_type_0 = const()[name = tensor("k_241_pad_type_0"), val = tensor("custom")]; + tensor k_241_pad_0 = const()[name = tensor("k_241_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_6_attn1_to_k_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_6_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2146468928)))]; + tensor k_241_cast_fp16 = conv(dilations = var_26667, groups = var_21077, pad = k_241_pad_0, pad_type = k_241_pad_type_0, strides = var_26665, weight = mid_block_attentions_0_transformer_blocks_6_attn1_to_k_weight_to_fp16, x = hidden_states_245_cast_fp16)[name = tensor("k_241_cast_fp16")]; + tensor var_26671 = const()[name = tensor("op_26671"), val = tensor([1, 1])]; + tensor var_26673 = const()[name = tensor("op_26673"), val = tensor([1, 1])]; + tensor v_121_pad_type_0 = const()[name = tensor("v_121_pad_type_0"), val = tensor("custom")]; + tensor v_121_pad_0 = const()[name = tensor("v_121_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_6_attn1_to_v_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_6_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2149745792)))]; + tensor v_121_cast_fp16 = conv(dilations = var_26673, groups = var_21077, pad = v_121_pad_0, pad_type = v_121_pad_type_0, strides = var_26671, weight = mid_block_attentions_0_transformer_blocks_6_attn1_to_v_weight_to_fp16, x = hidden_states_245_cast_fp16)[name = tensor("v_121_cast_fp16")]; + tensor var_26677_begin_0 = const()[name = tensor("op_26677_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26677_end_0 = const()[name = tensor("op_26677_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_26677_end_mask_0 = const()[name = tensor("op_26677_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26677_cast_fp16 = slice_by_index(begin = var_26677_begin_0, end = var_26677_end_0, end_mask = var_26677_end_mask_0, x = q_121_cast_fp16)[name = tensor("op_26677_cast_fp16")]; + tensor var_26681_begin_0 = const()[name = tensor("op_26681_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_26681_end_0 = const()[name = tensor("op_26681_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_26681_end_mask_0 = const()[name = tensor("op_26681_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26681_cast_fp16 = slice_by_index(begin = var_26681_begin_0, end = var_26681_end_0, end_mask = var_26681_end_mask_0, x = q_121_cast_fp16)[name = tensor("op_26681_cast_fp16")]; + tensor var_26685_begin_0 = const()[name = tensor("op_26685_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_26685_end_0 = const()[name = tensor("op_26685_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_26685_end_mask_0 = const()[name = tensor("op_26685_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26685_cast_fp16 = slice_by_index(begin = var_26685_begin_0, end = var_26685_end_0, end_mask = var_26685_end_mask_0, x = q_121_cast_fp16)[name = tensor("op_26685_cast_fp16")]; + tensor var_26689_begin_0 = const()[name = tensor("op_26689_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_26689_end_0 = const()[name = tensor("op_26689_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_26689_end_mask_0 = const()[name = tensor("op_26689_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26689_cast_fp16 = slice_by_index(begin = var_26689_begin_0, end = var_26689_end_0, end_mask = var_26689_end_mask_0, x = q_121_cast_fp16)[name = tensor("op_26689_cast_fp16")]; + tensor var_26693_begin_0 = const()[name = tensor("op_26693_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_26693_end_0 = const()[name = tensor("op_26693_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_26693_end_mask_0 = const()[name = tensor("op_26693_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26693_cast_fp16 = slice_by_index(begin = var_26693_begin_0, end = var_26693_end_0, end_mask = var_26693_end_mask_0, x = q_121_cast_fp16)[name = tensor("op_26693_cast_fp16")]; + tensor var_26697_begin_0 = const()[name = tensor("op_26697_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_26697_end_0 = const()[name = tensor("op_26697_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_26697_end_mask_0 = const()[name = tensor("op_26697_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26697_cast_fp16 = slice_by_index(begin = var_26697_begin_0, end = var_26697_end_0, end_mask = var_26697_end_mask_0, x = q_121_cast_fp16)[name = tensor("op_26697_cast_fp16")]; + tensor var_26701_begin_0 = const()[name = tensor("op_26701_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_26701_end_0 = const()[name = tensor("op_26701_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_26701_end_mask_0 = const()[name = tensor("op_26701_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26701_cast_fp16 = slice_by_index(begin = var_26701_begin_0, end = var_26701_end_0, end_mask = var_26701_end_mask_0, x = q_121_cast_fp16)[name = tensor("op_26701_cast_fp16")]; + tensor var_26705_begin_0 = const()[name = tensor("op_26705_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_26705_end_0 = const()[name = tensor("op_26705_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_26705_end_mask_0 = const()[name = tensor("op_26705_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26705_cast_fp16 = slice_by_index(begin = var_26705_begin_0, end = var_26705_end_0, end_mask = var_26705_end_mask_0, x = q_121_cast_fp16)[name = tensor("op_26705_cast_fp16")]; + tensor var_26709_begin_0 = const()[name = tensor("op_26709_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_26709_end_0 = const()[name = tensor("op_26709_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_26709_end_mask_0 = const()[name = tensor("op_26709_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26709_cast_fp16 = slice_by_index(begin = var_26709_begin_0, end = var_26709_end_0, end_mask = var_26709_end_mask_0, x = q_121_cast_fp16)[name = tensor("op_26709_cast_fp16")]; + tensor var_26713_begin_0 = const()[name = tensor("op_26713_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_26713_end_0 = const()[name = tensor("op_26713_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_26713_end_mask_0 = const()[name = tensor("op_26713_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26713_cast_fp16 = slice_by_index(begin = var_26713_begin_0, end = var_26713_end_0, end_mask = var_26713_end_mask_0, x = q_121_cast_fp16)[name = tensor("op_26713_cast_fp16")]; + tensor var_26717_begin_0 = const()[name = tensor("op_26717_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_26717_end_0 = const()[name = tensor("op_26717_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_26717_end_mask_0 = const()[name = tensor("op_26717_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26717_cast_fp16 = slice_by_index(begin = var_26717_begin_0, end = var_26717_end_0, end_mask = var_26717_end_mask_0, x = q_121_cast_fp16)[name = tensor("op_26717_cast_fp16")]; + tensor var_26721_begin_0 = const()[name = tensor("op_26721_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_26721_end_0 = const()[name = tensor("op_26721_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_26721_end_mask_0 = const()[name = tensor("op_26721_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26721_cast_fp16 = slice_by_index(begin = var_26721_begin_0, end = var_26721_end_0, end_mask = var_26721_end_mask_0, x = q_121_cast_fp16)[name = tensor("op_26721_cast_fp16")]; + tensor var_26725_begin_0 = const()[name = tensor("op_26725_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_26725_end_0 = const()[name = tensor("op_26725_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_26725_end_mask_0 = const()[name = tensor("op_26725_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26725_cast_fp16 = slice_by_index(begin = var_26725_begin_0, end = var_26725_end_0, end_mask = var_26725_end_mask_0, x = q_121_cast_fp16)[name = tensor("op_26725_cast_fp16")]; + tensor var_26729_begin_0 = const()[name = tensor("op_26729_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_26729_end_0 = const()[name = tensor("op_26729_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_26729_end_mask_0 = const()[name = tensor("op_26729_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26729_cast_fp16 = slice_by_index(begin = var_26729_begin_0, end = var_26729_end_0, end_mask = var_26729_end_mask_0, x = q_121_cast_fp16)[name = tensor("op_26729_cast_fp16")]; + tensor var_26733_begin_0 = const()[name = tensor("op_26733_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_26733_end_0 = const()[name = tensor("op_26733_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_26733_end_mask_0 = const()[name = tensor("op_26733_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26733_cast_fp16 = slice_by_index(begin = var_26733_begin_0, end = var_26733_end_0, end_mask = var_26733_end_mask_0, x = q_121_cast_fp16)[name = tensor("op_26733_cast_fp16")]; + tensor var_26737_begin_0 = const()[name = tensor("op_26737_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_26737_end_0 = const()[name = tensor("op_26737_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_26737_end_mask_0 = const()[name = tensor("op_26737_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26737_cast_fp16 = slice_by_index(begin = var_26737_begin_0, end = var_26737_end_0, end_mask = var_26737_end_mask_0, x = q_121_cast_fp16)[name = tensor("op_26737_cast_fp16")]; + tensor var_26741_begin_0 = const()[name = tensor("op_26741_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_26741_end_0 = const()[name = tensor("op_26741_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_26741_end_mask_0 = const()[name = tensor("op_26741_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26741_cast_fp16 = slice_by_index(begin = var_26741_begin_0, end = var_26741_end_0, end_mask = var_26741_end_mask_0, x = q_121_cast_fp16)[name = tensor("op_26741_cast_fp16")]; + tensor var_26745_begin_0 = const()[name = tensor("op_26745_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_26745_end_0 = const()[name = tensor("op_26745_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_26745_end_mask_0 = const()[name = tensor("op_26745_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26745_cast_fp16 = slice_by_index(begin = var_26745_begin_0, end = var_26745_end_0, end_mask = var_26745_end_mask_0, x = q_121_cast_fp16)[name = tensor("op_26745_cast_fp16")]; + tensor var_26749_begin_0 = const()[name = tensor("op_26749_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_26749_end_0 = const()[name = tensor("op_26749_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_26749_end_mask_0 = const()[name = tensor("op_26749_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26749_cast_fp16 = slice_by_index(begin = var_26749_begin_0, end = var_26749_end_0, end_mask = var_26749_end_mask_0, x = q_121_cast_fp16)[name = tensor("op_26749_cast_fp16")]; + tensor var_26753_begin_0 = const()[name = tensor("op_26753_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_26753_end_0 = const()[name = tensor("op_26753_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_26753_end_mask_0 = const()[name = tensor("op_26753_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26753_cast_fp16 = slice_by_index(begin = var_26753_begin_0, end = var_26753_end_0, end_mask = var_26753_end_mask_0, x = q_121_cast_fp16)[name = tensor("op_26753_cast_fp16")]; + tensor k_243_perm_0 = const()[name = tensor("k_243_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_26760_begin_0 = const()[name = tensor("op_26760_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26760_end_0 = const()[name = tensor("op_26760_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_26760_end_mask_0 = const()[name = tensor("op_26760_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_243_cast_fp16 = transpose(perm = k_243_perm_0, x = k_241_cast_fp16)[name = tensor("transpose_79")]; + tensor var_26760_cast_fp16 = slice_by_index(begin = var_26760_begin_0, end = var_26760_end_0, end_mask = var_26760_end_mask_0, x = k_243_cast_fp16)[name = tensor("op_26760_cast_fp16")]; + tensor var_26764_begin_0 = const()[name = tensor("op_26764_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_26764_end_0 = const()[name = tensor("op_26764_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_26764_end_mask_0 = const()[name = tensor("op_26764_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26764_cast_fp16 = slice_by_index(begin = var_26764_begin_0, end = var_26764_end_0, end_mask = var_26764_end_mask_0, x = k_243_cast_fp16)[name = tensor("op_26764_cast_fp16")]; + tensor var_26768_begin_0 = const()[name = tensor("op_26768_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_26768_end_0 = const()[name = tensor("op_26768_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_26768_end_mask_0 = const()[name = tensor("op_26768_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26768_cast_fp16 = slice_by_index(begin = var_26768_begin_0, end = var_26768_end_0, end_mask = var_26768_end_mask_0, x = k_243_cast_fp16)[name = tensor("op_26768_cast_fp16")]; + tensor var_26772_begin_0 = const()[name = tensor("op_26772_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_26772_end_0 = const()[name = tensor("op_26772_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_26772_end_mask_0 = const()[name = tensor("op_26772_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26772_cast_fp16 = slice_by_index(begin = var_26772_begin_0, end = var_26772_end_0, end_mask = var_26772_end_mask_0, x = k_243_cast_fp16)[name = tensor("op_26772_cast_fp16")]; + tensor var_26776_begin_0 = const()[name = tensor("op_26776_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_26776_end_0 = const()[name = tensor("op_26776_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_26776_end_mask_0 = const()[name = tensor("op_26776_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26776_cast_fp16 = slice_by_index(begin = var_26776_begin_0, end = var_26776_end_0, end_mask = var_26776_end_mask_0, x = k_243_cast_fp16)[name = tensor("op_26776_cast_fp16")]; + tensor var_26780_begin_0 = const()[name = tensor("op_26780_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_26780_end_0 = const()[name = tensor("op_26780_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_26780_end_mask_0 = const()[name = tensor("op_26780_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26780_cast_fp16 = slice_by_index(begin = var_26780_begin_0, end = var_26780_end_0, end_mask = var_26780_end_mask_0, x = k_243_cast_fp16)[name = tensor("op_26780_cast_fp16")]; + tensor var_26784_begin_0 = const()[name = tensor("op_26784_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_26784_end_0 = const()[name = tensor("op_26784_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_26784_end_mask_0 = const()[name = tensor("op_26784_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26784_cast_fp16 = slice_by_index(begin = var_26784_begin_0, end = var_26784_end_0, end_mask = var_26784_end_mask_0, x = k_243_cast_fp16)[name = tensor("op_26784_cast_fp16")]; + tensor var_26788_begin_0 = const()[name = tensor("op_26788_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_26788_end_0 = const()[name = tensor("op_26788_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_26788_end_mask_0 = const()[name = tensor("op_26788_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26788_cast_fp16 = slice_by_index(begin = var_26788_begin_0, end = var_26788_end_0, end_mask = var_26788_end_mask_0, x = k_243_cast_fp16)[name = tensor("op_26788_cast_fp16")]; + tensor var_26792_begin_0 = const()[name = tensor("op_26792_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_26792_end_0 = const()[name = tensor("op_26792_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_26792_end_mask_0 = const()[name = tensor("op_26792_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26792_cast_fp16 = slice_by_index(begin = var_26792_begin_0, end = var_26792_end_0, end_mask = var_26792_end_mask_0, x = k_243_cast_fp16)[name = tensor("op_26792_cast_fp16")]; + tensor var_26796_begin_0 = const()[name = tensor("op_26796_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_26796_end_0 = const()[name = tensor("op_26796_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_26796_end_mask_0 = const()[name = tensor("op_26796_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26796_cast_fp16 = slice_by_index(begin = var_26796_begin_0, end = var_26796_end_0, end_mask = var_26796_end_mask_0, x = k_243_cast_fp16)[name = tensor("op_26796_cast_fp16")]; + tensor var_26800_begin_0 = const()[name = tensor("op_26800_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_26800_end_0 = const()[name = tensor("op_26800_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_26800_end_mask_0 = const()[name = tensor("op_26800_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26800_cast_fp16 = slice_by_index(begin = var_26800_begin_0, end = var_26800_end_0, end_mask = var_26800_end_mask_0, x = k_243_cast_fp16)[name = tensor("op_26800_cast_fp16")]; + tensor var_26804_begin_0 = const()[name = tensor("op_26804_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_26804_end_0 = const()[name = tensor("op_26804_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_26804_end_mask_0 = const()[name = tensor("op_26804_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26804_cast_fp16 = slice_by_index(begin = var_26804_begin_0, end = var_26804_end_0, end_mask = var_26804_end_mask_0, x = k_243_cast_fp16)[name = tensor("op_26804_cast_fp16")]; + tensor var_26808_begin_0 = const()[name = tensor("op_26808_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_26808_end_0 = const()[name = tensor("op_26808_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_26808_end_mask_0 = const()[name = tensor("op_26808_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26808_cast_fp16 = slice_by_index(begin = var_26808_begin_0, end = var_26808_end_0, end_mask = var_26808_end_mask_0, x = k_243_cast_fp16)[name = tensor("op_26808_cast_fp16")]; + tensor var_26812_begin_0 = const()[name = tensor("op_26812_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_26812_end_0 = const()[name = tensor("op_26812_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_26812_end_mask_0 = const()[name = tensor("op_26812_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26812_cast_fp16 = slice_by_index(begin = var_26812_begin_0, end = var_26812_end_0, end_mask = var_26812_end_mask_0, x = k_243_cast_fp16)[name = tensor("op_26812_cast_fp16")]; + tensor var_26816_begin_0 = const()[name = tensor("op_26816_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_26816_end_0 = const()[name = tensor("op_26816_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_26816_end_mask_0 = const()[name = tensor("op_26816_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26816_cast_fp16 = slice_by_index(begin = var_26816_begin_0, end = var_26816_end_0, end_mask = var_26816_end_mask_0, x = k_243_cast_fp16)[name = tensor("op_26816_cast_fp16")]; + tensor var_26820_begin_0 = const()[name = tensor("op_26820_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_26820_end_0 = const()[name = tensor("op_26820_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_26820_end_mask_0 = const()[name = tensor("op_26820_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26820_cast_fp16 = slice_by_index(begin = var_26820_begin_0, end = var_26820_end_0, end_mask = var_26820_end_mask_0, x = k_243_cast_fp16)[name = tensor("op_26820_cast_fp16")]; + tensor var_26824_begin_0 = const()[name = tensor("op_26824_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_26824_end_0 = const()[name = tensor("op_26824_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_26824_end_mask_0 = const()[name = tensor("op_26824_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26824_cast_fp16 = slice_by_index(begin = var_26824_begin_0, end = var_26824_end_0, end_mask = var_26824_end_mask_0, x = k_243_cast_fp16)[name = tensor("op_26824_cast_fp16")]; + tensor var_26828_begin_0 = const()[name = tensor("op_26828_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_26828_end_0 = const()[name = tensor("op_26828_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_26828_end_mask_0 = const()[name = tensor("op_26828_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26828_cast_fp16 = slice_by_index(begin = var_26828_begin_0, end = var_26828_end_0, end_mask = var_26828_end_mask_0, x = k_243_cast_fp16)[name = tensor("op_26828_cast_fp16")]; + tensor var_26832_begin_0 = const()[name = tensor("op_26832_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_26832_end_0 = const()[name = tensor("op_26832_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_26832_end_mask_0 = const()[name = tensor("op_26832_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26832_cast_fp16 = slice_by_index(begin = var_26832_begin_0, end = var_26832_end_0, end_mask = var_26832_end_mask_0, x = k_243_cast_fp16)[name = tensor("op_26832_cast_fp16")]; + tensor var_26836_begin_0 = const()[name = tensor("op_26836_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_26836_end_0 = const()[name = tensor("op_26836_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_26836_end_mask_0 = const()[name = tensor("op_26836_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26836_cast_fp16 = slice_by_index(begin = var_26836_begin_0, end = var_26836_end_0, end_mask = var_26836_end_mask_0, x = k_243_cast_fp16)[name = tensor("op_26836_cast_fp16")]; + tensor var_26838_begin_0 = const()[name = tensor("op_26838_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26838_end_0 = const()[name = tensor("op_26838_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_26838_end_mask_0 = const()[name = tensor("op_26838_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26838_cast_fp16 = slice_by_index(begin = var_26838_begin_0, end = var_26838_end_0, end_mask = var_26838_end_mask_0, x = v_121_cast_fp16)[name = tensor("op_26838_cast_fp16")]; + tensor var_26842_begin_0 = const()[name = tensor("op_26842_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_26842_end_0 = const()[name = tensor("op_26842_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_26842_end_mask_0 = const()[name = tensor("op_26842_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26842_cast_fp16 = slice_by_index(begin = var_26842_begin_0, end = var_26842_end_0, end_mask = var_26842_end_mask_0, x = v_121_cast_fp16)[name = tensor("op_26842_cast_fp16")]; + tensor var_26846_begin_0 = const()[name = tensor("op_26846_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_26846_end_0 = const()[name = tensor("op_26846_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_26846_end_mask_0 = const()[name = tensor("op_26846_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26846_cast_fp16 = slice_by_index(begin = var_26846_begin_0, end = var_26846_end_0, end_mask = var_26846_end_mask_0, x = v_121_cast_fp16)[name = tensor("op_26846_cast_fp16")]; + tensor var_26850_begin_0 = const()[name = tensor("op_26850_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_26850_end_0 = const()[name = tensor("op_26850_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_26850_end_mask_0 = const()[name = tensor("op_26850_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26850_cast_fp16 = slice_by_index(begin = var_26850_begin_0, end = var_26850_end_0, end_mask = var_26850_end_mask_0, x = v_121_cast_fp16)[name = tensor("op_26850_cast_fp16")]; + tensor var_26854_begin_0 = const()[name = tensor("op_26854_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_26854_end_0 = const()[name = tensor("op_26854_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_26854_end_mask_0 = const()[name = tensor("op_26854_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26854_cast_fp16 = slice_by_index(begin = var_26854_begin_0, end = var_26854_end_0, end_mask = var_26854_end_mask_0, x = v_121_cast_fp16)[name = tensor("op_26854_cast_fp16")]; + tensor var_26858_begin_0 = const()[name = tensor("op_26858_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_26858_end_0 = const()[name = tensor("op_26858_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_26858_end_mask_0 = const()[name = tensor("op_26858_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26858_cast_fp16 = slice_by_index(begin = var_26858_begin_0, end = var_26858_end_0, end_mask = var_26858_end_mask_0, x = v_121_cast_fp16)[name = tensor("op_26858_cast_fp16")]; + tensor var_26862_begin_0 = const()[name = tensor("op_26862_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_26862_end_0 = const()[name = tensor("op_26862_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_26862_end_mask_0 = const()[name = tensor("op_26862_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26862_cast_fp16 = slice_by_index(begin = var_26862_begin_0, end = var_26862_end_0, end_mask = var_26862_end_mask_0, x = v_121_cast_fp16)[name = tensor("op_26862_cast_fp16")]; + tensor var_26866_begin_0 = const()[name = tensor("op_26866_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_26866_end_0 = const()[name = tensor("op_26866_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_26866_end_mask_0 = const()[name = tensor("op_26866_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26866_cast_fp16 = slice_by_index(begin = var_26866_begin_0, end = var_26866_end_0, end_mask = var_26866_end_mask_0, x = v_121_cast_fp16)[name = tensor("op_26866_cast_fp16")]; + tensor var_26870_begin_0 = const()[name = tensor("op_26870_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_26870_end_0 = const()[name = tensor("op_26870_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_26870_end_mask_0 = const()[name = tensor("op_26870_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26870_cast_fp16 = slice_by_index(begin = var_26870_begin_0, end = var_26870_end_0, end_mask = var_26870_end_mask_0, x = v_121_cast_fp16)[name = tensor("op_26870_cast_fp16")]; + tensor var_26874_begin_0 = const()[name = tensor("op_26874_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_26874_end_0 = const()[name = tensor("op_26874_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_26874_end_mask_0 = const()[name = tensor("op_26874_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26874_cast_fp16 = slice_by_index(begin = var_26874_begin_0, end = var_26874_end_0, end_mask = var_26874_end_mask_0, x = v_121_cast_fp16)[name = tensor("op_26874_cast_fp16")]; + tensor var_26878_begin_0 = const()[name = tensor("op_26878_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_26878_end_0 = const()[name = tensor("op_26878_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_26878_end_mask_0 = const()[name = tensor("op_26878_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26878_cast_fp16 = slice_by_index(begin = var_26878_begin_0, end = var_26878_end_0, end_mask = var_26878_end_mask_0, x = v_121_cast_fp16)[name = tensor("op_26878_cast_fp16")]; + tensor var_26882_begin_0 = const()[name = tensor("op_26882_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_26882_end_0 = const()[name = tensor("op_26882_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_26882_end_mask_0 = const()[name = tensor("op_26882_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26882_cast_fp16 = slice_by_index(begin = var_26882_begin_0, end = var_26882_end_0, end_mask = var_26882_end_mask_0, x = v_121_cast_fp16)[name = tensor("op_26882_cast_fp16")]; + tensor var_26886_begin_0 = const()[name = tensor("op_26886_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_26886_end_0 = const()[name = tensor("op_26886_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_26886_end_mask_0 = const()[name = tensor("op_26886_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26886_cast_fp16 = slice_by_index(begin = var_26886_begin_0, end = var_26886_end_0, end_mask = var_26886_end_mask_0, x = v_121_cast_fp16)[name = tensor("op_26886_cast_fp16")]; + tensor var_26890_begin_0 = const()[name = tensor("op_26890_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_26890_end_0 = const()[name = tensor("op_26890_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_26890_end_mask_0 = const()[name = tensor("op_26890_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26890_cast_fp16 = slice_by_index(begin = var_26890_begin_0, end = var_26890_end_0, end_mask = var_26890_end_mask_0, x = v_121_cast_fp16)[name = tensor("op_26890_cast_fp16")]; + tensor var_26894_begin_0 = const()[name = tensor("op_26894_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_26894_end_0 = const()[name = tensor("op_26894_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_26894_end_mask_0 = const()[name = tensor("op_26894_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26894_cast_fp16 = slice_by_index(begin = var_26894_begin_0, end = var_26894_end_0, end_mask = var_26894_end_mask_0, x = v_121_cast_fp16)[name = tensor("op_26894_cast_fp16")]; + tensor var_26898_begin_0 = const()[name = tensor("op_26898_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_26898_end_0 = const()[name = tensor("op_26898_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_26898_end_mask_0 = const()[name = tensor("op_26898_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26898_cast_fp16 = slice_by_index(begin = var_26898_begin_0, end = var_26898_end_0, end_mask = var_26898_end_mask_0, x = v_121_cast_fp16)[name = tensor("op_26898_cast_fp16")]; + tensor var_26902_begin_0 = const()[name = tensor("op_26902_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_26902_end_0 = const()[name = tensor("op_26902_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_26902_end_mask_0 = const()[name = tensor("op_26902_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26902_cast_fp16 = slice_by_index(begin = var_26902_begin_0, end = var_26902_end_0, end_mask = var_26902_end_mask_0, x = v_121_cast_fp16)[name = tensor("op_26902_cast_fp16")]; + tensor var_26906_begin_0 = const()[name = tensor("op_26906_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_26906_end_0 = const()[name = tensor("op_26906_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_26906_end_mask_0 = const()[name = tensor("op_26906_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26906_cast_fp16 = slice_by_index(begin = var_26906_begin_0, end = var_26906_end_0, end_mask = var_26906_end_mask_0, x = v_121_cast_fp16)[name = tensor("op_26906_cast_fp16")]; + tensor var_26910_begin_0 = const()[name = tensor("op_26910_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_26910_end_0 = const()[name = tensor("op_26910_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_26910_end_mask_0 = const()[name = tensor("op_26910_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26910_cast_fp16 = slice_by_index(begin = var_26910_begin_0, end = var_26910_end_0, end_mask = var_26910_end_mask_0, x = v_121_cast_fp16)[name = tensor("op_26910_cast_fp16")]; + tensor var_26914_begin_0 = const()[name = tensor("op_26914_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_26914_end_0 = const()[name = tensor("op_26914_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_26914_end_mask_0 = const()[name = tensor("op_26914_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26914_cast_fp16 = slice_by_index(begin = var_26914_begin_0, end = var_26914_end_0, end_mask = var_26914_end_mask_0, x = v_121_cast_fp16)[name = tensor("op_26914_cast_fp16")]; + tensor var_26918_equation_0 = const()[name = tensor("op_26918_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26918_cast_fp16 = einsum(equation = var_26918_equation_0, values = (var_26760_cast_fp16, var_26677_cast_fp16))[name = tensor("op_26918_cast_fp16")]; + tensor var_26919_to_fp16 = const()[name = tensor("op_26919_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2241_cast_fp16 = mul(x = var_26918_cast_fp16, y = var_26919_to_fp16)[name = tensor("aw_2241_cast_fp16")]; + tensor var_26922_equation_0 = const()[name = tensor("op_26922_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26922_cast_fp16 = einsum(equation = var_26922_equation_0, values = (var_26764_cast_fp16, var_26681_cast_fp16))[name = tensor("op_26922_cast_fp16")]; + tensor var_26923_to_fp16 = const()[name = tensor("op_26923_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2243_cast_fp16 = mul(x = var_26922_cast_fp16, y = var_26923_to_fp16)[name = tensor("aw_2243_cast_fp16")]; + tensor var_26926_equation_0 = const()[name = tensor("op_26926_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26926_cast_fp16 = einsum(equation = var_26926_equation_0, values = (var_26768_cast_fp16, var_26685_cast_fp16))[name = tensor("op_26926_cast_fp16")]; + tensor var_26927_to_fp16 = const()[name = tensor("op_26927_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2245_cast_fp16 = mul(x = var_26926_cast_fp16, y = var_26927_to_fp16)[name = tensor("aw_2245_cast_fp16")]; + tensor var_26930_equation_0 = const()[name = tensor("op_26930_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26930_cast_fp16 = einsum(equation = var_26930_equation_0, values = (var_26772_cast_fp16, var_26689_cast_fp16))[name = tensor("op_26930_cast_fp16")]; + tensor var_26931_to_fp16 = const()[name = tensor("op_26931_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2247_cast_fp16 = mul(x = var_26930_cast_fp16, y = var_26931_to_fp16)[name = tensor("aw_2247_cast_fp16")]; + tensor var_26934_equation_0 = const()[name = tensor("op_26934_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26934_cast_fp16 = einsum(equation = var_26934_equation_0, values = (var_26776_cast_fp16, var_26693_cast_fp16))[name = tensor("op_26934_cast_fp16")]; + tensor var_26935_to_fp16 = const()[name = tensor("op_26935_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2249_cast_fp16 = mul(x = var_26934_cast_fp16, y = var_26935_to_fp16)[name = tensor("aw_2249_cast_fp16")]; + tensor var_26938_equation_0 = const()[name = tensor("op_26938_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26938_cast_fp16 = einsum(equation = var_26938_equation_0, values = (var_26780_cast_fp16, var_26697_cast_fp16))[name = tensor("op_26938_cast_fp16")]; + tensor var_26939_to_fp16 = const()[name = tensor("op_26939_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2251_cast_fp16 = mul(x = var_26938_cast_fp16, y = var_26939_to_fp16)[name = tensor("aw_2251_cast_fp16")]; + tensor var_26942_equation_0 = const()[name = tensor("op_26942_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26942_cast_fp16 = einsum(equation = var_26942_equation_0, values = (var_26784_cast_fp16, var_26701_cast_fp16))[name = tensor("op_26942_cast_fp16")]; + tensor var_26943_to_fp16 = const()[name = tensor("op_26943_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2253_cast_fp16 = mul(x = var_26942_cast_fp16, y = var_26943_to_fp16)[name = tensor("aw_2253_cast_fp16")]; + tensor var_26946_equation_0 = const()[name = tensor("op_26946_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26946_cast_fp16 = einsum(equation = var_26946_equation_0, values = (var_26788_cast_fp16, var_26705_cast_fp16))[name = tensor("op_26946_cast_fp16")]; + tensor var_26947_to_fp16 = const()[name = tensor("op_26947_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2255_cast_fp16 = mul(x = var_26946_cast_fp16, y = var_26947_to_fp16)[name = tensor("aw_2255_cast_fp16")]; + tensor var_26950_equation_0 = const()[name = tensor("op_26950_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26950_cast_fp16 = einsum(equation = var_26950_equation_0, values = (var_26792_cast_fp16, var_26709_cast_fp16))[name = tensor("op_26950_cast_fp16")]; + tensor var_26951_to_fp16 = const()[name = tensor("op_26951_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2257_cast_fp16 = mul(x = var_26950_cast_fp16, y = var_26951_to_fp16)[name = tensor("aw_2257_cast_fp16")]; + tensor var_26954_equation_0 = const()[name = tensor("op_26954_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26954_cast_fp16 = einsum(equation = var_26954_equation_0, values = (var_26796_cast_fp16, var_26713_cast_fp16))[name = tensor("op_26954_cast_fp16")]; + tensor var_26955_to_fp16 = const()[name = tensor("op_26955_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2259_cast_fp16 = mul(x = var_26954_cast_fp16, y = var_26955_to_fp16)[name = tensor("aw_2259_cast_fp16")]; + tensor var_26958_equation_0 = const()[name = tensor("op_26958_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26958_cast_fp16 = einsum(equation = var_26958_equation_0, values = (var_26800_cast_fp16, var_26717_cast_fp16))[name = tensor("op_26958_cast_fp16")]; + tensor var_26959_to_fp16 = const()[name = tensor("op_26959_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2261_cast_fp16 = mul(x = var_26958_cast_fp16, y = var_26959_to_fp16)[name = tensor("aw_2261_cast_fp16")]; + tensor var_26962_equation_0 = const()[name = tensor("op_26962_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26962_cast_fp16 = einsum(equation = var_26962_equation_0, values = (var_26804_cast_fp16, var_26721_cast_fp16))[name = tensor("op_26962_cast_fp16")]; + tensor var_26963_to_fp16 = const()[name = tensor("op_26963_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2263_cast_fp16 = mul(x = var_26962_cast_fp16, y = var_26963_to_fp16)[name = tensor("aw_2263_cast_fp16")]; + tensor var_26966_equation_0 = const()[name = tensor("op_26966_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26966_cast_fp16 = einsum(equation = var_26966_equation_0, values = (var_26808_cast_fp16, var_26725_cast_fp16))[name = tensor("op_26966_cast_fp16")]; + tensor var_26967_to_fp16 = const()[name = tensor("op_26967_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2265_cast_fp16 = mul(x = var_26966_cast_fp16, y = var_26967_to_fp16)[name = tensor("aw_2265_cast_fp16")]; + tensor var_26970_equation_0 = const()[name = tensor("op_26970_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26970_cast_fp16 = einsum(equation = var_26970_equation_0, values = (var_26812_cast_fp16, var_26729_cast_fp16))[name = tensor("op_26970_cast_fp16")]; + tensor var_26971_to_fp16 = const()[name = tensor("op_26971_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2267_cast_fp16 = mul(x = var_26970_cast_fp16, y = var_26971_to_fp16)[name = tensor("aw_2267_cast_fp16")]; + tensor var_26974_equation_0 = const()[name = tensor("op_26974_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26974_cast_fp16 = einsum(equation = var_26974_equation_0, values = (var_26816_cast_fp16, var_26733_cast_fp16))[name = tensor("op_26974_cast_fp16")]; + tensor var_26975_to_fp16 = const()[name = tensor("op_26975_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2269_cast_fp16 = mul(x = var_26974_cast_fp16, y = var_26975_to_fp16)[name = tensor("aw_2269_cast_fp16")]; + tensor var_26978_equation_0 = const()[name = tensor("op_26978_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26978_cast_fp16 = einsum(equation = var_26978_equation_0, values = (var_26820_cast_fp16, var_26737_cast_fp16))[name = tensor("op_26978_cast_fp16")]; + tensor var_26979_to_fp16 = const()[name = tensor("op_26979_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2271_cast_fp16 = mul(x = var_26978_cast_fp16, y = var_26979_to_fp16)[name = tensor("aw_2271_cast_fp16")]; + tensor var_26982_equation_0 = const()[name = tensor("op_26982_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26982_cast_fp16 = einsum(equation = var_26982_equation_0, values = (var_26824_cast_fp16, var_26741_cast_fp16))[name = tensor("op_26982_cast_fp16")]; + tensor var_26983_to_fp16 = const()[name = tensor("op_26983_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2273_cast_fp16 = mul(x = var_26982_cast_fp16, y = var_26983_to_fp16)[name = tensor("aw_2273_cast_fp16")]; + tensor var_26986_equation_0 = const()[name = tensor("op_26986_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26986_cast_fp16 = einsum(equation = var_26986_equation_0, values = (var_26828_cast_fp16, var_26745_cast_fp16))[name = tensor("op_26986_cast_fp16")]; + tensor var_26987_to_fp16 = const()[name = tensor("op_26987_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2275_cast_fp16 = mul(x = var_26986_cast_fp16, y = var_26987_to_fp16)[name = tensor("aw_2275_cast_fp16")]; + tensor var_26990_equation_0 = const()[name = tensor("op_26990_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26990_cast_fp16 = einsum(equation = var_26990_equation_0, values = (var_26832_cast_fp16, var_26749_cast_fp16))[name = tensor("op_26990_cast_fp16")]; + tensor var_26991_to_fp16 = const()[name = tensor("op_26991_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2277_cast_fp16 = mul(x = var_26990_cast_fp16, y = var_26991_to_fp16)[name = tensor("aw_2277_cast_fp16")]; + tensor var_26994_equation_0 = const()[name = tensor("op_26994_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_26994_cast_fp16 = einsum(equation = var_26994_equation_0, values = (var_26836_cast_fp16, var_26753_cast_fp16))[name = tensor("op_26994_cast_fp16")]; + tensor var_26995_to_fp16 = const()[name = tensor("op_26995_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2279_cast_fp16 = mul(x = var_26994_cast_fp16, y = var_26995_to_fp16)[name = tensor("aw_2279_cast_fp16")]; + tensor var_26997_cast_fp16 = softmax(axis = var_21077, x = aw_2241_cast_fp16)[name = tensor("op_26997_cast_fp16")]; + tensor var_26998_cast_fp16 = softmax(axis = var_21077, x = aw_2243_cast_fp16)[name = tensor("op_26998_cast_fp16")]; + tensor var_26999_cast_fp16 = softmax(axis = var_21077, x = aw_2245_cast_fp16)[name = tensor("op_26999_cast_fp16")]; + tensor var_27000_cast_fp16 = softmax(axis = var_21077, x = aw_2247_cast_fp16)[name = tensor("op_27000_cast_fp16")]; + tensor var_27001_cast_fp16 = softmax(axis = var_21077, x = aw_2249_cast_fp16)[name = tensor("op_27001_cast_fp16")]; + tensor var_27002_cast_fp16 = softmax(axis = var_21077, x = aw_2251_cast_fp16)[name = tensor("op_27002_cast_fp16")]; + tensor var_27003_cast_fp16 = softmax(axis = var_21077, x = aw_2253_cast_fp16)[name = tensor("op_27003_cast_fp16")]; + tensor var_27004_cast_fp16 = softmax(axis = var_21077, x = aw_2255_cast_fp16)[name = tensor("op_27004_cast_fp16")]; + tensor var_27005_cast_fp16 = softmax(axis = var_21077, x = aw_2257_cast_fp16)[name = tensor("op_27005_cast_fp16")]; + tensor var_27006_cast_fp16 = softmax(axis = var_21077, x = aw_2259_cast_fp16)[name = tensor("op_27006_cast_fp16")]; + tensor var_27007_cast_fp16 = softmax(axis = var_21077, x = aw_2261_cast_fp16)[name = tensor("op_27007_cast_fp16")]; + tensor var_27008_cast_fp16 = softmax(axis = var_21077, x = aw_2263_cast_fp16)[name = tensor("op_27008_cast_fp16")]; + tensor var_27009_cast_fp16 = softmax(axis = var_21077, x = aw_2265_cast_fp16)[name = tensor("op_27009_cast_fp16")]; + tensor var_27010_cast_fp16 = softmax(axis = var_21077, x = aw_2267_cast_fp16)[name = tensor("op_27010_cast_fp16")]; + tensor var_27011_cast_fp16 = softmax(axis = var_21077, x = aw_2269_cast_fp16)[name = tensor("op_27011_cast_fp16")]; + tensor var_27012_cast_fp16 = softmax(axis = var_21077, x = aw_2271_cast_fp16)[name = tensor("op_27012_cast_fp16")]; + tensor var_27013_cast_fp16 = softmax(axis = var_21077, x = aw_2273_cast_fp16)[name = tensor("op_27013_cast_fp16")]; + tensor var_27014_cast_fp16 = softmax(axis = var_21077, x = aw_2275_cast_fp16)[name = tensor("op_27014_cast_fp16")]; + tensor var_27015_cast_fp16 = softmax(axis = var_21077, x = aw_2277_cast_fp16)[name = tensor("op_27015_cast_fp16")]; + tensor var_27016_cast_fp16 = softmax(axis = var_21077, x = aw_2279_cast_fp16)[name = tensor("op_27016_cast_fp16")]; + tensor var_27018_equation_0 = const()[name = tensor("op_27018_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27018_cast_fp16 = einsum(equation = var_27018_equation_0, values = (var_26838_cast_fp16, var_26997_cast_fp16))[name = tensor("op_27018_cast_fp16")]; + tensor var_27020_equation_0 = const()[name = tensor("op_27020_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27020_cast_fp16 = einsum(equation = var_27020_equation_0, values = (var_26842_cast_fp16, var_26998_cast_fp16))[name = tensor("op_27020_cast_fp16")]; + tensor var_27022_equation_0 = const()[name = tensor("op_27022_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27022_cast_fp16 = einsum(equation = var_27022_equation_0, values = (var_26846_cast_fp16, var_26999_cast_fp16))[name = tensor("op_27022_cast_fp16")]; + tensor var_27024_equation_0 = const()[name = tensor("op_27024_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27024_cast_fp16 = einsum(equation = var_27024_equation_0, values = (var_26850_cast_fp16, var_27000_cast_fp16))[name = tensor("op_27024_cast_fp16")]; + tensor var_27026_equation_0 = const()[name = tensor("op_27026_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27026_cast_fp16 = einsum(equation = var_27026_equation_0, values = (var_26854_cast_fp16, var_27001_cast_fp16))[name = tensor("op_27026_cast_fp16")]; + tensor var_27028_equation_0 = const()[name = tensor("op_27028_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27028_cast_fp16 = einsum(equation = var_27028_equation_0, values = (var_26858_cast_fp16, var_27002_cast_fp16))[name = tensor("op_27028_cast_fp16")]; + tensor var_27030_equation_0 = const()[name = tensor("op_27030_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27030_cast_fp16 = einsum(equation = var_27030_equation_0, values = (var_26862_cast_fp16, var_27003_cast_fp16))[name = tensor("op_27030_cast_fp16")]; + tensor var_27032_equation_0 = const()[name = tensor("op_27032_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27032_cast_fp16 = einsum(equation = var_27032_equation_0, values = (var_26866_cast_fp16, var_27004_cast_fp16))[name = tensor("op_27032_cast_fp16")]; + tensor var_27034_equation_0 = const()[name = tensor("op_27034_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27034_cast_fp16 = einsum(equation = var_27034_equation_0, values = (var_26870_cast_fp16, var_27005_cast_fp16))[name = tensor("op_27034_cast_fp16")]; + tensor var_27036_equation_0 = const()[name = tensor("op_27036_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27036_cast_fp16 = einsum(equation = var_27036_equation_0, values = (var_26874_cast_fp16, var_27006_cast_fp16))[name = tensor("op_27036_cast_fp16")]; + tensor var_27038_equation_0 = const()[name = tensor("op_27038_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27038_cast_fp16 = einsum(equation = var_27038_equation_0, values = (var_26878_cast_fp16, var_27007_cast_fp16))[name = tensor("op_27038_cast_fp16")]; + tensor var_27040_equation_0 = const()[name = tensor("op_27040_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27040_cast_fp16 = einsum(equation = var_27040_equation_0, values = (var_26882_cast_fp16, var_27008_cast_fp16))[name = tensor("op_27040_cast_fp16")]; + tensor var_27042_equation_0 = const()[name = tensor("op_27042_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27042_cast_fp16 = einsum(equation = var_27042_equation_0, values = (var_26886_cast_fp16, var_27009_cast_fp16))[name = tensor("op_27042_cast_fp16")]; + tensor var_27044_equation_0 = const()[name = tensor("op_27044_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27044_cast_fp16 = einsum(equation = var_27044_equation_0, values = (var_26890_cast_fp16, var_27010_cast_fp16))[name = tensor("op_27044_cast_fp16")]; + tensor var_27046_equation_0 = const()[name = tensor("op_27046_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27046_cast_fp16 = einsum(equation = var_27046_equation_0, values = (var_26894_cast_fp16, var_27011_cast_fp16))[name = tensor("op_27046_cast_fp16")]; + tensor var_27048_equation_0 = const()[name = tensor("op_27048_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27048_cast_fp16 = einsum(equation = var_27048_equation_0, values = (var_26898_cast_fp16, var_27012_cast_fp16))[name = tensor("op_27048_cast_fp16")]; + tensor var_27050_equation_0 = const()[name = tensor("op_27050_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27050_cast_fp16 = einsum(equation = var_27050_equation_0, values = (var_26902_cast_fp16, var_27013_cast_fp16))[name = tensor("op_27050_cast_fp16")]; + tensor var_27052_equation_0 = const()[name = tensor("op_27052_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27052_cast_fp16 = einsum(equation = var_27052_equation_0, values = (var_26906_cast_fp16, var_27014_cast_fp16))[name = tensor("op_27052_cast_fp16")]; + tensor var_27054_equation_0 = const()[name = tensor("op_27054_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27054_cast_fp16 = einsum(equation = var_27054_equation_0, values = (var_26910_cast_fp16, var_27015_cast_fp16))[name = tensor("op_27054_cast_fp16")]; + tensor var_27056_equation_0 = const()[name = tensor("op_27056_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27056_cast_fp16 = einsum(equation = var_27056_equation_0, values = (var_26914_cast_fp16, var_27016_cast_fp16))[name = tensor("op_27056_cast_fp16")]; + tensor input_375_interleave_0 = const()[name = tensor("input_375_interleave_0"), val = tensor(false)]; + tensor input_375_cast_fp16 = concat(axis = var_21077, interleave = input_375_interleave_0, values = (var_27018_cast_fp16, var_27020_cast_fp16, var_27022_cast_fp16, var_27024_cast_fp16, var_27026_cast_fp16, var_27028_cast_fp16, var_27030_cast_fp16, var_27032_cast_fp16, var_27034_cast_fp16, var_27036_cast_fp16, var_27038_cast_fp16, var_27040_cast_fp16, var_27042_cast_fp16, var_27044_cast_fp16, var_27046_cast_fp16, var_27048_cast_fp16, var_27050_cast_fp16, var_27052_cast_fp16, var_27054_cast_fp16, var_27056_cast_fp16))[name = tensor("input_375_cast_fp16")]; + tensor var_27062 = const()[name = tensor("op_27062"), val = tensor([1, 1])]; + tensor var_27064 = const()[name = tensor("op_27064"), val = tensor([1, 1])]; + tensor var_27066_pad_type_0 = const()[name = tensor("op_27066_pad_type_0"), val = tensor("custom")]; + tensor var_27066_pad_0 = const()[name = tensor("op_27066_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_6_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_6_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2153022656)))]; + tensor mid_block_attentions_0_transformer_blocks_6_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_6_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2156299520)))]; + tensor var_27066_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_6_attn1_to_out_0_bias_to_fp16, dilations = var_27064, groups = var_21077, pad = var_27066_pad_0, pad_type = var_27066_pad_type_0, strides = var_27062, weight = mid_block_attentions_0_transformer_blocks_6_attn1_to_out_0_weight_to_fp16, x = input_375_cast_fp16)[name = tensor("op_27066_cast_fp16")]; + tensor inputs_183_cast_fp16 = add(x = var_27066_cast_fp16, y = inputs_181_cast_fp16)[name = tensor("inputs_183_cast_fp16")]; + tensor hidden_states_247_axes_0 = const()[name = tensor("hidden_states_247_axes_0"), val = tensor([1])]; + tensor hidden_states_247_gamma_0_to_fp16 = const()[name = tensor("hidden_states_247_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2156302144)))]; + tensor hidden_states_247_beta_0_to_fp16 = const()[name = tensor("hidden_states_247_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2156304768)))]; + tensor var_27076_to_fp16 = const()[name = tensor("op_27076_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_247_cast_fp16 = layer_norm(axes = hidden_states_247_axes_0, beta = hidden_states_247_beta_0_to_fp16, epsilon = var_27076_to_fp16, gamma = hidden_states_247_gamma_0_to_fp16, x = inputs_183_cast_fp16)[name = tensor("hidden_states_247_cast_fp16")]; + tensor var_27091 = const()[name = tensor("op_27091"), val = tensor([1, 1])]; + tensor var_27093 = const()[name = tensor("op_27093"), val = tensor([1, 1])]; + tensor q_123_pad_type_0 = const()[name = tensor("q_123_pad_type_0"), val = tensor("custom")]; + tensor q_123_pad_0 = const()[name = tensor("q_123_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_6_attn2_to_q_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_6_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2156307392)))]; + tensor q_123_cast_fp16 = conv(dilations = var_27093, groups = var_21077, pad = q_123_pad_0, pad_type = q_123_pad_type_0, strides = var_27091, weight = mid_block_attentions_0_transformer_blocks_6_attn2_to_q_weight_to_fp16, x = hidden_states_247_cast_fp16)[name = tensor("q_123_cast_fp16")]; + tensor var_27097 = const()[name = tensor("op_27097"), val = tensor([1, 1])]; + tensor var_27099 = const()[name = tensor("op_27099"), val = tensor([1, 1])]; + tensor k_245_pad_type_0 = const()[name = tensor("k_245_pad_type_0"), val = tensor("custom")]; + tensor k_245_pad_0 = const()[name = tensor("k_245_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_6_attn2_to_k_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_6_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2159584256)))]; + tensor k_245_cast_fp16 = conv(dilations = var_27099, groups = var_21077, pad = k_245_pad_0, pad_type = k_245_pad_type_0, strides = var_27097, weight = mid_block_attentions_0_transformer_blocks_6_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_245_cast_fp16")]; + tensor var_27103 = const()[name = tensor("op_27103"), val = tensor([1, 1])]; + tensor var_27105 = const()[name = tensor("op_27105"), val = tensor([1, 1])]; + tensor v_123_pad_type_0 = const()[name = tensor("v_123_pad_type_0"), val = tensor("custom")]; + tensor v_123_pad_0 = const()[name = tensor("v_123_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_6_attn2_to_v_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_6_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2164827200)))]; + tensor v_123_cast_fp16 = conv(dilations = var_27105, groups = var_21077, pad = v_123_pad_0, pad_type = v_123_pad_type_0, strides = var_27103, weight = mid_block_attentions_0_transformer_blocks_6_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_123_cast_fp16")]; + tensor var_27109_begin_0 = const()[name = tensor("op_27109_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_27109_end_0 = const()[name = tensor("op_27109_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_27109_end_mask_0 = const()[name = tensor("op_27109_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27109_cast_fp16 = slice_by_index(begin = var_27109_begin_0, end = var_27109_end_0, end_mask = var_27109_end_mask_0, x = q_123_cast_fp16)[name = tensor("op_27109_cast_fp16")]; + tensor var_27113_begin_0 = const()[name = tensor("op_27113_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_27113_end_0 = const()[name = tensor("op_27113_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_27113_end_mask_0 = const()[name = tensor("op_27113_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27113_cast_fp16 = slice_by_index(begin = var_27113_begin_0, end = var_27113_end_0, end_mask = var_27113_end_mask_0, x = q_123_cast_fp16)[name = tensor("op_27113_cast_fp16")]; + tensor var_27117_begin_0 = const()[name = tensor("op_27117_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_27117_end_0 = const()[name = tensor("op_27117_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_27117_end_mask_0 = const()[name = tensor("op_27117_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27117_cast_fp16 = slice_by_index(begin = var_27117_begin_0, end = var_27117_end_0, end_mask = var_27117_end_mask_0, x = q_123_cast_fp16)[name = tensor("op_27117_cast_fp16")]; + tensor var_27121_begin_0 = const()[name = tensor("op_27121_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_27121_end_0 = const()[name = tensor("op_27121_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_27121_end_mask_0 = const()[name = tensor("op_27121_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27121_cast_fp16 = slice_by_index(begin = var_27121_begin_0, end = var_27121_end_0, end_mask = var_27121_end_mask_0, x = q_123_cast_fp16)[name = tensor("op_27121_cast_fp16")]; + tensor var_27125_begin_0 = const()[name = tensor("op_27125_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_27125_end_0 = const()[name = tensor("op_27125_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_27125_end_mask_0 = const()[name = tensor("op_27125_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27125_cast_fp16 = slice_by_index(begin = var_27125_begin_0, end = var_27125_end_0, end_mask = var_27125_end_mask_0, x = q_123_cast_fp16)[name = tensor("op_27125_cast_fp16")]; + tensor var_27129_begin_0 = const()[name = tensor("op_27129_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_27129_end_0 = const()[name = tensor("op_27129_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_27129_end_mask_0 = const()[name = tensor("op_27129_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27129_cast_fp16 = slice_by_index(begin = var_27129_begin_0, end = var_27129_end_0, end_mask = var_27129_end_mask_0, x = q_123_cast_fp16)[name = tensor("op_27129_cast_fp16")]; + tensor var_27133_begin_0 = const()[name = tensor("op_27133_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_27133_end_0 = const()[name = tensor("op_27133_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_27133_end_mask_0 = const()[name = tensor("op_27133_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27133_cast_fp16 = slice_by_index(begin = var_27133_begin_0, end = var_27133_end_0, end_mask = var_27133_end_mask_0, x = q_123_cast_fp16)[name = tensor("op_27133_cast_fp16")]; + tensor var_27137_begin_0 = const()[name = tensor("op_27137_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_27137_end_0 = const()[name = tensor("op_27137_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_27137_end_mask_0 = const()[name = tensor("op_27137_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27137_cast_fp16 = slice_by_index(begin = var_27137_begin_0, end = var_27137_end_0, end_mask = var_27137_end_mask_0, x = q_123_cast_fp16)[name = tensor("op_27137_cast_fp16")]; + tensor var_27141_begin_0 = const()[name = tensor("op_27141_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_27141_end_0 = const()[name = tensor("op_27141_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_27141_end_mask_0 = const()[name = tensor("op_27141_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27141_cast_fp16 = slice_by_index(begin = var_27141_begin_0, end = var_27141_end_0, end_mask = var_27141_end_mask_0, x = q_123_cast_fp16)[name = tensor("op_27141_cast_fp16")]; + tensor var_27145_begin_0 = const()[name = tensor("op_27145_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_27145_end_0 = const()[name = tensor("op_27145_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_27145_end_mask_0 = const()[name = tensor("op_27145_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27145_cast_fp16 = slice_by_index(begin = var_27145_begin_0, end = var_27145_end_0, end_mask = var_27145_end_mask_0, x = q_123_cast_fp16)[name = tensor("op_27145_cast_fp16")]; + tensor var_27149_begin_0 = const()[name = tensor("op_27149_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_27149_end_0 = const()[name = tensor("op_27149_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_27149_end_mask_0 = const()[name = tensor("op_27149_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27149_cast_fp16 = slice_by_index(begin = var_27149_begin_0, end = var_27149_end_0, end_mask = var_27149_end_mask_0, x = q_123_cast_fp16)[name = tensor("op_27149_cast_fp16")]; + tensor var_27153_begin_0 = const()[name = tensor("op_27153_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_27153_end_0 = const()[name = tensor("op_27153_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_27153_end_mask_0 = const()[name = tensor("op_27153_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27153_cast_fp16 = slice_by_index(begin = var_27153_begin_0, end = var_27153_end_0, end_mask = var_27153_end_mask_0, x = q_123_cast_fp16)[name = tensor("op_27153_cast_fp16")]; + tensor var_27157_begin_0 = const()[name = tensor("op_27157_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_27157_end_0 = const()[name = tensor("op_27157_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_27157_end_mask_0 = const()[name = tensor("op_27157_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27157_cast_fp16 = slice_by_index(begin = var_27157_begin_0, end = var_27157_end_0, end_mask = var_27157_end_mask_0, x = q_123_cast_fp16)[name = tensor("op_27157_cast_fp16")]; + tensor var_27161_begin_0 = const()[name = tensor("op_27161_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_27161_end_0 = const()[name = tensor("op_27161_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_27161_end_mask_0 = const()[name = tensor("op_27161_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27161_cast_fp16 = slice_by_index(begin = var_27161_begin_0, end = var_27161_end_0, end_mask = var_27161_end_mask_0, x = q_123_cast_fp16)[name = tensor("op_27161_cast_fp16")]; + tensor var_27165_begin_0 = const()[name = tensor("op_27165_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_27165_end_0 = const()[name = tensor("op_27165_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_27165_end_mask_0 = const()[name = tensor("op_27165_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27165_cast_fp16 = slice_by_index(begin = var_27165_begin_0, end = var_27165_end_0, end_mask = var_27165_end_mask_0, x = q_123_cast_fp16)[name = tensor("op_27165_cast_fp16")]; + tensor var_27169_begin_0 = const()[name = tensor("op_27169_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_27169_end_0 = const()[name = tensor("op_27169_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_27169_end_mask_0 = const()[name = tensor("op_27169_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27169_cast_fp16 = slice_by_index(begin = var_27169_begin_0, end = var_27169_end_0, end_mask = var_27169_end_mask_0, x = q_123_cast_fp16)[name = tensor("op_27169_cast_fp16")]; + tensor var_27173_begin_0 = const()[name = tensor("op_27173_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_27173_end_0 = const()[name = tensor("op_27173_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_27173_end_mask_0 = const()[name = tensor("op_27173_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27173_cast_fp16 = slice_by_index(begin = var_27173_begin_0, end = var_27173_end_0, end_mask = var_27173_end_mask_0, x = q_123_cast_fp16)[name = tensor("op_27173_cast_fp16")]; + tensor var_27177_begin_0 = const()[name = tensor("op_27177_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_27177_end_0 = const()[name = tensor("op_27177_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_27177_end_mask_0 = const()[name = tensor("op_27177_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27177_cast_fp16 = slice_by_index(begin = var_27177_begin_0, end = var_27177_end_0, end_mask = var_27177_end_mask_0, x = q_123_cast_fp16)[name = tensor("op_27177_cast_fp16")]; + tensor var_27181_begin_0 = const()[name = tensor("op_27181_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_27181_end_0 = const()[name = tensor("op_27181_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_27181_end_mask_0 = const()[name = tensor("op_27181_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27181_cast_fp16 = slice_by_index(begin = var_27181_begin_0, end = var_27181_end_0, end_mask = var_27181_end_mask_0, x = q_123_cast_fp16)[name = tensor("op_27181_cast_fp16")]; + tensor var_27185_begin_0 = const()[name = tensor("op_27185_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_27185_end_0 = const()[name = tensor("op_27185_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_27185_end_mask_0 = const()[name = tensor("op_27185_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27185_cast_fp16 = slice_by_index(begin = var_27185_begin_0, end = var_27185_end_0, end_mask = var_27185_end_mask_0, x = q_123_cast_fp16)[name = tensor("op_27185_cast_fp16")]; + tensor k_247_perm_0 = const()[name = tensor("k_247_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_27192_begin_0 = const()[name = tensor("op_27192_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_27192_end_0 = const()[name = tensor("op_27192_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_27192_end_mask_0 = const()[name = tensor("op_27192_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_247_cast_fp16 = transpose(perm = k_247_perm_0, x = k_245_cast_fp16)[name = tensor("transpose_78")]; + tensor var_27192_cast_fp16 = slice_by_index(begin = var_27192_begin_0, end = var_27192_end_0, end_mask = var_27192_end_mask_0, x = k_247_cast_fp16)[name = tensor("op_27192_cast_fp16")]; + tensor var_27196_begin_0 = const()[name = tensor("op_27196_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_27196_end_0 = const()[name = tensor("op_27196_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_27196_end_mask_0 = const()[name = tensor("op_27196_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27196_cast_fp16 = slice_by_index(begin = var_27196_begin_0, end = var_27196_end_0, end_mask = var_27196_end_mask_0, x = k_247_cast_fp16)[name = tensor("op_27196_cast_fp16")]; + tensor var_27200_begin_0 = const()[name = tensor("op_27200_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_27200_end_0 = const()[name = tensor("op_27200_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_27200_end_mask_0 = const()[name = tensor("op_27200_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27200_cast_fp16 = slice_by_index(begin = var_27200_begin_0, end = var_27200_end_0, end_mask = var_27200_end_mask_0, x = k_247_cast_fp16)[name = tensor("op_27200_cast_fp16")]; + tensor var_27204_begin_0 = const()[name = tensor("op_27204_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_27204_end_0 = const()[name = tensor("op_27204_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_27204_end_mask_0 = const()[name = tensor("op_27204_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27204_cast_fp16 = slice_by_index(begin = var_27204_begin_0, end = var_27204_end_0, end_mask = var_27204_end_mask_0, x = k_247_cast_fp16)[name = tensor("op_27204_cast_fp16")]; + tensor var_27208_begin_0 = const()[name = tensor("op_27208_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_27208_end_0 = const()[name = tensor("op_27208_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_27208_end_mask_0 = const()[name = tensor("op_27208_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27208_cast_fp16 = slice_by_index(begin = var_27208_begin_0, end = var_27208_end_0, end_mask = var_27208_end_mask_0, x = k_247_cast_fp16)[name = tensor("op_27208_cast_fp16")]; + tensor var_27212_begin_0 = const()[name = tensor("op_27212_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_27212_end_0 = const()[name = tensor("op_27212_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_27212_end_mask_0 = const()[name = tensor("op_27212_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27212_cast_fp16 = slice_by_index(begin = var_27212_begin_0, end = var_27212_end_0, end_mask = var_27212_end_mask_0, x = k_247_cast_fp16)[name = tensor("op_27212_cast_fp16")]; + tensor var_27216_begin_0 = const()[name = tensor("op_27216_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_27216_end_0 = const()[name = tensor("op_27216_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_27216_end_mask_0 = const()[name = tensor("op_27216_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27216_cast_fp16 = slice_by_index(begin = var_27216_begin_0, end = var_27216_end_0, end_mask = var_27216_end_mask_0, x = k_247_cast_fp16)[name = tensor("op_27216_cast_fp16")]; + tensor var_27220_begin_0 = const()[name = tensor("op_27220_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_27220_end_0 = const()[name = tensor("op_27220_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_27220_end_mask_0 = const()[name = tensor("op_27220_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27220_cast_fp16 = slice_by_index(begin = var_27220_begin_0, end = var_27220_end_0, end_mask = var_27220_end_mask_0, x = k_247_cast_fp16)[name = tensor("op_27220_cast_fp16")]; + tensor var_27224_begin_0 = const()[name = tensor("op_27224_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_27224_end_0 = const()[name = tensor("op_27224_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_27224_end_mask_0 = const()[name = tensor("op_27224_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27224_cast_fp16 = slice_by_index(begin = var_27224_begin_0, end = var_27224_end_0, end_mask = var_27224_end_mask_0, x = k_247_cast_fp16)[name = tensor("op_27224_cast_fp16")]; + tensor var_27228_begin_0 = const()[name = tensor("op_27228_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_27228_end_0 = const()[name = tensor("op_27228_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_27228_end_mask_0 = const()[name = tensor("op_27228_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27228_cast_fp16 = slice_by_index(begin = var_27228_begin_0, end = var_27228_end_0, end_mask = var_27228_end_mask_0, x = k_247_cast_fp16)[name = tensor("op_27228_cast_fp16")]; + tensor var_27232_begin_0 = const()[name = tensor("op_27232_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_27232_end_0 = const()[name = tensor("op_27232_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_27232_end_mask_0 = const()[name = tensor("op_27232_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27232_cast_fp16 = slice_by_index(begin = var_27232_begin_0, end = var_27232_end_0, end_mask = var_27232_end_mask_0, x = k_247_cast_fp16)[name = tensor("op_27232_cast_fp16")]; + tensor var_27236_begin_0 = const()[name = tensor("op_27236_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_27236_end_0 = const()[name = tensor("op_27236_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_27236_end_mask_0 = const()[name = tensor("op_27236_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27236_cast_fp16 = slice_by_index(begin = var_27236_begin_0, end = var_27236_end_0, end_mask = var_27236_end_mask_0, x = k_247_cast_fp16)[name = tensor("op_27236_cast_fp16")]; + tensor var_27240_begin_0 = const()[name = tensor("op_27240_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_27240_end_0 = const()[name = tensor("op_27240_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_27240_end_mask_0 = const()[name = tensor("op_27240_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27240_cast_fp16 = slice_by_index(begin = var_27240_begin_0, end = var_27240_end_0, end_mask = var_27240_end_mask_0, x = k_247_cast_fp16)[name = tensor("op_27240_cast_fp16")]; + tensor var_27244_begin_0 = const()[name = tensor("op_27244_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_27244_end_0 = const()[name = tensor("op_27244_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_27244_end_mask_0 = const()[name = tensor("op_27244_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27244_cast_fp16 = slice_by_index(begin = var_27244_begin_0, end = var_27244_end_0, end_mask = var_27244_end_mask_0, x = k_247_cast_fp16)[name = tensor("op_27244_cast_fp16")]; + tensor var_27248_begin_0 = const()[name = tensor("op_27248_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_27248_end_0 = const()[name = tensor("op_27248_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_27248_end_mask_0 = const()[name = tensor("op_27248_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27248_cast_fp16 = slice_by_index(begin = var_27248_begin_0, end = var_27248_end_0, end_mask = var_27248_end_mask_0, x = k_247_cast_fp16)[name = tensor("op_27248_cast_fp16")]; + tensor var_27252_begin_0 = const()[name = tensor("op_27252_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_27252_end_0 = const()[name = tensor("op_27252_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_27252_end_mask_0 = const()[name = tensor("op_27252_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27252_cast_fp16 = slice_by_index(begin = var_27252_begin_0, end = var_27252_end_0, end_mask = var_27252_end_mask_0, x = k_247_cast_fp16)[name = tensor("op_27252_cast_fp16")]; + tensor var_27256_begin_0 = const()[name = tensor("op_27256_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_27256_end_0 = const()[name = tensor("op_27256_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_27256_end_mask_0 = const()[name = tensor("op_27256_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27256_cast_fp16 = slice_by_index(begin = var_27256_begin_0, end = var_27256_end_0, end_mask = var_27256_end_mask_0, x = k_247_cast_fp16)[name = tensor("op_27256_cast_fp16")]; + tensor var_27260_begin_0 = const()[name = tensor("op_27260_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_27260_end_0 = const()[name = tensor("op_27260_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_27260_end_mask_0 = const()[name = tensor("op_27260_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27260_cast_fp16 = slice_by_index(begin = var_27260_begin_0, end = var_27260_end_0, end_mask = var_27260_end_mask_0, x = k_247_cast_fp16)[name = tensor("op_27260_cast_fp16")]; + tensor var_27264_begin_0 = const()[name = tensor("op_27264_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_27264_end_0 = const()[name = tensor("op_27264_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_27264_end_mask_0 = const()[name = tensor("op_27264_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27264_cast_fp16 = slice_by_index(begin = var_27264_begin_0, end = var_27264_end_0, end_mask = var_27264_end_mask_0, x = k_247_cast_fp16)[name = tensor("op_27264_cast_fp16")]; + tensor var_27268_begin_0 = const()[name = tensor("op_27268_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_27268_end_0 = const()[name = tensor("op_27268_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_27268_end_mask_0 = const()[name = tensor("op_27268_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27268_cast_fp16 = slice_by_index(begin = var_27268_begin_0, end = var_27268_end_0, end_mask = var_27268_end_mask_0, x = k_247_cast_fp16)[name = tensor("op_27268_cast_fp16")]; + tensor var_27270_begin_0 = const()[name = tensor("op_27270_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_27270_end_0 = const()[name = tensor("op_27270_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_27270_end_mask_0 = const()[name = tensor("op_27270_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27270_cast_fp16 = slice_by_index(begin = var_27270_begin_0, end = var_27270_end_0, end_mask = var_27270_end_mask_0, x = v_123_cast_fp16)[name = tensor("op_27270_cast_fp16")]; + tensor var_27274_begin_0 = const()[name = tensor("op_27274_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_27274_end_0 = const()[name = tensor("op_27274_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_27274_end_mask_0 = const()[name = tensor("op_27274_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27274_cast_fp16 = slice_by_index(begin = var_27274_begin_0, end = var_27274_end_0, end_mask = var_27274_end_mask_0, x = v_123_cast_fp16)[name = tensor("op_27274_cast_fp16")]; + tensor var_27278_begin_0 = const()[name = tensor("op_27278_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_27278_end_0 = const()[name = tensor("op_27278_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_27278_end_mask_0 = const()[name = tensor("op_27278_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27278_cast_fp16 = slice_by_index(begin = var_27278_begin_0, end = var_27278_end_0, end_mask = var_27278_end_mask_0, x = v_123_cast_fp16)[name = tensor("op_27278_cast_fp16")]; + tensor var_27282_begin_0 = const()[name = tensor("op_27282_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_27282_end_0 = const()[name = tensor("op_27282_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_27282_end_mask_0 = const()[name = tensor("op_27282_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27282_cast_fp16 = slice_by_index(begin = var_27282_begin_0, end = var_27282_end_0, end_mask = var_27282_end_mask_0, x = v_123_cast_fp16)[name = tensor("op_27282_cast_fp16")]; + tensor var_27286_begin_0 = const()[name = tensor("op_27286_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_27286_end_0 = const()[name = tensor("op_27286_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_27286_end_mask_0 = const()[name = tensor("op_27286_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27286_cast_fp16 = slice_by_index(begin = var_27286_begin_0, end = var_27286_end_0, end_mask = var_27286_end_mask_0, x = v_123_cast_fp16)[name = tensor("op_27286_cast_fp16")]; + tensor var_27290_begin_0 = const()[name = tensor("op_27290_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_27290_end_0 = const()[name = tensor("op_27290_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_27290_end_mask_0 = const()[name = tensor("op_27290_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27290_cast_fp16 = slice_by_index(begin = var_27290_begin_0, end = var_27290_end_0, end_mask = var_27290_end_mask_0, x = v_123_cast_fp16)[name = tensor("op_27290_cast_fp16")]; + tensor var_27294_begin_0 = const()[name = tensor("op_27294_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_27294_end_0 = const()[name = tensor("op_27294_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_27294_end_mask_0 = const()[name = tensor("op_27294_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27294_cast_fp16 = slice_by_index(begin = var_27294_begin_0, end = var_27294_end_0, end_mask = var_27294_end_mask_0, x = v_123_cast_fp16)[name = tensor("op_27294_cast_fp16")]; + tensor var_27298_begin_0 = const()[name = tensor("op_27298_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_27298_end_0 = const()[name = tensor("op_27298_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_27298_end_mask_0 = const()[name = tensor("op_27298_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27298_cast_fp16 = slice_by_index(begin = var_27298_begin_0, end = var_27298_end_0, end_mask = var_27298_end_mask_0, x = v_123_cast_fp16)[name = tensor("op_27298_cast_fp16")]; + tensor var_27302_begin_0 = const()[name = tensor("op_27302_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_27302_end_0 = const()[name = tensor("op_27302_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_27302_end_mask_0 = const()[name = tensor("op_27302_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27302_cast_fp16 = slice_by_index(begin = var_27302_begin_0, end = var_27302_end_0, end_mask = var_27302_end_mask_0, x = v_123_cast_fp16)[name = tensor("op_27302_cast_fp16")]; + tensor var_27306_begin_0 = const()[name = tensor("op_27306_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_27306_end_0 = const()[name = tensor("op_27306_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_27306_end_mask_0 = const()[name = tensor("op_27306_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27306_cast_fp16 = slice_by_index(begin = var_27306_begin_0, end = var_27306_end_0, end_mask = var_27306_end_mask_0, x = v_123_cast_fp16)[name = tensor("op_27306_cast_fp16")]; + tensor var_27310_begin_0 = const()[name = tensor("op_27310_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_27310_end_0 = const()[name = tensor("op_27310_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_27310_end_mask_0 = const()[name = tensor("op_27310_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27310_cast_fp16 = slice_by_index(begin = var_27310_begin_0, end = var_27310_end_0, end_mask = var_27310_end_mask_0, x = v_123_cast_fp16)[name = tensor("op_27310_cast_fp16")]; + tensor var_27314_begin_0 = const()[name = tensor("op_27314_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_27314_end_0 = const()[name = tensor("op_27314_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_27314_end_mask_0 = const()[name = tensor("op_27314_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27314_cast_fp16 = slice_by_index(begin = var_27314_begin_0, end = var_27314_end_0, end_mask = var_27314_end_mask_0, x = v_123_cast_fp16)[name = tensor("op_27314_cast_fp16")]; + tensor var_27318_begin_0 = const()[name = tensor("op_27318_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_27318_end_0 = const()[name = tensor("op_27318_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_27318_end_mask_0 = const()[name = tensor("op_27318_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27318_cast_fp16 = slice_by_index(begin = var_27318_begin_0, end = var_27318_end_0, end_mask = var_27318_end_mask_0, x = v_123_cast_fp16)[name = tensor("op_27318_cast_fp16")]; + tensor var_27322_begin_0 = const()[name = tensor("op_27322_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_27322_end_0 = const()[name = tensor("op_27322_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_27322_end_mask_0 = const()[name = tensor("op_27322_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27322_cast_fp16 = slice_by_index(begin = var_27322_begin_0, end = var_27322_end_0, end_mask = var_27322_end_mask_0, x = v_123_cast_fp16)[name = tensor("op_27322_cast_fp16")]; + tensor var_27326_begin_0 = const()[name = tensor("op_27326_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_27326_end_0 = const()[name = tensor("op_27326_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_27326_end_mask_0 = const()[name = tensor("op_27326_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27326_cast_fp16 = slice_by_index(begin = var_27326_begin_0, end = var_27326_end_0, end_mask = var_27326_end_mask_0, x = v_123_cast_fp16)[name = tensor("op_27326_cast_fp16")]; + tensor var_27330_begin_0 = const()[name = tensor("op_27330_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_27330_end_0 = const()[name = tensor("op_27330_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_27330_end_mask_0 = const()[name = tensor("op_27330_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27330_cast_fp16 = slice_by_index(begin = var_27330_begin_0, end = var_27330_end_0, end_mask = var_27330_end_mask_0, x = v_123_cast_fp16)[name = tensor("op_27330_cast_fp16")]; + tensor var_27334_begin_0 = const()[name = tensor("op_27334_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_27334_end_0 = const()[name = tensor("op_27334_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_27334_end_mask_0 = const()[name = tensor("op_27334_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27334_cast_fp16 = slice_by_index(begin = var_27334_begin_0, end = var_27334_end_0, end_mask = var_27334_end_mask_0, x = v_123_cast_fp16)[name = tensor("op_27334_cast_fp16")]; + tensor var_27338_begin_0 = const()[name = tensor("op_27338_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_27338_end_0 = const()[name = tensor("op_27338_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_27338_end_mask_0 = const()[name = tensor("op_27338_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27338_cast_fp16 = slice_by_index(begin = var_27338_begin_0, end = var_27338_end_0, end_mask = var_27338_end_mask_0, x = v_123_cast_fp16)[name = tensor("op_27338_cast_fp16")]; + tensor var_27342_begin_0 = const()[name = tensor("op_27342_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_27342_end_0 = const()[name = tensor("op_27342_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_27342_end_mask_0 = const()[name = tensor("op_27342_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27342_cast_fp16 = slice_by_index(begin = var_27342_begin_0, end = var_27342_end_0, end_mask = var_27342_end_mask_0, x = v_123_cast_fp16)[name = tensor("op_27342_cast_fp16")]; + tensor var_27346_begin_0 = const()[name = tensor("op_27346_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_27346_end_0 = const()[name = tensor("op_27346_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_27346_end_mask_0 = const()[name = tensor("op_27346_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27346_cast_fp16 = slice_by_index(begin = var_27346_begin_0, end = var_27346_end_0, end_mask = var_27346_end_mask_0, x = v_123_cast_fp16)[name = tensor("op_27346_cast_fp16")]; + tensor var_27350_equation_0 = const()[name = tensor("op_27350_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27350_cast_fp16 = einsum(equation = var_27350_equation_0, values = (var_27192_cast_fp16, var_27109_cast_fp16))[name = tensor("op_27350_cast_fp16")]; + tensor var_27351_to_fp16 = const()[name = tensor("op_27351_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2281_cast_fp16 = mul(x = var_27350_cast_fp16, y = var_27351_to_fp16)[name = tensor("aw_2281_cast_fp16")]; + tensor var_27354_equation_0 = const()[name = tensor("op_27354_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27354_cast_fp16 = einsum(equation = var_27354_equation_0, values = (var_27196_cast_fp16, var_27113_cast_fp16))[name = tensor("op_27354_cast_fp16")]; + tensor var_27355_to_fp16 = const()[name = tensor("op_27355_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2283_cast_fp16 = mul(x = var_27354_cast_fp16, y = var_27355_to_fp16)[name = tensor("aw_2283_cast_fp16")]; + tensor var_27358_equation_0 = const()[name = tensor("op_27358_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27358_cast_fp16 = einsum(equation = var_27358_equation_0, values = (var_27200_cast_fp16, var_27117_cast_fp16))[name = tensor("op_27358_cast_fp16")]; + tensor var_27359_to_fp16 = const()[name = tensor("op_27359_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2285_cast_fp16 = mul(x = var_27358_cast_fp16, y = var_27359_to_fp16)[name = tensor("aw_2285_cast_fp16")]; + tensor var_27362_equation_0 = const()[name = tensor("op_27362_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27362_cast_fp16 = einsum(equation = var_27362_equation_0, values = (var_27204_cast_fp16, var_27121_cast_fp16))[name = tensor("op_27362_cast_fp16")]; + tensor var_27363_to_fp16 = const()[name = tensor("op_27363_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2287_cast_fp16 = mul(x = var_27362_cast_fp16, y = var_27363_to_fp16)[name = tensor("aw_2287_cast_fp16")]; + tensor var_27366_equation_0 = const()[name = tensor("op_27366_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27366_cast_fp16 = einsum(equation = var_27366_equation_0, values = (var_27208_cast_fp16, var_27125_cast_fp16))[name = tensor("op_27366_cast_fp16")]; + tensor var_27367_to_fp16 = const()[name = tensor("op_27367_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2289_cast_fp16 = mul(x = var_27366_cast_fp16, y = var_27367_to_fp16)[name = tensor("aw_2289_cast_fp16")]; + tensor var_27370_equation_0 = const()[name = tensor("op_27370_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27370_cast_fp16 = einsum(equation = var_27370_equation_0, values = (var_27212_cast_fp16, var_27129_cast_fp16))[name = tensor("op_27370_cast_fp16")]; + tensor var_27371_to_fp16 = const()[name = tensor("op_27371_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2291_cast_fp16 = mul(x = var_27370_cast_fp16, y = var_27371_to_fp16)[name = tensor("aw_2291_cast_fp16")]; + tensor var_27374_equation_0 = const()[name = tensor("op_27374_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27374_cast_fp16 = einsum(equation = var_27374_equation_0, values = (var_27216_cast_fp16, var_27133_cast_fp16))[name = tensor("op_27374_cast_fp16")]; + tensor var_27375_to_fp16 = const()[name = tensor("op_27375_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2293_cast_fp16 = mul(x = var_27374_cast_fp16, y = var_27375_to_fp16)[name = tensor("aw_2293_cast_fp16")]; + tensor var_27378_equation_0 = const()[name = tensor("op_27378_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27378_cast_fp16 = einsum(equation = var_27378_equation_0, values = (var_27220_cast_fp16, var_27137_cast_fp16))[name = tensor("op_27378_cast_fp16")]; + tensor var_27379_to_fp16 = const()[name = tensor("op_27379_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2295_cast_fp16 = mul(x = var_27378_cast_fp16, y = var_27379_to_fp16)[name = tensor("aw_2295_cast_fp16")]; + tensor var_27382_equation_0 = const()[name = tensor("op_27382_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27382_cast_fp16 = einsum(equation = var_27382_equation_0, values = (var_27224_cast_fp16, var_27141_cast_fp16))[name = tensor("op_27382_cast_fp16")]; + tensor var_27383_to_fp16 = const()[name = tensor("op_27383_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2297_cast_fp16 = mul(x = var_27382_cast_fp16, y = var_27383_to_fp16)[name = tensor("aw_2297_cast_fp16")]; + tensor var_27386_equation_0 = const()[name = tensor("op_27386_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27386_cast_fp16 = einsum(equation = var_27386_equation_0, values = (var_27228_cast_fp16, var_27145_cast_fp16))[name = tensor("op_27386_cast_fp16")]; + tensor var_27387_to_fp16 = const()[name = tensor("op_27387_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2299_cast_fp16 = mul(x = var_27386_cast_fp16, y = var_27387_to_fp16)[name = tensor("aw_2299_cast_fp16")]; + tensor var_27390_equation_0 = const()[name = tensor("op_27390_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27390_cast_fp16 = einsum(equation = var_27390_equation_0, values = (var_27232_cast_fp16, var_27149_cast_fp16))[name = tensor("op_27390_cast_fp16")]; + tensor var_27391_to_fp16 = const()[name = tensor("op_27391_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2301_cast_fp16 = mul(x = var_27390_cast_fp16, y = var_27391_to_fp16)[name = tensor("aw_2301_cast_fp16")]; + tensor var_27394_equation_0 = const()[name = tensor("op_27394_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27394_cast_fp16 = einsum(equation = var_27394_equation_0, values = (var_27236_cast_fp16, var_27153_cast_fp16))[name = tensor("op_27394_cast_fp16")]; + tensor var_27395_to_fp16 = const()[name = tensor("op_27395_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2303_cast_fp16 = mul(x = var_27394_cast_fp16, y = var_27395_to_fp16)[name = tensor("aw_2303_cast_fp16")]; + tensor var_27398_equation_0 = const()[name = tensor("op_27398_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27398_cast_fp16 = einsum(equation = var_27398_equation_0, values = (var_27240_cast_fp16, var_27157_cast_fp16))[name = tensor("op_27398_cast_fp16")]; + tensor var_27399_to_fp16 = const()[name = tensor("op_27399_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2305_cast_fp16 = mul(x = var_27398_cast_fp16, y = var_27399_to_fp16)[name = tensor("aw_2305_cast_fp16")]; + tensor var_27402_equation_0 = const()[name = tensor("op_27402_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27402_cast_fp16 = einsum(equation = var_27402_equation_0, values = (var_27244_cast_fp16, var_27161_cast_fp16))[name = tensor("op_27402_cast_fp16")]; + tensor var_27403_to_fp16 = const()[name = tensor("op_27403_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2307_cast_fp16 = mul(x = var_27402_cast_fp16, y = var_27403_to_fp16)[name = tensor("aw_2307_cast_fp16")]; + tensor var_27406_equation_0 = const()[name = tensor("op_27406_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27406_cast_fp16 = einsum(equation = var_27406_equation_0, values = (var_27248_cast_fp16, var_27165_cast_fp16))[name = tensor("op_27406_cast_fp16")]; + tensor var_27407_to_fp16 = const()[name = tensor("op_27407_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2309_cast_fp16 = mul(x = var_27406_cast_fp16, y = var_27407_to_fp16)[name = tensor("aw_2309_cast_fp16")]; + tensor var_27410_equation_0 = const()[name = tensor("op_27410_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27410_cast_fp16 = einsum(equation = var_27410_equation_0, values = (var_27252_cast_fp16, var_27169_cast_fp16))[name = tensor("op_27410_cast_fp16")]; + tensor var_27411_to_fp16 = const()[name = tensor("op_27411_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2311_cast_fp16 = mul(x = var_27410_cast_fp16, y = var_27411_to_fp16)[name = tensor("aw_2311_cast_fp16")]; + tensor var_27414_equation_0 = const()[name = tensor("op_27414_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27414_cast_fp16 = einsum(equation = var_27414_equation_0, values = (var_27256_cast_fp16, var_27173_cast_fp16))[name = tensor("op_27414_cast_fp16")]; + tensor var_27415_to_fp16 = const()[name = tensor("op_27415_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2313_cast_fp16 = mul(x = var_27414_cast_fp16, y = var_27415_to_fp16)[name = tensor("aw_2313_cast_fp16")]; + tensor var_27418_equation_0 = const()[name = tensor("op_27418_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27418_cast_fp16 = einsum(equation = var_27418_equation_0, values = (var_27260_cast_fp16, var_27177_cast_fp16))[name = tensor("op_27418_cast_fp16")]; + tensor var_27419_to_fp16 = const()[name = tensor("op_27419_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2315_cast_fp16 = mul(x = var_27418_cast_fp16, y = var_27419_to_fp16)[name = tensor("aw_2315_cast_fp16")]; + tensor var_27422_equation_0 = const()[name = tensor("op_27422_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27422_cast_fp16 = einsum(equation = var_27422_equation_0, values = (var_27264_cast_fp16, var_27181_cast_fp16))[name = tensor("op_27422_cast_fp16")]; + tensor var_27423_to_fp16 = const()[name = tensor("op_27423_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2317_cast_fp16 = mul(x = var_27422_cast_fp16, y = var_27423_to_fp16)[name = tensor("aw_2317_cast_fp16")]; + tensor var_27426_equation_0 = const()[name = tensor("op_27426_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27426_cast_fp16 = einsum(equation = var_27426_equation_0, values = (var_27268_cast_fp16, var_27185_cast_fp16))[name = tensor("op_27426_cast_fp16")]; + tensor var_27427_to_fp16 = const()[name = tensor("op_27427_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2319_cast_fp16 = mul(x = var_27426_cast_fp16, y = var_27427_to_fp16)[name = tensor("aw_2319_cast_fp16")]; + tensor var_27429_cast_fp16 = softmax(axis = var_21077, x = aw_2281_cast_fp16)[name = tensor("op_27429_cast_fp16")]; + tensor var_27430_cast_fp16 = softmax(axis = var_21077, x = aw_2283_cast_fp16)[name = tensor("op_27430_cast_fp16")]; + tensor var_27431_cast_fp16 = softmax(axis = var_21077, x = aw_2285_cast_fp16)[name = tensor("op_27431_cast_fp16")]; + tensor var_27432_cast_fp16 = softmax(axis = var_21077, x = aw_2287_cast_fp16)[name = tensor("op_27432_cast_fp16")]; + tensor var_27433_cast_fp16 = softmax(axis = var_21077, x = aw_2289_cast_fp16)[name = tensor("op_27433_cast_fp16")]; + tensor var_27434_cast_fp16 = softmax(axis = var_21077, x = aw_2291_cast_fp16)[name = tensor("op_27434_cast_fp16")]; + tensor var_27435_cast_fp16 = softmax(axis = var_21077, x = aw_2293_cast_fp16)[name = tensor("op_27435_cast_fp16")]; + tensor var_27436_cast_fp16 = softmax(axis = var_21077, x = aw_2295_cast_fp16)[name = tensor("op_27436_cast_fp16")]; + tensor var_27437_cast_fp16 = softmax(axis = var_21077, x = aw_2297_cast_fp16)[name = tensor("op_27437_cast_fp16")]; + tensor var_27438_cast_fp16 = softmax(axis = var_21077, x = aw_2299_cast_fp16)[name = tensor("op_27438_cast_fp16")]; + tensor var_27439_cast_fp16 = softmax(axis = var_21077, x = aw_2301_cast_fp16)[name = tensor("op_27439_cast_fp16")]; + tensor var_27440_cast_fp16 = softmax(axis = var_21077, x = aw_2303_cast_fp16)[name = tensor("op_27440_cast_fp16")]; + tensor var_27441_cast_fp16 = softmax(axis = var_21077, x = aw_2305_cast_fp16)[name = tensor("op_27441_cast_fp16")]; + tensor var_27442_cast_fp16 = softmax(axis = var_21077, x = aw_2307_cast_fp16)[name = tensor("op_27442_cast_fp16")]; + tensor var_27443_cast_fp16 = softmax(axis = var_21077, x = aw_2309_cast_fp16)[name = tensor("op_27443_cast_fp16")]; + tensor var_27444_cast_fp16 = softmax(axis = var_21077, x = aw_2311_cast_fp16)[name = tensor("op_27444_cast_fp16")]; + tensor var_27445_cast_fp16 = softmax(axis = var_21077, x = aw_2313_cast_fp16)[name = tensor("op_27445_cast_fp16")]; + tensor var_27446_cast_fp16 = softmax(axis = var_21077, x = aw_2315_cast_fp16)[name = tensor("op_27446_cast_fp16")]; + tensor var_27447_cast_fp16 = softmax(axis = var_21077, x = aw_2317_cast_fp16)[name = tensor("op_27447_cast_fp16")]; + tensor var_27448_cast_fp16 = softmax(axis = var_21077, x = aw_2319_cast_fp16)[name = tensor("op_27448_cast_fp16")]; + tensor var_27450_equation_0 = const()[name = tensor("op_27450_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27450_cast_fp16 = einsum(equation = var_27450_equation_0, values = (var_27270_cast_fp16, var_27429_cast_fp16))[name = tensor("op_27450_cast_fp16")]; + tensor var_27452_equation_0 = const()[name = tensor("op_27452_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27452_cast_fp16 = einsum(equation = var_27452_equation_0, values = (var_27274_cast_fp16, var_27430_cast_fp16))[name = tensor("op_27452_cast_fp16")]; + tensor var_27454_equation_0 = const()[name = tensor("op_27454_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27454_cast_fp16 = einsum(equation = var_27454_equation_0, values = (var_27278_cast_fp16, var_27431_cast_fp16))[name = tensor("op_27454_cast_fp16")]; + tensor var_27456_equation_0 = const()[name = tensor("op_27456_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27456_cast_fp16 = einsum(equation = var_27456_equation_0, values = (var_27282_cast_fp16, var_27432_cast_fp16))[name = tensor("op_27456_cast_fp16")]; + tensor var_27458_equation_0 = const()[name = tensor("op_27458_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27458_cast_fp16 = einsum(equation = var_27458_equation_0, values = (var_27286_cast_fp16, var_27433_cast_fp16))[name = tensor("op_27458_cast_fp16")]; + tensor var_27460_equation_0 = const()[name = tensor("op_27460_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27460_cast_fp16 = einsum(equation = var_27460_equation_0, values = (var_27290_cast_fp16, var_27434_cast_fp16))[name = tensor("op_27460_cast_fp16")]; + tensor var_27462_equation_0 = const()[name = tensor("op_27462_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27462_cast_fp16 = einsum(equation = var_27462_equation_0, values = (var_27294_cast_fp16, var_27435_cast_fp16))[name = tensor("op_27462_cast_fp16")]; + tensor var_27464_equation_0 = const()[name = tensor("op_27464_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27464_cast_fp16 = einsum(equation = var_27464_equation_0, values = (var_27298_cast_fp16, var_27436_cast_fp16))[name = tensor("op_27464_cast_fp16")]; + tensor var_27466_equation_0 = const()[name = tensor("op_27466_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27466_cast_fp16 = einsum(equation = var_27466_equation_0, values = (var_27302_cast_fp16, var_27437_cast_fp16))[name = tensor("op_27466_cast_fp16")]; + tensor var_27468_equation_0 = const()[name = tensor("op_27468_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27468_cast_fp16 = einsum(equation = var_27468_equation_0, values = (var_27306_cast_fp16, var_27438_cast_fp16))[name = tensor("op_27468_cast_fp16")]; + tensor var_27470_equation_0 = const()[name = tensor("op_27470_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27470_cast_fp16 = einsum(equation = var_27470_equation_0, values = (var_27310_cast_fp16, var_27439_cast_fp16))[name = tensor("op_27470_cast_fp16")]; + tensor var_27472_equation_0 = const()[name = tensor("op_27472_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27472_cast_fp16 = einsum(equation = var_27472_equation_0, values = (var_27314_cast_fp16, var_27440_cast_fp16))[name = tensor("op_27472_cast_fp16")]; + tensor var_27474_equation_0 = const()[name = tensor("op_27474_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27474_cast_fp16 = einsum(equation = var_27474_equation_0, values = (var_27318_cast_fp16, var_27441_cast_fp16))[name = tensor("op_27474_cast_fp16")]; + tensor var_27476_equation_0 = const()[name = tensor("op_27476_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27476_cast_fp16 = einsum(equation = var_27476_equation_0, values = (var_27322_cast_fp16, var_27442_cast_fp16))[name = tensor("op_27476_cast_fp16")]; + tensor var_27478_equation_0 = const()[name = tensor("op_27478_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27478_cast_fp16 = einsum(equation = var_27478_equation_0, values = (var_27326_cast_fp16, var_27443_cast_fp16))[name = tensor("op_27478_cast_fp16")]; + tensor var_27480_equation_0 = const()[name = tensor("op_27480_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27480_cast_fp16 = einsum(equation = var_27480_equation_0, values = (var_27330_cast_fp16, var_27444_cast_fp16))[name = tensor("op_27480_cast_fp16")]; + tensor var_27482_equation_0 = const()[name = tensor("op_27482_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27482_cast_fp16 = einsum(equation = var_27482_equation_0, values = (var_27334_cast_fp16, var_27445_cast_fp16))[name = tensor("op_27482_cast_fp16")]; + tensor var_27484_equation_0 = const()[name = tensor("op_27484_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27484_cast_fp16 = einsum(equation = var_27484_equation_0, values = (var_27338_cast_fp16, var_27446_cast_fp16))[name = tensor("op_27484_cast_fp16")]; + tensor var_27486_equation_0 = const()[name = tensor("op_27486_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27486_cast_fp16 = einsum(equation = var_27486_equation_0, values = (var_27342_cast_fp16, var_27447_cast_fp16))[name = tensor("op_27486_cast_fp16")]; + tensor var_27488_equation_0 = const()[name = tensor("op_27488_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27488_cast_fp16 = einsum(equation = var_27488_equation_0, values = (var_27346_cast_fp16, var_27448_cast_fp16))[name = tensor("op_27488_cast_fp16")]; + tensor input_377_interleave_0 = const()[name = tensor("input_377_interleave_0"), val = tensor(false)]; + tensor input_377_cast_fp16 = concat(axis = var_21077, interleave = input_377_interleave_0, values = (var_27450_cast_fp16, var_27452_cast_fp16, var_27454_cast_fp16, var_27456_cast_fp16, var_27458_cast_fp16, var_27460_cast_fp16, var_27462_cast_fp16, var_27464_cast_fp16, var_27466_cast_fp16, var_27468_cast_fp16, var_27470_cast_fp16, var_27472_cast_fp16, var_27474_cast_fp16, var_27476_cast_fp16, var_27478_cast_fp16, var_27480_cast_fp16, var_27482_cast_fp16, var_27484_cast_fp16, var_27486_cast_fp16, var_27488_cast_fp16))[name = tensor("input_377_cast_fp16")]; + tensor var_27494 = const()[name = tensor("op_27494"), val = tensor([1, 1])]; + tensor var_27496 = const()[name = tensor("op_27496"), val = tensor([1, 1])]; + tensor var_27498_pad_type_0 = const()[name = tensor("op_27498_pad_type_0"), val = tensor("custom")]; + tensor var_27498_pad_0 = const()[name = tensor("op_27498_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_6_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_6_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2170070144)))]; + tensor mid_block_attentions_0_transformer_blocks_6_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_6_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2173347008)))]; + tensor var_27498_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_6_attn2_to_out_0_bias_to_fp16, dilations = var_27496, groups = var_21077, pad = var_27498_pad_0, pad_type = var_27498_pad_type_0, strides = var_27494, weight = mid_block_attentions_0_transformer_blocks_6_attn2_to_out_0_weight_to_fp16, x = input_377_cast_fp16)[name = tensor("op_27498_cast_fp16")]; + tensor inputs_185_cast_fp16 = add(x = var_27498_cast_fp16, y = inputs_183_cast_fp16)[name = tensor("inputs_185_cast_fp16")]; + tensor input_379_axes_0 = const()[name = tensor("input_379_axes_0"), val = tensor([1])]; + tensor input_379_gamma_0_to_fp16 = const()[name = tensor("input_379_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2173349632)))]; + tensor input_379_beta_0_to_fp16 = const()[name = tensor("input_379_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2173352256)))]; + tensor var_27508_to_fp16 = const()[name = tensor("op_27508_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_379_cast_fp16 = layer_norm(axes = input_379_axes_0, beta = input_379_beta_0_to_fp16, epsilon = var_27508_to_fp16, gamma = input_379_gamma_0_to_fp16, x = inputs_185_cast_fp16)[name = tensor("input_379_cast_fp16")]; + tensor var_27524 = const()[name = tensor("op_27524"), val = tensor([1, 1])]; + tensor var_27526 = const()[name = tensor("op_27526"), val = tensor([1, 1])]; + tensor var_27528_pad_type_0 = const()[name = tensor("op_27528_pad_type_0"), val = tensor("custom")]; + tensor var_27528_pad_0 = const()[name = tensor("op_27528_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_6_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_6_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2173354880)))]; + tensor mid_block_attentions_0_transformer_blocks_6_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_6_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2199569344)))]; + tensor var_27528_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_6_ff_net_0_proj_bias_to_fp16, dilations = var_27526, groups = var_21077, pad = var_27528_pad_0, pad_type = var_27528_pad_type_0, strides = var_27524, weight = mid_block_attentions_0_transformer_blocks_6_ff_net_0_proj_weight_to_fp16, x = input_379_cast_fp16)[name = tensor("op_27528_cast_fp16")]; + tensor var_27529_split_sizes_0 = const()[name = tensor("op_27529_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_27529_axis_0 = const()[name = tensor("op_27529_axis_0"), val = tensor(1)]; + tensor var_27529_cast_fp16_0, tensor var_27529_cast_fp16_1 = split(axis = var_27529_axis_0, split_sizes = var_27529_split_sizes_0, x = var_27528_cast_fp16)[name = tensor("op_27529_cast_fp16")]; + tensor var_27531_mode_0 = const()[name = tensor("op_27531_mode_0"), val = tensor("EXACT")]; + tensor var_27531_cast_fp16 = gelu(mode = var_27531_mode_0, x = var_27529_cast_fp16_1)[name = tensor("op_27531_cast_fp16")]; + tensor input_381_cast_fp16 = mul(x = var_27529_cast_fp16_0, y = var_27531_cast_fp16)[name = tensor("input_381_cast_fp16")]; + tensor var_27535 = const()[name = tensor("op_27535"), val = tensor([1, 1])]; + tensor var_27537 = const()[name = tensor("op_27537"), val = tensor([1, 1])]; + tensor var_27539_pad_type_0 = const()[name = tensor("op_27539_pad_type_0"), val = tensor("custom")]; + tensor var_27539_pad_0 = const()[name = tensor("op_27539_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_6_ff_net_2_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_6_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2199589888)))]; + tensor mid_block_attentions_0_transformer_blocks_6_ff_net_2_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_6_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2212697152)))]; + tensor var_27539_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_6_ff_net_2_bias_to_fp16, dilations = var_27537, groups = var_21077, pad = var_27539_pad_0, pad_type = var_27539_pad_type_0, strides = var_27535, weight = mid_block_attentions_0_transformer_blocks_6_ff_net_2_weight_to_fp16, x = input_381_cast_fp16)[name = tensor("op_27539_cast_fp16")]; + tensor inputs_187_cast_fp16 = add(x = var_27539_cast_fp16, y = inputs_185_cast_fp16)[name = tensor("inputs_187_cast_fp16")]; + tensor hidden_states_251_axes_0 = const()[name = tensor("hidden_states_251_axes_0"), val = tensor([1])]; + tensor hidden_states_251_gamma_0_to_fp16 = const()[name = tensor("hidden_states_251_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2212699776)))]; + tensor hidden_states_251_beta_0_to_fp16 = const()[name = tensor("hidden_states_251_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2212702400)))]; + tensor var_27555_to_fp16 = const()[name = tensor("op_27555_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_251_cast_fp16 = layer_norm(axes = hidden_states_251_axes_0, beta = hidden_states_251_beta_0_to_fp16, epsilon = var_27555_to_fp16, gamma = hidden_states_251_gamma_0_to_fp16, x = inputs_187_cast_fp16)[name = tensor("hidden_states_251_cast_fp16")]; + tensor var_27570 = const()[name = tensor("op_27570"), val = tensor([1, 1])]; + tensor var_27572 = const()[name = tensor("op_27572"), val = tensor([1, 1])]; + tensor q_125_pad_type_0 = const()[name = tensor("q_125_pad_type_0"), val = tensor("custom")]; + tensor q_125_pad_0 = const()[name = tensor("q_125_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_7_attn1_to_q_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_7_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2212705024)))]; + tensor q_125_cast_fp16 = conv(dilations = var_27572, groups = var_21077, pad = q_125_pad_0, pad_type = q_125_pad_type_0, strides = var_27570, weight = mid_block_attentions_0_transformer_blocks_7_attn1_to_q_weight_to_fp16, x = hidden_states_251_cast_fp16)[name = tensor("q_125_cast_fp16")]; + tensor var_27576 = const()[name = tensor("op_27576"), val = tensor([1, 1])]; + tensor var_27578 = const()[name = tensor("op_27578"), val = tensor([1, 1])]; + tensor k_249_pad_type_0 = const()[name = tensor("k_249_pad_type_0"), val = tensor("custom")]; + tensor k_249_pad_0 = const()[name = tensor("k_249_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_7_attn1_to_k_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_7_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2215981888)))]; + tensor k_249_cast_fp16 = conv(dilations = var_27578, groups = var_21077, pad = k_249_pad_0, pad_type = k_249_pad_type_0, strides = var_27576, weight = mid_block_attentions_0_transformer_blocks_7_attn1_to_k_weight_to_fp16, x = hidden_states_251_cast_fp16)[name = tensor("k_249_cast_fp16")]; + tensor var_27582 = const()[name = tensor("op_27582"), val = tensor([1, 1])]; + tensor var_27584 = const()[name = tensor("op_27584"), val = tensor([1, 1])]; + tensor v_125_pad_type_0 = const()[name = tensor("v_125_pad_type_0"), val = tensor("custom")]; + tensor v_125_pad_0 = const()[name = tensor("v_125_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_7_attn1_to_v_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_7_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2219258752)))]; + tensor v_125_cast_fp16 = conv(dilations = var_27584, groups = var_21077, pad = v_125_pad_0, pad_type = v_125_pad_type_0, strides = var_27582, weight = mid_block_attentions_0_transformer_blocks_7_attn1_to_v_weight_to_fp16, x = hidden_states_251_cast_fp16)[name = tensor("v_125_cast_fp16")]; + tensor var_27588_begin_0 = const()[name = tensor("op_27588_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_27588_end_0 = const()[name = tensor("op_27588_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_27588_end_mask_0 = const()[name = tensor("op_27588_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27588_cast_fp16 = slice_by_index(begin = var_27588_begin_0, end = var_27588_end_0, end_mask = var_27588_end_mask_0, x = q_125_cast_fp16)[name = tensor("op_27588_cast_fp16")]; + tensor var_27592_begin_0 = const()[name = tensor("op_27592_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_27592_end_0 = const()[name = tensor("op_27592_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_27592_end_mask_0 = const()[name = tensor("op_27592_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27592_cast_fp16 = slice_by_index(begin = var_27592_begin_0, end = var_27592_end_0, end_mask = var_27592_end_mask_0, x = q_125_cast_fp16)[name = tensor("op_27592_cast_fp16")]; + tensor var_27596_begin_0 = const()[name = tensor("op_27596_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_27596_end_0 = const()[name = tensor("op_27596_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_27596_end_mask_0 = const()[name = tensor("op_27596_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27596_cast_fp16 = slice_by_index(begin = var_27596_begin_0, end = var_27596_end_0, end_mask = var_27596_end_mask_0, x = q_125_cast_fp16)[name = tensor("op_27596_cast_fp16")]; + tensor var_27600_begin_0 = const()[name = tensor("op_27600_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_27600_end_0 = const()[name = tensor("op_27600_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_27600_end_mask_0 = const()[name = tensor("op_27600_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27600_cast_fp16 = slice_by_index(begin = var_27600_begin_0, end = var_27600_end_0, end_mask = var_27600_end_mask_0, x = q_125_cast_fp16)[name = tensor("op_27600_cast_fp16")]; + tensor var_27604_begin_0 = const()[name = tensor("op_27604_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_27604_end_0 = const()[name = tensor("op_27604_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_27604_end_mask_0 = const()[name = tensor("op_27604_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27604_cast_fp16 = slice_by_index(begin = var_27604_begin_0, end = var_27604_end_0, end_mask = var_27604_end_mask_0, x = q_125_cast_fp16)[name = tensor("op_27604_cast_fp16")]; + tensor var_27608_begin_0 = const()[name = tensor("op_27608_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_27608_end_0 = const()[name = tensor("op_27608_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_27608_end_mask_0 = const()[name = tensor("op_27608_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27608_cast_fp16 = slice_by_index(begin = var_27608_begin_0, end = var_27608_end_0, end_mask = var_27608_end_mask_0, x = q_125_cast_fp16)[name = tensor("op_27608_cast_fp16")]; + tensor var_27612_begin_0 = const()[name = tensor("op_27612_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_27612_end_0 = const()[name = tensor("op_27612_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_27612_end_mask_0 = const()[name = tensor("op_27612_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27612_cast_fp16 = slice_by_index(begin = var_27612_begin_0, end = var_27612_end_0, end_mask = var_27612_end_mask_0, x = q_125_cast_fp16)[name = tensor("op_27612_cast_fp16")]; + tensor var_27616_begin_0 = const()[name = tensor("op_27616_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_27616_end_0 = const()[name = tensor("op_27616_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_27616_end_mask_0 = const()[name = tensor("op_27616_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27616_cast_fp16 = slice_by_index(begin = var_27616_begin_0, end = var_27616_end_0, end_mask = var_27616_end_mask_0, x = q_125_cast_fp16)[name = tensor("op_27616_cast_fp16")]; + tensor var_27620_begin_0 = const()[name = tensor("op_27620_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_27620_end_0 = const()[name = tensor("op_27620_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_27620_end_mask_0 = const()[name = tensor("op_27620_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27620_cast_fp16 = slice_by_index(begin = var_27620_begin_0, end = var_27620_end_0, end_mask = var_27620_end_mask_0, x = q_125_cast_fp16)[name = tensor("op_27620_cast_fp16")]; + tensor var_27624_begin_0 = const()[name = tensor("op_27624_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_27624_end_0 = const()[name = tensor("op_27624_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_27624_end_mask_0 = const()[name = tensor("op_27624_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27624_cast_fp16 = slice_by_index(begin = var_27624_begin_0, end = var_27624_end_0, end_mask = var_27624_end_mask_0, x = q_125_cast_fp16)[name = tensor("op_27624_cast_fp16")]; + tensor var_27628_begin_0 = const()[name = tensor("op_27628_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_27628_end_0 = const()[name = tensor("op_27628_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_27628_end_mask_0 = const()[name = tensor("op_27628_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27628_cast_fp16 = slice_by_index(begin = var_27628_begin_0, end = var_27628_end_0, end_mask = var_27628_end_mask_0, x = q_125_cast_fp16)[name = tensor("op_27628_cast_fp16")]; + tensor var_27632_begin_0 = const()[name = tensor("op_27632_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_27632_end_0 = const()[name = tensor("op_27632_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_27632_end_mask_0 = const()[name = tensor("op_27632_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27632_cast_fp16 = slice_by_index(begin = var_27632_begin_0, end = var_27632_end_0, end_mask = var_27632_end_mask_0, x = q_125_cast_fp16)[name = tensor("op_27632_cast_fp16")]; + tensor var_27636_begin_0 = const()[name = tensor("op_27636_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_27636_end_0 = const()[name = tensor("op_27636_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_27636_end_mask_0 = const()[name = tensor("op_27636_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27636_cast_fp16 = slice_by_index(begin = var_27636_begin_0, end = var_27636_end_0, end_mask = var_27636_end_mask_0, x = q_125_cast_fp16)[name = tensor("op_27636_cast_fp16")]; + tensor var_27640_begin_0 = const()[name = tensor("op_27640_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_27640_end_0 = const()[name = tensor("op_27640_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_27640_end_mask_0 = const()[name = tensor("op_27640_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27640_cast_fp16 = slice_by_index(begin = var_27640_begin_0, end = var_27640_end_0, end_mask = var_27640_end_mask_0, x = q_125_cast_fp16)[name = tensor("op_27640_cast_fp16")]; + tensor var_27644_begin_0 = const()[name = tensor("op_27644_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_27644_end_0 = const()[name = tensor("op_27644_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_27644_end_mask_0 = const()[name = tensor("op_27644_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27644_cast_fp16 = slice_by_index(begin = var_27644_begin_0, end = var_27644_end_0, end_mask = var_27644_end_mask_0, x = q_125_cast_fp16)[name = tensor("op_27644_cast_fp16")]; + tensor var_27648_begin_0 = const()[name = tensor("op_27648_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_27648_end_0 = const()[name = tensor("op_27648_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_27648_end_mask_0 = const()[name = tensor("op_27648_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27648_cast_fp16 = slice_by_index(begin = var_27648_begin_0, end = var_27648_end_0, end_mask = var_27648_end_mask_0, x = q_125_cast_fp16)[name = tensor("op_27648_cast_fp16")]; + tensor var_27652_begin_0 = const()[name = tensor("op_27652_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_27652_end_0 = const()[name = tensor("op_27652_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_27652_end_mask_0 = const()[name = tensor("op_27652_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27652_cast_fp16 = slice_by_index(begin = var_27652_begin_0, end = var_27652_end_0, end_mask = var_27652_end_mask_0, x = q_125_cast_fp16)[name = tensor("op_27652_cast_fp16")]; + tensor var_27656_begin_0 = const()[name = tensor("op_27656_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_27656_end_0 = const()[name = tensor("op_27656_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_27656_end_mask_0 = const()[name = tensor("op_27656_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27656_cast_fp16 = slice_by_index(begin = var_27656_begin_0, end = var_27656_end_0, end_mask = var_27656_end_mask_0, x = q_125_cast_fp16)[name = tensor("op_27656_cast_fp16")]; + tensor var_27660_begin_0 = const()[name = tensor("op_27660_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_27660_end_0 = const()[name = tensor("op_27660_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_27660_end_mask_0 = const()[name = tensor("op_27660_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27660_cast_fp16 = slice_by_index(begin = var_27660_begin_0, end = var_27660_end_0, end_mask = var_27660_end_mask_0, x = q_125_cast_fp16)[name = tensor("op_27660_cast_fp16")]; + tensor var_27664_begin_0 = const()[name = tensor("op_27664_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_27664_end_0 = const()[name = tensor("op_27664_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_27664_end_mask_0 = const()[name = tensor("op_27664_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27664_cast_fp16 = slice_by_index(begin = var_27664_begin_0, end = var_27664_end_0, end_mask = var_27664_end_mask_0, x = q_125_cast_fp16)[name = tensor("op_27664_cast_fp16")]; + tensor k_251_perm_0 = const()[name = tensor("k_251_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_27671_begin_0 = const()[name = tensor("op_27671_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_27671_end_0 = const()[name = tensor("op_27671_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_27671_end_mask_0 = const()[name = tensor("op_27671_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_251_cast_fp16 = transpose(perm = k_251_perm_0, x = k_249_cast_fp16)[name = tensor("transpose_77")]; + tensor var_27671_cast_fp16 = slice_by_index(begin = var_27671_begin_0, end = var_27671_end_0, end_mask = var_27671_end_mask_0, x = k_251_cast_fp16)[name = tensor("op_27671_cast_fp16")]; + tensor var_27675_begin_0 = const()[name = tensor("op_27675_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_27675_end_0 = const()[name = tensor("op_27675_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_27675_end_mask_0 = const()[name = tensor("op_27675_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27675_cast_fp16 = slice_by_index(begin = var_27675_begin_0, end = var_27675_end_0, end_mask = var_27675_end_mask_0, x = k_251_cast_fp16)[name = tensor("op_27675_cast_fp16")]; + tensor var_27679_begin_0 = const()[name = tensor("op_27679_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_27679_end_0 = const()[name = tensor("op_27679_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_27679_end_mask_0 = const()[name = tensor("op_27679_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27679_cast_fp16 = slice_by_index(begin = var_27679_begin_0, end = var_27679_end_0, end_mask = var_27679_end_mask_0, x = k_251_cast_fp16)[name = tensor("op_27679_cast_fp16")]; + tensor var_27683_begin_0 = const()[name = tensor("op_27683_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_27683_end_0 = const()[name = tensor("op_27683_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_27683_end_mask_0 = const()[name = tensor("op_27683_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27683_cast_fp16 = slice_by_index(begin = var_27683_begin_0, end = var_27683_end_0, end_mask = var_27683_end_mask_0, x = k_251_cast_fp16)[name = tensor("op_27683_cast_fp16")]; + tensor var_27687_begin_0 = const()[name = tensor("op_27687_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_27687_end_0 = const()[name = tensor("op_27687_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_27687_end_mask_0 = const()[name = tensor("op_27687_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27687_cast_fp16 = slice_by_index(begin = var_27687_begin_0, end = var_27687_end_0, end_mask = var_27687_end_mask_0, x = k_251_cast_fp16)[name = tensor("op_27687_cast_fp16")]; + tensor var_27691_begin_0 = const()[name = tensor("op_27691_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_27691_end_0 = const()[name = tensor("op_27691_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_27691_end_mask_0 = const()[name = tensor("op_27691_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27691_cast_fp16 = slice_by_index(begin = var_27691_begin_0, end = var_27691_end_0, end_mask = var_27691_end_mask_0, x = k_251_cast_fp16)[name = tensor("op_27691_cast_fp16")]; + tensor var_27695_begin_0 = const()[name = tensor("op_27695_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_27695_end_0 = const()[name = tensor("op_27695_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_27695_end_mask_0 = const()[name = tensor("op_27695_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27695_cast_fp16 = slice_by_index(begin = var_27695_begin_0, end = var_27695_end_0, end_mask = var_27695_end_mask_0, x = k_251_cast_fp16)[name = tensor("op_27695_cast_fp16")]; + tensor var_27699_begin_0 = const()[name = tensor("op_27699_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_27699_end_0 = const()[name = tensor("op_27699_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_27699_end_mask_0 = const()[name = tensor("op_27699_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27699_cast_fp16 = slice_by_index(begin = var_27699_begin_0, end = var_27699_end_0, end_mask = var_27699_end_mask_0, x = k_251_cast_fp16)[name = tensor("op_27699_cast_fp16")]; + tensor var_27703_begin_0 = const()[name = tensor("op_27703_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_27703_end_0 = const()[name = tensor("op_27703_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_27703_end_mask_0 = const()[name = tensor("op_27703_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27703_cast_fp16 = slice_by_index(begin = var_27703_begin_0, end = var_27703_end_0, end_mask = var_27703_end_mask_0, x = k_251_cast_fp16)[name = tensor("op_27703_cast_fp16")]; + tensor var_27707_begin_0 = const()[name = tensor("op_27707_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_27707_end_0 = const()[name = tensor("op_27707_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_27707_end_mask_0 = const()[name = tensor("op_27707_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27707_cast_fp16 = slice_by_index(begin = var_27707_begin_0, end = var_27707_end_0, end_mask = var_27707_end_mask_0, x = k_251_cast_fp16)[name = tensor("op_27707_cast_fp16")]; + tensor var_27711_begin_0 = const()[name = tensor("op_27711_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_27711_end_0 = const()[name = tensor("op_27711_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_27711_end_mask_0 = const()[name = tensor("op_27711_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27711_cast_fp16 = slice_by_index(begin = var_27711_begin_0, end = var_27711_end_0, end_mask = var_27711_end_mask_0, x = k_251_cast_fp16)[name = tensor("op_27711_cast_fp16")]; + tensor var_27715_begin_0 = const()[name = tensor("op_27715_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_27715_end_0 = const()[name = tensor("op_27715_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_27715_end_mask_0 = const()[name = tensor("op_27715_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27715_cast_fp16 = slice_by_index(begin = var_27715_begin_0, end = var_27715_end_0, end_mask = var_27715_end_mask_0, x = k_251_cast_fp16)[name = tensor("op_27715_cast_fp16")]; + tensor var_27719_begin_0 = const()[name = tensor("op_27719_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_27719_end_0 = const()[name = tensor("op_27719_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_27719_end_mask_0 = const()[name = tensor("op_27719_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27719_cast_fp16 = slice_by_index(begin = var_27719_begin_0, end = var_27719_end_0, end_mask = var_27719_end_mask_0, x = k_251_cast_fp16)[name = tensor("op_27719_cast_fp16")]; + tensor var_27723_begin_0 = const()[name = tensor("op_27723_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_27723_end_0 = const()[name = tensor("op_27723_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_27723_end_mask_0 = const()[name = tensor("op_27723_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27723_cast_fp16 = slice_by_index(begin = var_27723_begin_0, end = var_27723_end_0, end_mask = var_27723_end_mask_0, x = k_251_cast_fp16)[name = tensor("op_27723_cast_fp16")]; + tensor var_27727_begin_0 = const()[name = tensor("op_27727_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_27727_end_0 = const()[name = tensor("op_27727_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_27727_end_mask_0 = const()[name = tensor("op_27727_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27727_cast_fp16 = slice_by_index(begin = var_27727_begin_0, end = var_27727_end_0, end_mask = var_27727_end_mask_0, x = k_251_cast_fp16)[name = tensor("op_27727_cast_fp16")]; + tensor var_27731_begin_0 = const()[name = tensor("op_27731_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_27731_end_0 = const()[name = tensor("op_27731_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_27731_end_mask_0 = const()[name = tensor("op_27731_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27731_cast_fp16 = slice_by_index(begin = var_27731_begin_0, end = var_27731_end_0, end_mask = var_27731_end_mask_0, x = k_251_cast_fp16)[name = tensor("op_27731_cast_fp16")]; + tensor var_27735_begin_0 = const()[name = tensor("op_27735_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_27735_end_0 = const()[name = tensor("op_27735_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_27735_end_mask_0 = const()[name = tensor("op_27735_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27735_cast_fp16 = slice_by_index(begin = var_27735_begin_0, end = var_27735_end_0, end_mask = var_27735_end_mask_0, x = k_251_cast_fp16)[name = tensor("op_27735_cast_fp16")]; + tensor var_27739_begin_0 = const()[name = tensor("op_27739_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_27739_end_0 = const()[name = tensor("op_27739_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_27739_end_mask_0 = const()[name = tensor("op_27739_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27739_cast_fp16 = slice_by_index(begin = var_27739_begin_0, end = var_27739_end_0, end_mask = var_27739_end_mask_0, x = k_251_cast_fp16)[name = tensor("op_27739_cast_fp16")]; + tensor var_27743_begin_0 = const()[name = tensor("op_27743_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_27743_end_0 = const()[name = tensor("op_27743_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_27743_end_mask_0 = const()[name = tensor("op_27743_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27743_cast_fp16 = slice_by_index(begin = var_27743_begin_0, end = var_27743_end_0, end_mask = var_27743_end_mask_0, x = k_251_cast_fp16)[name = tensor("op_27743_cast_fp16")]; + tensor var_27747_begin_0 = const()[name = tensor("op_27747_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_27747_end_0 = const()[name = tensor("op_27747_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_27747_end_mask_0 = const()[name = tensor("op_27747_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27747_cast_fp16 = slice_by_index(begin = var_27747_begin_0, end = var_27747_end_0, end_mask = var_27747_end_mask_0, x = k_251_cast_fp16)[name = tensor("op_27747_cast_fp16")]; + tensor var_27749_begin_0 = const()[name = tensor("op_27749_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_27749_end_0 = const()[name = tensor("op_27749_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_27749_end_mask_0 = const()[name = tensor("op_27749_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27749_cast_fp16 = slice_by_index(begin = var_27749_begin_0, end = var_27749_end_0, end_mask = var_27749_end_mask_0, x = v_125_cast_fp16)[name = tensor("op_27749_cast_fp16")]; + tensor var_27753_begin_0 = const()[name = tensor("op_27753_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_27753_end_0 = const()[name = tensor("op_27753_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_27753_end_mask_0 = const()[name = tensor("op_27753_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27753_cast_fp16 = slice_by_index(begin = var_27753_begin_0, end = var_27753_end_0, end_mask = var_27753_end_mask_0, x = v_125_cast_fp16)[name = tensor("op_27753_cast_fp16")]; + tensor var_27757_begin_0 = const()[name = tensor("op_27757_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_27757_end_0 = const()[name = tensor("op_27757_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_27757_end_mask_0 = const()[name = tensor("op_27757_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27757_cast_fp16 = slice_by_index(begin = var_27757_begin_0, end = var_27757_end_0, end_mask = var_27757_end_mask_0, x = v_125_cast_fp16)[name = tensor("op_27757_cast_fp16")]; + tensor var_27761_begin_0 = const()[name = tensor("op_27761_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_27761_end_0 = const()[name = tensor("op_27761_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_27761_end_mask_0 = const()[name = tensor("op_27761_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27761_cast_fp16 = slice_by_index(begin = var_27761_begin_0, end = var_27761_end_0, end_mask = var_27761_end_mask_0, x = v_125_cast_fp16)[name = tensor("op_27761_cast_fp16")]; + tensor var_27765_begin_0 = const()[name = tensor("op_27765_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_27765_end_0 = const()[name = tensor("op_27765_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_27765_end_mask_0 = const()[name = tensor("op_27765_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27765_cast_fp16 = slice_by_index(begin = var_27765_begin_0, end = var_27765_end_0, end_mask = var_27765_end_mask_0, x = v_125_cast_fp16)[name = tensor("op_27765_cast_fp16")]; + tensor var_27769_begin_0 = const()[name = tensor("op_27769_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_27769_end_0 = const()[name = tensor("op_27769_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_27769_end_mask_0 = const()[name = tensor("op_27769_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27769_cast_fp16 = slice_by_index(begin = var_27769_begin_0, end = var_27769_end_0, end_mask = var_27769_end_mask_0, x = v_125_cast_fp16)[name = tensor("op_27769_cast_fp16")]; + tensor var_27773_begin_0 = const()[name = tensor("op_27773_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_27773_end_0 = const()[name = tensor("op_27773_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_27773_end_mask_0 = const()[name = tensor("op_27773_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27773_cast_fp16 = slice_by_index(begin = var_27773_begin_0, end = var_27773_end_0, end_mask = var_27773_end_mask_0, x = v_125_cast_fp16)[name = tensor("op_27773_cast_fp16")]; + tensor var_27777_begin_0 = const()[name = tensor("op_27777_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_27777_end_0 = const()[name = tensor("op_27777_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_27777_end_mask_0 = const()[name = tensor("op_27777_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27777_cast_fp16 = slice_by_index(begin = var_27777_begin_0, end = var_27777_end_0, end_mask = var_27777_end_mask_0, x = v_125_cast_fp16)[name = tensor("op_27777_cast_fp16")]; + tensor var_27781_begin_0 = const()[name = tensor("op_27781_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_27781_end_0 = const()[name = tensor("op_27781_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_27781_end_mask_0 = const()[name = tensor("op_27781_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27781_cast_fp16 = slice_by_index(begin = var_27781_begin_0, end = var_27781_end_0, end_mask = var_27781_end_mask_0, x = v_125_cast_fp16)[name = tensor("op_27781_cast_fp16")]; + tensor var_27785_begin_0 = const()[name = tensor("op_27785_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_27785_end_0 = const()[name = tensor("op_27785_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_27785_end_mask_0 = const()[name = tensor("op_27785_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27785_cast_fp16 = slice_by_index(begin = var_27785_begin_0, end = var_27785_end_0, end_mask = var_27785_end_mask_0, x = v_125_cast_fp16)[name = tensor("op_27785_cast_fp16")]; + tensor var_27789_begin_0 = const()[name = tensor("op_27789_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_27789_end_0 = const()[name = tensor("op_27789_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_27789_end_mask_0 = const()[name = tensor("op_27789_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27789_cast_fp16 = slice_by_index(begin = var_27789_begin_0, end = var_27789_end_0, end_mask = var_27789_end_mask_0, x = v_125_cast_fp16)[name = tensor("op_27789_cast_fp16")]; + tensor var_27793_begin_0 = const()[name = tensor("op_27793_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_27793_end_0 = const()[name = tensor("op_27793_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_27793_end_mask_0 = const()[name = tensor("op_27793_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27793_cast_fp16 = slice_by_index(begin = var_27793_begin_0, end = var_27793_end_0, end_mask = var_27793_end_mask_0, x = v_125_cast_fp16)[name = tensor("op_27793_cast_fp16")]; + tensor var_27797_begin_0 = const()[name = tensor("op_27797_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_27797_end_0 = const()[name = tensor("op_27797_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_27797_end_mask_0 = const()[name = tensor("op_27797_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27797_cast_fp16 = slice_by_index(begin = var_27797_begin_0, end = var_27797_end_0, end_mask = var_27797_end_mask_0, x = v_125_cast_fp16)[name = tensor("op_27797_cast_fp16")]; + tensor var_27801_begin_0 = const()[name = tensor("op_27801_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_27801_end_0 = const()[name = tensor("op_27801_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_27801_end_mask_0 = const()[name = tensor("op_27801_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27801_cast_fp16 = slice_by_index(begin = var_27801_begin_0, end = var_27801_end_0, end_mask = var_27801_end_mask_0, x = v_125_cast_fp16)[name = tensor("op_27801_cast_fp16")]; + tensor var_27805_begin_0 = const()[name = tensor("op_27805_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_27805_end_0 = const()[name = tensor("op_27805_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_27805_end_mask_0 = const()[name = tensor("op_27805_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27805_cast_fp16 = slice_by_index(begin = var_27805_begin_0, end = var_27805_end_0, end_mask = var_27805_end_mask_0, x = v_125_cast_fp16)[name = tensor("op_27805_cast_fp16")]; + tensor var_27809_begin_0 = const()[name = tensor("op_27809_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_27809_end_0 = const()[name = tensor("op_27809_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_27809_end_mask_0 = const()[name = tensor("op_27809_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27809_cast_fp16 = slice_by_index(begin = var_27809_begin_0, end = var_27809_end_0, end_mask = var_27809_end_mask_0, x = v_125_cast_fp16)[name = tensor("op_27809_cast_fp16")]; + tensor var_27813_begin_0 = const()[name = tensor("op_27813_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_27813_end_0 = const()[name = tensor("op_27813_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_27813_end_mask_0 = const()[name = tensor("op_27813_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27813_cast_fp16 = slice_by_index(begin = var_27813_begin_0, end = var_27813_end_0, end_mask = var_27813_end_mask_0, x = v_125_cast_fp16)[name = tensor("op_27813_cast_fp16")]; + tensor var_27817_begin_0 = const()[name = tensor("op_27817_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_27817_end_0 = const()[name = tensor("op_27817_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_27817_end_mask_0 = const()[name = tensor("op_27817_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27817_cast_fp16 = slice_by_index(begin = var_27817_begin_0, end = var_27817_end_0, end_mask = var_27817_end_mask_0, x = v_125_cast_fp16)[name = tensor("op_27817_cast_fp16")]; + tensor var_27821_begin_0 = const()[name = tensor("op_27821_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_27821_end_0 = const()[name = tensor("op_27821_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_27821_end_mask_0 = const()[name = tensor("op_27821_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27821_cast_fp16 = slice_by_index(begin = var_27821_begin_0, end = var_27821_end_0, end_mask = var_27821_end_mask_0, x = v_125_cast_fp16)[name = tensor("op_27821_cast_fp16")]; + tensor var_27825_begin_0 = const()[name = tensor("op_27825_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_27825_end_0 = const()[name = tensor("op_27825_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_27825_end_mask_0 = const()[name = tensor("op_27825_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27825_cast_fp16 = slice_by_index(begin = var_27825_begin_0, end = var_27825_end_0, end_mask = var_27825_end_mask_0, x = v_125_cast_fp16)[name = tensor("op_27825_cast_fp16")]; + tensor var_27829_equation_0 = const()[name = tensor("op_27829_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27829_cast_fp16 = einsum(equation = var_27829_equation_0, values = (var_27671_cast_fp16, var_27588_cast_fp16))[name = tensor("op_27829_cast_fp16")]; + tensor var_27830_to_fp16 = const()[name = tensor("op_27830_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2321_cast_fp16 = mul(x = var_27829_cast_fp16, y = var_27830_to_fp16)[name = tensor("aw_2321_cast_fp16")]; + tensor var_27833_equation_0 = const()[name = tensor("op_27833_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27833_cast_fp16 = einsum(equation = var_27833_equation_0, values = (var_27675_cast_fp16, var_27592_cast_fp16))[name = tensor("op_27833_cast_fp16")]; + tensor var_27834_to_fp16 = const()[name = tensor("op_27834_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2323_cast_fp16 = mul(x = var_27833_cast_fp16, y = var_27834_to_fp16)[name = tensor("aw_2323_cast_fp16")]; + tensor var_27837_equation_0 = const()[name = tensor("op_27837_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27837_cast_fp16 = einsum(equation = var_27837_equation_0, values = (var_27679_cast_fp16, var_27596_cast_fp16))[name = tensor("op_27837_cast_fp16")]; + tensor var_27838_to_fp16 = const()[name = tensor("op_27838_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2325_cast_fp16 = mul(x = var_27837_cast_fp16, y = var_27838_to_fp16)[name = tensor("aw_2325_cast_fp16")]; + tensor var_27841_equation_0 = const()[name = tensor("op_27841_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27841_cast_fp16 = einsum(equation = var_27841_equation_0, values = (var_27683_cast_fp16, var_27600_cast_fp16))[name = tensor("op_27841_cast_fp16")]; + tensor var_27842_to_fp16 = const()[name = tensor("op_27842_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2327_cast_fp16 = mul(x = var_27841_cast_fp16, y = var_27842_to_fp16)[name = tensor("aw_2327_cast_fp16")]; + tensor var_27845_equation_0 = const()[name = tensor("op_27845_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27845_cast_fp16 = einsum(equation = var_27845_equation_0, values = (var_27687_cast_fp16, var_27604_cast_fp16))[name = tensor("op_27845_cast_fp16")]; + tensor var_27846_to_fp16 = const()[name = tensor("op_27846_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2329_cast_fp16 = mul(x = var_27845_cast_fp16, y = var_27846_to_fp16)[name = tensor("aw_2329_cast_fp16")]; + tensor var_27849_equation_0 = const()[name = tensor("op_27849_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27849_cast_fp16 = einsum(equation = var_27849_equation_0, values = (var_27691_cast_fp16, var_27608_cast_fp16))[name = tensor("op_27849_cast_fp16")]; + tensor var_27850_to_fp16 = const()[name = tensor("op_27850_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2331_cast_fp16 = mul(x = var_27849_cast_fp16, y = var_27850_to_fp16)[name = tensor("aw_2331_cast_fp16")]; + tensor var_27853_equation_0 = const()[name = tensor("op_27853_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27853_cast_fp16 = einsum(equation = var_27853_equation_0, values = (var_27695_cast_fp16, var_27612_cast_fp16))[name = tensor("op_27853_cast_fp16")]; + tensor var_27854_to_fp16 = const()[name = tensor("op_27854_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2333_cast_fp16 = mul(x = var_27853_cast_fp16, y = var_27854_to_fp16)[name = tensor("aw_2333_cast_fp16")]; + tensor var_27857_equation_0 = const()[name = tensor("op_27857_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27857_cast_fp16 = einsum(equation = var_27857_equation_0, values = (var_27699_cast_fp16, var_27616_cast_fp16))[name = tensor("op_27857_cast_fp16")]; + tensor var_27858_to_fp16 = const()[name = tensor("op_27858_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2335_cast_fp16 = mul(x = var_27857_cast_fp16, y = var_27858_to_fp16)[name = tensor("aw_2335_cast_fp16")]; + tensor var_27861_equation_0 = const()[name = tensor("op_27861_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27861_cast_fp16 = einsum(equation = var_27861_equation_0, values = (var_27703_cast_fp16, var_27620_cast_fp16))[name = tensor("op_27861_cast_fp16")]; + tensor var_27862_to_fp16 = const()[name = tensor("op_27862_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2337_cast_fp16 = mul(x = var_27861_cast_fp16, y = var_27862_to_fp16)[name = tensor("aw_2337_cast_fp16")]; + tensor var_27865_equation_0 = const()[name = tensor("op_27865_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27865_cast_fp16 = einsum(equation = var_27865_equation_0, values = (var_27707_cast_fp16, var_27624_cast_fp16))[name = tensor("op_27865_cast_fp16")]; + tensor var_27866_to_fp16 = const()[name = tensor("op_27866_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2339_cast_fp16 = mul(x = var_27865_cast_fp16, y = var_27866_to_fp16)[name = tensor("aw_2339_cast_fp16")]; + tensor var_27869_equation_0 = const()[name = tensor("op_27869_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27869_cast_fp16 = einsum(equation = var_27869_equation_0, values = (var_27711_cast_fp16, var_27628_cast_fp16))[name = tensor("op_27869_cast_fp16")]; + tensor var_27870_to_fp16 = const()[name = tensor("op_27870_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2341_cast_fp16 = mul(x = var_27869_cast_fp16, y = var_27870_to_fp16)[name = tensor("aw_2341_cast_fp16")]; + tensor var_27873_equation_0 = const()[name = tensor("op_27873_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27873_cast_fp16 = einsum(equation = var_27873_equation_0, values = (var_27715_cast_fp16, var_27632_cast_fp16))[name = tensor("op_27873_cast_fp16")]; + tensor var_27874_to_fp16 = const()[name = tensor("op_27874_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2343_cast_fp16 = mul(x = var_27873_cast_fp16, y = var_27874_to_fp16)[name = tensor("aw_2343_cast_fp16")]; + tensor var_27877_equation_0 = const()[name = tensor("op_27877_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27877_cast_fp16 = einsum(equation = var_27877_equation_0, values = (var_27719_cast_fp16, var_27636_cast_fp16))[name = tensor("op_27877_cast_fp16")]; + tensor var_27878_to_fp16 = const()[name = tensor("op_27878_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2345_cast_fp16 = mul(x = var_27877_cast_fp16, y = var_27878_to_fp16)[name = tensor("aw_2345_cast_fp16")]; + tensor var_27881_equation_0 = const()[name = tensor("op_27881_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27881_cast_fp16 = einsum(equation = var_27881_equation_0, values = (var_27723_cast_fp16, var_27640_cast_fp16))[name = tensor("op_27881_cast_fp16")]; + tensor var_27882_to_fp16 = const()[name = tensor("op_27882_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2347_cast_fp16 = mul(x = var_27881_cast_fp16, y = var_27882_to_fp16)[name = tensor("aw_2347_cast_fp16")]; + tensor var_27885_equation_0 = const()[name = tensor("op_27885_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27885_cast_fp16 = einsum(equation = var_27885_equation_0, values = (var_27727_cast_fp16, var_27644_cast_fp16))[name = tensor("op_27885_cast_fp16")]; + tensor var_27886_to_fp16 = const()[name = tensor("op_27886_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2349_cast_fp16 = mul(x = var_27885_cast_fp16, y = var_27886_to_fp16)[name = tensor("aw_2349_cast_fp16")]; + tensor var_27889_equation_0 = const()[name = tensor("op_27889_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27889_cast_fp16 = einsum(equation = var_27889_equation_0, values = (var_27731_cast_fp16, var_27648_cast_fp16))[name = tensor("op_27889_cast_fp16")]; + tensor var_27890_to_fp16 = const()[name = tensor("op_27890_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2351_cast_fp16 = mul(x = var_27889_cast_fp16, y = var_27890_to_fp16)[name = tensor("aw_2351_cast_fp16")]; + tensor var_27893_equation_0 = const()[name = tensor("op_27893_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27893_cast_fp16 = einsum(equation = var_27893_equation_0, values = (var_27735_cast_fp16, var_27652_cast_fp16))[name = tensor("op_27893_cast_fp16")]; + tensor var_27894_to_fp16 = const()[name = tensor("op_27894_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2353_cast_fp16 = mul(x = var_27893_cast_fp16, y = var_27894_to_fp16)[name = tensor("aw_2353_cast_fp16")]; + tensor var_27897_equation_0 = const()[name = tensor("op_27897_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27897_cast_fp16 = einsum(equation = var_27897_equation_0, values = (var_27739_cast_fp16, var_27656_cast_fp16))[name = tensor("op_27897_cast_fp16")]; + tensor var_27898_to_fp16 = const()[name = tensor("op_27898_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2355_cast_fp16 = mul(x = var_27897_cast_fp16, y = var_27898_to_fp16)[name = tensor("aw_2355_cast_fp16")]; + tensor var_27901_equation_0 = const()[name = tensor("op_27901_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27901_cast_fp16 = einsum(equation = var_27901_equation_0, values = (var_27743_cast_fp16, var_27660_cast_fp16))[name = tensor("op_27901_cast_fp16")]; + tensor var_27902_to_fp16 = const()[name = tensor("op_27902_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2357_cast_fp16 = mul(x = var_27901_cast_fp16, y = var_27902_to_fp16)[name = tensor("aw_2357_cast_fp16")]; + tensor var_27905_equation_0 = const()[name = tensor("op_27905_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27905_cast_fp16 = einsum(equation = var_27905_equation_0, values = (var_27747_cast_fp16, var_27664_cast_fp16))[name = tensor("op_27905_cast_fp16")]; + tensor var_27906_to_fp16 = const()[name = tensor("op_27906_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2359_cast_fp16 = mul(x = var_27905_cast_fp16, y = var_27906_to_fp16)[name = tensor("aw_2359_cast_fp16")]; + tensor var_27908_cast_fp16 = softmax(axis = var_21077, x = aw_2321_cast_fp16)[name = tensor("op_27908_cast_fp16")]; + tensor var_27909_cast_fp16 = softmax(axis = var_21077, x = aw_2323_cast_fp16)[name = tensor("op_27909_cast_fp16")]; + tensor var_27910_cast_fp16 = softmax(axis = var_21077, x = aw_2325_cast_fp16)[name = tensor("op_27910_cast_fp16")]; + tensor var_27911_cast_fp16 = softmax(axis = var_21077, x = aw_2327_cast_fp16)[name = tensor("op_27911_cast_fp16")]; + tensor var_27912_cast_fp16 = softmax(axis = var_21077, x = aw_2329_cast_fp16)[name = tensor("op_27912_cast_fp16")]; + tensor var_27913_cast_fp16 = softmax(axis = var_21077, x = aw_2331_cast_fp16)[name = tensor("op_27913_cast_fp16")]; + tensor var_27914_cast_fp16 = softmax(axis = var_21077, x = aw_2333_cast_fp16)[name = tensor("op_27914_cast_fp16")]; + tensor var_27915_cast_fp16 = softmax(axis = var_21077, x = aw_2335_cast_fp16)[name = tensor("op_27915_cast_fp16")]; + tensor var_27916_cast_fp16 = softmax(axis = var_21077, x = aw_2337_cast_fp16)[name = tensor("op_27916_cast_fp16")]; + tensor var_27917_cast_fp16 = softmax(axis = var_21077, x = aw_2339_cast_fp16)[name = tensor("op_27917_cast_fp16")]; + tensor var_27918_cast_fp16 = softmax(axis = var_21077, x = aw_2341_cast_fp16)[name = tensor("op_27918_cast_fp16")]; + tensor var_27919_cast_fp16 = softmax(axis = var_21077, x = aw_2343_cast_fp16)[name = tensor("op_27919_cast_fp16")]; + tensor var_27920_cast_fp16 = softmax(axis = var_21077, x = aw_2345_cast_fp16)[name = tensor("op_27920_cast_fp16")]; + tensor var_27921_cast_fp16 = softmax(axis = var_21077, x = aw_2347_cast_fp16)[name = tensor("op_27921_cast_fp16")]; + tensor var_27922_cast_fp16 = softmax(axis = var_21077, x = aw_2349_cast_fp16)[name = tensor("op_27922_cast_fp16")]; + tensor var_27923_cast_fp16 = softmax(axis = var_21077, x = aw_2351_cast_fp16)[name = tensor("op_27923_cast_fp16")]; + tensor var_27924_cast_fp16 = softmax(axis = var_21077, x = aw_2353_cast_fp16)[name = tensor("op_27924_cast_fp16")]; + tensor var_27925_cast_fp16 = softmax(axis = var_21077, x = aw_2355_cast_fp16)[name = tensor("op_27925_cast_fp16")]; + tensor var_27926_cast_fp16 = softmax(axis = var_21077, x = aw_2357_cast_fp16)[name = tensor("op_27926_cast_fp16")]; + tensor var_27927_cast_fp16 = softmax(axis = var_21077, x = aw_2359_cast_fp16)[name = tensor("op_27927_cast_fp16")]; + tensor var_27929_equation_0 = const()[name = tensor("op_27929_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27929_cast_fp16 = einsum(equation = var_27929_equation_0, values = (var_27749_cast_fp16, var_27908_cast_fp16))[name = tensor("op_27929_cast_fp16")]; + tensor var_27931_equation_0 = const()[name = tensor("op_27931_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27931_cast_fp16 = einsum(equation = var_27931_equation_0, values = (var_27753_cast_fp16, var_27909_cast_fp16))[name = tensor("op_27931_cast_fp16")]; + tensor var_27933_equation_0 = const()[name = tensor("op_27933_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27933_cast_fp16 = einsum(equation = var_27933_equation_0, values = (var_27757_cast_fp16, var_27910_cast_fp16))[name = tensor("op_27933_cast_fp16")]; + tensor var_27935_equation_0 = const()[name = tensor("op_27935_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27935_cast_fp16 = einsum(equation = var_27935_equation_0, values = (var_27761_cast_fp16, var_27911_cast_fp16))[name = tensor("op_27935_cast_fp16")]; + tensor var_27937_equation_0 = const()[name = tensor("op_27937_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27937_cast_fp16 = einsum(equation = var_27937_equation_0, values = (var_27765_cast_fp16, var_27912_cast_fp16))[name = tensor("op_27937_cast_fp16")]; + tensor var_27939_equation_0 = const()[name = tensor("op_27939_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27939_cast_fp16 = einsum(equation = var_27939_equation_0, values = (var_27769_cast_fp16, var_27913_cast_fp16))[name = tensor("op_27939_cast_fp16")]; + tensor var_27941_equation_0 = const()[name = tensor("op_27941_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27941_cast_fp16 = einsum(equation = var_27941_equation_0, values = (var_27773_cast_fp16, var_27914_cast_fp16))[name = tensor("op_27941_cast_fp16")]; + tensor var_27943_equation_0 = const()[name = tensor("op_27943_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27943_cast_fp16 = einsum(equation = var_27943_equation_0, values = (var_27777_cast_fp16, var_27915_cast_fp16))[name = tensor("op_27943_cast_fp16")]; + tensor var_27945_equation_0 = const()[name = tensor("op_27945_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27945_cast_fp16 = einsum(equation = var_27945_equation_0, values = (var_27781_cast_fp16, var_27916_cast_fp16))[name = tensor("op_27945_cast_fp16")]; + tensor var_27947_equation_0 = const()[name = tensor("op_27947_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27947_cast_fp16 = einsum(equation = var_27947_equation_0, values = (var_27785_cast_fp16, var_27917_cast_fp16))[name = tensor("op_27947_cast_fp16")]; + tensor var_27949_equation_0 = const()[name = tensor("op_27949_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27949_cast_fp16 = einsum(equation = var_27949_equation_0, values = (var_27789_cast_fp16, var_27918_cast_fp16))[name = tensor("op_27949_cast_fp16")]; + tensor var_27951_equation_0 = const()[name = tensor("op_27951_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27951_cast_fp16 = einsum(equation = var_27951_equation_0, values = (var_27793_cast_fp16, var_27919_cast_fp16))[name = tensor("op_27951_cast_fp16")]; + tensor var_27953_equation_0 = const()[name = tensor("op_27953_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27953_cast_fp16 = einsum(equation = var_27953_equation_0, values = (var_27797_cast_fp16, var_27920_cast_fp16))[name = tensor("op_27953_cast_fp16")]; + tensor var_27955_equation_0 = const()[name = tensor("op_27955_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27955_cast_fp16 = einsum(equation = var_27955_equation_0, values = (var_27801_cast_fp16, var_27921_cast_fp16))[name = tensor("op_27955_cast_fp16")]; + tensor var_27957_equation_0 = const()[name = tensor("op_27957_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27957_cast_fp16 = einsum(equation = var_27957_equation_0, values = (var_27805_cast_fp16, var_27922_cast_fp16))[name = tensor("op_27957_cast_fp16")]; + tensor var_27959_equation_0 = const()[name = tensor("op_27959_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27959_cast_fp16 = einsum(equation = var_27959_equation_0, values = (var_27809_cast_fp16, var_27923_cast_fp16))[name = tensor("op_27959_cast_fp16")]; + tensor var_27961_equation_0 = const()[name = tensor("op_27961_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27961_cast_fp16 = einsum(equation = var_27961_equation_0, values = (var_27813_cast_fp16, var_27924_cast_fp16))[name = tensor("op_27961_cast_fp16")]; + tensor var_27963_equation_0 = const()[name = tensor("op_27963_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27963_cast_fp16 = einsum(equation = var_27963_equation_0, values = (var_27817_cast_fp16, var_27925_cast_fp16))[name = tensor("op_27963_cast_fp16")]; + tensor var_27965_equation_0 = const()[name = tensor("op_27965_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27965_cast_fp16 = einsum(equation = var_27965_equation_0, values = (var_27821_cast_fp16, var_27926_cast_fp16))[name = tensor("op_27965_cast_fp16")]; + tensor var_27967_equation_0 = const()[name = tensor("op_27967_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27967_cast_fp16 = einsum(equation = var_27967_equation_0, values = (var_27825_cast_fp16, var_27927_cast_fp16))[name = tensor("op_27967_cast_fp16")]; + tensor input_383_interleave_0 = const()[name = tensor("input_383_interleave_0"), val = tensor(false)]; + tensor input_383_cast_fp16 = concat(axis = var_21077, interleave = input_383_interleave_0, values = (var_27929_cast_fp16, var_27931_cast_fp16, var_27933_cast_fp16, var_27935_cast_fp16, var_27937_cast_fp16, var_27939_cast_fp16, var_27941_cast_fp16, var_27943_cast_fp16, var_27945_cast_fp16, var_27947_cast_fp16, var_27949_cast_fp16, var_27951_cast_fp16, var_27953_cast_fp16, var_27955_cast_fp16, var_27957_cast_fp16, var_27959_cast_fp16, var_27961_cast_fp16, var_27963_cast_fp16, var_27965_cast_fp16, var_27967_cast_fp16))[name = tensor("input_383_cast_fp16")]; + tensor var_27973 = const()[name = tensor("op_27973"), val = tensor([1, 1])]; + tensor var_27975 = const()[name = tensor("op_27975"), val = tensor([1, 1])]; + tensor var_27977_pad_type_0 = const()[name = tensor("op_27977_pad_type_0"), val = tensor("custom")]; + tensor var_27977_pad_0 = const()[name = tensor("op_27977_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_7_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_7_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2222535616)))]; + tensor mid_block_attentions_0_transformer_blocks_7_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_7_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2225812480)))]; + tensor var_27977_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_7_attn1_to_out_0_bias_to_fp16, dilations = var_27975, groups = var_21077, pad = var_27977_pad_0, pad_type = var_27977_pad_type_0, strides = var_27973, weight = mid_block_attentions_0_transformer_blocks_7_attn1_to_out_0_weight_to_fp16, x = input_383_cast_fp16)[name = tensor("op_27977_cast_fp16")]; + tensor inputs_189_cast_fp16 = add(x = var_27977_cast_fp16, y = inputs_187_cast_fp16)[name = tensor("inputs_189_cast_fp16")]; + tensor hidden_states_253_axes_0 = const()[name = tensor("hidden_states_253_axes_0"), val = tensor([1])]; + tensor hidden_states_253_gamma_0_to_fp16 = const()[name = tensor("hidden_states_253_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2225815104)))]; + tensor hidden_states_253_beta_0_to_fp16 = const()[name = tensor("hidden_states_253_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2225817728)))]; + tensor var_27987_to_fp16 = const()[name = tensor("op_27987_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_253_cast_fp16 = layer_norm(axes = hidden_states_253_axes_0, beta = hidden_states_253_beta_0_to_fp16, epsilon = var_27987_to_fp16, gamma = hidden_states_253_gamma_0_to_fp16, x = inputs_189_cast_fp16)[name = tensor("hidden_states_253_cast_fp16")]; + tensor var_28002 = const()[name = tensor("op_28002"), val = tensor([1, 1])]; + tensor var_28004 = const()[name = tensor("op_28004"), val = tensor([1, 1])]; + tensor q_127_pad_type_0 = const()[name = tensor("q_127_pad_type_0"), val = tensor("custom")]; + tensor q_127_pad_0 = const()[name = tensor("q_127_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_7_attn2_to_q_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_7_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2225820352)))]; + tensor q_127_cast_fp16 = conv(dilations = var_28004, groups = var_21077, pad = q_127_pad_0, pad_type = q_127_pad_type_0, strides = var_28002, weight = mid_block_attentions_0_transformer_blocks_7_attn2_to_q_weight_to_fp16, x = hidden_states_253_cast_fp16)[name = tensor("q_127_cast_fp16")]; + tensor var_28008 = const()[name = tensor("op_28008"), val = tensor([1, 1])]; + tensor var_28010 = const()[name = tensor("op_28010"), val = tensor([1, 1])]; + tensor k_253_pad_type_0 = const()[name = tensor("k_253_pad_type_0"), val = tensor("custom")]; + tensor k_253_pad_0 = const()[name = tensor("k_253_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_7_attn2_to_k_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_7_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2229097216)))]; + tensor k_253_cast_fp16 = conv(dilations = var_28010, groups = var_21077, pad = k_253_pad_0, pad_type = k_253_pad_type_0, strides = var_28008, weight = mid_block_attentions_0_transformer_blocks_7_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_253_cast_fp16")]; + tensor var_28014 = const()[name = tensor("op_28014"), val = tensor([1, 1])]; + tensor var_28016 = const()[name = tensor("op_28016"), val = tensor([1, 1])]; + tensor v_127_pad_type_0 = const()[name = tensor("v_127_pad_type_0"), val = tensor("custom")]; + tensor v_127_pad_0 = const()[name = tensor("v_127_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_7_attn2_to_v_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_7_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2234340160)))]; + tensor v_127_cast_fp16 = conv(dilations = var_28016, groups = var_21077, pad = v_127_pad_0, pad_type = v_127_pad_type_0, strides = var_28014, weight = mid_block_attentions_0_transformer_blocks_7_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_127_cast_fp16")]; + tensor var_28020_begin_0 = const()[name = tensor("op_28020_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28020_end_0 = const()[name = tensor("op_28020_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_28020_end_mask_0 = const()[name = tensor("op_28020_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28020_cast_fp16 = slice_by_index(begin = var_28020_begin_0, end = var_28020_end_0, end_mask = var_28020_end_mask_0, x = q_127_cast_fp16)[name = tensor("op_28020_cast_fp16")]; + tensor var_28024_begin_0 = const()[name = tensor("op_28024_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_28024_end_0 = const()[name = tensor("op_28024_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_28024_end_mask_0 = const()[name = tensor("op_28024_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28024_cast_fp16 = slice_by_index(begin = var_28024_begin_0, end = var_28024_end_0, end_mask = var_28024_end_mask_0, x = q_127_cast_fp16)[name = tensor("op_28024_cast_fp16")]; + tensor var_28028_begin_0 = const()[name = tensor("op_28028_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_28028_end_0 = const()[name = tensor("op_28028_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_28028_end_mask_0 = const()[name = tensor("op_28028_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28028_cast_fp16 = slice_by_index(begin = var_28028_begin_0, end = var_28028_end_0, end_mask = var_28028_end_mask_0, x = q_127_cast_fp16)[name = tensor("op_28028_cast_fp16")]; + tensor var_28032_begin_0 = const()[name = tensor("op_28032_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_28032_end_0 = const()[name = tensor("op_28032_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_28032_end_mask_0 = const()[name = tensor("op_28032_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28032_cast_fp16 = slice_by_index(begin = var_28032_begin_0, end = var_28032_end_0, end_mask = var_28032_end_mask_0, x = q_127_cast_fp16)[name = tensor("op_28032_cast_fp16")]; + tensor var_28036_begin_0 = const()[name = tensor("op_28036_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_28036_end_0 = const()[name = tensor("op_28036_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_28036_end_mask_0 = const()[name = tensor("op_28036_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28036_cast_fp16 = slice_by_index(begin = var_28036_begin_0, end = var_28036_end_0, end_mask = var_28036_end_mask_0, x = q_127_cast_fp16)[name = tensor("op_28036_cast_fp16")]; + tensor var_28040_begin_0 = const()[name = tensor("op_28040_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_28040_end_0 = const()[name = tensor("op_28040_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_28040_end_mask_0 = const()[name = tensor("op_28040_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28040_cast_fp16 = slice_by_index(begin = var_28040_begin_0, end = var_28040_end_0, end_mask = var_28040_end_mask_0, x = q_127_cast_fp16)[name = tensor("op_28040_cast_fp16")]; + tensor var_28044_begin_0 = const()[name = tensor("op_28044_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_28044_end_0 = const()[name = tensor("op_28044_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_28044_end_mask_0 = const()[name = tensor("op_28044_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28044_cast_fp16 = slice_by_index(begin = var_28044_begin_0, end = var_28044_end_0, end_mask = var_28044_end_mask_0, x = q_127_cast_fp16)[name = tensor("op_28044_cast_fp16")]; + tensor var_28048_begin_0 = const()[name = tensor("op_28048_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_28048_end_0 = const()[name = tensor("op_28048_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_28048_end_mask_0 = const()[name = tensor("op_28048_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28048_cast_fp16 = slice_by_index(begin = var_28048_begin_0, end = var_28048_end_0, end_mask = var_28048_end_mask_0, x = q_127_cast_fp16)[name = tensor("op_28048_cast_fp16")]; + tensor var_28052_begin_0 = const()[name = tensor("op_28052_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_28052_end_0 = const()[name = tensor("op_28052_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_28052_end_mask_0 = const()[name = tensor("op_28052_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28052_cast_fp16 = slice_by_index(begin = var_28052_begin_0, end = var_28052_end_0, end_mask = var_28052_end_mask_0, x = q_127_cast_fp16)[name = tensor("op_28052_cast_fp16")]; + tensor var_28056_begin_0 = const()[name = tensor("op_28056_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_28056_end_0 = const()[name = tensor("op_28056_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_28056_end_mask_0 = const()[name = tensor("op_28056_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28056_cast_fp16 = slice_by_index(begin = var_28056_begin_0, end = var_28056_end_0, end_mask = var_28056_end_mask_0, x = q_127_cast_fp16)[name = tensor("op_28056_cast_fp16")]; + tensor var_28060_begin_0 = const()[name = tensor("op_28060_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_28060_end_0 = const()[name = tensor("op_28060_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_28060_end_mask_0 = const()[name = tensor("op_28060_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28060_cast_fp16 = slice_by_index(begin = var_28060_begin_0, end = var_28060_end_0, end_mask = var_28060_end_mask_0, x = q_127_cast_fp16)[name = tensor("op_28060_cast_fp16")]; + tensor var_28064_begin_0 = const()[name = tensor("op_28064_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_28064_end_0 = const()[name = tensor("op_28064_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_28064_end_mask_0 = const()[name = tensor("op_28064_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28064_cast_fp16 = slice_by_index(begin = var_28064_begin_0, end = var_28064_end_0, end_mask = var_28064_end_mask_0, x = q_127_cast_fp16)[name = tensor("op_28064_cast_fp16")]; + tensor var_28068_begin_0 = const()[name = tensor("op_28068_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_28068_end_0 = const()[name = tensor("op_28068_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_28068_end_mask_0 = const()[name = tensor("op_28068_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28068_cast_fp16 = slice_by_index(begin = var_28068_begin_0, end = var_28068_end_0, end_mask = var_28068_end_mask_0, x = q_127_cast_fp16)[name = tensor("op_28068_cast_fp16")]; + tensor var_28072_begin_0 = const()[name = tensor("op_28072_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_28072_end_0 = const()[name = tensor("op_28072_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_28072_end_mask_0 = const()[name = tensor("op_28072_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28072_cast_fp16 = slice_by_index(begin = var_28072_begin_0, end = var_28072_end_0, end_mask = var_28072_end_mask_0, x = q_127_cast_fp16)[name = tensor("op_28072_cast_fp16")]; + tensor var_28076_begin_0 = const()[name = tensor("op_28076_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_28076_end_0 = const()[name = tensor("op_28076_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_28076_end_mask_0 = const()[name = tensor("op_28076_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28076_cast_fp16 = slice_by_index(begin = var_28076_begin_0, end = var_28076_end_0, end_mask = var_28076_end_mask_0, x = q_127_cast_fp16)[name = tensor("op_28076_cast_fp16")]; + tensor var_28080_begin_0 = const()[name = tensor("op_28080_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_28080_end_0 = const()[name = tensor("op_28080_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_28080_end_mask_0 = const()[name = tensor("op_28080_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28080_cast_fp16 = slice_by_index(begin = var_28080_begin_0, end = var_28080_end_0, end_mask = var_28080_end_mask_0, x = q_127_cast_fp16)[name = tensor("op_28080_cast_fp16")]; + tensor var_28084_begin_0 = const()[name = tensor("op_28084_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_28084_end_0 = const()[name = tensor("op_28084_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_28084_end_mask_0 = const()[name = tensor("op_28084_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28084_cast_fp16 = slice_by_index(begin = var_28084_begin_0, end = var_28084_end_0, end_mask = var_28084_end_mask_0, x = q_127_cast_fp16)[name = tensor("op_28084_cast_fp16")]; + tensor var_28088_begin_0 = const()[name = tensor("op_28088_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_28088_end_0 = const()[name = tensor("op_28088_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_28088_end_mask_0 = const()[name = tensor("op_28088_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28088_cast_fp16 = slice_by_index(begin = var_28088_begin_0, end = var_28088_end_0, end_mask = var_28088_end_mask_0, x = q_127_cast_fp16)[name = tensor("op_28088_cast_fp16")]; + tensor var_28092_begin_0 = const()[name = tensor("op_28092_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_28092_end_0 = const()[name = tensor("op_28092_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_28092_end_mask_0 = const()[name = tensor("op_28092_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28092_cast_fp16 = slice_by_index(begin = var_28092_begin_0, end = var_28092_end_0, end_mask = var_28092_end_mask_0, x = q_127_cast_fp16)[name = tensor("op_28092_cast_fp16")]; + tensor var_28096_begin_0 = const()[name = tensor("op_28096_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_28096_end_0 = const()[name = tensor("op_28096_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_28096_end_mask_0 = const()[name = tensor("op_28096_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28096_cast_fp16 = slice_by_index(begin = var_28096_begin_0, end = var_28096_end_0, end_mask = var_28096_end_mask_0, x = q_127_cast_fp16)[name = tensor("op_28096_cast_fp16")]; + tensor k_255_perm_0 = const()[name = tensor("k_255_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_28103_begin_0 = const()[name = tensor("op_28103_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28103_end_0 = const()[name = tensor("op_28103_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_28103_end_mask_0 = const()[name = tensor("op_28103_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_255_cast_fp16 = transpose(perm = k_255_perm_0, x = k_253_cast_fp16)[name = tensor("transpose_76")]; + tensor var_28103_cast_fp16 = slice_by_index(begin = var_28103_begin_0, end = var_28103_end_0, end_mask = var_28103_end_mask_0, x = k_255_cast_fp16)[name = tensor("op_28103_cast_fp16")]; + tensor var_28107_begin_0 = const()[name = tensor("op_28107_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_28107_end_0 = const()[name = tensor("op_28107_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_28107_end_mask_0 = const()[name = tensor("op_28107_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28107_cast_fp16 = slice_by_index(begin = var_28107_begin_0, end = var_28107_end_0, end_mask = var_28107_end_mask_0, x = k_255_cast_fp16)[name = tensor("op_28107_cast_fp16")]; + tensor var_28111_begin_0 = const()[name = tensor("op_28111_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_28111_end_0 = const()[name = tensor("op_28111_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_28111_end_mask_0 = const()[name = tensor("op_28111_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28111_cast_fp16 = slice_by_index(begin = var_28111_begin_0, end = var_28111_end_0, end_mask = var_28111_end_mask_0, x = k_255_cast_fp16)[name = tensor("op_28111_cast_fp16")]; + tensor var_28115_begin_0 = const()[name = tensor("op_28115_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_28115_end_0 = const()[name = tensor("op_28115_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_28115_end_mask_0 = const()[name = tensor("op_28115_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28115_cast_fp16 = slice_by_index(begin = var_28115_begin_0, end = var_28115_end_0, end_mask = var_28115_end_mask_0, x = k_255_cast_fp16)[name = tensor("op_28115_cast_fp16")]; + tensor var_28119_begin_0 = const()[name = tensor("op_28119_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_28119_end_0 = const()[name = tensor("op_28119_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_28119_end_mask_0 = const()[name = tensor("op_28119_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28119_cast_fp16 = slice_by_index(begin = var_28119_begin_0, end = var_28119_end_0, end_mask = var_28119_end_mask_0, x = k_255_cast_fp16)[name = tensor("op_28119_cast_fp16")]; + tensor var_28123_begin_0 = const()[name = tensor("op_28123_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_28123_end_0 = const()[name = tensor("op_28123_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_28123_end_mask_0 = const()[name = tensor("op_28123_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28123_cast_fp16 = slice_by_index(begin = var_28123_begin_0, end = var_28123_end_0, end_mask = var_28123_end_mask_0, x = k_255_cast_fp16)[name = tensor("op_28123_cast_fp16")]; + tensor var_28127_begin_0 = const()[name = tensor("op_28127_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_28127_end_0 = const()[name = tensor("op_28127_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_28127_end_mask_0 = const()[name = tensor("op_28127_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28127_cast_fp16 = slice_by_index(begin = var_28127_begin_0, end = var_28127_end_0, end_mask = var_28127_end_mask_0, x = k_255_cast_fp16)[name = tensor("op_28127_cast_fp16")]; + tensor var_28131_begin_0 = const()[name = tensor("op_28131_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_28131_end_0 = const()[name = tensor("op_28131_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_28131_end_mask_0 = const()[name = tensor("op_28131_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28131_cast_fp16 = slice_by_index(begin = var_28131_begin_0, end = var_28131_end_0, end_mask = var_28131_end_mask_0, x = k_255_cast_fp16)[name = tensor("op_28131_cast_fp16")]; + tensor var_28135_begin_0 = const()[name = tensor("op_28135_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_28135_end_0 = const()[name = tensor("op_28135_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_28135_end_mask_0 = const()[name = tensor("op_28135_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28135_cast_fp16 = slice_by_index(begin = var_28135_begin_0, end = var_28135_end_0, end_mask = var_28135_end_mask_0, x = k_255_cast_fp16)[name = tensor("op_28135_cast_fp16")]; + tensor var_28139_begin_0 = const()[name = tensor("op_28139_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_28139_end_0 = const()[name = tensor("op_28139_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_28139_end_mask_0 = const()[name = tensor("op_28139_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28139_cast_fp16 = slice_by_index(begin = var_28139_begin_0, end = var_28139_end_0, end_mask = var_28139_end_mask_0, x = k_255_cast_fp16)[name = tensor("op_28139_cast_fp16")]; + tensor var_28143_begin_0 = const()[name = tensor("op_28143_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_28143_end_0 = const()[name = tensor("op_28143_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_28143_end_mask_0 = const()[name = tensor("op_28143_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28143_cast_fp16 = slice_by_index(begin = var_28143_begin_0, end = var_28143_end_0, end_mask = var_28143_end_mask_0, x = k_255_cast_fp16)[name = tensor("op_28143_cast_fp16")]; + tensor var_28147_begin_0 = const()[name = tensor("op_28147_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_28147_end_0 = const()[name = tensor("op_28147_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_28147_end_mask_0 = const()[name = tensor("op_28147_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28147_cast_fp16 = slice_by_index(begin = var_28147_begin_0, end = var_28147_end_0, end_mask = var_28147_end_mask_0, x = k_255_cast_fp16)[name = tensor("op_28147_cast_fp16")]; + tensor var_28151_begin_0 = const()[name = tensor("op_28151_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_28151_end_0 = const()[name = tensor("op_28151_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_28151_end_mask_0 = const()[name = tensor("op_28151_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28151_cast_fp16 = slice_by_index(begin = var_28151_begin_0, end = var_28151_end_0, end_mask = var_28151_end_mask_0, x = k_255_cast_fp16)[name = tensor("op_28151_cast_fp16")]; + tensor var_28155_begin_0 = const()[name = tensor("op_28155_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_28155_end_0 = const()[name = tensor("op_28155_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_28155_end_mask_0 = const()[name = tensor("op_28155_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28155_cast_fp16 = slice_by_index(begin = var_28155_begin_0, end = var_28155_end_0, end_mask = var_28155_end_mask_0, x = k_255_cast_fp16)[name = tensor("op_28155_cast_fp16")]; + tensor var_28159_begin_0 = const()[name = tensor("op_28159_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_28159_end_0 = const()[name = tensor("op_28159_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_28159_end_mask_0 = const()[name = tensor("op_28159_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28159_cast_fp16 = slice_by_index(begin = var_28159_begin_0, end = var_28159_end_0, end_mask = var_28159_end_mask_0, x = k_255_cast_fp16)[name = tensor("op_28159_cast_fp16")]; + tensor var_28163_begin_0 = const()[name = tensor("op_28163_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_28163_end_0 = const()[name = tensor("op_28163_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_28163_end_mask_0 = const()[name = tensor("op_28163_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28163_cast_fp16 = slice_by_index(begin = var_28163_begin_0, end = var_28163_end_0, end_mask = var_28163_end_mask_0, x = k_255_cast_fp16)[name = tensor("op_28163_cast_fp16")]; + tensor var_28167_begin_0 = const()[name = tensor("op_28167_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_28167_end_0 = const()[name = tensor("op_28167_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_28167_end_mask_0 = const()[name = tensor("op_28167_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28167_cast_fp16 = slice_by_index(begin = var_28167_begin_0, end = var_28167_end_0, end_mask = var_28167_end_mask_0, x = k_255_cast_fp16)[name = tensor("op_28167_cast_fp16")]; + tensor var_28171_begin_0 = const()[name = tensor("op_28171_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_28171_end_0 = const()[name = tensor("op_28171_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_28171_end_mask_0 = const()[name = tensor("op_28171_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28171_cast_fp16 = slice_by_index(begin = var_28171_begin_0, end = var_28171_end_0, end_mask = var_28171_end_mask_0, x = k_255_cast_fp16)[name = tensor("op_28171_cast_fp16")]; + tensor var_28175_begin_0 = const()[name = tensor("op_28175_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_28175_end_0 = const()[name = tensor("op_28175_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_28175_end_mask_0 = const()[name = tensor("op_28175_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28175_cast_fp16 = slice_by_index(begin = var_28175_begin_0, end = var_28175_end_0, end_mask = var_28175_end_mask_0, x = k_255_cast_fp16)[name = tensor("op_28175_cast_fp16")]; + tensor var_28179_begin_0 = const()[name = tensor("op_28179_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_28179_end_0 = const()[name = tensor("op_28179_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_28179_end_mask_0 = const()[name = tensor("op_28179_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28179_cast_fp16 = slice_by_index(begin = var_28179_begin_0, end = var_28179_end_0, end_mask = var_28179_end_mask_0, x = k_255_cast_fp16)[name = tensor("op_28179_cast_fp16")]; + tensor var_28181_begin_0 = const()[name = tensor("op_28181_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28181_end_0 = const()[name = tensor("op_28181_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_28181_end_mask_0 = const()[name = tensor("op_28181_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28181_cast_fp16 = slice_by_index(begin = var_28181_begin_0, end = var_28181_end_0, end_mask = var_28181_end_mask_0, x = v_127_cast_fp16)[name = tensor("op_28181_cast_fp16")]; + tensor var_28185_begin_0 = const()[name = tensor("op_28185_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_28185_end_0 = const()[name = tensor("op_28185_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_28185_end_mask_0 = const()[name = tensor("op_28185_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28185_cast_fp16 = slice_by_index(begin = var_28185_begin_0, end = var_28185_end_0, end_mask = var_28185_end_mask_0, x = v_127_cast_fp16)[name = tensor("op_28185_cast_fp16")]; + tensor var_28189_begin_0 = const()[name = tensor("op_28189_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_28189_end_0 = const()[name = tensor("op_28189_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_28189_end_mask_0 = const()[name = tensor("op_28189_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28189_cast_fp16 = slice_by_index(begin = var_28189_begin_0, end = var_28189_end_0, end_mask = var_28189_end_mask_0, x = v_127_cast_fp16)[name = tensor("op_28189_cast_fp16")]; + tensor var_28193_begin_0 = const()[name = tensor("op_28193_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_28193_end_0 = const()[name = tensor("op_28193_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_28193_end_mask_0 = const()[name = tensor("op_28193_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28193_cast_fp16 = slice_by_index(begin = var_28193_begin_0, end = var_28193_end_0, end_mask = var_28193_end_mask_0, x = v_127_cast_fp16)[name = tensor("op_28193_cast_fp16")]; + tensor var_28197_begin_0 = const()[name = tensor("op_28197_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_28197_end_0 = const()[name = tensor("op_28197_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_28197_end_mask_0 = const()[name = tensor("op_28197_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28197_cast_fp16 = slice_by_index(begin = var_28197_begin_0, end = var_28197_end_0, end_mask = var_28197_end_mask_0, x = v_127_cast_fp16)[name = tensor("op_28197_cast_fp16")]; + tensor var_28201_begin_0 = const()[name = tensor("op_28201_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_28201_end_0 = const()[name = tensor("op_28201_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_28201_end_mask_0 = const()[name = tensor("op_28201_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28201_cast_fp16 = slice_by_index(begin = var_28201_begin_0, end = var_28201_end_0, end_mask = var_28201_end_mask_0, x = v_127_cast_fp16)[name = tensor("op_28201_cast_fp16")]; + tensor var_28205_begin_0 = const()[name = tensor("op_28205_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_28205_end_0 = const()[name = tensor("op_28205_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_28205_end_mask_0 = const()[name = tensor("op_28205_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28205_cast_fp16 = slice_by_index(begin = var_28205_begin_0, end = var_28205_end_0, end_mask = var_28205_end_mask_0, x = v_127_cast_fp16)[name = tensor("op_28205_cast_fp16")]; + tensor var_28209_begin_0 = const()[name = tensor("op_28209_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_28209_end_0 = const()[name = tensor("op_28209_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_28209_end_mask_0 = const()[name = tensor("op_28209_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28209_cast_fp16 = slice_by_index(begin = var_28209_begin_0, end = var_28209_end_0, end_mask = var_28209_end_mask_0, x = v_127_cast_fp16)[name = tensor("op_28209_cast_fp16")]; + tensor var_28213_begin_0 = const()[name = tensor("op_28213_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_28213_end_0 = const()[name = tensor("op_28213_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_28213_end_mask_0 = const()[name = tensor("op_28213_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28213_cast_fp16 = slice_by_index(begin = var_28213_begin_0, end = var_28213_end_0, end_mask = var_28213_end_mask_0, x = v_127_cast_fp16)[name = tensor("op_28213_cast_fp16")]; + tensor var_28217_begin_0 = const()[name = tensor("op_28217_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_28217_end_0 = const()[name = tensor("op_28217_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_28217_end_mask_0 = const()[name = tensor("op_28217_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28217_cast_fp16 = slice_by_index(begin = var_28217_begin_0, end = var_28217_end_0, end_mask = var_28217_end_mask_0, x = v_127_cast_fp16)[name = tensor("op_28217_cast_fp16")]; + tensor var_28221_begin_0 = const()[name = tensor("op_28221_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_28221_end_0 = const()[name = tensor("op_28221_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_28221_end_mask_0 = const()[name = tensor("op_28221_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28221_cast_fp16 = slice_by_index(begin = var_28221_begin_0, end = var_28221_end_0, end_mask = var_28221_end_mask_0, x = v_127_cast_fp16)[name = tensor("op_28221_cast_fp16")]; + tensor var_28225_begin_0 = const()[name = tensor("op_28225_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_28225_end_0 = const()[name = tensor("op_28225_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_28225_end_mask_0 = const()[name = tensor("op_28225_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28225_cast_fp16 = slice_by_index(begin = var_28225_begin_0, end = var_28225_end_0, end_mask = var_28225_end_mask_0, x = v_127_cast_fp16)[name = tensor("op_28225_cast_fp16")]; + tensor var_28229_begin_0 = const()[name = tensor("op_28229_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_28229_end_0 = const()[name = tensor("op_28229_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_28229_end_mask_0 = const()[name = tensor("op_28229_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28229_cast_fp16 = slice_by_index(begin = var_28229_begin_0, end = var_28229_end_0, end_mask = var_28229_end_mask_0, x = v_127_cast_fp16)[name = tensor("op_28229_cast_fp16")]; + tensor var_28233_begin_0 = const()[name = tensor("op_28233_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_28233_end_0 = const()[name = tensor("op_28233_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_28233_end_mask_0 = const()[name = tensor("op_28233_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28233_cast_fp16 = slice_by_index(begin = var_28233_begin_0, end = var_28233_end_0, end_mask = var_28233_end_mask_0, x = v_127_cast_fp16)[name = tensor("op_28233_cast_fp16")]; + tensor var_28237_begin_0 = const()[name = tensor("op_28237_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_28237_end_0 = const()[name = tensor("op_28237_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_28237_end_mask_0 = const()[name = tensor("op_28237_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28237_cast_fp16 = slice_by_index(begin = var_28237_begin_0, end = var_28237_end_0, end_mask = var_28237_end_mask_0, x = v_127_cast_fp16)[name = tensor("op_28237_cast_fp16")]; + tensor var_28241_begin_0 = const()[name = tensor("op_28241_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_28241_end_0 = const()[name = tensor("op_28241_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_28241_end_mask_0 = const()[name = tensor("op_28241_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28241_cast_fp16 = slice_by_index(begin = var_28241_begin_0, end = var_28241_end_0, end_mask = var_28241_end_mask_0, x = v_127_cast_fp16)[name = tensor("op_28241_cast_fp16")]; + tensor var_28245_begin_0 = const()[name = tensor("op_28245_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_28245_end_0 = const()[name = tensor("op_28245_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_28245_end_mask_0 = const()[name = tensor("op_28245_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28245_cast_fp16 = slice_by_index(begin = var_28245_begin_0, end = var_28245_end_0, end_mask = var_28245_end_mask_0, x = v_127_cast_fp16)[name = tensor("op_28245_cast_fp16")]; + tensor var_28249_begin_0 = const()[name = tensor("op_28249_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_28249_end_0 = const()[name = tensor("op_28249_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_28249_end_mask_0 = const()[name = tensor("op_28249_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28249_cast_fp16 = slice_by_index(begin = var_28249_begin_0, end = var_28249_end_0, end_mask = var_28249_end_mask_0, x = v_127_cast_fp16)[name = tensor("op_28249_cast_fp16")]; + tensor var_28253_begin_0 = const()[name = tensor("op_28253_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_28253_end_0 = const()[name = tensor("op_28253_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_28253_end_mask_0 = const()[name = tensor("op_28253_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28253_cast_fp16 = slice_by_index(begin = var_28253_begin_0, end = var_28253_end_0, end_mask = var_28253_end_mask_0, x = v_127_cast_fp16)[name = tensor("op_28253_cast_fp16")]; + tensor var_28257_begin_0 = const()[name = tensor("op_28257_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_28257_end_0 = const()[name = tensor("op_28257_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_28257_end_mask_0 = const()[name = tensor("op_28257_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28257_cast_fp16 = slice_by_index(begin = var_28257_begin_0, end = var_28257_end_0, end_mask = var_28257_end_mask_0, x = v_127_cast_fp16)[name = tensor("op_28257_cast_fp16")]; + tensor var_28261_equation_0 = const()[name = tensor("op_28261_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28261_cast_fp16 = einsum(equation = var_28261_equation_0, values = (var_28103_cast_fp16, var_28020_cast_fp16))[name = tensor("op_28261_cast_fp16")]; + tensor var_28262_to_fp16 = const()[name = tensor("op_28262_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2361_cast_fp16 = mul(x = var_28261_cast_fp16, y = var_28262_to_fp16)[name = tensor("aw_2361_cast_fp16")]; + tensor var_28265_equation_0 = const()[name = tensor("op_28265_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28265_cast_fp16 = einsum(equation = var_28265_equation_0, values = (var_28107_cast_fp16, var_28024_cast_fp16))[name = tensor("op_28265_cast_fp16")]; + tensor var_28266_to_fp16 = const()[name = tensor("op_28266_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2363_cast_fp16 = mul(x = var_28265_cast_fp16, y = var_28266_to_fp16)[name = tensor("aw_2363_cast_fp16")]; + tensor var_28269_equation_0 = const()[name = tensor("op_28269_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28269_cast_fp16 = einsum(equation = var_28269_equation_0, values = (var_28111_cast_fp16, var_28028_cast_fp16))[name = tensor("op_28269_cast_fp16")]; + tensor var_28270_to_fp16 = const()[name = tensor("op_28270_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2365_cast_fp16 = mul(x = var_28269_cast_fp16, y = var_28270_to_fp16)[name = tensor("aw_2365_cast_fp16")]; + tensor var_28273_equation_0 = const()[name = tensor("op_28273_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28273_cast_fp16 = einsum(equation = var_28273_equation_0, values = (var_28115_cast_fp16, var_28032_cast_fp16))[name = tensor("op_28273_cast_fp16")]; + tensor var_28274_to_fp16 = const()[name = tensor("op_28274_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2367_cast_fp16 = mul(x = var_28273_cast_fp16, y = var_28274_to_fp16)[name = tensor("aw_2367_cast_fp16")]; + tensor var_28277_equation_0 = const()[name = tensor("op_28277_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28277_cast_fp16 = einsum(equation = var_28277_equation_0, values = (var_28119_cast_fp16, var_28036_cast_fp16))[name = tensor("op_28277_cast_fp16")]; + tensor var_28278_to_fp16 = const()[name = tensor("op_28278_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2369_cast_fp16 = mul(x = var_28277_cast_fp16, y = var_28278_to_fp16)[name = tensor("aw_2369_cast_fp16")]; + tensor var_28281_equation_0 = const()[name = tensor("op_28281_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28281_cast_fp16 = einsum(equation = var_28281_equation_0, values = (var_28123_cast_fp16, var_28040_cast_fp16))[name = tensor("op_28281_cast_fp16")]; + tensor var_28282_to_fp16 = const()[name = tensor("op_28282_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2371_cast_fp16 = mul(x = var_28281_cast_fp16, y = var_28282_to_fp16)[name = tensor("aw_2371_cast_fp16")]; + tensor var_28285_equation_0 = const()[name = tensor("op_28285_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28285_cast_fp16 = einsum(equation = var_28285_equation_0, values = (var_28127_cast_fp16, var_28044_cast_fp16))[name = tensor("op_28285_cast_fp16")]; + tensor var_28286_to_fp16 = const()[name = tensor("op_28286_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2373_cast_fp16 = mul(x = var_28285_cast_fp16, y = var_28286_to_fp16)[name = tensor("aw_2373_cast_fp16")]; + tensor var_28289_equation_0 = const()[name = tensor("op_28289_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28289_cast_fp16 = einsum(equation = var_28289_equation_0, values = (var_28131_cast_fp16, var_28048_cast_fp16))[name = tensor("op_28289_cast_fp16")]; + tensor var_28290_to_fp16 = const()[name = tensor("op_28290_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2375_cast_fp16 = mul(x = var_28289_cast_fp16, y = var_28290_to_fp16)[name = tensor("aw_2375_cast_fp16")]; + tensor var_28293_equation_0 = const()[name = tensor("op_28293_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28293_cast_fp16 = einsum(equation = var_28293_equation_0, values = (var_28135_cast_fp16, var_28052_cast_fp16))[name = tensor("op_28293_cast_fp16")]; + tensor var_28294_to_fp16 = const()[name = tensor("op_28294_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2377_cast_fp16 = mul(x = var_28293_cast_fp16, y = var_28294_to_fp16)[name = tensor("aw_2377_cast_fp16")]; + tensor var_28297_equation_0 = const()[name = tensor("op_28297_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28297_cast_fp16 = einsum(equation = var_28297_equation_0, values = (var_28139_cast_fp16, var_28056_cast_fp16))[name = tensor("op_28297_cast_fp16")]; + tensor var_28298_to_fp16 = const()[name = tensor("op_28298_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2379_cast_fp16 = mul(x = var_28297_cast_fp16, y = var_28298_to_fp16)[name = tensor("aw_2379_cast_fp16")]; + tensor var_28301_equation_0 = const()[name = tensor("op_28301_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28301_cast_fp16 = einsum(equation = var_28301_equation_0, values = (var_28143_cast_fp16, var_28060_cast_fp16))[name = tensor("op_28301_cast_fp16")]; + tensor var_28302_to_fp16 = const()[name = tensor("op_28302_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2381_cast_fp16 = mul(x = var_28301_cast_fp16, y = var_28302_to_fp16)[name = tensor("aw_2381_cast_fp16")]; + tensor var_28305_equation_0 = const()[name = tensor("op_28305_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28305_cast_fp16 = einsum(equation = var_28305_equation_0, values = (var_28147_cast_fp16, var_28064_cast_fp16))[name = tensor("op_28305_cast_fp16")]; + tensor var_28306_to_fp16 = const()[name = tensor("op_28306_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2383_cast_fp16 = mul(x = var_28305_cast_fp16, y = var_28306_to_fp16)[name = tensor("aw_2383_cast_fp16")]; + tensor var_28309_equation_0 = const()[name = tensor("op_28309_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28309_cast_fp16 = einsum(equation = var_28309_equation_0, values = (var_28151_cast_fp16, var_28068_cast_fp16))[name = tensor("op_28309_cast_fp16")]; + tensor var_28310_to_fp16 = const()[name = tensor("op_28310_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2385_cast_fp16 = mul(x = var_28309_cast_fp16, y = var_28310_to_fp16)[name = tensor("aw_2385_cast_fp16")]; + tensor var_28313_equation_0 = const()[name = tensor("op_28313_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28313_cast_fp16 = einsum(equation = var_28313_equation_0, values = (var_28155_cast_fp16, var_28072_cast_fp16))[name = tensor("op_28313_cast_fp16")]; + tensor var_28314_to_fp16 = const()[name = tensor("op_28314_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2387_cast_fp16 = mul(x = var_28313_cast_fp16, y = var_28314_to_fp16)[name = tensor("aw_2387_cast_fp16")]; + tensor var_28317_equation_0 = const()[name = tensor("op_28317_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28317_cast_fp16 = einsum(equation = var_28317_equation_0, values = (var_28159_cast_fp16, var_28076_cast_fp16))[name = tensor("op_28317_cast_fp16")]; + tensor var_28318_to_fp16 = const()[name = tensor("op_28318_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2389_cast_fp16 = mul(x = var_28317_cast_fp16, y = var_28318_to_fp16)[name = tensor("aw_2389_cast_fp16")]; + tensor var_28321_equation_0 = const()[name = tensor("op_28321_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28321_cast_fp16 = einsum(equation = var_28321_equation_0, values = (var_28163_cast_fp16, var_28080_cast_fp16))[name = tensor("op_28321_cast_fp16")]; + tensor var_28322_to_fp16 = const()[name = tensor("op_28322_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2391_cast_fp16 = mul(x = var_28321_cast_fp16, y = var_28322_to_fp16)[name = tensor("aw_2391_cast_fp16")]; + tensor var_28325_equation_0 = const()[name = tensor("op_28325_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28325_cast_fp16 = einsum(equation = var_28325_equation_0, values = (var_28167_cast_fp16, var_28084_cast_fp16))[name = tensor("op_28325_cast_fp16")]; + tensor var_28326_to_fp16 = const()[name = tensor("op_28326_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2393_cast_fp16 = mul(x = var_28325_cast_fp16, y = var_28326_to_fp16)[name = tensor("aw_2393_cast_fp16")]; + tensor var_28329_equation_0 = const()[name = tensor("op_28329_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28329_cast_fp16 = einsum(equation = var_28329_equation_0, values = (var_28171_cast_fp16, var_28088_cast_fp16))[name = tensor("op_28329_cast_fp16")]; + tensor var_28330_to_fp16 = const()[name = tensor("op_28330_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2395_cast_fp16 = mul(x = var_28329_cast_fp16, y = var_28330_to_fp16)[name = tensor("aw_2395_cast_fp16")]; + tensor var_28333_equation_0 = const()[name = tensor("op_28333_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28333_cast_fp16 = einsum(equation = var_28333_equation_0, values = (var_28175_cast_fp16, var_28092_cast_fp16))[name = tensor("op_28333_cast_fp16")]; + tensor var_28334_to_fp16 = const()[name = tensor("op_28334_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2397_cast_fp16 = mul(x = var_28333_cast_fp16, y = var_28334_to_fp16)[name = tensor("aw_2397_cast_fp16")]; + tensor var_28337_equation_0 = const()[name = tensor("op_28337_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28337_cast_fp16 = einsum(equation = var_28337_equation_0, values = (var_28179_cast_fp16, var_28096_cast_fp16))[name = tensor("op_28337_cast_fp16")]; + tensor var_28338_to_fp16 = const()[name = tensor("op_28338_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2399_cast_fp16 = mul(x = var_28337_cast_fp16, y = var_28338_to_fp16)[name = tensor("aw_2399_cast_fp16")]; + tensor var_28340_cast_fp16 = softmax(axis = var_21077, x = aw_2361_cast_fp16)[name = tensor("op_28340_cast_fp16")]; + tensor var_28341_cast_fp16 = softmax(axis = var_21077, x = aw_2363_cast_fp16)[name = tensor("op_28341_cast_fp16")]; + tensor var_28342_cast_fp16 = softmax(axis = var_21077, x = aw_2365_cast_fp16)[name = tensor("op_28342_cast_fp16")]; + tensor var_28343_cast_fp16 = softmax(axis = var_21077, x = aw_2367_cast_fp16)[name = tensor("op_28343_cast_fp16")]; + tensor var_28344_cast_fp16 = softmax(axis = var_21077, x = aw_2369_cast_fp16)[name = tensor("op_28344_cast_fp16")]; + tensor var_28345_cast_fp16 = softmax(axis = var_21077, x = aw_2371_cast_fp16)[name = tensor("op_28345_cast_fp16")]; + tensor var_28346_cast_fp16 = softmax(axis = var_21077, x = aw_2373_cast_fp16)[name = tensor("op_28346_cast_fp16")]; + tensor var_28347_cast_fp16 = softmax(axis = var_21077, x = aw_2375_cast_fp16)[name = tensor("op_28347_cast_fp16")]; + tensor var_28348_cast_fp16 = softmax(axis = var_21077, x = aw_2377_cast_fp16)[name = tensor("op_28348_cast_fp16")]; + tensor var_28349_cast_fp16 = softmax(axis = var_21077, x = aw_2379_cast_fp16)[name = tensor("op_28349_cast_fp16")]; + tensor var_28350_cast_fp16 = softmax(axis = var_21077, x = aw_2381_cast_fp16)[name = tensor("op_28350_cast_fp16")]; + tensor var_28351_cast_fp16 = softmax(axis = var_21077, x = aw_2383_cast_fp16)[name = tensor("op_28351_cast_fp16")]; + tensor var_28352_cast_fp16 = softmax(axis = var_21077, x = aw_2385_cast_fp16)[name = tensor("op_28352_cast_fp16")]; + tensor var_28353_cast_fp16 = softmax(axis = var_21077, x = aw_2387_cast_fp16)[name = tensor("op_28353_cast_fp16")]; + tensor var_28354_cast_fp16 = softmax(axis = var_21077, x = aw_2389_cast_fp16)[name = tensor("op_28354_cast_fp16")]; + tensor var_28355_cast_fp16 = softmax(axis = var_21077, x = aw_2391_cast_fp16)[name = tensor("op_28355_cast_fp16")]; + tensor var_28356_cast_fp16 = softmax(axis = var_21077, x = aw_2393_cast_fp16)[name = tensor("op_28356_cast_fp16")]; + tensor var_28357_cast_fp16 = softmax(axis = var_21077, x = aw_2395_cast_fp16)[name = tensor("op_28357_cast_fp16")]; + tensor var_28358_cast_fp16 = softmax(axis = var_21077, x = aw_2397_cast_fp16)[name = tensor("op_28358_cast_fp16")]; + tensor var_28359_cast_fp16 = softmax(axis = var_21077, x = aw_2399_cast_fp16)[name = tensor("op_28359_cast_fp16")]; + tensor var_28361_equation_0 = const()[name = tensor("op_28361_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28361_cast_fp16 = einsum(equation = var_28361_equation_0, values = (var_28181_cast_fp16, var_28340_cast_fp16))[name = tensor("op_28361_cast_fp16")]; + tensor var_28363_equation_0 = const()[name = tensor("op_28363_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28363_cast_fp16 = einsum(equation = var_28363_equation_0, values = (var_28185_cast_fp16, var_28341_cast_fp16))[name = tensor("op_28363_cast_fp16")]; + tensor var_28365_equation_0 = const()[name = tensor("op_28365_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28365_cast_fp16 = einsum(equation = var_28365_equation_0, values = (var_28189_cast_fp16, var_28342_cast_fp16))[name = tensor("op_28365_cast_fp16")]; + tensor var_28367_equation_0 = const()[name = tensor("op_28367_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28367_cast_fp16 = einsum(equation = var_28367_equation_0, values = (var_28193_cast_fp16, var_28343_cast_fp16))[name = tensor("op_28367_cast_fp16")]; + tensor var_28369_equation_0 = const()[name = tensor("op_28369_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28369_cast_fp16 = einsum(equation = var_28369_equation_0, values = (var_28197_cast_fp16, var_28344_cast_fp16))[name = tensor("op_28369_cast_fp16")]; + tensor var_28371_equation_0 = const()[name = tensor("op_28371_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28371_cast_fp16 = einsum(equation = var_28371_equation_0, values = (var_28201_cast_fp16, var_28345_cast_fp16))[name = tensor("op_28371_cast_fp16")]; + tensor var_28373_equation_0 = const()[name = tensor("op_28373_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28373_cast_fp16 = einsum(equation = var_28373_equation_0, values = (var_28205_cast_fp16, var_28346_cast_fp16))[name = tensor("op_28373_cast_fp16")]; + tensor var_28375_equation_0 = const()[name = tensor("op_28375_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28375_cast_fp16 = einsum(equation = var_28375_equation_0, values = (var_28209_cast_fp16, var_28347_cast_fp16))[name = tensor("op_28375_cast_fp16")]; + tensor var_28377_equation_0 = const()[name = tensor("op_28377_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28377_cast_fp16 = einsum(equation = var_28377_equation_0, values = (var_28213_cast_fp16, var_28348_cast_fp16))[name = tensor("op_28377_cast_fp16")]; + tensor var_28379_equation_0 = const()[name = tensor("op_28379_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28379_cast_fp16 = einsum(equation = var_28379_equation_0, values = (var_28217_cast_fp16, var_28349_cast_fp16))[name = tensor("op_28379_cast_fp16")]; + tensor var_28381_equation_0 = const()[name = tensor("op_28381_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28381_cast_fp16 = einsum(equation = var_28381_equation_0, values = (var_28221_cast_fp16, var_28350_cast_fp16))[name = tensor("op_28381_cast_fp16")]; + tensor var_28383_equation_0 = const()[name = tensor("op_28383_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28383_cast_fp16 = einsum(equation = var_28383_equation_0, values = (var_28225_cast_fp16, var_28351_cast_fp16))[name = tensor("op_28383_cast_fp16")]; + tensor var_28385_equation_0 = const()[name = tensor("op_28385_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28385_cast_fp16 = einsum(equation = var_28385_equation_0, values = (var_28229_cast_fp16, var_28352_cast_fp16))[name = tensor("op_28385_cast_fp16")]; + tensor var_28387_equation_0 = const()[name = tensor("op_28387_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28387_cast_fp16 = einsum(equation = var_28387_equation_0, values = (var_28233_cast_fp16, var_28353_cast_fp16))[name = tensor("op_28387_cast_fp16")]; + tensor var_28389_equation_0 = const()[name = tensor("op_28389_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28389_cast_fp16 = einsum(equation = var_28389_equation_0, values = (var_28237_cast_fp16, var_28354_cast_fp16))[name = tensor("op_28389_cast_fp16")]; + tensor var_28391_equation_0 = const()[name = tensor("op_28391_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28391_cast_fp16 = einsum(equation = var_28391_equation_0, values = (var_28241_cast_fp16, var_28355_cast_fp16))[name = tensor("op_28391_cast_fp16")]; + tensor var_28393_equation_0 = const()[name = tensor("op_28393_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28393_cast_fp16 = einsum(equation = var_28393_equation_0, values = (var_28245_cast_fp16, var_28356_cast_fp16))[name = tensor("op_28393_cast_fp16")]; + tensor var_28395_equation_0 = const()[name = tensor("op_28395_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28395_cast_fp16 = einsum(equation = var_28395_equation_0, values = (var_28249_cast_fp16, var_28357_cast_fp16))[name = tensor("op_28395_cast_fp16")]; + tensor var_28397_equation_0 = const()[name = tensor("op_28397_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28397_cast_fp16 = einsum(equation = var_28397_equation_0, values = (var_28253_cast_fp16, var_28358_cast_fp16))[name = tensor("op_28397_cast_fp16")]; + tensor var_28399_equation_0 = const()[name = tensor("op_28399_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28399_cast_fp16 = einsum(equation = var_28399_equation_0, values = (var_28257_cast_fp16, var_28359_cast_fp16))[name = tensor("op_28399_cast_fp16")]; + tensor input_385_interleave_0 = const()[name = tensor("input_385_interleave_0"), val = tensor(false)]; + tensor input_385_cast_fp16 = concat(axis = var_21077, interleave = input_385_interleave_0, values = (var_28361_cast_fp16, var_28363_cast_fp16, var_28365_cast_fp16, var_28367_cast_fp16, var_28369_cast_fp16, var_28371_cast_fp16, var_28373_cast_fp16, var_28375_cast_fp16, var_28377_cast_fp16, var_28379_cast_fp16, var_28381_cast_fp16, var_28383_cast_fp16, var_28385_cast_fp16, var_28387_cast_fp16, var_28389_cast_fp16, var_28391_cast_fp16, var_28393_cast_fp16, var_28395_cast_fp16, var_28397_cast_fp16, var_28399_cast_fp16))[name = tensor("input_385_cast_fp16")]; + tensor var_28405 = const()[name = tensor("op_28405"), val = tensor([1, 1])]; + tensor var_28407 = const()[name = tensor("op_28407"), val = tensor([1, 1])]; + tensor var_28409_pad_type_0 = const()[name = tensor("op_28409_pad_type_0"), val = tensor("custom")]; + tensor var_28409_pad_0 = const()[name = tensor("op_28409_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_7_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_7_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2239583104)))]; + tensor mid_block_attentions_0_transformer_blocks_7_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_7_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2242859968)))]; + tensor var_28409_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_7_attn2_to_out_0_bias_to_fp16, dilations = var_28407, groups = var_21077, pad = var_28409_pad_0, pad_type = var_28409_pad_type_0, strides = var_28405, weight = mid_block_attentions_0_transformer_blocks_7_attn2_to_out_0_weight_to_fp16, x = input_385_cast_fp16)[name = tensor("op_28409_cast_fp16")]; + tensor inputs_191_cast_fp16 = add(x = var_28409_cast_fp16, y = inputs_189_cast_fp16)[name = tensor("inputs_191_cast_fp16")]; + tensor input_387_axes_0 = const()[name = tensor("input_387_axes_0"), val = tensor([1])]; + tensor input_387_gamma_0_to_fp16 = const()[name = tensor("input_387_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2242862592)))]; + tensor input_387_beta_0_to_fp16 = const()[name = tensor("input_387_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2242865216)))]; + tensor var_28419_to_fp16 = const()[name = tensor("op_28419_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_387_cast_fp16 = layer_norm(axes = input_387_axes_0, beta = input_387_beta_0_to_fp16, epsilon = var_28419_to_fp16, gamma = input_387_gamma_0_to_fp16, x = inputs_191_cast_fp16)[name = tensor("input_387_cast_fp16")]; + tensor var_28435 = const()[name = tensor("op_28435"), val = tensor([1, 1])]; + tensor var_28437 = const()[name = tensor("op_28437"), val = tensor([1, 1])]; + tensor var_28439_pad_type_0 = const()[name = tensor("op_28439_pad_type_0"), val = tensor("custom")]; + tensor var_28439_pad_0 = const()[name = tensor("op_28439_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_7_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_7_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2242867840)))]; + tensor mid_block_attentions_0_transformer_blocks_7_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_7_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2269082304)))]; + tensor var_28439_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_7_ff_net_0_proj_bias_to_fp16, dilations = var_28437, groups = var_21077, pad = var_28439_pad_0, pad_type = var_28439_pad_type_0, strides = var_28435, weight = mid_block_attentions_0_transformer_blocks_7_ff_net_0_proj_weight_to_fp16, x = input_387_cast_fp16)[name = tensor("op_28439_cast_fp16")]; + tensor var_28440_split_sizes_0 = const()[name = tensor("op_28440_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_28440_axis_0 = const()[name = tensor("op_28440_axis_0"), val = tensor(1)]; + tensor var_28440_cast_fp16_0, tensor var_28440_cast_fp16_1 = split(axis = var_28440_axis_0, split_sizes = var_28440_split_sizes_0, x = var_28439_cast_fp16)[name = tensor("op_28440_cast_fp16")]; + tensor var_28442_mode_0 = const()[name = tensor("op_28442_mode_0"), val = tensor("EXACT")]; + tensor var_28442_cast_fp16 = gelu(mode = var_28442_mode_0, x = var_28440_cast_fp16_1)[name = tensor("op_28442_cast_fp16")]; + tensor input_389_cast_fp16 = mul(x = var_28440_cast_fp16_0, y = var_28442_cast_fp16)[name = tensor("input_389_cast_fp16")]; + tensor var_28446 = const()[name = tensor("op_28446"), val = tensor([1, 1])]; + tensor var_28448 = const()[name = tensor("op_28448"), val = tensor([1, 1])]; + tensor var_28450_pad_type_0 = const()[name = tensor("op_28450_pad_type_0"), val = tensor("custom")]; + tensor var_28450_pad_0 = const()[name = tensor("op_28450_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_7_ff_net_2_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_7_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2269102848)))]; + tensor mid_block_attentions_0_transformer_blocks_7_ff_net_2_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_7_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2282210112)))]; + tensor var_28450_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_7_ff_net_2_bias_to_fp16, dilations = var_28448, groups = var_21077, pad = var_28450_pad_0, pad_type = var_28450_pad_type_0, strides = var_28446, weight = mid_block_attentions_0_transformer_blocks_7_ff_net_2_weight_to_fp16, x = input_389_cast_fp16)[name = tensor("op_28450_cast_fp16")]; + tensor inputs_193_cast_fp16 = add(x = var_28450_cast_fp16, y = inputs_191_cast_fp16)[name = tensor("inputs_193_cast_fp16")]; + tensor hidden_states_257_axes_0 = const()[name = tensor("hidden_states_257_axes_0"), val = tensor([1])]; + tensor hidden_states_257_gamma_0_to_fp16 = const()[name = tensor("hidden_states_257_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2282212736)))]; + tensor hidden_states_257_beta_0_to_fp16 = const()[name = tensor("hidden_states_257_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2282215360)))]; + tensor var_28466_to_fp16 = const()[name = tensor("op_28466_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_257_cast_fp16 = layer_norm(axes = hidden_states_257_axes_0, beta = hidden_states_257_beta_0_to_fp16, epsilon = var_28466_to_fp16, gamma = hidden_states_257_gamma_0_to_fp16, x = inputs_193_cast_fp16)[name = tensor("hidden_states_257_cast_fp16")]; + tensor var_28481 = const()[name = tensor("op_28481"), val = tensor([1, 1])]; + tensor var_28483 = const()[name = tensor("op_28483"), val = tensor([1, 1])]; + tensor q_129_pad_type_0 = const()[name = tensor("q_129_pad_type_0"), val = tensor("custom")]; + tensor q_129_pad_0 = const()[name = tensor("q_129_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_8_attn1_to_q_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_8_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2282217984)))]; + tensor q_129_cast_fp16 = conv(dilations = var_28483, groups = var_21077, pad = q_129_pad_0, pad_type = q_129_pad_type_0, strides = var_28481, weight = mid_block_attentions_0_transformer_blocks_8_attn1_to_q_weight_to_fp16, x = hidden_states_257_cast_fp16)[name = tensor("q_129_cast_fp16")]; + tensor var_28487 = const()[name = tensor("op_28487"), val = tensor([1, 1])]; + tensor var_28489 = const()[name = tensor("op_28489"), val = tensor([1, 1])]; + tensor k_257_pad_type_0 = const()[name = tensor("k_257_pad_type_0"), val = tensor("custom")]; + tensor k_257_pad_0 = const()[name = tensor("k_257_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_8_attn1_to_k_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_8_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2285494848)))]; + tensor k_257_cast_fp16 = conv(dilations = var_28489, groups = var_21077, pad = k_257_pad_0, pad_type = k_257_pad_type_0, strides = var_28487, weight = mid_block_attentions_0_transformer_blocks_8_attn1_to_k_weight_to_fp16, x = hidden_states_257_cast_fp16)[name = tensor("k_257_cast_fp16")]; + tensor var_28493 = const()[name = tensor("op_28493"), val = tensor([1, 1])]; + tensor var_28495 = const()[name = tensor("op_28495"), val = tensor([1, 1])]; + tensor v_129_pad_type_0 = const()[name = tensor("v_129_pad_type_0"), val = tensor("custom")]; + tensor v_129_pad_0 = const()[name = tensor("v_129_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_8_attn1_to_v_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_8_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2288771712)))]; + tensor v_129_cast_fp16 = conv(dilations = var_28495, groups = var_21077, pad = v_129_pad_0, pad_type = v_129_pad_type_0, strides = var_28493, weight = mid_block_attentions_0_transformer_blocks_8_attn1_to_v_weight_to_fp16, x = hidden_states_257_cast_fp16)[name = tensor("v_129_cast_fp16")]; + tensor var_28499_begin_0 = const()[name = tensor("op_28499_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28499_end_0 = const()[name = tensor("op_28499_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_28499_end_mask_0 = const()[name = tensor("op_28499_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28499_cast_fp16 = slice_by_index(begin = var_28499_begin_0, end = var_28499_end_0, end_mask = var_28499_end_mask_0, x = q_129_cast_fp16)[name = tensor("op_28499_cast_fp16")]; + tensor var_28503_begin_0 = const()[name = tensor("op_28503_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_28503_end_0 = const()[name = tensor("op_28503_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_28503_end_mask_0 = const()[name = tensor("op_28503_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28503_cast_fp16 = slice_by_index(begin = var_28503_begin_0, end = var_28503_end_0, end_mask = var_28503_end_mask_0, x = q_129_cast_fp16)[name = tensor("op_28503_cast_fp16")]; + tensor var_28507_begin_0 = const()[name = tensor("op_28507_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_28507_end_0 = const()[name = tensor("op_28507_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_28507_end_mask_0 = const()[name = tensor("op_28507_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28507_cast_fp16 = slice_by_index(begin = var_28507_begin_0, end = var_28507_end_0, end_mask = var_28507_end_mask_0, x = q_129_cast_fp16)[name = tensor("op_28507_cast_fp16")]; + tensor var_28511_begin_0 = const()[name = tensor("op_28511_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_28511_end_0 = const()[name = tensor("op_28511_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_28511_end_mask_0 = const()[name = tensor("op_28511_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28511_cast_fp16 = slice_by_index(begin = var_28511_begin_0, end = var_28511_end_0, end_mask = var_28511_end_mask_0, x = q_129_cast_fp16)[name = tensor("op_28511_cast_fp16")]; + tensor var_28515_begin_0 = const()[name = tensor("op_28515_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_28515_end_0 = const()[name = tensor("op_28515_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_28515_end_mask_0 = const()[name = tensor("op_28515_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28515_cast_fp16 = slice_by_index(begin = var_28515_begin_0, end = var_28515_end_0, end_mask = var_28515_end_mask_0, x = q_129_cast_fp16)[name = tensor("op_28515_cast_fp16")]; + tensor var_28519_begin_0 = const()[name = tensor("op_28519_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_28519_end_0 = const()[name = tensor("op_28519_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_28519_end_mask_0 = const()[name = tensor("op_28519_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28519_cast_fp16 = slice_by_index(begin = var_28519_begin_0, end = var_28519_end_0, end_mask = var_28519_end_mask_0, x = q_129_cast_fp16)[name = tensor("op_28519_cast_fp16")]; + tensor var_28523_begin_0 = const()[name = tensor("op_28523_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_28523_end_0 = const()[name = tensor("op_28523_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_28523_end_mask_0 = const()[name = tensor("op_28523_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28523_cast_fp16 = slice_by_index(begin = var_28523_begin_0, end = var_28523_end_0, end_mask = var_28523_end_mask_0, x = q_129_cast_fp16)[name = tensor("op_28523_cast_fp16")]; + tensor var_28527_begin_0 = const()[name = tensor("op_28527_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_28527_end_0 = const()[name = tensor("op_28527_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_28527_end_mask_0 = const()[name = tensor("op_28527_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28527_cast_fp16 = slice_by_index(begin = var_28527_begin_0, end = var_28527_end_0, end_mask = var_28527_end_mask_0, x = q_129_cast_fp16)[name = tensor("op_28527_cast_fp16")]; + tensor var_28531_begin_0 = const()[name = tensor("op_28531_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_28531_end_0 = const()[name = tensor("op_28531_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_28531_end_mask_0 = const()[name = tensor("op_28531_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28531_cast_fp16 = slice_by_index(begin = var_28531_begin_0, end = var_28531_end_0, end_mask = var_28531_end_mask_0, x = q_129_cast_fp16)[name = tensor("op_28531_cast_fp16")]; + tensor var_28535_begin_0 = const()[name = tensor("op_28535_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_28535_end_0 = const()[name = tensor("op_28535_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_28535_end_mask_0 = const()[name = tensor("op_28535_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28535_cast_fp16 = slice_by_index(begin = var_28535_begin_0, end = var_28535_end_0, end_mask = var_28535_end_mask_0, x = q_129_cast_fp16)[name = tensor("op_28535_cast_fp16")]; + tensor var_28539_begin_0 = const()[name = tensor("op_28539_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_28539_end_0 = const()[name = tensor("op_28539_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_28539_end_mask_0 = const()[name = tensor("op_28539_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28539_cast_fp16 = slice_by_index(begin = var_28539_begin_0, end = var_28539_end_0, end_mask = var_28539_end_mask_0, x = q_129_cast_fp16)[name = tensor("op_28539_cast_fp16")]; + tensor var_28543_begin_0 = const()[name = tensor("op_28543_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_28543_end_0 = const()[name = tensor("op_28543_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_28543_end_mask_0 = const()[name = tensor("op_28543_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28543_cast_fp16 = slice_by_index(begin = var_28543_begin_0, end = var_28543_end_0, end_mask = var_28543_end_mask_0, x = q_129_cast_fp16)[name = tensor("op_28543_cast_fp16")]; + tensor var_28547_begin_0 = const()[name = tensor("op_28547_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_28547_end_0 = const()[name = tensor("op_28547_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_28547_end_mask_0 = const()[name = tensor("op_28547_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28547_cast_fp16 = slice_by_index(begin = var_28547_begin_0, end = var_28547_end_0, end_mask = var_28547_end_mask_0, x = q_129_cast_fp16)[name = tensor("op_28547_cast_fp16")]; + tensor var_28551_begin_0 = const()[name = tensor("op_28551_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_28551_end_0 = const()[name = tensor("op_28551_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_28551_end_mask_0 = const()[name = tensor("op_28551_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28551_cast_fp16 = slice_by_index(begin = var_28551_begin_0, end = var_28551_end_0, end_mask = var_28551_end_mask_0, x = q_129_cast_fp16)[name = tensor("op_28551_cast_fp16")]; + tensor var_28555_begin_0 = const()[name = tensor("op_28555_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_28555_end_0 = const()[name = tensor("op_28555_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_28555_end_mask_0 = const()[name = tensor("op_28555_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28555_cast_fp16 = slice_by_index(begin = var_28555_begin_0, end = var_28555_end_0, end_mask = var_28555_end_mask_0, x = q_129_cast_fp16)[name = tensor("op_28555_cast_fp16")]; + tensor var_28559_begin_0 = const()[name = tensor("op_28559_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_28559_end_0 = const()[name = tensor("op_28559_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_28559_end_mask_0 = const()[name = tensor("op_28559_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28559_cast_fp16 = slice_by_index(begin = var_28559_begin_0, end = var_28559_end_0, end_mask = var_28559_end_mask_0, x = q_129_cast_fp16)[name = tensor("op_28559_cast_fp16")]; + tensor var_28563_begin_0 = const()[name = tensor("op_28563_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_28563_end_0 = const()[name = tensor("op_28563_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_28563_end_mask_0 = const()[name = tensor("op_28563_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28563_cast_fp16 = slice_by_index(begin = var_28563_begin_0, end = var_28563_end_0, end_mask = var_28563_end_mask_0, x = q_129_cast_fp16)[name = tensor("op_28563_cast_fp16")]; + tensor var_28567_begin_0 = const()[name = tensor("op_28567_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_28567_end_0 = const()[name = tensor("op_28567_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_28567_end_mask_0 = const()[name = tensor("op_28567_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28567_cast_fp16 = slice_by_index(begin = var_28567_begin_0, end = var_28567_end_0, end_mask = var_28567_end_mask_0, x = q_129_cast_fp16)[name = tensor("op_28567_cast_fp16")]; + tensor var_28571_begin_0 = const()[name = tensor("op_28571_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_28571_end_0 = const()[name = tensor("op_28571_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_28571_end_mask_0 = const()[name = tensor("op_28571_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28571_cast_fp16 = slice_by_index(begin = var_28571_begin_0, end = var_28571_end_0, end_mask = var_28571_end_mask_0, x = q_129_cast_fp16)[name = tensor("op_28571_cast_fp16")]; + tensor var_28575_begin_0 = const()[name = tensor("op_28575_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_28575_end_0 = const()[name = tensor("op_28575_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_28575_end_mask_0 = const()[name = tensor("op_28575_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28575_cast_fp16 = slice_by_index(begin = var_28575_begin_0, end = var_28575_end_0, end_mask = var_28575_end_mask_0, x = q_129_cast_fp16)[name = tensor("op_28575_cast_fp16")]; + tensor k_259_perm_0 = const()[name = tensor("k_259_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_28582_begin_0 = const()[name = tensor("op_28582_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28582_end_0 = const()[name = tensor("op_28582_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_28582_end_mask_0 = const()[name = tensor("op_28582_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_259_cast_fp16 = transpose(perm = k_259_perm_0, x = k_257_cast_fp16)[name = tensor("transpose_75")]; + tensor var_28582_cast_fp16 = slice_by_index(begin = var_28582_begin_0, end = var_28582_end_0, end_mask = var_28582_end_mask_0, x = k_259_cast_fp16)[name = tensor("op_28582_cast_fp16")]; + tensor var_28586_begin_0 = const()[name = tensor("op_28586_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_28586_end_0 = const()[name = tensor("op_28586_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_28586_end_mask_0 = const()[name = tensor("op_28586_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28586_cast_fp16 = slice_by_index(begin = var_28586_begin_0, end = var_28586_end_0, end_mask = var_28586_end_mask_0, x = k_259_cast_fp16)[name = tensor("op_28586_cast_fp16")]; + tensor var_28590_begin_0 = const()[name = tensor("op_28590_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_28590_end_0 = const()[name = tensor("op_28590_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_28590_end_mask_0 = const()[name = tensor("op_28590_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28590_cast_fp16 = slice_by_index(begin = var_28590_begin_0, end = var_28590_end_0, end_mask = var_28590_end_mask_0, x = k_259_cast_fp16)[name = tensor("op_28590_cast_fp16")]; + tensor var_28594_begin_0 = const()[name = tensor("op_28594_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_28594_end_0 = const()[name = tensor("op_28594_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_28594_end_mask_0 = const()[name = tensor("op_28594_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28594_cast_fp16 = slice_by_index(begin = var_28594_begin_0, end = var_28594_end_0, end_mask = var_28594_end_mask_0, x = k_259_cast_fp16)[name = tensor("op_28594_cast_fp16")]; + tensor var_28598_begin_0 = const()[name = tensor("op_28598_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_28598_end_0 = const()[name = tensor("op_28598_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_28598_end_mask_0 = const()[name = tensor("op_28598_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28598_cast_fp16 = slice_by_index(begin = var_28598_begin_0, end = var_28598_end_0, end_mask = var_28598_end_mask_0, x = k_259_cast_fp16)[name = tensor("op_28598_cast_fp16")]; + tensor var_28602_begin_0 = const()[name = tensor("op_28602_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_28602_end_0 = const()[name = tensor("op_28602_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_28602_end_mask_0 = const()[name = tensor("op_28602_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28602_cast_fp16 = slice_by_index(begin = var_28602_begin_0, end = var_28602_end_0, end_mask = var_28602_end_mask_0, x = k_259_cast_fp16)[name = tensor("op_28602_cast_fp16")]; + tensor var_28606_begin_0 = const()[name = tensor("op_28606_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_28606_end_0 = const()[name = tensor("op_28606_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_28606_end_mask_0 = const()[name = tensor("op_28606_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28606_cast_fp16 = slice_by_index(begin = var_28606_begin_0, end = var_28606_end_0, end_mask = var_28606_end_mask_0, x = k_259_cast_fp16)[name = tensor("op_28606_cast_fp16")]; + tensor var_28610_begin_0 = const()[name = tensor("op_28610_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_28610_end_0 = const()[name = tensor("op_28610_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_28610_end_mask_0 = const()[name = tensor("op_28610_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28610_cast_fp16 = slice_by_index(begin = var_28610_begin_0, end = var_28610_end_0, end_mask = var_28610_end_mask_0, x = k_259_cast_fp16)[name = tensor("op_28610_cast_fp16")]; + tensor var_28614_begin_0 = const()[name = tensor("op_28614_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_28614_end_0 = const()[name = tensor("op_28614_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_28614_end_mask_0 = const()[name = tensor("op_28614_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28614_cast_fp16 = slice_by_index(begin = var_28614_begin_0, end = var_28614_end_0, end_mask = var_28614_end_mask_0, x = k_259_cast_fp16)[name = tensor("op_28614_cast_fp16")]; + tensor var_28618_begin_0 = const()[name = tensor("op_28618_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_28618_end_0 = const()[name = tensor("op_28618_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_28618_end_mask_0 = const()[name = tensor("op_28618_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28618_cast_fp16 = slice_by_index(begin = var_28618_begin_0, end = var_28618_end_0, end_mask = var_28618_end_mask_0, x = k_259_cast_fp16)[name = tensor("op_28618_cast_fp16")]; + tensor var_28622_begin_0 = const()[name = tensor("op_28622_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_28622_end_0 = const()[name = tensor("op_28622_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_28622_end_mask_0 = const()[name = tensor("op_28622_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28622_cast_fp16 = slice_by_index(begin = var_28622_begin_0, end = var_28622_end_0, end_mask = var_28622_end_mask_0, x = k_259_cast_fp16)[name = tensor("op_28622_cast_fp16")]; + tensor var_28626_begin_0 = const()[name = tensor("op_28626_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_28626_end_0 = const()[name = tensor("op_28626_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_28626_end_mask_0 = const()[name = tensor("op_28626_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28626_cast_fp16 = slice_by_index(begin = var_28626_begin_0, end = var_28626_end_0, end_mask = var_28626_end_mask_0, x = k_259_cast_fp16)[name = tensor("op_28626_cast_fp16")]; + tensor var_28630_begin_0 = const()[name = tensor("op_28630_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_28630_end_0 = const()[name = tensor("op_28630_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_28630_end_mask_0 = const()[name = tensor("op_28630_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28630_cast_fp16 = slice_by_index(begin = var_28630_begin_0, end = var_28630_end_0, end_mask = var_28630_end_mask_0, x = k_259_cast_fp16)[name = tensor("op_28630_cast_fp16")]; + tensor var_28634_begin_0 = const()[name = tensor("op_28634_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_28634_end_0 = const()[name = tensor("op_28634_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_28634_end_mask_0 = const()[name = tensor("op_28634_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28634_cast_fp16 = slice_by_index(begin = var_28634_begin_0, end = var_28634_end_0, end_mask = var_28634_end_mask_0, x = k_259_cast_fp16)[name = tensor("op_28634_cast_fp16")]; + tensor var_28638_begin_0 = const()[name = tensor("op_28638_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_28638_end_0 = const()[name = tensor("op_28638_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_28638_end_mask_0 = const()[name = tensor("op_28638_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28638_cast_fp16 = slice_by_index(begin = var_28638_begin_0, end = var_28638_end_0, end_mask = var_28638_end_mask_0, x = k_259_cast_fp16)[name = tensor("op_28638_cast_fp16")]; + tensor var_28642_begin_0 = const()[name = tensor("op_28642_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_28642_end_0 = const()[name = tensor("op_28642_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_28642_end_mask_0 = const()[name = tensor("op_28642_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28642_cast_fp16 = slice_by_index(begin = var_28642_begin_0, end = var_28642_end_0, end_mask = var_28642_end_mask_0, x = k_259_cast_fp16)[name = tensor("op_28642_cast_fp16")]; + tensor var_28646_begin_0 = const()[name = tensor("op_28646_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_28646_end_0 = const()[name = tensor("op_28646_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_28646_end_mask_0 = const()[name = tensor("op_28646_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28646_cast_fp16 = slice_by_index(begin = var_28646_begin_0, end = var_28646_end_0, end_mask = var_28646_end_mask_0, x = k_259_cast_fp16)[name = tensor("op_28646_cast_fp16")]; + tensor var_28650_begin_0 = const()[name = tensor("op_28650_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_28650_end_0 = const()[name = tensor("op_28650_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_28650_end_mask_0 = const()[name = tensor("op_28650_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28650_cast_fp16 = slice_by_index(begin = var_28650_begin_0, end = var_28650_end_0, end_mask = var_28650_end_mask_0, x = k_259_cast_fp16)[name = tensor("op_28650_cast_fp16")]; + tensor var_28654_begin_0 = const()[name = tensor("op_28654_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_28654_end_0 = const()[name = tensor("op_28654_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_28654_end_mask_0 = const()[name = tensor("op_28654_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28654_cast_fp16 = slice_by_index(begin = var_28654_begin_0, end = var_28654_end_0, end_mask = var_28654_end_mask_0, x = k_259_cast_fp16)[name = tensor("op_28654_cast_fp16")]; + tensor var_28658_begin_0 = const()[name = tensor("op_28658_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_28658_end_0 = const()[name = tensor("op_28658_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_28658_end_mask_0 = const()[name = tensor("op_28658_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28658_cast_fp16 = slice_by_index(begin = var_28658_begin_0, end = var_28658_end_0, end_mask = var_28658_end_mask_0, x = k_259_cast_fp16)[name = tensor("op_28658_cast_fp16")]; + tensor var_28660_begin_0 = const()[name = tensor("op_28660_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28660_end_0 = const()[name = tensor("op_28660_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_28660_end_mask_0 = const()[name = tensor("op_28660_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28660_cast_fp16 = slice_by_index(begin = var_28660_begin_0, end = var_28660_end_0, end_mask = var_28660_end_mask_0, x = v_129_cast_fp16)[name = tensor("op_28660_cast_fp16")]; + tensor var_28664_begin_0 = const()[name = tensor("op_28664_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_28664_end_0 = const()[name = tensor("op_28664_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_28664_end_mask_0 = const()[name = tensor("op_28664_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28664_cast_fp16 = slice_by_index(begin = var_28664_begin_0, end = var_28664_end_0, end_mask = var_28664_end_mask_0, x = v_129_cast_fp16)[name = tensor("op_28664_cast_fp16")]; + tensor var_28668_begin_0 = const()[name = tensor("op_28668_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_28668_end_0 = const()[name = tensor("op_28668_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_28668_end_mask_0 = const()[name = tensor("op_28668_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28668_cast_fp16 = slice_by_index(begin = var_28668_begin_0, end = var_28668_end_0, end_mask = var_28668_end_mask_0, x = v_129_cast_fp16)[name = tensor("op_28668_cast_fp16")]; + tensor var_28672_begin_0 = const()[name = tensor("op_28672_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_28672_end_0 = const()[name = tensor("op_28672_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_28672_end_mask_0 = const()[name = tensor("op_28672_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28672_cast_fp16 = slice_by_index(begin = var_28672_begin_0, end = var_28672_end_0, end_mask = var_28672_end_mask_0, x = v_129_cast_fp16)[name = tensor("op_28672_cast_fp16")]; + tensor var_28676_begin_0 = const()[name = tensor("op_28676_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_28676_end_0 = const()[name = tensor("op_28676_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_28676_end_mask_0 = const()[name = tensor("op_28676_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28676_cast_fp16 = slice_by_index(begin = var_28676_begin_0, end = var_28676_end_0, end_mask = var_28676_end_mask_0, x = v_129_cast_fp16)[name = tensor("op_28676_cast_fp16")]; + tensor var_28680_begin_0 = const()[name = tensor("op_28680_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_28680_end_0 = const()[name = tensor("op_28680_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_28680_end_mask_0 = const()[name = tensor("op_28680_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28680_cast_fp16 = slice_by_index(begin = var_28680_begin_0, end = var_28680_end_0, end_mask = var_28680_end_mask_0, x = v_129_cast_fp16)[name = tensor("op_28680_cast_fp16")]; + tensor var_28684_begin_0 = const()[name = tensor("op_28684_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_28684_end_0 = const()[name = tensor("op_28684_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_28684_end_mask_0 = const()[name = tensor("op_28684_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28684_cast_fp16 = slice_by_index(begin = var_28684_begin_0, end = var_28684_end_0, end_mask = var_28684_end_mask_0, x = v_129_cast_fp16)[name = tensor("op_28684_cast_fp16")]; + tensor var_28688_begin_0 = const()[name = tensor("op_28688_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_28688_end_0 = const()[name = tensor("op_28688_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_28688_end_mask_0 = const()[name = tensor("op_28688_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28688_cast_fp16 = slice_by_index(begin = var_28688_begin_0, end = var_28688_end_0, end_mask = var_28688_end_mask_0, x = v_129_cast_fp16)[name = tensor("op_28688_cast_fp16")]; + tensor var_28692_begin_0 = const()[name = tensor("op_28692_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_28692_end_0 = const()[name = tensor("op_28692_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_28692_end_mask_0 = const()[name = tensor("op_28692_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28692_cast_fp16 = slice_by_index(begin = var_28692_begin_0, end = var_28692_end_0, end_mask = var_28692_end_mask_0, x = v_129_cast_fp16)[name = tensor("op_28692_cast_fp16")]; + tensor var_28696_begin_0 = const()[name = tensor("op_28696_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_28696_end_0 = const()[name = tensor("op_28696_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_28696_end_mask_0 = const()[name = tensor("op_28696_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28696_cast_fp16 = slice_by_index(begin = var_28696_begin_0, end = var_28696_end_0, end_mask = var_28696_end_mask_0, x = v_129_cast_fp16)[name = tensor("op_28696_cast_fp16")]; + tensor var_28700_begin_0 = const()[name = tensor("op_28700_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_28700_end_0 = const()[name = tensor("op_28700_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_28700_end_mask_0 = const()[name = tensor("op_28700_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28700_cast_fp16 = slice_by_index(begin = var_28700_begin_0, end = var_28700_end_0, end_mask = var_28700_end_mask_0, x = v_129_cast_fp16)[name = tensor("op_28700_cast_fp16")]; + tensor var_28704_begin_0 = const()[name = tensor("op_28704_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_28704_end_0 = const()[name = tensor("op_28704_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_28704_end_mask_0 = const()[name = tensor("op_28704_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28704_cast_fp16 = slice_by_index(begin = var_28704_begin_0, end = var_28704_end_0, end_mask = var_28704_end_mask_0, x = v_129_cast_fp16)[name = tensor("op_28704_cast_fp16")]; + tensor var_28708_begin_0 = const()[name = tensor("op_28708_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_28708_end_0 = const()[name = tensor("op_28708_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_28708_end_mask_0 = const()[name = tensor("op_28708_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28708_cast_fp16 = slice_by_index(begin = var_28708_begin_0, end = var_28708_end_0, end_mask = var_28708_end_mask_0, x = v_129_cast_fp16)[name = tensor("op_28708_cast_fp16")]; + tensor var_28712_begin_0 = const()[name = tensor("op_28712_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_28712_end_0 = const()[name = tensor("op_28712_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_28712_end_mask_0 = const()[name = tensor("op_28712_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28712_cast_fp16 = slice_by_index(begin = var_28712_begin_0, end = var_28712_end_0, end_mask = var_28712_end_mask_0, x = v_129_cast_fp16)[name = tensor("op_28712_cast_fp16")]; + tensor var_28716_begin_0 = const()[name = tensor("op_28716_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_28716_end_0 = const()[name = tensor("op_28716_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_28716_end_mask_0 = const()[name = tensor("op_28716_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28716_cast_fp16 = slice_by_index(begin = var_28716_begin_0, end = var_28716_end_0, end_mask = var_28716_end_mask_0, x = v_129_cast_fp16)[name = tensor("op_28716_cast_fp16")]; + tensor var_28720_begin_0 = const()[name = tensor("op_28720_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_28720_end_0 = const()[name = tensor("op_28720_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_28720_end_mask_0 = const()[name = tensor("op_28720_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28720_cast_fp16 = slice_by_index(begin = var_28720_begin_0, end = var_28720_end_0, end_mask = var_28720_end_mask_0, x = v_129_cast_fp16)[name = tensor("op_28720_cast_fp16")]; + tensor var_28724_begin_0 = const()[name = tensor("op_28724_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_28724_end_0 = const()[name = tensor("op_28724_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_28724_end_mask_0 = const()[name = tensor("op_28724_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28724_cast_fp16 = slice_by_index(begin = var_28724_begin_0, end = var_28724_end_0, end_mask = var_28724_end_mask_0, x = v_129_cast_fp16)[name = tensor("op_28724_cast_fp16")]; + tensor var_28728_begin_0 = const()[name = tensor("op_28728_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_28728_end_0 = const()[name = tensor("op_28728_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_28728_end_mask_0 = const()[name = tensor("op_28728_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28728_cast_fp16 = slice_by_index(begin = var_28728_begin_0, end = var_28728_end_0, end_mask = var_28728_end_mask_0, x = v_129_cast_fp16)[name = tensor("op_28728_cast_fp16")]; + tensor var_28732_begin_0 = const()[name = tensor("op_28732_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_28732_end_0 = const()[name = tensor("op_28732_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_28732_end_mask_0 = const()[name = tensor("op_28732_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28732_cast_fp16 = slice_by_index(begin = var_28732_begin_0, end = var_28732_end_0, end_mask = var_28732_end_mask_0, x = v_129_cast_fp16)[name = tensor("op_28732_cast_fp16")]; + tensor var_28736_begin_0 = const()[name = tensor("op_28736_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_28736_end_0 = const()[name = tensor("op_28736_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_28736_end_mask_0 = const()[name = tensor("op_28736_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28736_cast_fp16 = slice_by_index(begin = var_28736_begin_0, end = var_28736_end_0, end_mask = var_28736_end_mask_0, x = v_129_cast_fp16)[name = tensor("op_28736_cast_fp16")]; + tensor var_28740_equation_0 = const()[name = tensor("op_28740_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28740_cast_fp16 = einsum(equation = var_28740_equation_0, values = (var_28582_cast_fp16, var_28499_cast_fp16))[name = tensor("op_28740_cast_fp16")]; + tensor var_28741_to_fp16 = const()[name = tensor("op_28741_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2401_cast_fp16 = mul(x = var_28740_cast_fp16, y = var_28741_to_fp16)[name = tensor("aw_2401_cast_fp16")]; + tensor var_28744_equation_0 = const()[name = tensor("op_28744_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28744_cast_fp16 = einsum(equation = var_28744_equation_0, values = (var_28586_cast_fp16, var_28503_cast_fp16))[name = tensor("op_28744_cast_fp16")]; + tensor var_28745_to_fp16 = const()[name = tensor("op_28745_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2403_cast_fp16 = mul(x = var_28744_cast_fp16, y = var_28745_to_fp16)[name = tensor("aw_2403_cast_fp16")]; + tensor var_28748_equation_0 = const()[name = tensor("op_28748_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28748_cast_fp16 = einsum(equation = var_28748_equation_0, values = (var_28590_cast_fp16, var_28507_cast_fp16))[name = tensor("op_28748_cast_fp16")]; + tensor var_28749_to_fp16 = const()[name = tensor("op_28749_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2405_cast_fp16 = mul(x = var_28748_cast_fp16, y = var_28749_to_fp16)[name = tensor("aw_2405_cast_fp16")]; + tensor var_28752_equation_0 = const()[name = tensor("op_28752_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28752_cast_fp16 = einsum(equation = var_28752_equation_0, values = (var_28594_cast_fp16, var_28511_cast_fp16))[name = tensor("op_28752_cast_fp16")]; + tensor var_28753_to_fp16 = const()[name = tensor("op_28753_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2407_cast_fp16 = mul(x = var_28752_cast_fp16, y = var_28753_to_fp16)[name = tensor("aw_2407_cast_fp16")]; + tensor var_28756_equation_0 = const()[name = tensor("op_28756_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28756_cast_fp16 = einsum(equation = var_28756_equation_0, values = (var_28598_cast_fp16, var_28515_cast_fp16))[name = tensor("op_28756_cast_fp16")]; + tensor var_28757_to_fp16 = const()[name = tensor("op_28757_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2409_cast_fp16 = mul(x = var_28756_cast_fp16, y = var_28757_to_fp16)[name = tensor("aw_2409_cast_fp16")]; + tensor var_28760_equation_0 = const()[name = tensor("op_28760_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28760_cast_fp16 = einsum(equation = var_28760_equation_0, values = (var_28602_cast_fp16, var_28519_cast_fp16))[name = tensor("op_28760_cast_fp16")]; + tensor var_28761_to_fp16 = const()[name = tensor("op_28761_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2411_cast_fp16 = mul(x = var_28760_cast_fp16, y = var_28761_to_fp16)[name = tensor("aw_2411_cast_fp16")]; + tensor var_28764_equation_0 = const()[name = tensor("op_28764_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28764_cast_fp16 = einsum(equation = var_28764_equation_0, values = (var_28606_cast_fp16, var_28523_cast_fp16))[name = tensor("op_28764_cast_fp16")]; + tensor var_28765_to_fp16 = const()[name = tensor("op_28765_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2413_cast_fp16 = mul(x = var_28764_cast_fp16, y = var_28765_to_fp16)[name = tensor("aw_2413_cast_fp16")]; + tensor var_28768_equation_0 = const()[name = tensor("op_28768_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28768_cast_fp16 = einsum(equation = var_28768_equation_0, values = (var_28610_cast_fp16, var_28527_cast_fp16))[name = tensor("op_28768_cast_fp16")]; + tensor var_28769_to_fp16 = const()[name = tensor("op_28769_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2415_cast_fp16 = mul(x = var_28768_cast_fp16, y = var_28769_to_fp16)[name = tensor("aw_2415_cast_fp16")]; + tensor var_28772_equation_0 = const()[name = tensor("op_28772_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28772_cast_fp16 = einsum(equation = var_28772_equation_0, values = (var_28614_cast_fp16, var_28531_cast_fp16))[name = tensor("op_28772_cast_fp16")]; + tensor var_28773_to_fp16 = const()[name = tensor("op_28773_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2417_cast_fp16 = mul(x = var_28772_cast_fp16, y = var_28773_to_fp16)[name = tensor("aw_2417_cast_fp16")]; + tensor var_28776_equation_0 = const()[name = tensor("op_28776_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28776_cast_fp16 = einsum(equation = var_28776_equation_0, values = (var_28618_cast_fp16, var_28535_cast_fp16))[name = tensor("op_28776_cast_fp16")]; + tensor var_28777_to_fp16 = const()[name = tensor("op_28777_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2419_cast_fp16 = mul(x = var_28776_cast_fp16, y = var_28777_to_fp16)[name = tensor("aw_2419_cast_fp16")]; + tensor var_28780_equation_0 = const()[name = tensor("op_28780_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28780_cast_fp16 = einsum(equation = var_28780_equation_0, values = (var_28622_cast_fp16, var_28539_cast_fp16))[name = tensor("op_28780_cast_fp16")]; + tensor var_28781_to_fp16 = const()[name = tensor("op_28781_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2421_cast_fp16 = mul(x = var_28780_cast_fp16, y = var_28781_to_fp16)[name = tensor("aw_2421_cast_fp16")]; + tensor var_28784_equation_0 = const()[name = tensor("op_28784_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28784_cast_fp16 = einsum(equation = var_28784_equation_0, values = (var_28626_cast_fp16, var_28543_cast_fp16))[name = tensor("op_28784_cast_fp16")]; + tensor var_28785_to_fp16 = const()[name = tensor("op_28785_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2423_cast_fp16 = mul(x = var_28784_cast_fp16, y = var_28785_to_fp16)[name = tensor("aw_2423_cast_fp16")]; + tensor var_28788_equation_0 = const()[name = tensor("op_28788_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28788_cast_fp16 = einsum(equation = var_28788_equation_0, values = (var_28630_cast_fp16, var_28547_cast_fp16))[name = tensor("op_28788_cast_fp16")]; + tensor var_28789_to_fp16 = const()[name = tensor("op_28789_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2425_cast_fp16 = mul(x = var_28788_cast_fp16, y = var_28789_to_fp16)[name = tensor("aw_2425_cast_fp16")]; + tensor var_28792_equation_0 = const()[name = tensor("op_28792_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28792_cast_fp16 = einsum(equation = var_28792_equation_0, values = (var_28634_cast_fp16, var_28551_cast_fp16))[name = tensor("op_28792_cast_fp16")]; + tensor var_28793_to_fp16 = const()[name = tensor("op_28793_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2427_cast_fp16 = mul(x = var_28792_cast_fp16, y = var_28793_to_fp16)[name = tensor("aw_2427_cast_fp16")]; + tensor var_28796_equation_0 = const()[name = tensor("op_28796_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28796_cast_fp16 = einsum(equation = var_28796_equation_0, values = (var_28638_cast_fp16, var_28555_cast_fp16))[name = tensor("op_28796_cast_fp16")]; + tensor var_28797_to_fp16 = const()[name = tensor("op_28797_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2429_cast_fp16 = mul(x = var_28796_cast_fp16, y = var_28797_to_fp16)[name = tensor("aw_2429_cast_fp16")]; + tensor var_28800_equation_0 = const()[name = tensor("op_28800_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28800_cast_fp16 = einsum(equation = var_28800_equation_0, values = (var_28642_cast_fp16, var_28559_cast_fp16))[name = tensor("op_28800_cast_fp16")]; + tensor var_28801_to_fp16 = const()[name = tensor("op_28801_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2431_cast_fp16 = mul(x = var_28800_cast_fp16, y = var_28801_to_fp16)[name = tensor("aw_2431_cast_fp16")]; + tensor var_28804_equation_0 = const()[name = tensor("op_28804_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28804_cast_fp16 = einsum(equation = var_28804_equation_0, values = (var_28646_cast_fp16, var_28563_cast_fp16))[name = tensor("op_28804_cast_fp16")]; + tensor var_28805_to_fp16 = const()[name = tensor("op_28805_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2433_cast_fp16 = mul(x = var_28804_cast_fp16, y = var_28805_to_fp16)[name = tensor("aw_2433_cast_fp16")]; + tensor var_28808_equation_0 = const()[name = tensor("op_28808_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28808_cast_fp16 = einsum(equation = var_28808_equation_0, values = (var_28650_cast_fp16, var_28567_cast_fp16))[name = tensor("op_28808_cast_fp16")]; + tensor var_28809_to_fp16 = const()[name = tensor("op_28809_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2435_cast_fp16 = mul(x = var_28808_cast_fp16, y = var_28809_to_fp16)[name = tensor("aw_2435_cast_fp16")]; + tensor var_28812_equation_0 = const()[name = tensor("op_28812_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28812_cast_fp16 = einsum(equation = var_28812_equation_0, values = (var_28654_cast_fp16, var_28571_cast_fp16))[name = tensor("op_28812_cast_fp16")]; + tensor var_28813_to_fp16 = const()[name = tensor("op_28813_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2437_cast_fp16 = mul(x = var_28812_cast_fp16, y = var_28813_to_fp16)[name = tensor("aw_2437_cast_fp16")]; + tensor var_28816_equation_0 = const()[name = tensor("op_28816_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28816_cast_fp16 = einsum(equation = var_28816_equation_0, values = (var_28658_cast_fp16, var_28575_cast_fp16))[name = tensor("op_28816_cast_fp16")]; + tensor var_28817_to_fp16 = const()[name = tensor("op_28817_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2439_cast_fp16 = mul(x = var_28816_cast_fp16, y = var_28817_to_fp16)[name = tensor("aw_2439_cast_fp16")]; + tensor var_28819_cast_fp16 = softmax(axis = var_21077, x = aw_2401_cast_fp16)[name = tensor("op_28819_cast_fp16")]; + tensor var_28820_cast_fp16 = softmax(axis = var_21077, x = aw_2403_cast_fp16)[name = tensor("op_28820_cast_fp16")]; + tensor var_28821_cast_fp16 = softmax(axis = var_21077, x = aw_2405_cast_fp16)[name = tensor("op_28821_cast_fp16")]; + tensor var_28822_cast_fp16 = softmax(axis = var_21077, x = aw_2407_cast_fp16)[name = tensor("op_28822_cast_fp16")]; + tensor var_28823_cast_fp16 = softmax(axis = var_21077, x = aw_2409_cast_fp16)[name = tensor("op_28823_cast_fp16")]; + tensor var_28824_cast_fp16 = softmax(axis = var_21077, x = aw_2411_cast_fp16)[name = tensor("op_28824_cast_fp16")]; + tensor var_28825_cast_fp16 = softmax(axis = var_21077, x = aw_2413_cast_fp16)[name = tensor("op_28825_cast_fp16")]; + tensor var_28826_cast_fp16 = softmax(axis = var_21077, x = aw_2415_cast_fp16)[name = tensor("op_28826_cast_fp16")]; + tensor var_28827_cast_fp16 = softmax(axis = var_21077, x = aw_2417_cast_fp16)[name = tensor("op_28827_cast_fp16")]; + tensor var_28828_cast_fp16 = softmax(axis = var_21077, x = aw_2419_cast_fp16)[name = tensor("op_28828_cast_fp16")]; + tensor var_28829_cast_fp16 = softmax(axis = var_21077, x = aw_2421_cast_fp16)[name = tensor("op_28829_cast_fp16")]; + tensor var_28830_cast_fp16 = softmax(axis = var_21077, x = aw_2423_cast_fp16)[name = tensor("op_28830_cast_fp16")]; + tensor var_28831_cast_fp16 = softmax(axis = var_21077, x = aw_2425_cast_fp16)[name = tensor("op_28831_cast_fp16")]; + tensor var_28832_cast_fp16 = softmax(axis = var_21077, x = aw_2427_cast_fp16)[name = tensor("op_28832_cast_fp16")]; + tensor var_28833_cast_fp16 = softmax(axis = var_21077, x = aw_2429_cast_fp16)[name = tensor("op_28833_cast_fp16")]; + tensor var_28834_cast_fp16 = softmax(axis = var_21077, x = aw_2431_cast_fp16)[name = tensor("op_28834_cast_fp16")]; + tensor var_28835_cast_fp16 = softmax(axis = var_21077, x = aw_2433_cast_fp16)[name = tensor("op_28835_cast_fp16")]; + tensor var_28836_cast_fp16 = softmax(axis = var_21077, x = aw_2435_cast_fp16)[name = tensor("op_28836_cast_fp16")]; + tensor var_28837_cast_fp16 = softmax(axis = var_21077, x = aw_2437_cast_fp16)[name = tensor("op_28837_cast_fp16")]; + tensor var_28838_cast_fp16 = softmax(axis = var_21077, x = aw_2439_cast_fp16)[name = tensor("op_28838_cast_fp16")]; + tensor var_28840_equation_0 = const()[name = tensor("op_28840_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28840_cast_fp16 = einsum(equation = var_28840_equation_0, values = (var_28660_cast_fp16, var_28819_cast_fp16))[name = tensor("op_28840_cast_fp16")]; + tensor var_28842_equation_0 = const()[name = tensor("op_28842_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28842_cast_fp16 = einsum(equation = var_28842_equation_0, values = (var_28664_cast_fp16, var_28820_cast_fp16))[name = tensor("op_28842_cast_fp16")]; + tensor var_28844_equation_0 = const()[name = tensor("op_28844_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28844_cast_fp16 = einsum(equation = var_28844_equation_0, values = (var_28668_cast_fp16, var_28821_cast_fp16))[name = tensor("op_28844_cast_fp16")]; + tensor var_28846_equation_0 = const()[name = tensor("op_28846_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28846_cast_fp16 = einsum(equation = var_28846_equation_0, values = (var_28672_cast_fp16, var_28822_cast_fp16))[name = tensor("op_28846_cast_fp16")]; + tensor var_28848_equation_0 = const()[name = tensor("op_28848_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28848_cast_fp16 = einsum(equation = var_28848_equation_0, values = (var_28676_cast_fp16, var_28823_cast_fp16))[name = tensor("op_28848_cast_fp16")]; + tensor var_28850_equation_0 = const()[name = tensor("op_28850_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28850_cast_fp16 = einsum(equation = var_28850_equation_0, values = (var_28680_cast_fp16, var_28824_cast_fp16))[name = tensor("op_28850_cast_fp16")]; + tensor var_28852_equation_0 = const()[name = tensor("op_28852_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28852_cast_fp16 = einsum(equation = var_28852_equation_0, values = (var_28684_cast_fp16, var_28825_cast_fp16))[name = tensor("op_28852_cast_fp16")]; + tensor var_28854_equation_0 = const()[name = tensor("op_28854_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28854_cast_fp16 = einsum(equation = var_28854_equation_0, values = (var_28688_cast_fp16, var_28826_cast_fp16))[name = tensor("op_28854_cast_fp16")]; + tensor var_28856_equation_0 = const()[name = tensor("op_28856_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28856_cast_fp16 = einsum(equation = var_28856_equation_0, values = (var_28692_cast_fp16, var_28827_cast_fp16))[name = tensor("op_28856_cast_fp16")]; + tensor var_28858_equation_0 = const()[name = tensor("op_28858_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28858_cast_fp16 = einsum(equation = var_28858_equation_0, values = (var_28696_cast_fp16, var_28828_cast_fp16))[name = tensor("op_28858_cast_fp16")]; + tensor var_28860_equation_0 = const()[name = tensor("op_28860_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28860_cast_fp16 = einsum(equation = var_28860_equation_0, values = (var_28700_cast_fp16, var_28829_cast_fp16))[name = tensor("op_28860_cast_fp16")]; + tensor var_28862_equation_0 = const()[name = tensor("op_28862_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28862_cast_fp16 = einsum(equation = var_28862_equation_0, values = (var_28704_cast_fp16, var_28830_cast_fp16))[name = tensor("op_28862_cast_fp16")]; + tensor var_28864_equation_0 = const()[name = tensor("op_28864_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28864_cast_fp16 = einsum(equation = var_28864_equation_0, values = (var_28708_cast_fp16, var_28831_cast_fp16))[name = tensor("op_28864_cast_fp16")]; + tensor var_28866_equation_0 = const()[name = tensor("op_28866_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28866_cast_fp16 = einsum(equation = var_28866_equation_0, values = (var_28712_cast_fp16, var_28832_cast_fp16))[name = tensor("op_28866_cast_fp16")]; + tensor var_28868_equation_0 = const()[name = tensor("op_28868_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28868_cast_fp16 = einsum(equation = var_28868_equation_0, values = (var_28716_cast_fp16, var_28833_cast_fp16))[name = tensor("op_28868_cast_fp16")]; + tensor var_28870_equation_0 = const()[name = tensor("op_28870_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28870_cast_fp16 = einsum(equation = var_28870_equation_0, values = (var_28720_cast_fp16, var_28834_cast_fp16))[name = tensor("op_28870_cast_fp16")]; + tensor var_28872_equation_0 = const()[name = tensor("op_28872_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28872_cast_fp16 = einsum(equation = var_28872_equation_0, values = (var_28724_cast_fp16, var_28835_cast_fp16))[name = tensor("op_28872_cast_fp16")]; + tensor var_28874_equation_0 = const()[name = tensor("op_28874_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28874_cast_fp16 = einsum(equation = var_28874_equation_0, values = (var_28728_cast_fp16, var_28836_cast_fp16))[name = tensor("op_28874_cast_fp16")]; + tensor var_28876_equation_0 = const()[name = tensor("op_28876_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28876_cast_fp16 = einsum(equation = var_28876_equation_0, values = (var_28732_cast_fp16, var_28837_cast_fp16))[name = tensor("op_28876_cast_fp16")]; + tensor var_28878_equation_0 = const()[name = tensor("op_28878_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_28878_cast_fp16 = einsum(equation = var_28878_equation_0, values = (var_28736_cast_fp16, var_28838_cast_fp16))[name = tensor("op_28878_cast_fp16")]; + tensor input_391_interleave_0 = const()[name = tensor("input_391_interleave_0"), val = tensor(false)]; + tensor input_391_cast_fp16 = concat(axis = var_21077, interleave = input_391_interleave_0, values = (var_28840_cast_fp16, var_28842_cast_fp16, var_28844_cast_fp16, var_28846_cast_fp16, var_28848_cast_fp16, var_28850_cast_fp16, var_28852_cast_fp16, var_28854_cast_fp16, var_28856_cast_fp16, var_28858_cast_fp16, var_28860_cast_fp16, var_28862_cast_fp16, var_28864_cast_fp16, var_28866_cast_fp16, var_28868_cast_fp16, var_28870_cast_fp16, var_28872_cast_fp16, var_28874_cast_fp16, var_28876_cast_fp16, var_28878_cast_fp16))[name = tensor("input_391_cast_fp16")]; + tensor var_28884 = const()[name = tensor("op_28884"), val = tensor([1, 1])]; + tensor var_28886 = const()[name = tensor("op_28886"), val = tensor([1, 1])]; + tensor var_28888_pad_type_0 = const()[name = tensor("op_28888_pad_type_0"), val = tensor("custom")]; + tensor var_28888_pad_0 = const()[name = tensor("op_28888_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_8_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_8_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2292048576)))]; + tensor mid_block_attentions_0_transformer_blocks_8_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_8_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2295325440)))]; + tensor var_28888_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_8_attn1_to_out_0_bias_to_fp16, dilations = var_28886, groups = var_21077, pad = var_28888_pad_0, pad_type = var_28888_pad_type_0, strides = var_28884, weight = mid_block_attentions_0_transformer_blocks_8_attn1_to_out_0_weight_to_fp16, x = input_391_cast_fp16)[name = tensor("op_28888_cast_fp16")]; + tensor inputs_195_cast_fp16 = add(x = var_28888_cast_fp16, y = inputs_193_cast_fp16)[name = tensor("inputs_195_cast_fp16")]; + tensor hidden_states_259_axes_0 = const()[name = tensor("hidden_states_259_axes_0"), val = tensor([1])]; + tensor hidden_states_259_gamma_0_to_fp16 = const()[name = tensor("hidden_states_259_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2295328064)))]; + tensor hidden_states_259_beta_0_to_fp16 = const()[name = tensor("hidden_states_259_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2295330688)))]; + tensor var_28898_to_fp16 = const()[name = tensor("op_28898_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_259_cast_fp16 = layer_norm(axes = hidden_states_259_axes_0, beta = hidden_states_259_beta_0_to_fp16, epsilon = var_28898_to_fp16, gamma = hidden_states_259_gamma_0_to_fp16, x = inputs_195_cast_fp16)[name = tensor("hidden_states_259_cast_fp16")]; + tensor var_28913 = const()[name = tensor("op_28913"), val = tensor([1, 1])]; + tensor var_28915 = const()[name = tensor("op_28915"), val = tensor([1, 1])]; + tensor q_131_pad_type_0 = const()[name = tensor("q_131_pad_type_0"), val = tensor("custom")]; + tensor q_131_pad_0 = const()[name = tensor("q_131_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_8_attn2_to_q_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_8_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2295333312)))]; + tensor q_131_cast_fp16 = conv(dilations = var_28915, groups = var_21077, pad = q_131_pad_0, pad_type = q_131_pad_type_0, strides = var_28913, weight = mid_block_attentions_0_transformer_blocks_8_attn2_to_q_weight_to_fp16, x = hidden_states_259_cast_fp16)[name = tensor("q_131_cast_fp16")]; + tensor var_28919 = const()[name = tensor("op_28919"), val = tensor([1, 1])]; + tensor var_28921 = const()[name = tensor("op_28921"), val = tensor([1, 1])]; + tensor k_261_pad_type_0 = const()[name = tensor("k_261_pad_type_0"), val = tensor("custom")]; + tensor k_261_pad_0 = const()[name = tensor("k_261_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_8_attn2_to_k_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_8_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2298610176)))]; + tensor k_261_cast_fp16 = conv(dilations = var_28921, groups = var_21077, pad = k_261_pad_0, pad_type = k_261_pad_type_0, strides = var_28919, weight = mid_block_attentions_0_transformer_blocks_8_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_261_cast_fp16")]; + tensor var_28925 = const()[name = tensor("op_28925"), val = tensor([1, 1])]; + tensor var_28927 = const()[name = tensor("op_28927"), val = tensor([1, 1])]; + tensor v_131_pad_type_0 = const()[name = tensor("v_131_pad_type_0"), val = tensor("custom")]; + tensor v_131_pad_0 = const()[name = tensor("v_131_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_8_attn2_to_v_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_8_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2303853120)))]; + tensor v_131_cast_fp16 = conv(dilations = var_28927, groups = var_21077, pad = v_131_pad_0, pad_type = v_131_pad_type_0, strides = var_28925, weight = mid_block_attentions_0_transformer_blocks_8_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_131_cast_fp16")]; + tensor var_28931_begin_0 = const()[name = tensor("op_28931_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28931_end_0 = const()[name = tensor("op_28931_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_28931_end_mask_0 = const()[name = tensor("op_28931_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28931_cast_fp16 = slice_by_index(begin = var_28931_begin_0, end = var_28931_end_0, end_mask = var_28931_end_mask_0, x = q_131_cast_fp16)[name = tensor("op_28931_cast_fp16")]; + tensor var_28935_begin_0 = const()[name = tensor("op_28935_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_28935_end_0 = const()[name = tensor("op_28935_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_28935_end_mask_0 = const()[name = tensor("op_28935_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28935_cast_fp16 = slice_by_index(begin = var_28935_begin_0, end = var_28935_end_0, end_mask = var_28935_end_mask_0, x = q_131_cast_fp16)[name = tensor("op_28935_cast_fp16")]; + tensor var_28939_begin_0 = const()[name = tensor("op_28939_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_28939_end_0 = const()[name = tensor("op_28939_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_28939_end_mask_0 = const()[name = tensor("op_28939_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28939_cast_fp16 = slice_by_index(begin = var_28939_begin_0, end = var_28939_end_0, end_mask = var_28939_end_mask_0, x = q_131_cast_fp16)[name = tensor("op_28939_cast_fp16")]; + tensor var_28943_begin_0 = const()[name = tensor("op_28943_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_28943_end_0 = const()[name = tensor("op_28943_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_28943_end_mask_0 = const()[name = tensor("op_28943_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28943_cast_fp16 = slice_by_index(begin = var_28943_begin_0, end = var_28943_end_0, end_mask = var_28943_end_mask_0, x = q_131_cast_fp16)[name = tensor("op_28943_cast_fp16")]; + tensor var_28947_begin_0 = const()[name = tensor("op_28947_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_28947_end_0 = const()[name = tensor("op_28947_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_28947_end_mask_0 = const()[name = tensor("op_28947_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28947_cast_fp16 = slice_by_index(begin = var_28947_begin_0, end = var_28947_end_0, end_mask = var_28947_end_mask_0, x = q_131_cast_fp16)[name = tensor("op_28947_cast_fp16")]; + tensor var_28951_begin_0 = const()[name = tensor("op_28951_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_28951_end_0 = const()[name = tensor("op_28951_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_28951_end_mask_0 = const()[name = tensor("op_28951_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28951_cast_fp16 = slice_by_index(begin = var_28951_begin_0, end = var_28951_end_0, end_mask = var_28951_end_mask_0, x = q_131_cast_fp16)[name = tensor("op_28951_cast_fp16")]; + tensor var_28955_begin_0 = const()[name = tensor("op_28955_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_28955_end_0 = const()[name = tensor("op_28955_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_28955_end_mask_0 = const()[name = tensor("op_28955_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28955_cast_fp16 = slice_by_index(begin = var_28955_begin_0, end = var_28955_end_0, end_mask = var_28955_end_mask_0, x = q_131_cast_fp16)[name = tensor("op_28955_cast_fp16")]; + tensor var_28959_begin_0 = const()[name = tensor("op_28959_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_28959_end_0 = const()[name = tensor("op_28959_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_28959_end_mask_0 = const()[name = tensor("op_28959_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28959_cast_fp16 = slice_by_index(begin = var_28959_begin_0, end = var_28959_end_0, end_mask = var_28959_end_mask_0, x = q_131_cast_fp16)[name = tensor("op_28959_cast_fp16")]; + tensor var_28963_begin_0 = const()[name = tensor("op_28963_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_28963_end_0 = const()[name = tensor("op_28963_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_28963_end_mask_0 = const()[name = tensor("op_28963_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28963_cast_fp16 = slice_by_index(begin = var_28963_begin_0, end = var_28963_end_0, end_mask = var_28963_end_mask_0, x = q_131_cast_fp16)[name = tensor("op_28963_cast_fp16")]; + tensor var_28967_begin_0 = const()[name = tensor("op_28967_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_28967_end_0 = const()[name = tensor("op_28967_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_28967_end_mask_0 = const()[name = tensor("op_28967_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28967_cast_fp16 = slice_by_index(begin = var_28967_begin_0, end = var_28967_end_0, end_mask = var_28967_end_mask_0, x = q_131_cast_fp16)[name = tensor("op_28967_cast_fp16")]; + tensor var_28971_begin_0 = const()[name = tensor("op_28971_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_28971_end_0 = const()[name = tensor("op_28971_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_28971_end_mask_0 = const()[name = tensor("op_28971_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28971_cast_fp16 = slice_by_index(begin = var_28971_begin_0, end = var_28971_end_0, end_mask = var_28971_end_mask_0, x = q_131_cast_fp16)[name = tensor("op_28971_cast_fp16")]; + tensor var_28975_begin_0 = const()[name = tensor("op_28975_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_28975_end_0 = const()[name = tensor("op_28975_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_28975_end_mask_0 = const()[name = tensor("op_28975_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28975_cast_fp16 = slice_by_index(begin = var_28975_begin_0, end = var_28975_end_0, end_mask = var_28975_end_mask_0, x = q_131_cast_fp16)[name = tensor("op_28975_cast_fp16")]; + tensor var_28979_begin_0 = const()[name = tensor("op_28979_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_28979_end_0 = const()[name = tensor("op_28979_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_28979_end_mask_0 = const()[name = tensor("op_28979_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28979_cast_fp16 = slice_by_index(begin = var_28979_begin_0, end = var_28979_end_0, end_mask = var_28979_end_mask_0, x = q_131_cast_fp16)[name = tensor("op_28979_cast_fp16")]; + tensor var_28983_begin_0 = const()[name = tensor("op_28983_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_28983_end_0 = const()[name = tensor("op_28983_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_28983_end_mask_0 = const()[name = tensor("op_28983_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28983_cast_fp16 = slice_by_index(begin = var_28983_begin_0, end = var_28983_end_0, end_mask = var_28983_end_mask_0, x = q_131_cast_fp16)[name = tensor("op_28983_cast_fp16")]; + tensor var_28987_begin_0 = const()[name = tensor("op_28987_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_28987_end_0 = const()[name = tensor("op_28987_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_28987_end_mask_0 = const()[name = tensor("op_28987_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28987_cast_fp16 = slice_by_index(begin = var_28987_begin_0, end = var_28987_end_0, end_mask = var_28987_end_mask_0, x = q_131_cast_fp16)[name = tensor("op_28987_cast_fp16")]; + tensor var_28991_begin_0 = const()[name = tensor("op_28991_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_28991_end_0 = const()[name = tensor("op_28991_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_28991_end_mask_0 = const()[name = tensor("op_28991_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28991_cast_fp16 = slice_by_index(begin = var_28991_begin_0, end = var_28991_end_0, end_mask = var_28991_end_mask_0, x = q_131_cast_fp16)[name = tensor("op_28991_cast_fp16")]; + tensor var_28995_begin_0 = const()[name = tensor("op_28995_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_28995_end_0 = const()[name = tensor("op_28995_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_28995_end_mask_0 = const()[name = tensor("op_28995_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28995_cast_fp16 = slice_by_index(begin = var_28995_begin_0, end = var_28995_end_0, end_mask = var_28995_end_mask_0, x = q_131_cast_fp16)[name = tensor("op_28995_cast_fp16")]; + tensor var_28999_begin_0 = const()[name = tensor("op_28999_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_28999_end_0 = const()[name = tensor("op_28999_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_28999_end_mask_0 = const()[name = tensor("op_28999_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28999_cast_fp16 = slice_by_index(begin = var_28999_begin_0, end = var_28999_end_0, end_mask = var_28999_end_mask_0, x = q_131_cast_fp16)[name = tensor("op_28999_cast_fp16")]; + tensor var_29003_begin_0 = const()[name = tensor("op_29003_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_29003_end_0 = const()[name = tensor("op_29003_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_29003_end_mask_0 = const()[name = tensor("op_29003_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29003_cast_fp16 = slice_by_index(begin = var_29003_begin_0, end = var_29003_end_0, end_mask = var_29003_end_mask_0, x = q_131_cast_fp16)[name = tensor("op_29003_cast_fp16")]; + tensor var_29007_begin_0 = const()[name = tensor("op_29007_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_29007_end_0 = const()[name = tensor("op_29007_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_29007_end_mask_0 = const()[name = tensor("op_29007_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29007_cast_fp16 = slice_by_index(begin = var_29007_begin_0, end = var_29007_end_0, end_mask = var_29007_end_mask_0, x = q_131_cast_fp16)[name = tensor("op_29007_cast_fp16")]; + tensor k_263_perm_0 = const()[name = tensor("k_263_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_29014_begin_0 = const()[name = tensor("op_29014_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29014_end_0 = const()[name = tensor("op_29014_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_29014_end_mask_0 = const()[name = tensor("op_29014_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_263_cast_fp16 = transpose(perm = k_263_perm_0, x = k_261_cast_fp16)[name = tensor("transpose_74")]; + tensor var_29014_cast_fp16 = slice_by_index(begin = var_29014_begin_0, end = var_29014_end_0, end_mask = var_29014_end_mask_0, x = k_263_cast_fp16)[name = tensor("op_29014_cast_fp16")]; + tensor var_29018_begin_0 = const()[name = tensor("op_29018_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_29018_end_0 = const()[name = tensor("op_29018_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_29018_end_mask_0 = const()[name = tensor("op_29018_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29018_cast_fp16 = slice_by_index(begin = var_29018_begin_0, end = var_29018_end_0, end_mask = var_29018_end_mask_0, x = k_263_cast_fp16)[name = tensor("op_29018_cast_fp16")]; + tensor var_29022_begin_0 = const()[name = tensor("op_29022_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_29022_end_0 = const()[name = tensor("op_29022_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_29022_end_mask_0 = const()[name = tensor("op_29022_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29022_cast_fp16 = slice_by_index(begin = var_29022_begin_0, end = var_29022_end_0, end_mask = var_29022_end_mask_0, x = k_263_cast_fp16)[name = tensor("op_29022_cast_fp16")]; + tensor var_29026_begin_0 = const()[name = tensor("op_29026_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_29026_end_0 = const()[name = tensor("op_29026_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_29026_end_mask_0 = const()[name = tensor("op_29026_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29026_cast_fp16 = slice_by_index(begin = var_29026_begin_0, end = var_29026_end_0, end_mask = var_29026_end_mask_0, x = k_263_cast_fp16)[name = tensor("op_29026_cast_fp16")]; + tensor var_29030_begin_0 = const()[name = tensor("op_29030_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_29030_end_0 = const()[name = tensor("op_29030_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_29030_end_mask_0 = const()[name = tensor("op_29030_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29030_cast_fp16 = slice_by_index(begin = var_29030_begin_0, end = var_29030_end_0, end_mask = var_29030_end_mask_0, x = k_263_cast_fp16)[name = tensor("op_29030_cast_fp16")]; + tensor var_29034_begin_0 = const()[name = tensor("op_29034_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_29034_end_0 = const()[name = tensor("op_29034_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_29034_end_mask_0 = const()[name = tensor("op_29034_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29034_cast_fp16 = slice_by_index(begin = var_29034_begin_0, end = var_29034_end_0, end_mask = var_29034_end_mask_0, x = k_263_cast_fp16)[name = tensor("op_29034_cast_fp16")]; + tensor var_29038_begin_0 = const()[name = tensor("op_29038_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_29038_end_0 = const()[name = tensor("op_29038_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_29038_end_mask_0 = const()[name = tensor("op_29038_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29038_cast_fp16 = slice_by_index(begin = var_29038_begin_0, end = var_29038_end_0, end_mask = var_29038_end_mask_0, x = k_263_cast_fp16)[name = tensor("op_29038_cast_fp16")]; + tensor var_29042_begin_0 = const()[name = tensor("op_29042_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_29042_end_0 = const()[name = tensor("op_29042_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_29042_end_mask_0 = const()[name = tensor("op_29042_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29042_cast_fp16 = slice_by_index(begin = var_29042_begin_0, end = var_29042_end_0, end_mask = var_29042_end_mask_0, x = k_263_cast_fp16)[name = tensor("op_29042_cast_fp16")]; + tensor var_29046_begin_0 = const()[name = tensor("op_29046_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_29046_end_0 = const()[name = tensor("op_29046_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_29046_end_mask_0 = const()[name = tensor("op_29046_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29046_cast_fp16 = slice_by_index(begin = var_29046_begin_0, end = var_29046_end_0, end_mask = var_29046_end_mask_0, x = k_263_cast_fp16)[name = tensor("op_29046_cast_fp16")]; + tensor var_29050_begin_0 = const()[name = tensor("op_29050_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_29050_end_0 = const()[name = tensor("op_29050_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_29050_end_mask_0 = const()[name = tensor("op_29050_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29050_cast_fp16 = slice_by_index(begin = var_29050_begin_0, end = var_29050_end_0, end_mask = var_29050_end_mask_0, x = k_263_cast_fp16)[name = tensor("op_29050_cast_fp16")]; + tensor var_29054_begin_0 = const()[name = tensor("op_29054_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_29054_end_0 = const()[name = tensor("op_29054_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_29054_end_mask_0 = const()[name = tensor("op_29054_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29054_cast_fp16 = slice_by_index(begin = var_29054_begin_0, end = var_29054_end_0, end_mask = var_29054_end_mask_0, x = k_263_cast_fp16)[name = tensor("op_29054_cast_fp16")]; + tensor var_29058_begin_0 = const()[name = tensor("op_29058_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_29058_end_0 = const()[name = tensor("op_29058_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_29058_end_mask_0 = const()[name = tensor("op_29058_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29058_cast_fp16 = slice_by_index(begin = var_29058_begin_0, end = var_29058_end_0, end_mask = var_29058_end_mask_0, x = k_263_cast_fp16)[name = tensor("op_29058_cast_fp16")]; + tensor var_29062_begin_0 = const()[name = tensor("op_29062_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_29062_end_0 = const()[name = tensor("op_29062_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_29062_end_mask_0 = const()[name = tensor("op_29062_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29062_cast_fp16 = slice_by_index(begin = var_29062_begin_0, end = var_29062_end_0, end_mask = var_29062_end_mask_0, x = k_263_cast_fp16)[name = tensor("op_29062_cast_fp16")]; + tensor var_29066_begin_0 = const()[name = tensor("op_29066_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_29066_end_0 = const()[name = tensor("op_29066_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_29066_end_mask_0 = const()[name = tensor("op_29066_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29066_cast_fp16 = slice_by_index(begin = var_29066_begin_0, end = var_29066_end_0, end_mask = var_29066_end_mask_0, x = k_263_cast_fp16)[name = tensor("op_29066_cast_fp16")]; + tensor var_29070_begin_0 = const()[name = tensor("op_29070_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_29070_end_0 = const()[name = tensor("op_29070_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_29070_end_mask_0 = const()[name = tensor("op_29070_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29070_cast_fp16 = slice_by_index(begin = var_29070_begin_0, end = var_29070_end_0, end_mask = var_29070_end_mask_0, x = k_263_cast_fp16)[name = tensor("op_29070_cast_fp16")]; + tensor var_29074_begin_0 = const()[name = tensor("op_29074_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_29074_end_0 = const()[name = tensor("op_29074_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_29074_end_mask_0 = const()[name = tensor("op_29074_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29074_cast_fp16 = slice_by_index(begin = var_29074_begin_0, end = var_29074_end_0, end_mask = var_29074_end_mask_0, x = k_263_cast_fp16)[name = tensor("op_29074_cast_fp16")]; + tensor var_29078_begin_0 = const()[name = tensor("op_29078_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_29078_end_0 = const()[name = tensor("op_29078_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_29078_end_mask_0 = const()[name = tensor("op_29078_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29078_cast_fp16 = slice_by_index(begin = var_29078_begin_0, end = var_29078_end_0, end_mask = var_29078_end_mask_0, x = k_263_cast_fp16)[name = tensor("op_29078_cast_fp16")]; + tensor var_29082_begin_0 = const()[name = tensor("op_29082_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_29082_end_0 = const()[name = tensor("op_29082_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_29082_end_mask_0 = const()[name = tensor("op_29082_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29082_cast_fp16 = slice_by_index(begin = var_29082_begin_0, end = var_29082_end_0, end_mask = var_29082_end_mask_0, x = k_263_cast_fp16)[name = tensor("op_29082_cast_fp16")]; + tensor var_29086_begin_0 = const()[name = tensor("op_29086_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_29086_end_0 = const()[name = tensor("op_29086_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_29086_end_mask_0 = const()[name = tensor("op_29086_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29086_cast_fp16 = slice_by_index(begin = var_29086_begin_0, end = var_29086_end_0, end_mask = var_29086_end_mask_0, x = k_263_cast_fp16)[name = tensor("op_29086_cast_fp16")]; + tensor var_29090_begin_0 = const()[name = tensor("op_29090_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_29090_end_0 = const()[name = tensor("op_29090_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_29090_end_mask_0 = const()[name = tensor("op_29090_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29090_cast_fp16 = slice_by_index(begin = var_29090_begin_0, end = var_29090_end_0, end_mask = var_29090_end_mask_0, x = k_263_cast_fp16)[name = tensor("op_29090_cast_fp16")]; + tensor var_29092_begin_0 = const()[name = tensor("op_29092_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29092_end_0 = const()[name = tensor("op_29092_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_29092_end_mask_0 = const()[name = tensor("op_29092_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29092_cast_fp16 = slice_by_index(begin = var_29092_begin_0, end = var_29092_end_0, end_mask = var_29092_end_mask_0, x = v_131_cast_fp16)[name = tensor("op_29092_cast_fp16")]; + tensor var_29096_begin_0 = const()[name = tensor("op_29096_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_29096_end_0 = const()[name = tensor("op_29096_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_29096_end_mask_0 = const()[name = tensor("op_29096_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29096_cast_fp16 = slice_by_index(begin = var_29096_begin_0, end = var_29096_end_0, end_mask = var_29096_end_mask_0, x = v_131_cast_fp16)[name = tensor("op_29096_cast_fp16")]; + tensor var_29100_begin_0 = const()[name = tensor("op_29100_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_29100_end_0 = const()[name = tensor("op_29100_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_29100_end_mask_0 = const()[name = tensor("op_29100_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29100_cast_fp16 = slice_by_index(begin = var_29100_begin_0, end = var_29100_end_0, end_mask = var_29100_end_mask_0, x = v_131_cast_fp16)[name = tensor("op_29100_cast_fp16")]; + tensor var_29104_begin_0 = const()[name = tensor("op_29104_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_29104_end_0 = const()[name = tensor("op_29104_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_29104_end_mask_0 = const()[name = tensor("op_29104_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29104_cast_fp16 = slice_by_index(begin = var_29104_begin_0, end = var_29104_end_0, end_mask = var_29104_end_mask_0, x = v_131_cast_fp16)[name = tensor("op_29104_cast_fp16")]; + tensor var_29108_begin_0 = const()[name = tensor("op_29108_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_29108_end_0 = const()[name = tensor("op_29108_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_29108_end_mask_0 = const()[name = tensor("op_29108_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29108_cast_fp16 = slice_by_index(begin = var_29108_begin_0, end = var_29108_end_0, end_mask = var_29108_end_mask_0, x = v_131_cast_fp16)[name = tensor("op_29108_cast_fp16")]; + tensor var_29112_begin_0 = const()[name = tensor("op_29112_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_29112_end_0 = const()[name = tensor("op_29112_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_29112_end_mask_0 = const()[name = tensor("op_29112_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29112_cast_fp16 = slice_by_index(begin = var_29112_begin_0, end = var_29112_end_0, end_mask = var_29112_end_mask_0, x = v_131_cast_fp16)[name = tensor("op_29112_cast_fp16")]; + tensor var_29116_begin_0 = const()[name = tensor("op_29116_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_29116_end_0 = const()[name = tensor("op_29116_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_29116_end_mask_0 = const()[name = tensor("op_29116_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29116_cast_fp16 = slice_by_index(begin = var_29116_begin_0, end = var_29116_end_0, end_mask = var_29116_end_mask_0, x = v_131_cast_fp16)[name = tensor("op_29116_cast_fp16")]; + tensor var_29120_begin_0 = const()[name = tensor("op_29120_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_29120_end_0 = const()[name = tensor("op_29120_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_29120_end_mask_0 = const()[name = tensor("op_29120_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29120_cast_fp16 = slice_by_index(begin = var_29120_begin_0, end = var_29120_end_0, end_mask = var_29120_end_mask_0, x = v_131_cast_fp16)[name = tensor("op_29120_cast_fp16")]; + tensor var_29124_begin_0 = const()[name = tensor("op_29124_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_29124_end_0 = const()[name = tensor("op_29124_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_29124_end_mask_0 = const()[name = tensor("op_29124_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29124_cast_fp16 = slice_by_index(begin = var_29124_begin_0, end = var_29124_end_0, end_mask = var_29124_end_mask_0, x = v_131_cast_fp16)[name = tensor("op_29124_cast_fp16")]; + tensor var_29128_begin_0 = const()[name = tensor("op_29128_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_29128_end_0 = const()[name = tensor("op_29128_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_29128_end_mask_0 = const()[name = tensor("op_29128_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29128_cast_fp16 = slice_by_index(begin = var_29128_begin_0, end = var_29128_end_0, end_mask = var_29128_end_mask_0, x = v_131_cast_fp16)[name = tensor("op_29128_cast_fp16")]; + tensor var_29132_begin_0 = const()[name = tensor("op_29132_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_29132_end_0 = const()[name = tensor("op_29132_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_29132_end_mask_0 = const()[name = tensor("op_29132_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29132_cast_fp16 = slice_by_index(begin = var_29132_begin_0, end = var_29132_end_0, end_mask = var_29132_end_mask_0, x = v_131_cast_fp16)[name = tensor("op_29132_cast_fp16")]; + tensor var_29136_begin_0 = const()[name = tensor("op_29136_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_29136_end_0 = const()[name = tensor("op_29136_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_29136_end_mask_0 = const()[name = tensor("op_29136_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29136_cast_fp16 = slice_by_index(begin = var_29136_begin_0, end = var_29136_end_0, end_mask = var_29136_end_mask_0, x = v_131_cast_fp16)[name = tensor("op_29136_cast_fp16")]; + tensor var_29140_begin_0 = const()[name = tensor("op_29140_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_29140_end_0 = const()[name = tensor("op_29140_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_29140_end_mask_0 = const()[name = tensor("op_29140_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29140_cast_fp16 = slice_by_index(begin = var_29140_begin_0, end = var_29140_end_0, end_mask = var_29140_end_mask_0, x = v_131_cast_fp16)[name = tensor("op_29140_cast_fp16")]; + tensor var_29144_begin_0 = const()[name = tensor("op_29144_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_29144_end_0 = const()[name = tensor("op_29144_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_29144_end_mask_0 = const()[name = tensor("op_29144_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29144_cast_fp16 = slice_by_index(begin = var_29144_begin_0, end = var_29144_end_0, end_mask = var_29144_end_mask_0, x = v_131_cast_fp16)[name = tensor("op_29144_cast_fp16")]; + tensor var_29148_begin_0 = const()[name = tensor("op_29148_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_29148_end_0 = const()[name = tensor("op_29148_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_29148_end_mask_0 = const()[name = tensor("op_29148_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29148_cast_fp16 = slice_by_index(begin = var_29148_begin_0, end = var_29148_end_0, end_mask = var_29148_end_mask_0, x = v_131_cast_fp16)[name = tensor("op_29148_cast_fp16")]; + tensor var_29152_begin_0 = const()[name = tensor("op_29152_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_29152_end_0 = const()[name = tensor("op_29152_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_29152_end_mask_0 = const()[name = tensor("op_29152_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29152_cast_fp16 = slice_by_index(begin = var_29152_begin_0, end = var_29152_end_0, end_mask = var_29152_end_mask_0, x = v_131_cast_fp16)[name = tensor("op_29152_cast_fp16")]; + tensor var_29156_begin_0 = const()[name = tensor("op_29156_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_29156_end_0 = const()[name = tensor("op_29156_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_29156_end_mask_0 = const()[name = tensor("op_29156_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29156_cast_fp16 = slice_by_index(begin = var_29156_begin_0, end = var_29156_end_0, end_mask = var_29156_end_mask_0, x = v_131_cast_fp16)[name = tensor("op_29156_cast_fp16")]; + tensor var_29160_begin_0 = const()[name = tensor("op_29160_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_29160_end_0 = const()[name = tensor("op_29160_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_29160_end_mask_0 = const()[name = tensor("op_29160_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29160_cast_fp16 = slice_by_index(begin = var_29160_begin_0, end = var_29160_end_0, end_mask = var_29160_end_mask_0, x = v_131_cast_fp16)[name = tensor("op_29160_cast_fp16")]; + tensor var_29164_begin_0 = const()[name = tensor("op_29164_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_29164_end_0 = const()[name = tensor("op_29164_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_29164_end_mask_0 = const()[name = tensor("op_29164_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29164_cast_fp16 = slice_by_index(begin = var_29164_begin_0, end = var_29164_end_0, end_mask = var_29164_end_mask_0, x = v_131_cast_fp16)[name = tensor("op_29164_cast_fp16")]; + tensor var_29168_begin_0 = const()[name = tensor("op_29168_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_29168_end_0 = const()[name = tensor("op_29168_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_29168_end_mask_0 = const()[name = tensor("op_29168_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29168_cast_fp16 = slice_by_index(begin = var_29168_begin_0, end = var_29168_end_0, end_mask = var_29168_end_mask_0, x = v_131_cast_fp16)[name = tensor("op_29168_cast_fp16")]; + tensor var_29172_equation_0 = const()[name = tensor("op_29172_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29172_cast_fp16 = einsum(equation = var_29172_equation_0, values = (var_29014_cast_fp16, var_28931_cast_fp16))[name = tensor("op_29172_cast_fp16")]; + tensor var_29173_to_fp16 = const()[name = tensor("op_29173_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2441_cast_fp16 = mul(x = var_29172_cast_fp16, y = var_29173_to_fp16)[name = tensor("aw_2441_cast_fp16")]; + tensor var_29176_equation_0 = const()[name = tensor("op_29176_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29176_cast_fp16 = einsum(equation = var_29176_equation_0, values = (var_29018_cast_fp16, var_28935_cast_fp16))[name = tensor("op_29176_cast_fp16")]; + tensor var_29177_to_fp16 = const()[name = tensor("op_29177_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2443_cast_fp16 = mul(x = var_29176_cast_fp16, y = var_29177_to_fp16)[name = tensor("aw_2443_cast_fp16")]; + tensor var_29180_equation_0 = const()[name = tensor("op_29180_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29180_cast_fp16 = einsum(equation = var_29180_equation_0, values = (var_29022_cast_fp16, var_28939_cast_fp16))[name = tensor("op_29180_cast_fp16")]; + tensor var_29181_to_fp16 = const()[name = tensor("op_29181_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2445_cast_fp16 = mul(x = var_29180_cast_fp16, y = var_29181_to_fp16)[name = tensor("aw_2445_cast_fp16")]; + tensor var_29184_equation_0 = const()[name = tensor("op_29184_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29184_cast_fp16 = einsum(equation = var_29184_equation_0, values = (var_29026_cast_fp16, var_28943_cast_fp16))[name = tensor("op_29184_cast_fp16")]; + tensor var_29185_to_fp16 = const()[name = tensor("op_29185_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2447_cast_fp16 = mul(x = var_29184_cast_fp16, y = var_29185_to_fp16)[name = tensor("aw_2447_cast_fp16")]; + tensor var_29188_equation_0 = const()[name = tensor("op_29188_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29188_cast_fp16 = einsum(equation = var_29188_equation_0, values = (var_29030_cast_fp16, var_28947_cast_fp16))[name = tensor("op_29188_cast_fp16")]; + tensor var_29189_to_fp16 = const()[name = tensor("op_29189_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2449_cast_fp16 = mul(x = var_29188_cast_fp16, y = var_29189_to_fp16)[name = tensor("aw_2449_cast_fp16")]; + tensor var_29192_equation_0 = const()[name = tensor("op_29192_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29192_cast_fp16 = einsum(equation = var_29192_equation_0, values = (var_29034_cast_fp16, var_28951_cast_fp16))[name = tensor("op_29192_cast_fp16")]; + tensor var_29193_to_fp16 = const()[name = tensor("op_29193_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2451_cast_fp16 = mul(x = var_29192_cast_fp16, y = var_29193_to_fp16)[name = tensor("aw_2451_cast_fp16")]; + tensor var_29196_equation_0 = const()[name = tensor("op_29196_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29196_cast_fp16 = einsum(equation = var_29196_equation_0, values = (var_29038_cast_fp16, var_28955_cast_fp16))[name = tensor("op_29196_cast_fp16")]; + tensor var_29197_to_fp16 = const()[name = tensor("op_29197_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2453_cast_fp16 = mul(x = var_29196_cast_fp16, y = var_29197_to_fp16)[name = tensor("aw_2453_cast_fp16")]; + tensor var_29200_equation_0 = const()[name = tensor("op_29200_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29200_cast_fp16 = einsum(equation = var_29200_equation_0, values = (var_29042_cast_fp16, var_28959_cast_fp16))[name = tensor("op_29200_cast_fp16")]; + tensor var_29201_to_fp16 = const()[name = tensor("op_29201_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2455_cast_fp16 = mul(x = var_29200_cast_fp16, y = var_29201_to_fp16)[name = tensor("aw_2455_cast_fp16")]; + tensor var_29204_equation_0 = const()[name = tensor("op_29204_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29204_cast_fp16 = einsum(equation = var_29204_equation_0, values = (var_29046_cast_fp16, var_28963_cast_fp16))[name = tensor("op_29204_cast_fp16")]; + tensor var_29205_to_fp16 = const()[name = tensor("op_29205_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2457_cast_fp16 = mul(x = var_29204_cast_fp16, y = var_29205_to_fp16)[name = tensor("aw_2457_cast_fp16")]; + tensor var_29208_equation_0 = const()[name = tensor("op_29208_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29208_cast_fp16 = einsum(equation = var_29208_equation_0, values = (var_29050_cast_fp16, var_28967_cast_fp16))[name = tensor("op_29208_cast_fp16")]; + tensor var_29209_to_fp16 = const()[name = tensor("op_29209_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2459_cast_fp16 = mul(x = var_29208_cast_fp16, y = var_29209_to_fp16)[name = tensor("aw_2459_cast_fp16")]; + tensor var_29212_equation_0 = const()[name = tensor("op_29212_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29212_cast_fp16 = einsum(equation = var_29212_equation_0, values = (var_29054_cast_fp16, var_28971_cast_fp16))[name = tensor("op_29212_cast_fp16")]; + tensor var_29213_to_fp16 = const()[name = tensor("op_29213_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2461_cast_fp16 = mul(x = var_29212_cast_fp16, y = var_29213_to_fp16)[name = tensor("aw_2461_cast_fp16")]; + tensor var_29216_equation_0 = const()[name = tensor("op_29216_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29216_cast_fp16 = einsum(equation = var_29216_equation_0, values = (var_29058_cast_fp16, var_28975_cast_fp16))[name = tensor("op_29216_cast_fp16")]; + tensor var_29217_to_fp16 = const()[name = tensor("op_29217_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2463_cast_fp16 = mul(x = var_29216_cast_fp16, y = var_29217_to_fp16)[name = tensor("aw_2463_cast_fp16")]; + tensor var_29220_equation_0 = const()[name = tensor("op_29220_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29220_cast_fp16 = einsum(equation = var_29220_equation_0, values = (var_29062_cast_fp16, var_28979_cast_fp16))[name = tensor("op_29220_cast_fp16")]; + tensor var_29221_to_fp16 = const()[name = tensor("op_29221_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2465_cast_fp16 = mul(x = var_29220_cast_fp16, y = var_29221_to_fp16)[name = tensor("aw_2465_cast_fp16")]; + tensor var_29224_equation_0 = const()[name = tensor("op_29224_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29224_cast_fp16 = einsum(equation = var_29224_equation_0, values = (var_29066_cast_fp16, var_28983_cast_fp16))[name = tensor("op_29224_cast_fp16")]; + tensor var_29225_to_fp16 = const()[name = tensor("op_29225_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2467_cast_fp16 = mul(x = var_29224_cast_fp16, y = var_29225_to_fp16)[name = tensor("aw_2467_cast_fp16")]; + tensor var_29228_equation_0 = const()[name = tensor("op_29228_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29228_cast_fp16 = einsum(equation = var_29228_equation_0, values = (var_29070_cast_fp16, var_28987_cast_fp16))[name = tensor("op_29228_cast_fp16")]; + tensor var_29229_to_fp16 = const()[name = tensor("op_29229_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2469_cast_fp16 = mul(x = var_29228_cast_fp16, y = var_29229_to_fp16)[name = tensor("aw_2469_cast_fp16")]; + tensor var_29232_equation_0 = const()[name = tensor("op_29232_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29232_cast_fp16 = einsum(equation = var_29232_equation_0, values = (var_29074_cast_fp16, var_28991_cast_fp16))[name = tensor("op_29232_cast_fp16")]; + tensor var_29233_to_fp16 = const()[name = tensor("op_29233_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2471_cast_fp16 = mul(x = var_29232_cast_fp16, y = var_29233_to_fp16)[name = tensor("aw_2471_cast_fp16")]; + tensor var_29236_equation_0 = const()[name = tensor("op_29236_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29236_cast_fp16 = einsum(equation = var_29236_equation_0, values = (var_29078_cast_fp16, var_28995_cast_fp16))[name = tensor("op_29236_cast_fp16")]; + tensor var_29237_to_fp16 = const()[name = tensor("op_29237_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2473_cast_fp16 = mul(x = var_29236_cast_fp16, y = var_29237_to_fp16)[name = tensor("aw_2473_cast_fp16")]; + tensor var_29240_equation_0 = const()[name = tensor("op_29240_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29240_cast_fp16 = einsum(equation = var_29240_equation_0, values = (var_29082_cast_fp16, var_28999_cast_fp16))[name = tensor("op_29240_cast_fp16")]; + tensor var_29241_to_fp16 = const()[name = tensor("op_29241_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2475_cast_fp16 = mul(x = var_29240_cast_fp16, y = var_29241_to_fp16)[name = tensor("aw_2475_cast_fp16")]; + tensor var_29244_equation_0 = const()[name = tensor("op_29244_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29244_cast_fp16 = einsum(equation = var_29244_equation_0, values = (var_29086_cast_fp16, var_29003_cast_fp16))[name = tensor("op_29244_cast_fp16")]; + tensor var_29245_to_fp16 = const()[name = tensor("op_29245_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2477_cast_fp16 = mul(x = var_29244_cast_fp16, y = var_29245_to_fp16)[name = tensor("aw_2477_cast_fp16")]; + tensor var_29248_equation_0 = const()[name = tensor("op_29248_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29248_cast_fp16 = einsum(equation = var_29248_equation_0, values = (var_29090_cast_fp16, var_29007_cast_fp16))[name = tensor("op_29248_cast_fp16")]; + tensor var_29249_to_fp16 = const()[name = tensor("op_29249_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2479_cast_fp16 = mul(x = var_29248_cast_fp16, y = var_29249_to_fp16)[name = tensor("aw_2479_cast_fp16")]; + tensor var_29251_cast_fp16 = softmax(axis = var_21077, x = aw_2441_cast_fp16)[name = tensor("op_29251_cast_fp16")]; + tensor var_29252_cast_fp16 = softmax(axis = var_21077, x = aw_2443_cast_fp16)[name = tensor("op_29252_cast_fp16")]; + tensor var_29253_cast_fp16 = softmax(axis = var_21077, x = aw_2445_cast_fp16)[name = tensor("op_29253_cast_fp16")]; + tensor var_29254_cast_fp16 = softmax(axis = var_21077, x = aw_2447_cast_fp16)[name = tensor("op_29254_cast_fp16")]; + tensor var_29255_cast_fp16 = softmax(axis = var_21077, x = aw_2449_cast_fp16)[name = tensor("op_29255_cast_fp16")]; + tensor var_29256_cast_fp16 = softmax(axis = var_21077, x = aw_2451_cast_fp16)[name = tensor("op_29256_cast_fp16")]; + tensor var_29257_cast_fp16 = softmax(axis = var_21077, x = aw_2453_cast_fp16)[name = tensor("op_29257_cast_fp16")]; + tensor var_29258_cast_fp16 = softmax(axis = var_21077, x = aw_2455_cast_fp16)[name = tensor("op_29258_cast_fp16")]; + tensor var_29259_cast_fp16 = softmax(axis = var_21077, x = aw_2457_cast_fp16)[name = tensor("op_29259_cast_fp16")]; + tensor var_29260_cast_fp16 = softmax(axis = var_21077, x = aw_2459_cast_fp16)[name = tensor("op_29260_cast_fp16")]; + tensor var_29261_cast_fp16 = softmax(axis = var_21077, x = aw_2461_cast_fp16)[name = tensor("op_29261_cast_fp16")]; + tensor var_29262_cast_fp16 = softmax(axis = var_21077, x = aw_2463_cast_fp16)[name = tensor("op_29262_cast_fp16")]; + tensor var_29263_cast_fp16 = softmax(axis = var_21077, x = aw_2465_cast_fp16)[name = tensor("op_29263_cast_fp16")]; + tensor var_29264_cast_fp16 = softmax(axis = var_21077, x = aw_2467_cast_fp16)[name = tensor("op_29264_cast_fp16")]; + tensor var_29265_cast_fp16 = softmax(axis = var_21077, x = aw_2469_cast_fp16)[name = tensor("op_29265_cast_fp16")]; + tensor var_29266_cast_fp16 = softmax(axis = var_21077, x = aw_2471_cast_fp16)[name = tensor("op_29266_cast_fp16")]; + tensor var_29267_cast_fp16 = softmax(axis = var_21077, x = aw_2473_cast_fp16)[name = tensor("op_29267_cast_fp16")]; + tensor var_29268_cast_fp16 = softmax(axis = var_21077, x = aw_2475_cast_fp16)[name = tensor("op_29268_cast_fp16")]; + tensor var_29269_cast_fp16 = softmax(axis = var_21077, x = aw_2477_cast_fp16)[name = tensor("op_29269_cast_fp16")]; + tensor var_29270_cast_fp16 = softmax(axis = var_21077, x = aw_2479_cast_fp16)[name = tensor("op_29270_cast_fp16")]; + tensor var_29272_equation_0 = const()[name = tensor("op_29272_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29272_cast_fp16 = einsum(equation = var_29272_equation_0, values = (var_29092_cast_fp16, var_29251_cast_fp16))[name = tensor("op_29272_cast_fp16")]; + tensor var_29274_equation_0 = const()[name = tensor("op_29274_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29274_cast_fp16 = einsum(equation = var_29274_equation_0, values = (var_29096_cast_fp16, var_29252_cast_fp16))[name = tensor("op_29274_cast_fp16")]; + tensor var_29276_equation_0 = const()[name = tensor("op_29276_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29276_cast_fp16 = einsum(equation = var_29276_equation_0, values = (var_29100_cast_fp16, var_29253_cast_fp16))[name = tensor("op_29276_cast_fp16")]; + tensor var_29278_equation_0 = const()[name = tensor("op_29278_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29278_cast_fp16 = einsum(equation = var_29278_equation_0, values = (var_29104_cast_fp16, var_29254_cast_fp16))[name = tensor("op_29278_cast_fp16")]; + tensor var_29280_equation_0 = const()[name = tensor("op_29280_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29280_cast_fp16 = einsum(equation = var_29280_equation_0, values = (var_29108_cast_fp16, var_29255_cast_fp16))[name = tensor("op_29280_cast_fp16")]; + tensor var_29282_equation_0 = const()[name = tensor("op_29282_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29282_cast_fp16 = einsum(equation = var_29282_equation_0, values = (var_29112_cast_fp16, var_29256_cast_fp16))[name = tensor("op_29282_cast_fp16")]; + tensor var_29284_equation_0 = const()[name = tensor("op_29284_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29284_cast_fp16 = einsum(equation = var_29284_equation_0, values = (var_29116_cast_fp16, var_29257_cast_fp16))[name = tensor("op_29284_cast_fp16")]; + tensor var_29286_equation_0 = const()[name = tensor("op_29286_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29286_cast_fp16 = einsum(equation = var_29286_equation_0, values = (var_29120_cast_fp16, var_29258_cast_fp16))[name = tensor("op_29286_cast_fp16")]; + tensor var_29288_equation_0 = const()[name = tensor("op_29288_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29288_cast_fp16 = einsum(equation = var_29288_equation_0, values = (var_29124_cast_fp16, var_29259_cast_fp16))[name = tensor("op_29288_cast_fp16")]; + tensor var_29290_equation_0 = const()[name = tensor("op_29290_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29290_cast_fp16 = einsum(equation = var_29290_equation_0, values = (var_29128_cast_fp16, var_29260_cast_fp16))[name = tensor("op_29290_cast_fp16")]; + tensor var_29292_equation_0 = const()[name = tensor("op_29292_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29292_cast_fp16 = einsum(equation = var_29292_equation_0, values = (var_29132_cast_fp16, var_29261_cast_fp16))[name = tensor("op_29292_cast_fp16")]; + tensor var_29294_equation_0 = const()[name = tensor("op_29294_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29294_cast_fp16 = einsum(equation = var_29294_equation_0, values = (var_29136_cast_fp16, var_29262_cast_fp16))[name = tensor("op_29294_cast_fp16")]; + tensor var_29296_equation_0 = const()[name = tensor("op_29296_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29296_cast_fp16 = einsum(equation = var_29296_equation_0, values = (var_29140_cast_fp16, var_29263_cast_fp16))[name = tensor("op_29296_cast_fp16")]; + tensor var_29298_equation_0 = const()[name = tensor("op_29298_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29298_cast_fp16 = einsum(equation = var_29298_equation_0, values = (var_29144_cast_fp16, var_29264_cast_fp16))[name = tensor("op_29298_cast_fp16")]; + tensor var_29300_equation_0 = const()[name = tensor("op_29300_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29300_cast_fp16 = einsum(equation = var_29300_equation_0, values = (var_29148_cast_fp16, var_29265_cast_fp16))[name = tensor("op_29300_cast_fp16")]; + tensor var_29302_equation_0 = const()[name = tensor("op_29302_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29302_cast_fp16 = einsum(equation = var_29302_equation_0, values = (var_29152_cast_fp16, var_29266_cast_fp16))[name = tensor("op_29302_cast_fp16")]; + tensor var_29304_equation_0 = const()[name = tensor("op_29304_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29304_cast_fp16 = einsum(equation = var_29304_equation_0, values = (var_29156_cast_fp16, var_29267_cast_fp16))[name = tensor("op_29304_cast_fp16")]; + tensor var_29306_equation_0 = const()[name = tensor("op_29306_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29306_cast_fp16 = einsum(equation = var_29306_equation_0, values = (var_29160_cast_fp16, var_29268_cast_fp16))[name = tensor("op_29306_cast_fp16")]; + tensor var_29308_equation_0 = const()[name = tensor("op_29308_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29308_cast_fp16 = einsum(equation = var_29308_equation_0, values = (var_29164_cast_fp16, var_29269_cast_fp16))[name = tensor("op_29308_cast_fp16")]; + tensor var_29310_equation_0 = const()[name = tensor("op_29310_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29310_cast_fp16 = einsum(equation = var_29310_equation_0, values = (var_29168_cast_fp16, var_29270_cast_fp16))[name = tensor("op_29310_cast_fp16")]; + tensor input_393_interleave_0 = const()[name = tensor("input_393_interleave_0"), val = tensor(false)]; + tensor input_393_cast_fp16 = concat(axis = var_21077, interleave = input_393_interleave_0, values = (var_29272_cast_fp16, var_29274_cast_fp16, var_29276_cast_fp16, var_29278_cast_fp16, var_29280_cast_fp16, var_29282_cast_fp16, var_29284_cast_fp16, var_29286_cast_fp16, var_29288_cast_fp16, var_29290_cast_fp16, var_29292_cast_fp16, var_29294_cast_fp16, var_29296_cast_fp16, var_29298_cast_fp16, var_29300_cast_fp16, var_29302_cast_fp16, var_29304_cast_fp16, var_29306_cast_fp16, var_29308_cast_fp16, var_29310_cast_fp16))[name = tensor("input_393_cast_fp16")]; + tensor var_29316 = const()[name = tensor("op_29316"), val = tensor([1, 1])]; + tensor var_29318 = const()[name = tensor("op_29318"), val = tensor([1, 1])]; + tensor var_29320_pad_type_0 = const()[name = tensor("op_29320_pad_type_0"), val = tensor("custom")]; + tensor var_29320_pad_0 = const()[name = tensor("op_29320_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_8_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_8_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2309096064)))]; + tensor mid_block_attentions_0_transformer_blocks_8_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_8_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2312372928)))]; + tensor var_29320_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_8_attn2_to_out_0_bias_to_fp16, dilations = var_29318, groups = var_21077, pad = var_29320_pad_0, pad_type = var_29320_pad_type_0, strides = var_29316, weight = mid_block_attentions_0_transformer_blocks_8_attn2_to_out_0_weight_to_fp16, x = input_393_cast_fp16)[name = tensor("op_29320_cast_fp16")]; + tensor inputs_197_cast_fp16 = add(x = var_29320_cast_fp16, y = inputs_195_cast_fp16)[name = tensor("inputs_197_cast_fp16")]; + tensor input_395_axes_0 = const()[name = tensor("input_395_axes_0"), val = tensor([1])]; + tensor input_395_gamma_0_to_fp16 = const()[name = tensor("input_395_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2312375552)))]; + tensor input_395_beta_0_to_fp16 = const()[name = tensor("input_395_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2312378176)))]; + tensor var_29330_to_fp16 = const()[name = tensor("op_29330_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_395_cast_fp16 = layer_norm(axes = input_395_axes_0, beta = input_395_beta_0_to_fp16, epsilon = var_29330_to_fp16, gamma = input_395_gamma_0_to_fp16, x = inputs_197_cast_fp16)[name = tensor("input_395_cast_fp16")]; + tensor var_29346 = const()[name = tensor("op_29346"), val = tensor([1, 1])]; + tensor var_29348 = const()[name = tensor("op_29348"), val = tensor([1, 1])]; + tensor var_29350_pad_type_0 = const()[name = tensor("op_29350_pad_type_0"), val = tensor("custom")]; + tensor var_29350_pad_0 = const()[name = tensor("op_29350_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_8_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_8_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2312380800)))]; + tensor mid_block_attentions_0_transformer_blocks_8_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_8_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2338595264)))]; + tensor var_29350_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_8_ff_net_0_proj_bias_to_fp16, dilations = var_29348, groups = var_21077, pad = var_29350_pad_0, pad_type = var_29350_pad_type_0, strides = var_29346, weight = mid_block_attentions_0_transformer_blocks_8_ff_net_0_proj_weight_to_fp16, x = input_395_cast_fp16)[name = tensor("op_29350_cast_fp16")]; + tensor var_29351_split_sizes_0 = const()[name = tensor("op_29351_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_29351_axis_0 = const()[name = tensor("op_29351_axis_0"), val = tensor(1)]; + tensor var_29351_cast_fp16_0, tensor var_29351_cast_fp16_1 = split(axis = var_29351_axis_0, split_sizes = var_29351_split_sizes_0, x = var_29350_cast_fp16)[name = tensor("op_29351_cast_fp16")]; + tensor var_29353_mode_0 = const()[name = tensor("op_29353_mode_0"), val = tensor("EXACT")]; + tensor var_29353_cast_fp16 = gelu(mode = var_29353_mode_0, x = var_29351_cast_fp16_1)[name = tensor("op_29353_cast_fp16")]; + tensor input_397_cast_fp16 = mul(x = var_29351_cast_fp16_0, y = var_29353_cast_fp16)[name = tensor("input_397_cast_fp16")]; + tensor var_29357 = const()[name = tensor("op_29357"), val = tensor([1, 1])]; + tensor var_29359 = const()[name = tensor("op_29359"), val = tensor([1, 1])]; + tensor var_29361_pad_type_0 = const()[name = tensor("op_29361_pad_type_0"), val = tensor("custom")]; + tensor var_29361_pad_0 = const()[name = tensor("op_29361_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_8_ff_net_2_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_8_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2338615808)))]; + tensor mid_block_attentions_0_transformer_blocks_8_ff_net_2_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_8_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2351723072)))]; + tensor var_29361_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_8_ff_net_2_bias_to_fp16, dilations = var_29359, groups = var_21077, pad = var_29361_pad_0, pad_type = var_29361_pad_type_0, strides = var_29357, weight = mid_block_attentions_0_transformer_blocks_8_ff_net_2_weight_to_fp16, x = input_397_cast_fp16)[name = tensor("op_29361_cast_fp16")]; + tensor inputs_199_cast_fp16 = add(x = var_29361_cast_fp16, y = inputs_197_cast_fp16)[name = tensor("inputs_199_cast_fp16")]; + tensor hidden_states_263_axes_0 = const()[name = tensor("hidden_states_263_axes_0"), val = tensor([1])]; + tensor hidden_states_263_gamma_0_to_fp16 = const()[name = tensor("hidden_states_263_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2351725696)))]; + tensor hidden_states_263_beta_0_to_fp16 = const()[name = tensor("hidden_states_263_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2351728320)))]; + tensor var_29377_to_fp16 = const()[name = tensor("op_29377_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_263_cast_fp16 = layer_norm(axes = hidden_states_263_axes_0, beta = hidden_states_263_beta_0_to_fp16, epsilon = var_29377_to_fp16, gamma = hidden_states_263_gamma_0_to_fp16, x = inputs_199_cast_fp16)[name = tensor("hidden_states_263_cast_fp16")]; + tensor var_29392 = const()[name = tensor("op_29392"), val = tensor([1, 1])]; + tensor var_29394 = const()[name = tensor("op_29394"), val = tensor([1, 1])]; + tensor q_133_pad_type_0 = const()[name = tensor("q_133_pad_type_0"), val = tensor("custom")]; + tensor q_133_pad_0 = const()[name = tensor("q_133_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_9_attn1_to_q_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_9_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2351730944)))]; + tensor q_133_cast_fp16 = conv(dilations = var_29394, groups = var_21077, pad = q_133_pad_0, pad_type = q_133_pad_type_0, strides = var_29392, weight = mid_block_attentions_0_transformer_blocks_9_attn1_to_q_weight_to_fp16, x = hidden_states_263_cast_fp16)[name = tensor("q_133_cast_fp16")]; + tensor var_29398 = const()[name = tensor("op_29398"), val = tensor([1, 1])]; + tensor var_29400 = const()[name = tensor("op_29400"), val = tensor([1, 1])]; + tensor k_265_pad_type_0 = const()[name = tensor("k_265_pad_type_0"), val = tensor("custom")]; + tensor k_265_pad_0 = const()[name = tensor("k_265_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_9_attn1_to_k_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_9_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2355007808)))]; + tensor k_265_cast_fp16 = conv(dilations = var_29400, groups = var_21077, pad = k_265_pad_0, pad_type = k_265_pad_type_0, strides = var_29398, weight = mid_block_attentions_0_transformer_blocks_9_attn1_to_k_weight_to_fp16, x = hidden_states_263_cast_fp16)[name = tensor("k_265_cast_fp16")]; + tensor var_29404 = const()[name = tensor("op_29404"), val = tensor([1, 1])]; + tensor var_29406 = const()[name = tensor("op_29406"), val = tensor([1, 1])]; + tensor v_133_pad_type_0 = const()[name = tensor("v_133_pad_type_0"), val = tensor("custom")]; + tensor v_133_pad_0 = const()[name = tensor("v_133_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_9_attn1_to_v_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_9_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2358284672)))]; + tensor v_133_cast_fp16 = conv(dilations = var_29406, groups = var_21077, pad = v_133_pad_0, pad_type = v_133_pad_type_0, strides = var_29404, weight = mid_block_attentions_0_transformer_blocks_9_attn1_to_v_weight_to_fp16, x = hidden_states_263_cast_fp16)[name = tensor("v_133_cast_fp16")]; + tensor var_29410_begin_0 = const()[name = tensor("op_29410_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29410_end_0 = const()[name = tensor("op_29410_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_29410_end_mask_0 = const()[name = tensor("op_29410_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29410_cast_fp16 = slice_by_index(begin = var_29410_begin_0, end = var_29410_end_0, end_mask = var_29410_end_mask_0, x = q_133_cast_fp16)[name = tensor("op_29410_cast_fp16")]; + tensor var_29414_begin_0 = const()[name = tensor("op_29414_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_29414_end_0 = const()[name = tensor("op_29414_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_29414_end_mask_0 = const()[name = tensor("op_29414_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29414_cast_fp16 = slice_by_index(begin = var_29414_begin_0, end = var_29414_end_0, end_mask = var_29414_end_mask_0, x = q_133_cast_fp16)[name = tensor("op_29414_cast_fp16")]; + tensor var_29418_begin_0 = const()[name = tensor("op_29418_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_29418_end_0 = const()[name = tensor("op_29418_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_29418_end_mask_0 = const()[name = tensor("op_29418_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29418_cast_fp16 = slice_by_index(begin = var_29418_begin_0, end = var_29418_end_0, end_mask = var_29418_end_mask_0, x = q_133_cast_fp16)[name = tensor("op_29418_cast_fp16")]; + tensor var_29422_begin_0 = const()[name = tensor("op_29422_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_29422_end_0 = const()[name = tensor("op_29422_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_29422_end_mask_0 = const()[name = tensor("op_29422_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29422_cast_fp16 = slice_by_index(begin = var_29422_begin_0, end = var_29422_end_0, end_mask = var_29422_end_mask_0, x = q_133_cast_fp16)[name = tensor("op_29422_cast_fp16")]; + tensor var_29426_begin_0 = const()[name = tensor("op_29426_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_29426_end_0 = const()[name = tensor("op_29426_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_29426_end_mask_0 = const()[name = tensor("op_29426_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29426_cast_fp16 = slice_by_index(begin = var_29426_begin_0, end = var_29426_end_0, end_mask = var_29426_end_mask_0, x = q_133_cast_fp16)[name = tensor("op_29426_cast_fp16")]; + tensor var_29430_begin_0 = const()[name = tensor("op_29430_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_29430_end_0 = const()[name = tensor("op_29430_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_29430_end_mask_0 = const()[name = tensor("op_29430_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29430_cast_fp16 = slice_by_index(begin = var_29430_begin_0, end = var_29430_end_0, end_mask = var_29430_end_mask_0, x = q_133_cast_fp16)[name = tensor("op_29430_cast_fp16")]; + tensor var_29434_begin_0 = const()[name = tensor("op_29434_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_29434_end_0 = const()[name = tensor("op_29434_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_29434_end_mask_0 = const()[name = tensor("op_29434_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29434_cast_fp16 = slice_by_index(begin = var_29434_begin_0, end = var_29434_end_0, end_mask = var_29434_end_mask_0, x = q_133_cast_fp16)[name = tensor("op_29434_cast_fp16")]; + tensor var_29438_begin_0 = const()[name = tensor("op_29438_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_29438_end_0 = const()[name = tensor("op_29438_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_29438_end_mask_0 = const()[name = tensor("op_29438_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29438_cast_fp16 = slice_by_index(begin = var_29438_begin_0, end = var_29438_end_0, end_mask = var_29438_end_mask_0, x = q_133_cast_fp16)[name = tensor("op_29438_cast_fp16")]; + tensor var_29442_begin_0 = const()[name = tensor("op_29442_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_29442_end_0 = const()[name = tensor("op_29442_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_29442_end_mask_0 = const()[name = tensor("op_29442_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29442_cast_fp16 = slice_by_index(begin = var_29442_begin_0, end = var_29442_end_0, end_mask = var_29442_end_mask_0, x = q_133_cast_fp16)[name = tensor("op_29442_cast_fp16")]; + tensor var_29446_begin_0 = const()[name = tensor("op_29446_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_29446_end_0 = const()[name = tensor("op_29446_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_29446_end_mask_0 = const()[name = tensor("op_29446_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29446_cast_fp16 = slice_by_index(begin = var_29446_begin_0, end = var_29446_end_0, end_mask = var_29446_end_mask_0, x = q_133_cast_fp16)[name = tensor("op_29446_cast_fp16")]; + tensor var_29450_begin_0 = const()[name = tensor("op_29450_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_29450_end_0 = const()[name = tensor("op_29450_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_29450_end_mask_0 = const()[name = tensor("op_29450_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29450_cast_fp16 = slice_by_index(begin = var_29450_begin_0, end = var_29450_end_0, end_mask = var_29450_end_mask_0, x = q_133_cast_fp16)[name = tensor("op_29450_cast_fp16")]; + tensor var_29454_begin_0 = const()[name = tensor("op_29454_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_29454_end_0 = const()[name = tensor("op_29454_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_29454_end_mask_0 = const()[name = tensor("op_29454_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29454_cast_fp16 = slice_by_index(begin = var_29454_begin_0, end = var_29454_end_0, end_mask = var_29454_end_mask_0, x = q_133_cast_fp16)[name = tensor("op_29454_cast_fp16")]; + tensor var_29458_begin_0 = const()[name = tensor("op_29458_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_29458_end_0 = const()[name = tensor("op_29458_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_29458_end_mask_0 = const()[name = tensor("op_29458_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29458_cast_fp16 = slice_by_index(begin = var_29458_begin_0, end = var_29458_end_0, end_mask = var_29458_end_mask_0, x = q_133_cast_fp16)[name = tensor("op_29458_cast_fp16")]; + tensor var_29462_begin_0 = const()[name = tensor("op_29462_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_29462_end_0 = const()[name = tensor("op_29462_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_29462_end_mask_0 = const()[name = tensor("op_29462_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29462_cast_fp16 = slice_by_index(begin = var_29462_begin_0, end = var_29462_end_0, end_mask = var_29462_end_mask_0, x = q_133_cast_fp16)[name = tensor("op_29462_cast_fp16")]; + tensor var_29466_begin_0 = const()[name = tensor("op_29466_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_29466_end_0 = const()[name = tensor("op_29466_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_29466_end_mask_0 = const()[name = tensor("op_29466_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29466_cast_fp16 = slice_by_index(begin = var_29466_begin_0, end = var_29466_end_0, end_mask = var_29466_end_mask_0, x = q_133_cast_fp16)[name = tensor("op_29466_cast_fp16")]; + tensor var_29470_begin_0 = const()[name = tensor("op_29470_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_29470_end_0 = const()[name = tensor("op_29470_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_29470_end_mask_0 = const()[name = tensor("op_29470_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29470_cast_fp16 = slice_by_index(begin = var_29470_begin_0, end = var_29470_end_0, end_mask = var_29470_end_mask_0, x = q_133_cast_fp16)[name = tensor("op_29470_cast_fp16")]; + tensor var_29474_begin_0 = const()[name = tensor("op_29474_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_29474_end_0 = const()[name = tensor("op_29474_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_29474_end_mask_0 = const()[name = tensor("op_29474_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29474_cast_fp16 = slice_by_index(begin = var_29474_begin_0, end = var_29474_end_0, end_mask = var_29474_end_mask_0, x = q_133_cast_fp16)[name = tensor("op_29474_cast_fp16")]; + tensor var_29478_begin_0 = const()[name = tensor("op_29478_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_29478_end_0 = const()[name = tensor("op_29478_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_29478_end_mask_0 = const()[name = tensor("op_29478_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29478_cast_fp16 = slice_by_index(begin = var_29478_begin_0, end = var_29478_end_0, end_mask = var_29478_end_mask_0, x = q_133_cast_fp16)[name = tensor("op_29478_cast_fp16")]; + tensor var_29482_begin_0 = const()[name = tensor("op_29482_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_29482_end_0 = const()[name = tensor("op_29482_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_29482_end_mask_0 = const()[name = tensor("op_29482_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29482_cast_fp16 = slice_by_index(begin = var_29482_begin_0, end = var_29482_end_0, end_mask = var_29482_end_mask_0, x = q_133_cast_fp16)[name = tensor("op_29482_cast_fp16")]; + tensor var_29486_begin_0 = const()[name = tensor("op_29486_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_29486_end_0 = const()[name = tensor("op_29486_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_29486_end_mask_0 = const()[name = tensor("op_29486_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29486_cast_fp16 = slice_by_index(begin = var_29486_begin_0, end = var_29486_end_0, end_mask = var_29486_end_mask_0, x = q_133_cast_fp16)[name = tensor("op_29486_cast_fp16")]; + tensor k_267_perm_0 = const()[name = tensor("k_267_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_29493_begin_0 = const()[name = tensor("op_29493_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29493_end_0 = const()[name = tensor("op_29493_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_29493_end_mask_0 = const()[name = tensor("op_29493_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_267_cast_fp16 = transpose(perm = k_267_perm_0, x = k_265_cast_fp16)[name = tensor("transpose_73")]; + tensor var_29493_cast_fp16 = slice_by_index(begin = var_29493_begin_0, end = var_29493_end_0, end_mask = var_29493_end_mask_0, x = k_267_cast_fp16)[name = tensor("op_29493_cast_fp16")]; + tensor var_29497_begin_0 = const()[name = tensor("op_29497_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_29497_end_0 = const()[name = tensor("op_29497_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_29497_end_mask_0 = const()[name = tensor("op_29497_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29497_cast_fp16 = slice_by_index(begin = var_29497_begin_0, end = var_29497_end_0, end_mask = var_29497_end_mask_0, x = k_267_cast_fp16)[name = tensor("op_29497_cast_fp16")]; + tensor var_29501_begin_0 = const()[name = tensor("op_29501_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_29501_end_0 = const()[name = tensor("op_29501_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_29501_end_mask_0 = const()[name = tensor("op_29501_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29501_cast_fp16 = slice_by_index(begin = var_29501_begin_0, end = var_29501_end_0, end_mask = var_29501_end_mask_0, x = k_267_cast_fp16)[name = tensor("op_29501_cast_fp16")]; + tensor var_29505_begin_0 = const()[name = tensor("op_29505_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_29505_end_0 = const()[name = tensor("op_29505_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_29505_end_mask_0 = const()[name = tensor("op_29505_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29505_cast_fp16 = slice_by_index(begin = var_29505_begin_0, end = var_29505_end_0, end_mask = var_29505_end_mask_0, x = k_267_cast_fp16)[name = tensor("op_29505_cast_fp16")]; + tensor var_29509_begin_0 = const()[name = tensor("op_29509_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_29509_end_0 = const()[name = tensor("op_29509_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_29509_end_mask_0 = const()[name = tensor("op_29509_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29509_cast_fp16 = slice_by_index(begin = var_29509_begin_0, end = var_29509_end_0, end_mask = var_29509_end_mask_0, x = k_267_cast_fp16)[name = tensor("op_29509_cast_fp16")]; + tensor var_29513_begin_0 = const()[name = tensor("op_29513_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_29513_end_0 = const()[name = tensor("op_29513_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_29513_end_mask_0 = const()[name = tensor("op_29513_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29513_cast_fp16 = slice_by_index(begin = var_29513_begin_0, end = var_29513_end_0, end_mask = var_29513_end_mask_0, x = k_267_cast_fp16)[name = tensor("op_29513_cast_fp16")]; + tensor var_29517_begin_0 = const()[name = tensor("op_29517_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_29517_end_0 = const()[name = tensor("op_29517_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_29517_end_mask_0 = const()[name = tensor("op_29517_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29517_cast_fp16 = slice_by_index(begin = var_29517_begin_0, end = var_29517_end_0, end_mask = var_29517_end_mask_0, x = k_267_cast_fp16)[name = tensor("op_29517_cast_fp16")]; + tensor var_29521_begin_0 = const()[name = tensor("op_29521_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_29521_end_0 = const()[name = tensor("op_29521_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_29521_end_mask_0 = const()[name = tensor("op_29521_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29521_cast_fp16 = slice_by_index(begin = var_29521_begin_0, end = var_29521_end_0, end_mask = var_29521_end_mask_0, x = k_267_cast_fp16)[name = tensor("op_29521_cast_fp16")]; + tensor var_29525_begin_0 = const()[name = tensor("op_29525_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_29525_end_0 = const()[name = tensor("op_29525_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_29525_end_mask_0 = const()[name = tensor("op_29525_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29525_cast_fp16 = slice_by_index(begin = var_29525_begin_0, end = var_29525_end_0, end_mask = var_29525_end_mask_0, x = k_267_cast_fp16)[name = tensor("op_29525_cast_fp16")]; + tensor var_29529_begin_0 = const()[name = tensor("op_29529_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_29529_end_0 = const()[name = tensor("op_29529_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_29529_end_mask_0 = const()[name = tensor("op_29529_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29529_cast_fp16 = slice_by_index(begin = var_29529_begin_0, end = var_29529_end_0, end_mask = var_29529_end_mask_0, x = k_267_cast_fp16)[name = tensor("op_29529_cast_fp16")]; + tensor var_29533_begin_0 = const()[name = tensor("op_29533_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_29533_end_0 = const()[name = tensor("op_29533_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_29533_end_mask_0 = const()[name = tensor("op_29533_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29533_cast_fp16 = slice_by_index(begin = var_29533_begin_0, end = var_29533_end_0, end_mask = var_29533_end_mask_0, x = k_267_cast_fp16)[name = tensor("op_29533_cast_fp16")]; + tensor var_29537_begin_0 = const()[name = tensor("op_29537_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_29537_end_0 = const()[name = tensor("op_29537_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_29537_end_mask_0 = const()[name = tensor("op_29537_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29537_cast_fp16 = slice_by_index(begin = var_29537_begin_0, end = var_29537_end_0, end_mask = var_29537_end_mask_0, x = k_267_cast_fp16)[name = tensor("op_29537_cast_fp16")]; + tensor var_29541_begin_0 = const()[name = tensor("op_29541_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_29541_end_0 = const()[name = tensor("op_29541_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_29541_end_mask_0 = const()[name = tensor("op_29541_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29541_cast_fp16 = slice_by_index(begin = var_29541_begin_0, end = var_29541_end_0, end_mask = var_29541_end_mask_0, x = k_267_cast_fp16)[name = tensor("op_29541_cast_fp16")]; + tensor var_29545_begin_0 = const()[name = tensor("op_29545_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_29545_end_0 = const()[name = tensor("op_29545_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_29545_end_mask_0 = const()[name = tensor("op_29545_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29545_cast_fp16 = slice_by_index(begin = var_29545_begin_0, end = var_29545_end_0, end_mask = var_29545_end_mask_0, x = k_267_cast_fp16)[name = tensor("op_29545_cast_fp16")]; + tensor var_29549_begin_0 = const()[name = tensor("op_29549_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_29549_end_0 = const()[name = tensor("op_29549_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_29549_end_mask_0 = const()[name = tensor("op_29549_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29549_cast_fp16 = slice_by_index(begin = var_29549_begin_0, end = var_29549_end_0, end_mask = var_29549_end_mask_0, x = k_267_cast_fp16)[name = tensor("op_29549_cast_fp16")]; + tensor var_29553_begin_0 = const()[name = tensor("op_29553_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_29553_end_0 = const()[name = tensor("op_29553_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_29553_end_mask_0 = const()[name = tensor("op_29553_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29553_cast_fp16 = slice_by_index(begin = var_29553_begin_0, end = var_29553_end_0, end_mask = var_29553_end_mask_0, x = k_267_cast_fp16)[name = tensor("op_29553_cast_fp16")]; + tensor var_29557_begin_0 = const()[name = tensor("op_29557_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_29557_end_0 = const()[name = tensor("op_29557_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_29557_end_mask_0 = const()[name = tensor("op_29557_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29557_cast_fp16 = slice_by_index(begin = var_29557_begin_0, end = var_29557_end_0, end_mask = var_29557_end_mask_0, x = k_267_cast_fp16)[name = tensor("op_29557_cast_fp16")]; + tensor var_29561_begin_0 = const()[name = tensor("op_29561_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_29561_end_0 = const()[name = tensor("op_29561_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_29561_end_mask_0 = const()[name = tensor("op_29561_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29561_cast_fp16 = slice_by_index(begin = var_29561_begin_0, end = var_29561_end_0, end_mask = var_29561_end_mask_0, x = k_267_cast_fp16)[name = tensor("op_29561_cast_fp16")]; + tensor var_29565_begin_0 = const()[name = tensor("op_29565_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_29565_end_0 = const()[name = tensor("op_29565_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_29565_end_mask_0 = const()[name = tensor("op_29565_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29565_cast_fp16 = slice_by_index(begin = var_29565_begin_0, end = var_29565_end_0, end_mask = var_29565_end_mask_0, x = k_267_cast_fp16)[name = tensor("op_29565_cast_fp16")]; + tensor var_29569_begin_0 = const()[name = tensor("op_29569_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_29569_end_0 = const()[name = tensor("op_29569_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_29569_end_mask_0 = const()[name = tensor("op_29569_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29569_cast_fp16 = slice_by_index(begin = var_29569_begin_0, end = var_29569_end_0, end_mask = var_29569_end_mask_0, x = k_267_cast_fp16)[name = tensor("op_29569_cast_fp16")]; + tensor var_29571_begin_0 = const()[name = tensor("op_29571_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29571_end_0 = const()[name = tensor("op_29571_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_29571_end_mask_0 = const()[name = tensor("op_29571_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29571_cast_fp16 = slice_by_index(begin = var_29571_begin_0, end = var_29571_end_0, end_mask = var_29571_end_mask_0, x = v_133_cast_fp16)[name = tensor("op_29571_cast_fp16")]; + tensor var_29575_begin_0 = const()[name = tensor("op_29575_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_29575_end_0 = const()[name = tensor("op_29575_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_29575_end_mask_0 = const()[name = tensor("op_29575_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29575_cast_fp16 = slice_by_index(begin = var_29575_begin_0, end = var_29575_end_0, end_mask = var_29575_end_mask_0, x = v_133_cast_fp16)[name = tensor("op_29575_cast_fp16")]; + tensor var_29579_begin_0 = const()[name = tensor("op_29579_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_29579_end_0 = const()[name = tensor("op_29579_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_29579_end_mask_0 = const()[name = tensor("op_29579_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29579_cast_fp16 = slice_by_index(begin = var_29579_begin_0, end = var_29579_end_0, end_mask = var_29579_end_mask_0, x = v_133_cast_fp16)[name = tensor("op_29579_cast_fp16")]; + tensor var_29583_begin_0 = const()[name = tensor("op_29583_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_29583_end_0 = const()[name = tensor("op_29583_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_29583_end_mask_0 = const()[name = tensor("op_29583_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29583_cast_fp16 = slice_by_index(begin = var_29583_begin_0, end = var_29583_end_0, end_mask = var_29583_end_mask_0, x = v_133_cast_fp16)[name = tensor("op_29583_cast_fp16")]; + tensor var_29587_begin_0 = const()[name = tensor("op_29587_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_29587_end_0 = const()[name = tensor("op_29587_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_29587_end_mask_0 = const()[name = tensor("op_29587_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29587_cast_fp16 = slice_by_index(begin = var_29587_begin_0, end = var_29587_end_0, end_mask = var_29587_end_mask_0, x = v_133_cast_fp16)[name = tensor("op_29587_cast_fp16")]; + tensor var_29591_begin_0 = const()[name = tensor("op_29591_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_29591_end_0 = const()[name = tensor("op_29591_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_29591_end_mask_0 = const()[name = tensor("op_29591_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29591_cast_fp16 = slice_by_index(begin = var_29591_begin_0, end = var_29591_end_0, end_mask = var_29591_end_mask_0, x = v_133_cast_fp16)[name = tensor("op_29591_cast_fp16")]; + tensor var_29595_begin_0 = const()[name = tensor("op_29595_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_29595_end_0 = const()[name = tensor("op_29595_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_29595_end_mask_0 = const()[name = tensor("op_29595_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29595_cast_fp16 = slice_by_index(begin = var_29595_begin_0, end = var_29595_end_0, end_mask = var_29595_end_mask_0, x = v_133_cast_fp16)[name = tensor("op_29595_cast_fp16")]; + tensor var_29599_begin_0 = const()[name = tensor("op_29599_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_29599_end_0 = const()[name = tensor("op_29599_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_29599_end_mask_0 = const()[name = tensor("op_29599_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29599_cast_fp16 = slice_by_index(begin = var_29599_begin_0, end = var_29599_end_0, end_mask = var_29599_end_mask_0, x = v_133_cast_fp16)[name = tensor("op_29599_cast_fp16")]; + tensor var_29603_begin_0 = const()[name = tensor("op_29603_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_29603_end_0 = const()[name = tensor("op_29603_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_29603_end_mask_0 = const()[name = tensor("op_29603_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29603_cast_fp16 = slice_by_index(begin = var_29603_begin_0, end = var_29603_end_0, end_mask = var_29603_end_mask_0, x = v_133_cast_fp16)[name = tensor("op_29603_cast_fp16")]; + tensor var_29607_begin_0 = const()[name = tensor("op_29607_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_29607_end_0 = const()[name = tensor("op_29607_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_29607_end_mask_0 = const()[name = tensor("op_29607_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29607_cast_fp16 = slice_by_index(begin = var_29607_begin_0, end = var_29607_end_0, end_mask = var_29607_end_mask_0, x = v_133_cast_fp16)[name = tensor("op_29607_cast_fp16")]; + tensor var_29611_begin_0 = const()[name = tensor("op_29611_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_29611_end_0 = const()[name = tensor("op_29611_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_29611_end_mask_0 = const()[name = tensor("op_29611_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29611_cast_fp16 = slice_by_index(begin = var_29611_begin_0, end = var_29611_end_0, end_mask = var_29611_end_mask_0, x = v_133_cast_fp16)[name = tensor("op_29611_cast_fp16")]; + tensor var_29615_begin_0 = const()[name = tensor("op_29615_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_29615_end_0 = const()[name = tensor("op_29615_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_29615_end_mask_0 = const()[name = tensor("op_29615_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29615_cast_fp16 = slice_by_index(begin = var_29615_begin_0, end = var_29615_end_0, end_mask = var_29615_end_mask_0, x = v_133_cast_fp16)[name = tensor("op_29615_cast_fp16")]; + tensor var_29619_begin_0 = const()[name = tensor("op_29619_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_29619_end_0 = const()[name = tensor("op_29619_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_29619_end_mask_0 = const()[name = tensor("op_29619_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29619_cast_fp16 = slice_by_index(begin = var_29619_begin_0, end = var_29619_end_0, end_mask = var_29619_end_mask_0, x = v_133_cast_fp16)[name = tensor("op_29619_cast_fp16")]; + tensor var_29623_begin_0 = const()[name = tensor("op_29623_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_29623_end_0 = const()[name = tensor("op_29623_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_29623_end_mask_0 = const()[name = tensor("op_29623_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29623_cast_fp16 = slice_by_index(begin = var_29623_begin_0, end = var_29623_end_0, end_mask = var_29623_end_mask_0, x = v_133_cast_fp16)[name = tensor("op_29623_cast_fp16")]; + tensor var_29627_begin_0 = const()[name = tensor("op_29627_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_29627_end_0 = const()[name = tensor("op_29627_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_29627_end_mask_0 = const()[name = tensor("op_29627_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29627_cast_fp16 = slice_by_index(begin = var_29627_begin_0, end = var_29627_end_0, end_mask = var_29627_end_mask_0, x = v_133_cast_fp16)[name = tensor("op_29627_cast_fp16")]; + tensor var_29631_begin_0 = const()[name = tensor("op_29631_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_29631_end_0 = const()[name = tensor("op_29631_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_29631_end_mask_0 = const()[name = tensor("op_29631_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29631_cast_fp16 = slice_by_index(begin = var_29631_begin_0, end = var_29631_end_0, end_mask = var_29631_end_mask_0, x = v_133_cast_fp16)[name = tensor("op_29631_cast_fp16")]; + tensor var_29635_begin_0 = const()[name = tensor("op_29635_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_29635_end_0 = const()[name = tensor("op_29635_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_29635_end_mask_0 = const()[name = tensor("op_29635_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29635_cast_fp16 = slice_by_index(begin = var_29635_begin_0, end = var_29635_end_0, end_mask = var_29635_end_mask_0, x = v_133_cast_fp16)[name = tensor("op_29635_cast_fp16")]; + tensor var_29639_begin_0 = const()[name = tensor("op_29639_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_29639_end_0 = const()[name = tensor("op_29639_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_29639_end_mask_0 = const()[name = tensor("op_29639_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29639_cast_fp16 = slice_by_index(begin = var_29639_begin_0, end = var_29639_end_0, end_mask = var_29639_end_mask_0, x = v_133_cast_fp16)[name = tensor("op_29639_cast_fp16")]; + tensor var_29643_begin_0 = const()[name = tensor("op_29643_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_29643_end_0 = const()[name = tensor("op_29643_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_29643_end_mask_0 = const()[name = tensor("op_29643_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29643_cast_fp16 = slice_by_index(begin = var_29643_begin_0, end = var_29643_end_0, end_mask = var_29643_end_mask_0, x = v_133_cast_fp16)[name = tensor("op_29643_cast_fp16")]; + tensor var_29647_begin_0 = const()[name = tensor("op_29647_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_29647_end_0 = const()[name = tensor("op_29647_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_29647_end_mask_0 = const()[name = tensor("op_29647_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29647_cast_fp16 = slice_by_index(begin = var_29647_begin_0, end = var_29647_end_0, end_mask = var_29647_end_mask_0, x = v_133_cast_fp16)[name = tensor("op_29647_cast_fp16")]; + tensor var_29651_equation_0 = const()[name = tensor("op_29651_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29651_cast_fp16 = einsum(equation = var_29651_equation_0, values = (var_29493_cast_fp16, var_29410_cast_fp16))[name = tensor("op_29651_cast_fp16")]; + tensor var_29652_to_fp16 = const()[name = tensor("op_29652_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2481_cast_fp16 = mul(x = var_29651_cast_fp16, y = var_29652_to_fp16)[name = tensor("aw_2481_cast_fp16")]; + tensor var_29655_equation_0 = const()[name = tensor("op_29655_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29655_cast_fp16 = einsum(equation = var_29655_equation_0, values = (var_29497_cast_fp16, var_29414_cast_fp16))[name = tensor("op_29655_cast_fp16")]; + tensor var_29656_to_fp16 = const()[name = tensor("op_29656_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2483_cast_fp16 = mul(x = var_29655_cast_fp16, y = var_29656_to_fp16)[name = tensor("aw_2483_cast_fp16")]; + tensor var_29659_equation_0 = const()[name = tensor("op_29659_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29659_cast_fp16 = einsum(equation = var_29659_equation_0, values = (var_29501_cast_fp16, var_29418_cast_fp16))[name = tensor("op_29659_cast_fp16")]; + tensor var_29660_to_fp16 = const()[name = tensor("op_29660_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2485_cast_fp16 = mul(x = var_29659_cast_fp16, y = var_29660_to_fp16)[name = tensor("aw_2485_cast_fp16")]; + tensor var_29663_equation_0 = const()[name = tensor("op_29663_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29663_cast_fp16 = einsum(equation = var_29663_equation_0, values = (var_29505_cast_fp16, var_29422_cast_fp16))[name = tensor("op_29663_cast_fp16")]; + tensor var_29664_to_fp16 = const()[name = tensor("op_29664_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2487_cast_fp16 = mul(x = var_29663_cast_fp16, y = var_29664_to_fp16)[name = tensor("aw_2487_cast_fp16")]; + tensor var_29667_equation_0 = const()[name = tensor("op_29667_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29667_cast_fp16 = einsum(equation = var_29667_equation_0, values = (var_29509_cast_fp16, var_29426_cast_fp16))[name = tensor("op_29667_cast_fp16")]; + tensor var_29668_to_fp16 = const()[name = tensor("op_29668_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2489_cast_fp16 = mul(x = var_29667_cast_fp16, y = var_29668_to_fp16)[name = tensor("aw_2489_cast_fp16")]; + tensor var_29671_equation_0 = const()[name = tensor("op_29671_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29671_cast_fp16 = einsum(equation = var_29671_equation_0, values = (var_29513_cast_fp16, var_29430_cast_fp16))[name = tensor("op_29671_cast_fp16")]; + tensor var_29672_to_fp16 = const()[name = tensor("op_29672_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2491_cast_fp16 = mul(x = var_29671_cast_fp16, y = var_29672_to_fp16)[name = tensor("aw_2491_cast_fp16")]; + tensor var_29675_equation_0 = const()[name = tensor("op_29675_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29675_cast_fp16 = einsum(equation = var_29675_equation_0, values = (var_29517_cast_fp16, var_29434_cast_fp16))[name = tensor("op_29675_cast_fp16")]; + tensor var_29676_to_fp16 = const()[name = tensor("op_29676_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2493_cast_fp16 = mul(x = var_29675_cast_fp16, y = var_29676_to_fp16)[name = tensor("aw_2493_cast_fp16")]; + tensor var_29679_equation_0 = const()[name = tensor("op_29679_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29679_cast_fp16 = einsum(equation = var_29679_equation_0, values = (var_29521_cast_fp16, var_29438_cast_fp16))[name = tensor("op_29679_cast_fp16")]; + tensor var_29680_to_fp16 = const()[name = tensor("op_29680_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2495_cast_fp16 = mul(x = var_29679_cast_fp16, y = var_29680_to_fp16)[name = tensor("aw_2495_cast_fp16")]; + tensor var_29683_equation_0 = const()[name = tensor("op_29683_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29683_cast_fp16 = einsum(equation = var_29683_equation_0, values = (var_29525_cast_fp16, var_29442_cast_fp16))[name = tensor("op_29683_cast_fp16")]; + tensor var_29684_to_fp16 = const()[name = tensor("op_29684_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2497_cast_fp16 = mul(x = var_29683_cast_fp16, y = var_29684_to_fp16)[name = tensor("aw_2497_cast_fp16")]; + tensor var_29687_equation_0 = const()[name = tensor("op_29687_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29687_cast_fp16 = einsum(equation = var_29687_equation_0, values = (var_29529_cast_fp16, var_29446_cast_fp16))[name = tensor("op_29687_cast_fp16")]; + tensor var_29688_to_fp16 = const()[name = tensor("op_29688_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2499_cast_fp16 = mul(x = var_29687_cast_fp16, y = var_29688_to_fp16)[name = tensor("aw_2499_cast_fp16")]; + tensor var_29691_equation_0 = const()[name = tensor("op_29691_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29691_cast_fp16 = einsum(equation = var_29691_equation_0, values = (var_29533_cast_fp16, var_29450_cast_fp16))[name = tensor("op_29691_cast_fp16")]; + tensor var_29692_to_fp16 = const()[name = tensor("op_29692_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2501_cast_fp16 = mul(x = var_29691_cast_fp16, y = var_29692_to_fp16)[name = tensor("aw_2501_cast_fp16")]; + tensor var_29695_equation_0 = const()[name = tensor("op_29695_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29695_cast_fp16 = einsum(equation = var_29695_equation_0, values = (var_29537_cast_fp16, var_29454_cast_fp16))[name = tensor("op_29695_cast_fp16")]; + tensor var_29696_to_fp16 = const()[name = tensor("op_29696_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2503_cast_fp16 = mul(x = var_29695_cast_fp16, y = var_29696_to_fp16)[name = tensor("aw_2503_cast_fp16")]; + tensor var_29699_equation_0 = const()[name = tensor("op_29699_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29699_cast_fp16 = einsum(equation = var_29699_equation_0, values = (var_29541_cast_fp16, var_29458_cast_fp16))[name = tensor("op_29699_cast_fp16")]; + tensor var_29700_to_fp16 = const()[name = tensor("op_29700_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2505_cast_fp16 = mul(x = var_29699_cast_fp16, y = var_29700_to_fp16)[name = tensor("aw_2505_cast_fp16")]; + tensor var_29703_equation_0 = const()[name = tensor("op_29703_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29703_cast_fp16 = einsum(equation = var_29703_equation_0, values = (var_29545_cast_fp16, var_29462_cast_fp16))[name = tensor("op_29703_cast_fp16")]; + tensor var_29704_to_fp16 = const()[name = tensor("op_29704_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2507_cast_fp16 = mul(x = var_29703_cast_fp16, y = var_29704_to_fp16)[name = tensor("aw_2507_cast_fp16")]; + tensor var_29707_equation_0 = const()[name = tensor("op_29707_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29707_cast_fp16 = einsum(equation = var_29707_equation_0, values = (var_29549_cast_fp16, var_29466_cast_fp16))[name = tensor("op_29707_cast_fp16")]; + tensor var_29708_to_fp16 = const()[name = tensor("op_29708_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2509_cast_fp16 = mul(x = var_29707_cast_fp16, y = var_29708_to_fp16)[name = tensor("aw_2509_cast_fp16")]; + tensor var_29711_equation_0 = const()[name = tensor("op_29711_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29711_cast_fp16 = einsum(equation = var_29711_equation_0, values = (var_29553_cast_fp16, var_29470_cast_fp16))[name = tensor("op_29711_cast_fp16")]; + tensor var_29712_to_fp16 = const()[name = tensor("op_29712_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2511_cast_fp16 = mul(x = var_29711_cast_fp16, y = var_29712_to_fp16)[name = tensor("aw_2511_cast_fp16")]; + tensor var_29715_equation_0 = const()[name = tensor("op_29715_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29715_cast_fp16 = einsum(equation = var_29715_equation_0, values = (var_29557_cast_fp16, var_29474_cast_fp16))[name = tensor("op_29715_cast_fp16")]; + tensor var_29716_to_fp16 = const()[name = tensor("op_29716_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2513_cast_fp16 = mul(x = var_29715_cast_fp16, y = var_29716_to_fp16)[name = tensor("aw_2513_cast_fp16")]; + tensor var_29719_equation_0 = const()[name = tensor("op_29719_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29719_cast_fp16 = einsum(equation = var_29719_equation_0, values = (var_29561_cast_fp16, var_29478_cast_fp16))[name = tensor("op_29719_cast_fp16")]; + tensor var_29720_to_fp16 = const()[name = tensor("op_29720_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2515_cast_fp16 = mul(x = var_29719_cast_fp16, y = var_29720_to_fp16)[name = tensor("aw_2515_cast_fp16")]; + tensor var_29723_equation_0 = const()[name = tensor("op_29723_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29723_cast_fp16 = einsum(equation = var_29723_equation_0, values = (var_29565_cast_fp16, var_29482_cast_fp16))[name = tensor("op_29723_cast_fp16")]; + tensor var_29724_to_fp16 = const()[name = tensor("op_29724_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2517_cast_fp16 = mul(x = var_29723_cast_fp16, y = var_29724_to_fp16)[name = tensor("aw_2517_cast_fp16")]; + tensor var_29727_equation_0 = const()[name = tensor("op_29727_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29727_cast_fp16 = einsum(equation = var_29727_equation_0, values = (var_29569_cast_fp16, var_29486_cast_fp16))[name = tensor("op_29727_cast_fp16")]; + tensor var_29728_to_fp16 = const()[name = tensor("op_29728_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2519_cast_fp16 = mul(x = var_29727_cast_fp16, y = var_29728_to_fp16)[name = tensor("aw_2519_cast_fp16")]; + tensor var_29730_cast_fp16 = softmax(axis = var_21077, x = aw_2481_cast_fp16)[name = tensor("op_29730_cast_fp16")]; + tensor var_29731_cast_fp16 = softmax(axis = var_21077, x = aw_2483_cast_fp16)[name = tensor("op_29731_cast_fp16")]; + tensor var_29732_cast_fp16 = softmax(axis = var_21077, x = aw_2485_cast_fp16)[name = tensor("op_29732_cast_fp16")]; + tensor var_29733_cast_fp16 = softmax(axis = var_21077, x = aw_2487_cast_fp16)[name = tensor("op_29733_cast_fp16")]; + tensor var_29734_cast_fp16 = softmax(axis = var_21077, x = aw_2489_cast_fp16)[name = tensor("op_29734_cast_fp16")]; + tensor var_29735_cast_fp16 = softmax(axis = var_21077, x = aw_2491_cast_fp16)[name = tensor("op_29735_cast_fp16")]; + tensor var_29736_cast_fp16 = softmax(axis = var_21077, x = aw_2493_cast_fp16)[name = tensor("op_29736_cast_fp16")]; + tensor var_29737_cast_fp16 = softmax(axis = var_21077, x = aw_2495_cast_fp16)[name = tensor("op_29737_cast_fp16")]; + tensor var_29738_cast_fp16 = softmax(axis = var_21077, x = aw_2497_cast_fp16)[name = tensor("op_29738_cast_fp16")]; + tensor var_29739_cast_fp16 = softmax(axis = var_21077, x = aw_2499_cast_fp16)[name = tensor("op_29739_cast_fp16")]; + tensor var_29740_cast_fp16 = softmax(axis = var_21077, x = aw_2501_cast_fp16)[name = tensor("op_29740_cast_fp16")]; + tensor var_29741_cast_fp16 = softmax(axis = var_21077, x = aw_2503_cast_fp16)[name = tensor("op_29741_cast_fp16")]; + tensor var_29742_cast_fp16 = softmax(axis = var_21077, x = aw_2505_cast_fp16)[name = tensor("op_29742_cast_fp16")]; + tensor var_29743_cast_fp16 = softmax(axis = var_21077, x = aw_2507_cast_fp16)[name = tensor("op_29743_cast_fp16")]; + tensor var_29744_cast_fp16 = softmax(axis = var_21077, x = aw_2509_cast_fp16)[name = tensor("op_29744_cast_fp16")]; + tensor var_29745_cast_fp16 = softmax(axis = var_21077, x = aw_2511_cast_fp16)[name = tensor("op_29745_cast_fp16")]; + tensor var_29746_cast_fp16 = softmax(axis = var_21077, x = aw_2513_cast_fp16)[name = tensor("op_29746_cast_fp16")]; + tensor var_29747_cast_fp16 = softmax(axis = var_21077, x = aw_2515_cast_fp16)[name = tensor("op_29747_cast_fp16")]; + tensor var_29748_cast_fp16 = softmax(axis = var_21077, x = aw_2517_cast_fp16)[name = tensor("op_29748_cast_fp16")]; + tensor var_29749_cast_fp16 = softmax(axis = var_21077, x = aw_2519_cast_fp16)[name = tensor("op_29749_cast_fp16")]; + tensor var_29751_equation_0 = const()[name = tensor("op_29751_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29751_cast_fp16 = einsum(equation = var_29751_equation_0, values = (var_29571_cast_fp16, var_29730_cast_fp16))[name = tensor("op_29751_cast_fp16")]; + tensor var_29753_equation_0 = const()[name = tensor("op_29753_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29753_cast_fp16 = einsum(equation = var_29753_equation_0, values = (var_29575_cast_fp16, var_29731_cast_fp16))[name = tensor("op_29753_cast_fp16")]; + tensor var_29755_equation_0 = const()[name = tensor("op_29755_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29755_cast_fp16 = einsum(equation = var_29755_equation_0, values = (var_29579_cast_fp16, var_29732_cast_fp16))[name = tensor("op_29755_cast_fp16")]; + tensor var_29757_equation_0 = const()[name = tensor("op_29757_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29757_cast_fp16 = einsum(equation = var_29757_equation_0, values = (var_29583_cast_fp16, var_29733_cast_fp16))[name = tensor("op_29757_cast_fp16")]; + tensor var_29759_equation_0 = const()[name = tensor("op_29759_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29759_cast_fp16 = einsum(equation = var_29759_equation_0, values = (var_29587_cast_fp16, var_29734_cast_fp16))[name = tensor("op_29759_cast_fp16")]; + tensor var_29761_equation_0 = const()[name = tensor("op_29761_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29761_cast_fp16 = einsum(equation = var_29761_equation_0, values = (var_29591_cast_fp16, var_29735_cast_fp16))[name = tensor("op_29761_cast_fp16")]; + tensor var_29763_equation_0 = const()[name = tensor("op_29763_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29763_cast_fp16 = einsum(equation = var_29763_equation_0, values = (var_29595_cast_fp16, var_29736_cast_fp16))[name = tensor("op_29763_cast_fp16")]; + tensor var_29765_equation_0 = const()[name = tensor("op_29765_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29765_cast_fp16 = einsum(equation = var_29765_equation_0, values = (var_29599_cast_fp16, var_29737_cast_fp16))[name = tensor("op_29765_cast_fp16")]; + tensor var_29767_equation_0 = const()[name = tensor("op_29767_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29767_cast_fp16 = einsum(equation = var_29767_equation_0, values = (var_29603_cast_fp16, var_29738_cast_fp16))[name = tensor("op_29767_cast_fp16")]; + tensor var_29769_equation_0 = const()[name = tensor("op_29769_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29769_cast_fp16 = einsum(equation = var_29769_equation_0, values = (var_29607_cast_fp16, var_29739_cast_fp16))[name = tensor("op_29769_cast_fp16")]; + tensor var_29771_equation_0 = const()[name = tensor("op_29771_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29771_cast_fp16 = einsum(equation = var_29771_equation_0, values = (var_29611_cast_fp16, var_29740_cast_fp16))[name = tensor("op_29771_cast_fp16")]; + tensor var_29773_equation_0 = const()[name = tensor("op_29773_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29773_cast_fp16 = einsum(equation = var_29773_equation_0, values = (var_29615_cast_fp16, var_29741_cast_fp16))[name = tensor("op_29773_cast_fp16")]; + tensor var_29775_equation_0 = const()[name = tensor("op_29775_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29775_cast_fp16 = einsum(equation = var_29775_equation_0, values = (var_29619_cast_fp16, var_29742_cast_fp16))[name = tensor("op_29775_cast_fp16")]; + tensor var_29777_equation_0 = const()[name = tensor("op_29777_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29777_cast_fp16 = einsum(equation = var_29777_equation_0, values = (var_29623_cast_fp16, var_29743_cast_fp16))[name = tensor("op_29777_cast_fp16")]; + tensor var_29779_equation_0 = const()[name = tensor("op_29779_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29779_cast_fp16 = einsum(equation = var_29779_equation_0, values = (var_29627_cast_fp16, var_29744_cast_fp16))[name = tensor("op_29779_cast_fp16")]; + tensor var_29781_equation_0 = const()[name = tensor("op_29781_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29781_cast_fp16 = einsum(equation = var_29781_equation_0, values = (var_29631_cast_fp16, var_29745_cast_fp16))[name = tensor("op_29781_cast_fp16")]; + tensor var_29783_equation_0 = const()[name = tensor("op_29783_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29783_cast_fp16 = einsum(equation = var_29783_equation_0, values = (var_29635_cast_fp16, var_29746_cast_fp16))[name = tensor("op_29783_cast_fp16")]; + tensor var_29785_equation_0 = const()[name = tensor("op_29785_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29785_cast_fp16 = einsum(equation = var_29785_equation_0, values = (var_29639_cast_fp16, var_29747_cast_fp16))[name = tensor("op_29785_cast_fp16")]; + tensor var_29787_equation_0 = const()[name = tensor("op_29787_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29787_cast_fp16 = einsum(equation = var_29787_equation_0, values = (var_29643_cast_fp16, var_29748_cast_fp16))[name = tensor("op_29787_cast_fp16")]; + tensor var_29789_equation_0 = const()[name = tensor("op_29789_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29789_cast_fp16 = einsum(equation = var_29789_equation_0, values = (var_29647_cast_fp16, var_29749_cast_fp16))[name = tensor("op_29789_cast_fp16")]; + tensor input_399_interleave_0 = const()[name = tensor("input_399_interleave_0"), val = tensor(false)]; + tensor input_399_cast_fp16 = concat(axis = var_21077, interleave = input_399_interleave_0, values = (var_29751_cast_fp16, var_29753_cast_fp16, var_29755_cast_fp16, var_29757_cast_fp16, var_29759_cast_fp16, var_29761_cast_fp16, var_29763_cast_fp16, var_29765_cast_fp16, var_29767_cast_fp16, var_29769_cast_fp16, var_29771_cast_fp16, var_29773_cast_fp16, var_29775_cast_fp16, var_29777_cast_fp16, var_29779_cast_fp16, var_29781_cast_fp16, var_29783_cast_fp16, var_29785_cast_fp16, var_29787_cast_fp16, var_29789_cast_fp16))[name = tensor("input_399_cast_fp16")]; + tensor var_29795 = const()[name = tensor("op_29795"), val = tensor([1, 1])]; + tensor var_29797 = const()[name = tensor("op_29797"), val = tensor([1, 1])]; + tensor var_29799_pad_type_0 = const()[name = tensor("op_29799_pad_type_0"), val = tensor("custom")]; + tensor var_29799_pad_0 = const()[name = tensor("op_29799_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_9_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_9_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2361561536)))]; + tensor mid_block_attentions_0_transformer_blocks_9_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_9_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2364838400)))]; + tensor var_29799_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_9_attn1_to_out_0_bias_to_fp16, dilations = var_29797, groups = var_21077, pad = var_29799_pad_0, pad_type = var_29799_pad_type_0, strides = var_29795, weight = mid_block_attentions_0_transformer_blocks_9_attn1_to_out_0_weight_to_fp16, x = input_399_cast_fp16)[name = tensor("op_29799_cast_fp16")]; + tensor inputs_201_cast_fp16 = add(x = var_29799_cast_fp16, y = inputs_199_cast_fp16)[name = tensor("inputs_201_cast_fp16")]; + tensor hidden_states_265_axes_0 = const()[name = tensor("hidden_states_265_axes_0"), val = tensor([1])]; + tensor hidden_states_265_gamma_0_to_fp16 = const()[name = tensor("hidden_states_265_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2364841024)))]; + tensor hidden_states_265_beta_0_to_fp16 = const()[name = tensor("hidden_states_265_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2364843648)))]; + tensor var_29809_to_fp16 = const()[name = tensor("op_29809_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_265_cast_fp16 = layer_norm(axes = hidden_states_265_axes_0, beta = hidden_states_265_beta_0_to_fp16, epsilon = var_29809_to_fp16, gamma = hidden_states_265_gamma_0_to_fp16, x = inputs_201_cast_fp16)[name = tensor("hidden_states_265_cast_fp16")]; + tensor var_29824 = const()[name = tensor("op_29824"), val = tensor([1, 1])]; + tensor var_29826 = const()[name = tensor("op_29826"), val = tensor([1, 1])]; + tensor q_135_pad_type_0 = const()[name = tensor("q_135_pad_type_0"), val = tensor("custom")]; + tensor q_135_pad_0 = const()[name = tensor("q_135_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_9_attn2_to_q_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_9_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2364846272)))]; + tensor q_135_cast_fp16 = conv(dilations = var_29826, groups = var_21077, pad = q_135_pad_0, pad_type = q_135_pad_type_0, strides = var_29824, weight = mid_block_attentions_0_transformer_blocks_9_attn2_to_q_weight_to_fp16, x = hidden_states_265_cast_fp16)[name = tensor("q_135_cast_fp16")]; + tensor var_29830 = const()[name = tensor("op_29830"), val = tensor([1, 1])]; + tensor var_29832 = const()[name = tensor("op_29832"), val = tensor([1, 1])]; + tensor k_269_pad_type_0 = const()[name = tensor("k_269_pad_type_0"), val = tensor("custom")]; + tensor k_269_pad_0 = const()[name = tensor("k_269_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_9_attn2_to_k_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_9_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2368123136)))]; + tensor k_269_cast_fp16 = conv(dilations = var_29832, groups = var_21077, pad = k_269_pad_0, pad_type = k_269_pad_type_0, strides = var_29830, weight = mid_block_attentions_0_transformer_blocks_9_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_269_cast_fp16")]; + tensor var_29836 = const()[name = tensor("op_29836"), val = tensor([1, 1])]; + tensor var_29838 = const()[name = tensor("op_29838"), val = tensor([1, 1])]; + tensor v_135_pad_type_0 = const()[name = tensor("v_135_pad_type_0"), val = tensor("custom")]; + tensor v_135_pad_0 = const()[name = tensor("v_135_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_9_attn2_to_v_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_9_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2373366080)))]; + tensor v_135_cast_fp16 = conv(dilations = var_29838, groups = var_21077, pad = v_135_pad_0, pad_type = v_135_pad_type_0, strides = var_29836, weight = mid_block_attentions_0_transformer_blocks_9_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_135_cast_fp16")]; + tensor var_29842_begin_0 = const()[name = tensor("op_29842_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29842_end_0 = const()[name = tensor("op_29842_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_29842_end_mask_0 = const()[name = tensor("op_29842_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29842_cast_fp16 = slice_by_index(begin = var_29842_begin_0, end = var_29842_end_0, end_mask = var_29842_end_mask_0, x = q_135_cast_fp16)[name = tensor("op_29842_cast_fp16")]; + tensor var_29846_begin_0 = const()[name = tensor("op_29846_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_29846_end_0 = const()[name = tensor("op_29846_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_29846_end_mask_0 = const()[name = tensor("op_29846_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29846_cast_fp16 = slice_by_index(begin = var_29846_begin_0, end = var_29846_end_0, end_mask = var_29846_end_mask_0, x = q_135_cast_fp16)[name = tensor("op_29846_cast_fp16")]; + tensor var_29850_begin_0 = const()[name = tensor("op_29850_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_29850_end_0 = const()[name = tensor("op_29850_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_29850_end_mask_0 = const()[name = tensor("op_29850_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29850_cast_fp16 = slice_by_index(begin = var_29850_begin_0, end = var_29850_end_0, end_mask = var_29850_end_mask_0, x = q_135_cast_fp16)[name = tensor("op_29850_cast_fp16")]; + tensor var_29854_begin_0 = const()[name = tensor("op_29854_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_29854_end_0 = const()[name = tensor("op_29854_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_29854_end_mask_0 = const()[name = tensor("op_29854_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29854_cast_fp16 = slice_by_index(begin = var_29854_begin_0, end = var_29854_end_0, end_mask = var_29854_end_mask_0, x = q_135_cast_fp16)[name = tensor("op_29854_cast_fp16")]; + tensor var_29858_begin_0 = const()[name = tensor("op_29858_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_29858_end_0 = const()[name = tensor("op_29858_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_29858_end_mask_0 = const()[name = tensor("op_29858_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29858_cast_fp16 = slice_by_index(begin = var_29858_begin_0, end = var_29858_end_0, end_mask = var_29858_end_mask_0, x = q_135_cast_fp16)[name = tensor("op_29858_cast_fp16")]; + tensor var_29862_begin_0 = const()[name = tensor("op_29862_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_29862_end_0 = const()[name = tensor("op_29862_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_29862_end_mask_0 = const()[name = tensor("op_29862_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29862_cast_fp16 = slice_by_index(begin = var_29862_begin_0, end = var_29862_end_0, end_mask = var_29862_end_mask_0, x = q_135_cast_fp16)[name = tensor("op_29862_cast_fp16")]; + tensor var_29866_begin_0 = const()[name = tensor("op_29866_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_29866_end_0 = const()[name = tensor("op_29866_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_29866_end_mask_0 = const()[name = tensor("op_29866_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29866_cast_fp16 = slice_by_index(begin = var_29866_begin_0, end = var_29866_end_0, end_mask = var_29866_end_mask_0, x = q_135_cast_fp16)[name = tensor("op_29866_cast_fp16")]; + tensor var_29870_begin_0 = const()[name = tensor("op_29870_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_29870_end_0 = const()[name = tensor("op_29870_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_29870_end_mask_0 = const()[name = tensor("op_29870_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29870_cast_fp16 = slice_by_index(begin = var_29870_begin_0, end = var_29870_end_0, end_mask = var_29870_end_mask_0, x = q_135_cast_fp16)[name = tensor("op_29870_cast_fp16")]; + tensor var_29874_begin_0 = const()[name = tensor("op_29874_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_29874_end_0 = const()[name = tensor("op_29874_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_29874_end_mask_0 = const()[name = tensor("op_29874_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29874_cast_fp16 = slice_by_index(begin = var_29874_begin_0, end = var_29874_end_0, end_mask = var_29874_end_mask_0, x = q_135_cast_fp16)[name = tensor("op_29874_cast_fp16")]; + tensor var_29878_begin_0 = const()[name = tensor("op_29878_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_29878_end_0 = const()[name = tensor("op_29878_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_29878_end_mask_0 = const()[name = tensor("op_29878_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29878_cast_fp16 = slice_by_index(begin = var_29878_begin_0, end = var_29878_end_0, end_mask = var_29878_end_mask_0, x = q_135_cast_fp16)[name = tensor("op_29878_cast_fp16")]; + tensor var_29882_begin_0 = const()[name = tensor("op_29882_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_29882_end_0 = const()[name = tensor("op_29882_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_29882_end_mask_0 = const()[name = tensor("op_29882_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29882_cast_fp16 = slice_by_index(begin = var_29882_begin_0, end = var_29882_end_0, end_mask = var_29882_end_mask_0, x = q_135_cast_fp16)[name = tensor("op_29882_cast_fp16")]; + tensor var_29886_begin_0 = const()[name = tensor("op_29886_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_29886_end_0 = const()[name = tensor("op_29886_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_29886_end_mask_0 = const()[name = tensor("op_29886_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29886_cast_fp16 = slice_by_index(begin = var_29886_begin_0, end = var_29886_end_0, end_mask = var_29886_end_mask_0, x = q_135_cast_fp16)[name = tensor("op_29886_cast_fp16")]; + tensor var_29890_begin_0 = const()[name = tensor("op_29890_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_29890_end_0 = const()[name = tensor("op_29890_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_29890_end_mask_0 = const()[name = tensor("op_29890_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29890_cast_fp16 = slice_by_index(begin = var_29890_begin_0, end = var_29890_end_0, end_mask = var_29890_end_mask_0, x = q_135_cast_fp16)[name = tensor("op_29890_cast_fp16")]; + tensor var_29894_begin_0 = const()[name = tensor("op_29894_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_29894_end_0 = const()[name = tensor("op_29894_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_29894_end_mask_0 = const()[name = tensor("op_29894_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29894_cast_fp16 = slice_by_index(begin = var_29894_begin_0, end = var_29894_end_0, end_mask = var_29894_end_mask_0, x = q_135_cast_fp16)[name = tensor("op_29894_cast_fp16")]; + tensor var_29898_begin_0 = const()[name = tensor("op_29898_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_29898_end_0 = const()[name = tensor("op_29898_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_29898_end_mask_0 = const()[name = tensor("op_29898_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29898_cast_fp16 = slice_by_index(begin = var_29898_begin_0, end = var_29898_end_0, end_mask = var_29898_end_mask_0, x = q_135_cast_fp16)[name = tensor("op_29898_cast_fp16")]; + tensor var_29902_begin_0 = const()[name = tensor("op_29902_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_29902_end_0 = const()[name = tensor("op_29902_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_29902_end_mask_0 = const()[name = tensor("op_29902_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29902_cast_fp16 = slice_by_index(begin = var_29902_begin_0, end = var_29902_end_0, end_mask = var_29902_end_mask_0, x = q_135_cast_fp16)[name = tensor("op_29902_cast_fp16")]; + tensor var_29906_begin_0 = const()[name = tensor("op_29906_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_29906_end_0 = const()[name = tensor("op_29906_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_29906_end_mask_0 = const()[name = tensor("op_29906_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29906_cast_fp16 = slice_by_index(begin = var_29906_begin_0, end = var_29906_end_0, end_mask = var_29906_end_mask_0, x = q_135_cast_fp16)[name = tensor("op_29906_cast_fp16")]; + tensor var_29910_begin_0 = const()[name = tensor("op_29910_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_29910_end_0 = const()[name = tensor("op_29910_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_29910_end_mask_0 = const()[name = tensor("op_29910_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29910_cast_fp16 = slice_by_index(begin = var_29910_begin_0, end = var_29910_end_0, end_mask = var_29910_end_mask_0, x = q_135_cast_fp16)[name = tensor("op_29910_cast_fp16")]; + tensor var_29914_begin_0 = const()[name = tensor("op_29914_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_29914_end_0 = const()[name = tensor("op_29914_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_29914_end_mask_0 = const()[name = tensor("op_29914_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29914_cast_fp16 = slice_by_index(begin = var_29914_begin_0, end = var_29914_end_0, end_mask = var_29914_end_mask_0, x = q_135_cast_fp16)[name = tensor("op_29914_cast_fp16")]; + tensor var_29918_begin_0 = const()[name = tensor("op_29918_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_29918_end_0 = const()[name = tensor("op_29918_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_29918_end_mask_0 = const()[name = tensor("op_29918_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29918_cast_fp16 = slice_by_index(begin = var_29918_begin_0, end = var_29918_end_0, end_mask = var_29918_end_mask_0, x = q_135_cast_fp16)[name = tensor("op_29918_cast_fp16")]; + tensor k_271_perm_0 = const()[name = tensor("k_271_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_29925_begin_0 = const()[name = tensor("op_29925_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29925_end_0 = const()[name = tensor("op_29925_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_29925_end_mask_0 = const()[name = tensor("op_29925_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_271_cast_fp16 = transpose(perm = k_271_perm_0, x = k_269_cast_fp16)[name = tensor("transpose_72")]; + tensor var_29925_cast_fp16 = slice_by_index(begin = var_29925_begin_0, end = var_29925_end_0, end_mask = var_29925_end_mask_0, x = k_271_cast_fp16)[name = tensor("op_29925_cast_fp16")]; + tensor var_29929_begin_0 = const()[name = tensor("op_29929_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_29929_end_0 = const()[name = tensor("op_29929_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_29929_end_mask_0 = const()[name = tensor("op_29929_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29929_cast_fp16 = slice_by_index(begin = var_29929_begin_0, end = var_29929_end_0, end_mask = var_29929_end_mask_0, x = k_271_cast_fp16)[name = tensor("op_29929_cast_fp16")]; + tensor var_29933_begin_0 = const()[name = tensor("op_29933_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_29933_end_0 = const()[name = tensor("op_29933_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_29933_end_mask_0 = const()[name = tensor("op_29933_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29933_cast_fp16 = slice_by_index(begin = var_29933_begin_0, end = var_29933_end_0, end_mask = var_29933_end_mask_0, x = k_271_cast_fp16)[name = tensor("op_29933_cast_fp16")]; + tensor var_29937_begin_0 = const()[name = tensor("op_29937_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_29937_end_0 = const()[name = tensor("op_29937_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_29937_end_mask_0 = const()[name = tensor("op_29937_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29937_cast_fp16 = slice_by_index(begin = var_29937_begin_0, end = var_29937_end_0, end_mask = var_29937_end_mask_0, x = k_271_cast_fp16)[name = tensor("op_29937_cast_fp16")]; + tensor var_29941_begin_0 = const()[name = tensor("op_29941_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_29941_end_0 = const()[name = tensor("op_29941_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_29941_end_mask_0 = const()[name = tensor("op_29941_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29941_cast_fp16 = slice_by_index(begin = var_29941_begin_0, end = var_29941_end_0, end_mask = var_29941_end_mask_0, x = k_271_cast_fp16)[name = tensor("op_29941_cast_fp16")]; + tensor var_29945_begin_0 = const()[name = tensor("op_29945_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_29945_end_0 = const()[name = tensor("op_29945_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_29945_end_mask_0 = const()[name = tensor("op_29945_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29945_cast_fp16 = slice_by_index(begin = var_29945_begin_0, end = var_29945_end_0, end_mask = var_29945_end_mask_0, x = k_271_cast_fp16)[name = tensor("op_29945_cast_fp16")]; + tensor var_29949_begin_0 = const()[name = tensor("op_29949_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_29949_end_0 = const()[name = tensor("op_29949_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_29949_end_mask_0 = const()[name = tensor("op_29949_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29949_cast_fp16 = slice_by_index(begin = var_29949_begin_0, end = var_29949_end_0, end_mask = var_29949_end_mask_0, x = k_271_cast_fp16)[name = tensor("op_29949_cast_fp16")]; + tensor var_29953_begin_0 = const()[name = tensor("op_29953_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_29953_end_0 = const()[name = tensor("op_29953_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_29953_end_mask_0 = const()[name = tensor("op_29953_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29953_cast_fp16 = slice_by_index(begin = var_29953_begin_0, end = var_29953_end_0, end_mask = var_29953_end_mask_0, x = k_271_cast_fp16)[name = tensor("op_29953_cast_fp16")]; + tensor var_29957_begin_0 = const()[name = tensor("op_29957_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_29957_end_0 = const()[name = tensor("op_29957_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_29957_end_mask_0 = const()[name = tensor("op_29957_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29957_cast_fp16 = slice_by_index(begin = var_29957_begin_0, end = var_29957_end_0, end_mask = var_29957_end_mask_0, x = k_271_cast_fp16)[name = tensor("op_29957_cast_fp16")]; + tensor var_29961_begin_0 = const()[name = tensor("op_29961_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_29961_end_0 = const()[name = tensor("op_29961_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_29961_end_mask_0 = const()[name = tensor("op_29961_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29961_cast_fp16 = slice_by_index(begin = var_29961_begin_0, end = var_29961_end_0, end_mask = var_29961_end_mask_0, x = k_271_cast_fp16)[name = tensor("op_29961_cast_fp16")]; + tensor var_29965_begin_0 = const()[name = tensor("op_29965_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_29965_end_0 = const()[name = tensor("op_29965_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_29965_end_mask_0 = const()[name = tensor("op_29965_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29965_cast_fp16 = slice_by_index(begin = var_29965_begin_0, end = var_29965_end_0, end_mask = var_29965_end_mask_0, x = k_271_cast_fp16)[name = tensor("op_29965_cast_fp16")]; + tensor var_29969_begin_0 = const()[name = tensor("op_29969_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_29969_end_0 = const()[name = tensor("op_29969_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_29969_end_mask_0 = const()[name = tensor("op_29969_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29969_cast_fp16 = slice_by_index(begin = var_29969_begin_0, end = var_29969_end_0, end_mask = var_29969_end_mask_0, x = k_271_cast_fp16)[name = tensor("op_29969_cast_fp16")]; + tensor var_29973_begin_0 = const()[name = tensor("op_29973_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_29973_end_0 = const()[name = tensor("op_29973_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_29973_end_mask_0 = const()[name = tensor("op_29973_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29973_cast_fp16 = slice_by_index(begin = var_29973_begin_0, end = var_29973_end_0, end_mask = var_29973_end_mask_0, x = k_271_cast_fp16)[name = tensor("op_29973_cast_fp16")]; + tensor var_29977_begin_0 = const()[name = tensor("op_29977_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_29977_end_0 = const()[name = tensor("op_29977_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_29977_end_mask_0 = const()[name = tensor("op_29977_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29977_cast_fp16 = slice_by_index(begin = var_29977_begin_0, end = var_29977_end_0, end_mask = var_29977_end_mask_0, x = k_271_cast_fp16)[name = tensor("op_29977_cast_fp16")]; + tensor var_29981_begin_0 = const()[name = tensor("op_29981_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_29981_end_0 = const()[name = tensor("op_29981_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_29981_end_mask_0 = const()[name = tensor("op_29981_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29981_cast_fp16 = slice_by_index(begin = var_29981_begin_0, end = var_29981_end_0, end_mask = var_29981_end_mask_0, x = k_271_cast_fp16)[name = tensor("op_29981_cast_fp16")]; + tensor var_29985_begin_0 = const()[name = tensor("op_29985_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_29985_end_0 = const()[name = tensor("op_29985_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_29985_end_mask_0 = const()[name = tensor("op_29985_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29985_cast_fp16 = slice_by_index(begin = var_29985_begin_0, end = var_29985_end_0, end_mask = var_29985_end_mask_0, x = k_271_cast_fp16)[name = tensor("op_29985_cast_fp16")]; + tensor var_29989_begin_0 = const()[name = tensor("op_29989_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_29989_end_0 = const()[name = tensor("op_29989_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_29989_end_mask_0 = const()[name = tensor("op_29989_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29989_cast_fp16 = slice_by_index(begin = var_29989_begin_0, end = var_29989_end_0, end_mask = var_29989_end_mask_0, x = k_271_cast_fp16)[name = tensor("op_29989_cast_fp16")]; + tensor var_29993_begin_0 = const()[name = tensor("op_29993_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_29993_end_0 = const()[name = tensor("op_29993_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_29993_end_mask_0 = const()[name = tensor("op_29993_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29993_cast_fp16 = slice_by_index(begin = var_29993_begin_0, end = var_29993_end_0, end_mask = var_29993_end_mask_0, x = k_271_cast_fp16)[name = tensor("op_29993_cast_fp16")]; + tensor var_29997_begin_0 = const()[name = tensor("op_29997_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_29997_end_0 = const()[name = tensor("op_29997_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_29997_end_mask_0 = const()[name = tensor("op_29997_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29997_cast_fp16 = slice_by_index(begin = var_29997_begin_0, end = var_29997_end_0, end_mask = var_29997_end_mask_0, x = k_271_cast_fp16)[name = tensor("op_29997_cast_fp16")]; + tensor var_30001_begin_0 = const()[name = tensor("op_30001_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_30001_end_0 = const()[name = tensor("op_30001_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_30001_end_mask_0 = const()[name = tensor("op_30001_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30001_cast_fp16 = slice_by_index(begin = var_30001_begin_0, end = var_30001_end_0, end_mask = var_30001_end_mask_0, x = k_271_cast_fp16)[name = tensor("op_30001_cast_fp16")]; + tensor var_30003_begin_0 = const()[name = tensor("op_30003_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_30003_end_0 = const()[name = tensor("op_30003_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_30003_end_mask_0 = const()[name = tensor("op_30003_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30003_cast_fp16 = slice_by_index(begin = var_30003_begin_0, end = var_30003_end_0, end_mask = var_30003_end_mask_0, x = v_135_cast_fp16)[name = tensor("op_30003_cast_fp16")]; + tensor var_30007_begin_0 = const()[name = tensor("op_30007_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_30007_end_0 = const()[name = tensor("op_30007_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_30007_end_mask_0 = const()[name = tensor("op_30007_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30007_cast_fp16 = slice_by_index(begin = var_30007_begin_0, end = var_30007_end_0, end_mask = var_30007_end_mask_0, x = v_135_cast_fp16)[name = tensor("op_30007_cast_fp16")]; + tensor var_30011_begin_0 = const()[name = tensor("op_30011_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_30011_end_0 = const()[name = tensor("op_30011_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_30011_end_mask_0 = const()[name = tensor("op_30011_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30011_cast_fp16 = slice_by_index(begin = var_30011_begin_0, end = var_30011_end_0, end_mask = var_30011_end_mask_0, x = v_135_cast_fp16)[name = tensor("op_30011_cast_fp16")]; + tensor var_30015_begin_0 = const()[name = tensor("op_30015_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_30015_end_0 = const()[name = tensor("op_30015_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_30015_end_mask_0 = const()[name = tensor("op_30015_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30015_cast_fp16 = slice_by_index(begin = var_30015_begin_0, end = var_30015_end_0, end_mask = var_30015_end_mask_0, x = v_135_cast_fp16)[name = tensor("op_30015_cast_fp16")]; + tensor var_30019_begin_0 = const()[name = tensor("op_30019_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_30019_end_0 = const()[name = tensor("op_30019_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_30019_end_mask_0 = const()[name = tensor("op_30019_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30019_cast_fp16 = slice_by_index(begin = var_30019_begin_0, end = var_30019_end_0, end_mask = var_30019_end_mask_0, x = v_135_cast_fp16)[name = tensor("op_30019_cast_fp16")]; + tensor var_30023_begin_0 = const()[name = tensor("op_30023_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_30023_end_0 = const()[name = tensor("op_30023_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_30023_end_mask_0 = const()[name = tensor("op_30023_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30023_cast_fp16 = slice_by_index(begin = var_30023_begin_0, end = var_30023_end_0, end_mask = var_30023_end_mask_0, x = v_135_cast_fp16)[name = tensor("op_30023_cast_fp16")]; + tensor var_30027_begin_0 = const()[name = tensor("op_30027_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_30027_end_0 = const()[name = tensor("op_30027_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_30027_end_mask_0 = const()[name = tensor("op_30027_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30027_cast_fp16 = slice_by_index(begin = var_30027_begin_0, end = var_30027_end_0, end_mask = var_30027_end_mask_0, x = v_135_cast_fp16)[name = tensor("op_30027_cast_fp16")]; + tensor var_30031_begin_0 = const()[name = tensor("op_30031_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_30031_end_0 = const()[name = tensor("op_30031_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_30031_end_mask_0 = const()[name = tensor("op_30031_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30031_cast_fp16 = slice_by_index(begin = var_30031_begin_0, end = var_30031_end_0, end_mask = var_30031_end_mask_0, x = v_135_cast_fp16)[name = tensor("op_30031_cast_fp16")]; + tensor var_30035_begin_0 = const()[name = tensor("op_30035_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_30035_end_0 = const()[name = tensor("op_30035_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_30035_end_mask_0 = const()[name = tensor("op_30035_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30035_cast_fp16 = slice_by_index(begin = var_30035_begin_0, end = var_30035_end_0, end_mask = var_30035_end_mask_0, x = v_135_cast_fp16)[name = tensor("op_30035_cast_fp16")]; + tensor var_30039_begin_0 = const()[name = tensor("op_30039_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_30039_end_0 = const()[name = tensor("op_30039_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_30039_end_mask_0 = const()[name = tensor("op_30039_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30039_cast_fp16 = slice_by_index(begin = var_30039_begin_0, end = var_30039_end_0, end_mask = var_30039_end_mask_0, x = v_135_cast_fp16)[name = tensor("op_30039_cast_fp16")]; + tensor var_30043_begin_0 = const()[name = tensor("op_30043_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_30043_end_0 = const()[name = tensor("op_30043_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_30043_end_mask_0 = const()[name = tensor("op_30043_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30043_cast_fp16 = slice_by_index(begin = var_30043_begin_0, end = var_30043_end_0, end_mask = var_30043_end_mask_0, x = v_135_cast_fp16)[name = tensor("op_30043_cast_fp16")]; + tensor var_30047_begin_0 = const()[name = tensor("op_30047_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_30047_end_0 = const()[name = tensor("op_30047_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_30047_end_mask_0 = const()[name = tensor("op_30047_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30047_cast_fp16 = slice_by_index(begin = var_30047_begin_0, end = var_30047_end_0, end_mask = var_30047_end_mask_0, x = v_135_cast_fp16)[name = tensor("op_30047_cast_fp16")]; + tensor var_30051_begin_0 = const()[name = tensor("op_30051_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_30051_end_0 = const()[name = tensor("op_30051_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_30051_end_mask_0 = const()[name = tensor("op_30051_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30051_cast_fp16 = slice_by_index(begin = var_30051_begin_0, end = var_30051_end_0, end_mask = var_30051_end_mask_0, x = v_135_cast_fp16)[name = tensor("op_30051_cast_fp16")]; + tensor var_30055_begin_0 = const()[name = tensor("op_30055_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_30055_end_0 = const()[name = tensor("op_30055_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_30055_end_mask_0 = const()[name = tensor("op_30055_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30055_cast_fp16 = slice_by_index(begin = var_30055_begin_0, end = var_30055_end_0, end_mask = var_30055_end_mask_0, x = v_135_cast_fp16)[name = tensor("op_30055_cast_fp16")]; + tensor var_30059_begin_0 = const()[name = tensor("op_30059_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_30059_end_0 = const()[name = tensor("op_30059_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_30059_end_mask_0 = const()[name = tensor("op_30059_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30059_cast_fp16 = slice_by_index(begin = var_30059_begin_0, end = var_30059_end_0, end_mask = var_30059_end_mask_0, x = v_135_cast_fp16)[name = tensor("op_30059_cast_fp16")]; + tensor var_30063_begin_0 = const()[name = tensor("op_30063_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_30063_end_0 = const()[name = tensor("op_30063_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_30063_end_mask_0 = const()[name = tensor("op_30063_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30063_cast_fp16 = slice_by_index(begin = var_30063_begin_0, end = var_30063_end_0, end_mask = var_30063_end_mask_0, x = v_135_cast_fp16)[name = tensor("op_30063_cast_fp16")]; + tensor var_30067_begin_0 = const()[name = tensor("op_30067_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_30067_end_0 = const()[name = tensor("op_30067_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_30067_end_mask_0 = const()[name = tensor("op_30067_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30067_cast_fp16 = slice_by_index(begin = var_30067_begin_0, end = var_30067_end_0, end_mask = var_30067_end_mask_0, x = v_135_cast_fp16)[name = tensor("op_30067_cast_fp16")]; + tensor var_30071_begin_0 = const()[name = tensor("op_30071_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_30071_end_0 = const()[name = tensor("op_30071_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_30071_end_mask_0 = const()[name = tensor("op_30071_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30071_cast_fp16 = slice_by_index(begin = var_30071_begin_0, end = var_30071_end_0, end_mask = var_30071_end_mask_0, x = v_135_cast_fp16)[name = tensor("op_30071_cast_fp16")]; + tensor var_30075_begin_0 = const()[name = tensor("op_30075_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_30075_end_0 = const()[name = tensor("op_30075_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_30075_end_mask_0 = const()[name = tensor("op_30075_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30075_cast_fp16 = slice_by_index(begin = var_30075_begin_0, end = var_30075_end_0, end_mask = var_30075_end_mask_0, x = v_135_cast_fp16)[name = tensor("op_30075_cast_fp16")]; + tensor var_30079_begin_0 = const()[name = tensor("op_30079_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_30079_end_0 = const()[name = tensor("op_30079_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_30079_end_mask_0 = const()[name = tensor("op_30079_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30079_cast_fp16 = slice_by_index(begin = var_30079_begin_0, end = var_30079_end_0, end_mask = var_30079_end_mask_0, x = v_135_cast_fp16)[name = tensor("op_30079_cast_fp16")]; + tensor var_30083_equation_0 = const()[name = tensor("op_30083_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30083_cast_fp16 = einsum(equation = var_30083_equation_0, values = (var_29925_cast_fp16, var_29842_cast_fp16))[name = tensor("op_30083_cast_fp16")]; + tensor var_30084_to_fp16 = const()[name = tensor("op_30084_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2521_cast_fp16 = mul(x = var_30083_cast_fp16, y = var_30084_to_fp16)[name = tensor("aw_2521_cast_fp16")]; + tensor var_30087_equation_0 = const()[name = tensor("op_30087_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30087_cast_fp16 = einsum(equation = var_30087_equation_0, values = (var_29929_cast_fp16, var_29846_cast_fp16))[name = tensor("op_30087_cast_fp16")]; + tensor var_30088_to_fp16 = const()[name = tensor("op_30088_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2523_cast_fp16 = mul(x = var_30087_cast_fp16, y = var_30088_to_fp16)[name = tensor("aw_2523_cast_fp16")]; + tensor var_30091_equation_0 = const()[name = tensor("op_30091_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30091_cast_fp16 = einsum(equation = var_30091_equation_0, values = (var_29933_cast_fp16, var_29850_cast_fp16))[name = tensor("op_30091_cast_fp16")]; + tensor var_30092_to_fp16 = const()[name = tensor("op_30092_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2525_cast_fp16 = mul(x = var_30091_cast_fp16, y = var_30092_to_fp16)[name = tensor("aw_2525_cast_fp16")]; + tensor var_30095_equation_0 = const()[name = tensor("op_30095_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30095_cast_fp16 = einsum(equation = var_30095_equation_0, values = (var_29937_cast_fp16, var_29854_cast_fp16))[name = tensor("op_30095_cast_fp16")]; + tensor var_30096_to_fp16 = const()[name = tensor("op_30096_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2527_cast_fp16 = mul(x = var_30095_cast_fp16, y = var_30096_to_fp16)[name = tensor("aw_2527_cast_fp16")]; + tensor var_30099_equation_0 = const()[name = tensor("op_30099_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30099_cast_fp16 = einsum(equation = var_30099_equation_0, values = (var_29941_cast_fp16, var_29858_cast_fp16))[name = tensor("op_30099_cast_fp16")]; + tensor var_30100_to_fp16 = const()[name = tensor("op_30100_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2529_cast_fp16 = mul(x = var_30099_cast_fp16, y = var_30100_to_fp16)[name = tensor("aw_2529_cast_fp16")]; + tensor var_30103_equation_0 = const()[name = tensor("op_30103_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30103_cast_fp16 = einsum(equation = var_30103_equation_0, values = (var_29945_cast_fp16, var_29862_cast_fp16))[name = tensor("op_30103_cast_fp16")]; + tensor var_30104_to_fp16 = const()[name = tensor("op_30104_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2531_cast_fp16 = mul(x = var_30103_cast_fp16, y = var_30104_to_fp16)[name = tensor("aw_2531_cast_fp16")]; + tensor var_30107_equation_0 = const()[name = tensor("op_30107_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30107_cast_fp16 = einsum(equation = var_30107_equation_0, values = (var_29949_cast_fp16, var_29866_cast_fp16))[name = tensor("op_30107_cast_fp16")]; + tensor var_30108_to_fp16 = const()[name = tensor("op_30108_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2533_cast_fp16 = mul(x = var_30107_cast_fp16, y = var_30108_to_fp16)[name = tensor("aw_2533_cast_fp16")]; + tensor var_30111_equation_0 = const()[name = tensor("op_30111_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30111_cast_fp16 = einsum(equation = var_30111_equation_0, values = (var_29953_cast_fp16, var_29870_cast_fp16))[name = tensor("op_30111_cast_fp16")]; + tensor var_30112_to_fp16 = const()[name = tensor("op_30112_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2535_cast_fp16 = mul(x = var_30111_cast_fp16, y = var_30112_to_fp16)[name = tensor("aw_2535_cast_fp16")]; + tensor var_30115_equation_0 = const()[name = tensor("op_30115_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30115_cast_fp16 = einsum(equation = var_30115_equation_0, values = (var_29957_cast_fp16, var_29874_cast_fp16))[name = tensor("op_30115_cast_fp16")]; + tensor var_30116_to_fp16 = const()[name = tensor("op_30116_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2537_cast_fp16 = mul(x = var_30115_cast_fp16, y = var_30116_to_fp16)[name = tensor("aw_2537_cast_fp16")]; + tensor var_30119_equation_0 = const()[name = tensor("op_30119_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30119_cast_fp16 = einsum(equation = var_30119_equation_0, values = (var_29961_cast_fp16, var_29878_cast_fp16))[name = tensor("op_30119_cast_fp16")]; + tensor var_30120_to_fp16 = const()[name = tensor("op_30120_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2539_cast_fp16 = mul(x = var_30119_cast_fp16, y = var_30120_to_fp16)[name = tensor("aw_2539_cast_fp16")]; + tensor var_30123_equation_0 = const()[name = tensor("op_30123_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30123_cast_fp16 = einsum(equation = var_30123_equation_0, values = (var_29965_cast_fp16, var_29882_cast_fp16))[name = tensor("op_30123_cast_fp16")]; + tensor var_30124_to_fp16 = const()[name = tensor("op_30124_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2541_cast_fp16 = mul(x = var_30123_cast_fp16, y = var_30124_to_fp16)[name = tensor("aw_2541_cast_fp16")]; + tensor var_30127_equation_0 = const()[name = tensor("op_30127_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30127_cast_fp16 = einsum(equation = var_30127_equation_0, values = (var_29969_cast_fp16, var_29886_cast_fp16))[name = tensor("op_30127_cast_fp16")]; + tensor var_30128_to_fp16 = const()[name = tensor("op_30128_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2543_cast_fp16 = mul(x = var_30127_cast_fp16, y = var_30128_to_fp16)[name = tensor("aw_2543_cast_fp16")]; + tensor var_30131_equation_0 = const()[name = tensor("op_30131_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30131_cast_fp16 = einsum(equation = var_30131_equation_0, values = (var_29973_cast_fp16, var_29890_cast_fp16))[name = tensor("op_30131_cast_fp16")]; + tensor var_30132_to_fp16 = const()[name = tensor("op_30132_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2545_cast_fp16 = mul(x = var_30131_cast_fp16, y = var_30132_to_fp16)[name = tensor("aw_2545_cast_fp16")]; + tensor var_30135_equation_0 = const()[name = tensor("op_30135_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30135_cast_fp16 = einsum(equation = var_30135_equation_0, values = (var_29977_cast_fp16, var_29894_cast_fp16))[name = tensor("op_30135_cast_fp16")]; + tensor var_30136_to_fp16 = const()[name = tensor("op_30136_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2547_cast_fp16 = mul(x = var_30135_cast_fp16, y = var_30136_to_fp16)[name = tensor("aw_2547_cast_fp16")]; + tensor var_30139_equation_0 = const()[name = tensor("op_30139_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30139_cast_fp16 = einsum(equation = var_30139_equation_0, values = (var_29981_cast_fp16, var_29898_cast_fp16))[name = tensor("op_30139_cast_fp16")]; + tensor var_30140_to_fp16 = const()[name = tensor("op_30140_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2549_cast_fp16 = mul(x = var_30139_cast_fp16, y = var_30140_to_fp16)[name = tensor("aw_2549_cast_fp16")]; + tensor var_30143_equation_0 = const()[name = tensor("op_30143_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30143_cast_fp16 = einsum(equation = var_30143_equation_0, values = (var_29985_cast_fp16, var_29902_cast_fp16))[name = tensor("op_30143_cast_fp16")]; + tensor var_30144_to_fp16 = const()[name = tensor("op_30144_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2551_cast_fp16 = mul(x = var_30143_cast_fp16, y = var_30144_to_fp16)[name = tensor("aw_2551_cast_fp16")]; + tensor var_30147_equation_0 = const()[name = tensor("op_30147_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30147_cast_fp16 = einsum(equation = var_30147_equation_0, values = (var_29989_cast_fp16, var_29906_cast_fp16))[name = tensor("op_30147_cast_fp16")]; + tensor var_30148_to_fp16 = const()[name = tensor("op_30148_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2553_cast_fp16 = mul(x = var_30147_cast_fp16, y = var_30148_to_fp16)[name = tensor("aw_2553_cast_fp16")]; + tensor var_30151_equation_0 = const()[name = tensor("op_30151_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30151_cast_fp16 = einsum(equation = var_30151_equation_0, values = (var_29993_cast_fp16, var_29910_cast_fp16))[name = tensor("op_30151_cast_fp16")]; + tensor var_30152_to_fp16 = const()[name = tensor("op_30152_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2555_cast_fp16 = mul(x = var_30151_cast_fp16, y = var_30152_to_fp16)[name = tensor("aw_2555_cast_fp16")]; + tensor var_30155_equation_0 = const()[name = tensor("op_30155_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30155_cast_fp16 = einsum(equation = var_30155_equation_0, values = (var_29997_cast_fp16, var_29914_cast_fp16))[name = tensor("op_30155_cast_fp16")]; + tensor var_30156_to_fp16 = const()[name = tensor("op_30156_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2557_cast_fp16 = mul(x = var_30155_cast_fp16, y = var_30156_to_fp16)[name = tensor("aw_2557_cast_fp16")]; + tensor var_30159_equation_0 = const()[name = tensor("op_30159_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30159_cast_fp16 = einsum(equation = var_30159_equation_0, values = (var_30001_cast_fp16, var_29918_cast_fp16))[name = tensor("op_30159_cast_fp16")]; + tensor var_30160_to_fp16 = const()[name = tensor("op_30160_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2559_cast_fp16 = mul(x = var_30159_cast_fp16, y = var_30160_to_fp16)[name = tensor("aw_2559_cast_fp16")]; + tensor var_30162_cast_fp16 = softmax(axis = var_21077, x = aw_2521_cast_fp16)[name = tensor("op_30162_cast_fp16")]; + tensor var_30163_cast_fp16 = softmax(axis = var_21077, x = aw_2523_cast_fp16)[name = tensor("op_30163_cast_fp16")]; + tensor var_30164_cast_fp16 = softmax(axis = var_21077, x = aw_2525_cast_fp16)[name = tensor("op_30164_cast_fp16")]; + tensor var_30165_cast_fp16 = softmax(axis = var_21077, x = aw_2527_cast_fp16)[name = tensor("op_30165_cast_fp16")]; + tensor var_30166_cast_fp16 = softmax(axis = var_21077, x = aw_2529_cast_fp16)[name = tensor("op_30166_cast_fp16")]; + tensor var_30167_cast_fp16 = softmax(axis = var_21077, x = aw_2531_cast_fp16)[name = tensor("op_30167_cast_fp16")]; + tensor var_30168_cast_fp16 = softmax(axis = var_21077, x = aw_2533_cast_fp16)[name = tensor("op_30168_cast_fp16")]; + tensor var_30169_cast_fp16 = softmax(axis = var_21077, x = aw_2535_cast_fp16)[name = tensor("op_30169_cast_fp16")]; + tensor var_30170_cast_fp16 = softmax(axis = var_21077, x = aw_2537_cast_fp16)[name = tensor("op_30170_cast_fp16")]; + tensor var_30171_cast_fp16 = softmax(axis = var_21077, x = aw_2539_cast_fp16)[name = tensor("op_30171_cast_fp16")]; + tensor var_30172_cast_fp16 = softmax(axis = var_21077, x = aw_2541_cast_fp16)[name = tensor("op_30172_cast_fp16")]; + tensor var_30173_cast_fp16 = softmax(axis = var_21077, x = aw_2543_cast_fp16)[name = tensor("op_30173_cast_fp16")]; + tensor var_30174_cast_fp16 = softmax(axis = var_21077, x = aw_2545_cast_fp16)[name = tensor("op_30174_cast_fp16")]; + tensor var_30175_cast_fp16 = softmax(axis = var_21077, x = aw_2547_cast_fp16)[name = tensor("op_30175_cast_fp16")]; + tensor var_30176_cast_fp16 = softmax(axis = var_21077, x = aw_2549_cast_fp16)[name = tensor("op_30176_cast_fp16")]; + tensor var_30177_cast_fp16 = softmax(axis = var_21077, x = aw_2551_cast_fp16)[name = tensor("op_30177_cast_fp16")]; + tensor var_30178_cast_fp16 = softmax(axis = var_21077, x = aw_2553_cast_fp16)[name = tensor("op_30178_cast_fp16")]; + tensor var_30179_cast_fp16 = softmax(axis = var_21077, x = aw_2555_cast_fp16)[name = tensor("op_30179_cast_fp16")]; + tensor var_30180_cast_fp16 = softmax(axis = var_21077, x = aw_2557_cast_fp16)[name = tensor("op_30180_cast_fp16")]; + tensor var_30181_cast_fp16 = softmax(axis = var_21077, x = aw_2559_cast_fp16)[name = tensor("op_30181_cast_fp16")]; + tensor var_30183_equation_0 = const()[name = tensor("op_30183_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30183_cast_fp16 = einsum(equation = var_30183_equation_0, values = (var_30003_cast_fp16, var_30162_cast_fp16))[name = tensor("op_30183_cast_fp16")]; + tensor var_30185_equation_0 = const()[name = tensor("op_30185_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30185_cast_fp16 = einsum(equation = var_30185_equation_0, values = (var_30007_cast_fp16, var_30163_cast_fp16))[name = tensor("op_30185_cast_fp16")]; + tensor var_30187_equation_0 = const()[name = tensor("op_30187_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30187_cast_fp16 = einsum(equation = var_30187_equation_0, values = (var_30011_cast_fp16, var_30164_cast_fp16))[name = tensor("op_30187_cast_fp16")]; + tensor var_30189_equation_0 = const()[name = tensor("op_30189_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30189_cast_fp16 = einsum(equation = var_30189_equation_0, values = (var_30015_cast_fp16, var_30165_cast_fp16))[name = tensor("op_30189_cast_fp16")]; + tensor var_30191_equation_0 = const()[name = tensor("op_30191_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30191_cast_fp16 = einsum(equation = var_30191_equation_0, values = (var_30019_cast_fp16, var_30166_cast_fp16))[name = tensor("op_30191_cast_fp16")]; + tensor var_30193_equation_0 = const()[name = tensor("op_30193_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30193_cast_fp16 = einsum(equation = var_30193_equation_0, values = (var_30023_cast_fp16, var_30167_cast_fp16))[name = tensor("op_30193_cast_fp16")]; + tensor var_30195_equation_0 = const()[name = tensor("op_30195_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30195_cast_fp16 = einsum(equation = var_30195_equation_0, values = (var_30027_cast_fp16, var_30168_cast_fp16))[name = tensor("op_30195_cast_fp16")]; + tensor var_30197_equation_0 = const()[name = tensor("op_30197_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30197_cast_fp16 = einsum(equation = var_30197_equation_0, values = (var_30031_cast_fp16, var_30169_cast_fp16))[name = tensor("op_30197_cast_fp16")]; + tensor var_30199_equation_0 = const()[name = tensor("op_30199_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30199_cast_fp16 = einsum(equation = var_30199_equation_0, values = (var_30035_cast_fp16, var_30170_cast_fp16))[name = tensor("op_30199_cast_fp16")]; + tensor var_30201_equation_0 = const()[name = tensor("op_30201_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30201_cast_fp16 = einsum(equation = var_30201_equation_0, values = (var_30039_cast_fp16, var_30171_cast_fp16))[name = tensor("op_30201_cast_fp16")]; + tensor var_30203_equation_0 = const()[name = tensor("op_30203_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30203_cast_fp16 = einsum(equation = var_30203_equation_0, values = (var_30043_cast_fp16, var_30172_cast_fp16))[name = tensor("op_30203_cast_fp16")]; + tensor var_30205_equation_0 = const()[name = tensor("op_30205_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30205_cast_fp16 = einsum(equation = var_30205_equation_0, values = (var_30047_cast_fp16, var_30173_cast_fp16))[name = tensor("op_30205_cast_fp16")]; + tensor var_30207_equation_0 = const()[name = tensor("op_30207_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30207_cast_fp16 = einsum(equation = var_30207_equation_0, values = (var_30051_cast_fp16, var_30174_cast_fp16))[name = tensor("op_30207_cast_fp16")]; + tensor var_30209_equation_0 = const()[name = tensor("op_30209_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30209_cast_fp16 = einsum(equation = var_30209_equation_0, values = (var_30055_cast_fp16, var_30175_cast_fp16))[name = tensor("op_30209_cast_fp16")]; + tensor var_30211_equation_0 = const()[name = tensor("op_30211_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30211_cast_fp16 = einsum(equation = var_30211_equation_0, values = (var_30059_cast_fp16, var_30176_cast_fp16))[name = tensor("op_30211_cast_fp16")]; + tensor var_30213_equation_0 = const()[name = tensor("op_30213_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30213_cast_fp16 = einsum(equation = var_30213_equation_0, values = (var_30063_cast_fp16, var_30177_cast_fp16))[name = tensor("op_30213_cast_fp16")]; + tensor var_30215_equation_0 = const()[name = tensor("op_30215_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30215_cast_fp16 = einsum(equation = var_30215_equation_0, values = (var_30067_cast_fp16, var_30178_cast_fp16))[name = tensor("op_30215_cast_fp16")]; + tensor var_30217_equation_0 = const()[name = tensor("op_30217_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30217_cast_fp16 = einsum(equation = var_30217_equation_0, values = (var_30071_cast_fp16, var_30179_cast_fp16))[name = tensor("op_30217_cast_fp16")]; + tensor var_30219_equation_0 = const()[name = tensor("op_30219_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30219_cast_fp16 = einsum(equation = var_30219_equation_0, values = (var_30075_cast_fp16, var_30180_cast_fp16))[name = tensor("op_30219_cast_fp16")]; + tensor var_30221_equation_0 = const()[name = tensor("op_30221_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30221_cast_fp16 = einsum(equation = var_30221_equation_0, values = (var_30079_cast_fp16, var_30181_cast_fp16))[name = tensor("op_30221_cast_fp16")]; + tensor input_401_interleave_0 = const()[name = tensor("input_401_interleave_0"), val = tensor(false)]; + tensor input_401_cast_fp16 = concat(axis = var_21077, interleave = input_401_interleave_0, values = (var_30183_cast_fp16, var_30185_cast_fp16, var_30187_cast_fp16, var_30189_cast_fp16, var_30191_cast_fp16, var_30193_cast_fp16, var_30195_cast_fp16, var_30197_cast_fp16, var_30199_cast_fp16, var_30201_cast_fp16, var_30203_cast_fp16, var_30205_cast_fp16, var_30207_cast_fp16, var_30209_cast_fp16, var_30211_cast_fp16, var_30213_cast_fp16, var_30215_cast_fp16, var_30217_cast_fp16, var_30219_cast_fp16, var_30221_cast_fp16))[name = tensor("input_401_cast_fp16")]; + tensor var_30227 = const()[name = tensor("op_30227"), val = tensor([1, 1])]; + tensor var_30229 = const()[name = tensor("op_30229"), val = tensor([1, 1])]; + tensor var_30231_pad_type_0 = const()[name = tensor("op_30231_pad_type_0"), val = tensor("custom")]; + tensor var_30231_pad_0 = const()[name = tensor("op_30231_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_9_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_9_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2378609024)))]; + tensor mid_block_attentions_0_transformer_blocks_9_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_9_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2381885888)))]; + tensor var_30231_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_9_attn2_to_out_0_bias_to_fp16, dilations = var_30229, groups = var_21077, pad = var_30231_pad_0, pad_type = var_30231_pad_type_0, strides = var_30227, weight = mid_block_attentions_0_transformer_blocks_9_attn2_to_out_0_weight_to_fp16, x = input_401_cast_fp16)[name = tensor("op_30231_cast_fp16")]; + tensor inputs_203_cast_fp16 = add(x = var_30231_cast_fp16, y = inputs_201_cast_fp16)[name = tensor("inputs_203_cast_fp16")]; + tensor input_403_axes_0 = const()[name = tensor("input_403_axes_0"), val = tensor([1])]; + tensor input_403_gamma_0_to_fp16 = const()[name = tensor("input_403_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2381888512)))]; + tensor input_403_beta_0_to_fp16 = const()[name = tensor("input_403_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2381891136)))]; + tensor var_30241_to_fp16 = const()[name = tensor("op_30241_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_403_cast_fp16 = layer_norm(axes = input_403_axes_0, beta = input_403_beta_0_to_fp16, epsilon = var_30241_to_fp16, gamma = input_403_gamma_0_to_fp16, x = inputs_203_cast_fp16)[name = tensor("input_403_cast_fp16")]; + tensor var_30257 = const()[name = tensor("op_30257"), val = tensor([1, 1])]; + tensor var_30259 = const()[name = tensor("op_30259"), val = tensor([1, 1])]; + tensor var_30261_pad_type_0 = const()[name = tensor("op_30261_pad_type_0"), val = tensor("custom")]; + tensor var_30261_pad_0 = const()[name = tensor("op_30261_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_9_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_9_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2381893760)))]; + tensor mid_block_attentions_0_transformer_blocks_9_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_9_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2408108224)))]; + tensor var_30261_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_9_ff_net_0_proj_bias_to_fp16, dilations = var_30259, groups = var_21077, pad = var_30261_pad_0, pad_type = var_30261_pad_type_0, strides = var_30257, weight = mid_block_attentions_0_transformer_blocks_9_ff_net_0_proj_weight_to_fp16, x = input_403_cast_fp16)[name = tensor("op_30261_cast_fp16")]; + tensor var_30262_split_sizes_0 = const()[name = tensor("op_30262_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_30262_axis_0 = const()[name = tensor("op_30262_axis_0"), val = tensor(1)]; + tensor var_30262_cast_fp16_0, tensor var_30262_cast_fp16_1 = split(axis = var_30262_axis_0, split_sizes = var_30262_split_sizes_0, x = var_30261_cast_fp16)[name = tensor("op_30262_cast_fp16")]; + tensor var_30264_mode_0 = const()[name = tensor("op_30264_mode_0"), val = tensor("EXACT")]; + tensor var_30264_cast_fp16 = gelu(mode = var_30264_mode_0, x = var_30262_cast_fp16_1)[name = tensor("op_30264_cast_fp16")]; + tensor input_405_cast_fp16 = mul(x = var_30262_cast_fp16_0, y = var_30264_cast_fp16)[name = tensor("input_405_cast_fp16")]; + tensor var_30268 = const()[name = tensor("op_30268"), val = tensor([1, 1])]; + tensor var_30270 = const()[name = tensor("op_30270"), val = tensor([1, 1])]; + tensor var_30272_pad_type_0 = const()[name = tensor("op_30272_pad_type_0"), val = tensor("custom")]; + tensor var_30272_pad_0 = const()[name = tensor("op_30272_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_9_ff_net_2_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_9_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2408128768)))]; + tensor mid_block_attentions_0_transformer_blocks_9_ff_net_2_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_9_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2421236032)))]; + tensor var_30272_cast_fp16 = conv(bias = mid_block_attentions_0_transformer_blocks_9_ff_net_2_bias_to_fp16, dilations = var_30270, groups = var_21077, pad = var_30272_pad_0, pad_type = var_30272_pad_type_0, strides = var_30268, weight = mid_block_attentions_0_transformer_blocks_9_ff_net_2_weight_to_fp16, x = input_405_cast_fp16)[name = tensor("op_30272_cast_fp16")]; + tensor hidden_states_269_cast_fp16 = add(x = var_30272_cast_fp16, y = inputs_203_cast_fp16)[name = tensor("hidden_states_269_cast_fp16")]; + tensor var_30274 = const()[name = tensor("op_30274"), val = tensor([2, 1280, 32, 32])]; + tensor input_407_cast_fp16 = reshape(shape = var_30274, x = hidden_states_269_cast_fp16)[name = tensor("input_407_cast_fp16")]; + tensor var_30278 = const()[name = tensor("op_30278"), val = tensor([1, 1])]; + tensor var_30280 = const()[name = tensor("op_30280"), val = tensor([1, 1])]; + tensor hidden_states_271_pad_type_0 = const()[name = tensor("hidden_states_271_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_271_pad_0 = const()[name = tensor("hidden_states_271_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_proj_out_weight_to_fp16 = const()[name = tensor("mid_block_attentions_0_proj_out_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2421238656)))]; + tensor mid_block_attentions_0_proj_out_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_proj_out_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2424515520)))]; + tensor hidden_states_271_cast_fp16 = conv(bias = mid_block_attentions_0_proj_out_bias_to_fp16, dilations = var_30280, groups = var_21077, pad = hidden_states_271_pad_0, pad_type = hidden_states_271_pad_type_0, strides = var_30278, weight = mid_block_attentions_0_proj_out_weight_to_fp16, x = input_407_cast_fp16)[name = tensor("hidden_states_271_cast_fp16")]; + tensor input_409_cast_fp16 = add(x = hidden_states_271_cast_fp16, y = hidden_states_205_cast_fp16)[name = tensor("input_409_cast_fp16")]; + tensor reshape_76_shape_0 = const()[name = tensor("reshape_76_shape_0"), val = tensor([2, 32, 40, 32, 32])]; + tensor reshape_76_cast_fp16 = reshape(shape = reshape_76_shape_0, x = input_409_cast_fp16)[name = tensor("reshape_76_cast_fp16")]; + tensor reduce_mean_57_axes_0 = const()[name = tensor("reduce_mean_57_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_57_keep_dims_0 = const()[name = tensor("reduce_mean_57_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_57_cast_fp16 = reduce_mean(axes = reduce_mean_57_axes_0, keep_dims = reduce_mean_57_keep_dims_0, x = reshape_76_cast_fp16)[name = tensor("reduce_mean_57_cast_fp16")]; + tensor sub_38_cast_fp16 = sub(x = reshape_76_cast_fp16, y = reduce_mean_57_cast_fp16)[name = tensor("sub_38_cast_fp16")]; + tensor square_19_cast_fp16 = square(x = sub_38_cast_fp16)[name = tensor("square_19_cast_fp16")]; + tensor reduce_mean_59_axes_0 = const()[name = tensor("reduce_mean_59_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_59_keep_dims_0 = const()[name = tensor("reduce_mean_59_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_59_cast_fp16 = reduce_mean(axes = reduce_mean_59_axes_0, keep_dims = reduce_mean_59_keep_dims_0, x = square_19_cast_fp16)[name = tensor("reduce_mean_59_cast_fp16")]; + tensor add_38_y_0_to_fp16 = const()[name = tensor("add_38_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_38_cast_fp16 = add(x = reduce_mean_59_cast_fp16, y = add_38_y_0_to_fp16)[name = tensor("add_38_cast_fp16")]; + tensor sqrt_19_cast_fp16 = sqrt(x = add_38_cast_fp16)[name = tensor("sqrt_19_cast_fp16")]; + tensor real_div_19_cast_fp16 = real_div(x = sub_38_cast_fp16, y = sqrt_19_cast_fp16)[name = tensor("real_div_19_cast_fp16")]; + tensor reshape_77_shape_0 = const()[name = tensor("reshape_77_shape_0"), val = tensor([2, 1280, 32, 32])]; + tensor reshape_77_cast_fp16 = reshape(shape = reshape_77_shape_0, x = real_div_19_cast_fp16)[name = tensor("reshape_77_cast_fp16")]; + tensor add_39_gamma_0_to_fp16 = const()[name = tensor("add_39_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2424518144)))]; + tensor add_39_beta_0_to_fp16 = const()[name = tensor("add_39_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2424520768)))]; + tensor add_39_epsilon_0_to_fp16 = const()[name = tensor("add_39_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_39_cast_fp16 = batch_norm(beta = add_39_beta_0_to_fp16, epsilon = add_39_epsilon_0_to_fp16, gamma = add_39_gamma_0_to_fp16, mean = add_23_mean_0_to_fp16, variance = add_23_variance_0_to_fp16, x = reshape_77_cast_fp16)[name = tensor("add_39_cast_fp16")]; + tensor input_413_cast_fp16 = silu(x = add_39_cast_fp16)[name = tensor("input_413_cast_fp16")]; + tensor var_30295 = const()[name = tensor("op_30295"), val = tensor([1, 1])]; + tensor var_30297 = const()[name = tensor("op_30297"), val = tensor([1, 1])]; + tensor hidden_states_273_pad_type_0 = const()[name = tensor("hidden_states_273_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_273_pad_0 = const()[name = tensor("hidden_states_273_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor mid_block_resnets_1_conv1_weight_to_fp16 = const()[name = tensor("mid_block_resnets_1_conv1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2424523392)))]; + tensor mid_block_resnets_1_conv1_bias_to_fp16 = const()[name = tensor("mid_block_resnets_1_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2454014656)))]; + tensor hidden_states_273_cast_fp16 = conv(bias = mid_block_resnets_1_conv1_bias_to_fp16, dilations = var_30297, groups = var_21077, pad = hidden_states_273_pad_0, pad_type = hidden_states_273_pad_type_0, strides = var_30295, weight = mid_block_resnets_1_conv1_weight_to_fp16, x = input_413_cast_fp16)[name = tensor("hidden_states_273_cast_fp16")]; + tensor var_30303 = const()[name = tensor("op_30303"), val = tensor([1, 1])]; + tensor var_30305 = const()[name = tensor("op_30305"), val = tensor([1, 1])]; + tensor temb_15_pad_type_0 = const()[name = tensor("temb_15_pad_type_0"), val = tensor("custom")]; + tensor temb_15_pad_0 = const()[name = tensor("temb_15_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_resnets_1_time_emb_proj_weight_to_fp16 = const()[name = tensor("mid_block_resnets_1_time_emb_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2454017280)))]; + tensor mid_block_resnets_1_time_emb_proj_bias_to_fp16 = const()[name = tensor("mid_block_resnets_1_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2457294144)))]; + tensor temb_15_cast_fp16 = conv(bias = mid_block_resnets_1_time_emb_proj_bias_to_fp16, dilations = var_30305, groups = var_21077, pad = temb_15_pad_0, pad_type = temb_15_pad_type_0, strides = var_30303, weight = mid_block_resnets_1_time_emb_proj_weight_to_fp16, x = input_21_cast_fp16)[name = tensor("temb_15_cast_fp16")]; + tensor input_417_cast_fp16 = add(x = hidden_states_273_cast_fp16, y = temb_15_cast_fp16)[name = tensor("input_417_cast_fp16")]; + tensor reshape_80_shape_0 = const()[name = tensor("reshape_80_shape_0"), val = tensor([2, 32, 40, 32, 32])]; + tensor reshape_80_cast_fp16 = reshape(shape = reshape_80_shape_0, x = input_417_cast_fp16)[name = tensor("reshape_80_cast_fp16")]; + tensor reduce_mean_60_axes_0 = const()[name = tensor("reduce_mean_60_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_60_keep_dims_0 = const()[name = tensor("reduce_mean_60_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_60_cast_fp16 = reduce_mean(axes = reduce_mean_60_axes_0, keep_dims = reduce_mean_60_keep_dims_0, x = reshape_80_cast_fp16)[name = tensor("reduce_mean_60_cast_fp16")]; + tensor sub_40_cast_fp16 = sub(x = reshape_80_cast_fp16, y = reduce_mean_60_cast_fp16)[name = tensor("sub_40_cast_fp16")]; + tensor square_20_cast_fp16 = square(x = sub_40_cast_fp16)[name = tensor("square_20_cast_fp16")]; + tensor reduce_mean_62_axes_0 = const()[name = tensor("reduce_mean_62_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_62_keep_dims_0 = const()[name = tensor("reduce_mean_62_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_62_cast_fp16 = reduce_mean(axes = reduce_mean_62_axes_0, keep_dims = reduce_mean_62_keep_dims_0, x = square_20_cast_fp16)[name = tensor("reduce_mean_62_cast_fp16")]; + tensor add_40_y_0_to_fp16 = const()[name = tensor("add_40_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_40_cast_fp16 = add(x = reduce_mean_62_cast_fp16, y = add_40_y_0_to_fp16)[name = tensor("add_40_cast_fp16")]; + tensor sqrt_20_cast_fp16 = sqrt(x = add_40_cast_fp16)[name = tensor("sqrt_20_cast_fp16")]; + tensor real_div_20_cast_fp16 = real_div(x = sub_40_cast_fp16, y = sqrt_20_cast_fp16)[name = tensor("real_div_20_cast_fp16")]; + tensor reshape_81_shape_0 = const()[name = tensor("reshape_81_shape_0"), val = tensor([2, 1280, 32, 32])]; + tensor reshape_81_cast_fp16 = reshape(shape = reshape_81_shape_0, x = real_div_20_cast_fp16)[name = tensor("reshape_81_cast_fp16")]; + tensor add_41_gamma_0_to_fp16 = const()[name = tensor("add_41_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2457296768)))]; + tensor add_41_beta_0_to_fp16 = const()[name = tensor("add_41_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2457299392)))]; + tensor add_41_epsilon_0_to_fp16 = const()[name = tensor("add_41_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_41_cast_fp16 = batch_norm(beta = add_41_beta_0_to_fp16, epsilon = add_41_epsilon_0_to_fp16, gamma = add_41_gamma_0_to_fp16, mean = add_23_mean_0_to_fp16, variance = add_23_variance_0_to_fp16, x = reshape_81_cast_fp16)[name = tensor("add_41_cast_fp16")]; + tensor input_421_cast_fp16 = silu(x = add_41_cast_fp16)[name = tensor("input_421_cast_fp16")]; + tensor var_30315 = const()[name = tensor("op_30315"), val = tensor([1, 1])]; + tensor var_30317 = const()[name = tensor("op_30317"), val = tensor([1, 1])]; + tensor hidden_states_275_pad_type_0 = const()[name = tensor("hidden_states_275_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_275_pad_0 = const()[name = tensor("hidden_states_275_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor mid_block_resnets_1_conv2_weight_to_fp16 = const()[name = tensor("mid_block_resnets_1_conv2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2457302016)))]; + tensor mid_block_resnets_1_conv2_bias_to_fp16 = const()[name = tensor("mid_block_resnets_1_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2486793280)))]; + tensor hidden_states_275_cast_fp16 = conv(bias = mid_block_resnets_1_conv2_bias_to_fp16, dilations = var_30317, groups = var_21077, pad = hidden_states_275_pad_0, pad_type = hidden_states_275_pad_type_0, strides = var_30315, weight = mid_block_resnets_1_conv2_weight_to_fp16, x = input_421_cast_fp16)[name = tensor("hidden_states_275_cast_fp16")]; + tensor hidden_states_277_cast_fp16 = add(x = input_409_cast_fp16, y = hidden_states_275_cast_fp16)[name = tensor("hidden_states_277_cast_fp16")]; + tensor var_30355 = const()[name = tensor("op_30355"), val = tensor(1)]; + tensor input_423_interleave_0 = const()[name = tensor("input_423_interleave_0"), val = tensor(false)]; + tensor input_423_cast_fp16 = concat(axis = var_30355, interleave = input_423_interleave_0, values = (hidden_states_277_cast_fp16, input_311_cast_fp16))[name = tensor("input_423_cast_fp16")]; + tensor reshape_84_shape_0 = const()[name = tensor("reshape_84_shape_0"), val = tensor([2, 32, 80, 32, 32])]; + tensor reshape_84_cast_fp16 = reshape(shape = reshape_84_shape_0, x = input_423_cast_fp16)[name = tensor("reshape_84_cast_fp16")]; + tensor reduce_mean_63_axes_0 = const()[name = tensor("reduce_mean_63_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_63_keep_dims_0 = const()[name = tensor("reduce_mean_63_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_63_cast_fp16 = reduce_mean(axes = reduce_mean_63_axes_0, keep_dims = reduce_mean_63_keep_dims_0, x = reshape_84_cast_fp16)[name = tensor("reduce_mean_63_cast_fp16")]; + tensor sub_42_cast_fp16 = sub(x = reshape_84_cast_fp16, y = reduce_mean_63_cast_fp16)[name = tensor("sub_42_cast_fp16")]; + tensor square_21_cast_fp16 = square(x = sub_42_cast_fp16)[name = tensor("square_21_cast_fp16")]; + tensor reduce_mean_65_axes_0 = const()[name = tensor("reduce_mean_65_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_65_keep_dims_0 = const()[name = tensor("reduce_mean_65_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_65_cast_fp16 = reduce_mean(axes = reduce_mean_65_axes_0, keep_dims = reduce_mean_65_keep_dims_0, x = square_21_cast_fp16)[name = tensor("reduce_mean_65_cast_fp16")]; + tensor add_42_y_0_to_fp16 = const()[name = tensor("add_42_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_42_cast_fp16 = add(x = reduce_mean_65_cast_fp16, y = add_42_y_0_to_fp16)[name = tensor("add_42_cast_fp16")]; + tensor sqrt_21_cast_fp16 = sqrt(x = add_42_cast_fp16)[name = tensor("sqrt_21_cast_fp16")]; + tensor real_div_21_cast_fp16 = real_div(x = sub_42_cast_fp16, y = sqrt_21_cast_fp16)[name = tensor("real_div_21_cast_fp16")]; + tensor reshape_85_shape_0 = const()[name = tensor("reshape_85_shape_0"), val = tensor([2, 2560, 32, 32])]; + tensor reshape_85_cast_fp16 = reshape(shape = reshape_85_shape_0, x = real_div_21_cast_fp16)[name = tensor("reshape_85_cast_fp16")]; + tensor add_43_mean_0_to_fp16 = const()[name = tensor("add_43_mean_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2486795904)))]; + tensor add_43_variance_0_to_fp16 = const()[name = tensor("add_43_variance_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2486801088)))]; + tensor add_43_gamma_0_to_fp16 = const()[name = tensor("add_43_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2486806272)))]; + tensor add_43_beta_0_to_fp16 = const()[name = tensor("add_43_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2486811456)))]; + tensor add_43_epsilon_0_to_fp16 = const()[name = tensor("add_43_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_43_cast_fp16 = batch_norm(beta = add_43_beta_0_to_fp16, epsilon = add_43_epsilon_0_to_fp16, gamma = add_43_gamma_0_to_fp16, mean = add_43_mean_0_to_fp16, variance = add_43_variance_0_to_fp16, x = reshape_85_cast_fp16)[name = tensor("add_43_cast_fp16")]; + tensor input_427_cast_fp16 = silu(x = add_43_cast_fp16)[name = tensor("input_427_cast_fp16")]; + tensor var_30384 = const()[name = tensor("op_30384"), val = tensor([1, 1])]; + tensor var_30386 = const()[name = tensor("op_30386"), val = tensor([1, 1])]; + tensor hidden_states_279_pad_type_0 = const()[name = tensor("hidden_states_279_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_279_pad_0 = const()[name = tensor("hidden_states_279_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor up_blocks_0_resnets_0_conv1_weight_to_fp16 = const()[name = tensor("up_blocks_0_resnets_0_conv1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2486816640)))]; + tensor up_blocks_0_resnets_0_conv1_bias_to_fp16 = const()[name = tensor("up_blocks_0_resnets_0_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2545799104)))]; + tensor hidden_states_279_cast_fp16 = conv(bias = up_blocks_0_resnets_0_conv1_bias_to_fp16, dilations = var_30386, groups = var_30355, pad = hidden_states_279_pad_0, pad_type = hidden_states_279_pad_type_0, strides = var_30384, weight = up_blocks_0_resnets_0_conv1_weight_to_fp16, x = input_427_cast_fp16)[name = tensor("hidden_states_279_cast_fp16")]; + tensor var_30392 = const()[name = tensor("op_30392"), val = tensor([1, 1])]; + tensor var_30394 = const()[name = tensor("op_30394"), val = tensor([1, 1])]; + tensor temb_17_pad_type_0 = const()[name = tensor("temb_17_pad_type_0"), val = tensor("custom")]; + tensor temb_17_pad_0 = const()[name = tensor("temb_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_resnets_0_time_emb_proj_weight_to_fp16 = const()[name = tensor("up_blocks_0_resnets_0_time_emb_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2545801728)))]; + tensor up_blocks_0_resnets_0_time_emb_proj_bias_to_fp16 = const()[name = tensor("up_blocks_0_resnets_0_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2549078592)))]; + tensor temb_17_cast_fp16 = conv(bias = up_blocks_0_resnets_0_time_emb_proj_bias_to_fp16, dilations = var_30394, groups = var_30355, pad = temb_17_pad_0, pad_type = temb_17_pad_type_0, strides = var_30392, weight = up_blocks_0_resnets_0_time_emb_proj_weight_to_fp16, x = input_21_cast_fp16)[name = tensor("temb_17_cast_fp16")]; + tensor input_431_cast_fp16 = add(x = hidden_states_279_cast_fp16, y = temb_17_cast_fp16)[name = tensor("input_431_cast_fp16")]; + tensor reshape_88_shape_0 = const()[name = tensor("reshape_88_shape_0"), val = tensor([2, 32, 40, 32, 32])]; + tensor reshape_88_cast_fp16 = reshape(shape = reshape_88_shape_0, x = input_431_cast_fp16)[name = tensor("reshape_88_cast_fp16")]; + tensor reduce_mean_66_axes_0 = const()[name = tensor("reduce_mean_66_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_66_keep_dims_0 = const()[name = tensor("reduce_mean_66_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_66_cast_fp16 = reduce_mean(axes = reduce_mean_66_axes_0, keep_dims = reduce_mean_66_keep_dims_0, x = reshape_88_cast_fp16)[name = tensor("reduce_mean_66_cast_fp16")]; + tensor sub_44_cast_fp16 = sub(x = reshape_88_cast_fp16, y = reduce_mean_66_cast_fp16)[name = tensor("sub_44_cast_fp16")]; + tensor square_22_cast_fp16 = square(x = sub_44_cast_fp16)[name = tensor("square_22_cast_fp16")]; + tensor reduce_mean_68_axes_0 = const()[name = tensor("reduce_mean_68_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_68_keep_dims_0 = const()[name = tensor("reduce_mean_68_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_68_cast_fp16 = reduce_mean(axes = reduce_mean_68_axes_0, keep_dims = reduce_mean_68_keep_dims_0, x = square_22_cast_fp16)[name = tensor("reduce_mean_68_cast_fp16")]; + tensor add_44_y_0_to_fp16 = const()[name = tensor("add_44_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_44_cast_fp16 = add(x = reduce_mean_68_cast_fp16, y = add_44_y_0_to_fp16)[name = tensor("add_44_cast_fp16")]; + tensor sqrt_22_cast_fp16 = sqrt(x = add_44_cast_fp16)[name = tensor("sqrt_22_cast_fp16")]; + tensor real_div_22_cast_fp16 = real_div(x = sub_44_cast_fp16, y = sqrt_22_cast_fp16)[name = tensor("real_div_22_cast_fp16")]; + tensor reshape_89_shape_0 = const()[name = tensor("reshape_89_shape_0"), val = tensor([2, 1280, 32, 32])]; + tensor reshape_89_cast_fp16 = reshape(shape = reshape_89_shape_0, x = real_div_22_cast_fp16)[name = tensor("reshape_89_cast_fp16")]; + tensor add_45_gamma_0_to_fp16 = const()[name = tensor("add_45_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2549081216)))]; + tensor add_45_beta_0_to_fp16 = const()[name = tensor("add_45_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2549083840)))]; + tensor add_45_epsilon_0_to_fp16 = const()[name = tensor("add_45_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_45_cast_fp16 = batch_norm(beta = add_45_beta_0_to_fp16, epsilon = add_45_epsilon_0_to_fp16, gamma = add_45_gamma_0_to_fp16, mean = add_23_mean_0_to_fp16, variance = add_23_variance_0_to_fp16, x = reshape_89_cast_fp16)[name = tensor("add_45_cast_fp16")]; + tensor input_435_cast_fp16 = silu(x = add_45_cast_fp16)[name = tensor("input_435_cast_fp16")]; + tensor var_30404 = const()[name = tensor("op_30404"), val = tensor([1, 1])]; + tensor var_30406 = const()[name = tensor("op_30406"), val = tensor([1, 1])]; + tensor hidden_states_281_pad_type_0 = const()[name = tensor("hidden_states_281_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_281_pad_0 = const()[name = tensor("hidden_states_281_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor up_blocks_0_resnets_0_conv2_weight_to_fp16 = const()[name = tensor("up_blocks_0_resnets_0_conv2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2549086464)))]; + tensor up_blocks_0_resnets_0_conv2_bias_to_fp16 = const()[name = tensor("up_blocks_0_resnets_0_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2578577728)))]; + tensor hidden_states_281_cast_fp16 = conv(bias = up_blocks_0_resnets_0_conv2_bias_to_fp16, dilations = var_30406, groups = var_30355, pad = hidden_states_281_pad_0, pad_type = hidden_states_281_pad_type_0, strides = var_30404, weight = up_blocks_0_resnets_0_conv2_weight_to_fp16, x = input_435_cast_fp16)[name = tensor("hidden_states_281_cast_fp16")]; + tensor var_30411 = const()[name = tensor("op_30411"), val = tensor([1, 1])]; + tensor var_30413 = const()[name = tensor("op_30413"), val = tensor([1, 1])]; + tensor x_5_pad_type_0 = const()[name = tensor("x_5_pad_type_0"), val = tensor("custom")]; + tensor x_5_pad_0 = const()[name = tensor("x_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_resnets_0_conv_shortcut_weight_to_fp16 = const()[name = tensor("up_blocks_0_resnets_0_conv_shortcut_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2578580352)))]; + tensor up_blocks_0_resnets_0_conv_shortcut_bias_to_fp16 = const()[name = tensor("up_blocks_0_resnets_0_conv_shortcut_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2585134016)))]; + tensor x_5_cast_fp16 = conv(bias = up_blocks_0_resnets_0_conv_shortcut_bias_to_fp16, dilations = var_30413, groups = var_30355, pad = x_5_pad_0, pad_type = x_5_pad_type_0, strides = var_30411, weight = up_blocks_0_resnets_0_conv_shortcut_weight_to_fp16, x = input_423_cast_fp16)[name = tensor("x_5_cast_fp16")]; + tensor hidden_states_283_cast_fp16 = add(x = x_5_cast_fp16, y = hidden_states_281_cast_fp16)[name = tensor("hidden_states_283_cast_fp16")]; + tensor reshape_92_shape_0 = const()[name = tensor("reshape_92_shape_0"), val = tensor([2, 32, 40, 32, 32])]; + tensor reshape_92_cast_fp16 = reshape(shape = reshape_92_shape_0, x = hidden_states_283_cast_fp16)[name = tensor("reshape_92_cast_fp16")]; + tensor reduce_mean_69_axes_0 = const()[name = tensor("reduce_mean_69_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_69_keep_dims_0 = const()[name = tensor("reduce_mean_69_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_69_cast_fp16 = reduce_mean(axes = reduce_mean_69_axes_0, keep_dims = reduce_mean_69_keep_dims_0, x = reshape_92_cast_fp16)[name = tensor("reduce_mean_69_cast_fp16")]; + tensor sub_46_cast_fp16 = sub(x = reshape_92_cast_fp16, y = reduce_mean_69_cast_fp16)[name = tensor("sub_46_cast_fp16")]; + tensor square_23_cast_fp16 = square(x = sub_46_cast_fp16)[name = tensor("square_23_cast_fp16")]; + tensor reduce_mean_71_axes_0 = const()[name = tensor("reduce_mean_71_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_71_keep_dims_0 = const()[name = tensor("reduce_mean_71_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_71_cast_fp16 = reduce_mean(axes = reduce_mean_71_axes_0, keep_dims = reduce_mean_71_keep_dims_0, x = square_23_cast_fp16)[name = tensor("reduce_mean_71_cast_fp16")]; + tensor add_46_y_0_to_fp16 = const()[name = tensor("add_46_y_0_to_fp16"), val = tensor(0x1.1p-20)]; + tensor add_46_cast_fp16 = add(x = reduce_mean_71_cast_fp16, y = add_46_y_0_to_fp16)[name = tensor("add_46_cast_fp16")]; + tensor sqrt_23_cast_fp16 = sqrt(x = add_46_cast_fp16)[name = tensor("sqrt_23_cast_fp16")]; + tensor real_div_23_cast_fp16 = real_div(x = sub_46_cast_fp16, y = sqrt_23_cast_fp16)[name = tensor("real_div_23_cast_fp16")]; + tensor reshape_93_shape_0 = const()[name = tensor("reshape_93_shape_0"), val = tensor([2, 1280, 32, 32])]; + tensor reshape_93_cast_fp16 = reshape(shape = reshape_93_shape_0, x = real_div_23_cast_fp16)[name = tensor("reshape_93_cast_fp16")]; + tensor add_47_gamma_0_to_fp16 = const()[name = tensor("add_47_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2585136640)))]; + tensor add_47_beta_0_to_fp16 = const()[name = tensor("add_47_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2585139264)))]; + tensor add_47_epsilon_0_to_fp16 = const()[name = tensor("add_47_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_47_cast_fp16 = batch_norm(beta = add_47_beta_0_to_fp16, epsilon = add_47_epsilon_0_to_fp16, gamma = add_47_gamma_0_to_fp16, mean = add_23_mean_0_to_fp16, variance = add_23_variance_0_to_fp16, x = reshape_93_cast_fp16)[name = tensor("add_47_cast_fp16")]; + tensor var_30451 = const()[name = tensor("op_30451"), val = tensor([1, 1])]; + tensor var_30453 = const()[name = tensor("op_30453"), val = tensor([1, 1])]; + tensor hidden_states_285_pad_type_0 = const()[name = tensor("hidden_states_285_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_285_pad_0 = const()[name = tensor("hidden_states_285_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_proj_in_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_proj_in_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2585141888)))]; + tensor up_blocks_0_attentions_0_proj_in_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_proj_in_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2588418752)))]; + tensor hidden_states_285_cast_fp16 = conv(bias = up_blocks_0_attentions_0_proj_in_bias_to_fp16, dilations = var_30453, groups = var_30355, pad = hidden_states_285_pad_0, pad_type = hidden_states_285_pad_type_0, strides = var_30451, weight = up_blocks_0_attentions_0_proj_in_weight_to_fp16, x = add_47_cast_fp16)[name = tensor("hidden_states_285_cast_fp16")]; + tensor var_30458 = const()[name = tensor("op_30458"), val = tensor([2, 1280, 1, 1024])]; + tensor inputs_205_cast_fp16 = reshape(shape = var_30458, x = hidden_states_285_cast_fp16)[name = tensor("inputs_205_cast_fp16")]; + tensor hidden_states_287_axes_0 = const()[name = tensor("hidden_states_287_axes_0"), val = tensor([1])]; + tensor hidden_states_287_gamma_0_to_fp16 = const()[name = tensor("hidden_states_287_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2588421376)))]; + tensor hidden_states_287_beta_0_to_fp16 = const()[name = tensor("hidden_states_287_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2588424000)))]; + tensor var_30474_to_fp16 = const()[name = tensor("op_30474_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_287_cast_fp16 = layer_norm(axes = hidden_states_287_axes_0, beta = hidden_states_287_beta_0_to_fp16, epsilon = var_30474_to_fp16, gamma = hidden_states_287_gamma_0_to_fp16, x = inputs_205_cast_fp16)[name = tensor("hidden_states_287_cast_fp16")]; + tensor var_30489 = const()[name = tensor("op_30489"), val = tensor([1, 1])]; + tensor var_30491 = const()[name = tensor("op_30491"), val = tensor([1, 1])]; + tensor q_137_pad_type_0 = const()[name = tensor("q_137_pad_type_0"), val = tensor("custom")]; + tensor q_137_pad_0 = const()[name = tensor("q_137_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2588426624)))]; + tensor q_137_cast_fp16 = conv(dilations = var_30491, groups = var_30355, pad = q_137_pad_0, pad_type = q_137_pad_type_0, strides = var_30489, weight = up_blocks_0_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16, x = hidden_states_287_cast_fp16)[name = tensor("q_137_cast_fp16")]; + tensor var_30495 = const()[name = tensor("op_30495"), val = tensor([1, 1])]; + tensor var_30497 = const()[name = tensor("op_30497"), val = tensor([1, 1])]; + tensor k_273_pad_type_0 = const()[name = tensor("k_273_pad_type_0"), val = tensor("custom")]; + tensor k_273_pad_0 = const()[name = tensor("k_273_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2591703488)))]; + tensor k_273_cast_fp16 = conv(dilations = var_30497, groups = var_30355, pad = k_273_pad_0, pad_type = k_273_pad_type_0, strides = var_30495, weight = up_blocks_0_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16, x = hidden_states_287_cast_fp16)[name = tensor("k_273_cast_fp16")]; + tensor var_30501 = const()[name = tensor("op_30501"), val = tensor([1, 1])]; + tensor var_30503 = const()[name = tensor("op_30503"), val = tensor([1, 1])]; + tensor v_137_pad_type_0 = const()[name = tensor("v_137_pad_type_0"), val = tensor("custom")]; + tensor v_137_pad_0 = const()[name = tensor("v_137_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2594980352)))]; + tensor v_137_cast_fp16 = conv(dilations = var_30503, groups = var_30355, pad = v_137_pad_0, pad_type = v_137_pad_type_0, strides = var_30501, weight = up_blocks_0_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16, x = hidden_states_287_cast_fp16)[name = tensor("v_137_cast_fp16")]; + tensor var_30507_begin_0 = const()[name = tensor("op_30507_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_30507_end_0 = const()[name = tensor("op_30507_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_30507_end_mask_0 = const()[name = tensor("op_30507_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30507_cast_fp16 = slice_by_index(begin = var_30507_begin_0, end = var_30507_end_0, end_mask = var_30507_end_mask_0, x = q_137_cast_fp16)[name = tensor("op_30507_cast_fp16")]; + tensor var_30511_begin_0 = const()[name = tensor("op_30511_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_30511_end_0 = const()[name = tensor("op_30511_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_30511_end_mask_0 = const()[name = tensor("op_30511_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30511_cast_fp16 = slice_by_index(begin = var_30511_begin_0, end = var_30511_end_0, end_mask = var_30511_end_mask_0, x = q_137_cast_fp16)[name = tensor("op_30511_cast_fp16")]; + tensor var_30515_begin_0 = const()[name = tensor("op_30515_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_30515_end_0 = const()[name = tensor("op_30515_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_30515_end_mask_0 = const()[name = tensor("op_30515_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30515_cast_fp16 = slice_by_index(begin = var_30515_begin_0, end = var_30515_end_0, end_mask = var_30515_end_mask_0, x = q_137_cast_fp16)[name = tensor("op_30515_cast_fp16")]; + tensor var_30519_begin_0 = const()[name = tensor("op_30519_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_30519_end_0 = const()[name = tensor("op_30519_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_30519_end_mask_0 = const()[name = tensor("op_30519_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30519_cast_fp16 = slice_by_index(begin = var_30519_begin_0, end = var_30519_end_0, end_mask = var_30519_end_mask_0, x = q_137_cast_fp16)[name = tensor("op_30519_cast_fp16")]; + tensor var_30523_begin_0 = const()[name = tensor("op_30523_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_30523_end_0 = const()[name = tensor("op_30523_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_30523_end_mask_0 = const()[name = tensor("op_30523_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30523_cast_fp16 = slice_by_index(begin = var_30523_begin_0, end = var_30523_end_0, end_mask = var_30523_end_mask_0, x = q_137_cast_fp16)[name = tensor("op_30523_cast_fp16")]; + tensor var_30527_begin_0 = const()[name = tensor("op_30527_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_30527_end_0 = const()[name = tensor("op_30527_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_30527_end_mask_0 = const()[name = tensor("op_30527_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30527_cast_fp16 = slice_by_index(begin = var_30527_begin_0, end = var_30527_end_0, end_mask = var_30527_end_mask_0, x = q_137_cast_fp16)[name = tensor("op_30527_cast_fp16")]; + tensor var_30531_begin_0 = const()[name = tensor("op_30531_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_30531_end_0 = const()[name = tensor("op_30531_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_30531_end_mask_0 = const()[name = tensor("op_30531_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30531_cast_fp16 = slice_by_index(begin = var_30531_begin_0, end = var_30531_end_0, end_mask = var_30531_end_mask_0, x = q_137_cast_fp16)[name = tensor("op_30531_cast_fp16")]; + tensor var_30535_begin_0 = const()[name = tensor("op_30535_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_30535_end_0 = const()[name = tensor("op_30535_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_30535_end_mask_0 = const()[name = tensor("op_30535_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30535_cast_fp16 = slice_by_index(begin = var_30535_begin_0, end = var_30535_end_0, end_mask = var_30535_end_mask_0, x = q_137_cast_fp16)[name = tensor("op_30535_cast_fp16")]; + tensor var_30539_begin_0 = const()[name = tensor("op_30539_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_30539_end_0 = const()[name = tensor("op_30539_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_30539_end_mask_0 = const()[name = tensor("op_30539_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30539_cast_fp16 = slice_by_index(begin = var_30539_begin_0, end = var_30539_end_0, end_mask = var_30539_end_mask_0, x = q_137_cast_fp16)[name = tensor("op_30539_cast_fp16")]; + tensor var_30543_begin_0 = const()[name = tensor("op_30543_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_30543_end_0 = const()[name = tensor("op_30543_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_30543_end_mask_0 = const()[name = tensor("op_30543_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30543_cast_fp16 = slice_by_index(begin = var_30543_begin_0, end = var_30543_end_0, end_mask = var_30543_end_mask_0, x = q_137_cast_fp16)[name = tensor("op_30543_cast_fp16")]; + tensor var_30547_begin_0 = const()[name = tensor("op_30547_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_30547_end_0 = const()[name = tensor("op_30547_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_30547_end_mask_0 = const()[name = tensor("op_30547_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30547_cast_fp16 = slice_by_index(begin = var_30547_begin_0, end = var_30547_end_0, end_mask = var_30547_end_mask_0, x = q_137_cast_fp16)[name = tensor("op_30547_cast_fp16")]; + tensor var_30551_begin_0 = const()[name = tensor("op_30551_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_30551_end_0 = const()[name = tensor("op_30551_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_30551_end_mask_0 = const()[name = tensor("op_30551_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30551_cast_fp16 = slice_by_index(begin = var_30551_begin_0, end = var_30551_end_0, end_mask = var_30551_end_mask_0, x = q_137_cast_fp16)[name = tensor("op_30551_cast_fp16")]; + tensor var_30555_begin_0 = const()[name = tensor("op_30555_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_30555_end_0 = const()[name = tensor("op_30555_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_30555_end_mask_0 = const()[name = tensor("op_30555_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30555_cast_fp16 = slice_by_index(begin = var_30555_begin_0, end = var_30555_end_0, end_mask = var_30555_end_mask_0, x = q_137_cast_fp16)[name = tensor("op_30555_cast_fp16")]; + tensor var_30559_begin_0 = const()[name = tensor("op_30559_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_30559_end_0 = const()[name = tensor("op_30559_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_30559_end_mask_0 = const()[name = tensor("op_30559_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30559_cast_fp16 = slice_by_index(begin = var_30559_begin_0, end = var_30559_end_0, end_mask = var_30559_end_mask_0, x = q_137_cast_fp16)[name = tensor("op_30559_cast_fp16")]; + tensor var_30563_begin_0 = const()[name = tensor("op_30563_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_30563_end_0 = const()[name = tensor("op_30563_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_30563_end_mask_0 = const()[name = tensor("op_30563_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30563_cast_fp16 = slice_by_index(begin = var_30563_begin_0, end = var_30563_end_0, end_mask = var_30563_end_mask_0, x = q_137_cast_fp16)[name = tensor("op_30563_cast_fp16")]; + tensor var_30567_begin_0 = const()[name = tensor("op_30567_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_30567_end_0 = const()[name = tensor("op_30567_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_30567_end_mask_0 = const()[name = tensor("op_30567_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30567_cast_fp16 = slice_by_index(begin = var_30567_begin_0, end = var_30567_end_0, end_mask = var_30567_end_mask_0, x = q_137_cast_fp16)[name = tensor("op_30567_cast_fp16")]; + tensor var_30571_begin_0 = const()[name = tensor("op_30571_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_30571_end_0 = const()[name = tensor("op_30571_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_30571_end_mask_0 = const()[name = tensor("op_30571_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30571_cast_fp16 = slice_by_index(begin = var_30571_begin_0, end = var_30571_end_0, end_mask = var_30571_end_mask_0, x = q_137_cast_fp16)[name = tensor("op_30571_cast_fp16")]; + tensor var_30575_begin_0 = const()[name = tensor("op_30575_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_30575_end_0 = const()[name = tensor("op_30575_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_30575_end_mask_0 = const()[name = tensor("op_30575_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30575_cast_fp16 = slice_by_index(begin = var_30575_begin_0, end = var_30575_end_0, end_mask = var_30575_end_mask_0, x = q_137_cast_fp16)[name = tensor("op_30575_cast_fp16")]; + tensor var_30579_begin_0 = const()[name = tensor("op_30579_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_30579_end_0 = const()[name = tensor("op_30579_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_30579_end_mask_0 = const()[name = tensor("op_30579_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30579_cast_fp16 = slice_by_index(begin = var_30579_begin_0, end = var_30579_end_0, end_mask = var_30579_end_mask_0, x = q_137_cast_fp16)[name = tensor("op_30579_cast_fp16")]; + tensor var_30583_begin_0 = const()[name = tensor("op_30583_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_30583_end_0 = const()[name = tensor("op_30583_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_30583_end_mask_0 = const()[name = tensor("op_30583_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30583_cast_fp16 = slice_by_index(begin = var_30583_begin_0, end = var_30583_end_0, end_mask = var_30583_end_mask_0, x = q_137_cast_fp16)[name = tensor("op_30583_cast_fp16")]; + tensor k_275_perm_0 = const()[name = tensor("k_275_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_30590_begin_0 = const()[name = tensor("op_30590_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_30590_end_0 = const()[name = tensor("op_30590_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_30590_end_mask_0 = const()[name = tensor("op_30590_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_275_cast_fp16 = transpose(perm = k_275_perm_0, x = k_273_cast_fp16)[name = tensor("transpose_71")]; + tensor var_30590_cast_fp16 = slice_by_index(begin = var_30590_begin_0, end = var_30590_end_0, end_mask = var_30590_end_mask_0, x = k_275_cast_fp16)[name = tensor("op_30590_cast_fp16")]; + tensor var_30594_begin_0 = const()[name = tensor("op_30594_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_30594_end_0 = const()[name = tensor("op_30594_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_30594_end_mask_0 = const()[name = tensor("op_30594_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30594_cast_fp16 = slice_by_index(begin = var_30594_begin_0, end = var_30594_end_0, end_mask = var_30594_end_mask_0, x = k_275_cast_fp16)[name = tensor("op_30594_cast_fp16")]; + tensor var_30598_begin_0 = const()[name = tensor("op_30598_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_30598_end_0 = const()[name = tensor("op_30598_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_30598_end_mask_0 = const()[name = tensor("op_30598_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30598_cast_fp16 = slice_by_index(begin = var_30598_begin_0, end = var_30598_end_0, end_mask = var_30598_end_mask_0, x = k_275_cast_fp16)[name = tensor("op_30598_cast_fp16")]; + tensor var_30602_begin_0 = const()[name = tensor("op_30602_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_30602_end_0 = const()[name = tensor("op_30602_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_30602_end_mask_0 = const()[name = tensor("op_30602_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30602_cast_fp16 = slice_by_index(begin = var_30602_begin_0, end = var_30602_end_0, end_mask = var_30602_end_mask_0, x = k_275_cast_fp16)[name = tensor("op_30602_cast_fp16")]; + tensor var_30606_begin_0 = const()[name = tensor("op_30606_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_30606_end_0 = const()[name = tensor("op_30606_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_30606_end_mask_0 = const()[name = tensor("op_30606_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30606_cast_fp16 = slice_by_index(begin = var_30606_begin_0, end = var_30606_end_0, end_mask = var_30606_end_mask_0, x = k_275_cast_fp16)[name = tensor("op_30606_cast_fp16")]; + tensor var_30610_begin_0 = const()[name = tensor("op_30610_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_30610_end_0 = const()[name = tensor("op_30610_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_30610_end_mask_0 = const()[name = tensor("op_30610_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30610_cast_fp16 = slice_by_index(begin = var_30610_begin_0, end = var_30610_end_0, end_mask = var_30610_end_mask_0, x = k_275_cast_fp16)[name = tensor("op_30610_cast_fp16")]; + tensor var_30614_begin_0 = const()[name = tensor("op_30614_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_30614_end_0 = const()[name = tensor("op_30614_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_30614_end_mask_0 = const()[name = tensor("op_30614_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30614_cast_fp16 = slice_by_index(begin = var_30614_begin_0, end = var_30614_end_0, end_mask = var_30614_end_mask_0, x = k_275_cast_fp16)[name = tensor("op_30614_cast_fp16")]; + tensor var_30618_begin_0 = const()[name = tensor("op_30618_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_30618_end_0 = const()[name = tensor("op_30618_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_30618_end_mask_0 = const()[name = tensor("op_30618_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30618_cast_fp16 = slice_by_index(begin = var_30618_begin_0, end = var_30618_end_0, end_mask = var_30618_end_mask_0, x = k_275_cast_fp16)[name = tensor("op_30618_cast_fp16")]; + tensor var_30622_begin_0 = const()[name = tensor("op_30622_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_30622_end_0 = const()[name = tensor("op_30622_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_30622_end_mask_0 = const()[name = tensor("op_30622_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30622_cast_fp16 = slice_by_index(begin = var_30622_begin_0, end = var_30622_end_0, end_mask = var_30622_end_mask_0, x = k_275_cast_fp16)[name = tensor("op_30622_cast_fp16")]; + tensor var_30626_begin_0 = const()[name = tensor("op_30626_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_30626_end_0 = const()[name = tensor("op_30626_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_30626_end_mask_0 = const()[name = tensor("op_30626_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30626_cast_fp16 = slice_by_index(begin = var_30626_begin_0, end = var_30626_end_0, end_mask = var_30626_end_mask_0, x = k_275_cast_fp16)[name = tensor("op_30626_cast_fp16")]; + tensor var_30630_begin_0 = const()[name = tensor("op_30630_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_30630_end_0 = const()[name = tensor("op_30630_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_30630_end_mask_0 = const()[name = tensor("op_30630_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30630_cast_fp16 = slice_by_index(begin = var_30630_begin_0, end = var_30630_end_0, end_mask = var_30630_end_mask_0, x = k_275_cast_fp16)[name = tensor("op_30630_cast_fp16")]; + tensor var_30634_begin_0 = const()[name = tensor("op_30634_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_30634_end_0 = const()[name = tensor("op_30634_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_30634_end_mask_0 = const()[name = tensor("op_30634_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30634_cast_fp16 = slice_by_index(begin = var_30634_begin_0, end = var_30634_end_0, end_mask = var_30634_end_mask_0, x = k_275_cast_fp16)[name = tensor("op_30634_cast_fp16")]; + tensor var_30638_begin_0 = const()[name = tensor("op_30638_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_30638_end_0 = const()[name = tensor("op_30638_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_30638_end_mask_0 = const()[name = tensor("op_30638_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30638_cast_fp16 = slice_by_index(begin = var_30638_begin_0, end = var_30638_end_0, end_mask = var_30638_end_mask_0, x = k_275_cast_fp16)[name = tensor("op_30638_cast_fp16")]; + tensor var_30642_begin_0 = const()[name = tensor("op_30642_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_30642_end_0 = const()[name = tensor("op_30642_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_30642_end_mask_0 = const()[name = tensor("op_30642_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30642_cast_fp16 = slice_by_index(begin = var_30642_begin_0, end = var_30642_end_0, end_mask = var_30642_end_mask_0, x = k_275_cast_fp16)[name = tensor("op_30642_cast_fp16")]; + tensor var_30646_begin_0 = const()[name = tensor("op_30646_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_30646_end_0 = const()[name = tensor("op_30646_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_30646_end_mask_0 = const()[name = tensor("op_30646_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30646_cast_fp16 = slice_by_index(begin = var_30646_begin_0, end = var_30646_end_0, end_mask = var_30646_end_mask_0, x = k_275_cast_fp16)[name = tensor("op_30646_cast_fp16")]; + tensor var_30650_begin_0 = const()[name = tensor("op_30650_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_30650_end_0 = const()[name = tensor("op_30650_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_30650_end_mask_0 = const()[name = tensor("op_30650_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30650_cast_fp16 = slice_by_index(begin = var_30650_begin_0, end = var_30650_end_0, end_mask = var_30650_end_mask_0, x = k_275_cast_fp16)[name = tensor("op_30650_cast_fp16")]; + tensor var_30654_begin_0 = const()[name = tensor("op_30654_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_30654_end_0 = const()[name = tensor("op_30654_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_30654_end_mask_0 = const()[name = tensor("op_30654_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30654_cast_fp16 = slice_by_index(begin = var_30654_begin_0, end = var_30654_end_0, end_mask = var_30654_end_mask_0, x = k_275_cast_fp16)[name = tensor("op_30654_cast_fp16")]; + tensor var_30658_begin_0 = const()[name = tensor("op_30658_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_30658_end_0 = const()[name = tensor("op_30658_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_30658_end_mask_0 = const()[name = tensor("op_30658_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30658_cast_fp16 = slice_by_index(begin = var_30658_begin_0, end = var_30658_end_0, end_mask = var_30658_end_mask_0, x = k_275_cast_fp16)[name = tensor("op_30658_cast_fp16")]; + tensor var_30662_begin_0 = const()[name = tensor("op_30662_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_30662_end_0 = const()[name = tensor("op_30662_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_30662_end_mask_0 = const()[name = tensor("op_30662_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30662_cast_fp16 = slice_by_index(begin = var_30662_begin_0, end = var_30662_end_0, end_mask = var_30662_end_mask_0, x = k_275_cast_fp16)[name = tensor("op_30662_cast_fp16")]; + tensor var_30666_begin_0 = const()[name = tensor("op_30666_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_30666_end_0 = const()[name = tensor("op_30666_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_30666_end_mask_0 = const()[name = tensor("op_30666_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30666_cast_fp16 = slice_by_index(begin = var_30666_begin_0, end = var_30666_end_0, end_mask = var_30666_end_mask_0, x = k_275_cast_fp16)[name = tensor("op_30666_cast_fp16")]; + tensor var_30668_begin_0 = const()[name = tensor("op_30668_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_30668_end_0 = const()[name = tensor("op_30668_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_30668_end_mask_0 = const()[name = tensor("op_30668_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30668_cast_fp16 = slice_by_index(begin = var_30668_begin_0, end = var_30668_end_0, end_mask = var_30668_end_mask_0, x = v_137_cast_fp16)[name = tensor("op_30668_cast_fp16")]; + tensor var_30672_begin_0 = const()[name = tensor("op_30672_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_30672_end_0 = const()[name = tensor("op_30672_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_30672_end_mask_0 = const()[name = tensor("op_30672_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30672_cast_fp16 = slice_by_index(begin = var_30672_begin_0, end = var_30672_end_0, end_mask = var_30672_end_mask_0, x = v_137_cast_fp16)[name = tensor("op_30672_cast_fp16")]; + tensor var_30676_begin_0 = const()[name = tensor("op_30676_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_30676_end_0 = const()[name = tensor("op_30676_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_30676_end_mask_0 = const()[name = tensor("op_30676_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30676_cast_fp16 = slice_by_index(begin = var_30676_begin_0, end = var_30676_end_0, end_mask = var_30676_end_mask_0, x = v_137_cast_fp16)[name = tensor("op_30676_cast_fp16")]; + tensor var_30680_begin_0 = const()[name = tensor("op_30680_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_30680_end_0 = const()[name = tensor("op_30680_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_30680_end_mask_0 = const()[name = tensor("op_30680_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30680_cast_fp16 = slice_by_index(begin = var_30680_begin_0, end = var_30680_end_0, end_mask = var_30680_end_mask_0, x = v_137_cast_fp16)[name = tensor("op_30680_cast_fp16")]; + tensor var_30684_begin_0 = const()[name = tensor("op_30684_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_30684_end_0 = const()[name = tensor("op_30684_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_30684_end_mask_0 = const()[name = tensor("op_30684_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30684_cast_fp16 = slice_by_index(begin = var_30684_begin_0, end = var_30684_end_0, end_mask = var_30684_end_mask_0, x = v_137_cast_fp16)[name = tensor("op_30684_cast_fp16")]; + tensor var_30688_begin_0 = const()[name = tensor("op_30688_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_30688_end_0 = const()[name = tensor("op_30688_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_30688_end_mask_0 = const()[name = tensor("op_30688_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30688_cast_fp16 = slice_by_index(begin = var_30688_begin_0, end = var_30688_end_0, end_mask = var_30688_end_mask_0, x = v_137_cast_fp16)[name = tensor("op_30688_cast_fp16")]; + tensor var_30692_begin_0 = const()[name = tensor("op_30692_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_30692_end_0 = const()[name = tensor("op_30692_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_30692_end_mask_0 = const()[name = tensor("op_30692_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30692_cast_fp16 = slice_by_index(begin = var_30692_begin_0, end = var_30692_end_0, end_mask = var_30692_end_mask_0, x = v_137_cast_fp16)[name = tensor("op_30692_cast_fp16")]; + tensor var_30696_begin_0 = const()[name = tensor("op_30696_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_30696_end_0 = const()[name = tensor("op_30696_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_30696_end_mask_0 = const()[name = tensor("op_30696_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30696_cast_fp16 = slice_by_index(begin = var_30696_begin_0, end = var_30696_end_0, end_mask = var_30696_end_mask_0, x = v_137_cast_fp16)[name = tensor("op_30696_cast_fp16")]; + tensor var_30700_begin_0 = const()[name = tensor("op_30700_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_30700_end_0 = const()[name = tensor("op_30700_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_30700_end_mask_0 = const()[name = tensor("op_30700_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30700_cast_fp16 = slice_by_index(begin = var_30700_begin_0, end = var_30700_end_0, end_mask = var_30700_end_mask_0, x = v_137_cast_fp16)[name = tensor("op_30700_cast_fp16")]; + tensor var_30704_begin_0 = const()[name = tensor("op_30704_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_30704_end_0 = const()[name = tensor("op_30704_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_30704_end_mask_0 = const()[name = tensor("op_30704_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30704_cast_fp16 = slice_by_index(begin = var_30704_begin_0, end = var_30704_end_0, end_mask = var_30704_end_mask_0, x = v_137_cast_fp16)[name = tensor("op_30704_cast_fp16")]; + tensor var_30708_begin_0 = const()[name = tensor("op_30708_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_30708_end_0 = const()[name = tensor("op_30708_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_30708_end_mask_0 = const()[name = tensor("op_30708_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30708_cast_fp16 = slice_by_index(begin = var_30708_begin_0, end = var_30708_end_0, end_mask = var_30708_end_mask_0, x = v_137_cast_fp16)[name = tensor("op_30708_cast_fp16")]; + tensor var_30712_begin_0 = const()[name = tensor("op_30712_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_30712_end_0 = const()[name = tensor("op_30712_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_30712_end_mask_0 = const()[name = tensor("op_30712_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30712_cast_fp16 = slice_by_index(begin = var_30712_begin_0, end = var_30712_end_0, end_mask = var_30712_end_mask_0, x = v_137_cast_fp16)[name = tensor("op_30712_cast_fp16")]; + tensor var_30716_begin_0 = const()[name = tensor("op_30716_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_30716_end_0 = const()[name = tensor("op_30716_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_30716_end_mask_0 = const()[name = tensor("op_30716_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30716_cast_fp16 = slice_by_index(begin = var_30716_begin_0, end = var_30716_end_0, end_mask = var_30716_end_mask_0, x = v_137_cast_fp16)[name = tensor("op_30716_cast_fp16")]; + tensor var_30720_begin_0 = const()[name = tensor("op_30720_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_30720_end_0 = const()[name = tensor("op_30720_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_30720_end_mask_0 = const()[name = tensor("op_30720_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30720_cast_fp16 = slice_by_index(begin = var_30720_begin_0, end = var_30720_end_0, end_mask = var_30720_end_mask_0, x = v_137_cast_fp16)[name = tensor("op_30720_cast_fp16")]; + tensor var_30724_begin_0 = const()[name = tensor("op_30724_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_30724_end_0 = const()[name = tensor("op_30724_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_30724_end_mask_0 = const()[name = tensor("op_30724_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30724_cast_fp16 = slice_by_index(begin = var_30724_begin_0, end = var_30724_end_0, end_mask = var_30724_end_mask_0, x = v_137_cast_fp16)[name = tensor("op_30724_cast_fp16")]; + tensor var_30728_begin_0 = const()[name = tensor("op_30728_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_30728_end_0 = const()[name = tensor("op_30728_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_30728_end_mask_0 = const()[name = tensor("op_30728_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30728_cast_fp16 = slice_by_index(begin = var_30728_begin_0, end = var_30728_end_0, end_mask = var_30728_end_mask_0, x = v_137_cast_fp16)[name = tensor("op_30728_cast_fp16")]; + tensor var_30732_begin_0 = const()[name = tensor("op_30732_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_30732_end_0 = const()[name = tensor("op_30732_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_30732_end_mask_0 = const()[name = tensor("op_30732_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30732_cast_fp16 = slice_by_index(begin = var_30732_begin_0, end = var_30732_end_0, end_mask = var_30732_end_mask_0, x = v_137_cast_fp16)[name = tensor("op_30732_cast_fp16")]; + tensor var_30736_begin_0 = const()[name = tensor("op_30736_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_30736_end_0 = const()[name = tensor("op_30736_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_30736_end_mask_0 = const()[name = tensor("op_30736_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30736_cast_fp16 = slice_by_index(begin = var_30736_begin_0, end = var_30736_end_0, end_mask = var_30736_end_mask_0, x = v_137_cast_fp16)[name = tensor("op_30736_cast_fp16")]; + tensor var_30740_begin_0 = const()[name = tensor("op_30740_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_30740_end_0 = const()[name = tensor("op_30740_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_30740_end_mask_0 = const()[name = tensor("op_30740_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30740_cast_fp16 = slice_by_index(begin = var_30740_begin_0, end = var_30740_end_0, end_mask = var_30740_end_mask_0, x = v_137_cast_fp16)[name = tensor("op_30740_cast_fp16")]; + tensor var_30744_begin_0 = const()[name = tensor("op_30744_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_30744_end_0 = const()[name = tensor("op_30744_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_30744_end_mask_0 = const()[name = tensor("op_30744_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30744_cast_fp16 = slice_by_index(begin = var_30744_begin_0, end = var_30744_end_0, end_mask = var_30744_end_mask_0, x = v_137_cast_fp16)[name = tensor("op_30744_cast_fp16")]; + tensor var_30748_equation_0 = const()[name = tensor("op_30748_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30748_cast_fp16 = einsum(equation = var_30748_equation_0, values = (var_30590_cast_fp16, var_30507_cast_fp16))[name = tensor("op_30748_cast_fp16")]; + tensor var_30749_to_fp16 = const()[name = tensor("op_30749_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2561_cast_fp16 = mul(x = var_30748_cast_fp16, y = var_30749_to_fp16)[name = tensor("aw_2561_cast_fp16")]; + tensor var_30752_equation_0 = const()[name = tensor("op_30752_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30752_cast_fp16 = einsum(equation = var_30752_equation_0, values = (var_30594_cast_fp16, var_30511_cast_fp16))[name = tensor("op_30752_cast_fp16")]; + tensor var_30753_to_fp16 = const()[name = tensor("op_30753_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2563_cast_fp16 = mul(x = var_30752_cast_fp16, y = var_30753_to_fp16)[name = tensor("aw_2563_cast_fp16")]; + tensor var_30756_equation_0 = const()[name = tensor("op_30756_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30756_cast_fp16 = einsum(equation = var_30756_equation_0, values = (var_30598_cast_fp16, var_30515_cast_fp16))[name = tensor("op_30756_cast_fp16")]; + tensor var_30757_to_fp16 = const()[name = tensor("op_30757_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2565_cast_fp16 = mul(x = var_30756_cast_fp16, y = var_30757_to_fp16)[name = tensor("aw_2565_cast_fp16")]; + tensor var_30760_equation_0 = const()[name = tensor("op_30760_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30760_cast_fp16 = einsum(equation = var_30760_equation_0, values = (var_30602_cast_fp16, var_30519_cast_fp16))[name = tensor("op_30760_cast_fp16")]; + tensor var_30761_to_fp16 = const()[name = tensor("op_30761_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2567_cast_fp16 = mul(x = var_30760_cast_fp16, y = var_30761_to_fp16)[name = tensor("aw_2567_cast_fp16")]; + tensor var_30764_equation_0 = const()[name = tensor("op_30764_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30764_cast_fp16 = einsum(equation = var_30764_equation_0, values = (var_30606_cast_fp16, var_30523_cast_fp16))[name = tensor("op_30764_cast_fp16")]; + tensor var_30765_to_fp16 = const()[name = tensor("op_30765_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2569_cast_fp16 = mul(x = var_30764_cast_fp16, y = var_30765_to_fp16)[name = tensor("aw_2569_cast_fp16")]; + tensor var_30768_equation_0 = const()[name = tensor("op_30768_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30768_cast_fp16 = einsum(equation = var_30768_equation_0, values = (var_30610_cast_fp16, var_30527_cast_fp16))[name = tensor("op_30768_cast_fp16")]; + tensor var_30769_to_fp16 = const()[name = tensor("op_30769_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2571_cast_fp16 = mul(x = var_30768_cast_fp16, y = var_30769_to_fp16)[name = tensor("aw_2571_cast_fp16")]; + tensor var_30772_equation_0 = const()[name = tensor("op_30772_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30772_cast_fp16 = einsum(equation = var_30772_equation_0, values = (var_30614_cast_fp16, var_30531_cast_fp16))[name = tensor("op_30772_cast_fp16")]; + tensor var_30773_to_fp16 = const()[name = tensor("op_30773_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2573_cast_fp16 = mul(x = var_30772_cast_fp16, y = var_30773_to_fp16)[name = tensor("aw_2573_cast_fp16")]; + tensor var_30776_equation_0 = const()[name = tensor("op_30776_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30776_cast_fp16 = einsum(equation = var_30776_equation_0, values = (var_30618_cast_fp16, var_30535_cast_fp16))[name = tensor("op_30776_cast_fp16")]; + tensor var_30777_to_fp16 = const()[name = tensor("op_30777_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2575_cast_fp16 = mul(x = var_30776_cast_fp16, y = var_30777_to_fp16)[name = tensor("aw_2575_cast_fp16")]; + tensor var_30780_equation_0 = const()[name = tensor("op_30780_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30780_cast_fp16 = einsum(equation = var_30780_equation_0, values = (var_30622_cast_fp16, var_30539_cast_fp16))[name = tensor("op_30780_cast_fp16")]; + tensor var_30781_to_fp16 = const()[name = tensor("op_30781_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2577_cast_fp16 = mul(x = var_30780_cast_fp16, y = var_30781_to_fp16)[name = tensor("aw_2577_cast_fp16")]; + tensor var_30784_equation_0 = const()[name = tensor("op_30784_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30784_cast_fp16 = einsum(equation = var_30784_equation_0, values = (var_30626_cast_fp16, var_30543_cast_fp16))[name = tensor("op_30784_cast_fp16")]; + tensor var_30785_to_fp16 = const()[name = tensor("op_30785_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2579_cast_fp16 = mul(x = var_30784_cast_fp16, y = var_30785_to_fp16)[name = tensor("aw_2579_cast_fp16")]; + tensor var_30788_equation_0 = const()[name = tensor("op_30788_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30788_cast_fp16 = einsum(equation = var_30788_equation_0, values = (var_30630_cast_fp16, var_30547_cast_fp16))[name = tensor("op_30788_cast_fp16")]; + tensor var_30789_to_fp16 = const()[name = tensor("op_30789_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2581_cast_fp16 = mul(x = var_30788_cast_fp16, y = var_30789_to_fp16)[name = tensor("aw_2581_cast_fp16")]; + tensor var_30792_equation_0 = const()[name = tensor("op_30792_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30792_cast_fp16 = einsum(equation = var_30792_equation_0, values = (var_30634_cast_fp16, var_30551_cast_fp16))[name = tensor("op_30792_cast_fp16")]; + tensor var_30793_to_fp16 = const()[name = tensor("op_30793_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2583_cast_fp16 = mul(x = var_30792_cast_fp16, y = var_30793_to_fp16)[name = tensor("aw_2583_cast_fp16")]; + tensor var_30796_equation_0 = const()[name = tensor("op_30796_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30796_cast_fp16 = einsum(equation = var_30796_equation_0, values = (var_30638_cast_fp16, var_30555_cast_fp16))[name = tensor("op_30796_cast_fp16")]; + tensor var_30797_to_fp16 = const()[name = tensor("op_30797_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2585_cast_fp16 = mul(x = var_30796_cast_fp16, y = var_30797_to_fp16)[name = tensor("aw_2585_cast_fp16")]; + tensor var_30800_equation_0 = const()[name = tensor("op_30800_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30800_cast_fp16 = einsum(equation = var_30800_equation_0, values = (var_30642_cast_fp16, var_30559_cast_fp16))[name = tensor("op_30800_cast_fp16")]; + tensor var_30801_to_fp16 = const()[name = tensor("op_30801_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2587_cast_fp16 = mul(x = var_30800_cast_fp16, y = var_30801_to_fp16)[name = tensor("aw_2587_cast_fp16")]; + tensor var_30804_equation_0 = const()[name = tensor("op_30804_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30804_cast_fp16 = einsum(equation = var_30804_equation_0, values = (var_30646_cast_fp16, var_30563_cast_fp16))[name = tensor("op_30804_cast_fp16")]; + tensor var_30805_to_fp16 = const()[name = tensor("op_30805_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2589_cast_fp16 = mul(x = var_30804_cast_fp16, y = var_30805_to_fp16)[name = tensor("aw_2589_cast_fp16")]; + tensor var_30808_equation_0 = const()[name = tensor("op_30808_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30808_cast_fp16 = einsum(equation = var_30808_equation_0, values = (var_30650_cast_fp16, var_30567_cast_fp16))[name = tensor("op_30808_cast_fp16")]; + tensor var_30809_to_fp16 = const()[name = tensor("op_30809_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2591_cast_fp16 = mul(x = var_30808_cast_fp16, y = var_30809_to_fp16)[name = tensor("aw_2591_cast_fp16")]; + tensor var_30812_equation_0 = const()[name = tensor("op_30812_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30812_cast_fp16 = einsum(equation = var_30812_equation_0, values = (var_30654_cast_fp16, var_30571_cast_fp16))[name = tensor("op_30812_cast_fp16")]; + tensor var_30813_to_fp16 = const()[name = tensor("op_30813_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2593_cast_fp16 = mul(x = var_30812_cast_fp16, y = var_30813_to_fp16)[name = tensor("aw_2593_cast_fp16")]; + tensor var_30816_equation_0 = const()[name = tensor("op_30816_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30816_cast_fp16 = einsum(equation = var_30816_equation_0, values = (var_30658_cast_fp16, var_30575_cast_fp16))[name = tensor("op_30816_cast_fp16")]; + tensor var_30817_to_fp16 = const()[name = tensor("op_30817_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2595_cast_fp16 = mul(x = var_30816_cast_fp16, y = var_30817_to_fp16)[name = tensor("aw_2595_cast_fp16")]; + tensor var_30820_equation_0 = const()[name = tensor("op_30820_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30820_cast_fp16 = einsum(equation = var_30820_equation_0, values = (var_30662_cast_fp16, var_30579_cast_fp16))[name = tensor("op_30820_cast_fp16")]; + tensor var_30821_to_fp16 = const()[name = tensor("op_30821_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2597_cast_fp16 = mul(x = var_30820_cast_fp16, y = var_30821_to_fp16)[name = tensor("aw_2597_cast_fp16")]; + tensor var_30824_equation_0 = const()[name = tensor("op_30824_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30824_cast_fp16 = einsum(equation = var_30824_equation_0, values = (var_30666_cast_fp16, var_30583_cast_fp16))[name = tensor("op_30824_cast_fp16")]; + tensor var_30825_to_fp16 = const()[name = tensor("op_30825_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2599_cast_fp16 = mul(x = var_30824_cast_fp16, y = var_30825_to_fp16)[name = tensor("aw_2599_cast_fp16")]; + tensor var_30827_cast_fp16 = softmax(axis = var_30355, x = aw_2561_cast_fp16)[name = tensor("op_30827_cast_fp16")]; + tensor var_30828_cast_fp16 = softmax(axis = var_30355, x = aw_2563_cast_fp16)[name = tensor("op_30828_cast_fp16")]; + tensor var_30829_cast_fp16 = softmax(axis = var_30355, x = aw_2565_cast_fp16)[name = tensor("op_30829_cast_fp16")]; + tensor var_30830_cast_fp16 = softmax(axis = var_30355, x = aw_2567_cast_fp16)[name = tensor("op_30830_cast_fp16")]; + tensor var_30831_cast_fp16 = softmax(axis = var_30355, x = aw_2569_cast_fp16)[name = tensor("op_30831_cast_fp16")]; + tensor var_30832_cast_fp16 = softmax(axis = var_30355, x = aw_2571_cast_fp16)[name = tensor("op_30832_cast_fp16")]; + tensor var_30833_cast_fp16 = softmax(axis = var_30355, x = aw_2573_cast_fp16)[name = tensor("op_30833_cast_fp16")]; + tensor var_30834_cast_fp16 = softmax(axis = var_30355, x = aw_2575_cast_fp16)[name = tensor("op_30834_cast_fp16")]; + tensor var_30835_cast_fp16 = softmax(axis = var_30355, x = aw_2577_cast_fp16)[name = tensor("op_30835_cast_fp16")]; + tensor var_30836_cast_fp16 = softmax(axis = var_30355, x = aw_2579_cast_fp16)[name = tensor("op_30836_cast_fp16")]; + tensor var_30837_cast_fp16 = softmax(axis = var_30355, x = aw_2581_cast_fp16)[name = tensor("op_30837_cast_fp16")]; + tensor var_30838_cast_fp16 = softmax(axis = var_30355, x = aw_2583_cast_fp16)[name = tensor("op_30838_cast_fp16")]; + tensor var_30839_cast_fp16 = softmax(axis = var_30355, x = aw_2585_cast_fp16)[name = tensor("op_30839_cast_fp16")]; + tensor var_30840_cast_fp16 = softmax(axis = var_30355, x = aw_2587_cast_fp16)[name = tensor("op_30840_cast_fp16")]; + tensor var_30841_cast_fp16 = softmax(axis = var_30355, x = aw_2589_cast_fp16)[name = tensor("op_30841_cast_fp16")]; + tensor var_30842_cast_fp16 = softmax(axis = var_30355, x = aw_2591_cast_fp16)[name = tensor("op_30842_cast_fp16")]; + tensor var_30843_cast_fp16 = softmax(axis = var_30355, x = aw_2593_cast_fp16)[name = tensor("op_30843_cast_fp16")]; + tensor var_30844_cast_fp16 = softmax(axis = var_30355, x = aw_2595_cast_fp16)[name = tensor("op_30844_cast_fp16")]; + tensor var_30845_cast_fp16 = softmax(axis = var_30355, x = aw_2597_cast_fp16)[name = tensor("op_30845_cast_fp16")]; + tensor var_30846_cast_fp16 = softmax(axis = var_30355, x = aw_2599_cast_fp16)[name = tensor("op_30846_cast_fp16")]; + tensor var_30848_equation_0 = const()[name = tensor("op_30848_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30848_cast_fp16 = einsum(equation = var_30848_equation_0, values = (var_30668_cast_fp16, var_30827_cast_fp16))[name = tensor("op_30848_cast_fp16")]; + tensor var_30850_equation_0 = const()[name = tensor("op_30850_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30850_cast_fp16 = einsum(equation = var_30850_equation_0, values = (var_30672_cast_fp16, var_30828_cast_fp16))[name = tensor("op_30850_cast_fp16")]; + tensor var_30852_equation_0 = const()[name = tensor("op_30852_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30852_cast_fp16 = einsum(equation = var_30852_equation_0, values = (var_30676_cast_fp16, var_30829_cast_fp16))[name = tensor("op_30852_cast_fp16")]; + tensor var_30854_equation_0 = const()[name = tensor("op_30854_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30854_cast_fp16 = einsum(equation = var_30854_equation_0, values = (var_30680_cast_fp16, var_30830_cast_fp16))[name = tensor("op_30854_cast_fp16")]; + tensor var_30856_equation_0 = const()[name = tensor("op_30856_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30856_cast_fp16 = einsum(equation = var_30856_equation_0, values = (var_30684_cast_fp16, var_30831_cast_fp16))[name = tensor("op_30856_cast_fp16")]; + tensor var_30858_equation_0 = const()[name = tensor("op_30858_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30858_cast_fp16 = einsum(equation = var_30858_equation_0, values = (var_30688_cast_fp16, var_30832_cast_fp16))[name = tensor("op_30858_cast_fp16")]; + tensor var_30860_equation_0 = const()[name = tensor("op_30860_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30860_cast_fp16 = einsum(equation = var_30860_equation_0, values = (var_30692_cast_fp16, var_30833_cast_fp16))[name = tensor("op_30860_cast_fp16")]; + tensor var_30862_equation_0 = const()[name = tensor("op_30862_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30862_cast_fp16 = einsum(equation = var_30862_equation_0, values = (var_30696_cast_fp16, var_30834_cast_fp16))[name = tensor("op_30862_cast_fp16")]; + tensor var_30864_equation_0 = const()[name = tensor("op_30864_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30864_cast_fp16 = einsum(equation = var_30864_equation_0, values = (var_30700_cast_fp16, var_30835_cast_fp16))[name = tensor("op_30864_cast_fp16")]; + tensor var_30866_equation_0 = const()[name = tensor("op_30866_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30866_cast_fp16 = einsum(equation = var_30866_equation_0, values = (var_30704_cast_fp16, var_30836_cast_fp16))[name = tensor("op_30866_cast_fp16")]; + tensor var_30868_equation_0 = const()[name = tensor("op_30868_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30868_cast_fp16 = einsum(equation = var_30868_equation_0, values = (var_30708_cast_fp16, var_30837_cast_fp16))[name = tensor("op_30868_cast_fp16")]; + tensor var_30870_equation_0 = const()[name = tensor("op_30870_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30870_cast_fp16 = einsum(equation = var_30870_equation_0, values = (var_30712_cast_fp16, var_30838_cast_fp16))[name = tensor("op_30870_cast_fp16")]; + tensor var_30872_equation_0 = const()[name = tensor("op_30872_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30872_cast_fp16 = einsum(equation = var_30872_equation_0, values = (var_30716_cast_fp16, var_30839_cast_fp16))[name = tensor("op_30872_cast_fp16")]; + tensor var_30874_equation_0 = const()[name = tensor("op_30874_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30874_cast_fp16 = einsum(equation = var_30874_equation_0, values = (var_30720_cast_fp16, var_30840_cast_fp16))[name = tensor("op_30874_cast_fp16")]; + tensor var_30876_equation_0 = const()[name = tensor("op_30876_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30876_cast_fp16 = einsum(equation = var_30876_equation_0, values = (var_30724_cast_fp16, var_30841_cast_fp16))[name = tensor("op_30876_cast_fp16")]; + tensor var_30878_equation_0 = const()[name = tensor("op_30878_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30878_cast_fp16 = einsum(equation = var_30878_equation_0, values = (var_30728_cast_fp16, var_30842_cast_fp16))[name = tensor("op_30878_cast_fp16")]; + tensor var_30880_equation_0 = const()[name = tensor("op_30880_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30880_cast_fp16 = einsum(equation = var_30880_equation_0, values = (var_30732_cast_fp16, var_30843_cast_fp16))[name = tensor("op_30880_cast_fp16")]; + tensor var_30882_equation_0 = const()[name = tensor("op_30882_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30882_cast_fp16 = einsum(equation = var_30882_equation_0, values = (var_30736_cast_fp16, var_30844_cast_fp16))[name = tensor("op_30882_cast_fp16")]; + tensor var_30884_equation_0 = const()[name = tensor("op_30884_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30884_cast_fp16 = einsum(equation = var_30884_equation_0, values = (var_30740_cast_fp16, var_30845_cast_fp16))[name = tensor("op_30884_cast_fp16")]; + tensor var_30886_equation_0 = const()[name = tensor("op_30886_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30886_cast_fp16 = einsum(equation = var_30886_equation_0, values = (var_30744_cast_fp16, var_30846_cast_fp16))[name = tensor("op_30886_cast_fp16")]; + tensor input_439_interleave_0 = const()[name = tensor("input_439_interleave_0"), val = tensor(false)]; + tensor input_439_cast_fp16 = concat(axis = var_30355, interleave = input_439_interleave_0, values = (var_30848_cast_fp16, var_30850_cast_fp16, var_30852_cast_fp16, var_30854_cast_fp16, var_30856_cast_fp16, var_30858_cast_fp16, var_30860_cast_fp16, var_30862_cast_fp16, var_30864_cast_fp16, var_30866_cast_fp16, var_30868_cast_fp16, var_30870_cast_fp16, var_30872_cast_fp16, var_30874_cast_fp16, var_30876_cast_fp16, var_30878_cast_fp16, var_30880_cast_fp16, var_30882_cast_fp16, var_30884_cast_fp16, var_30886_cast_fp16))[name = tensor("input_439_cast_fp16")]; + tensor var_30892 = const()[name = tensor("op_30892"), val = tensor([1, 1])]; + tensor var_30894 = const()[name = tensor("op_30894"), val = tensor([1, 1])]; + tensor var_30896_pad_type_0 = const()[name = tensor("op_30896_pad_type_0"), val = tensor("custom")]; + tensor var_30896_pad_0 = const()[name = tensor("op_30896_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2598257216)))]; + tensor up_blocks_0_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2601534080)))]; + tensor var_30896_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_30894, groups = var_30355, pad = var_30896_pad_0, pad_type = var_30896_pad_type_0, strides = var_30892, weight = up_blocks_0_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16, x = input_439_cast_fp16)[name = tensor("op_30896_cast_fp16")]; + tensor inputs_207_cast_fp16 = add(x = var_30896_cast_fp16, y = inputs_205_cast_fp16)[name = tensor("inputs_207_cast_fp16")]; + tensor hidden_states_289_axes_0 = const()[name = tensor("hidden_states_289_axes_0"), val = tensor([1])]; + tensor hidden_states_289_gamma_0_to_fp16 = const()[name = tensor("hidden_states_289_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2601536704)))]; + tensor hidden_states_289_beta_0_to_fp16 = const()[name = tensor("hidden_states_289_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2601539328)))]; + tensor var_30906_to_fp16 = const()[name = tensor("op_30906_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_289_cast_fp16 = layer_norm(axes = hidden_states_289_axes_0, beta = hidden_states_289_beta_0_to_fp16, epsilon = var_30906_to_fp16, gamma = hidden_states_289_gamma_0_to_fp16, x = inputs_207_cast_fp16)[name = tensor("hidden_states_289_cast_fp16")]; + tensor var_30921 = const()[name = tensor("op_30921"), val = tensor([1, 1])]; + tensor var_30923 = const()[name = tensor("op_30923"), val = tensor([1, 1])]; + tensor q_139_pad_type_0 = const()[name = tensor("q_139_pad_type_0"), val = tensor("custom")]; + tensor q_139_pad_0 = const()[name = tensor("q_139_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2601541952)))]; + tensor q_139_cast_fp16 = conv(dilations = var_30923, groups = var_30355, pad = q_139_pad_0, pad_type = q_139_pad_type_0, strides = var_30921, weight = up_blocks_0_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16, x = hidden_states_289_cast_fp16)[name = tensor("q_139_cast_fp16")]; + tensor var_30927 = const()[name = tensor("op_30927"), val = tensor([1, 1])]; + tensor var_30929 = const()[name = tensor("op_30929"), val = tensor([1, 1])]; + tensor k_277_pad_type_0 = const()[name = tensor("k_277_pad_type_0"), val = tensor("custom")]; + tensor k_277_pad_0 = const()[name = tensor("k_277_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2604818816)))]; + tensor k_277_cast_fp16 = conv(dilations = var_30929, groups = var_30355, pad = k_277_pad_0, pad_type = k_277_pad_type_0, strides = var_30927, weight = up_blocks_0_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_277_cast_fp16")]; + tensor var_30933 = const()[name = tensor("op_30933"), val = tensor([1, 1])]; + tensor var_30935 = const()[name = tensor("op_30935"), val = tensor([1, 1])]; + tensor v_139_pad_type_0 = const()[name = tensor("v_139_pad_type_0"), val = tensor("custom")]; + tensor v_139_pad_0 = const()[name = tensor("v_139_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2610061760)))]; + tensor v_139_cast_fp16 = conv(dilations = var_30935, groups = var_30355, pad = v_139_pad_0, pad_type = v_139_pad_type_0, strides = var_30933, weight = up_blocks_0_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_139_cast_fp16")]; + tensor var_30939_begin_0 = const()[name = tensor("op_30939_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_30939_end_0 = const()[name = tensor("op_30939_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_30939_end_mask_0 = const()[name = tensor("op_30939_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30939_cast_fp16 = slice_by_index(begin = var_30939_begin_0, end = var_30939_end_0, end_mask = var_30939_end_mask_0, x = q_139_cast_fp16)[name = tensor("op_30939_cast_fp16")]; + tensor var_30943_begin_0 = const()[name = tensor("op_30943_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_30943_end_0 = const()[name = tensor("op_30943_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_30943_end_mask_0 = const()[name = tensor("op_30943_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30943_cast_fp16 = slice_by_index(begin = var_30943_begin_0, end = var_30943_end_0, end_mask = var_30943_end_mask_0, x = q_139_cast_fp16)[name = tensor("op_30943_cast_fp16")]; + tensor var_30947_begin_0 = const()[name = tensor("op_30947_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_30947_end_0 = const()[name = tensor("op_30947_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_30947_end_mask_0 = const()[name = tensor("op_30947_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30947_cast_fp16 = slice_by_index(begin = var_30947_begin_0, end = var_30947_end_0, end_mask = var_30947_end_mask_0, x = q_139_cast_fp16)[name = tensor("op_30947_cast_fp16")]; + tensor var_30951_begin_0 = const()[name = tensor("op_30951_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_30951_end_0 = const()[name = tensor("op_30951_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_30951_end_mask_0 = const()[name = tensor("op_30951_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30951_cast_fp16 = slice_by_index(begin = var_30951_begin_0, end = var_30951_end_0, end_mask = var_30951_end_mask_0, x = q_139_cast_fp16)[name = tensor("op_30951_cast_fp16")]; + tensor var_30955_begin_0 = const()[name = tensor("op_30955_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_30955_end_0 = const()[name = tensor("op_30955_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_30955_end_mask_0 = const()[name = tensor("op_30955_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30955_cast_fp16 = slice_by_index(begin = var_30955_begin_0, end = var_30955_end_0, end_mask = var_30955_end_mask_0, x = q_139_cast_fp16)[name = tensor("op_30955_cast_fp16")]; + tensor var_30959_begin_0 = const()[name = tensor("op_30959_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_30959_end_0 = const()[name = tensor("op_30959_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_30959_end_mask_0 = const()[name = tensor("op_30959_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30959_cast_fp16 = slice_by_index(begin = var_30959_begin_0, end = var_30959_end_0, end_mask = var_30959_end_mask_0, x = q_139_cast_fp16)[name = tensor("op_30959_cast_fp16")]; + tensor var_30963_begin_0 = const()[name = tensor("op_30963_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_30963_end_0 = const()[name = tensor("op_30963_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_30963_end_mask_0 = const()[name = tensor("op_30963_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30963_cast_fp16 = slice_by_index(begin = var_30963_begin_0, end = var_30963_end_0, end_mask = var_30963_end_mask_0, x = q_139_cast_fp16)[name = tensor("op_30963_cast_fp16")]; + tensor var_30967_begin_0 = const()[name = tensor("op_30967_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_30967_end_0 = const()[name = tensor("op_30967_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_30967_end_mask_0 = const()[name = tensor("op_30967_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30967_cast_fp16 = slice_by_index(begin = var_30967_begin_0, end = var_30967_end_0, end_mask = var_30967_end_mask_0, x = q_139_cast_fp16)[name = tensor("op_30967_cast_fp16")]; + tensor var_30971_begin_0 = const()[name = tensor("op_30971_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_30971_end_0 = const()[name = tensor("op_30971_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_30971_end_mask_0 = const()[name = tensor("op_30971_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30971_cast_fp16 = slice_by_index(begin = var_30971_begin_0, end = var_30971_end_0, end_mask = var_30971_end_mask_0, x = q_139_cast_fp16)[name = tensor("op_30971_cast_fp16")]; + tensor var_30975_begin_0 = const()[name = tensor("op_30975_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_30975_end_0 = const()[name = tensor("op_30975_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_30975_end_mask_0 = const()[name = tensor("op_30975_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30975_cast_fp16 = slice_by_index(begin = var_30975_begin_0, end = var_30975_end_0, end_mask = var_30975_end_mask_0, x = q_139_cast_fp16)[name = tensor("op_30975_cast_fp16")]; + tensor var_30979_begin_0 = const()[name = tensor("op_30979_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_30979_end_0 = const()[name = tensor("op_30979_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_30979_end_mask_0 = const()[name = tensor("op_30979_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30979_cast_fp16 = slice_by_index(begin = var_30979_begin_0, end = var_30979_end_0, end_mask = var_30979_end_mask_0, x = q_139_cast_fp16)[name = tensor("op_30979_cast_fp16")]; + tensor var_30983_begin_0 = const()[name = tensor("op_30983_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_30983_end_0 = const()[name = tensor("op_30983_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_30983_end_mask_0 = const()[name = tensor("op_30983_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30983_cast_fp16 = slice_by_index(begin = var_30983_begin_0, end = var_30983_end_0, end_mask = var_30983_end_mask_0, x = q_139_cast_fp16)[name = tensor("op_30983_cast_fp16")]; + tensor var_30987_begin_0 = const()[name = tensor("op_30987_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_30987_end_0 = const()[name = tensor("op_30987_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_30987_end_mask_0 = const()[name = tensor("op_30987_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30987_cast_fp16 = slice_by_index(begin = var_30987_begin_0, end = var_30987_end_0, end_mask = var_30987_end_mask_0, x = q_139_cast_fp16)[name = tensor("op_30987_cast_fp16")]; + tensor var_30991_begin_0 = const()[name = tensor("op_30991_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_30991_end_0 = const()[name = tensor("op_30991_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_30991_end_mask_0 = const()[name = tensor("op_30991_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30991_cast_fp16 = slice_by_index(begin = var_30991_begin_0, end = var_30991_end_0, end_mask = var_30991_end_mask_0, x = q_139_cast_fp16)[name = tensor("op_30991_cast_fp16")]; + tensor var_30995_begin_0 = const()[name = tensor("op_30995_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_30995_end_0 = const()[name = tensor("op_30995_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_30995_end_mask_0 = const()[name = tensor("op_30995_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30995_cast_fp16 = slice_by_index(begin = var_30995_begin_0, end = var_30995_end_0, end_mask = var_30995_end_mask_0, x = q_139_cast_fp16)[name = tensor("op_30995_cast_fp16")]; + tensor var_30999_begin_0 = const()[name = tensor("op_30999_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_30999_end_0 = const()[name = tensor("op_30999_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_30999_end_mask_0 = const()[name = tensor("op_30999_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30999_cast_fp16 = slice_by_index(begin = var_30999_begin_0, end = var_30999_end_0, end_mask = var_30999_end_mask_0, x = q_139_cast_fp16)[name = tensor("op_30999_cast_fp16")]; + tensor var_31003_begin_0 = const()[name = tensor("op_31003_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_31003_end_0 = const()[name = tensor("op_31003_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_31003_end_mask_0 = const()[name = tensor("op_31003_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31003_cast_fp16 = slice_by_index(begin = var_31003_begin_0, end = var_31003_end_0, end_mask = var_31003_end_mask_0, x = q_139_cast_fp16)[name = tensor("op_31003_cast_fp16")]; + tensor var_31007_begin_0 = const()[name = tensor("op_31007_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_31007_end_0 = const()[name = tensor("op_31007_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_31007_end_mask_0 = const()[name = tensor("op_31007_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31007_cast_fp16 = slice_by_index(begin = var_31007_begin_0, end = var_31007_end_0, end_mask = var_31007_end_mask_0, x = q_139_cast_fp16)[name = tensor("op_31007_cast_fp16")]; + tensor var_31011_begin_0 = const()[name = tensor("op_31011_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_31011_end_0 = const()[name = tensor("op_31011_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_31011_end_mask_0 = const()[name = tensor("op_31011_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31011_cast_fp16 = slice_by_index(begin = var_31011_begin_0, end = var_31011_end_0, end_mask = var_31011_end_mask_0, x = q_139_cast_fp16)[name = tensor("op_31011_cast_fp16")]; + tensor var_31015_begin_0 = const()[name = tensor("op_31015_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_31015_end_0 = const()[name = tensor("op_31015_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_31015_end_mask_0 = const()[name = tensor("op_31015_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31015_cast_fp16 = slice_by_index(begin = var_31015_begin_0, end = var_31015_end_0, end_mask = var_31015_end_mask_0, x = q_139_cast_fp16)[name = tensor("op_31015_cast_fp16")]; + tensor k_279_perm_0 = const()[name = tensor("k_279_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_31022_begin_0 = const()[name = tensor("op_31022_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_31022_end_0 = const()[name = tensor("op_31022_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_31022_end_mask_0 = const()[name = tensor("op_31022_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_279_cast_fp16 = transpose(perm = k_279_perm_0, x = k_277_cast_fp16)[name = tensor("transpose_70")]; + tensor var_31022_cast_fp16 = slice_by_index(begin = var_31022_begin_0, end = var_31022_end_0, end_mask = var_31022_end_mask_0, x = k_279_cast_fp16)[name = tensor("op_31022_cast_fp16")]; + tensor var_31026_begin_0 = const()[name = tensor("op_31026_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_31026_end_0 = const()[name = tensor("op_31026_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_31026_end_mask_0 = const()[name = tensor("op_31026_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31026_cast_fp16 = slice_by_index(begin = var_31026_begin_0, end = var_31026_end_0, end_mask = var_31026_end_mask_0, x = k_279_cast_fp16)[name = tensor("op_31026_cast_fp16")]; + tensor var_31030_begin_0 = const()[name = tensor("op_31030_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_31030_end_0 = const()[name = tensor("op_31030_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_31030_end_mask_0 = const()[name = tensor("op_31030_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31030_cast_fp16 = slice_by_index(begin = var_31030_begin_0, end = var_31030_end_0, end_mask = var_31030_end_mask_0, x = k_279_cast_fp16)[name = tensor("op_31030_cast_fp16")]; + tensor var_31034_begin_0 = const()[name = tensor("op_31034_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_31034_end_0 = const()[name = tensor("op_31034_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_31034_end_mask_0 = const()[name = tensor("op_31034_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31034_cast_fp16 = slice_by_index(begin = var_31034_begin_0, end = var_31034_end_0, end_mask = var_31034_end_mask_0, x = k_279_cast_fp16)[name = tensor("op_31034_cast_fp16")]; + tensor var_31038_begin_0 = const()[name = tensor("op_31038_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_31038_end_0 = const()[name = tensor("op_31038_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_31038_end_mask_0 = const()[name = tensor("op_31038_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31038_cast_fp16 = slice_by_index(begin = var_31038_begin_0, end = var_31038_end_0, end_mask = var_31038_end_mask_0, x = k_279_cast_fp16)[name = tensor("op_31038_cast_fp16")]; + tensor var_31042_begin_0 = const()[name = tensor("op_31042_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_31042_end_0 = const()[name = tensor("op_31042_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_31042_end_mask_0 = const()[name = tensor("op_31042_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31042_cast_fp16 = slice_by_index(begin = var_31042_begin_0, end = var_31042_end_0, end_mask = var_31042_end_mask_0, x = k_279_cast_fp16)[name = tensor("op_31042_cast_fp16")]; + tensor var_31046_begin_0 = const()[name = tensor("op_31046_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_31046_end_0 = const()[name = tensor("op_31046_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_31046_end_mask_0 = const()[name = tensor("op_31046_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31046_cast_fp16 = slice_by_index(begin = var_31046_begin_0, end = var_31046_end_0, end_mask = var_31046_end_mask_0, x = k_279_cast_fp16)[name = tensor("op_31046_cast_fp16")]; + tensor var_31050_begin_0 = const()[name = tensor("op_31050_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_31050_end_0 = const()[name = tensor("op_31050_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_31050_end_mask_0 = const()[name = tensor("op_31050_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31050_cast_fp16 = slice_by_index(begin = var_31050_begin_0, end = var_31050_end_0, end_mask = var_31050_end_mask_0, x = k_279_cast_fp16)[name = tensor("op_31050_cast_fp16")]; + tensor var_31054_begin_0 = const()[name = tensor("op_31054_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_31054_end_0 = const()[name = tensor("op_31054_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_31054_end_mask_0 = const()[name = tensor("op_31054_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31054_cast_fp16 = slice_by_index(begin = var_31054_begin_0, end = var_31054_end_0, end_mask = var_31054_end_mask_0, x = k_279_cast_fp16)[name = tensor("op_31054_cast_fp16")]; + tensor var_31058_begin_0 = const()[name = tensor("op_31058_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_31058_end_0 = const()[name = tensor("op_31058_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_31058_end_mask_0 = const()[name = tensor("op_31058_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31058_cast_fp16 = slice_by_index(begin = var_31058_begin_0, end = var_31058_end_0, end_mask = var_31058_end_mask_0, x = k_279_cast_fp16)[name = tensor("op_31058_cast_fp16")]; + tensor var_31062_begin_0 = const()[name = tensor("op_31062_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_31062_end_0 = const()[name = tensor("op_31062_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_31062_end_mask_0 = const()[name = tensor("op_31062_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31062_cast_fp16 = slice_by_index(begin = var_31062_begin_0, end = var_31062_end_0, end_mask = var_31062_end_mask_0, x = k_279_cast_fp16)[name = tensor("op_31062_cast_fp16")]; + tensor var_31066_begin_0 = const()[name = tensor("op_31066_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_31066_end_0 = const()[name = tensor("op_31066_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_31066_end_mask_0 = const()[name = tensor("op_31066_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31066_cast_fp16 = slice_by_index(begin = var_31066_begin_0, end = var_31066_end_0, end_mask = var_31066_end_mask_0, x = k_279_cast_fp16)[name = tensor("op_31066_cast_fp16")]; + tensor var_31070_begin_0 = const()[name = tensor("op_31070_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_31070_end_0 = const()[name = tensor("op_31070_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_31070_end_mask_0 = const()[name = tensor("op_31070_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31070_cast_fp16 = slice_by_index(begin = var_31070_begin_0, end = var_31070_end_0, end_mask = var_31070_end_mask_0, x = k_279_cast_fp16)[name = tensor("op_31070_cast_fp16")]; + tensor var_31074_begin_0 = const()[name = tensor("op_31074_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_31074_end_0 = const()[name = tensor("op_31074_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_31074_end_mask_0 = const()[name = tensor("op_31074_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31074_cast_fp16 = slice_by_index(begin = var_31074_begin_0, end = var_31074_end_0, end_mask = var_31074_end_mask_0, x = k_279_cast_fp16)[name = tensor("op_31074_cast_fp16")]; + tensor var_31078_begin_0 = const()[name = tensor("op_31078_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_31078_end_0 = const()[name = tensor("op_31078_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_31078_end_mask_0 = const()[name = tensor("op_31078_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31078_cast_fp16 = slice_by_index(begin = var_31078_begin_0, end = var_31078_end_0, end_mask = var_31078_end_mask_0, x = k_279_cast_fp16)[name = tensor("op_31078_cast_fp16")]; + tensor var_31082_begin_0 = const()[name = tensor("op_31082_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_31082_end_0 = const()[name = tensor("op_31082_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_31082_end_mask_0 = const()[name = tensor("op_31082_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31082_cast_fp16 = slice_by_index(begin = var_31082_begin_0, end = var_31082_end_0, end_mask = var_31082_end_mask_0, x = k_279_cast_fp16)[name = tensor("op_31082_cast_fp16")]; + tensor var_31086_begin_0 = const()[name = tensor("op_31086_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_31086_end_0 = const()[name = tensor("op_31086_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_31086_end_mask_0 = const()[name = tensor("op_31086_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31086_cast_fp16 = slice_by_index(begin = var_31086_begin_0, end = var_31086_end_0, end_mask = var_31086_end_mask_0, x = k_279_cast_fp16)[name = tensor("op_31086_cast_fp16")]; + tensor var_31090_begin_0 = const()[name = tensor("op_31090_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_31090_end_0 = const()[name = tensor("op_31090_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_31090_end_mask_0 = const()[name = tensor("op_31090_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31090_cast_fp16 = slice_by_index(begin = var_31090_begin_0, end = var_31090_end_0, end_mask = var_31090_end_mask_0, x = k_279_cast_fp16)[name = tensor("op_31090_cast_fp16")]; + tensor var_31094_begin_0 = const()[name = tensor("op_31094_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_31094_end_0 = const()[name = tensor("op_31094_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_31094_end_mask_0 = const()[name = tensor("op_31094_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31094_cast_fp16 = slice_by_index(begin = var_31094_begin_0, end = var_31094_end_0, end_mask = var_31094_end_mask_0, x = k_279_cast_fp16)[name = tensor("op_31094_cast_fp16")]; + tensor var_31098_begin_0 = const()[name = tensor("op_31098_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_31098_end_0 = const()[name = tensor("op_31098_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_31098_end_mask_0 = const()[name = tensor("op_31098_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31098_cast_fp16 = slice_by_index(begin = var_31098_begin_0, end = var_31098_end_0, end_mask = var_31098_end_mask_0, x = k_279_cast_fp16)[name = tensor("op_31098_cast_fp16")]; + tensor var_31100_begin_0 = const()[name = tensor("op_31100_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_31100_end_0 = const()[name = tensor("op_31100_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_31100_end_mask_0 = const()[name = tensor("op_31100_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31100_cast_fp16 = slice_by_index(begin = var_31100_begin_0, end = var_31100_end_0, end_mask = var_31100_end_mask_0, x = v_139_cast_fp16)[name = tensor("op_31100_cast_fp16")]; + tensor var_31104_begin_0 = const()[name = tensor("op_31104_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_31104_end_0 = const()[name = tensor("op_31104_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_31104_end_mask_0 = const()[name = tensor("op_31104_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31104_cast_fp16 = slice_by_index(begin = var_31104_begin_0, end = var_31104_end_0, end_mask = var_31104_end_mask_0, x = v_139_cast_fp16)[name = tensor("op_31104_cast_fp16")]; + tensor var_31108_begin_0 = const()[name = tensor("op_31108_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_31108_end_0 = const()[name = tensor("op_31108_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_31108_end_mask_0 = const()[name = tensor("op_31108_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31108_cast_fp16 = slice_by_index(begin = var_31108_begin_0, end = var_31108_end_0, end_mask = var_31108_end_mask_0, x = v_139_cast_fp16)[name = tensor("op_31108_cast_fp16")]; + tensor var_31112_begin_0 = const()[name = tensor("op_31112_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_31112_end_0 = const()[name = tensor("op_31112_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_31112_end_mask_0 = const()[name = tensor("op_31112_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31112_cast_fp16 = slice_by_index(begin = var_31112_begin_0, end = var_31112_end_0, end_mask = var_31112_end_mask_0, x = v_139_cast_fp16)[name = tensor("op_31112_cast_fp16")]; + tensor var_31116_begin_0 = const()[name = tensor("op_31116_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_31116_end_0 = const()[name = tensor("op_31116_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_31116_end_mask_0 = const()[name = tensor("op_31116_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31116_cast_fp16 = slice_by_index(begin = var_31116_begin_0, end = var_31116_end_0, end_mask = var_31116_end_mask_0, x = v_139_cast_fp16)[name = tensor("op_31116_cast_fp16")]; + tensor var_31120_begin_0 = const()[name = tensor("op_31120_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_31120_end_0 = const()[name = tensor("op_31120_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_31120_end_mask_0 = const()[name = tensor("op_31120_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31120_cast_fp16 = slice_by_index(begin = var_31120_begin_0, end = var_31120_end_0, end_mask = var_31120_end_mask_0, x = v_139_cast_fp16)[name = tensor("op_31120_cast_fp16")]; + tensor var_31124_begin_0 = const()[name = tensor("op_31124_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_31124_end_0 = const()[name = tensor("op_31124_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_31124_end_mask_0 = const()[name = tensor("op_31124_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31124_cast_fp16 = slice_by_index(begin = var_31124_begin_0, end = var_31124_end_0, end_mask = var_31124_end_mask_0, x = v_139_cast_fp16)[name = tensor("op_31124_cast_fp16")]; + tensor var_31128_begin_0 = const()[name = tensor("op_31128_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_31128_end_0 = const()[name = tensor("op_31128_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_31128_end_mask_0 = const()[name = tensor("op_31128_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31128_cast_fp16 = slice_by_index(begin = var_31128_begin_0, end = var_31128_end_0, end_mask = var_31128_end_mask_0, x = v_139_cast_fp16)[name = tensor("op_31128_cast_fp16")]; + tensor var_31132_begin_0 = const()[name = tensor("op_31132_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_31132_end_0 = const()[name = tensor("op_31132_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_31132_end_mask_0 = const()[name = tensor("op_31132_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31132_cast_fp16 = slice_by_index(begin = var_31132_begin_0, end = var_31132_end_0, end_mask = var_31132_end_mask_0, x = v_139_cast_fp16)[name = tensor("op_31132_cast_fp16")]; + tensor var_31136_begin_0 = const()[name = tensor("op_31136_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_31136_end_0 = const()[name = tensor("op_31136_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_31136_end_mask_0 = const()[name = tensor("op_31136_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31136_cast_fp16 = slice_by_index(begin = var_31136_begin_0, end = var_31136_end_0, end_mask = var_31136_end_mask_0, x = v_139_cast_fp16)[name = tensor("op_31136_cast_fp16")]; + tensor var_31140_begin_0 = const()[name = tensor("op_31140_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_31140_end_0 = const()[name = tensor("op_31140_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_31140_end_mask_0 = const()[name = tensor("op_31140_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31140_cast_fp16 = slice_by_index(begin = var_31140_begin_0, end = var_31140_end_0, end_mask = var_31140_end_mask_0, x = v_139_cast_fp16)[name = tensor("op_31140_cast_fp16")]; + tensor var_31144_begin_0 = const()[name = tensor("op_31144_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_31144_end_0 = const()[name = tensor("op_31144_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_31144_end_mask_0 = const()[name = tensor("op_31144_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31144_cast_fp16 = slice_by_index(begin = var_31144_begin_0, end = var_31144_end_0, end_mask = var_31144_end_mask_0, x = v_139_cast_fp16)[name = tensor("op_31144_cast_fp16")]; + tensor var_31148_begin_0 = const()[name = tensor("op_31148_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_31148_end_0 = const()[name = tensor("op_31148_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_31148_end_mask_0 = const()[name = tensor("op_31148_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31148_cast_fp16 = slice_by_index(begin = var_31148_begin_0, end = var_31148_end_0, end_mask = var_31148_end_mask_0, x = v_139_cast_fp16)[name = tensor("op_31148_cast_fp16")]; + tensor var_31152_begin_0 = const()[name = tensor("op_31152_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_31152_end_0 = const()[name = tensor("op_31152_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_31152_end_mask_0 = const()[name = tensor("op_31152_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31152_cast_fp16 = slice_by_index(begin = var_31152_begin_0, end = var_31152_end_0, end_mask = var_31152_end_mask_0, x = v_139_cast_fp16)[name = tensor("op_31152_cast_fp16")]; + tensor var_31156_begin_0 = const()[name = tensor("op_31156_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_31156_end_0 = const()[name = tensor("op_31156_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_31156_end_mask_0 = const()[name = tensor("op_31156_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31156_cast_fp16 = slice_by_index(begin = var_31156_begin_0, end = var_31156_end_0, end_mask = var_31156_end_mask_0, x = v_139_cast_fp16)[name = tensor("op_31156_cast_fp16")]; + tensor var_31160_begin_0 = const()[name = tensor("op_31160_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_31160_end_0 = const()[name = tensor("op_31160_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_31160_end_mask_0 = const()[name = tensor("op_31160_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31160_cast_fp16 = slice_by_index(begin = var_31160_begin_0, end = var_31160_end_0, end_mask = var_31160_end_mask_0, x = v_139_cast_fp16)[name = tensor("op_31160_cast_fp16")]; + tensor var_31164_begin_0 = const()[name = tensor("op_31164_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_31164_end_0 = const()[name = tensor("op_31164_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_31164_end_mask_0 = const()[name = tensor("op_31164_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31164_cast_fp16 = slice_by_index(begin = var_31164_begin_0, end = var_31164_end_0, end_mask = var_31164_end_mask_0, x = v_139_cast_fp16)[name = tensor("op_31164_cast_fp16")]; + tensor var_31168_begin_0 = const()[name = tensor("op_31168_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_31168_end_0 = const()[name = tensor("op_31168_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_31168_end_mask_0 = const()[name = tensor("op_31168_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31168_cast_fp16 = slice_by_index(begin = var_31168_begin_0, end = var_31168_end_0, end_mask = var_31168_end_mask_0, x = v_139_cast_fp16)[name = tensor("op_31168_cast_fp16")]; + tensor var_31172_begin_0 = const()[name = tensor("op_31172_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_31172_end_0 = const()[name = tensor("op_31172_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_31172_end_mask_0 = const()[name = tensor("op_31172_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31172_cast_fp16 = slice_by_index(begin = var_31172_begin_0, end = var_31172_end_0, end_mask = var_31172_end_mask_0, x = v_139_cast_fp16)[name = tensor("op_31172_cast_fp16")]; + tensor var_31176_begin_0 = const()[name = tensor("op_31176_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_31176_end_0 = const()[name = tensor("op_31176_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_31176_end_mask_0 = const()[name = tensor("op_31176_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31176_cast_fp16 = slice_by_index(begin = var_31176_begin_0, end = var_31176_end_0, end_mask = var_31176_end_mask_0, x = v_139_cast_fp16)[name = tensor("op_31176_cast_fp16")]; + tensor var_31180_equation_0 = const()[name = tensor("op_31180_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_31180_cast_fp16 = einsum(equation = var_31180_equation_0, values = (var_31022_cast_fp16, var_30939_cast_fp16))[name = tensor("op_31180_cast_fp16")]; + tensor var_31181_to_fp16 = const()[name = tensor("op_31181_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2601_cast_fp16 = mul(x = var_31180_cast_fp16, y = var_31181_to_fp16)[name = tensor("aw_2601_cast_fp16")]; + tensor var_31184_equation_0 = const()[name = tensor("op_31184_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_31184_cast_fp16 = einsum(equation = var_31184_equation_0, values = (var_31026_cast_fp16, var_30943_cast_fp16))[name = tensor("op_31184_cast_fp16")]; + tensor var_31185_to_fp16 = const()[name = tensor("op_31185_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2603_cast_fp16 = mul(x = var_31184_cast_fp16, y = var_31185_to_fp16)[name = tensor("aw_2603_cast_fp16")]; + tensor var_31188_equation_0 = const()[name = tensor("op_31188_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_31188_cast_fp16 = einsum(equation = var_31188_equation_0, values = (var_31030_cast_fp16, var_30947_cast_fp16))[name = tensor("op_31188_cast_fp16")]; + tensor var_31189_to_fp16 = const()[name = tensor("op_31189_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2605_cast_fp16 = mul(x = var_31188_cast_fp16, y = var_31189_to_fp16)[name = tensor("aw_2605_cast_fp16")]; + tensor var_31192_equation_0 = const()[name = tensor("op_31192_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_31192_cast_fp16 = einsum(equation = var_31192_equation_0, values = (var_31034_cast_fp16, var_30951_cast_fp16))[name = tensor("op_31192_cast_fp16")]; + tensor var_31193_to_fp16 = const()[name = tensor("op_31193_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2607_cast_fp16 = mul(x = var_31192_cast_fp16, y = var_31193_to_fp16)[name = tensor("aw_2607_cast_fp16")]; + tensor var_31196_equation_0 = const()[name = tensor("op_31196_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_31196_cast_fp16 = einsum(equation = var_31196_equation_0, values = (var_31038_cast_fp16, var_30955_cast_fp16))[name = tensor("op_31196_cast_fp16")]; + tensor var_31197_to_fp16 = const()[name = tensor("op_31197_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2609_cast_fp16 = mul(x = var_31196_cast_fp16, y = var_31197_to_fp16)[name = tensor("aw_2609_cast_fp16")]; + tensor var_31200_equation_0 = const()[name = tensor("op_31200_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_31200_cast_fp16 = einsum(equation = var_31200_equation_0, values = (var_31042_cast_fp16, var_30959_cast_fp16))[name = tensor("op_31200_cast_fp16")]; + tensor var_31201_to_fp16 = const()[name = tensor("op_31201_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2611_cast_fp16 = mul(x = var_31200_cast_fp16, y = var_31201_to_fp16)[name = tensor("aw_2611_cast_fp16")]; + tensor var_31204_equation_0 = const()[name = tensor("op_31204_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_31204_cast_fp16 = einsum(equation = var_31204_equation_0, values = (var_31046_cast_fp16, var_30963_cast_fp16))[name = tensor("op_31204_cast_fp16")]; + tensor var_31205_to_fp16 = const()[name = tensor("op_31205_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2613_cast_fp16 = mul(x = var_31204_cast_fp16, y = var_31205_to_fp16)[name = tensor("aw_2613_cast_fp16")]; + tensor var_31208_equation_0 = const()[name = tensor("op_31208_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_31208_cast_fp16 = einsum(equation = var_31208_equation_0, values = (var_31050_cast_fp16, var_30967_cast_fp16))[name = tensor("op_31208_cast_fp16")]; + tensor var_31209_to_fp16 = const()[name = tensor("op_31209_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2615_cast_fp16 = mul(x = var_31208_cast_fp16, y = var_31209_to_fp16)[name = tensor("aw_2615_cast_fp16")]; + tensor var_31212_equation_0 = const()[name = tensor("op_31212_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_31212_cast_fp16 = einsum(equation = var_31212_equation_0, values = (var_31054_cast_fp16, var_30971_cast_fp16))[name = tensor("op_31212_cast_fp16")]; + tensor var_31213_to_fp16 = const()[name = tensor("op_31213_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2617_cast_fp16 = mul(x = var_31212_cast_fp16, y = var_31213_to_fp16)[name = tensor("aw_2617_cast_fp16")]; + tensor var_31216_equation_0 = const()[name = tensor("op_31216_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_31216_cast_fp16 = einsum(equation = var_31216_equation_0, values = (var_31058_cast_fp16, var_30975_cast_fp16))[name = tensor("op_31216_cast_fp16")]; + tensor var_31217_to_fp16 = const()[name = tensor("op_31217_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2619_cast_fp16 = mul(x = var_31216_cast_fp16, y = var_31217_to_fp16)[name = tensor("aw_2619_cast_fp16")]; + tensor var_31220_equation_0 = const()[name = tensor("op_31220_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_31220_cast_fp16 = einsum(equation = var_31220_equation_0, values = (var_31062_cast_fp16, var_30979_cast_fp16))[name = tensor("op_31220_cast_fp16")]; + tensor var_31221_to_fp16 = const()[name = tensor("op_31221_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2621_cast_fp16 = mul(x = var_31220_cast_fp16, y = var_31221_to_fp16)[name = tensor("aw_2621_cast_fp16")]; + tensor var_31224_equation_0 = const()[name = tensor("op_31224_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_31224_cast_fp16 = einsum(equation = var_31224_equation_0, values = (var_31066_cast_fp16, var_30983_cast_fp16))[name = tensor("op_31224_cast_fp16")]; + tensor var_31225_to_fp16 = const()[name = tensor("op_31225_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2623_cast_fp16 = mul(x = var_31224_cast_fp16, y = var_31225_to_fp16)[name = tensor("aw_2623_cast_fp16")]; + tensor var_31228_equation_0 = const()[name = tensor("op_31228_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_31228_cast_fp16 = einsum(equation = var_31228_equation_0, values = (var_31070_cast_fp16, var_30987_cast_fp16))[name = tensor("op_31228_cast_fp16")]; + tensor var_31229_to_fp16 = const()[name = tensor("op_31229_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2625_cast_fp16 = mul(x = var_31228_cast_fp16, y = var_31229_to_fp16)[name = tensor("aw_2625_cast_fp16")]; + tensor var_31232_equation_0 = const()[name = tensor("op_31232_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_31232_cast_fp16 = einsum(equation = var_31232_equation_0, values = (var_31074_cast_fp16, var_30991_cast_fp16))[name = tensor("op_31232_cast_fp16")]; + tensor var_31233_to_fp16 = const()[name = tensor("op_31233_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2627_cast_fp16 = mul(x = var_31232_cast_fp16, y = var_31233_to_fp16)[name = tensor("aw_2627_cast_fp16")]; + tensor var_31236_equation_0 = const()[name = tensor("op_31236_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_31236_cast_fp16 = einsum(equation = var_31236_equation_0, values = (var_31078_cast_fp16, var_30995_cast_fp16))[name = tensor("op_31236_cast_fp16")]; + tensor var_31237_to_fp16 = const()[name = tensor("op_31237_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2629_cast_fp16 = mul(x = var_31236_cast_fp16, y = var_31237_to_fp16)[name = tensor("aw_2629_cast_fp16")]; + tensor var_31240_equation_0 = const()[name = tensor("op_31240_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_31240_cast_fp16 = einsum(equation = var_31240_equation_0, values = (var_31082_cast_fp16, var_30999_cast_fp16))[name = tensor("op_31240_cast_fp16")]; + tensor var_31241_to_fp16 = const()[name = tensor("op_31241_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2631_cast_fp16 = mul(x = var_31240_cast_fp16, y = var_31241_to_fp16)[name = tensor("aw_2631_cast_fp16")]; + tensor var_31244_equation_0 = const()[name = tensor("op_31244_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_31244_cast_fp16 = einsum(equation = var_31244_equation_0, values = (var_31086_cast_fp16, var_31003_cast_fp16))[name = tensor("op_31244_cast_fp16")]; + tensor var_31245_to_fp16 = const()[name = tensor("op_31245_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2633_cast_fp16 = mul(x = var_31244_cast_fp16, y = var_31245_to_fp16)[name = tensor("aw_2633_cast_fp16")]; + tensor var_31248_equation_0 = const()[name = tensor("op_31248_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_31248_cast_fp16 = einsum(equation = var_31248_equation_0, values = (var_31090_cast_fp16, var_31007_cast_fp16))[name = tensor("op_31248_cast_fp16")]; + tensor var_31249_to_fp16 = const()[name = tensor("op_31249_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2635_cast_fp16 = mul(x = var_31248_cast_fp16, y = var_31249_to_fp16)[name = tensor("aw_2635_cast_fp16")]; + tensor var_31252_equation_0 = const()[name = tensor("op_31252_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_31252_cast_fp16 = einsum(equation = var_31252_equation_0, values = (var_31094_cast_fp16, var_31011_cast_fp16))[name = tensor("op_31252_cast_fp16")]; + tensor var_31253_to_fp16 = const()[name = tensor("op_31253_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2637_cast_fp16 = mul(x = var_31252_cast_fp16, y = var_31253_to_fp16)[name = tensor("aw_2637_cast_fp16")]; + tensor var_31256_equation_0 = const()[name = tensor("op_31256_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_31256_cast_fp16 = einsum(equation = var_31256_equation_0, values = (var_31098_cast_fp16, var_31015_cast_fp16))[name = tensor("op_31256_cast_fp16")]; + tensor var_31257_to_fp16 = const()[name = tensor("op_31257_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2639_cast_fp16 = mul(x = var_31256_cast_fp16, y = var_31257_to_fp16)[name = tensor("aw_2639_cast_fp16")]; + tensor var_31259_cast_fp16 = softmax(axis = var_30355, x = aw_2601_cast_fp16)[name = tensor("op_31259_cast_fp16")]; + tensor var_31260_cast_fp16 = softmax(axis = var_30355, x = aw_2603_cast_fp16)[name = tensor("op_31260_cast_fp16")]; + tensor var_31261_cast_fp16 = softmax(axis = var_30355, x = aw_2605_cast_fp16)[name = tensor("op_31261_cast_fp16")]; + tensor var_31262_cast_fp16 = softmax(axis = var_30355, x = aw_2607_cast_fp16)[name = tensor("op_31262_cast_fp16")]; + tensor var_31263_cast_fp16 = softmax(axis = var_30355, x = aw_2609_cast_fp16)[name = tensor("op_31263_cast_fp16")]; + tensor var_31264_cast_fp16 = softmax(axis = var_30355, x = aw_2611_cast_fp16)[name = tensor("op_31264_cast_fp16")]; + tensor var_31265_cast_fp16 = softmax(axis = var_30355, x = aw_2613_cast_fp16)[name = tensor("op_31265_cast_fp16")]; + tensor var_31266_cast_fp16 = softmax(axis = var_30355, x = aw_2615_cast_fp16)[name = tensor("op_31266_cast_fp16")]; + tensor var_31267_cast_fp16 = softmax(axis = var_30355, x = aw_2617_cast_fp16)[name = tensor("op_31267_cast_fp16")]; + tensor var_31268_cast_fp16 = softmax(axis = var_30355, x = aw_2619_cast_fp16)[name = tensor("op_31268_cast_fp16")]; + tensor var_31269_cast_fp16 = softmax(axis = var_30355, x = aw_2621_cast_fp16)[name = tensor("op_31269_cast_fp16")]; + tensor var_31270_cast_fp16 = softmax(axis = var_30355, x = aw_2623_cast_fp16)[name = tensor("op_31270_cast_fp16")]; + tensor var_31271_cast_fp16 = softmax(axis = var_30355, x = aw_2625_cast_fp16)[name = tensor("op_31271_cast_fp16")]; + tensor var_31272_cast_fp16 = softmax(axis = var_30355, x = aw_2627_cast_fp16)[name = tensor("op_31272_cast_fp16")]; + tensor var_31273_cast_fp16 = softmax(axis = var_30355, x = aw_2629_cast_fp16)[name = tensor("op_31273_cast_fp16")]; + tensor var_31274_cast_fp16 = softmax(axis = var_30355, x = aw_2631_cast_fp16)[name = tensor("op_31274_cast_fp16")]; + tensor var_31275_cast_fp16 = softmax(axis = var_30355, x = aw_2633_cast_fp16)[name = tensor("op_31275_cast_fp16")]; + tensor var_31276_cast_fp16 = softmax(axis = var_30355, x = aw_2635_cast_fp16)[name = tensor("op_31276_cast_fp16")]; + tensor var_31277_cast_fp16 = softmax(axis = var_30355, x = aw_2637_cast_fp16)[name = tensor("op_31277_cast_fp16")]; + tensor var_31278_cast_fp16 = softmax(axis = var_30355, x = aw_2639_cast_fp16)[name = tensor("op_31278_cast_fp16")]; + tensor var_31280_equation_0 = const()[name = tensor("op_31280_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31280_cast_fp16 = einsum(equation = var_31280_equation_0, values = (var_31100_cast_fp16, var_31259_cast_fp16))[name = tensor("op_31280_cast_fp16")]; + tensor var_31282_equation_0 = const()[name = tensor("op_31282_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31282_cast_fp16 = einsum(equation = var_31282_equation_0, values = (var_31104_cast_fp16, var_31260_cast_fp16))[name = tensor("op_31282_cast_fp16")]; + tensor var_31284_equation_0 = const()[name = tensor("op_31284_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31284_cast_fp16 = einsum(equation = var_31284_equation_0, values = (var_31108_cast_fp16, var_31261_cast_fp16))[name = tensor("op_31284_cast_fp16")]; + tensor var_31286_equation_0 = const()[name = tensor("op_31286_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31286_cast_fp16 = einsum(equation = var_31286_equation_0, values = (var_31112_cast_fp16, var_31262_cast_fp16))[name = tensor("op_31286_cast_fp16")]; + tensor var_31288_equation_0 = const()[name = tensor("op_31288_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31288_cast_fp16 = einsum(equation = var_31288_equation_0, values = (var_31116_cast_fp16, var_31263_cast_fp16))[name = tensor("op_31288_cast_fp16")]; + tensor var_31290_equation_0 = const()[name = tensor("op_31290_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31290_cast_fp16 = einsum(equation = var_31290_equation_0, values = (var_31120_cast_fp16, var_31264_cast_fp16))[name = tensor("op_31290_cast_fp16")]; + tensor var_31292_equation_0 = const()[name = tensor("op_31292_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31292_cast_fp16 = einsum(equation = var_31292_equation_0, values = (var_31124_cast_fp16, var_31265_cast_fp16))[name = tensor("op_31292_cast_fp16")]; + tensor var_31294_equation_0 = const()[name = tensor("op_31294_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31294_cast_fp16 = einsum(equation = var_31294_equation_0, values = (var_31128_cast_fp16, var_31266_cast_fp16))[name = tensor("op_31294_cast_fp16")]; + tensor var_31296_equation_0 = const()[name = tensor("op_31296_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31296_cast_fp16 = einsum(equation = var_31296_equation_0, values = (var_31132_cast_fp16, var_31267_cast_fp16))[name = tensor("op_31296_cast_fp16")]; + tensor var_31298_equation_0 = const()[name = tensor("op_31298_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31298_cast_fp16 = einsum(equation = var_31298_equation_0, values = (var_31136_cast_fp16, var_31268_cast_fp16))[name = tensor("op_31298_cast_fp16")]; + tensor var_31300_equation_0 = const()[name = tensor("op_31300_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31300_cast_fp16 = einsum(equation = var_31300_equation_0, values = (var_31140_cast_fp16, var_31269_cast_fp16))[name = tensor("op_31300_cast_fp16")]; + tensor var_31302_equation_0 = const()[name = tensor("op_31302_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31302_cast_fp16 = einsum(equation = var_31302_equation_0, values = (var_31144_cast_fp16, var_31270_cast_fp16))[name = tensor("op_31302_cast_fp16")]; + tensor var_31304_equation_0 = const()[name = tensor("op_31304_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31304_cast_fp16 = einsum(equation = var_31304_equation_0, values = (var_31148_cast_fp16, var_31271_cast_fp16))[name = tensor("op_31304_cast_fp16")]; + tensor var_31306_equation_0 = const()[name = tensor("op_31306_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31306_cast_fp16 = einsum(equation = var_31306_equation_0, values = (var_31152_cast_fp16, var_31272_cast_fp16))[name = tensor("op_31306_cast_fp16")]; + tensor var_31308_equation_0 = const()[name = tensor("op_31308_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31308_cast_fp16 = einsum(equation = var_31308_equation_0, values = (var_31156_cast_fp16, var_31273_cast_fp16))[name = tensor("op_31308_cast_fp16")]; + tensor var_31310_equation_0 = const()[name = tensor("op_31310_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31310_cast_fp16 = einsum(equation = var_31310_equation_0, values = (var_31160_cast_fp16, var_31274_cast_fp16))[name = tensor("op_31310_cast_fp16")]; + tensor var_31312_equation_0 = const()[name = tensor("op_31312_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31312_cast_fp16 = einsum(equation = var_31312_equation_0, values = (var_31164_cast_fp16, var_31275_cast_fp16))[name = tensor("op_31312_cast_fp16")]; + tensor var_31314_equation_0 = const()[name = tensor("op_31314_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31314_cast_fp16 = einsum(equation = var_31314_equation_0, values = (var_31168_cast_fp16, var_31276_cast_fp16))[name = tensor("op_31314_cast_fp16")]; + tensor var_31316_equation_0 = const()[name = tensor("op_31316_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31316_cast_fp16 = einsum(equation = var_31316_equation_0, values = (var_31172_cast_fp16, var_31277_cast_fp16))[name = tensor("op_31316_cast_fp16")]; + tensor var_31318_equation_0 = const()[name = tensor("op_31318_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31318_cast_fp16 = einsum(equation = var_31318_equation_0, values = (var_31176_cast_fp16, var_31278_cast_fp16))[name = tensor("op_31318_cast_fp16")]; + tensor input_441_interleave_0 = const()[name = tensor("input_441_interleave_0"), val = tensor(false)]; + tensor input_441_cast_fp16 = concat(axis = var_30355, interleave = input_441_interleave_0, values = (var_31280_cast_fp16, var_31282_cast_fp16, var_31284_cast_fp16, var_31286_cast_fp16, var_31288_cast_fp16, var_31290_cast_fp16, var_31292_cast_fp16, var_31294_cast_fp16, var_31296_cast_fp16, var_31298_cast_fp16, var_31300_cast_fp16, var_31302_cast_fp16, var_31304_cast_fp16, var_31306_cast_fp16, var_31308_cast_fp16, var_31310_cast_fp16, var_31312_cast_fp16, var_31314_cast_fp16, var_31316_cast_fp16, var_31318_cast_fp16))[name = tensor("input_441_cast_fp16")]; + tensor var_31324 = const()[name = tensor("op_31324"), val = tensor([1, 1])]; + tensor var_31326 = const()[name = tensor("op_31326"), val = tensor([1, 1])]; + tensor var_31328_pad_type_0 = const()[name = tensor("op_31328_pad_type_0"), val = tensor("custom")]; + tensor var_31328_pad_0 = const()[name = tensor("op_31328_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2615304704)))]; + tensor up_blocks_0_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2618581568)))]; + tensor var_31328_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_31326, groups = var_30355, pad = var_31328_pad_0, pad_type = var_31328_pad_type_0, strides = var_31324, weight = up_blocks_0_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16, x = input_441_cast_fp16)[name = tensor("op_31328_cast_fp16")]; + tensor inputs_209_cast_fp16 = add(x = var_31328_cast_fp16, y = inputs_207_cast_fp16)[name = tensor("inputs_209_cast_fp16")]; + tensor input_443_axes_0 = const()[name = tensor("input_443_axes_0"), val = tensor([1])]; + tensor input_443_gamma_0_to_fp16 = const()[name = tensor("input_443_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2618584192)))]; + tensor input_443_beta_0_to_fp16 = const()[name = tensor("input_443_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2618586816)))]; + tensor var_31338_to_fp16 = const()[name = tensor("op_31338_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_443_cast_fp16 = layer_norm(axes = input_443_axes_0, beta = input_443_beta_0_to_fp16, epsilon = var_31338_to_fp16, gamma = input_443_gamma_0_to_fp16, x = inputs_209_cast_fp16)[name = tensor("input_443_cast_fp16")]; + tensor var_31354 = const()[name = tensor("op_31354"), val = tensor([1, 1])]; + tensor var_31356 = const()[name = tensor("op_31356"), val = tensor([1, 1])]; + tensor var_31358_pad_type_0 = const()[name = tensor("op_31358_pad_type_0"), val = tensor("custom")]; + tensor var_31358_pad_0 = const()[name = tensor("op_31358_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2618589440)))]; + tensor up_blocks_0_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2644803904)))]; + tensor var_31358_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_31356, groups = var_30355, pad = var_31358_pad_0, pad_type = var_31358_pad_type_0, strides = var_31354, weight = up_blocks_0_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16, x = input_443_cast_fp16)[name = tensor("op_31358_cast_fp16")]; + tensor var_31359_split_sizes_0 = const()[name = tensor("op_31359_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_31359_axis_0 = const()[name = tensor("op_31359_axis_0"), val = tensor(1)]; + tensor var_31359_cast_fp16_0, tensor var_31359_cast_fp16_1 = split(axis = var_31359_axis_0, split_sizes = var_31359_split_sizes_0, x = var_31358_cast_fp16)[name = tensor("op_31359_cast_fp16")]; + tensor var_31361_mode_0 = const()[name = tensor("op_31361_mode_0"), val = tensor("EXACT")]; + tensor var_31361_cast_fp16 = gelu(mode = var_31361_mode_0, x = var_31359_cast_fp16_1)[name = tensor("op_31361_cast_fp16")]; + tensor input_445_cast_fp16 = mul(x = var_31359_cast_fp16_0, y = var_31361_cast_fp16)[name = tensor("input_445_cast_fp16")]; + tensor var_31365 = const()[name = tensor("op_31365"), val = tensor([1, 1])]; + tensor var_31367 = const()[name = tensor("op_31367"), val = tensor([1, 1])]; + tensor var_31369_pad_type_0 = const()[name = tensor("op_31369_pad_type_0"), val = tensor("custom")]; + tensor var_31369_pad_0 = const()[name = tensor("op_31369_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2644824448)))]; + tensor up_blocks_0_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2657931712)))]; + tensor var_31369_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_31367, groups = var_30355, pad = var_31369_pad_0, pad_type = var_31369_pad_type_0, strides = var_31365, weight = up_blocks_0_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16, x = input_445_cast_fp16)[name = tensor("op_31369_cast_fp16")]; + tensor inputs_211_cast_fp16 = add(x = var_31369_cast_fp16, y = inputs_209_cast_fp16)[name = tensor("inputs_211_cast_fp16")]; + tensor hidden_states_293_axes_0 = const()[name = tensor("hidden_states_293_axes_0"), val = tensor([1])]; + tensor hidden_states_293_gamma_0_to_fp16 = const()[name = tensor("hidden_states_293_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2657934336)))]; + tensor hidden_states_293_beta_0_to_fp16 = const()[name = tensor("hidden_states_293_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2657936960)))]; + tensor var_31385_to_fp16 = const()[name = tensor("op_31385_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_293_cast_fp16 = layer_norm(axes = hidden_states_293_axes_0, beta = hidden_states_293_beta_0_to_fp16, epsilon = var_31385_to_fp16, gamma = hidden_states_293_gamma_0_to_fp16, x = inputs_211_cast_fp16)[name = tensor("hidden_states_293_cast_fp16")]; + tensor var_31400 = const()[name = tensor("op_31400"), val = tensor([1, 1])]; + tensor var_31402 = const()[name = tensor("op_31402"), val = tensor([1, 1])]; + tensor q_141_pad_type_0 = const()[name = tensor("q_141_pad_type_0"), val = tensor("custom")]; + tensor q_141_pad_0 = const()[name = tensor("q_141_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_1_attn1_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_1_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2657939584)))]; + tensor q_141_cast_fp16 = conv(dilations = var_31402, groups = var_30355, pad = q_141_pad_0, pad_type = q_141_pad_type_0, strides = var_31400, weight = up_blocks_0_attentions_0_transformer_blocks_1_attn1_to_q_weight_to_fp16, x = hidden_states_293_cast_fp16)[name = tensor("q_141_cast_fp16")]; + tensor var_31406 = const()[name = tensor("op_31406"), val = tensor([1, 1])]; + tensor var_31408 = const()[name = tensor("op_31408"), val = tensor([1, 1])]; + tensor k_281_pad_type_0 = const()[name = tensor("k_281_pad_type_0"), val = tensor("custom")]; + tensor k_281_pad_0 = const()[name = tensor("k_281_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_1_attn1_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_1_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2661216448)))]; + tensor k_281_cast_fp16 = conv(dilations = var_31408, groups = var_30355, pad = k_281_pad_0, pad_type = k_281_pad_type_0, strides = var_31406, weight = up_blocks_0_attentions_0_transformer_blocks_1_attn1_to_k_weight_to_fp16, x = hidden_states_293_cast_fp16)[name = tensor("k_281_cast_fp16")]; + tensor var_31412 = const()[name = tensor("op_31412"), val = tensor([1, 1])]; + tensor var_31414 = const()[name = tensor("op_31414"), val = tensor([1, 1])]; + tensor v_141_pad_type_0 = const()[name = tensor("v_141_pad_type_0"), val = tensor("custom")]; + tensor v_141_pad_0 = const()[name = tensor("v_141_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_1_attn1_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_1_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2664493312)))]; + tensor v_141_cast_fp16 = conv(dilations = var_31414, groups = var_30355, pad = v_141_pad_0, pad_type = v_141_pad_type_0, strides = var_31412, weight = up_blocks_0_attentions_0_transformer_blocks_1_attn1_to_v_weight_to_fp16, x = hidden_states_293_cast_fp16)[name = tensor("v_141_cast_fp16")]; + tensor var_31418_begin_0 = const()[name = tensor("op_31418_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_31418_end_0 = const()[name = tensor("op_31418_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_31418_end_mask_0 = const()[name = tensor("op_31418_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31418_cast_fp16 = slice_by_index(begin = var_31418_begin_0, end = var_31418_end_0, end_mask = var_31418_end_mask_0, x = q_141_cast_fp16)[name = tensor("op_31418_cast_fp16")]; + tensor var_31422_begin_0 = const()[name = tensor("op_31422_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_31422_end_0 = const()[name = tensor("op_31422_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_31422_end_mask_0 = const()[name = tensor("op_31422_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31422_cast_fp16 = slice_by_index(begin = var_31422_begin_0, end = var_31422_end_0, end_mask = var_31422_end_mask_0, x = q_141_cast_fp16)[name = tensor("op_31422_cast_fp16")]; + tensor var_31426_begin_0 = const()[name = tensor("op_31426_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_31426_end_0 = const()[name = tensor("op_31426_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_31426_end_mask_0 = const()[name = tensor("op_31426_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31426_cast_fp16 = slice_by_index(begin = var_31426_begin_0, end = var_31426_end_0, end_mask = var_31426_end_mask_0, x = q_141_cast_fp16)[name = tensor("op_31426_cast_fp16")]; + tensor var_31430_begin_0 = const()[name = tensor("op_31430_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_31430_end_0 = const()[name = tensor("op_31430_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_31430_end_mask_0 = const()[name = tensor("op_31430_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31430_cast_fp16 = slice_by_index(begin = var_31430_begin_0, end = var_31430_end_0, end_mask = var_31430_end_mask_0, x = q_141_cast_fp16)[name = tensor("op_31430_cast_fp16")]; + tensor var_31434_begin_0 = const()[name = tensor("op_31434_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_31434_end_0 = const()[name = tensor("op_31434_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_31434_end_mask_0 = const()[name = tensor("op_31434_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31434_cast_fp16 = slice_by_index(begin = var_31434_begin_0, end = var_31434_end_0, end_mask = var_31434_end_mask_0, x = q_141_cast_fp16)[name = tensor("op_31434_cast_fp16")]; + tensor var_31438_begin_0 = const()[name = tensor("op_31438_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_31438_end_0 = const()[name = tensor("op_31438_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_31438_end_mask_0 = const()[name = tensor("op_31438_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31438_cast_fp16 = slice_by_index(begin = var_31438_begin_0, end = var_31438_end_0, end_mask = var_31438_end_mask_0, x = q_141_cast_fp16)[name = tensor("op_31438_cast_fp16")]; + tensor var_31442_begin_0 = const()[name = tensor("op_31442_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_31442_end_0 = const()[name = tensor("op_31442_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_31442_end_mask_0 = const()[name = tensor("op_31442_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31442_cast_fp16 = slice_by_index(begin = var_31442_begin_0, end = var_31442_end_0, end_mask = var_31442_end_mask_0, x = q_141_cast_fp16)[name = tensor("op_31442_cast_fp16")]; + tensor var_31446_begin_0 = const()[name = tensor("op_31446_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_31446_end_0 = const()[name = tensor("op_31446_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_31446_end_mask_0 = const()[name = tensor("op_31446_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31446_cast_fp16 = slice_by_index(begin = var_31446_begin_0, end = var_31446_end_0, end_mask = var_31446_end_mask_0, x = q_141_cast_fp16)[name = tensor("op_31446_cast_fp16")]; + tensor var_31450_begin_0 = const()[name = tensor("op_31450_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_31450_end_0 = const()[name = tensor("op_31450_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_31450_end_mask_0 = const()[name = tensor("op_31450_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31450_cast_fp16 = slice_by_index(begin = var_31450_begin_0, end = var_31450_end_0, end_mask = var_31450_end_mask_0, x = q_141_cast_fp16)[name = tensor("op_31450_cast_fp16")]; + tensor var_31454_begin_0 = const()[name = tensor("op_31454_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_31454_end_0 = const()[name = tensor("op_31454_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_31454_end_mask_0 = const()[name = tensor("op_31454_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31454_cast_fp16 = slice_by_index(begin = var_31454_begin_0, end = var_31454_end_0, end_mask = var_31454_end_mask_0, x = q_141_cast_fp16)[name = tensor("op_31454_cast_fp16")]; + tensor var_31458_begin_0 = const()[name = tensor("op_31458_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_31458_end_0 = const()[name = tensor("op_31458_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_31458_end_mask_0 = const()[name = tensor("op_31458_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31458_cast_fp16 = slice_by_index(begin = var_31458_begin_0, end = var_31458_end_0, end_mask = var_31458_end_mask_0, x = q_141_cast_fp16)[name = tensor("op_31458_cast_fp16")]; + tensor var_31462_begin_0 = const()[name = tensor("op_31462_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_31462_end_0 = const()[name = tensor("op_31462_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_31462_end_mask_0 = const()[name = tensor("op_31462_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31462_cast_fp16 = slice_by_index(begin = var_31462_begin_0, end = var_31462_end_0, end_mask = var_31462_end_mask_0, x = q_141_cast_fp16)[name = tensor("op_31462_cast_fp16")]; + tensor var_31466_begin_0 = const()[name = tensor("op_31466_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_31466_end_0 = const()[name = tensor("op_31466_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_31466_end_mask_0 = const()[name = tensor("op_31466_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31466_cast_fp16 = slice_by_index(begin = var_31466_begin_0, end = var_31466_end_0, end_mask = var_31466_end_mask_0, x = q_141_cast_fp16)[name = tensor("op_31466_cast_fp16")]; + tensor var_31470_begin_0 = const()[name = tensor("op_31470_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_31470_end_0 = const()[name = tensor("op_31470_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_31470_end_mask_0 = const()[name = tensor("op_31470_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31470_cast_fp16 = slice_by_index(begin = var_31470_begin_0, end = var_31470_end_0, end_mask = var_31470_end_mask_0, x = q_141_cast_fp16)[name = tensor("op_31470_cast_fp16")]; + tensor var_31474_begin_0 = const()[name = tensor("op_31474_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_31474_end_0 = const()[name = tensor("op_31474_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_31474_end_mask_0 = const()[name = tensor("op_31474_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31474_cast_fp16 = slice_by_index(begin = var_31474_begin_0, end = var_31474_end_0, end_mask = var_31474_end_mask_0, x = q_141_cast_fp16)[name = tensor("op_31474_cast_fp16")]; + tensor var_31478_begin_0 = const()[name = tensor("op_31478_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_31478_end_0 = const()[name = tensor("op_31478_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_31478_end_mask_0 = const()[name = tensor("op_31478_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31478_cast_fp16 = slice_by_index(begin = var_31478_begin_0, end = var_31478_end_0, end_mask = var_31478_end_mask_0, x = q_141_cast_fp16)[name = tensor("op_31478_cast_fp16")]; + tensor var_31482_begin_0 = const()[name = tensor("op_31482_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_31482_end_0 = const()[name = tensor("op_31482_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_31482_end_mask_0 = const()[name = tensor("op_31482_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31482_cast_fp16 = slice_by_index(begin = var_31482_begin_0, end = var_31482_end_0, end_mask = var_31482_end_mask_0, x = q_141_cast_fp16)[name = tensor("op_31482_cast_fp16")]; + tensor var_31486_begin_0 = const()[name = tensor("op_31486_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_31486_end_0 = const()[name = tensor("op_31486_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_31486_end_mask_0 = const()[name = tensor("op_31486_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31486_cast_fp16 = slice_by_index(begin = var_31486_begin_0, end = var_31486_end_0, end_mask = var_31486_end_mask_0, x = q_141_cast_fp16)[name = tensor("op_31486_cast_fp16")]; + tensor var_31490_begin_0 = const()[name = tensor("op_31490_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_31490_end_0 = const()[name = tensor("op_31490_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_31490_end_mask_0 = const()[name = tensor("op_31490_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31490_cast_fp16 = slice_by_index(begin = var_31490_begin_0, end = var_31490_end_0, end_mask = var_31490_end_mask_0, x = q_141_cast_fp16)[name = tensor("op_31490_cast_fp16")]; + tensor var_31494_begin_0 = const()[name = tensor("op_31494_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_31494_end_0 = const()[name = tensor("op_31494_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_31494_end_mask_0 = const()[name = tensor("op_31494_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31494_cast_fp16 = slice_by_index(begin = var_31494_begin_0, end = var_31494_end_0, end_mask = var_31494_end_mask_0, x = q_141_cast_fp16)[name = tensor("op_31494_cast_fp16")]; + tensor k_283_perm_0 = const()[name = tensor("k_283_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_31501_begin_0 = const()[name = tensor("op_31501_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_31501_end_0 = const()[name = tensor("op_31501_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_31501_end_mask_0 = const()[name = tensor("op_31501_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_283_cast_fp16 = transpose(perm = k_283_perm_0, x = k_281_cast_fp16)[name = tensor("transpose_69")]; + tensor var_31501_cast_fp16 = slice_by_index(begin = var_31501_begin_0, end = var_31501_end_0, end_mask = var_31501_end_mask_0, x = k_283_cast_fp16)[name = tensor("op_31501_cast_fp16")]; + tensor var_31505_begin_0 = const()[name = tensor("op_31505_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_31505_end_0 = const()[name = tensor("op_31505_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_31505_end_mask_0 = const()[name = tensor("op_31505_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31505_cast_fp16 = slice_by_index(begin = var_31505_begin_0, end = var_31505_end_0, end_mask = var_31505_end_mask_0, x = k_283_cast_fp16)[name = tensor("op_31505_cast_fp16")]; + tensor var_31509_begin_0 = const()[name = tensor("op_31509_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_31509_end_0 = const()[name = tensor("op_31509_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_31509_end_mask_0 = const()[name = tensor("op_31509_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31509_cast_fp16 = slice_by_index(begin = var_31509_begin_0, end = var_31509_end_0, end_mask = var_31509_end_mask_0, x = k_283_cast_fp16)[name = tensor("op_31509_cast_fp16")]; + tensor var_31513_begin_0 = const()[name = tensor("op_31513_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_31513_end_0 = const()[name = tensor("op_31513_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_31513_end_mask_0 = const()[name = tensor("op_31513_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31513_cast_fp16 = slice_by_index(begin = var_31513_begin_0, end = var_31513_end_0, end_mask = var_31513_end_mask_0, x = k_283_cast_fp16)[name = tensor("op_31513_cast_fp16")]; + tensor var_31517_begin_0 = const()[name = tensor("op_31517_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_31517_end_0 = const()[name = tensor("op_31517_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_31517_end_mask_0 = const()[name = tensor("op_31517_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31517_cast_fp16 = slice_by_index(begin = var_31517_begin_0, end = var_31517_end_0, end_mask = var_31517_end_mask_0, x = k_283_cast_fp16)[name = tensor("op_31517_cast_fp16")]; + tensor var_31521_begin_0 = const()[name = tensor("op_31521_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_31521_end_0 = const()[name = tensor("op_31521_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_31521_end_mask_0 = const()[name = tensor("op_31521_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31521_cast_fp16 = slice_by_index(begin = var_31521_begin_0, end = var_31521_end_0, end_mask = var_31521_end_mask_0, x = k_283_cast_fp16)[name = tensor("op_31521_cast_fp16")]; + tensor var_31525_begin_0 = const()[name = tensor("op_31525_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_31525_end_0 = const()[name = tensor("op_31525_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_31525_end_mask_0 = const()[name = tensor("op_31525_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31525_cast_fp16 = slice_by_index(begin = var_31525_begin_0, end = var_31525_end_0, end_mask = var_31525_end_mask_0, x = k_283_cast_fp16)[name = tensor("op_31525_cast_fp16")]; + tensor var_31529_begin_0 = const()[name = tensor("op_31529_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_31529_end_0 = const()[name = tensor("op_31529_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_31529_end_mask_0 = const()[name = tensor("op_31529_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31529_cast_fp16 = slice_by_index(begin = var_31529_begin_0, end = var_31529_end_0, end_mask = var_31529_end_mask_0, x = k_283_cast_fp16)[name = tensor("op_31529_cast_fp16")]; + tensor var_31533_begin_0 = const()[name = tensor("op_31533_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_31533_end_0 = const()[name = tensor("op_31533_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_31533_end_mask_0 = const()[name = tensor("op_31533_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31533_cast_fp16 = slice_by_index(begin = var_31533_begin_0, end = var_31533_end_0, end_mask = var_31533_end_mask_0, x = k_283_cast_fp16)[name = tensor("op_31533_cast_fp16")]; + tensor var_31537_begin_0 = const()[name = tensor("op_31537_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_31537_end_0 = const()[name = tensor("op_31537_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_31537_end_mask_0 = const()[name = tensor("op_31537_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31537_cast_fp16 = slice_by_index(begin = var_31537_begin_0, end = var_31537_end_0, end_mask = var_31537_end_mask_0, x = k_283_cast_fp16)[name = tensor("op_31537_cast_fp16")]; + tensor var_31541_begin_0 = const()[name = tensor("op_31541_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_31541_end_0 = const()[name = tensor("op_31541_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_31541_end_mask_0 = const()[name = tensor("op_31541_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31541_cast_fp16 = slice_by_index(begin = var_31541_begin_0, end = var_31541_end_0, end_mask = var_31541_end_mask_0, x = k_283_cast_fp16)[name = tensor("op_31541_cast_fp16")]; + tensor var_31545_begin_0 = const()[name = tensor("op_31545_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_31545_end_0 = const()[name = tensor("op_31545_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_31545_end_mask_0 = const()[name = tensor("op_31545_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31545_cast_fp16 = slice_by_index(begin = var_31545_begin_0, end = var_31545_end_0, end_mask = var_31545_end_mask_0, x = k_283_cast_fp16)[name = tensor("op_31545_cast_fp16")]; + tensor var_31549_begin_0 = const()[name = tensor("op_31549_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_31549_end_0 = const()[name = tensor("op_31549_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_31549_end_mask_0 = const()[name = tensor("op_31549_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31549_cast_fp16 = slice_by_index(begin = var_31549_begin_0, end = var_31549_end_0, end_mask = var_31549_end_mask_0, x = k_283_cast_fp16)[name = tensor("op_31549_cast_fp16")]; + tensor var_31553_begin_0 = const()[name = tensor("op_31553_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_31553_end_0 = const()[name = tensor("op_31553_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_31553_end_mask_0 = const()[name = tensor("op_31553_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31553_cast_fp16 = slice_by_index(begin = var_31553_begin_0, end = var_31553_end_0, end_mask = var_31553_end_mask_0, x = k_283_cast_fp16)[name = tensor("op_31553_cast_fp16")]; + tensor var_31557_begin_0 = const()[name = tensor("op_31557_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_31557_end_0 = const()[name = tensor("op_31557_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_31557_end_mask_0 = const()[name = tensor("op_31557_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31557_cast_fp16 = slice_by_index(begin = var_31557_begin_0, end = var_31557_end_0, end_mask = var_31557_end_mask_0, x = k_283_cast_fp16)[name = tensor("op_31557_cast_fp16")]; + tensor var_31561_begin_0 = const()[name = tensor("op_31561_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_31561_end_0 = const()[name = tensor("op_31561_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_31561_end_mask_0 = const()[name = tensor("op_31561_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31561_cast_fp16 = slice_by_index(begin = var_31561_begin_0, end = var_31561_end_0, end_mask = var_31561_end_mask_0, x = k_283_cast_fp16)[name = tensor("op_31561_cast_fp16")]; + tensor var_31565_begin_0 = const()[name = tensor("op_31565_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_31565_end_0 = const()[name = tensor("op_31565_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_31565_end_mask_0 = const()[name = tensor("op_31565_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31565_cast_fp16 = slice_by_index(begin = var_31565_begin_0, end = var_31565_end_0, end_mask = var_31565_end_mask_0, x = k_283_cast_fp16)[name = tensor("op_31565_cast_fp16")]; + tensor var_31569_begin_0 = const()[name = tensor("op_31569_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_31569_end_0 = const()[name = tensor("op_31569_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_31569_end_mask_0 = const()[name = tensor("op_31569_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31569_cast_fp16 = slice_by_index(begin = var_31569_begin_0, end = var_31569_end_0, end_mask = var_31569_end_mask_0, x = k_283_cast_fp16)[name = tensor("op_31569_cast_fp16")]; + tensor var_31573_begin_0 = const()[name = tensor("op_31573_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_31573_end_0 = const()[name = tensor("op_31573_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_31573_end_mask_0 = const()[name = tensor("op_31573_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31573_cast_fp16 = slice_by_index(begin = var_31573_begin_0, end = var_31573_end_0, end_mask = var_31573_end_mask_0, x = k_283_cast_fp16)[name = tensor("op_31573_cast_fp16")]; + tensor var_31577_begin_0 = const()[name = tensor("op_31577_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_31577_end_0 = const()[name = tensor("op_31577_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_31577_end_mask_0 = const()[name = tensor("op_31577_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31577_cast_fp16 = slice_by_index(begin = var_31577_begin_0, end = var_31577_end_0, end_mask = var_31577_end_mask_0, x = k_283_cast_fp16)[name = tensor("op_31577_cast_fp16")]; + tensor var_31579_begin_0 = const()[name = tensor("op_31579_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_31579_end_0 = const()[name = tensor("op_31579_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_31579_end_mask_0 = const()[name = tensor("op_31579_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31579_cast_fp16 = slice_by_index(begin = var_31579_begin_0, end = var_31579_end_0, end_mask = var_31579_end_mask_0, x = v_141_cast_fp16)[name = tensor("op_31579_cast_fp16")]; + tensor var_31583_begin_0 = const()[name = tensor("op_31583_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_31583_end_0 = const()[name = tensor("op_31583_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_31583_end_mask_0 = const()[name = tensor("op_31583_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31583_cast_fp16 = slice_by_index(begin = var_31583_begin_0, end = var_31583_end_0, end_mask = var_31583_end_mask_0, x = v_141_cast_fp16)[name = tensor("op_31583_cast_fp16")]; + tensor var_31587_begin_0 = const()[name = tensor("op_31587_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_31587_end_0 = const()[name = tensor("op_31587_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_31587_end_mask_0 = const()[name = tensor("op_31587_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31587_cast_fp16 = slice_by_index(begin = var_31587_begin_0, end = var_31587_end_0, end_mask = var_31587_end_mask_0, x = v_141_cast_fp16)[name = tensor("op_31587_cast_fp16")]; + tensor var_31591_begin_0 = const()[name = tensor("op_31591_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_31591_end_0 = const()[name = tensor("op_31591_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_31591_end_mask_0 = const()[name = tensor("op_31591_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31591_cast_fp16 = slice_by_index(begin = var_31591_begin_0, end = var_31591_end_0, end_mask = var_31591_end_mask_0, x = v_141_cast_fp16)[name = tensor("op_31591_cast_fp16")]; + tensor var_31595_begin_0 = const()[name = tensor("op_31595_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_31595_end_0 = const()[name = tensor("op_31595_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_31595_end_mask_0 = const()[name = tensor("op_31595_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31595_cast_fp16 = slice_by_index(begin = var_31595_begin_0, end = var_31595_end_0, end_mask = var_31595_end_mask_0, x = v_141_cast_fp16)[name = tensor("op_31595_cast_fp16")]; + tensor var_31599_begin_0 = const()[name = tensor("op_31599_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_31599_end_0 = const()[name = tensor("op_31599_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_31599_end_mask_0 = const()[name = tensor("op_31599_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31599_cast_fp16 = slice_by_index(begin = var_31599_begin_0, end = var_31599_end_0, end_mask = var_31599_end_mask_0, x = v_141_cast_fp16)[name = tensor("op_31599_cast_fp16")]; + tensor var_31603_begin_0 = const()[name = tensor("op_31603_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_31603_end_0 = const()[name = tensor("op_31603_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_31603_end_mask_0 = const()[name = tensor("op_31603_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31603_cast_fp16 = slice_by_index(begin = var_31603_begin_0, end = var_31603_end_0, end_mask = var_31603_end_mask_0, x = v_141_cast_fp16)[name = tensor("op_31603_cast_fp16")]; + tensor var_31607_begin_0 = const()[name = tensor("op_31607_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_31607_end_0 = const()[name = tensor("op_31607_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_31607_end_mask_0 = const()[name = tensor("op_31607_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31607_cast_fp16 = slice_by_index(begin = var_31607_begin_0, end = var_31607_end_0, end_mask = var_31607_end_mask_0, x = v_141_cast_fp16)[name = tensor("op_31607_cast_fp16")]; + tensor var_31611_begin_0 = const()[name = tensor("op_31611_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_31611_end_0 = const()[name = tensor("op_31611_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_31611_end_mask_0 = const()[name = tensor("op_31611_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31611_cast_fp16 = slice_by_index(begin = var_31611_begin_0, end = var_31611_end_0, end_mask = var_31611_end_mask_0, x = v_141_cast_fp16)[name = tensor("op_31611_cast_fp16")]; + tensor var_31615_begin_0 = const()[name = tensor("op_31615_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_31615_end_0 = const()[name = tensor("op_31615_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_31615_end_mask_0 = const()[name = tensor("op_31615_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31615_cast_fp16 = slice_by_index(begin = var_31615_begin_0, end = var_31615_end_0, end_mask = var_31615_end_mask_0, x = v_141_cast_fp16)[name = tensor("op_31615_cast_fp16")]; + tensor var_31619_begin_0 = const()[name = tensor("op_31619_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_31619_end_0 = const()[name = tensor("op_31619_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_31619_end_mask_0 = const()[name = tensor("op_31619_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31619_cast_fp16 = slice_by_index(begin = var_31619_begin_0, end = var_31619_end_0, end_mask = var_31619_end_mask_0, x = v_141_cast_fp16)[name = tensor("op_31619_cast_fp16")]; + tensor var_31623_begin_0 = const()[name = tensor("op_31623_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_31623_end_0 = const()[name = tensor("op_31623_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_31623_end_mask_0 = const()[name = tensor("op_31623_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31623_cast_fp16 = slice_by_index(begin = var_31623_begin_0, end = var_31623_end_0, end_mask = var_31623_end_mask_0, x = v_141_cast_fp16)[name = tensor("op_31623_cast_fp16")]; + tensor var_31627_begin_0 = const()[name = tensor("op_31627_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_31627_end_0 = const()[name = tensor("op_31627_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_31627_end_mask_0 = const()[name = tensor("op_31627_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31627_cast_fp16 = slice_by_index(begin = var_31627_begin_0, end = var_31627_end_0, end_mask = var_31627_end_mask_0, x = v_141_cast_fp16)[name = tensor("op_31627_cast_fp16")]; + tensor var_31631_begin_0 = const()[name = tensor("op_31631_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_31631_end_0 = const()[name = tensor("op_31631_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_31631_end_mask_0 = const()[name = tensor("op_31631_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31631_cast_fp16 = slice_by_index(begin = var_31631_begin_0, end = var_31631_end_0, end_mask = var_31631_end_mask_0, x = v_141_cast_fp16)[name = tensor("op_31631_cast_fp16")]; + tensor var_31635_begin_0 = const()[name = tensor("op_31635_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_31635_end_0 = const()[name = tensor("op_31635_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_31635_end_mask_0 = const()[name = tensor("op_31635_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31635_cast_fp16 = slice_by_index(begin = var_31635_begin_0, end = var_31635_end_0, end_mask = var_31635_end_mask_0, x = v_141_cast_fp16)[name = tensor("op_31635_cast_fp16")]; + tensor var_31639_begin_0 = const()[name = tensor("op_31639_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_31639_end_0 = const()[name = tensor("op_31639_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_31639_end_mask_0 = const()[name = tensor("op_31639_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31639_cast_fp16 = slice_by_index(begin = var_31639_begin_0, end = var_31639_end_0, end_mask = var_31639_end_mask_0, x = v_141_cast_fp16)[name = tensor("op_31639_cast_fp16")]; + tensor var_31643_begin_0 = const()[name = tensor("op_31643_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_31643_end_0 = const()[name = tensor("op_31643_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_31643_end_mask_0 = const()[name = tensor("op_31643_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31643_cast_fp16 = slice_by_index(begin = var_31643_begin_0, end = var_31643_end_0, end_mask = var_31643_end_mask_0, x = v_141_cast_fp16)[name = tensor("op_31643_cast_fp16")]; + tensor var_31647_begin_0 = const()[name = tensor("op_31647_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_31647_end_0 = const()[name = tensor("op_31647_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_31647_end_mask_0 = const()[name = tensor("op_31647_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31647_cast_fp16 = slice_by_index(begin = var_31647_begin_0, end = var_31647_end_0, end_mask = var_31647_end_mask_0, x = v_141_cast_fp16)[name = tensor("op_31647_cast_fp16")]; + tensor var_31651_begin_0 = const()[name = tensor("op_31651_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_31651_end_0 = const()[name = tensor("op_31651_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_31651_end_mask_0 = const()[name = tensor("op_31651_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31651_cast_fp16 = slice_by_index(begin = var_31651_begin_0, end = var_31651_end_0, end_mask = var_31651_end_mask_0, x = v_141_cast_fp16)[name = tensor("op_31651_cast_fp16")]; + tensor var_31655_begin_0 = const()[name = tensor("op_31655_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_31655_end_0 = const()[name = tensor("op_31655_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_31655_end_mask_0 = const()[name = tensor("op_31655_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31655_cast_fp16 = slice_by_index(begin = var_31655_begin_0, end = var_31655_end_0, end_mask = var_31655_end_mask_0, x = v_141_cast_fp16)[name = tensor("op_31655_cast_fp16")]; + tensor var_31659_equation_0 = const()[name = tensor("op_31659_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_31659_cast_fp16 = einsum(equation = var_31659_equation_0, values = (var_31501_cast_fp16, var_31418_cast_fp16))[name = tensor("op_31659_cast_fp16")]; + tensor var_31660_to_fp16 = const()[name = tensor("op_31660_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2641_cast_fp16 = mul(x = var_31659_cast_fp16, y = var_31660_to_fp16)[name = tensor("aw_2641_cast_fp16")]; + tensor var_31663_equation_0 = const()[name = tensor("op_31663_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_31663_cast_fp16 = einsum(equation = var_31663_equation_0, values = (var_31505_cast_fp16, var_31422_cast_fp16))[name = tensor("op_31663_cast_fp16")]; + tensor var_31664_to_fp16 = const()[name = tensor("op_31664_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2643_cast_fp16 = mul(x = var_31663_cast_fp16, y = var_31664_to_fp16)[name = tensor("aw_2643_cast_fp16")]; + tensor var_31667_equation_0 = const()[name = tensor("op_31667_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_31667_cast_fp16 = einsum(equation = var_31667_equation_0, values = (var_31509_cast_fp16, var_31426_cast_fp16))[name = tensor("op_31667_cast_fp16")]; + tensor var_31668_to_fp16 = const()[name = tensor("op_31668_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2645_cast_fp16 = mul(x = var_31667_cast_fp16, y = var_31668_to_fp16)[name = tensor("aw_2645_cast_fp16")]; + tensor var_31671_equation_0 = const()[name = tensor("op_31671_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_31671_cast_fp16 = einsum(equation = var_31671_equation_0, values = (var_31513_cast_fp16, var_31430_cast_fp16))[name = tensor("op_31671_cast_fp16")]; + tensor var_31672_to_fp16 = const()[name = tensor("op_31672_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2647_cast_fp16 = mul(x = var_31671_cast_fp16, y = var_31672_to_fp16)[name = tensor("aw_2647_cast_fp16")]; + tensor var_31675_equation_0 = const()[name = tensor("op_31675_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_31675_cast_fp16 = einsum(equation = var_31675_equation_0, values = (var_31517_cast_fp16, var_31434_cast_fp16))[name = tensor("op_31675_cast_fp16")]; + tensor var_31676_to_fp16 = const()[name = tensor("op_31676_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2649_cast_fp16 = mul(x = var_31675_cast_fp16, y = var_31676_to_fp16)[name = tensor("aw_2649_cast_fp16")]; + tensor var_31679_equation_0 = const()[name = tensor("op_31679_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_31679_cast_fp16 = einsum(equation = var_31679_equation_0, values = (var_31521_cast_fp16, var_31438_cast_fp16))[name = tensor("op_31679_cast_fp16")]; + tensor var_31680_to_fp16 = const()[name = tensor("op_31680_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2651_cast_fp16 = mul(x = var_31679_cast_fp16, y = var_31680_to_fp16)[name = tensor("aw_2651_cast_fp16")]; + tensor var_31683_equation_0 = const()[name = tensor("op_31683_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_31683_cast_fp16 = einsum(equation = var_31683_equation_0, values = (var_31525_cast_fp16, var_31442_cast_fp16))[name = tensor("op_31683_cast_fp16")]; + tensor var_31684_to_fp16 = const()[name = tensor("op_31684_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2653_cast_fp16 = mul(x = var_31683_cast_fp16, y = var_31684_to_fp16)[name = tensor("aw_2653_cast_fp16")]; + tensor var_31687_equation_0 = const()[name = tensor("op_31687_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_31687_cast_fp16 = einsum(equation = var_31687_equation_0, values = (var_31529_cast_fp16, var_31446_cast_fp16))[name = tensor("op_31687_cast_fp16")]; + tensor var_31688_to_fp16 = const()[name = tensor("op_31688_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2655_cast_fp16 = mul(x = var_31687_cast_fp16, y = var_31688_to_fp16)[name = tensor("aw_2655_cast_fp16")]; + tensor var_31691_equation_0 = const()[name = tensor("op_31691_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_31691_cast_fp16 = einsum(equation = var_31691_equation_0, values = (var_31533_cast_fp16, var_31450_cast_fp16))[name = tensor("op_31691_cast_fp16")]; + tensor var_31692_to_fp16 = const()[name = tensor("op_31692_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2657_cast_fp16 = mul(x = var_31691_cast_fp16, y = var_31692_to_fp16)[name = tensor("aw_2657_cast_fp16")]; + tensor var_31695_equation_0 = const()[name = tensor("op_31695_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_31695_cast_fp16 = einsum(equation = var_31695_equation_0, values = (var_31537_cast_fp16, var_31454_cast_fp16))[name = tensor("op_31695_cast_fp16")]; + tensor var_31696_to_fp16 = const()[name = tensor("op_31696_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2659_cast_fp16 = mul(x = var_31695_cast_fp16, y = var_31696_to_fp16)[name = tensor("aw_2659_cast_fp16")]; + tensor var_31699_equation_0 = const()[name = tensor("op_31699_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_31699_cast_fp16 = einsum(equation = var_31699_equation_0, values = (var_31541_cast_fp16, var_31458_cast_fp16))[name = tensor("op_31699_cast_fp16")]; + tensor var_31700_to_fp16 = const()[name = tensor("op_31700_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2661_cast_fp16 = mul(x = var_31699_cast_fp16, y = var_31700_to_fp16)[name = tensor("aw_2661_cast_fp16")]; + tensor var_31703_equation_0 = const()[name = tensor("op_31703_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_31703_cast_fp16 = einsum(equation = var_31703_equation_0, values = (var_31545_cast_fp16, var_31462_cast_fp16))[name = tensor("op_31703_cast_fp16")]; + tensor var_31704_to_fp16 = const()[name = tensor("op_31704_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2663_cast_fp16 = mul(x = var_31703_cast_fp16, y = var_31704_to_fp16)[name = tensor("aw_2663_cast_fp16")]; + tensor var_31707_equation_0 = const()[name = tensor("op_31707_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_31707_cast_fp16 = einsum(equation = var_31707_equation_0, values = (var_31549_cast_fp16, var_31466_cast_fp16))[name = tensor("op_31707_cast_fp16")]; + tensor var_31708_to_fp16 = const()[name = tensor("op_31708_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2665_cast_fp16 = mul(x = var_31707_cast_fp16, y = var_31708_to_fp16)[name = tensor("aw_2665_cast_fp16")]; + tensor var_31711_equation_0 = const()[name = tensor("op_31711_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_31711_cast_fp16 = einsum(equation = var_31711_equation_0, values = (var_31553_cast_fp16, var_31470_cast_fp16))[name = tensor("op_31711_cast_fp16")]; + tensor var_31712_to_fp16 = const()[name = tensor("op_31712_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2667_cast_fp16 = mul(x = var_31711_cast_fp16, y = var_31712_to_fp16)[name = tensor("aw_2667_cast_fp16")]; + tensor var_31715_equation_0 = const()[name = tensor("op_31715_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_31715_cast_fp16 = einsum(equation = var_31715_equation_0, values = (var_31557_cast_fp16, var_31474_cast_fp16))[name = tensor("op_31715_cast_fp16")]; + tensor var_31716_to_fp16 = const()[name = tensor("op_31716_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2669_cast_fp16 = mul(x = var_31715_cast_fp16, y = var_31716_to_fp16)[name = tensor("aw_2669_cast_fp16")]; + tensor var_31719_equation_0 = const()[name = tensor("op_31719_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_31719_cast_fp16 = einsum(equation = var_31719_equation_0, values = (var_31561_cast_fp16, var_31478_cast_fp16))[name = tensor("op_31719_cast_fp16")]; + tensor var_31720_to_fp16 = const()[name = tensor("op_31720_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2671_cast_fp16 = mul(x = var_31719_cast_fp16, y = var_31720_to_fp16)[name = tensor("aw_2671_cast_fp16")]; + tensor var_31723_equation_0 = const()[name = tensor("op_31723_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_31723_cast_fp16 = einsum(equation = var_31723_equation_0, values = (var_31565_cast_fp16, var_31482_cast_fp16))[name = tensor("op_31723_cast_fp16")]; + tensor var_31724_to_fp16 = const()[name = tensor("op_31724_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2673_cast_fp16 = mul(x = var_31723_cast_fp16, y = var_31724_to_fp16)[name = tensor("aw_2673_cast_fp16")]; + tensor var_31727_equation_0 = const()[name = tensor("op_31727_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_31727_cast_fp16 = einsum(equation = var_31727_equation_0, values = (var_31569_cast_fp16, var_31486_cast_fp16))[name = tensor("op_31727_cast_fp16")]; + tensor var_31728_to_fp16 = const()[name = tensor("op_31728_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2675_cast_fp16 = mul(x = var_31727_cast_fp16, y = var_31728_to_fp16)[name = tensor("aw_2675_cast_fp16")]; + tensor var_31731_equation_0 = const()[name = tensor("op_31731_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_31731_cast_fp16 = einsum(equation = var_31731_equation_0, values = (var_31573_cast_fp16, var_31490_cast_fp16))[name = tensor("op_31731_cast_fp16")]; + tensor var_31732_to_fp16 = const()[name = tensor("op_31732_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2677_cast_fp16 = mul(x = var_31731_cast_fp16, y = var_31732_to_fp16)[name = tensor("aw_2677_cast_fp16")]; + tensor var_31735_equation_0 = const()[name = tensor("op_31735_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_31735_cast_fp16 = einsum(equation = var_31735_equation_0, values = (var_31577_cast_fp16, var_31494_cast_fp16))[name = tensor("op_31735_cast_fp16")]; + tensor var_31736_to_fp16 = const()[name = tensor("op_31736_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2679_cast_fp16 = mul(x = var_31735_cast_fp16, y = var_31736_to_fp16)[name = tensor("aw_2679_cast_fp16")]; + tensor var_31738_cast_fp16 = softmax(axis = var_30355, x = aw_2641_cast_fp16)[name = tensor("op_31738_cast_fp16")]; + tensor var_31739_cast_fp16 = softmax(axis = var_30355, x = aw_2643_cast_fp16)[name = tensor("op_31739_cast_fp16")]; + tensor var_31740_cast_fp16 = softmax(axis = var_30355, x = aw_2645_cast_fp16)[name = tensor("op_31740_cast_fp16")]; + tensor var_31741_cast_fp16 = softmax(axis = var_30355, x = aw_2647_cast_fp16)[name = tensor("op_31741_cast_fp16")]; + tensor var_31742_cast_fp16 = softmax(axis = var_30355, x = aw_2649_cast_fp16)[name = tensor("op_31742_cast_fp16")]; + tensor var_31743_cast_fp16 = softmax(axis = var_30355, x = aw_2651_cast_fp16)[name = tensor("op_31743_cast_fp16")]; + tensor var_31744_cast_fp16 = softmax(axis = var_30355, x = aw_2653_cast_fp16)[name = tensor("op_31744_cast_fp16")]; + tensor var_31745_cast_fp16 = softmax(axis = var_30355, x = aw_2655_cast_fp16)[name = tensor("op_31745_cast_fp16")]; + tensor var_31746_cast_fp16 = softmax(axis = var_30355, x = aw_2657_cast_fp16)[name = tensor("op_31746_cast_fp16")]; + tensor var_31747_cast_fp16 = softmax(axis = var_30355, x = aw_2659_cast_fp16)[name = tensor("op_31747_cast_fp16")]; + tensor var_31748_cast_fp16 = softmax(axis = var_30355, x = aw_2661_cast_fp16)[name = tensor("op_31748_cast_fp16")]; + tensor var_31749_cast_fp16 = softmax(axis = var_30355, x = aw_2663_cast_fp16)[name = tensor("op_31749_cast_fp16")]; + tensor var_31750_cast_fp16 = softmax(axis = var_30355, x = aw_2665_cast_fp16)[name = tensor("op_31750_cast_fp16")]; + tensor var_31751_cast_fp16 = softmax(axis = var_30355, x = aw_2667_cast_fp16)[name = tensor("op_31751_cast_fp16")]; + tensor var_31752_cast_fp16 = softmax(axis = var_30355, x = aw_2669_cast_fp16)[name = tensor("op_31752_cast_fp16")]; + tensor var_31753_cast_fp16 = softmax(axis = var_30355, x = aw_2671_cast_fp16)[name = tensor("op_31753_cast_fp16")]; + tensor var_31754_cast_fp16 = softmax(axis = var_30355, x = aw_2673_cast_fp16)[name = tensor("op_31754_cast_fp16")]; + tensor var_31755_cast_fp16 = softmax(axis = var_30355, x = aw_2675_cast_fp16)[name = tensor("op_31755_cast_fp16")]; + tensor var_31756_cast_fp16 = softmax(axis = var_30355, x = aw_2677_cast_fp16)[name = tensor("op_31756_cast_fp16")]; + tensor var_31757_cast_fp16 = softmax(axis = var_30355, x = aw_2679_cast_fp16)[name = tensor("op_31757_cast_fp16")]; + tensor var_31759_equation_0 = const()[name = tensor("op_31759_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31759_cast_fp16 = einsum(equation = var_31759_equation_0, values = (var_31579_cast_fp16, var_31738_cast_fp16))[name = tensor("op_31759_cast_fp16")]; + tensor var_31761_equation_0 = const()[name = tensor("op_31761_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31761_cast_fp16 = einsum(equation = var_31761_equation_0, values = (var_31583_cast_fp16, var_31739_cast_fp16))[name = tensor("op_31761_cast_fp16")]; + tensor var_31763_equation_0 = const()[name = tensor("op_31763_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31763_cast_fp16 = einsum(equation = var_31763_equation_0, values = (var_31587_cast_fp16, var_31740_cast_fp16))[name = tensor("op_31763_cast_fp16")]; + tensor var_31765_equation_0 = const()[name = tensor("op_31765_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31765_cast_fp16 = einsum(equation = var_31765_equation_0, values = (var_31591_cast_fp16, var_31741_cast_fp16))[name = tensor("op_31765_cast_fp16")]; + tensor var_31767_equation_0 = const()[name = tensor("op_31767_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31767_cast_fp16 = einsum(equation = var_31767_equation_0, values = (var_31595_cast_fp16, var_31742_cast_fp16))[name = tensor("op_31767_cast_fp16")]; + tensor var_31769_equation_0 = const()[name = tensor("op_31769_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31769_cast_fp16 = einsum(equation = var_31769_equation_0, values = (var_31599_cast_fp16, var_31743_cast_fp16))[name = tensor("op_31769_cast_fp16")]; + tensor var_31771_equation_0 = const()[name = tensor("op_31771_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31771_cast_fp16 = einsum(equation = var_31771_equation_0, values = (var_31603_cast_fp16, var_31744_cast_fp16))[name = tensor("op_31771_cast_fp16")]; + tensor var_31773_equation_0 = const()[name = tensor("op_31773_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31773_cast_fp16 = einsum(equation = var_31773_equation_0, values = (var_31607_cast_fp16, var_31745_cast_fp16))[name = tensor("op_31773_cast_fp16")]; + tensor var_31775_equation_0 = const()[name = tensor("op_31775_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31775_cast_fp16 = einsum(equation = var_31775_equation_0, values = (var_31611_cast_fp16, var_31746_cast_fp16))[name = tensor("op_31775_cast_fp16")]; + tensor var_31777_equation_0 = const()[name = tensor("op_31777_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31777_cast_fp16 = einsum(equation = var_31777_equation_0, values = (var_31615_cast_fp16, var_31747_cast_fp16))[name = tensor("op_31777_cast_fp16")]; + tensor var_31779_equation_0 = const()[name = tensor("op_31779_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31779_cast_fp16 = einsum(equation = var_31779_equation_0, values = (var_31619_cast_fp16, var_31748_cast_fp16))[name = tensor("op_31779_cast_fp16")]; + tensor var_31781_equation_0 = const()[name = tensor("op_31781_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31781_cast_fp16 = einsum(equation = var_31781_equation_0, values = (var_31623_cast_fp16, var_31749_cast_fp16))[name = tensor("op_31781_cast_fp16")]; + tensor var_31783_equation_0 = const()[name = tensor("op_31783_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31783_cast_fp16 = einsum(equation = var_31783_equation_0, values = (var_31627_cast_fp16, var_31750_cast_fp16))[name = tensor("op_31783_cast_fp16")]; + tensor var_31785_equation_0 = const()[name = tensor("op_31785_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31785_cast_fp16 = einsum(equation = var_31785_equation_0, values = (var_31631_cast_fp16, var_31751_cast_fp16))[name = tensor("op_31785_cast_fp16")]; + tensor var_31787_equation_0 = const()[name = tensor("op_31787_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31787_cast_fp16 = einsum(equation = var_31787_equation_0, values = (var_31635_cast_fp16, var_31752_cast_fp16))[name = tensor("op_31787_cast_fp16")]; + tensor var_31789_equation_0 = const()[name = tensor("op_31789_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31789_cast_fp16 = einsum(equation = var_31789_equation_0, values = (var_31639_cast_fp16, var_31753_cast_fp16))[name = tensor("op_31789_cast_fp16")]; + tensor var_31791_equation_0 = const()[name = tensor("op_31791_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31791_cast_fp16 = einsum(equation = var_31791_equation_0, values = (var_31643_cast_fp16, var_31754_cast_fp16))[name = tensor("op_31791_cast_fp16")]; + tensor var_31793_equation_0 = const()[name = tensor("op_31793_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31793_cast_fp16 = einsum(equation = var_31793_equation_0, values = (var_31647_cast_fp16, var_31755_cast_fp16))[name = tensor("op_31793_cast_fp16")]; + tensor var_31795_equation_0 = const()[name = tensor("op_31795_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31795_cast_fp16 = einsum(equation = var_31795_equation_0, values = (var_31651_cast_fp16, var_31756_cast_fp16))[name = tensor("op_31795_cast_fp16")]; + tensor var_31797_equation_0 = const()[name = tensor("op_31797_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31797_cast_fp16 = einsum(equation = var_31797_equation_0, values = (var_31655_cast_fp16, var_31757_cast_fp16))[name = tensor("op_31797_cast_fp16")]; + tensor input_447_interleave_0 = const()[name = tensor("input_447_interleave_0"), val = tensor(false)]; + tensor input_447_cast_fp16 = concat(axis = var_30355, interleave = input_447_interleave_0, values = (var_31759_cast_fp16, var_31761_cast_fp16, var_31763_cast_fp16, var_31765_cast_fp16, var_31767_cast_fp16, var_31769_cast_fp16, var_31771_cast_fp16, var_31773_cast_fp16, var_31775_cast_fp16, var_31777_cast_fp16, var_31779_cast_fp16, var_31781_cast_fp16, var_31783_cast_fp16, var_31785_cast_fp16, var_31787_cast_fp16, var_31789_cast_fp16, var_31791_cast_fp16, var_31793_cast_fp16, var_31795_cast_fp16, var_31797_cast_fp16))[name = tensor("input_447_cast_fp16")]; + tensor var_31803 = const()[name = tensor("op_31803"), val = tensor([1, 1])]; + tensor var_31805 = const()[name = tensor("op_31805"), val = tensor([1, 1])]; + tensor var_31807_pad_type_0 = const()[name = tensor("op_31807_pad_type_0"), val = tensor("custom")]; + tensor var_31807_pad_0 = const()[name = tensor("op_31807_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_1_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_1_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2667770176)))]; + tensor up_blocks_0_attentions_0_transformer_blocks_1_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_1_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2671047040)))]; + tensor var_31807_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_1_attn1_to_out_0_bias_to_fp16, dilations = var_31805, groups = var_30355, pad = var_31807_pad_0, pad_type = var_31807_pad_type_0, strides = var_31803, weight = up_blocks_0_attentions_0_transformer_blocks_1_attn1_to_out_0_weight_to_fp16, x = input_447_cast_fp16)[name = tensor("op_31807_cast_fp16")]; + tensor inputs_213_cast_fp16 = add(x = var_31807_cast_fp16, y = inputs_211_cast_fp16)[name = tensor("inputs_213_cast_fp16")]; + tensor hidden_states_295_axes_0 = const()[name = tensor("hidden_states_295_axes_0"), val = tensor([1])]; + tensor hidden_states_295_gamma_0_to_fp16 = const()[name = tensor("hidden_states_295_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2671049664)))]; + tensor hidden_states_295_beta_0_to_fp16 = const()[name = tensor("hidden_states_295_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2671052288)))]; + tensor var_31817_to_fp16 = const()[name = tensor("op_31817_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_295_cast_fp16 = layer_norm(axes = hidden_states_295_axes_0, beta = hidden_states_295_beta_0_to_fp16, epsilon = var_31817_to_fp16, gamma = hidden_states_295_gamma_0_to_fp16, x = inputs_213_cast_fp16)[name = tensor("hidden_states_295_cast_fp16")]; + tensor var_31832 = const()[name = tensor("op_31832"), val = tensor([1, 1])]; + tensor var_31834 = const()[name = tensor("op_31834"), val = tensor([1, 1])]; + tensor q_143_pad_type_0 = const()[name = tensor("q_143_pad_type_0"), val = tensor("custom")]; + tensor q_143_pad_0 = const()[name = tensor("q_143_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_1_attn2_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_1_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2671054912)))]; + tensor q_143_cast_fp16 = conv(dilations = var_31834, groups = var_30355, pad = q_143_pad_0, pad_type = q_143_pad_type_0, strides = var_31832, weight = up_blocks_0_attentions_0_transformer_blocks_1_attn2_to_q_weight_to_fp16, x = hidden_states_295_cast_fp16)[name = tensor("q_143_cast_fp16")]; + tensor var_31838 = const()[name = tensor("op_31838"), val = tensor([1, 1])]; + tensor var_31840 = const()[name = tensor("op_31840"), val = tensor([1, 1])]; + tensor k_285_pad_type_0 = const()[name = tensor("k_285_pad_type_0"), val = tensor("custom")]; + tensor k_285_pad_0 = const()[name = tensor("k_285_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_1_attn2_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_1_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2674331776)))]; + tensor k_285_cast_fp16 = conv(dilations = var_31840, groups = var_30355, pad = k_285_pad_0, pad_type = k_285_pad_type_0, strides = var_31838, weight = up_blocks_0_attentions_0_transformer_blocks_1_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_285_cast_fp16")]; + tensor var_31844 = const()[name = tensor("op_31844"), val = tensor([1, 1])]; + tensor var_31846 = const()[name = tensor("op_31846"), val = tensor([1, 1])]; + tensor v_143_pad_type_0 = const()[name = tensor("v_143_pad_type_0"), val = tensor("custom")]; + tensor v_143_pad_0 = const()[name = tensor("v_143_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_1_attn2_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_1_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2679574720)))]; + tensor v_143_cast_fp16 = conv(dilations = var_31846, groups = var_30355, pad = v_143_pad_0, pad_type = v_143_pad_type_0, strides = var_31844, weight = up_blocks_0_attentions_0_transformer_blocks_1_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_143_cast_fp16")]; + tensor var_31850_begin_0 = const()[name = tensor("op_31850_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_31850_end_0 = const()[name = tensor("op_31850_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_31850_end_mask_0 = const()[name = tensor("op_31850_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31850_cast_fp16 = slice_by_index(begin = var_31850_begin_0, end = var_31850_end_0, end_mask = var_31850_end_mask_0, x = q_143_cast_fp16)[name = tensor("op_31850_cast_fp16")]; + tensor var_31854_begin_0 = const()[name = tensor("op_31854_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_31854_end_0 = const()[name = tensor("op_31854_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_31854_end_mask_0 = const()[name = tensor("op_31854_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31854_cast_fp16 = slice_by_index(begin = var_31854_begin_0, end = var_31854_end_0, end_mask = var_31854_end_mask_0, x = q_143_cast_fp16)[name = tensor("op_31854_cast_fp16")]; + tensor var_31858_begin_0 = const()[name = tensor("op_31858_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_31858_end_0 = const()[name = tensor("op_31858_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_31858_end_mask_0 = const()[name = tensor("op_31858_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31858_cast_fp16 = slice_by_index(begin = var_31858_begin_0, end = var_31858_end_0, end_mask = var_31858_end_mask_0, x = q_143_cast_fp16)[name = tensor("op_31858_cast_fp16")]; + tensor var_31862_begin_0 = const()[name = tensor("op_31862_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_31862_end_0 = const()[name = tensor("op_31862_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_31862_end_mask_0 = const()[name = tensor("op_31862_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31862_cast_fp16 = slice_by_index(begin = var_31862_begin_0, end = var_31862_end_0, end_mask = var_31862_end_mask_0, x = q_143_cast_fp16)[name = tensor("op_31862_cast_fp16")]; + tensor var_31866_begin_0 = const()[name = tensor("op_31866_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_31866_end_0 = const()[name = tensor("op_31866_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_31866_end_mask_0 = const()[name = tensor("op_31866_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31866_cast_fp16 = slice_by_index(begin = var_31866_begin_0, end = var_31866_end_0, end_mask = var_31866_end_mask_0, x = q_143_cast_fp16)[name = tensor("op_31866_cast_fp16")]; + tensor var_31870_begin_0 = const()[name = tensor("op_31870_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_31870_end_0 = const()[name = tensor("op_31870_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_31870_end_mask_0 = const()[name = tensor("op_31870_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31870_cast_fp16 = slice_by_index(begin = var_31870_begin_0, end = var_31870_end_0, end_mask = var_31870_end_mask_0, x = q_143_cast_fp16)[name = tensor("op_31870_cast_fp16")]; + tensor var_31874_begin_0 = const()[name = tensor("op_31874_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_31874_end_0 = const()[name = tensor("op_31874_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_31874_end_mask_0 = const()[name = tensor("op_31874_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31874_cast_fp16 = slice_by_index(begin = var_31874_begin_0, end = var_31874_end_0, end_mask = var_31874_end_mask_0, x = q_143_cast_fp16)[name = tensor("op_31874_cast_fp16")]; + tensor var_31878_begin_0 = const()[name = tensor("op_31878_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_31878_end_0 = const()[name = tensor("op_31878_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_31878_end_mask_0 = const()[name = tensor("op_31878_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31878_cast_fp16 = slice_by_index(begin = var_31878_begin_0, end = var_31878_end_0, end_mask = var_31878_end_mask_0, x = q_143_cast_fp16)[name = tensor("op_31878_cast_fp16")]; + tensor var_31882_begin_0 = const()[name = tensor("op_31882_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_31882_end_0 = const()[name = tensor("op_31882_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_31882_end_mask_0 = const()[name = tensor("op_31882_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31882_cast_fp16 = slice_by_index(begin = var_31882_begin_0, end = var_31882_end_0, end_mask = var_31882_end_mask_0, x = q_143_cast_fp16)[name = tensor("op_31882_cast_fp16")]; + tensor var_31886_begin_0 = const()[name = tensor("op_31886_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_31886_end_0 = const()[name = tensor("op_31886_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_31886_end_mask_0 = const()[name = tensor("op_31886_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31886_cast_fp16 = slice_by_index(begin = var_31886_begin_0, end = var_31886_end_0, end_mask = var_31886_end_mask_0, x = q_143_cast_fp16)[name = tensor("op_31886_cast_fp16")]; + tensor var_31890_begin_0 = const()[name = tensor("op_31890_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_31890_end_0 = const()[name = tensor("op_31890_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_31890_end_mask_0 = const()[name = tensor("op_31890_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31890_cast_fp16 = slice_by_index(begin = var_31890_begin_0, end = var_31890_end_0, end_mask = var_31890_end_mask_0, x = q_143_cast_fp16)[name = tensor("op_31890_cast_fp16")]; + tensor var_31894_begin_0 = const()[name = tensor("op_31894_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_31894_end_0 = const()[name = tensor("op_31894_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_31894_end_mask_0 = const()[name = tensor("op_31894_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31894_cast_fp16 = slice_by_index(begin = var_31894_begin_0, end = var_31894_end_0, end_mask = var_31894_end_mask_0, x = q_143_cast_fp16)[name = tensor("op_31894_cast_fp16")]; + tensor var_31898_begin_0 = const()[name = tensor("op_31898_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_31898_end_0 = const()[name = tensor("op_31898_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_31898_end_mask_0 = const()[name = tensor("op_31898_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31898_cast_fp16 = slice_by_index(begin = var_31898_begin_0, end = var_31898_end_0, end_mask = var_31898_end_mask_0, x = q_143_cast_fp16)[name = tensor("op_31898_cast_fp16")]; + tensor var_31902_begin_0 = const()[name = tensor("op_31902_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_31902_end_0 = const()[name = tensor("op_31902_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_31902_end_mask_0 = const()[name = tensor("op_31902_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31902_cast_fp16 = slice_by_index(begin = var_31902_begin_0, end = var_31902_end_0, end_mask = var_31902_end_mask_0, x = q_143_cast_fp16)[name = tensor("op_31902_cast_fp16")]; + tensor var_31906_begin_0 = const()[name = tensor("op_31906_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_31906_end_0 = const()[name = tensor("op_31906_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_31906_end_mask_0 = const()[name = tensor("op_31906_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31906_cast_fp16 = slice_by_index(begin = var_31906_begin_0, end = var_31906_end_0, end_mask = var_31906_end_mask_0, x = q_143_cast_fp16)[name = tensor("op_31906_cast_fp16")]; + tensor var_31910_begin_0 = const()[name = tensor("op_31910_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_31910_end_0 = const()[name = tensor("op_31910_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_31910_end_mask_0 = const()[name = tensor("op_31910_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31910_cast_fp16 = slice_by_index(begin = var_31910_begin_0, end = var_31910_end_0, end_mask = var_31910_end_mask_0, x = q_143_cast_fp16)[name = tensor("op_31910_cast_fp16")]; + tensor var_31914_begin_0 = const()[name = tensor("op_31914_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_31914_end_0 = const()[name = tensor("op_31914_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_31914_end_mask_0 = const()[name = tensor("op_31914_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31914_cast_fp16 = slice_by_index(begin = var_31914_begin_0, end = var_31914_end_0, end_mask = var_31914_end_mask_0, x = q_143_cast_fp16)[name = tensor("op_31914_cast_fp16")]; + tensor var_31918_begin_0 = const()[name = tensor("op_31918_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_31918_end_0 = const()[name = tensor("op_31918_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_31918_end_mask_0 = const()[name = tensor("op_31918_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31918_cast_fp16 = slice_by_index(begin = var_31918_begin_0, end = var_31918_end_0, end_mask = var_31918_end_mask_0, x = q_143_cast_fp16)[name = tensor("op_31918_cast_fp16")]; + tensor var_31922_begin_0 = const()[name = tensor("op_31922_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_31922_end_0 = const()[name = tensor("op_31922_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_31922_end_mask_0 = const()[name = tensor("op_31922_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31922_cast_fp16 = slice_by_index(begin = var_31922_begin_0, end = var_31922_end_0, end_mask = var_31922_end_mask_0, x = q_143_cast_fp16)[name = tensor("op_31922_cast_fp16")]; + tensor var_31926_begin_0 = const()[name = tensor("op_31926_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_31926_end_0 = const()[name = tensor("op_31926_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_31926_end_mask_0 = const()[name = tensor("op_31926_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31926_cast_fp16 = slice_by_index(begin = var_31926_begin_0, end = var_31926_end_0, end_mask = var_31926_end_mask_0, x = q_143_cast_fp16)[name = tensor("op_31926_cast_fp16")]; + tensor k_287_perm_0 = const()[name = tensor("k_287_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_31933_begin_0 = const()[name = tensor("op_31933_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_31933_end_0 = const()[name = tensor("op_31933_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_31933_end_mask_0 = const()[name = tensor("op_31933_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_287_cast_fp16 = transpose(perm = k_287_perm_0, x = k_285_cast_fp16)[name = tensor("transpose_68")]; + tensor var_31933_cast_fp16 = slice_by_index(begin = var_31933_begin_0, end = var_31933_end_0, end_mask = var_31933_end_mask_0, x = k_287_cast_fp16)[name = tensor("op_31933_cast_fp16")]; + tensor var_31937_begin_0 = const()[name = tensor("op_31937_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_31937_end_0 = const()[name = tensor("op_31937_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_31937_end_mask_0 = const()[name = tensor("op_31937_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31937_cast_fp16 = slice_by_index(begin = var_31937_begin_0, end = var_31937_end_0, end_mask = var_31937_end_mask_0, x = k_287_cast_fp16)[name = tensor("op_31937_cast_fp16")]; + tensor var_31941_begin_0 = const()[name = tensor("op_31941_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_31941_end_0 = const()[name = tensor("op_31941_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_31941_end_mask_0 = const()[name = tensor("op_31941_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31941_cast_fp16 = slice_by_index(begin = var_31941_begin_0, end = var_31941_end_0, end_mask = var_31941_end_mask_0, x = k_287_cast_fp16)[name = tensor("op_31941_cast_fp16")]; + tensor var_31945_begin_0 = const()[name = tensor("op_31945_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_31945_end_0 = const()[name = tensor("op_31945_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_31945_end_mask_0 = const()[name = tensor("op_31945_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31945_cast_fp16 = slice_by_index(begin = var_31945_begin_0, end = var_31945_end_0, end_mask = var_31945_end_mask_0, x = k_287_cast_fp16)[name = tensor("op_31945_cast_fp16")]; + tensor var_31949_begin_0 = const()[name = tensor("op_31949_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_31949_end_0 = const()[name = tensor("op_31949_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_31949_end_mask_0 = const()[name = tensor("op_31949_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31949_cast_fp16 = slice_by_index(begin = var_31949_begin_0, end = var_31949_end_0, end_mask = var_31949_end_mask_0, x = k_287_cast_fp16)[name = tensor("op_31949_cast_fp16")]; + tensor var_31953_begin_0 = const()[name = tensor("op_31953_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_31953_end_0 = const()[name = tensor("op_31953_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_31953_end_mask_0 = const()[name = tensor("op_31953_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31953_cast_fp16 = slice_by_index(begin = var_31953_begin_0, end = var_31953_end_0, end_mask = var_31953_end_mask_0, x = k_287_cast_fp16)[name = tensor("op_31953_cast_fp16")]; + tensor var_31957_begin_0 = const()[name = tensor("op_31957_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_31957_end_0 = const()[name = tensor("op_31957_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_31957_end_mask_0 = const()[name = tensor("op_31957_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31957_cast_fp16 = slice_by_index(begin = var_31957_begin_0, end = var_31957_end_0, end_mask = var_31957_end_mask_0, x = k_287_cast_fp16)[name = tensor("op_31957_cast_fp16")]; + tensor var_31961_begin_0 = const()[name = tensor("op_31961_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_31961_end_0 = const()[name = tensor("op_31961_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_31961_end_mask_0 = const()[name = tensor("op_31961_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31961_cast_fp16 = slice_by_index(begin = var_31961_begin_0, end = var_31961_end_0, end_mask = var_31961_end_mask_0, x = k_287_cast_fp16)[name = tensor("op_31961_cast_fp16")]; + tensor var_31965_begin_0 = const()[name = tensor("op_31965_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_31965_end_0 = const()[name = tensor("op_31965_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_31965_end_mask_0 = const()[name = tensor("op_31965_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31965_cast_fp16 = slice_by_index(begin = var_31965_begin_0, end = var_31965_end_0, end_mask = var_31965_end_mask_0, x = k_287_cast_fp16)[name = tensor("op_31965_cast_fp16")]; + tensor var_31969_begin_0 = const()[name = tensor("op_31969_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_31969_end_0 = const()[name = tensor("op_31969_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_31969_end_mask_0 = const()[name = tensor("op_31969_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31969_cast_fp16 = slice_by_index(begin = var_31969_begin_0, end = var_31969_end_0, end_mask = var_31969_end_mask_0, x = k_287_cast_fp16)[name = tensor("op_31969_cast_fp16")]; + tensor var_31973_begin_0 = const()[name = tensor("op_31973_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_31973_end_0 = const()[name = tensor("op_31973_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_31973_end_mask_0 = const()[name = tensor("op_31973_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31973_cast_fp16 = slice_by_index(begin = var_31973_begin_0, end = var_31973_end_0, end_mask = var_31973_end_mask_0, x = k_287_cast_fp16)[name = tensor("op_31973_cast_fp16")]; + tensor var_31977_begin_0 = const()[name = tensor("op_31977_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_31977_end_0 = const()[name = tensor("op_31977_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_31977_end_mask_0 = const()[name = tensor("op_31977_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31977_cast_fp16 = slice_by_index(begin = var_31977_begin_0, end = var_31977_end_0, end_mask = var_31977_end_mask_0, x = k_287_cast_fp16)[name = tensor("op_31977_cast_fp16")]; + tensor var_31981_begin_0 = const()[name = tensor("op_31981_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_31981_end_0 = const()[name = tensor("op_31981_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_31981_end_mask_0 = const()[name = tensor("op_31981_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31981_cast_fp16 = slice_by_index(begin = var_31981_begin_0, end = var_31981_end_0, end_mask = var_31981_end_mask_0, x = k_287_cast_fp16)[name = tensor("op_31981_cast_fp16")]; + tensor var_31985_begin_0 = const()[name = tensor("op_31985_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_31985_end_0 = const()[name = tensor("op_31985_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_31985_end_mask_0 = const()[name = tensor("op_31985_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31985_cast_fp16 = slice_by_index(begin = var_31985_begin_0, end = var_31985_end_0, end_mask = var_31985_end_mask_0, x = k_287_cast_fp16)[name = tensor("op_31985_cast_fp16")]; + tensor var_31989_begin_0 = const()[name = tensor("op_31989_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_31989_end_0 = const()[name = tensor("op_31989_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_31989_end_mask_0 = const()[name = tensor("op_31989_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31989_cast_fp16 = slice_by_index(begin = var_31989_begin_0, end = var_31989_end_0, end_mask = var_31989_end_mask_0, x = k_287_cast_fp16)[name = tensor("op_31989_cast_fp16")]; + tensor var_31993_begin_0 = const()[name = tensor("op_31993_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_31993_end_0 = const()[name = tensor("op_31993_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_31993_end_mask_0 = const()[name = tensor("op_31993_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31993_cast_fp16 = slice_by_index(begin = var_31993_begin_0, end = var_31993_end_0, end_mask = var_31993_end_mask_0, x = k_287_cast_fp16)[name = tensor("op_31993_cast_fp16")]; + tensor var_31997_begin_0 = const()[name = tensor("op_31997_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_31997_end_0 = const()[name = tensor("op_31997_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_31997_end_mask_0 = const()[name = tensor("op_31997_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31997_cast_fp16 = slice_by_index(begin = var_31997_begin_0, end = var_31997_end_0, end_mask = var_31997_end_mask_0, x = k_287_cast_fp16)[name = tensor("op_31997_cast_fp16")]; + tensor var_32001_begin_0 = const()[name = tensor("op_32001_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_32001_end_0 = const()[name = tensor("op_32001_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_32001_end_mask_0 = const()[name = tensor("op_32001_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32001_cast_fp16 = slice_by_index(begin = var_32001_begin_0, end = var_32001_end_0, end_mask = var_32001_end_mask_0, x = k_287_cast_fp16)[name = tensor("op_32001_cast_fp16")]; + tensor var_32005_begin_0 = const()[name = tensor("op_32005_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_32005_end_0 = const()[name = tensor("op_32005_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_32005_end_mask_0 = const()[name = tensor("op_32005_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32005_cast_fp16 = slice_by_index(begin = var_32005_begin_0, end = var_32005_end_0, end_mask = var_32005_end_mask_0, x = k_287_cast_fp16)[name = tensor("op_32005_cast_fp16")]; + tensor var_32009_begin_0 = const()[name = tensor("op_32009_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_32009_end_0 = const()[name = tensor("op_32009_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_32009_end_mask_0 = const()[name = tensor("op_32009_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32009_cast_fp16 = slice_by_index(begin = var_32009_begin_0, end = var_32009_end_0, end_mask = var_32009_end_mask_0, x = k_287_cast_fp16)[name = tensor("op_32009_cast_fp16")]; + tensor var_32011_begin_0 = const()[name = tensor("op_32011_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_32011_end_0 = const()[name = tensor("op_32011_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_32011_end_mask_0 = const()[name = tensor("op_32011_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32011_cast_fp16 = slice_by_index(begin = var_32011_begin_0, end = var_32011_end_0, end_mask = var_32011_end_mask_0, x = v_143_cast_fp16)[name = tensor("op_32011_cast_fp16")]; + tensor var_32015_begin_0 = const()[name = tensor("op_32015_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_32015_end_0 = const()[name = tensor("op_32015_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_32015_end_mask_0 = const()[name = tensor("op_32015_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32015_cast_fp16 = slice_by_index(begin = var_32015_begin_0, end = var_32015_end_0, end_mask = var_32015_end_mask_0, x = v_143_cast_fp16)[name = tensor("op_32015_cast_fp16")]; + tensor var_32019_begin_0 = const()[name = tensor("op_32019_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_32019_end_0 = const()[name = tensor("op_32019_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_32019_end_mask_0 = const()[name = tensor("op_32019_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32019_cast_fp16 = slice_by_index(begin = var_32019_begin_0, end = var_32019_end_0, end_mask = var_32019_end_mask_0, x = v_143_cast_fp16)[name = tensor("op_32019_cast_fp16")]; + tensor var_32023_begin_0 = const()[name = tensor("op_32023_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_32023_end_0 = const()[name = tensor("op_32023_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_32023_end_mask_0 = const()[name = tensor("op_32023_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32023_cast_fp16 = slice_by_index(begin = var_32023_begin_0, end = var_32023_end_0, end_mask = var_32023_end_mask_0, x = v_143_cast_fp16)[name = tensor("op_32023_cast_fp16")]; + tensor var_32027_begin_0 = const()[name = tensor("op_32027_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_32027_end_0 = const()[name = tensor("op_32027_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_32027_end_mask_0 = const()[name = tensor("op_32027_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32027_cast_fp16 = slice_by_index(begin = var_32027_begin_0, end = var_32027_end_0, end_mask = var_32027_end_mask_0, x = v_143_cast_fp16)[name = tensor("op_32027_cast_fp16")]; + tensor var_32031_begin_0 = const()[name = tensor("op_32031_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_32031_end_0 = const()[name = tensor("op_32031_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_32031_end_mask_0 = const()[name = tensor("op_32031_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32031_cast_fp16 = slice_by_index(begin = var_32031_begin_0, end = var_32031_end_0, end_mask = var_32031_end_mask_0, x = v_143_cast_fp16)[name = tensor("op_32031_cast_fp16")]; + tensor var_32035_begin_0 = const()[name = tensor("op_32035_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_32035_end_0 = const()[name = tensor("op_32035_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_32035_end_mask_0 = const()[name = tensor("op_32035_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32035_cast_fp16 = slice_by_index(begin = var_32035_begin_0, end = var_32035_end_0, end_mask = var_32035_end_mask_0, x = v_143_cast_fp16)[name = tensor("op_32035_cast_fp16")]; + tensor var_32039_begin_0 = const()[name = tensor("op_32039_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_32039_end_0 = const()[name = tensor("op_32039_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_32039_end_mask_0 = const()[name = tensor("op_32039_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32039_cast_fp16 = slice_by_index(begin = var_32039_begin_0, end = var_32039_end_0, end_mask = var_32039_end_mask_0, x = v_143_cast_fp16)[name = tensor("op_32039_cast_fp16")]; + tensor var_32043_begin_0 = const()[name = tensor("op_32043_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_32043_end_0 = const()[name = tensor("op_32043_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_32043_end_mask_0 = const()[name = tensor("op_32043_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32043_cast_fp16 = slice_by_index(begin = var_32043_begin_0, end = var_32043_end_0, end_mask = var_32043_end_mask_0, x = v_143_cast_fp16)[name = tensor("op_32043_cast_fp16")]; + tensor var_32047_begin_0 = const()[name = tensor("op_32047_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_32047_end_0 = const()[name = tensor("op_32047_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_32047_end_mask_0 = const()[name = tensor("op_32047_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32047_cast_fp16 = slice_by_index(begin = var_32047_begin_0, end = var_32047_end_0, end_mask = var_32047_end_mask_0, x = v_143_cast_fp16)[name = tensor("op_32047_cast_fp16")]; + tensor var_32051_begin_0 = const()[name = tensor("op_32051_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_32051_end_0 = const()[name = tensor("op_32051_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_32051_end_mask_0 = const()[name = tensor("op_32051_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32051_cast_fp16 = slice_by_index(begin = var_32051_begin_0, end = var_32051_end_0, end_mask = var_32051_end_mask_0, x = v_143_cast_fp16)[name = tensor("op_32051_cast_fp16")]; + tensor var_32055_begin_0 = const()[name = tensor("op_32055_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_32055_end_0 = const()[name = tensor("op_32055_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_32055_end_mask_0 = const()[name = tensor("op_32055_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32055_cast_fp16 = slice_by_index(begin = var_32055_begin_0, end = var_32055_end_0, end_mask = var_32055_end_mask_0, x = v_143_cast_fp16)[name = tensor("op_32055_cast_fp16")]; + tensor var_32059_begin_0 = const()[name = tensor("op_32059_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_32059_end_0 = const()[name = tensor("op_32059_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_32059_end_mask_0 = const()[name = tensor("op_32059_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32059_cast_fp16 = slice_by_index(begin = var_32059_begin_0, end = var_32059_end_0, end_mask = var_32059_end_mask_0, x = v_143_cast_fp16)[name = tensor("op_32059_cast_fp16")]; + tensor var_32063_begin_0 = const()[name = tensor("op_32063_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_32063_end_0 = const()[name = tensor("op_32063_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_32063_end_mask_0 = const()[name = tensor("op_32063_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32063_cast_fp16 = slice_by_index(begin = var_32063_begin_0, end = var_32063_end_0, end_mask = var_32063_end_mask_0, x = v_143_cast_fp16)[name = tensor("op_32063_cast_fp16")]; + tensor var_32067_begin_0 = const()[name = tensor("op_32067_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_32067_end_0 = const()[name = tensor("op_32067_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_32067_end_mask_0 = const()[name = tensor("op_32067_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32067_cast_fp16 = slice_by_index(begin = var_32067_begin_0, end = var_32067_end_0, end_mask = var_32067_end_mask_0, x = v_143_cast_fp16)[name = tensor("op_32067_cast_fp16")]; + tensor var_32071_begin_0 = const()[name = tensor("op_32071_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_32071_end_0 = const()[name = tensor("op_32071_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_32071_end_mask_0 = const()[name = tensor("op_32071_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32071_cast_fp16 = slice_by_index(begin = var_32071_begin_0, end = var_32071_end_0, end_mask = var_32071_end_mask_0, x = v_143_cast_fp16)[name = tensor("op_32071_cast_fp16")]; + tensor var_32075_begin_0 = const()[name = tensor("op_32075_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_32075_end_0 = const()[name = tensor("op_32075_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_32075_end_mask_0 = const()[name = tensor("op_32075_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32075_cast_fp16 = slice_by_index(begin = var_32075_begin_0, end = var_32075_end_0, end_mask = var_32075_end_mask_0, x = v_143_cast_fp16)[name = tensor("op_32075_cast_fp16")]; + tensor var_32079_begin_0 = const()[name = tensor("op_32079_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_32079_end_0 = const()[name = tensor("op_32079_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_32079_end_mask_0 = const()[name = tensor("op_32079_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32079_cast_fp16 = slice_by_index(begin = var_32079_begin_0, end = var_32079_end_0, end_mask = var_32079_end_mask_0, x = v_143_cast_fp16)[name = tensor("op_32079_cast_fp16")]; + tensor var_32083_begin_0 = const()[name = tensor("op_32083_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_32083_end_0 = const()[name = tensor("op_32083_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_32083_end_mask_0 = const()[name = tensor("op_32083_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32083_cast_fp16 = slice_by_index(begin = var_32083_begin_0, end = var_32083_end_0, end_mask = var_32083_end_mask_0, x = v_143_cast_fp16)[name = tensor("op_32083_cast_fp16")]; + tensor var_32087_begin_0 = const()[name = tensor("op_32087_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_32087_end_0 = const()[name = tensor("op_32087_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_32087_end_mask_0 = const()[name = tensor("op_32087_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32087_cast_fp16 = slice_by_index(begin = var_32087_begin_0, end = var_32087_end_0, end_mask = var_32087_end_mask_0, x = v_143_cast_fp16)[name = tensor("op_32087_cast_fp16")]; + tensor var_32091_equation_0 = const()[name = tensor("op_32091_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32091_cast_fp16 = einsum(equation = var_32091_equation_0, values = (var_31933_cast_fp16, var_31850_cast_fp16))[name = tensor("op_32091_cast_fp16")]; + tensor var_32092_to_fp16 = const()[name = tensor("op_32092_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2681_cast_fp16 = mul(x = var_32091_cast_fp16, y = var_32092_to_fp16)[name = tensor("aw_2681_cast_fp16")]; + tensor var_32095_equation_0 = const()[name = tensor("op_32095_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32095_cast_fp16 = einsum(equation = var_32095_equation_0, values = (var_31937_cast_fp16, var_31854_cast_fp16))[name = tensor("op_32095_cast_fp16")]; + tensor var_32096_to_fp16 = const()[name = tensor("op_32096_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2683_cast_fp16 = mul(x = var_32095_cast_fp16, y = var_32096_to_fp16)[name = tensor("aw_2683_cast_fp16")]; + tensor var_32099_equation_0 = const()[name = tensor("op_32099_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32099_cast_fp16 = einsum(equation = var_32099_equation_0, values = (var_31941_cast_fp16, var_31858_cast_fp16))[name = tensor("op_32099_cast_fp16")]; + tensor var_32100_to_fp16 = const()[name = tensor("op_32100_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2685_cast_fp16 = mul(x = var_32099_cast_fp16, y = var_32100_to_fp16)[name = tensor("aw_2685_cast_fp16")]; + tensor var_32103_equation_0 = const()[name = tensor("op_32103_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32103_cast_fp16 = einsum(equation = var_32103_equation_0, values = (var_31945_cast_fp16, var_31862_cast_fp16))[name = tensor("op_32103_cast_fp16")]; + tensor var_32104_to_fp16 = const()[name = tensor("op_32104_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2687_cast_fp16 = mul(x = var_32103_cast_fp16, y = var_32104_to_fp16)[name = tensor("aw_2687_cast_fp16")]; + tensor var_32107_equation_0 = const()[name = tensor("op_32107_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32107_cast_fp16 = einsum(equation = var_32107_equation_0, values = (var_31949_cast_fp16, var_31866_cast_fp16))[name = tensor("op_32107_cast_fp16")]; + tensor var_32108_to_fp16 = const()[name = tensor("op_32108_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2689_cast_fp16 = mul(x = var_32107_cast_fp16, y = var_32108_to_fp16)[name = tensor("aw_2689_cast_fp16")]; + tensor var_32111_equation_0 = const()[name = tensor("op_32111_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32111_cast_fp16 = einsum(equation = var_32111_equation_0, values = (var_31953_cast_fp16, var_31870_cast_fp16))[name = tensor("op_32111_cast_fp16")]; + tensor var_32112_to_fp16 = const()[name = tensor("op_32112_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2691_cast_fp16 = mul(x = var_32111_cast_fp16, y = var_32112_to_fp16)[name = tensor("aw_2691_cast_fp16")]; + tensor var_32115_equation_0 = const()[name = tensor("op_32115_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32115_cast_fp16 = einsum(equation = var_32115_equation_0, values = (var_31957_cast_fp16, var_31874_cast_fp16))[name = tensor("op_32115_cast_fp16")]; + tensor var_32116_to_fp16 = const()[name = tensor("op_32116_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2693_cast_fp16 = mul(x = var_32115_cast_fp16, y = var_32116_to_fp16)[name = tensor("aw_2693_cast_fp16")]; + tensor var_32119_equation_0 = const()[name = tensor("op_32119_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32119_cast_fp16 = einsum(equation = var_32119_equation_0, values = (var_31961_cast_fp16, var_31878_cast_fp16))[name = tensor("op_32119_cast_fp16")]; + tensor var_32120_to_fp16 = const()[name = tensor("op_32120_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2695_cast_fp16 = mul(x = var_32119_cast_fp16, y = var_32120_to_fp16)[name = tensor("aw_2695_cast_fp16")]; + tensor var_32123_equation_0 = const()[name = tensor("op_32123_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32123_cast_fp16 = einsum(equation = var_32123_equation_0, values = (var_31965_cast_fp16, var_31882_cast_fp16))[name = tensor("op_32123_cast_fp16")]; + tensor var_32124_to_fp16 = const()[name = tensor("op_32124_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2697_cast_fp16 = mul(x = var_32123_cast_fp16, y = var_32124_to_fp16)[name = tensor("aw_2697_cast_fp16")]; + tensor var_32127_equation_0 = const()[name = tensor("op_32127_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32127_cast_fp16 = einsum(equation = var_32127_equation_0, values = (var_31969_cast_fp16, var_31886_cast_fp16))[name = tensor("op_32127_cast_fp16")]; + tensor var_32128_to_fp16 = const()[name = tensor("op_32128_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2699_cast_fp16 = mul(x = var_32127_cast_fp16, y = var_32128_to_fp16)[name = tensor("aw_2699_cast_fp16")]; + tensor var_32131_equation_0 = const()[name = tensor("op_32131_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32131_cast_fp16 = einsum(equation = var_32131_equation_0, values = (var_31973_cast_fp16, var_31890_cast_fp16))[name = tensor("op_32131_cast_fp16")]; + tensor var_32132_to_fp16 = const()[name = tensor("op_32132_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2701_cast_fp16 = mul(x = var_32131_cast_fp16, y = var_32132_to_fp16)[name = tensor("aw_2701_cast_fp16")]; + tensor var_32135_equation_0 = const()[name = tensor("op_32135_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32135_cast_fp16 = einsum(equation = var_32135_equation_0, values = (var_31977_cast_fp16, var_31894_cast_fp16))[name = tensor("op_32135_cast_fp16")]; + tensor var_32136_to_fp16 = const()[name = tensor("op_32136_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2703_cast_fp16 = mul(x = var_32135_cast_fp16, y = var_32136_to_fp16)[name = tensor("aw_2703_cast_fp16")]; + tensor var_32139_equation_0 = const()[name = tensor("op_32139_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32139_cast_fp16 = einsum(equation = var_32139_equation_0, values = (var_31981_cast_fp16, var_31898_cast_fp16))[name = tensor("op_32139_cast_fp16")]; + tensor var_32140_to_fp16 = const()[name = tensor("op_32140_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2705_cast_fp16 = mul(x = var_32139_cast_fp16, y = var_32140_to_fp16)[name = tensor("aw_2705_cast_fp16")]; + tensor var_32143_equation_0 = const()[name = tensor("op_32143_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32143_cast_fp16 = einsum(equation = var_32143_equation_0, values = (var_31985_cast_fp16, var_31902_cast_fp16))[name = tensor("op_32143_cast_fp16")]; + tensor var_32144_to_fp16 = const()[name = tensor("op_32144_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2707_cast_fp16 = mul(x = var_32143_cast_fp16, y = var_32144_to_fp16)[name = tensor("aw_2707_cast_fp16")]; + tensor var_32147_equation_0 = const()[name = tensor("op_32147_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32147_cast_fp16 = einsum(equation = var_32147_equation_0, values = (var_31989_cast_fp16, var_31906_cast_fp16))[name = tensor("op_32147_cast_fp16")]; + tensor var_32148_to_fp16 = const()[name = tensor("op_32148_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2709_cast_fp16 = mul(x = var_32147_cast_fp16, y = var_32148_to_fp16)[name = tensor("aw_2709_cast_fp16")]; + tensor var_32151_equation_0 = const()[name = tensor("op_32151_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32151_cast_fp16 = einsum(equation = var_32151_equation_0, values = (var_31993_cast_fp16, var_31910_cast_fp16))[name = tensor("op_32151_cast_fp16")]; + tensor var_32152_to_fp16 = const()[name = tensor("op_32152_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2711_cast_fp16 = mul(x = var_32151_cast_fp16, y = var_32152_to_fp16)[name = tensor("aw_2711_cast_fp16")]; + tensor var_32155_equation_0 = const()[name = tensor("op_32155_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32155_cast_fp16 = einsum(equation = var_32155_equation_0, values = (var_31997_cast_fp16, var_31914_cast_fp16))[name = tensor("op_32155_cast_fp16")]; + tensor var_32156_to_fp16 = const()[name = tensor("op_32156_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2713_cast_fp16 = mul(x = var_32155_cast_fp16, y = var_32156_to_fp16)[name = tensor("aw_2713_cast_fp16")]; + tensor var_32159_equation_0 = const()[name = tensor("op_32159_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32159_cast_fp16 = einsum(equation = var_32159_equation_0, values = (var_32001_cast_fp16, var_31918_cast_fp16))[name = tensor("op_32159_cast_fp16")]; + tensor var_32160_to_fp16 = const()[name = tensor("op_32160_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2715_cast_fp16 = mul(x = var_32159_cast_fp16, y = var_32160_to_fp16)[name = tensor("aw_2715_cast_fp16")]; + tensor var_32163_equation_0 = const()[name = tensor("op_32163_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32163_cast_fp16 = einsum(equation = var_32163_equation_0, values = (var_32005_cast_fp16, var_31922_cast_fp16))[name = tensor("op_32163_cast_fp16")]; + tensor var_32164_to_fp16 = const()[name = tensor("op_32164_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2717_cast_fp16 = mul(x = var_32163_cast_fp16, y = var_32164_to_fp16)[name = tensor("aw_2717_cast_fp16")]; + tensor var_32167_equation_0 = const()[name = tensor("op_32167_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32167_cast_fp16 = einsum(equation = var_32167_equation_0, values = (var_32009_cast_fp16, var_31926_cast_fp16))[name = tensor("op_32167_cast_fp16")]; + tensor var_32168_to_fp16 = const()[name = tensor("op_32168_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2719_cast_fp16 = mul(x = var_32167_cast_fp16, y = var_32168_to_fp16)[name = tensor("aw_2719_cast_fp16")]; + tensor var_32170_cast_fp16 = softmax(axis = var_30355, x = aw_2681_cast_fp16)[name = tensor("op_32170_cast_fp16")]; + tensor var_32171_cast_fp16 = softmax(axis = var_30355, x = aw_2683_cast_fp16)[name = tensor("op_32171_cast_fp16")]; + tensor var_32172_cast_fp16 = softmax(axis = var_30355, x = aw_2685_cast_fp16)[name = tensor("op_32172_cast_fp16")]; + tensor var_32173_cast_fp16 = softmax(axis = var_30355, x = aw_2687_cast_fp16)[name = tensor("op_32173_cast_fp16")]; + tensor var_32174_cast_fp16 = softmax(axis = var_30355, x = aw_2689_cast_fp16)[name = tensor("op_32174_cast_fp16")]; + tensor var_32175_cast_fp16 = softmax(axis = var_30355, x = aw_2691_cast_fp16)[name = tensor("op_32175_cast_fp16")]; + tensor var_32176_cast_fp16 = softmax(axis = var_30355, x = aw_2693_cast_fp16)[name = tensor("op_32176_cast_fp16")]; + tensor var_32177_cast_fp16 = softmax(axis = var_30355, x = aw_2695_cast_fp16)[name = tensor("op_32177_cast_fp16")]; + tensor var_32178_cast_fp16 = softmax(axis = var_30355, x = aw_2697_cast_fp16)[name = tensor("op_32178_cast_fp16")]; + tensor var_32179_cast_fp16 = softmax(axis = var_30355, x = aw_2699_cast_fp16)[name = tensor("op_32179_cast_fp16")]; + tensor var_32180_cast_fp16 = softmax(axis = var_30355, x = aw_2701_cast_fp16)[name = tensor("op_32180_cast_fp16")]; + tensor var_32181_cast_fp16 = softmax(axis = var_30355, x = aw_2703_cast_fp16)[name = tensor("op_32181_cast_fp16")]; + tensor var_32182_cast_fp16 = softmax(axis = var_30355, x = aw_2705_cast_fp16)[name = tensor("op_32182_cast_fp16")]; + tensor var_32183_cast_fp16 = softmax(axis = var_30355, x = aw_2707_cast_fp16)[name = tensor("op_32183_cast_fp16")]; + tensor var_32184_cast_fp16 = softmax(axis = var_30355, x = aw_2709_cast_fp16)[name = tensor("op_32184_cast_fp16")]; + tensor var_32185_cast_fp16 = softmax(axis = var_30355, x = aw_2711_cast_fp16)[name = tensor("op_32185_cast_fp16")]; + tensor var_32186_cast_fp16 = softmax(axis = var_30355, x = aw_2713_cast_fp16)[name = tensor("op_32186_cast_fp16")]; + tensor var_32187_cast_fp16 = softmax(axis = var_30355, x = aw_2715_cast_fp16)[name = tensor("op_32187_cast_fp16")]; + tensor var_32188_cast_fp16 = softmax(axis = var_30355, x = aw_2717_cast_fp16)[name = tensor("op_32188_cast_fp16")]; + tensor var_32189_cast_fp16 = softmax(axis = var_30355, x = aw_2719_cast_fp16)[name = tensor("op_32189_cast_fp16")]; + tensor var_32191_equation_0 = const()[name = tensor("op_32191_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32191_cast_fp16 = einsum(equation = var_32191_equation_0, values = (var_32011_cast_fp16, var_32170_cast_fp16))[name = tensor("op_32191_cast_fp16")]; + tensor var_32193_equation_0 = const()[name = tensor("op_32193_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32193_cast_fp16 = einsum(equation = var_32193_equation_0, values = (var_32015_cast_fp16, var_32171_cast_fp16))[name = tensor("op_32193_cast_fp16")]; + tensor var_32195_equation_0 = const()[name = tensor("op_32195_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32195_cast_fp16 = einsum(equation = var_32195_equation_0, values = (var_32019_cast_fp16, var_32172_cast_fp16))[name = tensor("op_32195_cast_fp16")]; + tensor var_32197_equation_0 = const()[name = tensor("op_32197_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32197_cast_fp16 = einsum(equation = var_32197_equation_0, values = (var_32023_cast_fp16, var_32173_cast_fp16))[name = tensor("op_32197_cast_fp16")]; + tensor var_32199_equation_0 = const()[name = tensor("op_32199_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32199_cast_fp16 = einsum(equation = var_32199_equation_0, values = (var_32027_cast_fp16, var_32174_cast_fp16))[name = tensor("op_32199_cast_fp16")]; + tensor var_32201_equation_0 = const()[name = tensor("op_32201_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32201_cast_fp16 = einsum(equation = var_32201_equation_0, values = (var_32031_cast_fp16, var_32175_cast_fp16))[name = tensor("op_32201_cast_fp16")]; + tensor var_32203_equation_0 = const()[name = tensor("op_32203_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32203_cast_fp16 = einsum(equation = var_32203_equation_0, values = (var_32035_cast_fp16, var_32176_cast_fp16))[name = tensor("op_32203_cast_fp16")]; + tensor var_32205_equation_0 = const()[name = tensor("op_32205_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32205_cast_fp16 = einsum(equation = var_32205_equation_0, values = (var_32039_cast_fp16, var_32177_cast_fp16))[name = tensor("op_32205_cast_fp16")]; + tensor var_32207_equation_0 = const()[name = tensor("op_32207_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32207_cast_fp16 = einsum(equation = var_32207_equation_0, values = (var_32043_cast_fp16, var_32178_cast_fp16))[name = tensor("op_32207_cast_fp16")]; + tensor var_32209_equation_0 = const()[name = tensor("op_32209_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32209_cast_fp16 = einsum(equation = var_32209_equation_0, values = (var_32047_cast_fp16, var_32179_cast_fp16))[name = tensor("op_32209_cast_fp16")]; + tensor var_32211_equation_0 = const()[name = tensor("op_32211_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32211_cast_fp16 = einsum(equation = var_32211_equation_0, values = (var_32051_cast_fp16, var_32180_cast_fp16))[name = tensor("op_32211_cast_fp16")]; + tensor var_32213_equation_0 = const()[name = tensor("op_32213_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32213_cast_fp16 = einsum(equation = var_32213_equation_0, values = (var_32055_cast_fp16, var_32181_cast_fp16))[name = tensor("op_32213_cast_fp16")]; + tensor var_32215_equation_0 = const()[name = tensor("op_32215_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32215_cast_fp16 = einsum(equation = var_32215_equation_0, values = (var_32059_cast_fp16, var_32182_cast_fp16))[name = tensor("op_32215_cast_fp16")]; + tensor var_32217_equation_0 = const()[name = tensor("op_32217_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32217_cast_fp16 = einsum(equation = var_32217_equation_0, values = (var_32063_cast_fp16, var_32183_cast_fp16))[name = tensor("op_32217_cast_fp16")]; + tensor var_32219_equation_0 = const()[name = tensor("op_32219_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32219_cast_fp16 = einsum(equation = var_32219_equation_0, values = (var_32067_cast_fp16, var_32184_cast_fp16))[name = tensor("op_32219_cast_fp16")]; + tensor var_32221_equation_0 = const()[name = tensor("op_32221_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32221_cast_fp16 = einsum(equation = var_32221_equation_0, values = (var_32071_cast_fp16, var_32185_cast_fp16))[name = tensor("op_32221_cast_fp16")]; + tensor var_32223_equation_0 = const()[name = tensor("op_32223_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32223_cast_fp16 = einsum(equation = var_32223_equation_0, values = (var_32075_cast_fp16, var_32186_cast_fp16))[name = tensor("op_32223_cast_fp16")]; + tensor var_32225_equation_0 = const()[name = tensor("op_32225_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32225_cast_fp16 = einsum(equation = var_32225_equation_0, values = (var_32079_cast_fp16, var_32187_cast_fp16))[name = tensor("op_32225_cast_fp16")]; + tensor var_32227_equation_0 = const()[name = tensor("op_32227_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32227_cast_fp16 = einsum(equation = var_32227_equation_0, values = (var_32083_cast_fp16, var_32188_cast_fp16))[name = tensor("op_32227_cast_fp16")]; + tensor var_32229_equation_0 = const()[name = tensor("op_32229_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32229_cast_fp16 = einsum(equation = var_32229_equation_0, values = (var_32087_cast_fp16, var_32189_cast_fp16))[name = tensor("op_32229_cast_fp16")]; + tensor input_449_interleave_0 = const()[name = tensor("input_449_interleave_0"), val = tensor(false)]; + tensor input_449_cast_fp16 = concat(axis = var_30355, interleave = input_449_interleave_0, values = (var_32191_cast_fp16, var_32193_cast_fp16, var_32195_cast_fp16, var_32197_cast_fp16, var_32199_cast_fp16, var_32201_cast_fp16, var_32203_cast_fp16, var_32205_cast_fp16, var_32207_cast_fp16, var_32209_cast_fp16, var_32211_cast_fp16, var_32213_cast_fp16, var_32215_cast_fp16, var_32217_cast_fp16, var_32219_cast_fp16, var_32221_cast_fp16, var_32223_cast_fp16, var_32225_cast_fp16, var_32227_cast_fp16, var_32229_cast_fp16))[name = tensor("input_449_cast_fp16")]; + tensor var_32235 = const()[name = tensor("op_32235"), val = tensor([1, 1])]; + tensor var_32237 = const()[name = tensor("op_32237"), val = tensor([1, 1])]; + tensor var_32239_pad_type_0 = const()[name = tensor("op_32239_pad_type_0"), val = tensor("custom")]; + tensor var_32239_pad_0 = const()[name = tensor("op_32239_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_1_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_1_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2684817664)))]; + tensor up_blocks_0_attentions_0_transformer_blocks_1_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_1_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2688094528)))]; + tensor var_32239_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_1_attn2_to_out_0_bias_to_fp16, dilations = var_32237, groups = var_30355, pad = var_32239_pad_0, pad_type = var_32239_pad_type_0, strides = var_32235, weight = up_blocks_0_attentions_0_transformer_blocks_1_attn2_to_out_0_weight_to_fp16, x = input_449_cast_fp16)[name = tensor("op_32239_cast_fp16")]; + tensor inputs_215_cast_fp16 = add(x = var_32239_cast_fp16, y = inputs_213_cast_fp16)[name = tensor("inputs_215_cast_fp16")]; + tensor input_451_axes_0 = const()[name = tensor("input_451_axes_0"), val = tensor([1])]; + tensor input_451_gamma_0_to_fp16 = const()[name = tensor("input_451_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2688097152)))]; + tensor input_451_beta_0_to_fp16 = const()[name = tensor("input_451_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2688099776)))]; + tensor var_32249_to_fp16 = const()[name = tensor("op_32249_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_451_cast_fp16 = layer_norm(axes = input_451_axes_0, beta = input_451_beta_0_to_fp16, epsilon = var_32249_to_fp16, gamma = input_451_gamma_0_to_fp16, x = inputs_215_cast_fp16)[name = tensor("input_451_cast_fp16")]; + tensor var_32265 = const()[name = tensor("op_32265"), val = tensor([1, 1])]; + tensor var_32267 = const()[name = tensor("op_32267"), val = tensor([1, 1])]; + tensor var_32269_pad_type_0 = const()[name = tensor("op_32269_pad_type_0"), val = tensor("custom")]; + tensor var_32269_pad_0 = const()[name = tensor("op_32269_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_1_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_1_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2688102400)))]; + tensor up_blocks_0_attentions_0_transformer_blocks_1_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_1_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2714316864)))]; + tensor var_32269_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_1_ff_net_0_proj_bias_to_fp16, dilations = var_32267, groups = var_30355, pad = var_32269_pad_0, pad_type = var_32269_pad_type_0, strides = var_32265, weight = up_blocks_0_attentions_0_transformer_blocks_1_ff_net_0_proj_weight_to_fp16, x = input_451_cast_fp16)[name = tensor("op_32269_cast_fp16")]; + tensor var_32270_split_sizes_0 = const()[name = tensor("op_32270_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_32270_axis_0 = const()[name = tensor("op_32270_axis_0"), val = tensor(1)]; + tensor var_32270_cast_fp16_0, tensor var_32270_cast_fp16_1 = split(axis = var_32270_axis_0, split_sizes = var_32270_split_sizes_0, x = var_32269_cast_fp16)[name = tensor("op_32270_cast_fp16")]; + tensor var_32272_mode_0 = const()[name = tensor("op_32272_mode_0"), val = tensor("EXACT")]; + tensor var_32272_cast_fp16 = gelu(mode = var_32272_mode_0, x = var_32270_cast_fp16_1)[name = tensor("op_32272_cast_fp16")]; + tensor input_453_cast_fp16 = mul(x = var_32270_cast_fp16_0, y = var_32272_cast_fp16)[name = tensor("input_453_cast_fp16")]; + tensor var_32276 = const()[name = tensor("op_32276"), val = tensor([1, 1])]; + tensor var_32278 = const()[name = tensor("op_32278"), val = tensor([1, 1])]; + tensor var_32280_pad_type_0 = const()[name = tensor("op_32280_pad_type_0"), val = tensor("custom")]; + tensor var_32280_pad_0 = const()[name = tensor("op_32280_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_1_ff_net_2_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_1_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2714337408)))]; + tensor up_blocks_0_attentions_0_transformer_blocks_1_ff_net_2_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_1_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2727444672)))]; + tensor var_32280_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_1_ff_net_2_bias_to_fp16, dilations = var_32278, groups = var_30355, pad = var_32280_pad_0, pad_type = var_32280_pad_type_0, strides = var_32276, weight = up_blocks_0_attentions_0_transformer_blocks_1_ff_net_2_weight_to_fp16, x = input_453_cast_fp16)[name = tensor("op_32280_cast_fp16")]; + tensor inputs_217_cast_fp16 = add(x = var_32280_cast_fp16, y = inputs_215_cast_fp16)[name = tensor("inputs_217_cast_fp16")]; + tensor hidden_states_299_axes_0 = const()[name = tensor("hidden_states_299_axes_0"), val = tensor([1])]; + tensor hidden_states_299_gamma_0_to_fp16 = const()[name = tensor("hidden_states_299_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2727447296)))]; + tensor hidden_states_299_beta_0_to_fp16 = const()[name = tensor("hidden_states_299_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2727449920)))]; + tensor var_32296_to_fp16 = const()[name = tensor("op_32296_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_299_cast_fp16 = layer_norm(axes = hidden_states_299_axes_0, beta = hidden_states_299_beta_0_to_fp16, epsilon = var_32296_to_fp16, gamma = hidden_states_299_gamma_0_to_fp16, x = inputs_217_cast_fp16)[name = tensor("hidden_states_299_cast_fp16")]; + tensor var_32311 = const()[name = tensor("op_32311"), val = tensor([1, 1])]; + tensor var_32313 = const()[name = tensor("op_32313"), val = tensor([1, 1])]; + tensor q_145_pad_type_0 = const()[name = tensor("q_145_pad_type_0"), val = tensor("custom")]; + tensor q_145_pad_0 = const()[name = tensor("q_145_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_2_attn1_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_2_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2727452544)))]; + tensor q_145_cast_fp16 = conv(dilations = var_32313, groups = var_30355, pad = q_145_pad_0, pad_type = q_145_pad_type_0, strides = var_32311, weight = up_blocks_0_attentions_0_transformer_blocks_2_attn1_to_q_weight_to_fp16, x = hidden_states_299_cast_fp16)[name = tensor("q_145_cast_fp16")]; + tensor var_32317 = const()[name = tensor("op_32317"), val = tensor([1, 1])]; + tensor var_32319 = const()[name = tensor("op_32319"), val = tensor([1, 1])]; + tensor k_289_pad_type_0 = const()[name = tensor("k_289_pad_type_0"), val = tensor("custom")]; + tensor k_289_pad_0 = const()[name = tensor("k_289_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_2_attn1_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_2_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2730729408)))]; + tensor k_289_cast_fp16 = conv(dilations = var_32319, groups = var_30355, pad = k_289_pad_0, pad_type = k_289_pad_type_0, strides = var_32317, weight = up_blocks_0_attentions_0_transformer_blocks_2_attn1_to_k_weight_to_fp16, x = hidden_states_299_cast_fp16)[name = tensor("k_289_cast_fp16")]; + tensor var_32323 = const()[name = tensor("op_32323"), val = tensor([1, 1])]; + tensor var_32325 = const()[name = tensor("op_32325"), val = tensor([1, 1])]; + tensor v_145_pad_type_0 = const()[name = tensor("v_145_pad_type_0"), val = tensor("custom")]; + tensor v_145_pad_0 = const()[name = tensor("v_145_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_2_attn1_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_2_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2734006272)))]; + tensor v_145_cast_fp16 = conv(dilations = var_32325, groups = var_30355, pad = v_145_pad_0, pad_type = v_145_pad_type_0, strides = var_32323, weight = up_blocks_0_attentions_0_transformer_blocks_2_attn1_to_v_weight_to_fp16, x = hidden_states_299_cast_fp16)[name = tensor("v_145_cast_fp16")]; + tensor var_32329_begin_0 = const()[name = tensor("op_32329_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_32329_end_0 = const()[name = tensor("op_32329_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_32329_end_mask_0 = const()[name = tensor("op_32329_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32329_cast_fp16 = slice_by_index(begin = var_32329_begin_0, end = var_32329_end_0, end_mask = var_32329_end_mask_0, x = q_145_cast_fp16)[name = tensor("op_32329_cast_fp16")]; + tensor var_32333_begin_0 = const()[name = tensor("op_32333_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_32333_end_0 = const()[name = tensor("op_32333_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_32333_end_mask_0 = const()[name = tensor("op_32333_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32333_cast_fp16 = slice_by_index(begin = var_32333_begin_0, end = var_32333_end_0, end_mask = var_32333_end_mask_0, x = q_145_cast_fp16)[name = tensor("op_32333_cast_fp16")]; + tensor var_32337_begin_0 = const()[name = tensor("op_32337_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_32337_end_0 = const()[name = tensor("op_32337_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_32337_end_mask_0 = const()[name = tensor("op_32337_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32337_cast_fp16 = slice_by_index(begin = var_32337_begin_0, end = var_32337_end_0, end_mask = var_32337_end_mask_0, x = q_145_cast_fp16)[name = tensor("op_32337_cast_fp16")]; + tensor var_32341_begin_0 = const()[name = tensor("op_32341_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_32341_end_0 = const()[name = tensor("op_32341_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_32341_end_mask_0 = const()[name = tensor("op_32341_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32341_cast_fp16 = slice_by_index(begin = var_32341_begin_0, end = var_32341_end_0, end_mask = var_32341_end_mask_0, x = q_145_cast_fp16)[name = tensor("op_32341_cast_fp16")]; + tensor var_32345_begin_0 = const()[name = tensor("op_32345_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_32345_end_0 = const()[name = tensor("op_32345_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_32345_end_mask_0 = const()[name = tensor("op_32345_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32345_cast_fp16 = slice_by_index(begin = var_32345_begin_0, end = var_32345_end_0, end_mask = var_32345_end_mask_0, x = q_145_cast_fp16)[name = tensor("op_32345_cast_fp16")]; + tensor var_32349_begin_0 = const()[name = tensor("op_32349_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_32349_end_0 = const()[name = tensor("op_32349_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_32349_end_mask_0 = const()[name = tensor("op_32349_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32349_cast_fp16 = slice_by_index(begin = var_32349_begin_0, end = var_32349_end_0, end_mask = var_32349_end_mask_0, x = q_145_cast_fp16)[name = tensor("op_32349_cast_fp16")]; + tensor var_32353_begin_0 = const()[name = tensor("op_32353_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_32353_end_0 = const()[name = tensor("op_32353_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_32353_end_mask_0 = const()[name = tensor("op_32353_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32353_cast_fp16 = slice_by_index(begin = var_32353_begin_0, end = var_32353_end_0, end_mask = var_32353_end_mask_0, x = q_145_cast_fp16)[name = tensor("op_32353_cast_fp16")]; + tensor var_32357_begin_0 = const()[name = tensor("op_32357_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_32357_end_0 = const()[name = tensor("op_32357_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_32357_end_mask_0 = const()[name = tensor("op_32357_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32357_cast_fp16 = slice_by_index(begin = var_32357_begin_0, end = var_32357_end_0, end_mask = var_32357_end_mask_0, x = q_145_cast_fp16)[name = tensor("op_32357_cast_fp16")]; + tensor var_32361_begin_0 = const()[name = tensor("op_32361_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_32361_end_0 = const()[name = tensor("op_32361_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_32361_end_mask_0 = const()[name = tensor("op_32361_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32361_cast_fp16 = slice_by_index(begin = var_32361_begin_0, end = var_32361_end_0, end_mask = var_32361_end_mask_0, x = q_145_cast_fp16)[name = tensor("op_32361_cast_fp16")]; + tensor var_32365_begin_0 = const()[name = tensor("op_32365_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_32365_end_0 = const()[name = tensor("op_32365_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_32365_end_mask_0 = const()[name = tensor("op_32365_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32365_cast_fp16 = slice_by_index(begin = var_32365_begin_0, end = var_32365_end_0, end_mask = var_32365_end_mask_0, x = q_145_cast_fp16)[name = tensor("op_32365_cast_fp16")]; + tensor var_32369_begin_0 = const()[name = tensor("op_32369_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_32369_end_0 = const()[name = tensor("op_32369_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_32369_end_mask_0 = const()[name = tensor("op_32369_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32369_cast_fp16 = slice_by_index(begin = var_32369_begin_0, end = var_32369_end_0, end_mask = var_32369_end_mask_0, x = q_145_cast_fp16)[name = tensor("op_32369_cast_fp16")]; + tensor var_32373_begin_0 = const()[name = tensor("op_32373_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_32373_end_0 = const()[name = tensor("op_32373_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_32373_end_mask_0 = const()[name = tensor("op_32373_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32373_cast_fp16 = slice_by_index(begin = var_32373_begin_0, end = var_32373_end_0, end_mask = var_32373_end_mask_0, x = q_145_cast_fp16)[name = tensor("op_32373_cast_fp16")]; + tensor var_32377_begin_0 = const()[name = tensor("op_32377_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_32377_end_0 = const()[name = tensor("op_32377_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_32377_end_mask_0 = const()[name = tensor("op_32377_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32377_cast_fp16 = slice_by_index(begin = var_32377_begin_0, end = var_32377_end_0, end_mask = var_32377_end_mask_0, x = q_145_cast_fp16)[name = tensor("op_32377_cast_fp16")]; + tensor var_32381_begin_0 = const()[name = tensor("op_32381_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_32381_end_0 = const()[name = tensor("op_32381_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_32381_end_mask_0 = const()[name = tensor("op_32381_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32381_cast_fp16 = slice_by_index(begin = var_32381_begin_0, end = var_32381_end_0, end_mask = var_32381_end_mask_0, x = q_145_cast_fp16)[name = tensor("op_32381_cast_fp16")]; + tensor var_32385_begin_0 = const()[name = tensor("op_32385_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_32385_end_0 = const()[name = tensor("op_32385_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_32385_end_mask_0 = const()[name = tensor("op_32385_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32385_cast_fp16 = slice_by_index(begin = var_32385_begin_0, end = var_32385_end_0, end_mask = var_32385_end_mask_0, x = q_145_cast_fp16)[name = tensor("op_32385_cast_fp16")]; + tensor var_32389_begin_0 = const()[name = tensor("op_32389_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_32389_end_0 = const()[name = tensor("op_32389_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_32389_end_mask_0 = const()[name = tensor("op_32389_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32389_cast_fp16 = slice_by_index(begin = var_32389_begin_0, end = var_32389_end_0, end_mask = var_32389_end_mask_0, x = q_145_cast_fp16)[name = tensor("op_32389_cast_fp16")]; + tensor var_32393_begin_0 = const()[name = tensor("op_32393_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_32393_end_0 = const()[name = tensor("op_32393_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_32393_end_mask_0 = const()[name = tensor("op_32393_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32393_cast_fp16 = slice_by_index(begin = var_32393_begin_0, end = var_32393_end_0, end_mask = var_32393_end_mask_0, x = q_145_cast_fp16)[name = tensor("op_32393_cast_fp16")]; + tensor var_32397_begin_0 = const()[name = tensor("op_32397_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_32397_end_0 = const()[name = tensor("op_32397_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_32397_end_mask_0 = const()[name = tensor("op_32397_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32397_cast_fp16 = slice_by_index(begin = var_32397_begin_0, end = var_32397_end_0, end_mask = var_32397_end_mask_0, x = q_145_cast_fp16)[name = tensor("op_32397_cast_fp16")]; + tensor var_32401_begin_0 = const()[name = tensor("op_32401_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_32401_end_0 = const()[name = tensor("op_32401_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_32401_end_mask_0 = const()[name = tensor("op_32401_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32401_cast_fp16 = slice_by_index(begin = var_32401_begin_0, end = var_32401_end_0, end_mask = var_32401_end_mask_0, x = q_145_cast_fp16)[name = tensor("op_32401_cast_fp16")]; + tensor var_32405_begin_0 = const()[name = tensor("op_32405_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_32405_end_0 = const()[name = tensor("op_32405_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_32405_end_mask_0 = const()[name = tensor("op_32405_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32405_cast_fp16 = slice_by_index(begin = var_32405_begin_0, end = var_32405_end_0, end_mask = var_32405_end_mask_0, x = q_145_cast_fp16)[name = tensor("op_32405_cast_fp16")]; + tensor k_291_perm_0 = const()[name = tensor("k_291_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_32412_begin_0 = const()[name = tensor("op_32412_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_32412_end_0 = const()[name = tensor("op_32412_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_32412_end_mask_0 = const()[name = tensor("op_32412_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_291_cast_fp16 = transpose(perm = k_291_perm_0, x = k_289_cast_fp16)[name = tensor("transpose_67")]; + tensor var_32412_cast_fp16 = slice_by_index(begin = var_32412_begin_0, end = var_32412_end_0, end_mask = var_32412_end_mask_0, x = k_291_cast_fp16)[name = tensor("op_32412_cast_fp16")]; + tensor var_32416_begin_0 = const()[name = tensor("op_32416_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_32416_end_0 = const()[name = tensor("op_32416_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_32416_end_mask_0 = const()[name = tensor("op_32416_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32416_cast_fp16 = slice_by_index(begin = var_32416_begin_0, end = var_32416_end_0, end_mask = var_32416_end_mask_0, x = k_291_cast_fp16)[name = tensor("op_32416_cast_fp16")]; + tensor var_32420_begin_0 = const()[name = tensor("op_32420_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_32420_end_0 = const()[name = tensor("op_32420_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_32420_end_mask_0 = const()[name = tensor("op_32420_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32420_cast_fp16 = slice_by_index(begin = var_32420_begin_0, end = var_32420_end_0, end_mask = var_32420_end_mask_0, x = k_291_cast_fp16)[name = tensor("op_32420_cast_fp16")]; + tensor var_32424_begin_0 = const()[name = tensor("op_32424_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_32424_end_0 = const()[name = tensor("op_32424_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_32424_end_mask_0 = const()[name = tensor("op_32424_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32424_cast_fp16 = slice_by_index(begin = var_32424_begin_0, end = var_32424_end_0, end_mask = var_32424_end_mask_0, x = k_291_cast_fp16)[name = tensor("op_32424_cast_fp16")]; + tensor var_32428_begin_0 = const()[name = tensor("op_32428_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_32428_end_0 = const()[name = tensor("op_32428_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_32428_end_mask_0 = const()[name = tensor("op_32428_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32428_cast_fp16 = slice_by_index(begin = var_32428_begin_0, end = var_32428_end_0, end_mask = var_32428_end_mask_0, x = k_291_cast_fp16)[name = tensor("op_32428_cast_fp16")]; + tensor var_32432_begin_0 = const()[name = tensor("op_32432_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_32432_end_0 = const()[name = tensor("op_32432_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_32432_end_mask_0 = const()[name = tensor("op_32432_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32432_cast_fp16 = slice_by_index(begin = var_32432_begin_0, end = var_32432_end_0, end_mask = var_32432_end_mask_0, x = k_291_cast_fp16)[name = tensor("op_32432_cast_fp16")]; + tensor var_32436_begin_0 = const()[name = tensor("op_32436_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_32436_end_0 = const()[name = tensor("op_32436_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_32436_end_mask_0 = const()[name = tensor("op_32436_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32436_cast_fp16 = slice_by_index(begin = var_32436_begin_0, end = var_32436_end_0, end_mask = var_32436_end_mask_0, x = k_291_cast_fp16)[name = tensor("op_32436_cast_fp16")]; + tensor var_32440_begin_0 = const()[name = tensor("op_32440_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_32440_end_0 = const()[name = tensor("op_32440_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_32440_end_mask_0 = const()[name = tensor("op_32440_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32440_cast_fp16 = slice_by_index(begin = var_32440_begin_0, end = var_32440_end_0, end_mask = var_32440_end_mask_0, x = k_291_cast_fp16)[name = tensor("op_32440_cast_fp16")]; + tensor var_32444_begin_0 = const()[name = tensor("op_32444_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_32444_end_0 = const()[name = tensor("op_32444_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_32444_end_mask_0 = const()[name = tensor("op_32444_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32444_cast_fp16 = slice_by_index(begin = var_32444_begin_0, end = var_32444_end_0, end_mask = var_32444_end_mask_0, x = k_291_cast_fp16)[name = tensor("op_32444_cast_fp16")]; + tensor var_32448_begin_0 = const()[name = tensor("op_32448_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_32448_end_0 = const()[name = tensor("op_32448_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_32448_end_mask_0 = const()[name = tensor("op_32448_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32448_cast_fp16 = slice_by_index(begin = var_32448_begin_0, end = var_32448_end_0, end_mask = var_32448_end_mask_0, x = k_291_cast_fp16)[name = tensor("op_32448_cast_fp16")]; + tensor var_32452_begin_0 = const()[name = tensor("op_32452_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_32452_end_0 = const()[name = tensor("op_32452_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_32452_end_mask_0 = const()[name = tensor("op_32452_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32452_cast_fp16 = slice_by_index(begin = var_32452_begin_0, end = var_32452_end_0, end_mask = var_32452_end_mask_0, x = k_291_cast_fp16)[name = tensor("op_32452_cast_fp16")]; + tensor var_32456_begin_0 = const()[name = tensor("op_32456_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_32456_end_0 = const()[name = tensor("op_32456_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_32456_end_mask_0 = const()[name = tensor("op_32456_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32456_cast_fp16 = slice_by_index(begin = var_32456_begin_0, end = var_32456_end_0, end_mask = var_32456_end_mask_0, x = k_291_cast_fp16)[name = tensor("op_32456_cast_fp16")]; + tensor var_32460_begin_0 = const()[name = tensor("op_32460_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_32460_end_0 = const()[name = tensor("op_32460_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_32460_end_mask_0 = const()[name = tensor("op_32460_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32460_cast_fp16 = slice_by_index(begin = var_32460_begin_0, end = var_32460_end_0, end_mask = var_32460_end_mask_0, x = k_291_cast_fp16)[name = tensor("op_32460_cast_fp16")]; + tensor var_32464_begin_0 = const()[name = tensor("op_32464_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_32464_end_0 = const()[name = tensor("op_32464_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_32464_end_mask_0 = const()[name = tensor("op_32464_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32464_cast_fp16 = slice_by_index(begin = var_32464_begin_0, end = var_32464_end_0, end_mask = var_32464_end_mask_0, x = k_291_cast_fp16)[name = tensor("op_32464_cast_fp16")]; + tensor var_32468_begin_0 = const()[name = tensor("op_32468_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_32468_end_0 = const()[name = tensor("op_32468_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_32468_end_mask_0 = const()[name = tensor("op_32468_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32468_cast_fp16 = slice_by_index(begin = var_32468_begin_0, end = var_32468_end_0, end_mask = var_32468_end_mask_0, x = k_291_cast_fp16)[name = tensor("op_32468_cast_fp16")]; + tensor var_32472_begin_0 = const()[name = tensor("op_32472_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_32472_end_0 = const()[name = tensor("op_32472_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_32472_end_mask_0 = const()[name = tensor("op_32472_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32472_cast_fp16 = slice_by_index(begin = var_32472_begin_0, end = var_32472_end_0, end_mask = var_32472_end_mask_0, x = k_291_cast_fp16)[name = tensor("op_32472_cast_fp16")]; + tensor var_32476_begin_0 = const()[name = tensor("op_32476_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_32476_end_0 = const()[name = tensor("op_32476_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_32476_end_mask_0 = const()[name = tensor("op_32476_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32476_cast_fp16 = slice_by_index(begin = var_32476_begin_0, end = var_32476_end_0, end_mask = var_32476_end_mask_0, x = k_291_cast_fp16)[name = tensor("op_32476_cast_fp16")]; + tensor var_32480_begin_0 = const()[name = tensor("op_32480_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_32480_end_0 = const()[name = tensor("op_32480_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_32480_end_mask_0 = const()[name = tensor("op_32480_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32480_cast_fp16 = slice_by_index(begin = var_32480_begin_0, end = var_32480_end_0, end_mask = var_32480_end_mask_0, x = k_291_cast_fp16)[name = tensor("op_32480_cast_fp16")]; + tensor var_32484_begin_0 = const()[name = tensor("op_32484_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_32484_end_0 = const()[name = tensor("op_32484_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_32484_end_mask_0 = const()[name = tensor("op_32484_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32484_cast_fp16 = slice_by_index(begin = var_32484_begin_0, end = var_32484_end_0, end_mask = var_32484_end_mask_0, x = k_291_cast_fp16)[name = tensor("op_32484_cast_fp16")]; + tensor var_32488_begin_0 = const()[name = tensor("op_32488_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_32488_end_0 = const()[name = tensor("op_32488_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_32488_end_mask_0 = const()[name = tensor("op_32488_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32488_cast_fp16 = slice_by_index(begin = var_32488_begin_0, end = var_32488_end_0, end_mask = var_32488_end_mask_0, x = k_291_cast_fp16)[name = tensor("op_32488_cast_fp16")]; + tensor var_32490_begin_0 = const()[name = tensor("op_32490_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_32490_end_0 = const()[name = tensor("op_32490_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_32490_end_mask_0 = const()[name = tensor("op_32490_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32490_cast_fp16 = slice_by_index(begin = var_32490_begin_0, end = var_32490_end_0, end_mask = var_32490_end_mask_0, x = v_145_cast_fp16)[name = tensor("op_32490_cast_fp16")]; + tensor var_32494_begin_0 = const()[name = tensor("op_32494_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_32494_end_0 = const()[name = tensor("op_32494_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_32494_end_mask_0 = const()[name = tensor("op_32494_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32494_cast_fp16 = slice_by_index(begin = var_32494_begin_0, end = var_32494_end_0, end_mask = var_32494_end_mask_0, x = v_145_cast_fp16)[name = tensor("op_32494_cast_fp16")]; + tensor var_32498_begin_0 = const()[name = tensor("op_32498_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_32498_end_0 = const()[name = tensor("op_32498_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_32498_end_mask_0 = const()[name = tensor("op_32498_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32498_cast_fp16 = slice_by_index(begin = var_32498_begin_0, end = var_32498_end_0, end_mask = var_32498_end_mask_0, x = v_145_cast_fp16)[name = tensor("op_32498_cast_fp16")]; + tensor var_32502_begin_0 = const()[name = tensor("op_32502_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_32502_end_0 = const()[name = tensor("op_32502_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_32502_end_mask_0 = const()[name = tensor("op_32502_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32502_cast_fp16 = slice_by_index(begin = var_32502_begin_0, end = var_32502_end_0, end_mask = var_32502_end_mask_0, x = v_145_cast_fp16)[name = tensor("op_32502_cast_fp16")]; + tensor var_32506_begin_0 = const()[name = tensor("op_32506_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_32506_end_0 = const()[name = tensor("op_32506_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_32506_end_mask_0 = const()[name = tensor("op_32506_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32506_cast_fp16 = slice_by_index(begin = var_32506_begin_0, end = var_32506_end_0, end_mask = var_32506_end_mask_0, x = v_145_cast_fp16)[name = tensor("op_32506_cast_fp16")]; + tensor var_32510_begin_0 = const()[name = tensor("op_32510_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_32510_end_0 = const()[name = tensor("op_32510_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_32510_end_mask_0 = const()[name = tensor("op_32510_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32510_cast_fp16 = slice_by_index(begin = var_32510_begin_0, end = var_32510_end_0, end_mask = var_32510_end_mask_0, x = v_145_cast_fp16)[name = tensor("op_32510_cast_fp16")]; + tensor var_32514_begin_0 = const()[name = tensor("op_32514_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_32514_end_0 = const()[name = tensor("op_32514_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_32514_end_mask_0 = const()[name = tensor("op_32514_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32514_cast_fp16 = slice_by_index(begin = var_32514_begin_0, end = var_32514_end_0, end_mask = var_32514_end_mask_0, x = v_145_cast_fp16)[name = tensor("op_32514_cast_fp16")]; + tensor var_32518_begin_0 = const()[name = tensor("op_32518_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_32518_end_0 = const()[name = tensor("op_32518_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_32518_end_mask_0 = const()[name = tensor("op_32518_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32518_cast_fp16 = slice_by_index(begin = var_32518_begin_0, end = var_32518_end_0, end_mask = var_32518_end_mask_0, x = v_145_cast_fp16)[name = tensor("op_32518_cast_fp16")]; + tensor var_32522_begin_0 = const()[name = tensor("op_32522_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_32522_end_0 = const()[name = tensor("op_32522_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_32522_end_mask_0 = const()[name = tensor("op_32522_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32522_cast_fp16 = slice_by_index(begin = var_32522_begin_0, end = var_32522_end_0, end_mask = var_32522_end_mask_0, x = v_145_cast_fp16)[name = tensor("op_32522_cast_fp16")]; + tensor var_32526_begin_0 = const()[name = tensor("op_32526_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_32526_end_0 = const()[name = tensor("op_32526_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_32526_end_mask_0 = const()[name = tensor("op_32526_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32526_cast_fp16 = slice_by_index(begin = var_32526_begin_0, end = var_32526_end_0, end_mask = var_32526_end_mask_0, x = v_145_cast_fp16)[name = tensor("op_32526_cast_fp16")]; + tensor var_32530_begin_0 = const()[name = tensor("op_32530_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_32530_end_0 = const()[name = tensor("op_32530_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_32530_end_mask_0 = const()[name = tensor("op_32530_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32530_cast_fp16 = slice_by_index(begin = var_32530_begin_0, end = var_32530_end_0, end_mask = var_32530_end_mask_0, x = v_145_cast_fp16)[name = tensor("op_32530_cast_fp16")]; + tensor var_32534_begin_0 = const()[name = tensor("op_32534_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_32534_end_0 = const()[name = tensor("op_32534_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_32534_end_mask_0 = const()[name = tensor("op_32534_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32534_cast_fp16 = slice_by_index(begin = var_32534_begin_0, end = var_32534_end_0, end_mask = var_32534_end_mask_0, x = v_145_cast_fp16)[name = tensor("op_32534_cast_fp16")]; + tensor var_32538_begin_0 = const()[name = tensor("op_32538_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_32538_end_0 = const()[name = tensor("op_32538_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_32538_end_mask_0 = const()[name = tensor("op_32538_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32538_cast_fp16 = slice_by_index(begin = var_32538_begin_0, end = var_32538_end_0, end_mask = var_32538_end_mask_0, x = v_145_cast_fp16)[name = tensor("op_32538_cast_fp16")]; + tensor var_32542_begin_0 = const()[name = tensor("op_32542_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_32542_end_0 = const()[name = tensor("op_32542_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_32542_end_mask_0 = const()[name = tensor("op_32542_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32542_cast_fp16 = slice_by_index(begin = var_32542_begin_0, end = var_32542_end_0, end_mask = var_32542_end_mask_0, x = v_145_cast_fp16)[name = tensor("op_32542_cast_fp16")]; + tensor var_32546_begin_0 = const()[name = tensor("op_32546_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_32546_end_0 = const()[name = tensor("op_32546_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_32546_end_mask_0 = const()[name = tensor("op_32546_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32546_cast_fp16 = slice_by_index(begin = var_32546_begin_0, end = var_32546_end_0, end_mask = var_32546_end_mask_0, x = v_145_cast_fp16)[name = tensor("op_32546_cast_fp16")]; + tensor var_32550_begin_0 = const()[name = tensor("op_32550_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_32550_end_0 = const()[name = tensor("op_32550_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_32550_end_mask_0 = const()[name = tensor("op_32550_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32550_cast_fp16 = slice_by_index(begin = var_32550_begin_0, end = var_32550_end_0, end_mask = var_32550_end_mask_0, x = v_145_cast_fp16)[name = tensor("op_32550_cast_fp16")]; + tensor var_32554_begin_0 = const()[name = tensor("op_32554_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_32554_end_0 = const()[name = tensor("op_32554_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_32554_end_mask_0 = const()[name = tensor("op_32554_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32554_cast_fp16 = slice_by_index(begin = var_32554_begin_0, end = var_32554_end_0, end_mask = var_32554_end_mask_0, x = v_145_cast_fp16)[name = tensor("op_32554_cast_fp16")]; + tensor var_32558_begin_0 = const()[name = tensor("op_32558_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_32558_end_0 = const()[name = tensor("op_32558_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_32558_end_mask_0 = const()[name = tensor("op_32558_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32558_cast_fp16 = slice_by_index(begin = var_32558_begin_0, end = var_32558_end_0, end_mask = var_32558_end_mask_0, x = v_145_cast_fp16)[name = tensor("op_32558_cast_fp16")]; + tensor var_32562_begin_0 = const()[name = tensor("op_32562_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_32562_end_0 = const()[name = tensor("op_32562_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_32562_end_mask_0 = const()[name = tensor("op_32562_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32562_cast_fp16 = slice_by_index(begin = var_32562_begin_0, end = var_32562_end_0, end_mask = var_32562_end_mask_0, x = v_145_cast_fp16)[name = tensor("op_32562_cast_fp16")]; + tensor var_32566_begin_0 = const()[name = tensor("op_32566_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_32566_end_0 = const()[name = tensor("op_32566_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_32566_end_mask_0 = const()[name = tensor("op_32566_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32566_cast_fp16 = slice_by_index(begin = var_32566_begin_0, end = var_32566_end_0, end_mask = var_32566_end_mask_0, x = v_145_cast_fp16)[name = tensor("op_32566_cast_fp16")]; + tensor var_32570_equation_0 = const()[name = tensor("op_32570_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32570_cast_fp16 = einsum(equation = var_32570_equation_0, values = (var_32412_cast_fp16, var_32329_cast_fp16))[name = tensor("op_32570_cast_fp16")]; + tensor var_32571_to_fp16 = const()[name = tensor("op_32571_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2721_cast_fp16 = mul(x = var_32570_cast_fp16, y = var_32571_to_fp16)[name = tensor("aw_2721_cast_fp16")]; + tensor var_32574_equation_0 = const()[name = tensor("op_32574_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32574_cast_fp16 = einsum(equation = var_32574_equation_0, values = (var_32416_cast_fp16, var_32333_cast_fp16))[name = tensor("op_32574_cast_fp16")]; + tensor var_32575_to_fp16 = const()[name = tensor("op_32575_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2723_cast_fp16 = mul(x = var_32574_cast_fp16, y = var_32575_to_fp16)[name = tensor("aw_2723_cast_fp16")]; + tensor var_32578_equation_0 = const()[name = tensor("op_32578_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32578_cast_fp16 = einsum(equation = var_32578_equation_0, values = (var_32420_cast_fp16, var_32337_cast_fp16))[name = tensor("op_32578_cast_fp16")]; + tensor var_32579_to_fp16 = const()[name = tensor("op_32579_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2725_cast_fp16 = mul(x = var_32578_cast_fp16, y = var_32579_to_fp16)[name = tensor("aw_2725_cast_fp16")]; + tensor var_32582_equation_0 = const()[name = tensor("op_32582_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32582_cast_fp16 = einsum(equation = var_32582_equation_0, values = (var_32424_cast_fp16, var_32341_cast_fp16))[name = tensor("op_32582_cast_fp16")]; + tensor var_32583_to_fp16 = const()[name = tensor("op_32583_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2727_cast_fp16 = mul(x = var_32582_cast_fp16, y = var_32583_to_fp16)[name = tensor("aw_2727_cast_fp16")]; + tensor var_32586_equation_0 = const()[name = tensor("op_32586_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32586_cast_fp16 = einsum(equation = var_32586_equation_0, values = (var_32428_cast_fp16, var_32345_cast_fp16))[name = tensor("op_32586_cast_fp16")]; + tensor var_32587_to_fp16 = const()[name = tensor("op_32587_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2729_cast_fp16 = mul(x = var_32586_cast_fp16, y = var_32587_to_fp16)[name = tensor("aw_2729_cast_fp16")]; + tensor var_32590_equation_0 = const()[name = tensor("op_32590_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32590_cast_fp16 = einsum(equation = var_32590_equation_0, values = (var_32432_cast_fp16, var_32349_cast_fp16))[name = tensor("op_32590_cast_fp16")]; + tensor var_32591_to_fp16 = const()[name = tensor("op_32591_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2731_cast_fp16 = mul(x = var_32590_cast_fp16, y = var_32591_to_fp16)[name = tensor("aw_2731_cast_fp16")]; + tensor var_32594_equation_0 = const()[name = tensor("op_32594_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32594_cast_fp16 = einsum(equation = var_32594_equation_0, values = (var_32436_cast_fp16, var_32353_cast_fp16))[name = tensor("op_32594_cast_fp16")]; + tensor var_32595_to_fp16 = const()[name = tensor("op_32595_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2733_cast_fp16 = mul(x = var_32594_cast_fp16, y = var_32595_to_fp16)[name = tensor("aw_2733_cast_fp16")]; + tensor var_32598_equation_0 = const()[name = tensor("op_32598_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32598_cast_fp16 = einsum(equation = var_32598_equation_0, values = (var_32440_cast_fp16, var_32357_cast_fp16))[name = tensor("op_32598_cast_fp16")]; + tensor var_32599_to_fp16 = const()[name = tensor("op_32599_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2735_cast_fp16 = mul(x = var_32598_cast_fp16, y = var_32599_to_fp16)[name = tensor("aw_2735_cast_fp16")]; + tensor var_32602_equation_0 = const()[name = tensor("op_32602_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32602_cast_fp16 = einsum(equation = var_32602_equation_0, values = (var_32444_cast_fp16, var_32361_cast_fp16))[name = tensor("op_32602_cast_fp16")]; + tensor var_32603_to_fp16 = const()[name = tensor("op_32603_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2737_cast_fp16 = mul(x = var_32602_cast_fp16, y = var_32603_to_fp16)[name = tensor("aw_2737_cast_fp16")]; + tensor var_32606_equation_0 = const()[name = tensor("op_32606_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32606_cast_fp16 = einsum(equation = var_32606_equation_0, values = (var_32448_cast_fp16, var_32365_cast_fp16))[name = tensor("op_32606_cast_fp16")]; + tensor var_32607_to_fp16 = const()[name = tensor("op_32607_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2739_cast_fp16 = mul(x = var_32606_cast_fp16, y = var_32607_to_fp16)[name = tensor("aw_2739_cast_fp16")]; + tensor var_32610_equation_0 = const()[name = tensor("op_32610_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32610_cast_fp16 = einsum(equation = var_32610_equation_0, values = (var_32452_cast_fp16, var_32369_cast_fp16))[name = tensor("op_32610_cast_fp16")]; + tensor var_32611_to_fp16 = const()[name = tensor("op_32611_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2741_cast_fp16 = mul(x = var_32610_cast_fp16, y = var_32611_to_fp16)[name = tensor("aw_2741_cast_fp16")]; + tensor var_32614_equation_0 = const()[name = tensor("op_32614_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32614_cast_fp16 = einsum(equation = var_32614_equation_0, values = (var_32456_cast_fp16, var_32373_cast_fp16))[name = tensor("op_32614_cast_fp16")]; + tensor var_32615_to_fp16 = const()[name = tensor("op_32615_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2743_cast_fp16 = mul(x = var_32614_cast_fp16, y = var_32615_to_fp16)[name = tensor("aw_2743_cast_fp16")]; + tensor var_32618_equation_0 = const()[name = tensor("op_32618_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32618_cast_fp16 = einsum(equation = var_32618_equation_0, values = (var_32460_cast_fp16, var_32377_cast_fp16))[name = tensor("op_32618_cast_fp16")]; + tensor var_32619_to_fp16 = const()[name = tensor("op_32619_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2745_cast_fp16 = mul(x = var_32618_cast_fp16, y = var_32619_to_fp16)[name = tensor("aw_2745_cast_fp16")]; + tensor var_32622_equation_0 = const()[name = tensor("op_32622_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32622_cast_fp16 = einsum(equation = var_32622_equation_0, values = (var_32464_cast_fp16, var_32381_cast_fp16))[name = tensor("op_32622_cast_fp16")]; + tensor var_32623_to_fp16 = const()[name = tensor("op_32623_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2747_cast_fp16 = mul(x = var_32622_cast_fp16, y = var_32623_to_fp16)[name = tensor("aw_2747_cast_fp16")]; + tensor var_32626_equation_0 = const()[name = tensor("op_32626_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32626_cast_fp16 = einsum(equation = var_32626_equation_0, values = (var_32468_cast_fp16, var_32385_cast_fp16))[name = tensor("op_32626_cast_fp16")]; + tensor var_32627_to_fp16 = const()[name = tensor("op_32627_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2749_cast_fp16 = mul(x = var_32626_cast_fp16, y = var_32627_to_fp16)[name = tensor("aw_2749_cast_fp16")]; + tensor var_32630_equation_0 = const()[name = tensor("op_32630_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32630_cast_fp16 = einsum(equation = var_32630_equation_0, values = (var_32472_cast_fp16, var_32389_cast_fp16))[name = tensor("op_32630_cast_fp16")]; + tensor var_32631_to_fp16 = const()[name = tensor("op_32631_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2751_cast_fp16 = mul(x = var_32630_cast_fp16, y = var_32631_to_fp16)[name = tensor("aw_2751_cast_fp16")]; + tensor var_32634_equation_0 = const()[name = tensor("op_32634_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32634_cast_fp16 = einsum(equation = var_32634_equation_0, values = (var_32476_cast_fp16, var_32393_cast_fp16))[name = tensor("op_32634_cast_fp16")]; + tensor var_32635_to_fp16 = const()[name = tensor("op_32635_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2753_cast_fp16 = mul(x = var_32634_cast_fp16, y = var_32635_to_fp16)[name = tensor("aw_2753_cast_fp16")]; + tensor var_32638_equation_0 = const()[name = tensor("op_32638_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32638_cast_fp16 = einsum(equation = var_32638_equation_0, values = (var_32480_cast_fp16, var_32397_cast_fp16))[name = tensor("op_32638_cast_fp16")]; + tensor var_32639_to_fp16 = const()[name = tensor("op_32639_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2755_cast_fp16 = mul(x = var_32638_cast_fp16, y = var_32639_to_fp16)[name = tensor("aw_2755_cast_fp16")]; + tensor var_32642_equation_0 = const()[name = tensor("op_32642_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32642_cast_fp16 = einsum(equation = var_32642_equation_0, values = (var_32484_cast_fp16, var_32401_cast_fp16))[name = tensor("op_32642_cast_fp16")]; + tensor var_32643_to_fp16 = const()[name = tensor("op_32643_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2757_cast_fp16 = mul(x = var_32642_cast_fp16, y = var_32643_to_fp16)[name = tensor("aw_2757_cast_fp16")]; + tensor var_32646_equation_0 = const()[name = tensor("op_32646_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32646_cast_fp16 = einsum(equation = var_32646_equation_0, values = (var_32488_cast_fp16, var_32405_cast_fp16))[name = tensor("op_32646_cast_fp16")]; + tensor var_32647_to_fp16 = const()[name = tensor("op_32647_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2759_cast_fp16 = mul(x = var_32646_cast_fp16, y = var_32647_to_fp16)[name = tensor("aw_2759_cast_fp16")]; + tensor var_32649_cast_fp16 = softmax(axis = var_30355, x = aw_2721_cast_fp16)[name = tensor("op_32649_cast_fp16")]; + tensor var_32650_cast_fp16 = softmax(axis = var_30355, x = aw_2723_cast_fp16)[name = tensor("op_32650_cast_fp16")]; + tensor var_32651_cast_fp16 = softmax(axis = var_30355, x = aw_2725_cast_fp16)[name = tensor("op_32651_cast_fp16")]; + tensor var_32652_cast_fp16 = softmax(axis = var_30355, x = aw_2727_cast_fp16)[name = tensor("op_32652_cast_fp16")]; + tensor var_32653_cast_fp16 = softmax(axis = var_30355, x = aw_2729_cast_fp16)[name = tensor("op_32653_cast_fp16")]; + tensor var_32654_cast_fp16 = softmax(axis = var_30355, x = aw_2731_cast_fp16)[name = tensor("op_32654_cast_fp16")]; + tensor var_32655_cast_fp16 = softmax(axis = var_30355, x = aw_2733_cast_fp16)[name = tensor("op_32655_cast_fp16")]; + tensor var_32656_cast_fp16 = softmax(axis = var_30355, x = aw_2735_cast_fp16)[name = tensor("op_32656_cast_fp16")]; + tensor var_32657_cast_fp16 = softmax(axis = var_30355, x = aw_2737_cast_fp16)[name = tensor("op_32657_cast_fp16")]; + tensor var_32658_cast_fp16 = softmax(axis = var_30355, x = aw_2739_cast_fp16)[name = tensor("op_32658_cast_fp16")]; + tensor var_32659_cast_fp16 = softmax(axis = var_30355, x = aw_2741_cast_fp16)[name = tensor("op_32659_cast_fp16")]; + tensor var_32660_cast_fp16 = softmax(axis = var_30355, x = aw_2743_cast_fp16)[name = tensor("op_32660_cast_fp16")]; + tensor var_32661_cast_fp16 = softmax(axis = var_30355, x = aw_2745_cast_fp16)[name = tensor("op_32661_cast_fp16")]; + tensor var_32662_cast_fp16 = softmax(axis = var_30355, x = aw_2747_cast_fp16)[name = tensor("op_32662_cast_fp16")]; + tensor var_32663_cast_fp16 = softmax(axis = var_30355, x = aw_2749_cast_fp16)[name = tensor("op_32663_cast_fp16")]; + tensor var_32664_cast_fp16 = softmax(axis = var_30355, x = aw_2751_cast_fp16)[name = tensor("op_32664_cast_fp16")]; + tensor var_32665_cast_fp16 = softmax(axis = var_30355, x = aw_2753_cast_fp16)[name = tensor("op_32665_cast_fp16")]; + tensor var_32666_cast_fp16 = softmax(axis = var_30355, x = aw_2755_cast_fp16)[name = tensor("op_32666_cast_fp16")]; + tensor var_32667_cast_fp16 = softmax(axis = var_30355, x = aw_2757_cast_fp16)[name = tensor("op_32667_cast_fp16")]; + tensor var_32668_cast_fp16 = softmax(axis = var_30355, x = aw_2759_cast_fp16)[name = tensor("op_32668_cast_fp16")]; + tensor var_32670_equation_0 = const()[name = tensor("op_32670_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32670_cast_fp16 = einsum(equation = var_32670_equation_0, values = (var_32490_cast_fp16, var_32649_cast_fp16))[name = tensor("op_32670_cast_fp16")]; + tensor var_32672_equation_0 = const()[name = tensor("op_32672_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32672_cast_fp16 = einsum(equation = var_32672_equation_0, values = (var_32494_cast_fp16, var_32650_cast_fp16))[name = tensor("op_32672_cast_fp16")]; + tensor var_32674_equation_0 = const()[name = tensor("op_32674_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32674_cast_fp16 = einsum(equation = var_32674_equation_0, values = (var_32498_cast_fp16, var_32651_cast_fp16))[name = tensor("op_32674_cast_fp16")]; + tensor var_32676_equation_0 = const()[name = tensor("op_32676_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32676_cast_fp16 = einsum(equation = var_32676_equation_0, values = (var_32502_cast_fp16, var_32652_cast_fp16))[name = tensor("op_32676_cast_fp16")]; + tensor var_32678_equation_0 = const()[name = tensor("op_32678_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32678_cast_fp16 = einsum(equation = var_32678_equation_0, values = (var_32506_cast_fp16, var_32653_cast_fp16))[name = tensor("op_32678_cast_fp16")]; + tensor var_32680_equation_0 = const()[name = tensor("op_32680_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32680_cast_fp16 = einsum(equation = var_32680_equation_0, values = (var_32510_cast_fp16, var_32654_cast_fp16))[name = tensor("op_32680_cast_fp16")]; + tensor var_32682_equation_0 = const()[name = tensor("op_32682_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32682_cast_fp16 = einsum(equation = var_32682_equation_0, values = (var_32514_cast_fp16, var_32655_cast_fp16))[name = tensor("op_32682_cast_fp16")]; + tensor var_32684_equation_0 = const()[name = tensor("op_32684_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32684_cast_fp16 = einsum(equation = var_32684_equation_0, values = (var_32518_cast_fp16, var_32656_cast_fp16))[name = tensor("op_32684_cast_fp16")]; + tensor var_32686_equation_0 = const()[name = tensor("op_32686_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32686_cast_fp16 = einsum(equation = var_32686_equation_0, values = (var_32522_cast_fp16, var_32657_cast_fp16))[name = tensor("op_32686_cast_fp16")]; + tensor var_32688_equation_0 = const()[name = tensor("op_32688_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32688_cast_fp16 = einsum(equation = var_32688_equation_0, values = (var_32526_cast_fp16, var_32658_cast_fp16))[name = tensor("op_32688_cast_fp16")]; + tensor var_32690_equation_0 = const()[name = tensor("op_32690_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32690_cast_fp16 = einsum(equation = var_32690_equation_0, values = (var_32530_cast_fp16, var_32659_cast_fp16))[name = tensor("op_32690_cast_fp16")]; + tensor var_32692_equation_0 = const()[name = tensor("op_32692_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32692_cast_fp16 = einsum(equation = var_32692_equation_0, values = (var_32534_cast_fp16, var_32660_cast_fp16))[name = tensor("op_32692_cast_fp16")]; + tensor var_32694_equation_0 = const()[name = tensor("op_32694_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32694_cast_fp16 = einsum(equation = var_32694_equation_0, values = (var_32538_cast_fp16, var_32661_cast_fp16))[name = tensor("op_32694_cast_fp16")]; + tensor var_32696_equation_0 = const()[name = tensor("op_32696_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32696_cast_fp16 = einsum(equation = var_32696_equation_0, values = (var_32542_cast_fp16, var_32662_cast_fp16))[name = tensor("op_32696_cast_fp16")]; + tensor var_32698_equation_0 = const()[name = tensor("op_32698_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32698_cast_fp16 = einsum(equation = var_32698_equation_0, values = (var_32546_cast_fp16, var_32663_cast_fp16))[name = tensor("op_32698_cast_fp16")]; + tensor var_32700_equation_0 = const()[name = tensor("op_32700_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32700_cast_fp16 = einsum(equation = var_32700_equation_0, values = (var_32550_cast_fp16, var_32664_cast_fp16))[name = tensor("op_32700_cast_fp16")]; + tensor var_32702_equation_0 = const()[name = tensor("op_32702_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32702_cast_fp16 = einsum(equation = var_32702_equation_0, values = (var_32554_cast_fp16, var_32665_cast_fp16))[name = tensor("op_32702_cast_fp16")]; + tensor var_32704_equation_0 = const()[name = tensor("op_32704_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32704_cast_fp16 = einsum(equation = var_32704_equation_0, values = (var_32558_cast_fp16, var_32666_cast_fp16))[name = tensor("op_32704_cast_fp16")]; + tensor var_32706_equation_0 = const()[name = tensor("op_32706_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32706_cast_fp16 = einsum(equation = var_32706_equation_0, values = (var_32562_cast_fp16, var_32667_cast_fp16))[name = tensor("op_32706_cast_fp16")]; + tensor var_32708_equation_0 = const()[name = tensor("op_32708_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32708_cast_fp16 = einsum(equation = var_32708_equation_0, values = (var_32566_cast_fp16, var_32668_cast_fp16))[name = tensor("op_32708_cast_fp16")]; + tensor input_455_interleave_0 = const()[name = tensor("input_455_interleave_0"), val = tensor(false)]; + tensor input_455_cast_fp16 = concat(axis = var_30355, interleave = input_455_interleave_0, values = (var_32670_cast_fp16, var_32672_cast_fp16, var_32674_cast_fp16, var_32676_cast_fp16, var_32678_cast_fp16, var_32680_cast_fp16, var_32682_cast_fp16, var_32684_cast_fp16, var_32686_cast_fp16, var_32688_cast_fp16, var_32690_cast_fp16, var_32692_cast_fp16, var_32694_cast_fp16, var_32696_cast_fp16, var_32698_cast_fp16, var_32700_cast_fp16, var_32702_cast_fp16, var_32704_cast_fp16, var_32706_cast_fp16, var_32708_cast_fp16))[name = tensor("input_455_cast_fp16")]; + tensor var_32714 = const()[name = tensor("op_32714"), val = tensor([1, 1])]; + tensor var_32716 = const()[name = tensor("op_32716"), val = tensor([1, 1])]; + tensor var_32718_pad_type_0 = const()[name = tensor("op_32718_pad_type_0"), val = tensor("custom")]; + tensor var_32718_pad_0 = const()[name = tensor("op_32718_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_2_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_2_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2737283136)))]; + tensor up_blocks_0_attentions_0_transformer_blocks_2_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_2_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2740560000)))]; + tensor var_32718_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_2_attn1_to_out_0_bias_to_fp16, dilations = var_32716, groups = var_30355, pad = var_32718_pad_0, pad_type = var_32718_pad_type_0, strides = var_32714, weight = up_blocks_0_attentions_0_transformer_blocks_2_attn1_to_out_0_weight_to_fp16, x = input_455_cast_fp16)[name = tensor("op_32718_cast_fp16")]; + tensor inputs_219_cast_fp16 = add(x = var_32718_cast_fp16, y = inputs_217_cast_fp16)[name = tensor("inputs_219_cast_fp16")]; + tensor hidden_states_301_axes_0 = const()[name = tensor("hidden_states_301_axes_0"), val = tensor([1])]; + tensor hidden_states_301_gamma_0_to_fp16 = const()[name = tensor("hidden_states_301_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2740562624)))]; + tensor hidden_states_301_beta_0_to_fp16 = const()[name = tensor("hidden_states_301_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2740565248)))]; + tensor var_32728_to_fp16 = const()[name = tensor("op_32728_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_301_cast_fp16 = layer_norm(axes = hidden_states_301_axes_0, beta = hidden_states_301_beta_0_to_fp16, epsilon = var_32728_to_fp16, gamma = hidden_states_301_gamma_0_to_fp16, x = inputs_219_cast_fp16)[name = tensor("hidden_states_301_cast_fp16")]; + tensor var_32743 = const()[name = tensor("op_32743"), val = tensor([1, 1])]; + tensor var_32745 = const()[name = tensor("op_32745"), val = tensor([1, 1])]; + tensor q_147_pad_type_0 = const()[name = tensor("q_147_pad_type_0"), val = tensor("custom")]; + tensor q_147_pad_0 = const()[name = tensor("q_147_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_2_attn2_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_2_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2740567872)))]; + tensor q_147_cast_fp16 = conv(dilations = var_32745, groups = var_30355, pad = q_147_pad_0, pad_type = q_147_pad_type_0, strides = var_32743, weight = up_blocks_0_attentions_0_transformer_blocks_2_attn2_to_q_weight_to_fp16, x = hidden_states_301_cast_fp16)[name = tensor("q_147_cast_fp16")]; + tensor var_32749 = const()[name = tensor("op_32749"), val = tensor([1, 1])]; + tensor var_32751 = const()[name = tensor("op_32751"), val = tensor([1, 1])]; + tensor k_293_pad_type_0 = const()[name = tensor("k_293_pad_type_0"), val = tensor("custom")]; + tensor k_293_pad_0 = const()[name = tensor("k_293_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_2_attn2_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_2_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2743844736)))]; + tensor k_293_cast_fp16 = conv(dilations = var_32751, groups = var_30355, pad = k_293_pad_0, pad_type = k_293_pad_type_0, strides = var_32749, weight = up_blocks_0_attentions_0_transformer_blocks_2_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_293_cast_fp16")]; + tensor var_32755 = const()[name = tensor("op_32755"), val = tensor([1, 1])]; + tensor var_32757 = const()[name = tensor("op_32757"), val = tensor([1, 1])]; + tensor v_147_pad_type_0 = const()[name = tensor("v_147_pad_type_0"), val = tensor("custom")]; + tensor v_147_pad_0 = const()[name = tensor("v_147_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_2_attn2_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_2_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2749087680)))]; + tensor v_147_cast_fp16 = conv(dilations = var_32757, groups = var_30355, pad = v_147_pad_0, pad_type = v_147_pad_type_0, strides = var_32755, weight = up_blocks_0_attentions_0_transformer_blocks_2_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_147_cast_fp16")]; + tensor var_32761_begin_0 = const()[name = tensor("op_32761_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_32761_end_0 = const()[name = tensor("op_32761_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_32761_end_mask_0 = const()[name = tensor("op_32761_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32761_cast_fp16 = slice_by_index(begin = var_32761_begin_0, end = var_32761_end_0, end_mask = var_32761_end_mask_0, x = q_147_cast_fp16)[name = tensor("op_32761_cast_fp16")]; + tensor var_32765_begin_0 = const()[name = tensor("op_32765_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_32765_end_0 = const()[name = tensor("op_32765_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_32765_end_mask_0 = const()[name = tensor("op_32765_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32765_cast_fp16 = slice_by_index(begin = var_32765_begin_0, end = var_32765_end_0, end_mask = var_32765_end_mask_0, x = q_147_cast_fp16)[name = tensor("op_32765_cast_fp16")]; + tensor var_32769_begin_0 = const()[name = tensor("op_32769_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_32769_end_0 = const()[name = tensor("op_32769_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_32769_end_mask_0 = const()[name = tensor("op_32769_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32769_cast_fp16 = slice_by_index(begin = var_32769_begin_0, end = var_32769_end_0, end_mask = var_32769_end_mask_0, x = q_147_cast_fp16)[name = tensor("op_32769_cast_fp16")]; + tensor var_32773_begin_0 = const()[name = tensor("op_32773_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_32773_end_0 = const()[name = tensor("op_32773_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_32773_end_mask_0 = const()[name = tensor("op_32773_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32773_cast_fp16 = slice_by_index(begin = var_32773_begin_0, end = var_32773_end_0, end_mask = var_32773_end_mask_0, x = q_147_cast_fp16)[name = tensor("op_32773_cast_fp16")]; + tensor var_32777_begin_0 = const()[name = tensor("op_32777_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_32777_end_0 = const()[name = tensor("op_32777_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_32777_end_mask_0 = const()[name = tensor("op_32777_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32777_cast_fp16 = slice_by_index(begin = var_32777_begin_0, end = var_32777_end_0, end_mask = var_32777_end_mask_0, x = q_147_cast_fp16)[name = tensor("op_32777_cast_fp16")]; + tensor var_32781_begin_0 = const()[name = tensor("op_32781_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_32781_end_0 = const()[name = tensor("op_32781_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_32781_end_mask_0 = const()[name = tensor("op_32781_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32781_cast_fp16 = slice_by_index(begin = var_32781_begin_0, end = var_32781_end_0, end_mask = var_32781_end_mask_0, x = q_147_cast_fp16)[name = tensor("op_32781_cast_fp16")]; + tensor var_32785_begin_0 = const()[name = tensor("op_32785_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_32785_end_0 = const()[name = tensor("op_32785_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_32785_end_mask_0 = const()[name = tensor("op_32785_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32785_cast_fp16 = slice_by_index(begin = var_32785_begin_0, end = var_32785_end_0, end_mask = var_32785_end_mask_0, x = q_147_cast_fp16)[name = tensor("op_32785_cast_fp16")]; + tensor var_32789_begin_0 = const()[name = tensor("op_32789_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_32789_end_0 = const()[name = tensor("op_32789_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_32789_end_mask_0 = const()[name = tensor("op_32789_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32789_cast_fp16 = slice_by_index(begin = var_32789_begin_0, end = var_32789_end_0, end_mask = var_32789_end_mask_0, x = q_147_cast_fp16)[name = tensor("op_32789_cast_fp16")]; + tensor var_32793_begin_0 = const()[name = tensor("op_32793_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_32793_end_0 = const()[name = tensor("op_32793_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_32793_end_mask_0 = const()[name = tensor("op_32793_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32793_cast_fp16 = slice_by_index(begin = var_32793_begin_0, end = var_32793_end_0, end_mask = var_32793_end_mask_0, x = q_147_cast_fp16)[name = tensor("op_32793_cast_fp16")]; + tensor var_32797_begin_0 = const()[name = tensor("op_32797_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_32797_end_0 = const()[name = tensor("op_32797_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_32797_end_mask_0 = const()[name = tensor("op_32797_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32797_cast_fp16 = slice_by_index(begin = var_32797_begin_0, end = var_32797_end_0, end_mask = var_32797_end_mask_0, x = q_147_cast_fp16)[name = tensor("op_32797_cast_fp16")]; + tensor var_32801_begin_0 = const()[name = tensor("op_32801_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_32801_end_0 = const()[name = tensor("op_32801_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_32801_end_mask_0 = const()[name = tensor("op_32801_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32801_cast_fp16 = slice_by_index(begin = var_32801_begin_0, end = var_32801_end_0, end_mask = var_32801_end_mask_0, x = q_147_cast_fp16)[name = tensor("op_32801_cast_fp16")]; + tensor var_32805_begin_0 = const()[name = tensor("op_32805_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_32805_end_0 = const()[name = tensor("op_32805_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_32805_end_mask_0 = const()[name = tensor("op_32805_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32805_cast_fp16 = slice_by_index(begin = var_32805_begin_0, end = var_32805_end_0, end_mask = var_32805_end_mask_0, x = q_147_cast_fp16)[name = tensor("op_32805_cast_fp16")]; + tensor var_32809_begin_0 = const()[name = tensor("op_32809_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_32809_end_0 = const()[name = tensor("op_32809_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_32809_end_mask_0 = const()[name = tensor("op_32809_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32809_cast_fp16 = slice_by_index(begin = var_32809_begin_0, end = var_32809_end_0, end_mask = var_32809_end_mask_0, x = q_147_cast_fp16)[name = tensor("op_32809_cast_fp16")]; + tensor var_32813_begin_0 = const()[name = tensor("op_32813_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_32813_end_0 = const()[name = tensor("op_32813_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_32813_end_mask_0 = const()[name = tensor("op_32813_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32813_cast_fp16 = slice_by_index(begin = var_32813_begin_0, end = var_32813_end_0, end_mask = var_32813_end_mask_0, x = q_147_cast_fp16)[name = tensor("op_32813_cast_fp16")]; + tensor var_32817_begin_0 = const()[name = tensor("op_32817_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_32817_end_0 = const()[name = tensor("op_32817_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_32817_end_mask_0 = const()[name = tensor("op_32817_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32817_cast_fp16 = slice_by_index(begin = var_32817_begin_0, end = var_32817_end_0, end_mask = var_32817_end_mask_0, x = q_147_cast_fp16)[name = tensor("op_32817_cast_fp16")]; + tensor var_32821_begin_0 = const()[name = tensor("op_32821_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_32821_end_0 = const()[name = tensor("op_32821_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_32821_end_mask_0 = const()[name = tensor("op_32821_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32821_cast_fp16 = slice_by_index(begin = var_32821_begin_0, end = var_32821_end_0, end_mask = var_32821_end_mask_0, x = q_147_cast_fp16)[name = tensor("op_32821_cast_fp16")]; + tensor var_32825_begin_0 = const()[name = tensor("op_32825_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_32825_end_0 = const()[name = tensor("op_32825_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_32825_end_mask_0 = const()[name = tensor("op_32825_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32825_cast_fp16 = slice_by_index(begin = var_32825_begin_0, end = var_32825_end_0, end_mask = var_32825_end_mask_0, x = q_147_cast_fp16)[name = tensor("op_32825_cast_fp16")]; + tensor var_32829_begin_0 = const()[name = tensor("op_32829_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_32829_end_0 = const()[name = tensor("op_32829_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_32829_end_mask_0 = const()[name = tensor("op_32829_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32829_cast_fp16 = slice_by_index(begin = var_32829_begin_0, end = var_32829_end_0, end_mask = var_32829_end_mask_0, x = q_147_cast_fp16)[name = tensor("op_32829_cast_fp16")]; + tensor var_32833_begin_0 = const()[name = tensor("op_32833_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_32833_end_0 = const()[name = tensor("op_32833_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_32833_end_mask_0 = const()[name = tensor("op_32833_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32833_cast_fp16 = slice_by_index(begin = var_32833_begin_0, end = var_32833_end_0, end_mask = var_32833_end_mask_0, x = q_147_cast_fp16)[name = tensor("op_32833_cast_fp16")]; + tensor var_32837_begin_0 = const()[name = tensor("op_32837_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_32837_end_0 = const()[name = tensor("op_32837_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_32837_end_mask_0 = const()[name = tensor("op_32837_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32837_cast_fp16 = slice_by_index(begin = var_32837_begin_0, end = var_32837_end_0, end_mask = var_32837_end_mask_0, x = q_147_cast_fp16)[name = tensor("op_32837_cast_fp16")]; + tensor k_295_perm_0 = const()[name = tensor("k_295_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_32844_begin_0 = const()[name = tensor("op_32844_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_32844_end_0 = const()[name = tensor("op_32844_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_32844_end_mask_0 = const()[name = tensor("op_32844_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_295_cast_fp16 = transpose(perm = k_295_perm_0, x = k_293_cast_fp16)[name = tensor("transpose_66")]; + tensor var_32844_cast_fp16 = slice_by_index(begin = var_32844_begin_0, end = var_32844_end_0, end_mask = var_32844_end_mask_0, x = k_295_cast_fp16)[name = tensor("op_32844_cast_fp16")]; + tensor var_32848_begin_0 = const()[name = tensor("op_32848_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_32848_end_0 = const()[name = tensor("op_32848_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_32848_end_mask_0 = const()[name = tensor("op_32848_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32848_cast_fp16 = slice_by_index(begin = var_32848_begin_0, end = var_32848_end_0, end_mask = var_32848_end_mask_0, x = k_295_cast_fp16)[name = tensor("op_32848_cast_fp16")]; + tensor var_32852_begin_0 = const()[name = tensor("op_32852_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_32852_end_0 = const()[name = tensor("op_32852_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_32852_end_mask_0 = const()[name = tensor("op_32852_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32852_cast_fp16 = slice_by_index(begin = var_32852_begin_0, end = var_32852_end_0, end_mask = var_32852_end_mask_0, x = k_295_cast_fp16)[name = tensor("op_32852_cast_fp16")]; + tensor var_32856_begin_0 = const()[name = tensor("op_32856_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_32856_end_0 = const()[name = tensor("op_32856_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_32856_end_mask_0 = const()[name = tensor("op_32856_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32856_cast_fp16 = slice_by_index(begin = var_32856_begin_0, end = var_32856_end_0, end_mask = var_32856_end_mask_0, x = k_295_cast_fp16)[name = tensor("op_32856_cast_fp16")]; + tensor var_32860_begin_0 = const()[name = tensor("op_32860_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_32860_end_0 = const()[name = tensor("op_32860_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_32860_end_mask_0 = const()[name = tensor("op_32860_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32860_cast_fp16 = slice_by_index(begin = var_32860_begin_0, end = var_32860_end_0, end_mask = var_32860_end_mask_0, x = k_295_cast_fp16)[name = tensor("op_32860_cast_fp16")]; + tensor var_32864_begin_0 = const()[name = tensor("op_32864_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_32864_end_0 = const()[name = tensor("op_32864_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_32864_end_mask_0 = const()[name = tensor("op_32864_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32864_cast_fp16 = slice_by_index(begin = var_32864_begin_0, end = var_32864_end_0, end_mask = var_32864_end_mask_0, x = k_295_cast_fp16)[name = tensor("op_32864_cast_fp16")]; + tensor var_32868_begin_0 = const()[name = tensor("op_32868_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_32868_end_0 = const()[name = tensor("op_32868_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_32868_end_mask_0 = const()[name = tensor("op_32868_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32868_cast_fp16 = slice_by_index(begin = var_32868_begin_0, end = var_32868_end_0, end_mask = var_32868_end_mask_0, x = k_295_cast_fp16)[name = tensor("op_32868_cast_fp16")]; + tensor var_32872_begin_0 = const()[name = tensor("op_32872_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_32872_end_0 = const()[name = tensor("op_32872_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_32872_end_mask_0 = const()[name = tensor("op_32872_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32872_cast_fp16 = slice_by_index(begin = var_32872_begin_0, end = var_32872_end_0, end_mask = var_32872_end_mask_0, x = k_295_cast_fp16)[name = tensor("op_32872_cast_fp16")]; + tensor var_32876_begin_0 = const()[name = tensor("op_32876_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_32876_end_0 = const()[name = tensor("op_32876_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_32876_end_mask_0 = const()[name = tensor("op_32876_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32876_cast_fp16 = slice_by_index(begin = var_32876_begin_0, end = var_32876_end_0, end_mask = var_32876_end_mask_0, x = k_295_cast_fp16)[name = tensor("op_32876_cast_fp16")]; + tensor var_32880_begin_0 = const()[name = tensor("op_32880_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_32880_end_0 = const()[name = tensor("op_32880_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_32880_end_mask_0 = const()[name = tensor("op_32880_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32880_cast_fp16 = slice_by_index(begin = var_32880_begin_0, end = var_32880_end_0, end_mask = var_32880_end_mask_0, x = k_295_cast_fp16)[name = tensor("op_32880_cast_fp16")]; + tensor var_32884_begin_0 = const()[name = tensor("op_32884_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_32884_end_0 = const()[name = tensor("op_32884_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_32884_end_mask_0 = const()[name = tensor("op_32884_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32884_cast_fp16 = slice_by_index(begin = var_32884_begin_0, end = var_32884_end_0, end_mask = var_32884_end_mask_0, x = k_295_cast_fp16)[name = tensor("op_32884_cast_fp16")]; + tensor var_32888_begin_0 = const()[name = tensor("op_32888_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_32888_end_0 = const()[name = tensor("op_32888_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_32888_end_mask_0 = const()[name = tensor("op_32888_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32888_cast_fp16 = slice_by_index(begin = var_32888_begin_0, end = var_32888_end_0, end_mask = var_32888_end_mask_0, x = k_295_cast_fp16)[name = tensor("op_32888_cast_fp16")]; + tensor var_32892_begin_0 = const()[name = tensor("op_32892_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_32892_end_0 = const()[name = tensor("op_32892_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_32892_end_mask_0 = const()[name = tensor("op_32892_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32892_cast_fp16 = slice_by_index(begin = var_32892_begin_0, end = var_32892_end_0, end_mask = var_32892_end_mask_0, x = k_295_cast_fp16)[name = tensor("op_32892_cast_fp16")]; + tensor var_32896_begin_0 = const()[name = tensor("op_32896_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_32896_end_0 = const()[name = tensor("op_32896_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_32896_end_mask_0 = const()[name = tensor("op_32896_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32896_cast_fp16 = slice_by_index(begin = var_32896_begin_0, end = var_32896_end_0, end_mask = var_32896_end_mask_0, x = k_295_cast_fp16)[name = tensor("op_32896_cast_fp16")]; + tensor var_32900_begin_0 = const()[name = tensor("op_32900_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_32900_end_0 = const()[name = tensor("op_32900_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_32900_end_mask_0 = const()[name = tensor("op_32900_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32900_cast_fp16 = slice_by_index(begin = var_32900_begin_0, end = var_32900_end_0, end_mask = var_32900_end_mask_0, x = k_295_cast_fp16)[name = tensor("op_32900_cast_fp16")]; + tensor var_32904_begin_0 = const()[name = tensor("op_32904_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_32904_end_0 = const()[name = tensor("op_32904_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_32904_end_mask_0 = const()[name = tensor("op_32904_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32904_cast_fp16 = slice_by_index(begin = var_32904_begin_0, end = var_32904_end_0, end_mask = var_32904_end_mask_0, x = k_295_cast_fp16)[name = tensor("op_32904_cast_fp16")]; + tensor var_32908_begin_0 = const()[name = tensor("op_32908_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_32908_end_0 = const()[name = tensor("op_32908_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_32908_end_mask_0 = const()[name = tensor("op_32908_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32908_cast_fp16 = slice_by_index(begin = var_32908_begin_0, end = var_32908_end_0, end_mask = var_32908_end_mask_0, x = k_295_cast_fp16)[name = tensor("op_32908_cast_fp16")]; + tensor var_32912_begin_0 = const()[name = tensor("op_32912_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_32912_end_0 = const()[name = tensor("op_32912_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_32912_end_mask_0 = const()[name = tensor("op_32912_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32912_cast_fp16 = slice_by_index(begin = var_32912_begin_0, end = var_32912_end_0, end_mask = var_32912_end_mask_0, x = k_295_cast_fp16)[name = tensor("op_32912_cast_fp16")]; + tensor var_32916_begin_0 = const()[name = tensor("op_32916_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_32916_end_0 = const()[name = tensor("op_32916_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_32916_end_mask_0 = const()[name = tensor("op_32916_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32916_cast_fp16 = slice_by_index(begin = var_32916_begin_0, end = var_32916_end_0, end_mask = var_32916_end_mask_0, x = k_295_cast_fp16)[name = tensor("op_32916_cast_fp16")]; + tensor var_32920_begin_0 = const()[name = tensor("op_32920_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_32920_end_0 = const()[name = tensor("op_32920_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_32920_end_mask_0 = const()[name = tensor("op_32920_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32920_cast_fp16 = slice_by_index(begin = var_32920_begin_0, end = var_32920_end_0, end_mask = var_32920_end_mask_0, x = k_295_cast_fp16)[name = tensor("op_32920_cast_fp16")]; + tensor var_32922_begin_0 = const()[name = tensor("op_32922_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_32922_end_0 = const()[name = tensor("op_32922_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_32922_end_mask_0 = const()[name = tensor("op_32922_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32922_cast_fp16 = slice_by_index(begin = var_32922_begin_0, end = var_32922_end_0, end_mask = var_32922_end_mask_0, x = v_147_cast_fp16)[name = tensor("op_32922_cast_fp16")]; + tensor var_32926_begin_0 = const()[name = tensor("op_32926_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_32926_end_0 = const()[name = tensor("op_32926_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_32926_end_mask_0 = const()[name = tensor("op_32926_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32926_cast_fp16 = slice_by_index(begin = var_32926_begin_0, end = var_32926_end_0, end_mask = var_32926_end_mask_0, x = v_147_cast_fp16)[name = tensor("op_32926_cast_fp16")]; + tensor var_32930_begin_0 = const()[name = tensor("op_32930_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_32930_end_0 = const()[name = tensor("op_32930_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_32930_end_mask_0 = const()[name = tensor("op_32930_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32930_cast_fp16 = slice_by_index(begin = var_32930_begin_0, end = var_32930_end_0, end_mask = var_32930_end_mask_0, x = v_147_cast_fp16)[name = tensor("op_32930_cast_fp16")]; + tensor var_32934_begin_0 = const()[name = tensor("op_32934_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_32934_end_0 = const()[name = tensor("op_32934_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_32934_end_mask_0 = const()[name = tensor("op_32934_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32934_cast_fp16 = slice_by_index(begin = var_32934_begin_0, end = var_32934_end_0, end_mask = var_32934_end_mask_0, x = v_147_cast_fp16)[name = tensor("op_32934_cast_fp16")]; + tensor var_32938_begin_0 = const()[name = tensor("op_32938_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_32938_end_0 = const()[name = tensor("op_32938_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_32938_end_mask_0 = const()[name = tensor("op_32938_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32938_cast_fp16 = slice_by_index(begin = var_32938_begin_0, end = var_32938_end_0, end_mask = var_32938_end_mask_0, x = v_147_cast_fp16)[name = tensor("op_32938_cast_fp16")]; + tensor var_32942_begin_0 = const()[name = tensor("op_32942_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_32942_end_0 = const()[name = tensor("op_32942_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_32942_end_mask_0 = const()[name = tensor("op_32942_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32942_cast_fp16 = slice_by_index(begin = var_32942_begin_0, end = var_32942_end_0, end_mask = var_32942_end_mask_0, x = v_147_cast_fp16)[name = tensor("op_32942_cast_fp16")]; + tensor var_32946_begin_0 = const()[name = tensor("op_32946_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_32946_end_0 = const()[name = tensor("op_32946_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_32946_end_mask_0 = const()[name = tensor("op_32946_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32946_cast_fp16 = slice_by_index(begin = var_32946_begin_0, end = var_32946_end_0, end_mask = var_32946_end_mask_0, x = v_147_cast_fp16)[name = tensor("op_32946_cast_fp16")]; + tensor var_32950_begin_0 = const()[name = tensor("op_32950_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_32950_end_0 = const()[name = tensor("op_32950_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_32950_end_mask_0 = const()[name = tensor("op_32950_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32950_cast_fp16 = slice_by_index(begin = var_32950_begin_0, end = var_32950_end_0, end_mask = var_32950_end_mask_0, x = v_147_cast_fp16)[name = tensor("op_32950_cast_fp16")]; + tensor var_32954_begin_0 = const()[name = tensor("op_32954_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_32954_end_0 = const()[name = tensor("op_32954_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_32954_end_mask_0 = const()[name = tensor("op_32954_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32954_cast_fp16 = slice_by_index(begin = var_32954_begin_0, end = var_32954_end_0, end_mask = var_32954_end_mask_0, x = v_147_cast_fp16)[name = tensor("op_32954_cast_fp16")]; + tensor var_32958_begin_0 = const()[name = tensor("op_32958_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_32958_end_0 = const()[name = tensor("op_32958_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_32958_end_mask_0 = const()[name = tensor("op_32958_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32958_cast_fp16 = slice_by_index(begin = var_32958_begin_0, end = var_32958_end_0, end_mask = var_32958_end_mask_0, x = v_147_cast_fp16)[name = tensor("op_32958_cast_fp16")]; + tensor var_32962_begin_0 = const()[name = tensor("op_32962_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_32962_end_0 = const()[name = tensor("op_32962_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_32962_end_mask_0 = const()[name = tensor("op_32962_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32962_cast_fp16 = slice_by_index(begin = var_32962_begin_0, end = var_32962_end_0, end_mask = var_32962_end_mask_0, x = v_147_cast_fp16)[name = tensor("op_32962_cast_fp16")]; + tensor var_32966_begin_0 = const()[name = tensor("op_32966_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_32966_end_0 = const()[name = tensor("op_32966_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_32966_end_mask_0 = const()[name = tensor("op_32966_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32966_cast_fp16 = slice_by_index(begin = var_32966_begin_0, end = var_32966_end_0, end_mask = var_32966_end_mask_0, x = v_147_cast_fp16)[name = tensor("op_32966_cast_fp16")]; + tensor var_32970_begin_0 = const()[name = tensor("op_32970_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_32970_end_0 = const()[name = tensor("op_32970_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_32970_end_mask_0 = const()[name = tensor("op_32970_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32970_cast_fp16 = slice_by_index(begin = var_32970_begin_0, end = var_32970_end_0, end_mask = var_32970_end_mask_0, x = v_147_cast_fp16)[name = tensor("op_32970_cast_fp16")]; + tensor var_32974_begin_0 = const()[name = tensor("op_32974_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_32974_end_0 = const()[name = tensor("op_32974_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_32974_end_mask_0 = const()[name = tensor("op_32974_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32974_cast_fp16 = slice_by_index(begin = var_32974_begin_0, end = var_32974_end_0, end_mask = var_32974_end_mask_0, x = v_147_cast_fp16)[name = tensor("op_32974_cast_fp16")]; + tensor var_32978_begin_0 = const()[name = tensor("op_32978_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_32978_end_0 = const()[name = tensor("op_32978_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_32978_end_mask_0 = const()[name = tensor("op_32978_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32978_cast_fp16 = slice_by_index(begin = var_32978_begin_0, end = var_32978_end_0, end_mask = var_32978_end_mask_0, x = v_147_cast_fp16)[name = tensor("op_32978_cast_fp16")]; + tensor var_32982_begin_0 = const()[name = tensor("op_32982_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_32982_end_0 = const()[name = tensor("op_32982_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_32982_end_mask_0 = const()[name = tensor("op_32982_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32982_cast_fp16 = slice_by_index(begin = var_32982_begin_0, end = var_32982_end_0, end_mask = var_32982_end_mask_0, x = v_147_cast_fp16)[name = tensor("op_32982_cast_fp16")]; + tensor var_32986_begin_0 = const()[name = tensor("op_32986_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_32986_end_0 = const()[name = tensor("op_32986_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_32986_end_mask_0 = const()[name = tensor("op_32986_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32986_cast_fp16 = slice_by_index(begin = var_32986_begin_0, end = var_32986_end_0, end_mask = var_32986_end_mask_0, x = v_147_cast_fp16)[name = tensor("op_32986_cast_fp16")]; + tensor var_32990_begin_0 = const()[name = tensor("op_32990_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_32990_end_0 = const()[name = tensor("op_32990_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_32990_end_mask_0 = const()[name = tensor("op_32990_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32990_cast_fp16 = slice_by_index(begin = var_32990_begin_0, end = var_32990_end_0, end_mask = var_32990_end_mask_0, x = v_147_cast_fp16)[name = tensor("op_32990_cast_fp16")]; + tensor var_32994_begin_0 = const()[name = tensor("op_32994_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_32994_end_0 = const()[name = tensor("op_32994_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_32994_end_mask_0 = const()[name = tensor("op_32994_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32994_cast_fp16 = slice_by_index(begin = var_32994_begin_0, end = var_32994_end_0, end_mask = var_32994_end_mask_0, x = v_147_cast_fp16)[name = tensor("op_32994_cast_fp16")]; + tensor var_32998_begin_0 = const()[name = tensor("op_32998_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_32998_end_0 = const()[name = tensor("op_32998_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_32998_end_mask_0 = const()[name = tensor("op_32998_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32998_cast_fp16 = slice_by_index(begin = var_32998_begin_0, end = var_32998_end_0, end_mask = var_32998_end_mask_0, x = v_147_cast_fp16)[name = tensor("op_32998_cast_fp16")]; + tensor var_33002_equation_0 = const()[name = tensor("op_33002_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33002_cast_fp16 = einsum(equation = var_33002_equation_0, values = (var_32844_cast_fp16, var_32761_cast_fp16))[name = tensor("op_33002_cast_fp16")]; + tensor var_33003_to_fp16 = const()[name = tensor("op_33003_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2761_cast_fp16 = mul(x = var_33002_cast_fp16, y = var_33003_to_fp16)[name = tensor("aw_2761_cast_fp16")]; + tensor var_33006_equation_0 = const()[name = tensor("op_33006_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33006_cast_fp16 = einsum(equation = var_33006_equation_0, values = (var_32848_cast_fp16, var_32765_cast_fp16))[name = tensor("op_33006_cast_fp16")]; + tensor var_33007_to_fp16 = const()[name = tensor("op_33007_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2763_cast_fp16 = mul(x = var_33006_cast_fp16, y = var_33007_to_fp16)[name = tensor("aw_2763_cast_fp16")]; + tensor var_33010_equation_0 = const()[name = tensor("op_33010_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33010_cast_fp16 = einsum(equation = var_33010_equation_0, values = (var_32852_cast_fp16, var_32769_cast_fp16))[name = tensor("op_33010_cast_fp16")]; + tensor var_33011_to_fp16 = const()[name = tensor("op_33011_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2765_cast_fp16 = mul(x = var_33010_cast_fp16, y = var_33011_to_fp16)[name = tensor("aw_2765_cast_fp16")]; + tensor var_33014_equation_0 = const()[name = tensor("op_33014_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33014_cast_fp16 = einsum(equation = var_33014_equation_0, values = (var_32856_cast_fp16, var_32773_cast_fp16))[name = tensor("op_33014_cast_fp16")]; + tensor var_33015_to_fp16 = const()[name = tensor("op_33015_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2767_cast_fp16 = mul(x = var_33014_cast_fp16, y = var_33015_to_fp16)[name = tensor("aw_2767_cast_fp16")]; + tensor var_33018_equation_0 = const()[name = tensor("op_33018_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33018_cast_fp16 = einsum(equation = var_33018_equation_0, values = (var_32860_cast_fp16, var_32777_cast_fp16))[name = tensor("op_33018_cast_fp16")]; + tensor var_33019_to_fp16 = const()[name = tensor("op_33019_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2769_cast_fp16 = mul(x = var_33018_cast_fp16, y = var_33019_to_fp16)[name = tensor("aw_2769_cast_fp16")]; + tensor var_33022_equation_0 = const()[name = tensor("op_33022_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33022_cast_fp16 = einsum(equation = var_33022_equation_0, values = (var_32864_cast_fp16, var_32781_cast_fp16))[name = tensor("op_33022_cast_fp16")]; + tensor var_33023_to_fp16 = const()[name = tensor("op_33023_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2771_cast_fp16 = mul(x = var_33022_cast_fp16, y = var_33023_to_fp16)[name = tensor("aw_2771_cast_fp16")]; + tensor var_33026_equation_0 = const()[name = tensor("op_33026_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33026_cast_fp16 = einsum(equation = var_33026_equation_0, values = (var_32868_cast_fp16, var_32785_cast_fp16))[name = tensor("op_33026_cast_fp16")]; + tensor var_33027_to_fp16 = const()[name = tensor("op_33027_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2773_cast_fp16 = mul(x = var_33026_cast_fp16, y = var_33027_to_fp16)[name = tensor("aw_2773_cast_fp16")]; + tensor var_33030_equation_0 = const()[name = tensor("op_33030_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33030_cast_fp16 = einsum(equation = var_33030_equation_0, values = (var_32872_cast_fp16, var_32789_cast_fp16))[name = tensor("op_33030_cast_fp16")]; + tensor var_33031_to_fp16 = const()[name = tensor("op_33031_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2775_cast_fp16 = mul(x = var_33030_cast_fp16, y = var_33031_to_fp16)[name = tensor("aw_2775_cast_fp16")]; + tensor var_33034_equation_0 = const()[name = tensor("op_33034_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33034_cast_fp16 = einsum(equation = var_33034_equation_0, values = (var_32876_cast_fp16, var_32793_cast_fp16))[name = tensor("op_33034_cast_fp16")]; + tensor var_33035_to_fp16 = const()[name = tensor("op_33035_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2777_cast_fp16 = mul(x = var_33034_cast_fp16, y = var_33035_to_fp16)[name = tensor("aw_2777_cast_fp16")]; + tensor var_33038_equation_0 = const()[name = tensor("op_33038_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33038_cast_fp16 = einsum(equation = var_33038_equation_0, values = (var_32880_cast_fp16, var_32797_cast_fp16))[name = tensor("op_33038_cast_fp16")]; + tensor var_33039_to_fp16 = const()[name = tensor("op_33039_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2779_cast_fp16 = mul(x = var_33038_cast_fp16, y = var_33039_to_fp16)[name = tensor("aw_2779_cast_fp16")]; + tensor var_33042_equation_0 = const()[name = tensor("op_33042_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33042_cast_fp16 = einsum(equation = var_33042_equation_0, values = (var_32884_cast_fp16, var_32801_cast_fp16))[name = tensor("op_33042_cast_fp16")]; + tensor var_33043_to_fp16 = const()[name = tensor("op_33043_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2781_cast_fp16 = mul(x = var_33042_cast_fp16, y = var_33043_to_fp16)[name = tensor("aw_2781_cast_fp16")]; + tensor var_33046_equation_0 = const()[name = tensor("op_33046_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33046_cast_fp16 = einsum(equation = var_33046_equation_0, values = (var_32888_cast_fp16, var_32805_cast_fp16))[name = tensor("op_33046_cast_fp16")]; + tensor var_33047_to_fp16 = const()[name = tensor("op_33047_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2783_cast_fp16 = mul(x = var_33046_cast_fp16, y = var_33047_to_fp16)[name = tensor("aw_2783_cast_fp16")]; + tensor var_33050_equation_0 = const()[name = tensor("op_33050_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33050_cast_fp16 = einsum(equation = var_33050_equation_0, values = (var_32892_cast_fp16, var_32809_cast_fp16))[name = tensor("op_33050_cast_fp16")]; + tensor var_33051_to_fp16 = const()[name = tensor("op_33051_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2785_cast_fp16 = mul(x = var_33050_cast_fp16, y = var_33051_to_fp16)[name = tensor("aw_2785_cast_fp16")]; + tensor var_33054_equation_0 = const()[name = tensor("op_33054_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33054_cast_fp16 = einsum(equation = var_33054_equation_0, values = (var_32896_cast_fp16, var_32813_cast_fp16))[name = tensor("op_33054_cast_fp16")]; + tensor var_33055_to_fp16 = const()[name = tensor("op_33055_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2787_cast_fp16 = mul(x = var_33054_cast_fp16, y = var_33055_to_fp16)[name = tensor("aw_2787_cast_fp16")]; + tensor var_33058_equation_0 = const()[name = tensor("op_33058_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33058_cast_fp16 = einsum(equation = var_33058_equation_0, values = (var_32900_cast_fp16, var_32817_cast_fp16))[name = tensor("op_33058_cast_fp16")]; + tensor var_33059_to_fp16 = const()[name = tensor("op_33059_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2789_cast_fp16 = mul(x = var_33058_cast_fp16, y = var_33059_to_fp16)[name = tensor("aw_2789_cast_fp16")]; + tensor var_33062_equation_0 = const()[name = tensor("op_33062_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33062_cast_fp16 = einsum(equation = var_33062_equation_0, values = (var_32904_cast_fp16, var_32821_cast_fp16))[name = tensor("op_33062_cast_fp16")]; + tensor var_33063_to_fp16 = const()[name = tensor("op_33063_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2791_cast_fp16 = mul(x = var_33062_cast_fp16, y = var_33063_to_fp16)[name = tensor("aw_2791_cast_fp16")]; + tensor var_33066_equation_0 = const()[name = tensor("op_33066_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33066_cast_fp16 = einsum(equation = var_33066_equation_0, values = (var_32908_cast_fp16, var_32825_cast_fp16))[name = tensor("op_33066_cast_fp16")]; + tensor var_33067_to_fp16 = const()[name = tensor("op_33067_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2793_cast_fp16 = mul(x = var_33066_cast_fp16, y = var_33067_to_fp16)[name = tensor("aw_2793_cast_fp16")]; + tensor var_33070_equation_0 = const()[name = tensor("op_33070_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33070_cast_fp16 = einsum(equation = var_33070_equation_0, values = (var_32912_cast_fp16, var_32829_cast_fp16))[name = tensor("op_33070_cast_fp16")]; + tensor var_33071_to_fp16 = const()[name = tensor("op_33071_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2795_cast_fp16 = mul(x = var_33070_cast_fp16, y = var_33071_to_fp16)[name = tensor("aw_2795_cast_fp16")]; + tensor var_33074_equation_0 = const()[name = tensor("op_33074_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33074_cast_fp16 = einsum(equation = var_33074_equation_0, values = (var_32916_cast_fp16, var_32833_cast_fp16))[name = tensor("op_33074_cast_fp16")]; + tensor var_33075_to_fp16 = const()[name = tensor("op_33075_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2797_cast_fp16 = mul(x = var_33074_cast_fp16, y = var_33075_to_fp16)[name = tensor("aw_2797_cast_fp16")]; + tensor var_33078_equation_0 = const()[name = tensor("op_33078_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33078_cast_fp16 = einsum(equation = var_33078_equation_0, values = (var_32920_cast_fp16, var_32837_cast_fp16))[name = tensor("op_33078_cast_fp16")]; + tensor var_33079_to_fp16 = const()[name = tensor("op_33079_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2799_cast_fp16 = mul(x = var_33078_cast_fp16, y = var_33079_to_fp16)[name = tensor("aw_2799_cast_fp16")]; + tensor var_33081_cast_fp16 = softmax(axis = var_30355, x = aw_2761_cast_fp16)[name = tensor("op_33081_cast_fp16")]; + tensor var_33082_cast_fp16 = softmax(axis = var_30355, x = aw_2763_cast_fp16)[name = tensor("op_33082_cast_fp16")]; + tensor var_33083_cast_fp16 = softmax(axis = var_30355, x = aw_2765_cast_fp16)[name = tensor("op_33083_cast_fp16")]; + tensor var_33084_cast_fp16 = softmax(axis = var_30355, x = aw_2767_cast_fp16)[name = tensor("op_33084_cast_fp16")]; + tensor var_33085_cast_fp16 = softmax(axis = var_30355, x = aw_2769_cast_fp16)[name = tensor("op_33085_cast_fp16")]; + tensor var_33086_cast_fp16 = softmax(axis = var_30355, x = aw_2771_cast_fp16)[name = tensor("op_33086_cast_fp16")]; + tensor var_33087_cast_fp16 = softmax(axis = var_30355, x = aw_2773_cast_fp16)[name = tensor("op_33087_cast_fp16")]; + tensor var_33088_cast_fp16 = softmax(axis = var_30355, x = aw_2775_cast_fp16)[name = tensor("op_33088_cast_fp16")]; + tensor var_33089_cast_fp16 = softmax(axis = var_30355, x = aw_2777_cast_fp16)[name = tensor("op_33089_cast_fp16")]; + tensor var_33090_cast_fp16 = softmax(axis = var_30355, x = aw_2779_cast_fp16)[name = tensor("op_33090_cast_fp16")]; + tensor var_33091_cast_fp16 = softmax(axis = var_30355, x = aw_2781_cast_fp16)[name = tensor("op_33091_cast_fp16")]; + tensor var_33092_cast_fp16 = softmax(axis = var_30355, x = aw_2783_cast_fp16)[name = tensor("op_33092_cast_fp16")]; + tensor var_33093_cast_fp16 = softmax(axis = var_30355, x = aw_2785_cast_fp16)[name = tensor("op_33093_cast_fp16")]; + tensor var_33094_cast_fp16 = softmax(axis = var_30355, x = aw_2787_cast_fp16)[name = tensor("op_33094_cast_fp16")]; + tensor var_33095_cast_fp16 = softmax(axis = var_30355, x = aw_2789_cast_fp16)[name = tensor("op_33095_cast_fp16")]; + tensor var_33096_cast_fp16 = softmax(axis = var_30355, x = aw_2791_cast_fp16)[name = tensor("op_33096_cast_fp16")]; + tensor var_33097_cast_fp16 = softmax(axis = var_30355, x = aw_2793_cast_fp16)[name = tensor("op_33097_cast_fp16")]; + tensor var_33098_cast_fp16 = softmax(axis = var_30355, x = aw_2795_cast_fp16)[name = tensor("op_33098_cast_fp16")]; + tensor var_33099_cast_fp16 = softmax(axis = var_30355, x = aw_2797_cast_fp16)[name = tensor("op_33099_cast_fp16")]; + tensor var_33100_cast_fp16 = softmax(axis = var_30355, x = aw_2799_cast_fp16)[name = tensor("op_33100_cast_fp16")]; + tensor var_33102_equation_0 = const()[name = tensor("op_33102_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_33102_cast_fp16 = einsum(equation = var_33102_equation_0, values = (var_32922_cast_fp16, var_33081_cast_fp16))[name = tensor("op_33102_cast_fp16")]; + tensor var_33104_equation_0 = const()[name = tensor("op_33104_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_33104_cast_fp16 = einsum(equation = var_33104_equation_0, values = (var_32926_cast_fp16, var_33082_cast_fp16))[name = tensor("op_33104_cast_fp16")]; + tensor var_33106_equation_0 = const()[name = tensor("op_33106_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_33106_cast_fp16 = einsum(equation = var_33106_equation_0, values = (var_32930_cast_fp16, var_33083_cast_fp16))[name = tensor("op_33106_cast_fp16")]; + tensor var_33108_equation_0 = const()[name = tensor("op_33108_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_33108_cast_fp16 = einsum(equation = var_33108_equation_0, values = (var_32934_cast_fp16, var_33084_cast_fp16))[name = tensor("op_33108_cast_fp16")]; + tensor var_33110_equation_0 = const()[name = tensor("op_33110_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_33110_cast_fp16 = einsum(equation = var_33110_equation_0, values = (var_32938_cast_fp16, var_33085_cast_fp16))[name = tensor("op_33110_cast_fp16")]; + tensor var_33112_equation_0 = const()[name = tensor("op_33112_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_33112_cast_fp16 = einsum(equation = var_33112_equation_0, values = (var_32942_cast_fp16, var_33086_cast_fp16))[name = tensor("op_33112_cast_fp16")]; + tensor var_33114_equation_0 = const()[name = tensor("op_33114_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_33114_cast_fp16 = einsum(equation = var_33114_equation_0, values = (var_32946_cast_fp16, var_33087_cast_fp16))[name = tensor("op_33114_cast_fp16")]; + tensor var_33116_equation_0 = const()[name = tensor("op_33116_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_33116_cast_fp16 = einsum(equation = var_33116_equation_0, values = (var_32950_cast_fp16, var_33088_cast_fp16))[name = tensor("op_33116_cast_fp16")]; + tensor var_33118_equation_0 = const()[name = tensor("op_33118_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_33118_cast_fp16 = einsum(equation = var_33118_equation_0, values = (var_32954_cast_fp16, var_33089_cast_fp16))[name = tensor("op_33118_cast_fp16")]; + tensor var_33120_equation_0 = const()[name = tensor("op_33120_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_33120_cast_fp16 = einsum(equation = var_33120_equation_0, values = (var_32958_cast_fp16, var_33090_cast_fp16))[name = tensor("op_33120_cast_fp16")]; + tensor var_33122_equation_0 = const()[name = tensor("op_33122_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_33122_cast_fp16 = einsum(equation = var_33122_equation_0, values = (var_32962_cast_fp16, var_33091_cast_fp16))[name = tensor("op_33122_cast_fp16")]; + tensor var_33124_equation_0 = const()[name = tensor("op_33124_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_33124_cast_fp16 = einsum(equation = var_33124_equation_0, values = (var_32966_cast_fp16, var_33092_cast_fp16))[name = tensor("op_33124_cast_fp16")]; + tensor var_33126_equation_0 = const()[name = tensor("op_33126_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_33126_cast_fp16 = einsum(equation = var_33126_equation_0, values = (var_32970_cast_fp16, var_33093_cast_fp16))[name = tensor("op_33126_cast_fp16")]; + tensor var_33128_equation_0 = const()[name = tensor("op_33128_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_33128_cast_fp16 = einsum(equation = var_33128_equation_0, values = (var_32974_cast_fp16, var_33094_cast_fp16))[name = tensor("op_33128_cast_fp16")]; + tensor var_33130_equation_0 = const()[name = tensor("op_33130_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_33130_cast_fp16 = einsum(equation = var_33130_equation_0, values = (var_32978_cast_fp16, var_33095_cast_fp16))[name = tensor("op_33130_cast_fp16")]; + tensor var_33132_equation_0 = const()[name = tensor("op_33132_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_33132_cast_fp16 = einsum(equation = var_33132_equation_0, values = (var_32982_cast_fp16, var_33096_cast_fp16))[name = tensor("op_33132_cast_fp16")]; + tensor var_33134_equation_0 = const()[name = tensor("op_33134_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_33134_cast_fp16 = einsum(equation = var_33134_equation_0, values = (var_32986_cast_fp16, var_33097_cast_fp16))[name = tensor("op_33134_cast_fp16")]; + tensor var_33136_equation_0 = const()[name = tensor("op_33136_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_33136_cast_fp16 = einsum(equation = var_33136_equation_0, values = (var_32990_cast_fp16, var_33098_cast_fp16))[name = tensor("op_33136_cast_fp16")]; + tensor var_33138_equation_0 = const()[name = tensor("op_33138_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_33138_cast_fp16 = einsum(equation = var_33138_equation_0, values = (var_32994_cast_fp16, var_33099_cast_fp16))[name = tensor("op_33138_cast_fp16")]; + tensor var_33140_equation_0 = const()[name = tensor("op_33140_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_33140_cast_fp16 = einsum(equation = var_33140_equation_0, values = (var_32998_cast_fp16, var_33100_cast_fp16))[name = tensor("op_33140_cast_fp16")]; + tensor input_457_interleave_0 = const()[name = tensor("input_457_interleave_0"), val = tensor(false)]; + tensor input_457_cast_fp16 = concat(axis = var_30355, interleave = input_457_interleave_0, values = (var_33102_cast_fp16, var_33104_cast_fp16, var_33106_cast_fp16, var_33108_cast_fp16, var_33110_cast_fp16, var_33112_cast_fp16, var_33114_cast_fp16, var_33116_cast_fp16, var_33118_cast_fp16, var_33120_cast_fp16, var_33122_cast_fp16, var_33124_cast_fp16, var_33126_cast_fp16, var_33128_cast_fp16, var_33130_cast_fp16, var_33132_cast_fp16, var_33134_cast_fp16, var_33136_cast_fp16, var_33138_cast_fp16, var_33140_cast_fp16))[name = tensor("input_457_cast_fp16")]; + tensor var_33146 = const()[name = tensor("op_33146"), val = tensor([1, 1])]; + tensor var_33148 = const()[name = tensor("op_33148"), val = tensor([1, 1])]; + tensor var_33150_pad_type_0 = const()[name = tensor("op_33150_pad_type_0"), val = tensor("custom")]; + tensor var_33150_pad_0 = const()[name = tensor("op_33150_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_2_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_2_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2754330624)))]; + tensor up_blocks_0_attentions_0_transformer_blocks_2_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_2_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2757607488)))]; + tensor var_33150_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_2_attn2_to_out_0_bias_to_fp16, dilations = var_33148, groups = var_30355, pad = var_33150_pad_0, pad_type = var_33150_pad_type_0, strides = var_33146, weight = up_blocks_0_attentions_0_transformer_blocks_2_attn2_to_out_0_weight_to_fp16, x = input_457_cast_fp16)[name = tensor("op_33150_cast_fp16")]; + tensor inputs_221_cast_fp16 = add(x = var_33150_cast_fp16, y = inputs_219_cast_fp16)[name = tensor("inputs_221_cast_fp16")]; + tensor input_459_axes_0 = const()[name = tensor("input_459_axes_0"), val = tensor([1])]; + tensor input_459_gamma_0_to_fp16 = const()[name = tensor("input_459_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2757610112)))]; + tensor input_459_beta_0_to_fp16 = const()[name = tensor("input_459_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2757612736)))]; + tensor var_33160_to_fp16 = const()[name = tensor("op_33160_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_459_cast_fp16 = layer_norm(axes = input_459_axes_0, beta = input_459_beta_0_to_fp16, epsilon = var_33160_to_fp16, gamma = input_459_gamma_0_to_fp16, x = inputs_221_cast_fp16)[name = tensor("input_459_cast_fp16")]; + tensor var_33176 = const()[name = tensor("op_33176"), val = tensor([1, 1])]; + tensor var_33178 = const()[name = tensor("op_33178"), val = tensor([1, 1])]; + tensor var_33180_pad_type_0 = const()[name = tensor("op_33180_pad_type_0"), val = tensor("custom")]; + tensor var_33180_pad_0 = const()[name = tensor("op_33180_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_2_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_2_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2757615360)))]; + tensor up_blocks_0_attentions_0_transformer_blocks_2_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_2_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2783829824)))]; + tensor var_33180_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_2_ff_net_0_proj_bias_to_fp16, dilations = var_33178, groups = var_30355, pad = var_33180_pad_0, pad_type = var_33180_pad_type_0, strides = var_33176, weight = up_blocks_0_attentions_0_transformer_blocks_2_ff_net_0_proj_weight_to_fp16, x = input_459_cast_fp16)[name = tensor("op_33180_cast_fp16")]; + tensor var_33181_split_sizes_0 = const()[name = tensor("op_33181_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_33181_axis_0 = const()[name = tensor("op_33181_axis_0"), val = tensor(1)]; + tensor var_33181_cast_fp16_0, tensor var_33181_cast_fp16_1 = split(axis = var_33181_axis_0, split_sizes = var_33181_split_sizes_0, x = var_33180_cast_fp16)[name = tensor("op_33181_cast_fp16")]; + tensor var_33183_mode_0 = const()[name = tensor("op_33183_mode_0"), val = tensor("EXACT")]; + tensor var_33183_cast_fp16 = gelu(mode = var_33183_mode_0, x = var_33181_cast_fp16_1)[name = tensor("op_33183_cast_fp16")]; + tensor input_461_cast_fp16 = mul(x = var_33181_cast_fp16_0, y = var_33183_cast_fp16)[name = tensor("input_461_cast_fp16")]; + tensor var_33187 = const()[name = tensor("op_33187"), val = tensor([1, 1])]; + tensor var_33189 = const()[name = tensor("op_33189"), val = tensor([1, 1])]; + tensor var_33191_pad_type_0 = const()[name = tensor("op_33191_pad_type_0"), val = tensor("custom")]; + tensor var_33191_pad_0 = const()[name = tensor("op_33191_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_2_ff_net_2_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_2_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2783850368)))]; + tensor up_blocks_0_attentions_0_transformer_blocks_2_ff_net_2_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_2_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2796957632)))]; + tensor var_33191_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_2_ff_net_2_bias_to_fp16, dilations = var_33189, groups = var_30355, pad = var_33191_pad_0, pad_type = var_33191_pad_type_0, strides = var_33187, weight = up_blocks_0_attentions_0_transformer_blocks_2_ff_net_2_weight_to_fp16, x = input_461_cast_fp16)[name = tensor("op_33191_cast_fp16")]; + tensor inputs_223_cast_fp16 = add(x = var_33191_cast_fp16, y = inputs_221_cast_fp16)[name = tensor("inputs_223_cast_fp16")]; + tensor hidden_states_305_axes_0 = const()[name = tensor("hidden_states_305_axes_0"), val = tensor([1])]; + tensor hidden_states_305_gamma_0_to_fp16 = const()[name = tensor("hidden_states_305_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2796960256)))]; + tensor hidden_states_305_beta_0_to_fp16 = const()[name = tensor("hidden_states_305_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2796962880)))]; + tensor var_33207_to_fp16 = const()[name = tensor("op_33207_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_305_cast_fp16 = layer_norm(axes = hidden_states_305_axes_0, beta = hidden_states_305_beta_0_to_fp16, epsilon = var_33207_to_fp16, gamma = hidden_states_305_gamma_0_to_fp16, x = inputs_223_cast_fp16)[name = tensor("hidden_states_305_cast_fp16")]; + tensor var_33222 = const()[name = tensor("op_33222"), val = tensor([1, 1])]; + tensor var_33224 = const()[name = tensor("op_33224"), val = tensor([1, 1])]; + tensor q_149_pad_type_0 = const()[name = tensor("q_149_pad_type_0"), val = tensor("custom")]; + tensor q_149_pad_0 = const()[name = tensor("q_149_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_3_attn1_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_3_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2796965504)))]; + tensor q_149_cast_fp16 = conv(dilations = var_33224, groups = var_30355, pad = q_149_pad_0, pad_type = q_149_pad_type_0, strides = var_33222, weight = up_blocks_0_attentions_0_transformer_blocks_3_attn1_to_q_weight_to_fp16, x = hidden_states_305_cast_fp16)[name = tensor("q_149_cast_fp16")]; + tensor var_33228 = const()[name = tensor("op_33228"), val = tensor([1, 1])]; + tensor var_33230 = const()[name = tensor("op_33230"), val = tensor([1, 1])]; + tensor k_297_pad_type_0 = const()[name = tensor("k_297_pad_type_0"), val = tensor("custom")]; + tensor k_297_pad_0 = const()[name = tensor("k_297_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_3_attn1_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_3_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2800242368)))]; + tensor k_297_cast_fp16 = conv(dilations = var_33230, groups = var_30355, pad = k_297_pad_0, pad_type = k_297_pad_type_0, strides = var_33228, weight = up_blocks_0_attentions_0_transformer_blocks_3_attn1_to_k_weight_to_fp16, x = hidden_states_305_cast_fp16)[name = tensor("k_297_cast_fp16")]; + tensor var_33234 = const()[name = tensor("op_33234"), val = tensor([1, 1])]; + tensor var_33236 = const()[name = tensor("op_33236"), val = tensor([1, 1])]; + tensor v_149_pad_type_0 = const()[name = tensor("v_149_pad_type_0"), val = tensor("custom")]; + tensor v_149_pad_0 = const()[name = tensor("v_149_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_3_attn1_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_3_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2803519232)))]; + tensor v_149_cast_fp16 = conv(dilations = var_33236, groups = var_30355, pad = v_149_pad_0, pad_type = v_149_pad_type_0, strides = var_33234, weight = up_blocks_0_attentions_0_transformer_blocks_3_attn1_to_v_weight_to_fp16, x = hidden_states_305_cast_fp16)[name = tensor("v_149_cast_fp16")]; + tensor var_33240_begin_0 = const()[name = tensor("op_33240_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_33240_end_0 = const()[name = tensor("op_33240_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_33240_end_mask_0 = const()[name = tensor("op_33240_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33240_cast_fp16 = slice_by_index(begin = var_33240_begin_0, end = var_33240_end_0, end_mask = var_33240_end_mask_0, x = q_149_cast_fp16)[name = tensor("op_33240_cast_fp16")]; + tensor var_33244_begin_0 = const()[name = tensor("op_33244_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_33244_end_0 = const()[name = tensor("op_33244_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_33244_end_mask_0 = const()[name = tensor("op_33244_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33244_cast_fp16 = slice_by_index(begin = var_33244_begin_0, end = var_33244_end_0, end_mask = var_33244_end_mask_0, x = q_149_cast_fp16)[name = tensor("op_33244_cast_fp16")]; + tensor var_33248_begin_0 = const()[name = tensor("op_33248_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_33248_end_0 = const()[name = tensor("op_33248_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_33248_end_mask_0 = const()[name = tensor("op_33248_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33248_cast_fp16 = slice_by_index(begin = var_33248_begin_0, end = var_33248_end_0, end_mask = var_33248_end_mask_0, x = q_149_cast_fp16)[name = tensor("op_33248_cast_fp16")]; + tensor var_33252_begin_0 = const()[name = tensor("op_33252_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_33252_end_0 = const()[name = tensor("op_33252_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_33252_end_mask_0 = const()[name = tensor("op_33252_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33252_cast_fp16 = slice_by_index(begin = var_33252_begin_0, end = var_33252_end_0, end_mask = var_33252_end_mask_0, x = q_149_cast_fp16)[name = tensor("op_33252_cast_fp16")]; + tensor var_33256_begin_0 = const()[name = tensor("op_33256_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_33256_end_0 = const()[name = tensor("op_33256_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_33256_end_mask_0 = const()[name = tensor("op_33256_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33256_cast_fp16 = slice_by_index(begin = var_33256_begin_0, end = var_33256_end_0, end_mask = var_33256_end_mask_0, x = q_149_cast_fp16)[name = tensor("op_33256_cast_fp16")]; + tensor var_33260_begin_0 = const()[name = tensor("op_33260_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_33260_end_0 = const()[name = tensor("op_33260_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_33260_end_mask_0 = const()[name = tensor("op_33260_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33260_cast_fp16 = slice_by_index(begin = var_33260_begin_0, end = var_33260_end_0, end_mask = var_33260_end_mask_0, x = q_149_cast_fp16)[name = tensor("op_33260_cast_fp16")]; + tensor var_33264_begin_0 = const()[name = tensor("op_33264_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_33264_end_0 = const()[name = tensor("op_33264_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_33264_end_mask_0 = const()[name = tensor("op_33264_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33264_cast_fp16 = slice_by_index(begin = var_33264_begin_0, end = var_33264_end_0, end_mask = var_33264_end_mask_0, x = q_149_cast_fp16)[name = tensor("op_33264_cast_fp16")]; + tensor var_33268_begin_0 = const()[name = tensor("op_33268_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_33268_end_0 = const()[name = tensor("op_33268_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_33268_end_mask_0 = const()[name = tensor("op_33268_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33268_cast_fp16 = slice_by_index(begin = var_33268_begin_0, end = var_33268_end_0, end_mask = var_33268_end_mask_0, x = q_149_cast_fp16)[name = tensor("op_33268_cast_fp16")]; + tensor var_33272_begin_0 = const()[name = tensor("op_33272_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_33272_end_0 = const()[name = tensor("op_33272_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_33272_end_mask_0 = const()[name = tensor("op_33272_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33272_cast_fp16 = slice_by_index(begin = var_33272_begin_0, end = var_33272_end_0, end_mask = var_33272_end_mask_0, x = q_149_cast_fp16)[name = tensor("op_33272_cast_fp16")]; + tensor var_33276_begin_0 = const()[name = tensor("op_33276_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_33276_end_0 = const()[name = tensor("op_33276_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_33276_end_mask_0 = const()[name = tensor("op_33276_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33276_cast_fp16 = slice_by_index(begin = var_33276_begin_0, end = var_33276_end_0, end_mask = var_33276_end_mask_0, x = q_149_cast_fp16)[name = tensor("op_33276_cast_fp16")]; + tensor var_33280_begin_0 = const()[name = tensor("op_33280_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_33280_end_0 = const()[name = tensor("op_33280_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_33280_end_mask_0 = const()[name = tensor("op_33280_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33280_cast_fp16 = slice_by_index(begin = var_33280_begin_0, end = var_33280_end_0, end_mask = var_33280_end_mask_0, x = q_149_cast_fp16)[name = tensor("op_33280_cast_fp16")]; + tensor var_33284_begin_0 = const()[name = tensor("op_33284_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_33284_end_0 = const()[name = tensor("op_33284_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_33284_end_mask_0 = const()[name = tensor("op_33284_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33284_cast_fp16 = slice_by_index(begin = var_33284_begin_0, end = var_33284_end_0, end_mask = var_33284_end_mask_0, x = q_149_cast_fp16)[name = tensor("op_33284_cast_fp16")]; + tensor var_33288_begin_0 = const()[name = tensor("op_33288_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_33288_end_0 = const()[name = tensor("op_33288_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_33288_end_mask_0 = const()[name = tensor("op_33288_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33288_cast_fp16 = slice_by_index(begin = var_33288_begin_0, end = var_33288_end_0, end_mask = var_33288_end_mask_0, x = q_149_cast_fp16)[name = tensor("op_33288_cast_fp16")]; + tensor var_33292_begin_0 = const()[name = tensor("op_33292_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_33292_end_0 = const()[name = tensor("op_33292_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_33292_end_mask_0 = const()[name = tensor("op_33292_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33292_cast_fp16 = slice_by_index(begin = var_33292_begin_0, end = var_33292_end_0, end_mask = var_33292_end_mask_0, x = q_149_cast_fp16)[name = tensor("op_33292_cast_fp16")]; + tensor var_33296_begin_0 = const()[name = tensor("op_33296_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_33296_end_0 = const()[name = tensor("op_33296_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_33296_end_mask_0 = const()[name = tensor("op_33296_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33296_cast_fp16 = slice_by_index(begin = var_33296_begin_0, end = var_33296_end_0, end_mask = var_33296_end_mask_0, x = q_149_cast_fp16)[name = tensor("op_33296_cast_fp16")]; + tensor var_33300_begin_0 = const()[name = tensor("op_33300_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_33300_end_0 = const()[name = tensor("op_33300_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_33300_end_mask_0 = const()[name = tensor("op_33300_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33300_cast_fp16 = slice_by_index(begin = var_33300_begin_0, end = var_33300_end_0, end_mask = var_33300_end_mask_0, x = q_149_cast_fp16)[name = tensor("op_33300_cast_fp16")]; + tensor var_33304_begin_0 = const()[name = tensor("op_33304_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_33304_end_0 = const()[name = tensor("op_33304_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_33304_end_mask_0 = const()[name = tensor("op_33304_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33304_cast_fp16 = slice_by_index(begin = var_33304_begin_0, end = var_33304_end_0, end_mask = var_33304_end_mask_0, x = q_149_cast_fp16)[name = tensor("op_33304_cast_fp16")]; + tensor var_33308_begin_0 = const()[name = tensor("op_33308_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_33308_end_0 = const()[name = tensor("op_33308_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_33308_end_mask_0 = const()[name = tensor("op_33308_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33308_cast_fp16 = slice_by_index(begin = var_33308_begin_0, end = var_33308_end_0, end_mask = var_33308_end_mask_0, x = q_149_cast_fp16)[name = tensor("op_33308_cast_fp16")]; + tensor var_33312_begin_0 = const()[name = tensor("op_33312_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_33312_end_0 = const()[name = tensor("op_33312_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_33312_end_mask_0 = const()[name = tensor("op_33312_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33312_cast_fp16 = slice_by_index(begin = var_33312_begin_0, end = var_33312_end_0, end_mask = var_33312_end_mask_0, x = q_149_cast_fp16)[name = tensor("op_33312_cast_fp16")]; + tensor var_33316_begin_0 = const()[name = tensor("op_33316_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_33316_end_0 = const()[name = tensor("op_33316_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_33316_end_mask_0 = const()[name = tensor("op_33316_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33316_cast_fp16 = slice_by_index(begin = var_33316_begin_0, end = var_33316_end_0, end_mask = var_33316_end_mask_0, x = q_149_cast_fp16)[name = tensor("op_33316_cast_fp16")]; + tensor k_299_perm_0 = const()[name = tensor("k_299_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_33323_begin_0 = const()[name = tensor("op_33323_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_33323_end_0 = const()[name = tensor("op_33323_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_33323_end_mask_0 = const()[name = tensor("op_33323_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_299_cast_fp16 = transpose(perm = k_299_perm_0, x = k_297_cast_fp16)[name = tensor("transpose_65")]; + tensor var_33323_cast_fp16 = slice_by_index(begin = var_33323_begin_0, end = var_33323_end_0, end_mask = var_33323_end_mask_0, x = k_299_cast_fp16)[name = tensor("op_33323_cast_fp16")]; + tensor var_33327_begin_0 = const()[name = tensor("op_33327_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_33327_end_0 = const()[name = tensor("op_33327_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_33327_end_mask_0 = const()[name = tensor("op_33327_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33327_cast_fp16 = slice_by_index(begin = var_33327_begin_0, end = var_33327_end_0, end_mask = var_33327_end_mask_0, x = k_299_cast_fp16)[name = tensor("op_33327_cast_fp16")]; + tensor var_33331_begin_0 = const()[name = tensor("op_33331_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_33331_end_0 = const()[name = tensor("op_33331_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_33331_end_mask_0 = const()[name = tensor("op_33331_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33331_cast_fp16 = slice_by_index(begin = var_33331_begin_0, end = var_33331_end_0, end_mask = var_33331_end_mask_0, x = k_299_cast_fp16)[name = tensor("op_33331_cast_fp16")]; + tensor var_33335_begin_0 = const()[name = tensor("op_33335_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_33335_end_0 = const()[name = tensor("op_33335_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_33335_end_mask_0 = const()[name = tensor("op_33335_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33335_cast_fp16 = slice_by_index(begin = var_33335_begin_0, end = var_33335_end_0, end_mask = var_33335_end_mask_0, x = k_299_cast_fp16)[name = tensor("op_33335_cast_fp16")]; + tensor var_33339_begin_0 = const()[name = tensor("op_33339_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_33339_end_0 = const()[name = tensor("op_33339_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_33339_end_mask_0 = const()[name = tensor("op_33339_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33339_cast_fp16 = slice_by_index(begin = var_33339_begin_0, end = var_33339_end_0, end_mask = var_33339_end_mask_0, x = k_299_cast_fp16)[name = tensor("op_33339_cast_fp16")]; + tensor var_33343_begin_0 = const()[name = tensor("op_33343_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_33343_end_0 = const()[name = tensor("op_33343_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_33343_end_mask_0 = const()[name = tensor("op_33343_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33343_cast_fp16 = slice_by_index(begin = var_33343_begin_0, end = var_33343_end_0, end_mask = var_33343_end_mask_0, x = k_299_cast_fp16)[name = tensor("op_33343_cast_fp16")]; + tensor var_33347_begin_0 = const()[name = tensor("op_33347_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_33347_end_0 = const()[name = tensor("op_33347_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_33347_end_mask_0 = const()[name = tensor("op_33347_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33347_cast_fp16 = slice_by_index(begin = var_33347_begin_0, end = var_33347_end_0, end_mask = var_33347_end_mask_0, x = k_299_cast_fp16)[name = tensor("op_33347_cast_fp16")]; + tensor var_33351_begin_0 = const()[name = tensor("op_33351_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_33351_end_0 = const()[name = tensor("op_33351_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_33351_end_mask_0 = const()[name = tensor("op_33351_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33351_cast_fp16 = slice_by_index(begin = var_33351_begin_0, end = var_33351_end_0, end_mask = var_33351_end_mask_0, x = k_299_cast_fp16)[name = tensor("op_33351_cast_fp16")]; + tensor var_33355_begin_0 = const()[name = tensor("op_33355_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_33355_end_0 = const()[name = tensor("op_33355_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_33355_end_mask_0 = const()[name = tensor("op_33355_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33355_cast_fp16 = slice_by_index(begin = var_33355_begin_0, end = var_33355_end_0, end_mask = var_33355_end_mask_0, x = k_299_cast_fp16)[name = tensor("op_33355_cast_fp16")]; + tensor var_33359_begin_0 = const()[name = tensor("op_33359_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_33359_end_0 = const()[name = tensor("op_33359_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_33359_end_mask_0 = const()[name = tensor("op_33359_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33359_cast_fp16 = slice_by_index(begin = var_33359_begin_0, end = var_33359_end_0, end_mask = var_33359_end_mask_0, x = k_299_cast_fp16)[name = tensor("op_33359_cast_fp16")]; + tensor var_33363_begin_0 = const()[name = tensor("op_33363_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_33363_end_0 = const()[name = tensor("op_33363_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_33363_end_mask_0 = const()[name = tensor("op_33363_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33363_cast_fp16 = slice_by_index(begin = var_33363_begin_0, end = var_33363_end_0, end_mask = var_33363_end_mask_0, x = k_299_cast_fp16)[name = tensor("op_33363_cast_fp16")]; + tensor var_33367_begin_0 = const()[name = tensor("op_33367_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_33367_end_0 = const()[name = tensor("op_33367_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_33367_end_mask_0 = const()[name = tensor("op_33367_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33367_cast_fp16 = slice_by_index(begin = var_33367_begin_0, end = var_33367_end_0, end_mask = var_33367_end_mask_0, x = k_299_cast_fp16)[name = tensor("op_33367_cast_fp16")]; + tensor var_33371_begin_0 = const()[name = tensor("op_33371_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_33371_end_0 = const()[name = tensor("op_33371_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_33371_end_mask_0 = const()[name = tensor("op_33371_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33371_cast_fp16 = slice_by_index(begin = var_33371_begin_0, end = var_33371_end_0, end_mask = var_33371_end_mask_0, x = k_299_cast_fp16)[name = tensor("op_33371_cast_fp16")]; + tensor var_33375_begin_0 = const()[name = tensor("op_33375_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_33375_end_0 = const()[name = tensor("op_33375_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_33375_end_mask_0 = const()[name = tensor("op_33375_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33375_cast_fp16 = slice_by_index(begin = var_33375_begin_0, end = var_33375_end_0, end_mask = var_33375_end_mask_0, x = k_299_cast_fp16)[name = tensor("op_33375_cast_fp16")]; + tensor var_33379_begin_0 = const()[name = tensor("op_33379_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_33379_end_0 = const()[name = tensor("op_33379_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_33379_end_mask_0 = const()[name = tensor("op_33379_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33379_cast_fp16 = slice_by_index(begin = var_33379_begin_0, end = var_33379_end_0, end_mask = var_33379_end_mask_0, x = k_299_cast_fp16)[name = tensor("op_33379_cast_fp16")]; + tensor var_33383_begin_0 = const()[name = tensor("op_33383_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_33383_end_0 = const()[name = tensor("op_33383_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_33383_end_mask_0 = const()[name = tensor("op_33383_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33383_cast_fp16 = slice_by_index(begin = var_33383_begin_0, end = var_33383_end_0, end_mask = var_33383_end_mask_0, x = k_299_cast_fp16)[name = tensor("op_33383_cast_fp16")]; + tensor var_33387_begin_0 = const()[name = tensor("op_33387_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_33387_end_0 = const()[name = tensor("op_33387_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_33387_end_mask_0 = const()[name = tensor("op_33387_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33387_cast_fp16 = slice_by_index(begin = var_33387_begin_0, end = var_33387_end_0, end_mask = var_33387_end_mask_0, x = k_299_cast_fp16)[name = tensor("op_33387_cast_fp16")]; + tensor var_33391_begin_0 = const()[name = tensor("op_33391_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_33391_end_0 = const()[name = tensor("op_33391_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_33391_end_mask_0 = const()[name = tensor("op_33391_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33391_cast_fp16 = slice_by_index(begin = var_33391_begin_0, end = var_33391_end_0, end_mask = var_33391_end_mask_0, x = k_299_cast_fp16)[name = tensor("op_33391_cast_fp16")]; + tensor var_33395_begin_0 = const()[name = tensor("op_33395_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_33395_end_0 = const()[name = tensor("op_33395_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_33395_end_mask_0 = const()[name = tensor("op_33395_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33395_cast_fp16 = slice_by_index(begin = var_33395_begin_0, end = var_33395_end_0, end_mask = var_33395_end_mask_0, x = k_299_cast_fp16)[name = tensor("op_33395_cast_fp16")]; + tensor var_33399_begin_0 = const()[name = tensor("op_33399_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_33399_end_0 = const()[name = tensor("op_33399_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_33399_end_mask_0 = const()[name = tensor("op_33399_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33399_cast_fp16 = slice_by_index(begin = var_33399_begin_0, end = var_33399_end_0, end_mask = var_33399_end_mask_0, x = k_299_cast_fp16)[name = tensor("op_33399_cast_fp16")]; + tensor var_33401_begin_0 = const()[name = tensor("op_33401_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_33401_end_0 = const()[name = tensor("op_33401_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_33401_end_mask_0 = const()[name = tensor("op_33401_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33401_cast_fp16 = slice_by_index(begin = var_33401_begin_0, end = var_33401_end_0, end_mask = var_33401_end_mask_0, x = v_149_cast_fp16)[name = tensor("op_33401_cast_fp16")]; + tensor var_33405_begin_0 = const()[name = tensor("op_33405_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_33405_end_0 = const()[name = tensor("op_33405_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_33405_end_mask_0 = const()[name = tensor("op_33405_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33405_cast_fp16 = slice_by_index(begin = var_33405_begin_0, end = var_33405_end_0, end_mask = var_33405_end_mask_0, x = v_149_cast_fp16)[name = tensor("op_33405_cast_fp16")]; + tensor var_33409_begin_0 = const()[name = tensor("op_33409_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_33409_end_0 = const()[name = tensor("op_33409_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_33409_end_mask_0 = const()[name = tensor("op_33409_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33409_cast_fp16 = slice_by_index(begin = var_33409_begin_0, end = var_33409_end_0, end_mask = var_33409_end_mask_0, x = v_149_cast_fp16)[name = tensor("op_33409_cast_fp16")]; + tensor var_33413_begin_0 = const()[name = tensor("op_33413_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_33413_end_0 = const()[name = tensor("op_33413_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_33413_end_mask_0 = const()[name = tensor("op_33413_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33413_cast_fp16 = slice_by_index(begin = var_33413_begin_0, end = var_33413_end_0, end_mask = var_33413_end_mask_0, x = v_149_cast_fp16)[name = tensor("op_33413_cast_fp16")]; + tensor var_33417_begin_0 = const()[name = tensor("op_33417_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_33417_end_0 = const()[name = tensor("op_33417_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_33417_end_mask_0 = const()[name = tensor("op_33417_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33417_cast_fp16 = slice_by_index(begin = var_33417_begin_0, end = var_33417_end_0, end_mask = var_33417_end_mask_0, x = v_149_cast_fp16)[name = tensor("op_33417_cast_fp16")]; + tensor var_33421_begin_0 = const()[name = tensor("op_33421_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_33421_end_0 = const()[name = tensor("op_33421_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_33421_end_mask_0 = const()[name = tensor("op_33421_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33421_cast_fp16 = slice_by_index(begin = var_33421_begin_0, end = var_33421_end_0, end_mask = var_33421_end_mask_0, x = v_149_cast_fp16)[name = tensor("op_33421_cast_fp16")]; + tensor var_33425_begin_0 = const()[name = tensor("op_33425_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_33425_end_0 = const()[name = tensor("op_33425_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_33425_end_mask_0 = const()[name = tensor("op_33425_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33425_cast_fp16 = slice_by_index(begin = var_33425_begin_0, end = var_33425_end_0, end_mask = var_33425_end_mask_0, x = v_149_cast_fp16)[name = tensor("op_33425_cast_fp16")]; + tensor var_33429_begin_0 = const()[name = tensor("op_33429_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_33429_end_0 = const()[name = tensor("op_33429_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_33429_end_mask_0 = const()[name = tensor("op_33429_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33429_cast_fp16 = slice_by_index(begin = var_33429_begin_0, end = var_33429_end_0, end_mask = var_33429_end_mask_0, x = v_149_cast_fp16)[name = tensor("op_33429_cast_fp16")]; + tensor var_33433_begin_0 = const()[name = tensor("op_33433_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_33433_end_0 = const()[name = tensor("op_33433_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_33433_end_mask_0 = const()[name = tensor("op_33433_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33433_cast_fp16 = slice_by_index(begin = var_33433_begin_0, end = var_33433_end_0, end_mask = var_33433_end_mask_0, x = v_149_cast_fp16)[name = tensor("op_33433_cast_fp16")]; + tensor var_33437_begin_0 = const()[name = tensor("op_33437_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_33437_end_0 = const()[name = tensor("op_33437_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_33437_end_mask_0 = const()[name = tensor("op_33437_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33437_cast_fp16 = slice_by_index(begin = var_33437_begin_0, end = var_33437_end_0, end_mask = var_33437_end_mask_0, x = v_149_cast_fp16)[name = tensor("op_33437_cast_fp16")]; + tensor var_33441_begin_0 = const()[name = tensor("op_33441_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_33441_end_0 = const()[name = tensor("op_33441_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_33441_end_mask_0 = const()[name = tensor("op_33441_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33441_cast_fp16 = slice_by_index(begin = var_33441_begin_0, end = var_33441_end_0, end_mask = var_33441_end_mask_0, x = v_149_cast_fp16)[name = tensor("op_33441_cast_fp16")]; + tensor var_33445_begin_0 = const()[name = tensor("op_33445_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_33445_end_0 = const()[name = tensor("op_33445_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_33445_end_mask_0 = const()[name = tensor("op_33445_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33445_cast_fp16 = slice_by_index(begin = var_33445_begin_0, end = var_33445_end_0, end_mask = var_33445_end_mask_0, x = v_149_cast_fp16)[name = tensor("op_33445_cast_fp16")]; + tensor var_33449_begin_0 = const()[name = tensor("op_33449_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_33449_end_0 = const()[name = tensor("op_33449_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_33449_end_mask_0 = const()[name = tensor("op_33449_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33449_cast_fp16 = slice_by_index(begin = var_33449_begin_0, end = var_33449_end_0, end_mask = var_33449_end_mask_0, x = v_149_cast_fp16)[name = tensor("op_33449_cast_fp16")]; + tensor var_33453_begin_0 = const()[name = tensor("op_33453_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_33453_end_0 = const()[name = tensor("op_33453_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_33453_end_mask_0 = const()[name = tensor("op_33453_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33453_cast_fp16 = slice_by_index(begin = var_33453_begin_0, end = var_33453_end_0, end_mask = var_33453_end_mask_0, x = v_149_cast_fp16)[name = tensor("op_33453_cast_fp16")]; + tensor var_33457_begin_0 = const()[name = tensor("op_33457_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_33457_end_0 = const()[name = tensor("op_33457_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_33457_end_mask_0 = const()[name = tensor("op_33457_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33457_cast_fp16 = slice_by_index(begin = var_33457_begin_0, end = var_33457_end_0, end_mask = var_33457_end_mask_0, x = v_149_cast_fp16)[name = tensor("op_33457_cast_fp16")]; + tensor var_33461_begin_0 = const()[name = tensor("op_33461_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_33461_end_0 = const()[name = tensor("op_33461_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_33461_end_mask_0 = const()[name = tensor("op_33461_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33461_cast_fp16 = slice_by_index(begin = var_33461_begin_0, end = var_33461_end_0, end_mask = var_33461_end_mask_0, x = v_149_cast_fp16)[name = tensor("op_33461_cast_fp16")]; + tensor var_33465_begin_0 = const()[name = tensor("op_33465_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_33465_end_0 = const()[name = tensor("op_33465_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_33465_end_mask_0 = const()[name = tensor("op_33465_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33465_cast_fp16 = slice_by_index(begin = var_33465_begin_0, end = var_33465_end_0, end_mask = var_33465_end_mask_0, x = v_149_cast_fp16)[name = tensor("op_33465_cast_fp16")]; + tensor var_33469_begin_0 = const()[name = tensor("op_33469_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_33469_end_0 = const()[name = tensor("op_33469_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_33469_end_mask_0 = const()[name = tensor("op_33469_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33469_cast_fp16 = slice_by_index(begin = var_33469_begin_0, end = var_33469_end_0, end_mask = var_33469_end_mask_0, x = v_149_cast_fp16)[name = tensor("op_33469_cast_fp16")]; + tensor var_33473_begin_0 = const()[name = tensor("op_33473_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_33473_end_0 = const()[name = tensor("op_33473_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_33473_end_mask_0 = const()[name = tensor("op_33473_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33473_cast_fp16 = slice_by_index(begin = var_33473_begin_0, end = var_33473_end_0, end_mask = var_33473_end_mask_0, x = v_149_cast_fp16)[name = tensor("op_33473_cast_fp16")]; + tensor var_33477_begin_0 = const()[name = tensor("op_33477_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_33477_end_0 = const()[name = tensor("op_33477_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_33477_end_mask_0 = const()[name = tensor("op_33477_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33477_cast_fp16 = slice_by_index(begin = var_33477_begin_0, end = var_33477_end_0, end_mask = var_33477_end_mask_0, x = v_149_cast_fp16)[name = tensor("op_33477_cast_fp16")]; + tensor var_33481_equation_0 = const()[name = tensor("op_33481_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33481_cast_fp16 = einsum(equation = var_33481_equation_0, values = (var_33323_cast_fp16, var_33240_cast_fp16))[name = tensor("op_33481_cast_fp16")]; + tensor var_33482_to_fp16 = const()[name = tensor("op_33482_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2801_cast_fp16 = mul(x = var_33481_cast_fp16, y = var_33482_to_fp16)[name = tensor("aw_2801_cast_fp16")]; + tensor var_33485_equation_0 = const()[name = tensor("op_33485_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33485_cast_fp16 = einsum(equation = var_33485_equation_0, values = (var_33327_cast_fp16, var_33244_cast_fp16))[name = tensor("op_33485_cast_fp16")]; + tensor var_33486_to_fp16 = const()[name = tensor("op_33486_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2803_cast_fp16 = mul(x = var_33485_cast_fp16, y = var_33486_to_fp16)[name = tensor("aw_2803_cast_fp16")]; + tensor var_33489_equation_0 = const()[name = tensor("op_33489_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33489_cast_fp16 = einsum(equation = var_33489_equation_0, values = (var_33331_cast_fp16, var_33248_cast_fp16))[name = tensor("op_33489_cast_fp16")]; + tensor var_33490_to_fp16 = const()[name = tensor("op_33490_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2805_cast_fp16 = mul(x = var_33489_cast_fp16, y = var_33490_to_fp16)[name = tensor("aw_2805_cast_fp16")]; + tensor var_33493_equation_0 = const()[name = tensor("op_33493_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33493_cast_fp16 = einsum(equation = var_33493_equation_0, values = (var_33335_cast_fp16, var_33252_cast_fp16))[name = tensor("op_33493_cast_fp16")]; + tensor var_33494_to_fp16 = const()[name = tensor("op_33494_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2807_cast_fp16 = mul(x = var_33493_cast_fp16, y = var_33494_to_fp16)[name = tensor("aw_2807_cast_fp16")]; + tensor var_33497_equation_0 = const()[name = tensor("op_33497_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33497_cast_fp16 = einsum(equation = var_33497_equation_0, values = (var_33339_cast_fp16, var_33256_cast_fp16))[name = tensor("op_33497_cast_fp16")]; + tensor var_33498_to_fp16 = const()[name = tensor("op_33498_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2809_cast_fp16 = mul(x = var_33497_cast_fp16, y = var_33498_to_fp16)[name = tensor("aw_2809_cast_fp16")]; + tensor var_33501_equation_0 = const()[name = tensor("op_33501_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33501_cast_fp16 = einsum(equation = var_33501_equation_0, values = (var_33343_cast_fp16, var_33260_cast_fp16))[name = tensor("op_33501_cast_fp16")]; + tensor var_33502_to_fp16 = const()[name = tensor("op_33502_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2811_cast_fp16 = mul(x = var_33501_cast_fp16, y = var_33502_to_fp16)[name = tensor("aw_2811_cast_fp16")]; + tensor var_33505_equation_0 = const()[name = tensor("op_33505_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33505_cast_fp16 = einsum(equation = var_33505_equation_0, values = (var_33347_cast_fp16, var_33264_cast_fp16))[name = tensor("op_33505_cast_fp16")]; + tensor var_33506_to_fp16 = const()[name = tensor("op_33506_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2813_cast_fp16 = mul(x = var_33505_cast_fp16, y = var_33506_to_fp16)[name = tensor("aw_2813_cast_fp16")]; + tensor var_33509_equation_0 = const()[name = tensor("op_33509_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33509_cast_fp16 = einsum(equation = var_33509_equation_0, values = (var_33351_cast_fp16, var_33268_cast_fp16))[name = tensor("op_33509_cast_fp16")]; + tensor var_33510_to_fp16 = const()[name = tensor("op_33510_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2815_cast_fp16 = mul(x = var_33509_cast_fp16, y = var_33510_to_fp16)[name = tensor("aw_2815_cast_fp16")]; + tensor var_33513_equation_0 = const()[name = tensor("op_33513_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33513_cast_fp16 = einsum(equation = var_33513_equation_0, values = (var_33355_cast_fp16, var_33272_cast_fp16))[name = tensor("op_33513_cast_fp16")]; + tensor var_33514_to_fp16 = const()[name = tensor("op_33514_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2817_cast_fp16 = mul(x = var_33513_cast_fp16, y = var_33514_to_fp16)[name = tensor("aw_2817_cast_fp16")]; + tensor var_33517_equation_0 = const()[name = tensor("op_33517_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33517_cast_fp16 = einsum(equation = var_33517_equation_0, values = (var_33359_cast_fp16, var_33276_cast_fp16))[name = tensor("op_33517_cast_fp16")]; + tensor var_33518_to_fp16 = const()[name = tensor("op_33518_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2819_cast_fp16 = mul(x = var_33517_cast_fp16, y = var_33518_to_fp16)[name = tensor("aw_2819_cast_fp16")]; + tensor var_33521_equation_0 = const()[name = tensor("op_33521_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33521_cast_fp16 = einsum(equation = var_33521_equation_0, values = (var_33363_cast_fp16, var_33280_cast_fp16))[name = tensor("op_33521_cast_fp16")]; + tensor var_33522_to_fp16 = const()[name = tensor("op_33522_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2821_cast_fp16 = mul(x = var_33521_cast_fp16, y = var_33522_to_fp16)[name = tensor("aw_2821_cast_fp16")]; + tensor var_33525_equation_0 = const()[name = tensor("op_33525_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33525_cast_fp16 = einsum(equation = var_33525_equation_0, values = (var_33367_cast_fp16, var_33284_cast_fp16))[name = tensor("op_33525_cast_fp16")]; + tensor var_33526_to_fp16 = const()[name = tensor("op_33526_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2823_cast_fp16 = mul(x = var_33525_cast_fp16, y = var_33526_to_fp16)[name = tensor("aw_2823_cast_fp16")]; + tensor var_33529_equation_0 = const()[name = tensor("op_33529_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33529_cast_fp16 = einsum(equation = var_33529_equation_0, values = (var_33371_cast_fp16, var_33288_cast_fp16))[name = tensor("op_33529_cast_fp16")]; + tensor var_33530_to_fp16 = const()[name = tensor("op_33530_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2825_cast_fp16 = mul(x = var_33529_cast_fp16, y = var_33530_to_fp16)[name = tensor("aw_2825_cast_fp16")]; + tensor var_33533_equation_0 = const()[name = tensor("op_33533_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33533_cast_fp16 = einsum(equation = var_33533_equation_0, values = (var_33375_cast_fp16, var_33292_cast_fp16))[name = tensor("op_33533_cast_fp16")]; + tensor var_33534_to_fp16 = const()[name = tensor("op_33534_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2827_cast_fp16 = mul(x = var_33533_cast_fp16, y = var_33534_to_fp16)[name = tensor("aw_2827_cast_fp16")]; + tensor var_33537_equation_0 = const()[name = tensor("op_33537_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33537_cast_fp16 = einsum(equation = var_33537_equation_0, values = (var_33379_cast_fp16, var_33296_cast_fp16))[name = tensor("op_33537_cast_fp16")]; + tensor var_33538_to_fp16 = const()[name = tensor("op_33538_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2829_cast_fp16 = mul(x = var_33537_cast_fp16, y = var_33538_to_fp16)[name = tensor("aw_2829_cast_fp16")]; + tensor var_33541_equation_0 = const()[name = tensor("op_33541_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33541_cast_fp16 = einsum(equation = var_33541_equation_0, values = (var_33383_cast_fp16, var_33300_cast_fp16))[name = tensor("op_33541_cast_fp16")]; + tensor var_33542_to_fp16 = const()[name = tensor("op_33542_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2831_cast_fp16 = mul(x = var_33541_cast_fp16, y = var_33542_to_fp16)[name = tensor("aw_2831_cast_fp16")]; + tensor var_33545_equation_0 = const()[name = tensor("op_33545_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33545_cast_fp16 = einsum(equation = var_33545_equation_0, values = (var_33387_cast_fp16, var_33304_cast_fp16))[name = tensor("op_33545_cast_fp16")]; + tensor var_33546_to_fp16 = const()[name = tensor("op_33546_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2833_cast_fp16 = mul(x = var_33545_cast_fp16, y = var_33546_to_fp16)[name = tensor("aw_2833_cast_fp16")]; + tensor var_33549_equation_0 = const()[name = tensor("op_33549_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33549_cast_fp16 = einsum(equation = var_33549_equation_0, values = (var_33391_cast_fp16, var_33308_cast_fp16))[name = tensor("op_33549_cast_fp16")]; + tensor var_33550_to_fp16 = const()[name = tensor("op_33550_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2835_cast_fp16 = mul(x = var_33549_cast_fp16, y = var_33550_to_fp16)[name = tensor("aw_2835_cast_fp16")]; + tensor var_33553_equation_0 = const()[name = tensor("op_33553_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33553_cast_fp16 = einsum(equation = var_33553_equation_0, values = (var_33395_cast_fp16, var_33312_cast_fp16))[name = tensor("op_33553_cast_fp16")]; + tensor var_33554_to_fp16 = const()[name = tensor("op_33554_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2837_cast_fp16 = mul(x = var_33553_cast_fp16, y = var_33554_to_fp16)[name = tensor("aw_2837_cast_fp16")]; + tensor var_33557_equation_0 = const()[name = tensor("op_33557_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33557_cast_fp16 = einsum(equation = var_33557_equation_0, values = (var_33399_cast_fp16, var_33316_cast_fp16))[name = tensor("op_33557_cast_fp16")]; + tensor var_33558_to_fp16 = const()[name = tensor("op_33558_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2839_cast_fp16 = mul(x = var_33557_cast_fp16, y = var_33558_to_fp16)[name = tensor("aw_2839_cast_fp16")]; + tensor var_33560_cast_fp16 = softmax(axis = var_30355, x = aw_2801_cast_fp16)[name = tensor("op_33560_cast_fp16")]; + tensor var_33561_cast_fp16 = softmax(axis = var_30355, x = aw_2803_cast_fp16)[name = tensor("op_33561_cast_fp16")]; + tensor var_33562_cast_fp16 = softmax(axis = var_30355, x = aw_2805_cast_fp16)[name = tensor("op_33562_cast_fp16")]; + tensor var_33563_cast_fp16 = softmax(axis = var_30355, x = aw_2807_cast_fp16)[name = tensor("op_33563_cast_fp16")]; + tensor var_33564_cast_fp16 = softmax(axis = var_30355, x = aw_2809_cast_fp16)[name = tensor("op_33564_cast_fp16")]; + tensor var_33565_cast_fp16 = softmax(axis = var_30355, x = aw_2811_cast_fp16)[name = tensor("op_33565_cast_fp16")]; + tensor var_33566_cast_fp16 = softmax(axis = var_30355, x = aw_2813_cast_fp16)[name = tensor("op_33566_cast_fp16")]; + tensor var_33567_cast_fp16 = softmax(axis = var_30355, x = aw_2815_cast_fp16)[name = tensor("op_33567_cast_fp16")]; + tensor var_33568_cast_fp16 = softmax(axis = var_30355, x = aw_2817_cast_fp16)[name = tensor("op_33568_cast_fp16")]; + tensor var_33569_cast_fp16 = softmax(axis = var_30355, x = aw_2819_cast_fp16)[name = tensor("op_33569_cast_fp16")]; + tensor var_33570_cast_fp16 = softmax(axis = var_30355, x = aw_2821_cast_fp16)[name = tensor("op_33570_cast_fp16")]; + tensor var_33571_cast_fp16 = softmax(axis = var_30355, x = aw_2823_cast_fp16)[name = tensor("op_33571_cast_fp16")]; + tensor var_33572_cast_fp16 = softmax(axis = var_30355, x = aw_2825_cast_fp16)[name = tensor("op_33572_cast_fp16")]; + tensor var_33573_cast_fp16 = softmax(axis = var_30355, x = aw_2827_cast_fp16)[name = tensor("op_33573_cast_fp16")]; + tensor var_33574_cast_fp16 = softmax(axis = var_30355, x = aw_2829_cast_fp16)[name = tensor("op_33574_cast_fp16")]; + tensor var_33575_cast_fp16 = softmax(axis = var_30355, x = aw_2831_cast_fp16)[name = tensor("op_33575_cast_fp16")]; + tensor var_33576_cast_fp16 = softmax(axis = var_30355, x = aw_2833_cast_fp16)[name = tensor("op_33576_cast_fp16")]; + tensor var_33577_cast_fp16 = softmax(axis = var_30355, x = aw_2835_cast_fp16)[name = tensor("op_33577_cast_fp16")]; + tensor var_33578_cast_fp16 = softmax(axis = var_30355, x = aw_2837_cast_fp16)[name = tensor("op_33578_cast_fp16")]; + tensor var_33579_cast_fp16 = softmax(axis = var_30355, x = aw_2839_cast_fp16)[name = tensor("op_33579_cast_fp16")]; + tensor var_33581_equation_0 = const()[name = tensor("op_33581_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_33581_cast_fp16 = einsum(equation = var_33581_equation_0, values = (var_33401_cast_fp16, var_33560_cast_fp16))[name = tensor("op_33581_cast_fp16")]; + tensor var_33583_equation_0 = const()[name = tensor("op_33583_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_33583_cast_fp16 = einsum(equation = var_33583_equation_0, values = (var_33405_cast_fp16, var_33561_cast_fp16))[name = tensor("op_33583_cast_fp16")]; + tensor var_33585_equation_0 = const()[name = tensor("op_33585_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_33585_cast_fp16 = einsum(equation = var_33585_equation_0, values = (var_33409_cast_fp16, var_33562_cast_fp16))[name = tensor("op_33585_cast_fp16")]; + tensor var_33587_equation_0 = const()[name = tensor("op_33587_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_33587_cast_fp16 = einsum(equation = var_33587_equation_0, values = (var_33413_cast_fp16, var_33563_cast_fp16))[name = tensor("op_33587_cast_fp16")]; + tensor var_33589_equation_0 = const()[name = tensor("op_33589_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_33589_cast_fp16 = einsum(equation = var_33589_equation_0, values = (var_33417_cast_fp16, var_33564_cast_fp16))[name = tensor("op_33589_cast_fp16")]; + tensor var_33591_equation_0 = const()[name = tensor("op_33591_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_33591_cast_fp16 = einsum(equation = var_33591_equation_0, values = (var_33421_cast_fp16, var_33565_cast_fp16))[name = tensor("op_33591_cast_fp16")]; + tensor var_33593_equation_0 = const()[name = tensor("op_33593_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_33593_cast_fp16 = einsum(equation = var_33593_equation_0, values = (var_33425_cast_fp16, var_33566_cast_fp16))[name = tensor("op_33593_cast_fp16")]; + tensor var_33595_equation_0 = const()[name = tensor("op_33595_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_33595_cast_fp16 = einsum(equation = var_33595_equation_0, values = (var_33429_cast_fp16, var_33567_cast_fp16))[name = tensor("op_33595_cast_fp16")]; + tensor var_33597_equation_0 = const()[name = tensor("op_33597_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_33597_cast_fp16 = einsum(equation = var_33597_equation_0, values = (var_33433_cast_fp16, var_33568_cast_fp16))[name = tensor("op_33597_cast_fp16")]; + tensor var_33599_equation_0 = const()[name = tensor("op_33599_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_33599_cast_fp16 = einsum(equation = var_33599_equation_0, values = (var_33437_cast_fp16, var_33569_cast_fp16))[name = tensor("op_33599_cast_fp16")]; + tensor var_33601_equation_0 = const()[name = tensor("op_33601_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_33601_cast_fp16 = einsum(equation = var_33601_equation_0, values = (var_33441_cast_fp16, var_33570_cast_fp16))[name = tensor("op_33601_cast_fp16")]; + tensor var_33603_equation_0 = const()[name = tensor("op_33603_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_33603_cast_fp16 = einsum(equation = var_33603_equation_0, values = (var_33445_cast_fp16, var_33571_cast_fp16))[name = tensor("op_33603_cast_fp16")]; + tensor var_33605_equation_0 = const()[name = tensor("op_33605_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_33605_cast_fp16 = einsum(equation = var_33605_equation_0, values = (var_33449_cast_fp16, var_33572_cast_fp16))[name = tensor("op_33605_cast_fp16")]; + tensor var_33607_equation_0 = const()[name = tensor("op_33607_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_33607_cast_fp16 = einsum(equation = var_33607_equation_0, values = (var_33453_cast_fp16, var_33573_cast_fp16))[name = tensor("op_33607_cast_fp16")]; + tensor var_33609_equation_0 = const()[name = tensor("op_33609_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_33609_cast_fp16 = einsum(equation = var_33609_equation_0, values = (var_33457_cast_fp16, var_33574_cast_fp16))[name = tensor("op_33609_cast_fp16")]; + tensor var_33611_equation_0 = const()[name = tensor("op_33611_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_33611_cast_fp16 = einsum(equation = var_33611_equation_0, values = (var_33461_cast_fp16, var_33575_cast_fp16))[name = tensor("op_33611_cast_fp16")]; + tensor var_33613_equation_0 = const()[name = tensor("op_33613_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_33613_cast_fp16 = einsum(equation = var_33613_equation_0, values = (var_33465_cast_fp16, var_33576_cast_fp16))[name = tensor("op_33613_cast_fp16")]; + tensor var_33615_equation_0 = const()[name = tensor("op_33615_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_33615_cast_fp16 = einsum(equation = var_33615_equation_0, values = (var_33469_cast_fp16, var_33577_cast_fp16))[name = tensor("op_33615_cast_fp16")]; + tensor var_33617_equation_0 = const()[name = tensor("op_33617_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_33617_cast_fp16 = einsum(equation = var_33617_equation_0, values = (var_33473_cast_fp16, var_33578_cast_fp16))[name = tensor("op_33617_cast_fp16")]; + tensor var_33619_equation_0 = const()[name = tensor("op_33619_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_33619_cast_fp16 = einsum(equation = var_33619_equation_0, values = (var_33477_cast_fp16, var_33579_cast_fp16))[name = tensor("op_33619_cast_fp16")]; + tensor input_463_interleave_0 = const()[name = tensor("input_463_interleave_0"), val = tensor(false)]; + tensor input_463_cast_fp16 = concat(axis = var_30355, interleave = input_463_interleave_0, values = (var_33581_cast_fp16, var_33583_cast_fp16, var_33585_cast_fp16, var_33587_cast_fp16, var_33589_cast_fp16, var_33591_cast_fp16, var_33593_cast_fp16, var_33595_cast_fp16, var_33597_cast_fp16, var_33599_cast_fp16, var_33601_cast_fp16, var_33603_cast_fp16, var_33605_cast_fp16, var_33607_cast_fp16, var_33609_cast_fp16, var_33611_cast_fp16, var_33613_cast_fp16, var_33615_cast_fp16, var_33617_cast_fp16, var_33619_cast_fp16))[name = tensor("input_463_cast_fp16")]; + tensor var_33625 = const()[name = tensor("op_33625"), val = tensor([1, 1])]; + tensor var_33627 = const()[name = tensor("op_33627"), val = tensor([1, 1])]; + tensor var_33629_pad_type_0 = const()[name = tensor("op_33629_pad_type_0"), val = tensor("custom")]; + tensor var_33629_pad_0 = const()[name = tensor("op_33629_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_3_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_3_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2806796096)))]; + tensor up_blocks_0_attentions_0_transformer_blocks_3_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_3_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2810072960)))]; + tensor var_33629_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_3_attn1_to_out_0_bias_to_fp16, dilations = var_33627, groups = var_30355, pad = var_33629_pad_0, pad_type = var_33629_pad_type_0, strides = var_33625, weight = up_blocks_0_attentions_0_transformer_blocks_3_attn1_to_out_0_weight_to_fp16, x = input_463_cast_fp16)[name = tensor("op_33629_cast_fp16")]; + tensor inputs_225_cast_fp16 = add(x = var_33629_cast_fp16, y = inputs_223_cast_fp16)[name = tensor("inputs_225_cast_fp16")]; + tensor hidden_states_307_axes_0 = const()[name = tensor("hidden_states_307_axes_0"), val = tensor([1])]; + tensor hidden_states_307_gamma_0_to_fp16 = const()[name = tensor("hidden_states_307_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2810075584)))]; + tensor hidden_states_307_beta_0_to_fp16 = const()[name = tensor("hidden_states_307_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2810078208)))]; + tensor var_33639_to_fp16 = const()[name = tensor("op_33639_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_307_cast_fp16 = layer_norm(axes = hidden_states_307_axes_0, beta = hidden_states_307_beta_0_to_fp16, epsilon = var_33639_to_fp16, gamma = hidden_states_307_gamma_0_to_fp16, x = inputs_225_cast_fp16)[name = tensor("hidden_states_307_cast_fp16")]; + tensor var_33654 = const()[name = tensor("op_33654"), val = tensor([1, 1])]; + tensor var_33656 = const()[name = tensor("op_33656"), val = tensor([1, 1])]; + tensor q_151_pad_type_0 = const()[name = tensor("q_151_pad_type_0"), val = tensor("custom")]; + tensor q_151_pad_0 = const()[name = tensor("q_151_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_3_attn2_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_3_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2810080832)))]; + tensor q_151_cast_fp16 = conv(dilations = var_33656, groups = var_30355, pad = q_151_pad_0, pad_type = q_151_pad_type_0, strides = var_33654, weight = up_blocks_0_attentions_0_transformer_blocks_3_attn2_to_q_weight_to_fp16, x = hidden_states_307_cast_fp16)[name = tensor("q_151_cast_fp16")]; + tensor var_33660 = const()[name = tensor("op_33660"), val = tensor([1, 1])]; + tensor var_33662 = const()[name = tensor("op_33662"), val = tensor([1, 1])]; + tensor k_301_pad_type_0 = const()[name = tensor("k_301_pad_type_0"), val = tensor("custom")]; + tensor k_301_pad_0 = const()[name = tensor("k_301_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_3_attn2_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_3_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2813357696)))]; + tensor k_301_cast_fp16 = conv(dilations = var_33662, groups = var_30355, pad = k_301_pad_0, pad_type = k_301_pad_type_0, strides = var_33660, weight = up_blocks_0_attentions_0_transformer_blocks_3_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_301_cast_fp16")]; + tensor var_33666 = const()[name = tensor("op_33666"), val = tensor([1, 1])]; + tensor var_33668 = const()[name = tensor("op_33668"), val = tensor([1, 1])]; + tensor v_151_pad_type_0 = const()[name = tensor("v_151_pad_type_0"), val = tensor("custom")]; + tensor v_151_pad_0 = const()[name = tensor("v_151_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_3_attn2_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_3_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2818600640)))]; + tensor v_151_cast_fp16 = conv(dilations = var_33668, groups = var_30355, pad = v_151_pad_0, pad_type = v_151_pad_type_0, strides = var_33666, weight = up_blocks_0_attentions_0_transformer_blocks_3_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_151_cast_fp16")]; + tensor var_33672_begin_0 = const()[name = tensor("op_33672_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_33672_end_0 = const()[name = tensor("op_33672_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_33672_end_mask_0 = const()[name = tensor("op_33672_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33672_cast_fp16 = slice_by_index(begin = var_33672_begin_0, end = var_33672_end_0, end_mask = var_33672_end_mask_0, x = q_151_cast_fp16)[name = tensor("op_33672_cast_fp16")]; + tensor var_33676_begin_0 = const()[name = tensor("op_33676_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_33676_end_0 = const()[name = tensor("op_33676_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_33676_end_mask_0 = const()[name = tensor("op_33676_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33676_cast_fp16 = slice_by_index(begin = var_33676_begin_0, end = var_33676_end_0, end_mask = var_33676_end_mask_0, x = q_151_cast_fp16)[name = tensor("op_33676_cast_fp16")]; + tensor var_33680_begin_0 = const()[name = tensor("op_33680_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_33680_end_0 = const()[name = tensor("op_33680_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_33680_end_mask_0 = const()[name = tensor("op_33680_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33680_cast_fp16 = slice_by_index(begin = var_33680_begin_0, end = var_33680_end_0, end_mask = var_33680_end_mask_0, x = q_151_cast_fp16)[name = tensor("op_33680_cast_fp16")]; + tensor var_33684_begin_0 = const()[name = tensor("op_33684_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_33684_end_0 = const()[name = tensor("op_33684_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_33684_end_mask_0 = const()[name = tensor("op_33684_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33684_cast_fp16 = slice_by_index(begin = var_33684_begin_0, end = var_33684_end_0, end_mask = var_33684_end_mask_0, x = q_151_cast_fp16)[name = tensor("op_33684_cast_fp16")]; + tensor var_33688_begin_0 = const()[name = tensor("op_33688_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_33688_end_0 = const()[name = tensor("op_33688_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_33688_end_mask_0 = const()[name = tensor("op_33688_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33688_cast_fp16 = slice_by_index(begin = var_33688_begin_0, end = var_33688_end_0, end_mask = var_33688_end_mask_0, x = q_151_cast_fp16)[name = tensor("op_33688_cast_fp16")]; + tensor var_33692_begin_0 = const()[name = tensor("op_33692_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_33692_end_0 = const()[name = tensor("op_33692_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_33692_end_mask_0 = const()[name = tensor("op_33692_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33692_cast_fp16 = slice_by_index(begin = var_33692_begin_0, end = var_33692_end_0, end_mask = var_33692_end_mask_0, x = q_151_cast_fp16)[name = tensor("op_33692_cast_fp16")]; + tensor var_33696_begin_0 = const()[name = tensor("op_33696_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_33696_end_0 = const()[name = tensor("op_33696_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_33696_end_mask_0 = const()[name = tensor("op_33696_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33696_cast_fp16 = slice_by_index(begin = var_33696_begin_0, end = var_33696_end_0, end_mask = var_33696_end_mask_0, x = q_151_cast_fp16)[name = tensor("op_33696_cast_fp16")]; + tensor var_33700_begin_0 = const()[name = tensor("op_33700_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_33700_end_0 = const()[name = tensor("op_33700_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_33700_end_mask_0 = const()[name = tensor("op_33700_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33700_cast_fp16 = slice_by_index(begin = var_33700_begin_0, end = var_33700_end_0, end_mask = var_33700_end_mask_0, x = q_151_cast_fp16)[name = tensor("op_33700_cast_fp16")]; + tensor var_33704_begin_0 = const()[name = tensor("op_33704_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_33704_end_0 = const()[name = tensor("op_33704_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_33704_end_mask_0 = const()[name = tensor("op_33704_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33704_cast_fp16 = slice_by_index(begin = var_33704_begin_0, end = var_33704_end_0, end_mask = var_33704_end_mask_0, x = q_151_cast_fp16)[name = tensor("op_33704_cast_fp16")]; + tensor var_33708_begin_0 = const()[name = tensor("op_33708_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_33708_end_0 = const()[name = tensor("op_33708_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_33708_end_mask_0 = const()[name = tensor("op_33708_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33708_cast_fp16 = slice_by_index(begin = var_33708_begin_0, end = var_33708_end_0, end_mask = var_33708_end_mask_0, x = q_151_cast_fp16)[name = tensor("op_33708_cast_fp16")]; + tensor var_33712_begin_0 = const()[name = tensor("op_33712_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_33712_end_0 = const()[name = tensor("op_33712_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_33712_end_mask_0 = const()[name = tensor("op_33712_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33712_cast_fp16 = slice_by_index(begin = var_33712_begin_0, end = var_33712_end_0, end_mask = var_33712_end_mask_0, x = q_151_cast_fp16)[name = tensor("op_33712_cast_fp16")]; + tensor var_33716_begin_0 = const()[name = tensor("op_33716_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_33716_end_0 = const()[name = tensor("op_33716_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_33716_end_mask_0 = const()[name = tensor("op_33716_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33716_cast_fp16 = slice_by_index(begin = var_33716_begin_0, end = var_33716_end_0, end_mask = var_33716_end_mask_0, x = q_151_cast_fp16)[name = tensor("op_33716_cast_fp16")]; + tensor var_33720_begin_0 = const()[name = tensor("op_33720_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_33720_end_0 = const()[name = tensor("op_33720_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_33720_end_mask_0 = const()[name = tensor("op_33720_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33720_cast_fp16 = slice_by_index(begin = var_33720_begin_0, end = var_33720_end_0, end_mask = var_33720_end_mask_0, x = q_151_cast_fp16)[name = tensor("op_33720_cast_fp16")]; + tensor var_33724_begin_0 = const()[name = tensor("op_33724_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_33724_end_0 = const()[name = tensor("op_33724_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_33724_end_mask_0 = const()[name = tensor("op_33724_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33724_cast_fp16 = slice_by_index(begin = var_33724_begin_0, end = var_33724_end_0, end_mask = var_33724_end_mask_0, x = q_151_cast_fp16)[name = tensor("op_33724_cast_fp16")]; + tensor var_33728_begin_0 = const()[name = tensor("op_33728_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_33728_end_0 = const()[name = tensor("op_33728_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_33728_end_mask_0 = const()[name = tensor("op_33728_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33728_cast_fp16 = slice_by_index(begin = var_33728_begin_0, end = var_33728_end_0, end_mask = var_33728_end_mask_0, x = q_151_cast_fp16)[name = tensor("op_33728_cast_fp16")]; + tensor var_33732_begin_0 = const()[name = tensor("op_33732_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_33732_end_0 = const()[name = tensor("op_33732_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_33732_end_mask_0 = const()[name = tensor("op_33732_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33732_cast_fp16 = slice_by_index(begin = var_33732_begin_0, end = var_33732_end_0, end_mask = var_33732_end_mask_0, x = q_151_cast_fp16)[name = tensor("op_33732_cast_fp16")]; + tensor var_33736_begin_0 = const()[name = tensor("op_33736_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_33736_end_0 = const()[name = tensor("op_33736_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_33736_end_mask_0 = const()[name = tensor("op_33736_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33736_cast_fp16 = slice_by_index(begin = var_33736_begin_0, end = var_33736_end_0, end_mask = var_33736_end_mask_0, x = q_151_cast_fp16)[name = tensor("op_33736_cast_fp16")]; + tensor var_33740_begin_0 = const()[name = tensor("op_33740_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_33740_end_0 = const()[name = tensor("op_33740_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_33740_end_mask_0 = const()[name = tensor("op_33740_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33740_cast_fp16 = slice_by_index(begin = var_33740_begin_0, end = var_33740_end_0, end_mask = var_33740_end_mask_0, x = q_151_cast_fp16)[name = tensor("op_33740_cast_fp16")]; + tensor var_33744_begin_0 = const()[name = tensor("op_33744_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_33744_end_0 = const()[name = tensor("op_33744_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_33744_end_mask_0 = const()[name = tensor("op_33744_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33744_cast_fp16 = slice_by_index(begin = var_33744_begin_0, end = var_33744_end_0, end_mask = var_33744_end_mask_0, x = q_151_cast_fp16)[name = tensor("op_33744_cast_fp16")]; + tensor var_33748_begin_0 = const()[name = tensor("op_33748_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_33748_end_0 = const()[name = tensor("op_33748_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_33748_end_mask_0 = const()[name = tensor("op_33748_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33748_cast_fp16 = slice_by_index(begin = var_33748_begin_0, end = var_33748_end_0, end_mask = var_33748_end_mask_0, x = q_151_cast_fp16)[name = tensor("op_33748_cast_fp16")]; + tensor k_303_perm_0 = const()[name = tensor("k_303_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_33755_begin_0 = const()[name = tensor("op_33755_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_33755_end_0 = const()[name = tensor("op_33755_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_33755_end_mask_0 = const()[name = tensor("op_33755_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_303_cast_fp16 = transpose(perm = k_303_perm_0, x = k_301_cast_fp16)[name = tensor("transpose_64")]; + tensor var_33755_cast_fp16 = slice_by_index(begin = var_33755_begin_0, end = var_33755_end_0, end_mask = var_33755_end_mask_0, x = k_303_cast_fp16)[name = tensor("op_33755_cast_fp16")]; + tensor var_33759_begin_0 = const()[name = tensor("op_33759_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_33759_end_0 = const()[name = tensor("op_33759_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_33759_end_mask_0 = const()[name = tensor("op_33759_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33759_cast_fp16 = slice_by_index(begin = var_33759_begin_0, end = var_33759_end_0, end_mask = var_33759_end_mask_0, x = k_303_cast_fp16)[name = tensor("op_33759_cast_fp16")]; + tensor var_33763_begin_0 = const()[name = tensor("op_33763_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_33763_end_0 = const()[name = tensor("op_33763_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_33763_end_mask_0 = const()[name = tensor("op_33763_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33763_cast_fp16 = slice_by_index(begin = var_33763_begin_0, end = var_33763_end_0, end_mask = var_33763_end_mask_0, x = k_303_cast_fp16)[name = tensor("op_33763_cast_fp16")]; + tensor var_33767_begin_0 = const()[name = tensor("op_33767_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_33767_end_0 = const()[name = tensor("op_33767_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_33767_end_mask_0 = const()[name = tensor("op_33767_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33767_cast_fp16 = slice_by_index(begin = var_33767_begin_0, end = var_33767_end_0, end_mask = var_33767_end_mask_0, x = k_303_cast_fp16)[name = tensor("op_33767_cast_fp16")]; + tensor var_33771_begin_0 = const()[name = tensor("op_33771_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_33771_end_0 = const()[name = tensor("op_33771_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_33771_end_mask_0 = const()[name = tensor("op_33771_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33771_cast_fp16 = slice_by_index(begin = var_33771_begin_0, end = var_33771_end_0, end_mask = var_33771_end_mask_0, x = k_303_cast_fp16)[name = tensor("op_33771_cast_fp16")]; + tensor var_33775_begin_0 = const()[name = tensor("op_33775_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_33775_end_0 = const()[name = tensor("op_33775_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_33775_end_mask_0 = const()[name = tensor("op_33775_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33775_cast_fp16 = slice_by_index(begin = var_33775_begin_0, end = var_33775_end_0, end_mask = var_33775_end_mask_0, x = k_303_cast_fp16)[name = tensor("op_33775_cast_fp16")]; + tensor var_33779_begin_0 = const()[name = tensor("op_33779_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_33779_end_0 = const()[name = tensor("op_33779_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_33779_end_mask_0 = const()[name = tensor("op_33779_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33779_cast_fp16 = slice_by_index(begin = var_33779_begin_0, end = var_33779_end_0, end_mask = var_33779_end_mask_0, x = k_303_cast_fp16)[name = tensor("op_33779_cast_fp16")]; + tensor var_33783_begin_0 = const()[name = tensor("op_33783_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_33783_end_0 = const()[name = tensor("op_33783_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_33783_end_mask_0 = const()[name = tensor("op_33783_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33783_cast_fp16 = slice_by_index(begin = var_33783_begin_0, end = var_33783_end_0, end_mask = var_33783_end_mask_0, x = k_303_cast_fp16)[name = tensor("op_33783_cast_fp16")]; + tensor var_33787_begin_0 = const()[name = tensor("op_33787_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_33787_end_0 = const()[name = tensor("op_33787_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_33787_end_mask_0 = const()[name = tensor("op_33787_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33787_cast_fp16 = slice_by_index(begin = var_33787_begin_0, end = var_33787_end_0, end_mask = var_33787_end_mask_0, x = k_303_cast_fp16)[name = tensor("op_33787_cast_fp16")]; + tensor var_33791_begin_0 = const()[name = tensor("op_33791_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_33791_end_0 = const()[name = tensor("op_33791_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_33791_end_mask_0 = const()[name = tensor("op_33791_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33791_cast_fp16 = slice_by_index(begin = var_33791_begin_0, end = var_33791_end_0, end_mask = var_33791_end_mask_0, x = k_303_cast_fp16)[name = tensor("op_33791_cast_fp16")]; + tensor var_33795_begin_0 = const()[name = tensor("op_33795_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_33795_end_0 = const()[name = tensor("op_33795_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_33795_end_mask_0 = const()[name = tensor("op_33795_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33795_cast_fp16 = slice_by_index(begin = var_33795_begin_0, end = var_33795_end_0, end_mask = var_33795_end_mask_0, x = k_303_cast_fp16)[name = tensor("op_33795_cast_fp16")]; + tensor var_33799_begin_0 = const()[name = tensor("op_33799_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_33799_end_0 = const()[name = tensor("op_33799_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_33799_end_mask_0 = const()[name = tensor("op_33799_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33799_cast_fp16 = slice_by_index(begin = var_33799_begin_0, end = var_33799_end_0, end_mask = var_33799_end_mask_0, x = k_303_cast_fp16)[name = tensor("op_33799_cast_fp16")]; + tensor var_33803_begin_0 = const()[name = tensor("op_33803_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_33803_end_0 = const()[name = tensor("op_33803_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_33803_end_mask_0 = const()[name = tensor("op_33803_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33803_cast_fp16 = slice_by_index(begin = var_33803_begin_0, end = var_33803_end_0, end_mask = var_33803_end_mask_0, x = k_303_cast_fp16)[name = tensor("op_33803_cast_fp16")]; + tensor var_33807_begin_0 = const()[name = tensor("op_33807_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_33807_end_0 = const()[name = tensor("op_33807_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_33807_end_mask_0 = const()[name = tensor("op_33807_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33807_cast_fp16 = slice_by_index(begin = var_33807_begin_0, end = var_33807_end_0, end_mask = var_33807_end_mask_0, x = k_303_cast_fp16)[name = tensor("op_33807_cast_fp16")]; + tensor var_33811_begin_0 = const()[name = tensor("op_33811_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_33811_end_0 = const()[name = tensor("op_33811_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_33811_end_mask_0 = const()[name = tensor("op_33811_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33811_cast_fp16 = slice_by_index(begin = var_33811_begin_0, end = var_33811_end_0, end_mask = var_33811_end_mask_0, x = k_303_cast_fp16)[name = tensor("op_33811_cast_fp16")]; + tensor var_33815_begin_0 = const()[name = tensor("op_33815_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_33815_end_0 = const()[name = tensor("op_33815_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_33815_end_mask_0 = const()[name = tensor("op_33815_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33815_cast_fp16 = slice_by_index(begin = var_33815_begin_0, end = var_33815_end_0, end_mask = var_33815_end_mask_0, x = k_303_cast_fp16)[name = tensor("op_33815_cast_fp16")]; + tensor var_33819_begin_0 = const()[name = tensor("op_33819_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_33819_end_0 = const()[name = tensor("op_33819_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_33819_end_mask_0 = const()[name = tensor("op_33819_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33819_cast_fp16 = slice_by_index(begin = var_33819_begin_0, end = var_33819_end_0, end_mask = var_33819_end_mask_0, x = k_303_cast_fp16)[name = tensor("op_33819_cast_fp16")]; + tensor var_33823_begin_0 = const()[name = tensor("op_33823_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_33823_end_0 = const()[name = tensor("op_33823_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_33823_end_mask_0 = const()[name = tensor("op_33823_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33823_cast_fp16 = slice_by_index(begin = var_33823_begin_0, end = var_33823_end_0, end_mask = var_33823_end_mask_0, x = k_303_cast_fp16)[name = tensor("op_33823_cast_fp16")]; + tensor var_33827_begin_0 = const()[name = tensor("op_33827_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_33827_end_0 = const()[name = tensor("op_33827_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_33827_end_mask_0 = const()[name = tensor("op_33827_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33827_cast_fp16 = slice_by_index(begin = var_33827_begin_0, end = var_33827_end_0, end_mask = var_33827_end_mask_0, x = k_303_cast_fp16)[name = tensor("op_33827_cast_fp16")]; + tensor var_33831_begin_0 = const()[name = tensor("op_33831_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_33831_end_0 = const()[name = tensor("op_33831_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_33831_end_mask_0 = const()[name = tensor("op_33831_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33831_cast_fp16 = slice_by_index(begin = var_33831_begin_0, end = var_33831_end_0, end_mask = var_33831_end_mask_0, x = k_303_cast_fp16)[name = tensor("op_33831_cast_fp16")]; + tensor var_33833_begin_0 = const()[name = tensor("op_33833_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_33833_end_0 = const()[name = tensor("op_33833_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_33833_end_mask_0 = const()[name = tensor("op_33833_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33833_cast_fp16 = slice_by_index(begin = var_33833_begin_0, end = var_33833_end_0, end_mask = var_33833_end_mask_0, x = v_151_cast_fp16)[name = tensor("op_33833_cast_fp16")]; + tensor var_33837_begin_0 = const()[name = tensor("op_33837_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_33837_end_0 = const()[name = tensor("op_33837_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_33837_end_mask_0 = const()[name = tensor("op_33837_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33837_cast_fp16 = slice_by_index(begin = var_33837_begin_0, end = var_33837_end_0, end_mask = var_33837_end_mask_0, x = v_151_cast_fp16)[name = tensor("op_33837_cast_fp16")]; + tensor var_33841_begin_0 = const()[name = tensor("op_33841_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_33841_end_0 = const()[name = tensor("op_33841_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_33841_end_mask_0 = const()[name = tensor("op_33841_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33841_cast_fp16 = slice_by_index(begin = var_33841_begin_0, end = var_33841_end_0, end_mask = var_33841_end_mask_0, x = v_151_cast_fp16)[name = tensor("op_33841_cast_fp16")]; + tensor var_33845_begin_0 = const()[name = tensor("op_33845_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_33845_end_0 = const()[name = tensor("op_33845_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_33845_end_mask_0 = const()[name = tensor("op_33845_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33845_cast_fp16 = slice_by_index(begin = var_33845_begin_0, end = var_33845_end_0, end_mask = var_33845_end_mask_0, x = v_151_cast_fp16)[name = tensor("op_33845_cast_fp16")]; + tensor var_33849_begin_0 = const()[name = tensor("op_33849_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_33849_end_0 = const()[name = tensor("op_33849_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_33849_end_mask_0 = const()[name = tensor("op_33849_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33849_cast_fp16 = slice_by_index(begin = var_33849_begin_0, end = var_33849_end_0, end_mask = var_33849_end_mask_0, x = v_151_cast_fp16)[name = tensor("op_33849_cast_fp16")]; + tensor var_33853_begin_0 = const()[name = tensor("op_33853_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_33853_end_0 = const()[name = tensor("op_33853_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_33853_end_mask_0 = const()[name = tensor("op_33853_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33853_cast_fp16 = slice_by_index(begin = var_33853_begin_0, end = var_33853_end_0, end_mask = var_33853_end_mask_0, x = v_151_cast_fp16)[name = tensor("op_33853_cast_fp16")]; + tensor var_33857_begin_0 = const()[name = tensor("op_33857_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_33857_end_0 = const()[name = tensor("op_33857_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_33857_end_mask_0 = const()[name = tensor("op_33857_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33857_cast_fp16 = slice_by_index(begin = var_33857_begin_0, end = var_33857_end_0, end_mask = var_33857_end_mask_0, x = v_151_cast_fp16)[name = tensor("op_33857_cast_fp16")]; + tensor var_33861_begin_0 = const()[name = tensor("op_33861_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_33861_end_0 = const()[name = tensor("op_33861_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_33861_end_mask_0 = const()[name = tensor("op_33861_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33861_cast_fp16 = slice_by_index(begin = var_33861_begin_0, end = var_33861_end_0, end_mask = var_33861_end_mask_0, x = v_151_cast_fp16)[name = tensor("op_33861_cast_fp16")]; + tensor var_33865_begin_0 = const()[name = tensor("op_33865_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_33865_end_0 = const()[name = tensor("op_33865_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_33865_end_mask_0 = const()[name = tensor("op_33865_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33865_cast_fp16 = slice_by_index(begin = var_33865_begin_0, end = var_33865_end_0, end_mask = var_33865_end_mask_0, x = v_151_cast_fp16)[name = tensor("op_33865_cast_fp16")]; + tensor var_33869_begin_0 = const()[name = tensor("op_33869_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_33869_end_0 = const()[name = tensor("op_33869_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_33869_end_mask_0 = const()[name = tensor("op_33869_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33869_cast_fp16 = slice_by_index(begin = var_33869_begin_0, end = var_33869_end_0, end_mask = var_33869_end_mask_0, x = v_151_cast_fp16)[name = tensor("op_33869_cast_fp16")]; + tensor var_33873_begin_0 = const()[name = tensor("op_33873_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_33873_end_0 = const()[name = tensor("op_33873_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_33873_end_mask_0 = const()[name = tensor("op_33873_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33873_cast_fp16 = slice_by_index(begin = var_33873_begin_0, end = var_33873_end_0, end_mask = var_33873_end_mask_0, x = v_151_cast_fp16)[name = tensor("op_33873_cast_fp16")]; + tensor var_33877_begin_0 = const()[name = tensor("op_33877_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_33877_end_0 = const()[name = tensor("op_33877_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_33877_end_mask_0 = const()[name = tensor("op_33877_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33877_cast_fp16 = slice_by_index(begin = var_33877_begin_0, end = var_33877_end_0, end_mask = var_33877_end_mask_0, x = v_151_cast_fp16)[name = tensor("op_33877_cast_fp16")]; + tensor var_33881_begin_0 = const()[name = tensor("op_33881_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_33881_end_0 = const()[name = tensor("op_33881_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_33881_end_mask_0 = const()[name = tensor("op_33881_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33881_cast_fp16 = slice_by_index(begin = var_33881_begin_0, end = var_33881_end_0, end_mask = var_33881_end_mask_0, x = v_151_cast_fp16)[name = tensor("op_33881_cast_fp16")]; + tensor var_33885_begin_0 = const()[name = tensor("op_33885_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_33885_end_0 = const()[name = tensor("op_33885_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_33885_end_mask_0 = const()[name = tensor("op_33885_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33885_cast_fp16 = slice_by_index(begin = var_33885_begin_0, end = var_33885_end_0, end_mask = var_33885_end_mask_0, x = v_151_cast_fp16)[name = tensor("op_33885_cast_fp16")]; + tensor var_33889_begin_0 = const()[name = tensor("op_33889_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_33889_end_0 = const()[name = tensor("op_33889_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_33889_end_mask_0 = const()[name = tensor("op_33889_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33889_cast_fp16 = slice_by_index(begin = var_33889_begin_0, end = var_33889_end_0, end_mask = var_33889_end_mask_0, x = v_151_cast_fp16)[name = tensor("op_33889_cast_fp16")]; + tensor var_33893_begin_0 = const()[name = tensor("op_33893_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_33893_end_0 = const()[name = tensor("op_33893_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_33893_end_mask_0 = const()[name = tensor("op_33893_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33893_cast_fp16 = slice_by_index(begin = var_33893_begin_0, end = var_33893_end_0, end_mask = var_33893_end_mask_0, x = v_151_cast_fp16)[name = tensor("op_33893_cast_fp16")]; + tensor var_33897_begin_0 = const()[name = tensor("op_33897_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_33897_end_0 = const()[name = tensor("op_33897_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_33897_end_mask_0 = const()[name = tensor("op_33897_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33897_cast_fp16 = slice_by_index(begin = var_33897_begin_0, end = var_33897_end_0, end_mask = var_33897_end_mask_0, x = v_151_cast_fp16)[name = tensor("op_33897_cast_fp16")]; + tensor var_33901_begin_0 = const()[name = tensor("op_33901_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_33901_end_0 = const()[name = tensor("op_33901_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_33901_end_mask_0 = const()[name = tensor("op_33901_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33901_cast_fp16 = slice_by_index(begin = var_33901_begin_0, end = var_33901_end_0, end_mask = var_33901_end_mask_0, x = v_151_cast_fp16)[name = tensor("op_33901_cast_fp16")]; + tensor var_33905_begin_0 = const()[name = tensor("op_33905_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_33905_end_0 = const()[name = tensor("op_33905_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_33905_end_mask_0 = const()[name = tensor("op_33905_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33905_cast_fp16 = slice_by_index(begin = var_33905_begin_0, end = var_33905_end_0, end_mask = var_33905_end_mask_0, x = v_151_cast_fp16)[name = tensor("op_33905_cast_fp16")]; + tensor var_33909_begin_0 = const()[name = tensor("op_33909_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_33909_end_0 = const()[name = tensor("op_33909_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_33909_end_mask_0 = const()[name = tensor("op_33909_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33909_cast_fp16 = slice_by_index(begin = var_33909_begin_0, end = var_33909_end_0, end_mask = var_33909_end_mask_0, x = v_151_cast_fp16)[name = tensor("op_33909_cast_fp16")]; + tensor var_33913_equation_0 = const()[name = tensor("op_33913_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33913_cast_fp16 = einsum(equation = var_33913_equation_0, values = (var_33755_cast_fp16, var_33672_cast_fp16))[name = tensor("op_33913_cast_fp16")]; + tensor var_33914_to_fp16 = const()[name = tensor("op_33914_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2841_cast_fp16 = mul(x = var_33913_cast_fp16, y = var_33914_to_fp16)[name = tensor("aw_2841_cast_fp16")]; + tensor var_33917_equation_0 = const()[name = tensor("op_33917_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33917_cast_fp16 = einsum(equation = var_33917_equation_0, values = (var_33759_cast_fp16, var_33676_cast_fp16))[name = tensor("op_33917_cast_fp16")]; + tensor var_33918_to_fp16 = const()[name = tensor("op_33918_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2843_cast_fp16 = mul(x = var_33917_cast_fp16, y = var_33918_to_fp16)[name = tensor("aw_2843_cast_fp16")]; + tensor var_33921_equation_0 = const()[name = tensor("op_33921_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33921_cast_fp16 = einsum(equation = var_33921_equation_0, values = (var_33763_cast_fp16, var_33680_cast_fp16))[name = tensor("op_33921_cast_fp16")]; + tensor var_33922_to_fp16 = const()[name = tensor("op_33922_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2845_cast_fp16 = mul(x = var_33921_cast_fp16, y = var_33922_to_fp16)[name = tensor("aw_2845_cast_fp16")]; + tensor var_33925_equation_0 = const()[name = tensor("op_33925_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33925_cast_fp16 = einsum(equation = var_33925_equation_0, values = (var_33767_cast_fp16, var_33684_cast_fp16))[name = tensor("op_33925_cast_fp16")]; + tensor var_33926_to_fp16 = const()[name = tensor("op_33926_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2847_cast_fp16 = mul(x = var_33925_cast_fp16, y = var_33926_to_fp16)[name = tensor("aw_2847_cast_fp16")]; + tensor var_33929_equation_0 = const()[name = tensor("op_33929_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33929_cast_fp16 = einsum(equation = var_33929_equation_0, values = (var_33771_cast_fp16, var_33688_cast_fp16))[name = tensor("op_33929_cast_fp16")]; + tensor var_33930_to_fp16 = const()[name = tensor("op_33930_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2849_cast_fp16 = mul(x = var_33929_cast_fp16, y = var_33930_to_fp16)[name = tensor("aw_2849_cast_fp16")]; + tensor var_33933_equation_0 = const()[name = tensor("op_33933_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33933_cast_fp16 = einsum(equation = var_33933_equation_0, values = (var_33775_cast_fp16, var_33692_cast_fp16))[name = tensor("op_33933_cast_fp16")]; + tensor var_33934_to_fp16 = const()[name = tensor("op_33934_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2851_cast_fp16 = mul(x = var_33933_cast_fp16, y = var_33934_to_fp16)[name = tensor("aw_2851_cast_fp16")]; + tensor var_33937_equation_0 = const()[name = tensor("op_33937_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33937_cast_fp16 = einsum(equation = var_33937_equation_0, values = (var_33779_cast_fp16, var_33696_cast_fp16))[name = tensor("op_33937_cast_fp16")]; + tensor var_33938_to_fp16 = const()[name = tensor("op_33938_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2853_cast_fp16 = mul(x = var_33937_cast_fp16, y = var_33938_to_fp16)[name = tensor("aw_2853_cast_fp16")]; + tensor var_33941_equation_0 = const()[name = tensor("op_33941_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33941_cast_fp16 = einsum(equation = var_33941_equation_0, values = (var_33783_cast_fp16, var_33700_cast_fp16))[name = tensor("op_33941_cast_fp16")]; + tensor var_33942_to_fp16 = const()[name = tensor("op_33942_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2855_cast_fp16 = mul(x = var_33941_cast_fp16, y = var_33942_to_fp16)[name = tensor("aw_2855_cast_fp16")]; + tensor var_33945_equation_0 = const()[name = tensor("op_33945_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33945_cast_fp16 = einsum(equation = var_33945_equation_0, values = (var_33787_cast_fp16, var_33704_cast_fp16))[name = tensor("op_33945_cast_fp16")]; + tensor var_33946_to_fp16 = const()[name = tensor("op_33946_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2857_cast_fp16 = mul(x = var_33945_cast_fp16, y = var_33946_to_fp16)[name = tensor("aw_2857_cast_fp16")]; + tensor var_33949_equation_0 = const()[name = tensor("op_33949_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33949_cast_fp16 = einsum(equation = var_33949_equation_0, values = (var_33791_cast_fp16, var_33708_cast_fp16))[name = tensor("op_33949_cast_fp16")]; + tensor var_33950_to_fp16 = const()[name = tensor("op_33950_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2859_cast_fp16 = mul(x = var_33949_cast_fp16, y = var_33950_to_fp16)[name = tensor("aw_2859_cast_fp16")]; + tensor var_33953_equation_0 = const()[name = tensor("op_33953_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33953_cast_fp16 = einsum(equation = var_33953_equation_0, values = (var_33795_cast_fp16, var_33712_cast_fp16))[name = tensor("op_33953_cast_fp16")]; + tensor var_33954_to_fp16 = const()[name = tensor("op_33954_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2861_cast_fp16 = mul(x = var_33953_cast_fp16, y = var_33954_to_fp16)[name = tensor("aw_2861_cast_fp16")]; + tensor var_33957_equation_0 = const()[name = tensor("op_33957_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33957_cast_fp16 = einsum(equation = var_33957_equation_0, values = (var_33799_cast_fp16, var_33716_cast_fp16))[name = tensor("op_33957_cast_fp16")]; + tensor var_33958_to_fp16 = const()[name = tensor("op_33958_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2863_cast_fp16 = mul(x = var_33957_cast_fp16, y = var_33958_to_fp16)[name = tensor("aw_2863_cast_fp16")]; + tensor var_33961_equation_0 = const()[name = tensor("op_33961_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33961_cast_fp16 = einsum(equation = var_33961_equation_0, values = (var_33803_cast_fp16, var_33720_cast_fp16))[name = tensor("op_33961_cast_fp16")]; + tensor var_33962_to_fp16 = const()[name = tensor("op_33962_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2865_cast_fp16 = mul(x = var_33961_cast_fp16, y = var_33962_to_fp16)[name = tensor("aw_2865_cast_fp16")]; + tensor var_33965_equation_0 = const()[name = tensor("op_33965_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33965_cast_fp16 = einsum(equation = var_33965_equation_0, values = (var_33807_cast_fp16, var_33724_cast_fp16))[name = tensor("op_33965_cast_fp16")]; + tensor var_33966_to_fp16 = const()[name = tensor("op_33966_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2867_cast_fp16 = mul(x = var_33965_cast_fp16, y = var_33966_to_fp16)[name = tensor("aw_2867_cast_fp16")]; + tensor var_33969_equation_0 = const()[name = tensor("op_33969_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33969_cast_fp16 = einsum(equation = var_33969_equation_0, values = (var_33811_cast_fp16, var_33728_cast_fp16))[name = tensor("op_33969_cast_fp16")]; + tensor var_33970_to_fp16 = const()[name = tensor("op_33970_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2869_cast_fp16 = mul(x = var_33969_cast_fp16, y = var_33970_to_fp16)[name = tensor("aw_2869_cast_fp16")]; + tensor var_33973_equation_0 = const()[name = tensor("op_33973_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33973_cast_fp16 = einsum(equation = var_33973_equation_0, values = (var_33815_cast_fp16, var_33732_cast_fp16))[name = tensor("op_33973_cast_fp16")]; + tensor var_33974_to_fp16 = const()[name = tensor("op_33974_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2871_cast_fp16 = mul(x = var_33973_cast_fp16, y = var_33974_to_fp16)[name = tensor("aw_2871_cast_fp16")]; + tensor var_33977_equation_0 = const()[name = tensor("op_33977_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33977_cast_fp16 = einsum(equation = var_33977_equation_0, values = (var_33819_cast_fp16, var_33736_cast_fp16))[name = tensor("op_33977_cast_fp16")]; + tensor var_33978_to_fp16 = const()[name = tensor("op_33978_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2873_cast_fp16 = mul(x = var_33977_cast_fp16, y = var_33978_to_fp16)[name = tensor("aw_2873_cast_fp16")]; + tensor var_33981_equation_0 = const()[name = tensor("op_33981_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33981_cast_fp16 = einsum(equation = var_33981_equation_0, values = (var_33823_cast_fp16, var_33740_cast_fp16))[name = tensor("op_33981_cast_fp16")]; + tensor var_33982_to_fp16 = const()[name = tensor("op_33982_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2875_cast_fp16 = mul(x = var_33981_cast_fp16, y = var_33982_to_fp16)[name = tensor("aw_2875_cast_fp16")]; + tensor var_33985_equation_0 = const()[name = tensor("op_33985_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33985_cast_fp16 = einsum(equation = var_33985_equation_0, values = (var_33827_cast_fp16, var_33744_cast_fp16))[name = tensor("op_33985_cast_fp16")]; + tensor var_33986_to_fp16 = const()[name = tensor("op_33986_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2877_cast_fp16 = mul(x = var_33985_cast_fp16, y = var_33986_to_fp16)[name = tensor("aw_2877_cast_fp16")]; + tensor var_33989_equation_0 = const()[name = tensor("op_33989_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33989_cast_fp16 = einsum(equation = var_33989_equation_0, values = (var_33831_cast_fp16, var_33748_cast_fp16))[name = tensor("op_33989_cast_fp16")]; + tensor var_33990_to_fp16 = const()[name = tensor("op_33990_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2879_cast_fp16 = mul(x = var_33989_cast_fp16, y = var_33990_to_fp16)[name = tensor("aw_2879_cast_fp16")]; + tensor var_33992_cast_fp16 = softmax(axis = var_30355, x = aw_2841_cast_fp16)[name = tensor("op_33992_cast_fp16")]; + tensor var_33993_cast_fp16 = softmax(axis = var_30355, x = aw_2843_cast_fp16)[name = tensor("op_33993_cast_fp16")]; + tensor var_33994_cast_fp16 = softmax(axis = var_30355, x = aw_2845_cast_fp16)[name = tensor("op_33994_cast_fp16")]; + tensor var_33995_cast_fp16 = softmax(axis = var_30355, x = aw_2847_cast_fp16)[name = tensor("op_33995_cast_fp16")]; + tensor var_33996_cast_fp16 = softmax(axis = var_30355, x = aw_2849_cast_fp16)[name = tensor("op_33996_cast_fp16")]; + tensor var_33997_cast_fp16 = softmax(axis = var_30355, x = aw_2851_cast_fp16)[name = tensor("op_33997_cast_fp16")]; + tensor var_33998_cast_fp16 = softmax(axis = var_30355, x = aw_2853_cast_fp16)[name = tensor("op_33998_cast_fp16")]; + tensor var_33999_cast_fp16 = softmax(axis = var_30355, x = aw_2855_cast_fp16)[name = tensor("op_33999_cast_fp16")]; + tensor var_34000_cast_fp16 = softmax(axis = var_30355, x = aw_2857_cast_fp16)[name = tensor("op_34000_cast_fp16")]; + tensor var_34001_cast_fp16 = softmax(axis = var_30355, x = aw_2859_cast_fp16)[name = tensor("op_34001_cast_fp16")]; + tensor var_34002_cast_fp16 = softmax(axis = var_30355, x = aw_2861_cast_fp16)[name = tensor("op_34002_cast_fp16")]; + tensor var_34003_cast_fp16 = softmax(axis = var_30355, x = aw_2863_cast_fp16)[name = tensor("op_34003_cast_fp16")]; + tensor var_34004_cast_fp16 = softmax(axis = var_30355, x = aw_2865_cast_fp16)[name = tensor("op_34004_cast_fp16")]; + tensor var_34005_cast_fp16 = softmax(axis = var_30355, x = aw_2867_cast_fp16)[name = tensor("op_34005_cast_fp16")]; + tensor var_34006_cast_fp16 = softmax(axis = var_30355, x = aw_2869_cast_fp16)[name = tensor("op_34006_cast_fp16")]; + tensor var_34007_cast_fp16 = softmax(axis = var_30355, x = aw_2871_cast_fp16)[name = tensor("op_34007_cast_fp16")]; + tensor var_34008_cast_fp16 = softmax(axis = var_30355, x = aw_2873_cast_fp16)[name = tensor("op_34008_cast_fp16")]; + tensor var_34009_cast_fp16 = softmax(axis = var_30355, x = aw_2875_cast_fp16)[name = tensor("op_34009_cast_fp16")]; + tensor var_34010_cast_fp16 = softmax(axis = var_30355, x = aw_2877_cast_fp16)[name = tensor("op_34010_cast_fp16")]; + tensor var_34011_cast_fp16 = softmax(axis = var_30355, x = aw_2879_cast_fp16)[name = tensor("op_34011_cast_fp16")]; + tensor var_34013_equation_0 = const()[name = tensor("op_34013_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34013_cast_fp16 = einsum(equation = var_34013_equation_0, values = (var_33833_cast_fp16, var_33992_cast_fp16))[name = tensor("op_34013_cast_fp16")]; + tensor var_34015_equation_0 = const()[name = tensor("op_34015_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34015_cast_fp16 = einsum(equation = var_34015_equation_0, values = (var_33837_cast_fp16, var_33993_cast_fp16))[name = tensor("op_34015_cast_fp16")]; + tensor var_34017_equation_0 = const()[name = tensor("op_34017_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34017_cast_fp16 = einsum(equation = var_34017_equation_0, values = (var_33841_cast_fp16, var_33994_cast_fp16))[name = tensor("op_34017_cast_fp16")]; + tensor var_34019_equation_0 = const()[name = tensor("op_34019_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34019_cast_fp16 = einsum(equation = var_34019_equation_0, values = (var_33845_cast_fp16, var_33995_cast_fp16))[name = tensor("op_34019_cast_fp16")]; + tensor var_34021_equation_0 = const()[name = tensor("op_34021_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34021_cast_fp16 = einsum(equation = var_34021_equation_0, values = (var_33849_cast_fp16, var_33996_cast_fp16))[name = tensor("op_34021_cast_fp16")]; + tensor var_34023_equation_0 = const()[name = tensor("op_34023_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34023_cast_fp16 = einsum(equation = var_34023_equation_0, values = (var_33853_cast_fp16, var_33997_cast_fp16))[name = tensor("op_34023_cast_fp16")]; + tensor var_34025_equation_0 = const()[name = tensor("op_34025_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34025_cast_fp16 = einsum(equation = var_34025_equation_0, values = (var_33857_cast_fp16, var_33998_cast_fp16))[name = tensor("op_34025_cast_fp16")]; + tensor var_34027_equation_0 = const()[name = tensor("op_34027_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34027_cast_fp16 = einsum(equation = var_34027_equation_0, values = (var_33861_cast_fp16, var_33999_cast_fp16))[name = tensor("op_34027_cast_fp16")]; + tensor var_34029_equation_0 = const()[name = tensor("op_34029_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34029_cast_fp16 = einsum(equation = var_34029_equation_0, values = (var_33865_cast_fp16, var_34000_cast_fp16))[name = tensor("op_34029_cast_fp16")]; + tensor var_34031_equation_0 = const()[name = tensor("op_34031_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34031_cast_fp16 = einsum(equation = var_34031_equation_0, values = (var_33869_cast_fp16, var_34001_cast_fp16))[name = tensor("op_34031_cast_fp16")]; + tensor var_34033_equation_0 = const()[name = tensor("op_34033_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34033_cast_fp16 = einsum(equation = var_34033_equation_0, values = (var_33873_cast_fp16, var_34002_cast_fp16))[name = tensor("op_34033_cast_fp16")]; + tensor var_34035_equation_0 = const()[name = tensor("op_34035_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34035_cast_fp16 = einsum(equation = var_34035_equation_0, values = (var_33877_cast_fp16, var_34003_cast_fp16))[name = tensor("op_34035_cast_fp16")]; + tensor var_34037_equation_0 = const()[name = tensor("op_34037_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34037_cast_fp16 = einsum(equation = var_34037_equation_0, values = (var_33881_cast_fp16, var_34004_cast_fp16))[name = tensor("op_34037_cast_fp16")]; + tensor var_34039_equation_0 = const()[name = tensor("op_34039_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34039_cast_fp16 = einsum(equation = var_34039_equation_0, values = (var_33885_cast_fp16, var_34005_cast_fp16))[name = tensor("op_34039_cast_fp16")]; + tensor var_34041_equation_0 = const()[name = tensor("op_34041_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34041_cast_fp16 = einsum(equation = var_34041_equation_0, values = (var_33889_cast_fp16, var_34006_cast_fp16))[name = tensor("op_34041_cast_fp16")]; + tensor var_34043_equation_0 = const()[name = tensor("op_34043_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34043_cast_fp16 = einsum(equation = var_34043_equation_0, values = (var_33893_cast_fp16, var_34007_cast_fp16))[name = tensor("op_34043_cast_fp16")]; + tensor var_34045_equation_0 = const()[name = tensor("op_34045_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34045_cast_fp16 = einsum(equation = var_34045_equation_0, values = (var_33897_cast_fp16, var_34008_cast_fp16))[name = tensor("op_34045_cast_fp16")]; + tensor var_34047_equation_0 = const()[name = tensor("op_34047_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34047_cast_fp16 = einsum(equation = var_34047_equation_0, values = (var_33901_cast_fp16, var_34009_cast_fp16))[name = tensor("op_34047_cast_fp16")]; + tensor var_34049_equation_0 = const()[name = tensor("op_34049_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34049_cast_fp16 = einsum(equation = var_34049_equation_0, values = (var_33905_cast_fp16, var_34010_cast_fp16))[name = tensor("op_34049_cast_fp16")]; + tensor var_34051_equation_0 = const()[name = tensor("op_34051_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34051_cast_fp16 = einsum(equation = var_34051_equation_0, values = (var_33909_cast_fp16, var_34011_cast_fp16))[name = tensor("op_34051_cast_fp16")]; + tensor input_465_interleave_0 = const()[name = tensor("input_465_interleave_0"), val = tensor(false)]; + tensor input_465_cast_fp16 = concat(axis = var_30355, interleave = input_465_interleave_0, values = (var_34013_cast_fp16, var_34015_cast_fp16, var_34017_cast_fp16, var_34019_cast_fp16, var_34021_cast_fp16, var_34023_cast_fp16, var_34025_cast_fp16, var_34027_cast_fp16, var_34029_cast_fp16, var_34031_cast_fp16, var_34033_cast_fp16, var_34035_cast_fp16, var_34037_cast_fp16, var_34039_cast_fp16, var_34041_cast_fp16, var_34043_cast_fp16, var_34045_cast_fp16, var_34047_cast_fp16, var_34049_cast_fp16, var_34051_cast_fp16))[name = tensor("input_465_cast_fp16")]; + tensor var_34057 = const()[name = tensor("op_34057"), val = tensor([1, 1])]; + tensor var_34059 = const()[name = tensor("op_34059"), val = tensor([1, 1])]; + tensor var_34061_pad_type_0 = const()[name = tensor("op_34061_pad_type_0"), val = tensor("custom")]; + tensor var_34061_pad_0 = const()[name = tensor("op_34061_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_3_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_3_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2823843584)))]; + tensor up_blocks_0_attentions_0_transformer_blocks_3_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_3_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2827120448)))]; + tensor var_34061_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_3_attn2_to_out_0_bias_to_fp16, dilations = var_34059, groups = var_30355, pad = var_34061_pad_0, pad_type = var_34061_pad_type_0, strides = var_34057, weight = up_blocks_0_attentions_0_transformer_blocks_3_attn2_to_out_0_weight_to_fp16, x = input_465_cast_fp16)[name = tensor("op_34061_cast_fp16")]; + tensor inputs_227_cast_fp16 = add(x = var_34061_cast_fp16, y = inputs_225_cast_fp16)[name = tensor("inputs_227_cast_fp16")]; + tensor input_467_axes_0 = const()[name = tensor("input_467_axes_0"), val = tensor([1])]; + tensor input_467_gamma_0_to_fp16 = const()[name = tensor("input_467_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2827123072)))]; + tensor input_467_beta_0_to_fp16 = const()[name = tensor("input_467_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2827125696)))]; + tensor var_34071_to_fp16 = const()[name = tensor("op_34071_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_467_cast_fp16 = layer_norm(axes = input_467_axes_0, beta = input_467_beta_0_to_fp16, epsilon = var_34071_to_fp16, gamma = input_467_gamma_0_to_fp16, x = inputs_227_cast_fp16)[name = tensor("input_467_cast_fp16")]; + tensor var_34087 = const()[name = tensor("op_34087"), val = tensor([1, 1])]; + tensor var_34089 = const()[name = tensor("op_34089"), val = tensor([1, 1])]; + tensor var_34091_pad_type_0 = const()[name = tensor("op_34091_pad_type_0"), val = tensor("custom")]; + tensor var_34091_pad_0 = const()[name = tensor("op_34091_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_3_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_3_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2827128320)))]; + tensor up_blocks_0_attentions_0_transformer_blocks_3_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_3_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2853342784)))]; + tensor var_34091_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_3_ff_net_0_proj_bias_to_fp16, dilations = var_34089, groups = var_30355, pad = var_34091_pad_0, pad_type = var_34091_pad_type_0, strides = var_34087, weight = up_blocks_0_attentions_0_transformer_blocks_3_ff_net_0_proj_weight_to_fp16, x = input_467_cast_fp16)[name = tensor("op_34091_cast_fp16")]; + tensor var_34092_split_sizes_0 = const()[name = tensor("op_34092_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_34092_axis_0 = const()[name = tensor("op_34092_axis_0"), val = tensor(1)]; + tensor var_34092_cast_fp16_0, tensor var_34092_cast_fp16_1 = split(axis = var_34092_axis_0, split_sizes = var_34092_split_sizes_0, x = var_34091_cast_fp16)[name = tensor("op_34092_cast_fp16")]; + tensor var_34094_mode_0 = const()[name = tensor("op_34094_mode_0"), val = tensor("EXACT")]; + tensor var_34094_cast_fp16 = gelu(mode = var_34094_mode_0, x = var_34092_cast_fp16_1)[name = tensor("op_34094_cast_fp16")]; + tensor input_469_cast_fp16 = mul(x = var_34092_cast_fp16_0, y = var_34094_cast_fp16)[name = tensor("input_469_cast_fp16")]; + tensor var_34098 = const()[name = tensor("op_34098"), val = tensor([1, 1])]; + tensor var_34100 = const()[name = tensor("op_34100"), val = tensor([1, 1])]; + tensor var_34102_pad_type_0 = const()[name = tensor("op_34102_pad_type_0"), val = tensor("custom")]; + tensor var_34102_pad_0 = const()[name = tensor("op_34102_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_3_ff_net_2_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_3_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2853363328)))]; + tensor up_blocks_0_attentions_0_transformer_blocks_3_ff_net_2_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_3_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2866470592)))]; + tensor var_34102_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_3_ff_net_2_bias_to_fp16, dilations = var_34100, groups = var_30355, pad = var_34102_pad_0, pad_type = var_34102_pad_type_0, strides = var_34098, weight = up_blocks_0_attentions_0_transformer_blocks_3_ff_net_2_weight_to_fp16, x = input_469_cast_fp16)[name = tensor("op_34102_cast_fp16")]; + tensor inputs_229_cast_fp16 = add(x = var_34102_cast_fp16, y = inputs_227_cast_fp16)[name = tensor("inputs_229_cast_fp16")]; + tensor hidden_states_311_axes_0 = const()[name = tensor("hidden_states_311_axes_0"), val = tensor([1])]; + tensor hidden_states_311_gamma_0_to_fp16 = const()[name = tensor("hidden_states_311_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2866473216)))]; + tensor hidden_states_311_beta_0_to_fp16 = const()[name = tensor("hidden_states_311_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2866475840)))]; + tensor var_34118_to_fp16 = const()[name = tensor("op_34118_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_311_cast_fp16 = layer_norm(axes = hidden_states_311_axes_0, beta = hidden_states_311_beta_0_to_fp16, epsilon = var_34118_to_fp16, gamma = hidden_states_311_gamma_0_to_fp16, x = inputs_229_cast_fp16)[name = tensor("hidden_states_311_cast_fp16")]; + tensor var_34133 = const()[name = tensor("op_34133"), val = tensor([1, 1])]; + tensor var_34135 = const()[name = tensor("op_34135"), val = tensor([1, 1])]; + tensor q_153_pad_type_0 = const()[name = tensor("q_153_pad_type_0"), val = tensor("custom")]; + tensor q_153_pad_0 = const()[name = tensor("q_153_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_4_attn1_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_4_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2866478464)))]; + tensor q_153_cast_fp16 = conv(dilations = var_34135, groups = var_30355, pad = q_153_pad_0, pad_type = q_153_pad_type_0, strides = var_34133, weight = up_blocks_0_attentions_0_transformer_blocks_4_attn1_to_q_weight_to_fp16, x = hidden_states_311_cast_fp16)[name = tensor("q_153_cast_fp16")]; + tensor var_34139 = const()[name = tensor("op_34139"), val = tensor([1, 1])]; + tensor var_34141 = const()[name = tensor("op_34141"), val = tensor([1, 1])]; + tensor k_305_pad_type_0 = const()[name = tensor("k_305_pad_type_0"), val = tensor("custom")]; + tensor k_305_pad_0 = const()[name = tensor("k_305_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_4_attn1_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_4_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2869755328)))]; + tensor k_305_cast_fp16 = conv(dilations = var_34141, groups = var_30355, pad = k_305_pad_0, pad_type = k_305_pad_type_0, strides = var_34139, weight = up_blocks_0_attentions_0_transformer_blocks_4_attn1_to_k_weight_to_fp16, x = hidden_states_311_cast_fp16)[name = tensor("k_305_cast_fp16")]; + tensor var_34145 = const()[name = tensor("op_34145"), val = tensor([1, 1])]; + tensor var_34147 = const()[name = tensor("op_34147"), val = tensor([1, 1])]; + tensor v_153_pad_type_0 = const()[name = tensor("v_153_pad_type_0"), val = tensor("custom")]; + tensor v_153_pad_0 = const()[name = tensor("v_153_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_4_attn1_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_4_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2873032192)))]; + tensor v_153_cast_fp16 = conv(dilations = var_34147, groups = var_30355, pad = v_153_pad_0, pad_type = v_153_pad_type_0, strides = var_34145, weight = up_blocks_0_attentions_0_transformer_blocks_4_attn1_to_v_weight_to_fp16, x = hidden_states_311_cast_fp16)[name = tensor("v_153_cast_fp16")]; + tensor var_34151_begin_0 = const()[name = tensor("op_34151_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_34151_end_0 = const()[name = tensor("op_34151_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_34151_end_mask_0 = const()[name = tensor("op_34151_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34151_cast_fp16 = slice_by_index(begin = var_34151_begin_0, end = var_34151_end_0, end_mask = var_34151_end_mask_0, x = q_153_cast_fp16)[name = tensor("op_34151_cast_fp16")]; + tensor var_34155_begin_0 = const()[name = tensor("op_34155_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_34155_end_0 = const()[name = tensor("op_34155_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_34155_end_mask_0 = const()[name = tensor("op_34155_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34155_cast_fp16 = slice_by_index(begin = var_34155_begin_0, end = var_34155_end_0, end_mask = var_34155_end_mask_0, x = q_153_cast_fp16)[name = tensor("op_34155_cast_fp16")]; + tensor var_34159_begin_0 = const()[name = tensor("op_34159_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_34159_end_0 = const()[name = tensor("op_34159_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_34159_end_mask_0 = const()[name = tensor("op_34159_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34159_cast_fp16 = slice_by_index(begin = var_34159_begin_0, end = var_34159_end_0, end_mask = var_34159_end_mask_0, x = q_153_cast_fp16)[name = tensor("op_34159_cast_fp16")]; + tensor var_34163_begin_0 = const()[name = tensor("op_34163_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_34163_end_0 = const()[name = tensor("op_34163_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_34163_end_mask_0 = const()[name = tensor("op_34163_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34163_cast_fp16 = slice_by_index(begin = var_34163_begin_0, end = var_34163_end_0, end_mask = var_34163_end_mask_0, x = q_153_cast_fp16)[name = tensor("op_34163_cast_fp16")]; + tensor var_34167_begin_0 = const()[name = tensor("op_34167_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_34167_end_0 = const()[name = tensor("op_34167_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_34167_end_mask_0 = const()[name = tensor("op_34167_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34167_cast_fp16 = slice_by_index(begin = var_34167_begin_0, end = var_34167_end_0, end_mask = var_34167_end_mask_0, x = q_153_cast_fp16)[name = tensor("op_34167_cast_fp16")]; + tensor var_34171_begin_0 = const()[name = tensor("op_34171_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_34171_end_0 = const()[name = tensor("op_34171_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_34171_end_mask_0 = const()[name = tensor("op_34171_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34171_cast_fp16 = slice_by_index(begin = var_34171_begin_0, end = var_34171_end_0, end_mask = var_34171_end_mask_0, x = q_153_cast_fp16)[name = tensor("op_34171_cast_fp16")]; + tensor var_34175_begin_0 = const()[name = tensor("op_34175_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_34175_end_0 = const()[name = tensor("op_34175_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_34175_end_mask_0 = const()[name = tensor("op_34175_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34175_cast_fp16 = slice_by_index(begin = var_34175_begin_0, end = var_34175_end_0, end_mask = var_34175_end_mask_0, x = q_153_cast_fp16)[name = tensor("op_34175_cast_fp16")]; + tensor var_34179_begin_0 = const()[name = tensor("op_34179_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_34179_end_0 = const()[name = tensor("op_34179_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_34179_end_mask_0 = const()[name = tensor("op_34179_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34179_cast_fp16 = slice_by_index(begin = var_34179_begin_0, end = var_34179_end_0, end_mask = var_34179_end_mask_0, x = q_153_cast_fp16)[name = tensor("op_34179_cast_fp16")]; + tensor var_34183_begin_0 = const()[name = tensor("op_34183_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_34183_end_0 = const()[name = tensor("op_34183_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_34183_end_mask_0 = const()[name = tensor("op_34183_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34183_cast_fp16 = slice_by_index(begin = var_34183_begin_0, end = var_34183_end_0, end_mask = var_34183_end_mask_0, x = q_153_cast_fp16)[name = tensor("op_34183_cast_fp16")]; + tensor var_34187_begin_0 = const()[name = tensor("op_34187_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_34187_end_0 = const()[name = tensor("op_34187_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_34187_end_mask_0 = const()[name = tensor("op_34187_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34187_cast_fp16 = slice_by_index(begin = var_34187_begin_0, end = var_34187_end_0, end_mask = var_34187_end_mask_0, x = q_153_cast_fp16)[name = tensor("op_34187_cast_fp16")]; + tensor var_34191_begin_0 = const()[name = tensor("op_34191_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_34191_end_0 = const()[name = tensor("op_34191_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_34191_end_mask_0 = const()[name = tensor("op_34191_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34191_cast_fp16 = slice_by_index(begin = var_34191_begin_0, end = var_34191_end_0, end_mask = var_34191_end_mask_0, x = q_153_cast_fp16)[name = tensor("op_34191_cast_fp16")]; + tensor var_34195_begin_0 = const()[name = tensor("op_34195_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_34195_end_0 = const()[name = tensor("op_34195_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_34195_end_mask_0 = const()[name = tensor("op_34195_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34195_cast_fp16 = slice_by_index(begin = var_34195_begin_0, end = var_34195_end_0, end_mask = var_34195_end_mask_0, x = q_153_cast_fp16)[name = tensor("op_34195_cast_fp16")]; + tensor var_34199_begin_0 = const()[name = tensor("op_34199_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_34199_end_0 = const()[name = tensor("op_34199_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_34199_end_mask_0 = const()[name = tensor("op_34199_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34199_cast_fp16 = slice_by_index(begin = var_34199_begin_0, end = var_34199_end_0, end_mask = var_34199_end_mask_0, x = q_153_cast_fp16)[name = tensor("op_34199_cast_fp16")]; + tensor var_34203_begin_0 = const()[name = tensor("op_34203_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_34203_end_0 = const()[name = tensor("op_34203_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_34203_end_mask_0 = const()[name = tensor("op_34203_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34203_cast_fp16 = slice_by_index(begin = var_34203_begin_0, end = var_34203_end_0, end_mask = var_34203_end_mask_0, x = q_153_cast_fp16)[name = tensor("op_34203_cast_fp16")]; + tensor var_34207_begin_0 = const()[name = tensor("op_34207_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_34207_end_0 = const()[name = tensor("op_34207_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_34207_end_mask_0 = const()[name = tensor("op_34207_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34207_cast_fp16 = slice_by_index(begin = var_34207_begin_0, end = var_34207_end_0, end_mask = var_34207_end_mask_0, x = q_153_cast_fp16)[name = tensor("op_34207_cast_fp16")]; + tensor var_34211_begin_0 = const()[name = tensor("op_34211_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_34211_end_0 = const()[name = tensor("op_34211_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_34211_end_mask_0 = const()[name = tensor("op_34211_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34211_cast_fp16 = slice_by_index(begin = var_34211_begin_0, end = var_34211_end_0, end_mask = var_34211_end_mask_0, x = q_153_cast_fp16)[name = tensor("op_34211_cast_fp16")]; + tensor var_34215_begin_0 = const()[name = tensor("op_34215_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_34215_end_0 = const()[name = tensor("op_34215_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_34215_end_mask_0 = const()[name = tensor("op_34215_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34215_cast_fp16 = slice_by_index(begin = var_34215_begin_0, end = var_34215_end_0, end_mask = var_34215_end_mask_0, x = q_153_cast_fp16)[name = tensor("op_34215_cast_fp16")]; + tensor var_34219_begin_0 = const()[name = tensor("op_34219_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_34219_end_0 = const()[name = tensor("op_34219_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_34219_end_mask_0 = const()[name = tensor("op_34219_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34219_cast_fp16 = slice_by_index(begin = var_34219_begin_0, end = var_34219_end_0, end_mask = var_34219_end_mask_0, x = q_153_cast_fp16)[name = tensor("op_34219_cast_fp16")]; + tensor var_34223_begin_0 = const()[name = tensor("op_34223_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_34223_end_0 = const()[name = tensor("op_34223_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_34223_end_mask_0 = const()[name = tensor("op_34223_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34223_cast_fp16 = slice_by_index(begin = var_34223_begin_0, end = var_34223_end_0, end_mask = var_34223_end_mask_0, x = q_153_cast_fp16)[name = tensor("op_34223_cast_fp16")]; + tensor var_34227_begin_0 = const()[name = tensor("op_34227_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_34227_end_0 = const()[name = tensor("op_34227_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_34227_end_mask_0 = const()[name = tensor("op_34227_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34227_cast_fp16 = slice_by_index(begin = var_34227_begin_0, end = var_34227_end_0, end_mask = var_34227_end_mask_0, x = q_153_cast_fp16)[name = tensor("op_34227_cast_fp16")]; + tensor k_307_perm_0 = const()[name = tensor("k_307_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_34234_begin_0 = const()[name = tensor("op_34234_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_34234_end_0 = const()[name = tensor("op_34234_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_34234_end_mask_0 = const()[name = tensor("op_34234_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_307_cast_fp16 = transpose(perm = k_307_perm_0, x = k_305_cast_fp16)[name = tensor("transpose_63")]; + tensor var_34234_cast_fp16 = slice_by_index(begin = var_34234_begin_0, end = var_34234_end_0, end_mask = var_34234_end_mask_0, x = k_307_cast_fp16)[name = tensor("op_34234_cast_fp16")]; + tensor var_34238_begin_0 = const()[name = tensor("op_34238_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_34238_end_0 = const()[name = tensor("op_34238_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_34238_end_mask_0 = const()[name = tensor("op_34238_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34238_cast_fp16 = slice_by_index(begin = var_34238_begin_0, end = var_34238_end_0, end_mask = var_34238_end_mask_0, x = k_307_cast_fp16)[name = tensor("op_34238_cast_fp16")]; + tensor var_34242_begin_0 = const()[name = tensor("op_34242_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_34242_end_0 = const()[name = tensor("op_34242_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_34242_end_mask_0 = const()[name = tensor("op_34242_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34242_cast_fp16 = slice_by_index(begin = var_34242_begin_0, end = var_34242_end_0, end_mask = var_34242_end_mask_0, x = k_307_cast_fp16)[name = tensor("op_34242_cast_fp16")]; + tensor var_34246_begin_0 = const()[name = tensor("op_34246_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_34246_end_0 = const()[name = tensor("op_34246_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_34246_end_mask_0 = const()[name = tensor("op_34246_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34246_cast_fp16 = slice_by_index(begin = var_34246_begin_0, end = var_34246_end_0, end_mask = var_34246_end_mask_0, x = k_307_cast_fp16)[name = tensor("op_34246_cast_fp16")]; + tensor var_34250_begin_0 = const()[name = tensor("op_34250_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_34250_end_0 = const()[name = tensor("op_34250_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_34250_end_mask_0 = const()[name = tensor("op_34250_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34250_cast_fp16 = slice_by_index(begin = var_34250_begin_0, end = var_34250_end_0, end_mask = var_34250_end_mask_0, x = k_307_cast_fp16)[name = tensor("op_34250_cast_fp16")]; + tensor var_34254_begin_0 = const()[name = tensor("op_34254_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_34254_end_0 = const()[name = tensor("op_34254_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_34254_end_mask_0 = const()[name = tensor("op_34254_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34254_cast_fp16 = slice_by_index(begin = var_34254_begin_0, end = var_34254_end_0, end_mask = var_34254_end_mask_0, x = k_307_cast_fp16)[name = tensor("op_34254_cast_fp16")]; + tensor var_34258_begin_0 = const()[name = tensor("op_34258_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_34258_end_0 = const()[name = tensor("op_34258_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_34258_end_mask_0 = const()[name = tensor("op_34258_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34258_cast_fp16 = slice_by_index(begin = var_34258_begin_0, end = var_34258_end_0, end_mask = var_34258_end_mask_0, x = k_307_cast_fp16)[name = tensor("op_34258_cast_fp16")]; + tensor var_34262_begin_0 = const()[name = tensor("op_34262_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_34262_end_0 = const()[name = tensor("op_34262_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_34262_end_mask_0 = const()[name = tensor("op_34262_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34262_cast_fp16 = slice_by_index(begin = var_34262_begin_0, end = var_34262_end_0, end_mask = var_34262_end_mask_0, x = k_307_cast_fp16)[name = tensor("op_34262_cast_fp16")]; + tensor var_34266_begin_0 = const()[name = tensor("op_34266_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_34266_end_0 = const()[name = tensor("op_34266_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_34266_end_mask_0 = const()[name = tensor("op_34266_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34266_cast_fp16 = slice_by_index(begin = var_34266_begin_0, end = var_34266_end_0, end_mask = var_34266_end_mask_0, x = k_307_cast_fp16)[name = tensor("op_34266_cast_fp16")]; + tensor var_34270_begin_0 = const()[name = tensor("op_34270_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_34270_end_0 = const()[name = tensor("op_34270_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_34270_end_mask_0 = const()[name = tensor("op_34270_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34270_cast_fp16 = slice_by_index(begin = var_34270_begin_0, end = var_34270_end_0, end_mask = var_34270_end_mask_0, x = k_307_cast_fp16)[name = tensor("op_34270_cast_fp16")]; + tensor var_34274_begin_0 = const()[name = tensor("op_34274_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_34274_end_0 = const()[name = tensor("op_34274_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_34274_end_mask_0 = const()[name = tensor("op_34274_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34274_cast_fp16 = slice_by_index(begin = var_34274_begin_0, end = var_34274_end_0, end_mask = var_34274_end_mask_0, x = k_307_cast_fp16)[name = tensor("op_34274_cast_fp16")]; + tensor var_34278_begin_0 = const()[name = tensor("op_34278_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_34278_end_0 = const()[name = tensor("op_34278_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_34278_end_mask_0 = const()[name = tensor("op_34278_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34278_cast_fp16 = slice_by_index(begin = var_34278_begin_0, end = var_34278_end_0, end_mask = var_34278_end_mask_0, x = k_307_cast_fp16)[name = tensor("op_34278_cast_fp16")]; + tensor var_34282_begin_0 = const()[name = tensor("op_34282_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_34282_end_0 = const()[name = tensor("op_34282_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_34282_end_mask_0 = const()[name = tensor("op_34282_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34282_cast_fp16 = slice_by_index(begin = var_34282_begin_0, end = var_34282_end_0, end_mask = var_34282_end_mask_0, x = k_307_cast_fp16)[name = tensor("op_34282_cast_fp16")]; + tensor var_34286_begin_0 = const()[name = tensor("op_34286_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_34286_end_0 = const()[name = tensor("op_34286_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_34286_end_mask_0 = const()[name = tensor("op_34286_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34286_cast_fp16 = slice_by_index(begin = var_34286_begin_0, end = var_34286_end_0, end_mask = var_34286_end_mask_0, x = k_307_cast_fp16)[name = tensor("op_34286_cast_fp16")]; + tensor var_34290_begin_0 = const()[name = tensor("op_34290_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_34290_end_0 = const()[name = tensor("op_34290_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_34290_end_mask_0 = const()[name = tensor("op_34290_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34290_cast_fp16 = slice_by_index(begin = var_34290_begin_0, end = var_34290_end_0, end_mask = var_34290_end_mask_0, x = k_307_cast_fp16)[name = tensor("op_34290_cast_fp16")]; + tensor var_34294_begin_0 = const()[name = tensor("op_34294_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_34294_end_0 = const()[name = tensor("op_34294_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_34294_end_mask_0 = const()[name = tensor("op_34294_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34294_cast_fp16 = slice_by_index(begin = var_34294_begin_0, end = var_34294_end_0, end_mask = var_34294_end_mask_0, x = k_307_cast_fp16)[name = tensor("op_34294_cast_fp16")]; + tensor var_34298_begin_0 = const()[name = tensor("op_34298_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_34298_end_0 = const()[name = tensor("op_34298_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_34298_end_mask_0 = const()[name = tensor("op_34298_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34298_cast_fp16 = slice_by_index(begin = var_34298_begin_0, end = var_34298_end_0, end_mask = var_34298_end_mask_0, x = k_307_cast_fp16)[name = tensor("op_34298_cast_fp16")]; + tensor var_34302_begin_0 = const()[name = tensor("op_34302_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_34302_end_0 = const()[name = tensor("op_34302_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_34302_end_mask_0 = const()[name = tensor("op_34302_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34302_cast_fp16 = slice_by_index(begin = var_34302_begin_0, end = var_34302_end_0, end_mask = var_34302_end_mask_0, x = k_307_cast_fp16)[name = tensor("op_34302_cast_fp16")]; + tensor var_34306_begin_0 = const()[name = tensor("op_34306_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_34306_end_0 = const()[name = tensor("op_34306_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_34306_end_mask_0 = const()[name = tensor("op_34306_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34306_cast_fp16 = slice_by_index(begin = var_34306_begin_0, end = var_34306_end_0, end_mask = var_34306_end_mask_0, x = k_307_cast_fp16)[name = tensor("op_34306_cast_fp16")]; + tensor var_34310_begin_0 = const()[name = tensor("op_34310_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_34310_end_0 = const()[name = tensor("op_34310_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_34310_end_mask_0 = const()[name = tensor("op_34310_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34310_cast_fp16 = slice_by_index(begin = var_34310_begin_0, end = var_34310_end_0, end_mask = var_34310_end_mask_0, x = k_307_cast_fp16)[name = tensor("op_34310_cast_fp16")]; + tensor var_34312_begin_0 = const()[name = tensor("op_34312_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_34312_end_0 = const()[name = tensor("op_34312_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_34312_end_mask_0 = const()[name = tensor("op_34312_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34312_cast_fp16 = slice_by_index(begin = var_34312_begin_0, end = var_34312_end_0, end_mask = var_34312_end_mask_0, x = v_153_cast_fp16)[name = tensor("op_34312_cast_fp16")]; + tensor var_34316_begin_0 = const()[name = tensor("op_34316_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_34316_end_0 = const()[name = tensor("op_34316_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_34316_end_mask_0 = const()[name = tensor("op_34316_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34316_cast_fp16 = slice_by_index(begin = var_34316_begin_0, end = var_34316_end_0, end_mask = var_34316_end_mask_0, x = v_153_cast_fp16)[name = tensor("op_34316_cast_fp16")]; + tensor var_34320_begin_0 = const()[name = tensor("op_34320_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_34320_end_0 = const()[name = tensor("op_34320_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_34320_end_mask_0 = const()[name = tensor("op_34320_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34320_cast_fp16 = slice_by_index(begin = var_34320_begin_0, end = var_34320_end_0, end_mask = var_34320_end_mask_0, x = v_153_cast_fp16)[name = tensor("op_34320_cast_fp16")]; + tensor var_34324_begin_0 = const()[name = tensor("op_34324_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_34324_end_0 = const()[name = tensor("op_34324_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_34324_end_mask_0 = const()[name = tensor("op_34324_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34324_cast_fp16 = slice_by_index(begin = var_34324_begin_0, end = var_34324_end_0, end_mask = var_34324_end_mask_0, x = v_153_cast_fp16)[name = tensor("op_34324_cast_fp16")]; + tensor var_34328_begin_0 = const()[name = tensor("op_34328_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_34328_end_0 = const()[name = tensor("op_34328_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_34328_end_mask_0 = const()[name = tensor("op_34328_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34328_cast_fp16 = slice_by_index(begin = var_34328_begin_0, end = var_34328_end_0, end_mask = var_34328_end_mask_0, x = v_153_cast_fp16)[name = tensor("op_34328_cast_fp16")]; + tensor var_34332_begin_0 = const()[name = tensor("op_34332_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_34332_end_0 = const()[name = tensor("op_34332_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_34332_end_mask_0 = const()[name = tensor("op_34332_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34332_cast_fp16 = slice_by_index(begin = var_34332_begin_0, end = var_34332_end_0, end_mask = var_34332_end_mask_0, x = v_153_cast_fp16)[name = tensor("op_34332_cast_fp16")]; + tensor var_34336_begin_0 = const()[name = tensor("op_34336_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_34336_end_0 = const()[name = tensor("op_34336_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_34336_end_mask_0 = const()[name = tensor("op_34336_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34336_cast_fp16 = slice_by_index(begin = var_34336_begin_0, end = var_34336_end_0, end_mask = var_34336_end_mask_0, x = v_153_cast_fp16)[name = tensor("op_34336_cast_fp16")]; + tensor var_34340_begin_0 = const()[name = tensor("op_34340_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_34340_end_0 = const()[name = tensor("op_34340_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_34340_end_mask_0 = const()[name = tensor("op_34340_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34340_cast_fp16 = slice_by_index(begin = var_34340_begin_0, end = var_34340_end_0, end_mask = var_34340_end_mask_0, x = v_153_cast_fp16)[name = tensor("op_34340_cast_fp16")]; + tensor var_34344_begin_0 = const()[name = tensor("op_34344_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_34344_end_0 = const()[name = tensor("op_34344_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_34344_end_mask_0 = const()[name = tensor("op_34344_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34344_cast_fp16 = slice_by_index(begin = var_34344_begin_0, end = var_34344_end_0, end_mask = var_34344_end_mask_0, x = v_153_cast_fp16)[name = tensor("op_34344_cast_fp16")]; + tensor var_34348_begin_0 = const()[name = tensor("op_34348_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_34348_end_0 = const()[name = tensor("op_34348_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_34348_end_mask_0 = const()[name = tensor("op_34348_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34348_cast_fp16 = slice_by_index(begin = var_34348_begin_0, end = var_34348_end_0, end_mask = var_34348_end_mask_0, x = v_153_cast_fp16)[name = tensor("op_34348_cast_fp16")]; + tensor var_34352_begin_0 = const()[name = tensor("op_34352_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_34352_end_0 = const()[name = tensor("op_34352_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_34352_end_mask_0 = const()[name = tensor("op_34352_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34352_cast_fp16 = slice_by_index(begin = var_34352_begin_0, end = var_34352_end_0, end_mask = var_34352_end_mask_0, x = v_153_cast_fp16)[name = tensor("op_34352_cast_fp16")]; + tensor var_34356_begin_0 = const()[name = tensor("op_34356_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_34356_end_0 = const()[name = tensor("op_34356_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_34356_end_mask_0 = const()[name = tensor("op_34356_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34356_cast_fp16 = slice_by_index(begin = var_34356_begin_0, end = var_34356_end_0, end_mask = var_34356_end_mask_0, x = v_153_cast_fp16)[name = tensor("op_34356_cast_fp16")]; + tensor var_34360_begin_0 = const()[name = tensor("op_34360_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_34360_end_0 = const()[name = tensor("op_34360_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_34360_end_mask_0 = const()[name = tensor("op_34360_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34360_cast_fp16 = slice_by_index(begin = var_34360_begin_0, end = var_34360_end_0, end_mask = var_34360_end_mask_0, x = v_153_cast_fp16)[name = tensor("op_34360_cast_fp16")]; + tensor var_34364_begin_0 = const()[name = tensor("op_34364_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_34364_end_0 = const()[name = tensor("op_34364_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_34364_end_mask_0 = const()[name = tensor("op_34364_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34364_cast_fp16 = slice_by_index(begin = var_34364_begin_0, end = var_34364_end_0, end_mask = var_34364_end_mask_0, x = v_153_cast_fp16)[name = tensor("op_34364_cast_fp16")]; + tensor var_34368_begin_0 = const()[name = tensor("op_34368_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_34368_end_0 = const()[name = tensor("op_34368_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_34368_end_mask_0 = const()[name = tensor("op_34368_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34368_cast_fp16 = slice_by_index(begin = var_34368_begin_0, end = var_34368_end_0, end_mask = var_34368_end_mask_0, x = v_153_cast_fp16)[name = tensor("op_34368_cast_fp16")]; + tensor var_34372_begin_0 = const()[name = tensor("op_34372_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_34372_end_0 = const()[name = tensor("op_34372_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_34372_end_mask_0 = const()[name = tensor("op_34372_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34372_cast_fp16 = slice_by_index(begin = var_34372_begin_0, end = var_34372_end_0, end_mask = var_34372_end_mask_0, x = v_153_cast_fp16)[name = tensor("op_34372_cast_fp16")]; + tensor var_34376_begin_0 = const()[name = tensor("op_34376_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_34376_end_0 = const()[name = tensor("op_34376_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_34376_end_mask_0 = const()[name = tensor("op_34376_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34376_cast_fp16 = slice_by_index(begin = var_34376_begin_0, end = var_34376_end_0, end_mask = var_34376_end_mask_0, x = v_153_cast_fp16)[name = tensor("op_34376_cast_fp16")]; + tensor var_34380_begin_0 = const()[name = tensor("op_34380_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_34380_end_0 = const()[name = tensor("op_34380_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_34380_end_mask_0 = const()[name = tensor("op_34380_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34380_cast_fp16 = slice_by_index(begin = var_34380_begin_0, end = var_34380_end_0, end_mask = var_34380_end_mask_0, x = v_153_cast_fp16)[name = tensor("op_34380_cast_fp16")]; + tensor var_34384_begin_0 = const()[name = tensor("op_34384_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_34384_end_0 = const()[name = tensor("op_34384_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_34384_end_mask_0 = const()[name = tensor("op_34384_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34384_cast_fp16 = slice_by_index(begin = var_34384_begin_0, end = var_34384_end_0, end_mask = var_34384_end_mask_0, x = v_153_cast_fp16)[name = tensor("op_34384_cast_fp16")]; + tensor var_34388_begin_0 = const()[name = tensor("op_34388_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_34388_end_0 = const()[name = tensor("op_34388_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_34388_end_mask_0 = const()[name = tensor("op_34388_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34388_cast_fp16 = slice_by_index(begin = var_34388_begin_0, end = var_34388_end_0, end_mask = var_34388_end_mask_0, x = v_153_cast_fp16)[name = tensor("op_34388_cast_fp16")]; + tensor var_34392_equation_0 = const()[name = tensor("op_34392_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34392_cast_fp16 = einsum(equation = var_34392_equation_0, values = (var_34234_cast_fp16, var_34151_cast_fp16))[name = tensor("op_34392_cast_fp16")]; + tensor var_34393_to_fp16 = const()[name = tensor("op_34393_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2881_cast_fp16 = mul(x = var_34392_cast_fp16, y = var_34393_to_fp16)[name = tensor("aw_2881_cast_fp16")]; + tensor var_34396_equation_0 = const()[name = tensor("op_34396_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34396_cast_fp16 = einsum(equation = var_34396_equation_0, values = (var_34238_cast_fp16, var_34155_cast_fp16))[name = tensor("op_34396_cast_fp16")]; + tensor var_34397_to_fp16 = const()[name = tensor("op_34397_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2883_cast_fp16 = mul(x = var_34396_cast_fp16, y = var_34397_to_fp16)[name = tensor("aw_2883_cast_fp16")]; + tensor var_34400_equation_0 = const()[name = tensor("op_34400_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34400_cast_fp16 = einsum(equation = var_34400_equation_0, values = (var_34242_cast_fp16, var_34159_cast_fp16))[name = tensor("op_34400_cast_fp16")]; + tensor var_34401_to_fp16 = const()[name = tensor("op_34401_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2885_cast_fp16 = mul(x = var_34400_cast_fp16, y = var_34401_to_fp16)[name = tensor("aw_2885_cast_fp16")]; + tensor var_34404_equation_0 = const()[name = tensor("op_34404_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34404_cast_fp16 = einsum(equation = var_34404_equation_0, values = (var_34246_cast_fp16, var_34163_cast_fp16))[name = tensor("op_34404_cast_fp16")]; + tensor var_34405_to_fp16 = const()[name = tensor("op_34405_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2887_cast_fp16 = mul(x = var_34404_cast_fp16, y = var_34405_to_fp16)[name = tensor("aw_2887_cast_fp16")]; + tensor var_34408_equation_0 = const()[name = tensor("op_34408_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34408_cast_fp16 = einsum(equation = var_34408_equation_0, values = (var_34250_cast_fp16, var_34167_cast_fp16))[name = tensor("op_34408_cast_fp16")]; + tensor var_34409_to_fp16 = const()[name = tensor("op_34409_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2889_cast_fp16 = mul(x = var_34408_cast_fp16, y = var_34409_to_fp16)[name = tensor("aw_2889_cast_fp16")]; + tensor var_34412_equation_0 = const()[name = tensor("op_34412_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34412_cast_fp16 = einsum(equation = var_34412_equation_0, values = (var_34254_cast_fp16, var_34171_cast_fp16))[name = tensor("op_34412_cast_fp16")]; + tensor var_34413_to_fp16 = const()[name = tensor("op_34413_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2891_cast_fp16 = mul(x = var_34412_cast_fp16, y = var_34413_to_fp16)[name = tensor("aw_2891_cast_fp16")]; + tensor var_34416_equation_0 = const()[name = tensor("op_34416_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34416_cast_fp16 = einsum(equation = var_34416_equation_0, values = (var_34258_cast_fp16, var_34175_cast_fp16))[name = tensor("op_34416_cast_fp16")]; + tensor var_34417_to_fp16 = const()[name = tensor("op_34417_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2893_cast_fp16 = mul(x = var_34416_cast_fp16, y = var_34417_to_fp16)[name = tensor("aw_2893_cast_fp16")]; + tensor var_34420_equation_0 = const()[name = tensor("op_34420_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34420_cast_fp16 = einsum(equation = var_34420_equation_0, values = (var_34262_cast_fp16, var_34179_cast_fp16))[name = tensor("op_34420_cast_fp16")]; + tensor var_34421_to_fp16 = const()[name = tensor("op_34421_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2895_cast_fp16 = mul(x = var_34420_cast_fp16, y = var_34421_to_fp16)[name = tensor("aw_2895_cast_fp16")]; + tensor var_34424_equation_0 = const()[name = tensor("op_34424_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34424_cast_fp16 = einsum(equation = var_34424_equation_0, values = (var_34266_cast_fp16, var_34183_cast_fp16))[name = tensor("op_34424_cast_fp16")]; + tensor var_34425_to_fp16 = const()[name = tensor("op_34425_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2897_cast_fp16 = mul(x = var_34424_cast_fp16, y = var_34425_to_fp16)[name = tensor("aw_2897_cast_fp16")]; + tensor var_34428_equation_0 = const()[name = tensor("op_34428_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34428_cast_fp16 = einsum(equation = var_34428_equation_0, values = (var_34270_cast_fp16, var_34187_cast_fp16))[name = tensor("op_34428_cast_fp16")]; + tensor var_34429_to_fp16 = const()[name = tensor("op_34429_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2899_cast_fp16 = mul(x = var_34428_cast_fp16, y = var_34429_to_fp16)[name = tensor("aw_2899_cast_fp16")]; + tensor var_34432_equation_0 = const()[name = tensor("op_34432_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34432_cast_fp16 = einsum(equation = var_34432_equation_0, values = (var_34274_cast_fp16, var_34191_cast_fp16))[name = tensor("op_34432_cast_fp16")]; + tensor var_34433_to_fp16 = const()[name = tensor("op_34433_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2901_cast_fp16 = mul(x = var_34432_cast_fp16, y = var_34433_to_fp16)[name = tensor("aw_2901_cast_fp16")]; + tensor var_34436_equation_0 = const()[name = tensor("op_34436_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34436_cast_fp16 = einsum(equation = var_34436_equation_0, values = (var_34278_cast_fp16, var_34195_cast_fp16))[name = tensor("op_34436_cast_fp16")]; + tensor var_34437_to_fp16 = const()[name = tensor("op_34437_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2903_cast_fp16 = mul(x = var_34436_cast_fp16, y = var_34437_to_fp16)[name = tensor("aw_2903_cast_fp16")]; + tensor var_34440_equation_0 = const()[name = tensor("op_34440_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34440_cast_fp16 = einsum(equation = var_34440_equation_0, values = (var_34282_cast_fp16, var_34199_cast_fp16))[name = tensor("op_34440_cast_fp16")]; + tensor var_34441_to_fp16 = const()[name = tensor("op_34441_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2905_cast_fp16 = mul(x = var_34440_cast_fp16, y = var_34441_to_fp16)[name = tensor("aw_2905_cast_fp16")]; + tensor var_34444_equation_0 = const()[name = tensor("op_34444_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34444_cast_fp16 = einsum(equation = var_34444_equation_0, values = (var_34286_cast_fp16, var_34203_cast_fp16))[name = tensor("op_34444_cast_fp16")]; + tensor var_34445_to_fp16 = const()[name = tensor("op_34445_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2907_cast_fp16 = mul(x = var_34444_cast_fp16, y = var_34445_to_fp16)[name = tensor("aw_2907_cast_fp16")]; + tensor var_34448_equation_0 = const()[name = tensor("op_34448_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34448_cast_fp16 = einsum(equation = var_34448_equation_0, values = (var_34290_cast_fp16, var_34207_cast_fp16))[name = tensor("op_34448_cast_fp16")]; + tensor var_34449_to_fp16 = const()[name = tensor("op_34449_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2909_cast_fp16 = mul(x = var_34448_cast_fp16, y = var_34449_to_fp16)[name = tensor("aw_2909_cast_fp16")]; + tensor var_34452_equation_0 = const()[name = tensor("op_34452_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34452_cast_fp16 = einsum(equation = var_34452_equation_0, values = (var_34294_cast_fp16, var_34211_cast_fp16))[name = tensor("op_34452_cast_fp16")]; + tensor var_34453_to_fp16 = const()[name = tensor("op_34453_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2911_cast_fp16 = mul(x = var_34452_cast_fp16, y = var_34453_to_fp16)[name = tensor("aw_2911_cast_fp16")]; + tensor var_34456_equation_0 = const()[name = tensor("op_34456_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34456_cast_fp16 = einsum(equation = var_34456_equation_0, values = (var_34298_cast_fp16, var_34215_cast_fp16))[name = tensor("op_34456_cast_fp16")]; + tensor var_34457_to_fp16 = const()[name = tensor("op_34457_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2913_cast_fp16 = mul(x = var_34456_cast_fp16, y = var_34457_to_fp16)[name = tensor("aw_2913_cast_fp16")]; + tensor var_34460_equation_0 = const()[name = tensor("op_34460_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34460_cast_fp16 = einsum(equation = var_34460_equation_0, values = (var_34302_cast_fp16, var_34219_cast_fp16))[name = tensor("op_34460_cast_fp16")]; + tensor var_34461_to_fp16 = const()[name = tensor("op_34461_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2915_cast_fp16 = mul(x = var_34460_cast_fp16, y = var_34461_to_fp16)[name = tensor("aw_2915_cast_fp16")]; + tensor var_34464_equation_0 = const()[name = tensor("op_34464_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34464_cast_fp16 = einsum(equation = var_34464_equation_0, values = (var_34306_cast_fp16, var_34223_cast_fp16))[name = tensor("op_34464_cast_fp16")]; + tensor var_34465_to_fp16 = const()[name = tensor("op_34465_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2917_cast_fp16 = mul(x = var_34464_cast_fp16, y = var_34465_to_fp16)[name = tensor("aw_2917_cast_fp16")]; + tensor var_34468_equation_0 = const()[name = tensor("op_34468_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34468_cast_fp16 = einsum(equation = var_34468_equation_0, values = (var_34310_cast_fp16, var_34227_cast_fp16))[name = tensor("op_34468_cast_fp16")]; + tensor var_34469_to_fp16 = const()[name = tensor("op_34469_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2919_cast_fp16 = mul(x = var_34468_cast_fp16, y = var_34469_to_fp16)[name = tensor("aw_2919_cast_fp16")]; + tensor var_34471_cast_fp16 = softmax(axis = var_30355, x = aw_2881_cast_fp16)[name = tensor("op_34471_cast_fp16")]; + tensor var_34472_cast_fp16 = softmax(axis = var_30355, x = aw_2883_cast_fp16)[name = tensor("op_34472_cast_fp16")]; + tensor var_34473_cast_fp16 = softmax(axis = var_30355, x = aw_2885_cast_fp16)[name = tensor("op_34473_cast_fp16")]; + tensor var_34474_cast_fp16 = softmax(axis = var_30355, x = aw_2887_cast_fp16)[name = tensor("op_34474_cast_fp16")]; + tensor var_34475_cast_fp16 = softmax(axis = var_30355, x = aw_2889_cast_fp16)[name = tensor("op_34475_cast_fp16")]; + tensor var_34476_cast_fp16 = softmax(axis = var_30355, x = aw_2891_cast_fp16)[name = tensor("op_34476_cast_fp16")]; + tensor var_34477_cast_fp16 = softmax(axis = var_30355, x = aw_2893_cast_fp16)[name = tensor("op_34477_cast_fp16")]; + tensor var_34478_cast_fp16 = softmax(axis = var_30355, x = aw_2895_cast_fp16)[name = tensor("op_34478_cast_fp16")]; + tensor var_34479_cast_fp16 = softmax(axis = var_30355, x = aw_2897_cast_fp16)[name = tensor("op_34479_cast_fp16")]; + tensor var_34480_cast_fp16 = softmax(axis = var_30355, x = aw_2899_cast_fp16)[name = tensor("op_34480_cast_fp16")]; + tensor var_34481_cast_fp16 = softmax(axis = var_30355, x = aw_2901_cast_fp16)[name = tensor("op_34481_cast_fp16")]; + tensor var_34482_cast_fp16 = softmax(axis = var_30355, x = aw_2903_cast_fp16)[name = tensor("op_34482_cast_fp16")]; + tensor var_34483_cast_fp16 = softmax(axis = var_30355, x = aw_2905_cast_fp16)[name = tensor("op_34483_cast_fp16")]; + tensor var_34484_cast_fp16 = softmax(axis = var_30355, x = aw_2907_cast_fp16)[name = tensor("op_34484_cast_fp16")]; + tensor var_34485_cast_fp16 = softmax(axis = var_30355, x = aw_2909_cast_fp16)[name = tensor("op_34485_cast_fp16")]; + tensor var_34486_cast_fp16 = softmax(axis = var_30355, x = aw_2911_cast_fp16)[name = tensor("op_34486_cast_fp16")]; + tensor var_34487_cast_fp16 = softmax(axis = var_30355, x = aw_2913_cast_fp16)[name = tensor("op_34487_cast_fp16")]; + tensor var_34488_cast_fp16 = softmax(axis = var_30355, x = aw_2915_cast_fp16)[name = tensor("op_34488_cast_fp16")]; + tensor var_34489_cast_fp16 = softmax(axis = var_30355, x = aw_2917_cast_fp16)[name = tensor("op_34489_cast_fp16")]; + tensor var_34490_cast_fp16 = softmax(axis = var_30355, x = aw_2919_cast_fp16)[name = tensor("op_34490_cast_fp16")]; + tensor var_34492_equation_0 = const()[name = tensor("op_34492_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34492_cast_fp16 = einsum(equation = var_34492_equation_0, values = (var_34312_cast_fp16, var_34471_cast_fp16))[name = tensor("op_34492_cast_fp16")]; + tensor var_34494_equation_0 = const()[name = tensor("op_34494_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34494_cast_fp16 = einsum(equation = var_34494_equation_0, values = (var_34316_cast_fp16, var_34472_cast_fp16))[name = tensor("op_34494_cast_fp16")]; + tensor var_34496_equation_0 = const()[name = tensor("op_34496_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34496_cast_fp16 = einsum(equation = var_34496_equation_0, values = (var_34320_cast_fp16, var_34473_cast_fp16))[name = tensor("op_34496_cast_fp16")]; + tensor var_34498_equation_0 = const()[name = tensor("op_34498_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34498_cast_fp16 = einsum(equation = var_34498_equation_0, values = (var_34324_cast_fp16, var_34474_cast_fp16))[name = tensor("op_34498_cast_fp16")]; + tensor var_34500_equation_0 = const()[name = tensor("op_34500_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34500_cast_fp16 = einsum(equation = var_34500_equation_0, values = (var_34328_cast_fp16, var_34475_cast_fp16))[name = tensor("op_34500_cast_fp16")]; + tensor var_34502_equation_0 = const()[name = tensor("op_34502_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34502_cast_fp16 = einsum(equation = var_34502_equation_0, values = (var_34332_cast_fp16, var_34476_cast_fp16))[name = tensor("op_34502_cast_fp16")]; + tensor var_34504_equation_0 = const()[name = tensor("op_34504_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34504_cast_fp16 = einsum(equation = var_34504_equation_0, values = (var_34336_cast_fp16, var_34477_cast_fp16))[name = tensor("op_34504_cast_fp16")]; + tensor var_34506_equation_0 = const()[name = tensor("op_34506_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34506_cast_fp16 = einsum(equation = var_34506_equation_0, values = (var_34340_cast_fp16, var_34478_cast_fp16))[name = tensor("op_34506_cast_fp16")]; + tensor var_34508_equation_0 = const()[name = tensor("op_34508_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34508_cast_fp16 = einsum(equation = var_34508_equation_0, values = (var_34344_cast_fp16, var_34479_cast_fp16))[name = tensor("op_34508_cast_fp16")]; + tensor var_34510_equation_0 = const()[name = tensor("op_34510_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34510_cast_fp16 = einsum(equation = var_34510_equation_0, values = (var_34348_cast_fp16, var_34480_cast_fp16))[name = tensor("op_34510_cast_fp16")]; + tensor var_34512_equation_0 = const()[name = tensor("op_34512_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34512_cast_fp16 = einsum(equation = var_34512_equation_0, values = (var_34352_cast_fp16, var_34481_cast_fp16))[name = tensor("op_34512_cast_fp16")]; + tensor var_34514_equation_0 = const()[name = tensor("op_34514_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34514_cast_fp16 = einsum(equation = var_34514_equation_0, values = (var_34356_cast_fp16, var_34482_cast_fp16))[name = tensor("op_34514_cast_fp16")]; + tensor var_34516_equation_0 = const()[name = tensor("op_34516_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34516_cast_fp16 = einsum(equation = var_34516_equation_0, values = (var_34360_cast_fp16, var_34483_cast_fp16))[name = tensor("op_34516_cast_fp16")]; + tensor var_34518_equation_0 = const()[name = tensor("op_34518_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34518_cast_fp16 = einsum(equation = var_34518_equation_0, values = (var_34364_cast_fp16, var_34484_cast_fp16))[name = tensor("op_34518_cast_fp16")]; + tensor var_34520_equation_0 = const()[name = tensor("op_34520_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34520_cast_fp16 = einsum(equation = var_34520_equation_0, values = (var_34368_cast_fp16, var_34485_cast_fp16))[name = tensor("op_34520_cast_fp16")]; + tensor var_34522_equation_0 = const()[name = tensor("op_34522_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34522_cast_fp16 = einsum(equation = var_34522_equation_0, values = (var_34372_cast_fp16, var_34486_cast_fp16))[name = tensor("op_34522_cast_fp16")]; + tensor var_34524_equation_0 = const()[name = tensor("op_34524_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34524_cast_fp16 = einsum(equation = var_34524_equation_0, values = (var_34376_cast_fp16, var_34487_cast_fp16))[name = tensor("op_34524_cast_fp16")]; + tensor var_34526_equation_0 = const()[name = tensor("op_34526_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34526_cast_fp16 = einsum(equation = var_34526_equation_0, values = (var_34380_cast_fp16, var_34488_cast_fp16))[name = tensor("op_34526_cast_fp16")]; + tensor var_34528_equation_0 = const()[name = tensor("op_34528_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34528_cast_fp16 = einsum(equation = var_34528_equation_0, values = (var_34384_cast_fp16, var_34489_cast_fp16))[name = tensor("op_34528_cast_fp16")]; + tensor var_34530_equation_0 = const()[name = tensor("op_34530_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34530_cast_fp16 = einsum(equation = var_34530_equation_0, values = (var_34388_cast_fp16, var_34490_cast_fp16))[name = tensor("op_34530_cast_fp16")]; + tensor input_471_interleave_0 = const()[name = tensor("input_471_interleave_0"), val = tensor(false)]; + tensor input_471_cast_fp16 = concat(axis = var_30355, interleave = input_471_interleave_0, values = (var_34492_cast_fp16, var_34494_cast_fp16, var_34496_cast_fp16, var_34498_cast_fp16, var_34500_cast_fp16, var_34502_cast_fp16, var_34504_cast_fp16, var_34506_cast_fp16, var_34508_cast_fp16, var_34510_cast_fp16, var_34512_cast_fp16, var_34514_cast_fp16, var_34516_cast_fp16, var_34518_cast_fp16, var_34520_cast_fp16, var_34522_cast_fp16, var_34524_cast_fp16, var_34526_cast_fp16, var_34528_cast_fp16, var_34530_cast_fp16))[name = tensor("input_471_cast_fp16")]; + tensor var_34536 = const()[name = tensor("op_34536"), val = tensor([1, 1])]; + tensor var_34538 = const()[name = tensor("op_34538"), val = tensor([1, 1])]; + tensor var_34540_pad_type_0 = const()[name = tensor("op_34540_pad_type_0"), val = tensor("custom")]; + tensor var_34540_pad_0 = const()[name = tensor("op_34540_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_4_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_4_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2876309056)))]; + tensor up_blocks_0_attentions_0_transformer_blocks_4_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_4_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2879585920)))]; + tensor var_34540_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_4_attn1_to_out_0_bias_to_fp16, dilations = var_34538, groups = var_30355, pad = var_34540_pad_0, pad_type = var_34540_pad_type_0, strides = var_34536, weight = up_blocks_0_attentions_0_transformer_blocks_4_attn1_to_out_0_weight_to_fp16, x = input_471_cast_fp16)[name = tensor("op_34540_cast_fp16")]; + tensor inputs_231_cast_fp16 = add(x = var_34540_cast_fp16, y = inputs_229_cast_fp16)[name = tensor("inputs_231_cast_fp16")]; + tensor hidden_states_313_axes_0 = const()[name = tensor("hidden_states_313_axes_0"), val = tensor([1])]; + tensor hidden_states_313_gamma_0_to_fp16 = const()[name = tensor("hidden_states_313_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2879588544)))]; + tensor hidden_states_313_beta_0_to_fp16 = const()[name = tensor("hidden_states_313_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2879591168)))]; + tensor var_34550_to_fp16 = const()[name = tensor("op_34550_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_313_cast_fp16 = layer_norm(axes = hidden_states_313_axes_0, beta = hidden_states_313_beta_0_to_fp16, epsilon = var_34550_to_fp16, gamma = hidden_states_313_gamma_0_to_fp16, x = inputs_231_cast_fp16)[name = tensor("hidden_states_313_cast_fp16")]; + tensor var_34565 = const()[name = tensor("op_34565"), val = tensor([1, 1])]; + tensor var_34567 = const()[name = tensor("op_34567"), val = tensor([1, 1])]; + tensor q_155_pad_type_0 = const()[name = tensor("q_155_pad_type_0"), val = tensor("custom")]; + tensor q_155_pad_0 = const()[name = tensor("q_155_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_4_attn2_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_4_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2879593792)))]; + tensor q_155_cast_fp16 = conv(dilations = var_34567, groups = var_30355, pad = q_155_pad_0, pad_type = q_155_pad_type_0, strides = var_34565, weight = up_blocks_0_attentions_0_transformer_blocks_4_attn2_to_q_weight_to_fp16, x = hidden_states_313_cast_fp16)[name = tensor("q_155_cast_fp16")]; + tensor var_34571 = const()[name = tensor("op_34571"), val = tensor([1, 1])]; + tensor var_34573 = const()[name = tensor("op_34573"), val = tensor([1, 1])]; + tensor k_309_pad_type_0 = const()[name = tensor("k_309_pad_type_0"), val = tensor("custom")]; + tensor k_309_pad_0 = const()[name = tensor("k_309_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_4_attn2_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_4_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2882870656)))]; + tensor k_309_cast_fp16 = conv(dilations = var_34573, groups = var_30355, pad = k_309_pad_0, pad_type = k_309_pad_type_0, strides = var_34571, weight = up_blocks_0_attentions_0_transformer_blocks_4_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_309_cast_fp16")]; + tensor var_34577 = const()[name = tensor("op_34577"), val = tensor([1, 1])]; + tensor var_34579 = const()[name = tensor("op_34579"), val = tensor([1, 1])]; + tensor v_155_pad_type_0 = const()[name = tensor("v_155_pad_type_0"), val = tensor("custom")]; + tensor v_155_pad_0 = const()[name = tensor("v_155_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_4_attn2_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_4_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2888113600)))]; + tensor v_155_cast_fp16 = conv(dilations = var_34579, groups = var_30355, pad = v_155_pad_0, pad_type = v_155_pad_type_0, strides = var_34577, weight = up_blocks_0_attentions_0_transformer_blocks_4_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_155_cast_fp16")]; + tensor var_34583_begin_0 = const()[name = tensor("op_34583_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_34583_end_0 = const()[name = tensor("op_34583_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_34583_end_mask_0 = const()[name = tensor("op_34583_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34583_cast_fp16 = slice_by_index(begin = var_34583_begin_0, end = var_34583_end_0, end_mask = var_34583_end_mask_0, x = q_155_cast_fp16)[name = tensor("op_34583_cast_fp16")]; + tensor var_34587_begin_0 = const()[name = tensor("op_34587_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_34587_end_0 = const()[name = tensor("op_34587_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_34587_end_mask_0 = const()[name = tensor("op_34587_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34587_cast_fp16 = slice_by_index(begin = var_34587_begin_0, end = var_34587_end_0, end_mask = var_34587_end_mask_0, x = q_155_cast_fp16)[name = tensor("op_34587_cast_fp16")]; + tensor var_34591_begin_0 = const()[name = tensor("op_34591_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_34591_end_0 = const()[name = tensor("op_34591_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_34591_end_mask_0 = const()[name = tensor("op_34591_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34591_cast_fp16 = slice_by_index(begin = var_34591_begin_0, end = var_34591_end_0, end_mask = var_34591_end_mask_0, x = q_155_cast_fp16)[name = tensor("op_34591_cast_fp16")]; + tensor var_34595_begin_0 = const()[name = tensor("op_34595_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_34595_end_0 = const()[name = tensor("op_34595_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_34595_end_mask_0 = const()[name = tensor("op_34595_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34595_cast_fp16 = slice_by_index(begin = var_34595_begin_0, end = var_34595_end_0, end_mask = var_34595_end_mask_0, x = q_155_cast_fp16)[name = tensor("op_34595_cast_fp16")]; + tensor var_34599_begin_0 = const()[name = tensor("op_34599_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_34599_end_0 = const()[name = tensor("op_34599_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_34599_end_mask_0 = const()[name = tensor("op_34599_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34599_cast_fp16 = slice_by_index(begin = var_34599_begin_0, end = var_34599_end_0, end_mask = var_34599_end_mask_0, x = q_155_cast_fp16)[name = tensor("op_34599_cast_fp16")]; + tensor var_34603_begin_0 = const()[name = tensor("op_34603_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_34603_end_0 = const()[name = tensor("op_34603_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_34603_end_mask_0 = const()[name = tensor("op_34603_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34603_cast_fp16 = slice_by_index(begin = var_34603_begin_0, end = var_34603_end_0, end_mask = var_34603_end_mask_0, x = q_155_cast_fp16)[name = tensor("op_34603_cast_fp16")]; + tensor var_34607_begin_0 = const()[name = tensor("op_34607_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_34607_end_0 = const()[name = tensor("op_34607_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_34607_end_mask_0 = const()[name = tensor("op_34607_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34607_cast_fp16 = slice_by_index(begin = var_34607_begin_0, end = var_34607_end_0, end_mask = var_34607_end_mask_0, x = q_155_cast_fp16)[name = tensor("op_34607_cast_fp16")]; + tensor var_34611_begin_0 = const()[name = tensor("op_34611_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_34611_end_0 = const()[name = tensor("op_34611_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_34611_end_mask_0 = const()[name = tensor("op_34611_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34611_cast_fp16 = slice_by_index(begin = var_34611_begin_0, end = var_34611_end_0, end_mask = var_34611_end_mask_0, x = q_155_cast_fp16)[name = tensor("op_34611_cast_fp16")]; + tensor var_34615_begin_0 = const()[name = tensor("op_34615_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_34615_end_0 = const()[name = tensor("op_34615_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_34615_end_mask_0 = const()[name = tensor("op_34615_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34615_cast_fp16 = slice_by_index(begin = var_34615_begin_0, end = var_34615_end_0, end_mask = var_34615_end_mask_0, x = q_155_cast_fp16)[name = tensor("op_34615_cast_fp16")]; + tensor var_34619_begin_0 = const()[name = tensor("op_34619_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_34619_end_0 = const()[name = tensor("op_34619_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_34619_end_mask_0 = const()[name = tensor("op_34619_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34619_cast_fp16 = slice_by_index(begin = var_34619_begin_0, end = var_34619_end_0, end_mask = var_34619_end_mask_0, x = q_155_cast_fp16)[name = tensor("op_34619_cast_fp16")]; + tensor var_34623_begin_0 = const()[name = tensor("op_34623_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_34623_end_0 = const()[name = tensor("op_34623_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_34623_end_mask_0 = const()[name = tensor("op_34623_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34623_cast_fp16 = slice_by_index(begin = var_34623_begin_0, end = var_34623_end_0, end_mask = var_34623_end_mask_0, x = q_155_cast_fp16)[name = tensor("op_34623_cast_fp16")]; + tensor var_34627_begin_0 = const()[name = tensor("op_34627_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_34627_end_0 = const()[name = tensor("op_34627_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_34627_end_mask_0 = const()[name = tensor("op_34627_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34627_cast_fp16 = slice_by_index(begin = var_34627_begin_0, end = var_34627_end_0, end_mask = var_34627_end_mask_0, x = q_155_cast_fp16)[name = tensor("op_34627_cast_fp16")]; + tensor var_34631_begin_0 = const()[name = tensor("op_34631_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_34631_end_0 = const()[name = tensor("op_34631_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_34631_end_mask_0 = const()[name = tensor("op_34631_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34631_cast_fp16 = slice_by_index(begin = var_34631_begin_0, end = var_34631_end_0, end_mask = var_34631_end_mask_0, x = q_155_cast_fp16)[name = tensor("op_34631_cast_fp16")]; + tensor var_34635_begin_0 = const()[name = tensor("op_34635_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_34635_end_0 = const()[name = tensor("op_34635_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_34635_end_mask_0 = const()[name = tensor("op_34635_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34635_cast_fp16 = slice_by_index(begin = var_34635_begin_0, end = var_34635_end_0, end_mask = var_34635_end_mask_0, x = q_155_cast_fp16)[name = tensor("op_34635_cast_fp16")]; + tensor var_34639_begin_0 = const()[name = tensor("op_34639_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_34639_end_0 = const()[name = tensor("op_34639_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_34639_end_mask_0 = const()[name = tensor("op_34639_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34639_cast_fp16 = slice_by_index(begin = var_34639_begin_0, end = var_34639_end_0, end_mask = var_34639_end_mask_0, x = q_155_cast_fp16)[name = tensor("op_34639_cast_fp16")]; + tensor var_34643_begin_0 = const()[name = tensor("op_34643_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_34643_end_0 = const()[name = tensor("op_34643_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_34643_end_mask_0 = const()[name = tensor("op_34643_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34643_cast_fp16 = slice_by_index(begin = var_34643_begin_0, end = var_34643_end_0, end_mask = var_34643_end_mask_0, x = q_155_cast_fp16)[name = tensor("op_34643_cast_fp16")]; + tensor var_34647_begin_0 = const()[name = tensor("op_34647_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_34647_end_0 = const()[name = tensor("op_34647_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_34647_end_mask_0 = const()[name = tensor("op_34647_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34647_cast_fp16 = slice_by_index(begin = var_34647_begin_0, end = var_34647_end_0, end_mask = var_34647_end_mask_0, x = q_155_cast_fp16)[name = tensor("op_34647_cast_fp16")]; + tensor var_34651_begin_0 = const()[name = tensor("op_34651_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_34651_end_0 = const()[name = tensor("op_34651_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_34651_end_mask_0 = const()[name = tensor("op_34651_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34651_cast_fp16 = slice_by_index(begin = var_34651_begin_0, end = var_34651_end_0, end_mask = var_34651_end_mask_0, x = q_155_cast_fp16)[name = tensor("op_34651_cast_fp16")]; + tensor var_34655_begin_0 = const()[name = tensor("op_34655_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_34655_end_0 = const()[name = tensor("op_34655_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_34655_end_mask_0 = const()[name = tensor("op_34655_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34655_cast_fp16 = slice_by_index(begin = var_34655_begin_0, end = var_34655_end_0, end_mask = var_34655_end_mask_0, x = q_155_cast_fp16)[name = tensor("op_34655_cast_fp16")]; + tensor var_34659_begin_0 = const()[name = tensor("op_34659_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_34659_end_0 = const()[name = tensor("op_34659_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_34659_end_mask_0 = const()[name = tensor("op_34659_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34659_cast_fp16 = slice_by_index(begin = var_34659_begin_0, end = var_34659_end_0, end_mask = var_34659_end_mask_0, x = q_155_cast_fp16)[name = tensor("op_34659_cast_fp16")]; + tensor k_311_perm_0 = const()[name = tensor("k_311_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_34666_begin_0 = const()[name = tensor("op_34666_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_34666_end_0 = const()[name = tensor("op_34666_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_34666_end_mask_0 = const()[name = tensor("op_34666_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_311_cast_fp16 = transpose(perm = k_311_perm_0, x = k_309_cast_fp16)[name = tensor("transpose_62")]; + tensor var_34666_cast_fp16 = slice_by_index(begin = var_34666_begin_0, end = var_34666_end_0, end_mask = var_34666_end_mask_0, x = k_311_cast_fp16)[name = tensor("op_34666_cast_fp16")]; + tensor var_34670_begin_0 = const()[name = tensor("op_34670_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_34670_end_0 = const()[name = tensor("op_34670_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_34670_end_mask_0 = const()[name = tensor("op_34670_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34670_cast_fp16 = slice_by_index(begin = var_34670_begin_0, end = var_34670_end_0, end_mask = var_34670_end_mask_0, x = k_311_cast_fp16)[name = tensor("op_34670_cast_fp16")]; + tensor var_34674_begin_0 = const()[name = tensor("op_34674_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_34674_end_0 = const()[name = tensor("op_34674_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_34674_end_mask_0 = const()[name = tensor("op_34674_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34674_cast_fp16 = slice_by_index(begin = var_34674_begin_0, end = var_34674_end_0, end_mask = var_34674_end_mask_0, x = k_311_cast_fp16)[name = tensor("op_34674_cast_fp16")]; + tensor var_34678_begin_0 = const()[name = tensor("op_34678_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_34678_end_0 = const()[name = tensor("op_34678_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_34678_end_mask_0 = const()[name = tensor("op_34678_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34678_cast_fp16 = slice_by_index(begin = var_34678_begin_0, end = var_34678_end_0, end_mask = var_34678_end_mask_0, x = k_311_cast_fp16)[name = tensor("op_34678_cast_fp16")]; + tensor var_34682_begin_0 = const()[name = tensor("op_34682_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_34682_end_0 = const()[name = tensor("op_34682_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_34682_end_mask_0 = const()[name = tensor("op_34682_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34682_cast_fp16 = slice_by_index(begin = var_34682_begin_0, end = var_34682_end_0, end_mask = var_34682_end_mask_0, x = k_311_cast_fp16)[name = tensor("op_34682_cast_fp16")]; + tensor var_34686_begin_0 = const()[name = tensor("op_34686_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_34686_end_0 = const()[name = tensor("op_34686_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_34686_end_mask_0 = const()[name = tensor("op_34686_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34686_cast_fp16 = slice_by_index(begin = var_34686_begin_0, end = var_34686_end_0, end_mask = var_34686_end_mask_0, x = k_311_cast_fp16)[name = tensor("op_34686_cast_fp16")]; + tensor var_34690_begin_0 = const()[name = tensor("op_34690_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_34690_end_0 = const()[name = tensor("op_34690_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_34690_end_mask_0 = const()[name = tensor("op_34690_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34690_cast_fp16 = slice_by_index(begin = var_34690_begin_0, end = var_34690_end_0, end_mask = var_34690_end_mask_0, x = k_311_cast_fp16)[name = tensor("op_34690_cast_fp16")]; + tensor var_34694_begin_0 = const()[name = tensor("op_34694_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_34694_end_0 = const()[name = tensor("op_34694_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_34694_end_mask_0 = const()[name = tensor("op_34694_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34694_cast_fp16 = slice_by_index(begin = var_34694_begin_0, end = var_34694_end_0, end_mask = var_34694_end_mask_0, x = k_311_cast_fp16)[name = tensor("op_34694_cast_fp16")]; + tensor var_34698_begin_0 = const()[name = tensor("op_34698_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_34698_end_0 = const()[name = tensor("op_34698_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_34698_end_mask_0 = const()[name = tensor("op_34698_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34698_cast_fp16 = slice_by_index(begin = var_34698_begin_0, end = var_34698_end_0, end_mask = var_34698_end_mask_0, x = k_311_cast_fp16)[name = tensor("op_34698_cast_fp16")]; + tensor var_34702_begin_0 = const()[name = tensor("op_34702_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_34702_end_0 = const()[name = tensor("op_34702_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_34702_end_mask_0 = const()[name = tensor("op_34702_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34702_cast_fp16 = slice_by_index(begin = var_34702_begin_0, end = var_34702_end_0, end_mask = var_34702_end_mask_0, x = k_311_cast_fp16)[name = tensor("op_34702_cast_fp16")]; + tensor var_34706_begin_0 = const()[name = tensor("op_34706_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_34706_end_0 = const()[name = tensor("op_34706_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_34706_end_mask_0 = const()[name = tensor("op_34706_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34706_cast_fp16 = slice_by_index(begin = var_34706_begin_0, end = var_34706_end_0, end_mask = var_34706_end_mask_0, x = k_311_cast_fp16)[name = tensor("op_34706_cast_fp16")]; + tensor var_34710_begin_0 = const()[name = tensor("op_34710_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_34710_end_0 = const()[name = tensor("op_34710_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_34710_end_mask_0 = const()[name = tensor("op_34710_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34710_cast_fp16 = slice_by_index(begin = var_34710_begin_0, end = var_34710_end_0, end_mask = var_34710_end_mask_0, x = k_311_cast_fp16)[name = tensor("op_34710_cast_fp16")]; + tensor var_34714_begin_0 = const()[name = tensor("op_34714_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_34714_end_0 = const()[name = tensor("op_34714_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_34714_end_mask_0 = const()[name = tensor("op_34714_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34714_cast_fp16 = slice_by_index(begin = var_34714_begin_0, end = var_34714_end_0, end_mask = var_34714_end_mask_0, x = k_311_cast_fp16)[name = tensor("op_34714_cast_fp16")]; + tensor var_34718_begin_0 = const()[name = tensor("op_34718_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_34718_end_0 = const()[name = tensor("op_34718_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_34718_end_mask_0 = const()[name = tensor("op_34718_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34718_cast_fp16 = slice_by_index(begin = var_34718_begin_0, end = var_34718_end_0, end_mask = var_34718_end_mask_0, x = k_311_cast_fp16)[name = tensor("op_34718_cast_fp16")]; + tensor var_34722_begin_0 = const()[name = tensor("op_34722_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_34722_end_0 = const()[name = tensor("op_34722_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_34722_end_mask_0 = const()[name = tensor("op_34722_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34722_cast_fp16 = slice_by_index(begin = var_34722_begin_0, end = var_34722_end_0, end_mask = var_34722_end_mask_0, x = k_311_cast_fp16)[name = tensor("op_34722_cast_fp16")]; + tensor var_34726_begin_0 = const()[name = tensor("op_34726_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_34726_end_0 = const()[name = tensor("op_34726_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_34726_end_mask_0 = const()[name = tensor("op_34726_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34726_cast_fp16 = slice_by_index(begin = var_34726_begin_0, end = var_34726_end_0, end_mask = var_34726_end_mask_0, x = k_311_cast_fp16)[name = tensor("op_34726_cast_fp16")]; + tensor var_34730_begin_0 = const()[name = tensor("op_34730_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_34730_end_0 = const()[name = tensor("op_34730_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_34730_end_mask_0 = const()[name = tensor("op_34730_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34730_cast_fp16 = slice_by_index(begin = var_34730_begin_0, end = var_34730_end_0, end_mask = var_34730_end_mask_0, x = k_311_cast_fp16)[name = tensor("op_34730_cast_fp16")]; + tensor var_34734_begin_0 = const()[name = tensor("op_34734_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_34734_end_0 = const()[name = tensor("op_34734_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_34734_end_mask_0 = const()[name = tensor("op_34734_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34734_cast_fp16 = slice_by_index(begin = var_34734_begin_0, end = var_34734_end_0, end_mask = var_34734_end_mask_0, x = k_311_cast_fp16)[name = tensor("op_34734_cast_fp16")]; + tensor var_34738_begin_0 = const()[name = tensor("op_34738_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_34738_end_0 = const()[name = tensor("op_34738_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_34738_end_mask_0 = const()[name = tensor("op_34738_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34738_cast_fp16 = slice_by_index(begin = var_34738_begin_0, end = var_34738_end_0, end_mask = var_34738_end_mask_0, x = k_311_cast_fp16)[name = tensor("op_34738_cast_fp16")]; + tensor var_34742_begin_0 = const()[name = tensor("op_34742_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_34742_end_0 = const()[name = tensor("op_34742_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_34742_end_mask_0 = const()[name = tensor("op_34742_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34742_cast_fp16 = slice_by_index(begin = var_34742_begin_0, end = var_34742_end_0, end_mask = var_34742_end_mask_0, x = k_311_cast_fp16)[name = tensor("op_34742_cast_fp16")]; + tensor var_34744_begin_0 = const()[name = tensor("op_34744_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_34744_end_0 = const()[name = tensor("op_34744_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_34744_end_mask_0 = const()[name = tensor("op_34744_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34744_cast_fp16 = slice_by_index(begin = var_34744_begin_0, end = var_34744_end_0, end_mask = var_34744_end_mask_0, x = v_155_cast_fp16)[name = tensor("op_34744_cast_fp16")]; + tensor var_34748_begin_0 = const()[name = tensor("op_34748_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_34748_end_0 = const()[name = tensor("op_34748_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_34748_end_mask_0 = const()[name = tensor("op_34748_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34748_cast_fp16 = slice_by_index(begin = var_34748_begin_0, end = var_34748_end_0, end_mask = var_34748_end_mask_0, x = v_155_cast_fp16)[name = tensor("op_34748_cast_fp16")]; + tensor var_34752_begin_0 = const()[name = tensor("op_34752_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_34752_end_0 = const()[name = tensor("op_34752_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_34752_end_mask_0 = const()[name = tensor("op_34752_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34752_cast_fp16 = slice_by_index(begin = var_34752_begin_0, end = var_34752_end_0, end_mask = var_34752_end_mask_0, x = v_155_cast_fp16)[name = tensor("op_34752_cast_fp16")]; + tensor var_34756_begin_0 = const()[name = tensor("op_34756_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_34756_end_0 = const()[name = tensor("op_34756_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_34756_end_mask_0 = const()[name = tensor("op_34756_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34756_cast_fp16 = slice_by_index(begin = var_34756_begin_0, end = var_34756_end_0, end_mask = var_34756_end_mask_0, x = v_155_cast_fp16)[name = tensor("op_34756_cast_fp16")]; + tensor var_34760_begin_0 = const()[name = tensor("op_34760_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_34760_end_0 = const()[name = tensor("op_34760_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_34760_end_mask_0 = const()[name = tensor("op_34760_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34760_cast_fp16 = slice_by_index(begin = var_34760_begin_0, end = var_34760_end_0, end_mask = var_34760_end_mask_0, x = v_155_cast_fp16)[name = tensor("op_34760_cast_fp16")]; + tensor var_34764_begin_0 = const()[name = tensor("op_34764_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_34764_end_0 = const()[name = tensor("op_34764_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_34764_end_mask_0 = const()[name = tensor("op_34764_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34764_cast_fp16 = slice_by_index(begin = var_34764_begin_0, end = var_34764_end_0, end_mask = var_34764_end_mask_0, x = v_155_cast_fp16)[name = tensor("op_34764_cast_fp16")]; + tensor var_34768_begin_0 = const()[name = tensor("op_34768_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_34768_end_0 = const()[name = tensor("op_34768_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_34768_end_mask_0 = const()[name = tensor("op_34768_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34768_cast_fp16 = slice_by_index(begin = var_34768_begin_0, end = var_34768_end_0, end_mask = var_34768_end_mask_0, x = v_155_cast_fp16)[name = tensor("op_34768_cast_fp16")]; + tensor var_34772_begin_0 = const()[name = tensor("op_34772_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_34772_end_0 = const()[name = tensor("op_34772_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_34772_end_mask_0 = const()[name = tensor("op_34772_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34772_cast_fp16 = slice_by_index(begin = var_34772_begin_0, end = var_34772_end_0, end_mask = var_34772_end_mask_0, x = v_155_cast_fp16)[name = tensor("op_34772_cast_fp16")]; + tensor var_34776_begin_0 = const()[name = tensor("op_34776_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_34776_end_0 = const()[name = tensor("op_34776_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_34776_end_mask_0 = const()[name = tensor("op_34776_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34776_cast_fp16 = slice_by_index(begin = var_34776_begin_0, end = var_34776_end_0, end_mask = var_34776_end_mask_0, x = v_155_cast_fp16)[name = tensor("op_34776_cast_fp16")]; + tensor var_34780_begin_0 = const()[name = tensor("op_34780_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_34780_end_0 = const()[name = tensor("op_34780_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_34780_end_mask_0 = const()[name = tensor("op_34780_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34780_cast_fp16 = slice_by_index(begin = var_34780_begin_0, end = var_34780_end_0, end_mask = var_34780_end_mask_0, x = v_155_cast_fp16)[name = tensor("op_34780_cast_fp16")]; + tensor var_34784_begin_0 = const()[name = tensor("op_34784_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_34784_end_0 = const()[name = tensor("op_34784_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_34784_end_mask_0 = const()[name = tensor("op_34784_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34784_cast_fp16 = slice_by_index(begin = var_34784_begin_0, end = var_34784_end_0, end_mask = var_34784_end_mask_0, x = v_155_cast_fp16)[name = tensor("op_34784_cast_fp16")]; + tensor var_34788_begin_0 = const()[name = tensor("op_34788_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_34788_end_0 = const()[name = tensor("op_34788_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_34788_end_mask_0 = const()[name = tensor("op_34788_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34788_cast_fp16 = slice_by_index(begin = var_34788_begin_0, end = var_34788_end_0, end_mask = var_34788_end_mask_0, x = v_155_cast_fp16)[name = tensor("op_34788_cast_fp16")]; + tensor var_34792_begin_0 = const()[name = tensor("op_34792_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_34792_end_0 = const()[name = tensor("op_34792_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_34792_end_mask_0 = const()[name = tensor("op_34792_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34792_cast_fp16 = slice_by_index(begin = var_34792_begin_0, end = var_34792_end_0, end_mask = var_34792_end_mask_0, x = v_155_cast_fp16)[name = tensor("op_34792_cast_fp16")]; + tensor var_34796_begin_0 = const()[name = tensor("op_34796_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_34796_end_0 = const()[name = tensor("op_34796_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_34796_end_mask_0 = const()[name = tensor("op_34796_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34796_cast_fp16 = slice_by_index(begin = var_34796_begin_0, end = var_34796_end_0, end_mask = var_34796_end_mask_0, x = v_155_cast_fp16)[name = tensor("op_34796_cast_fp16")]; + tensor var_34800_begin_0 = const()[name = tensor("op_34800_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_34800_end_0 = const()[name = tensor("op_34800_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_34800_end_mask_0 = const()[name = tensor("op_34800_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34800_cast_fp16 = slice_by_index(begin = var_34800_begin_0, end = var_34800_end_0, end_mask = var_34800_end_mask_0, x = v_155_cast_fp16)[name = tensor("op_34800_cast_fp16")]; + tensor var_34804_begin_0 = const()[name = tensor("op_34804_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_34804_end_0 = const()[name = tensor("op_34804_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_34804_end_mask_0 = const()[name = tensor("op_34804_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34804_cast_fp16 = slice_by_index(begin = var_34804_begin_0, end = var_34804_end_0, end_mask = var_34804_end_mask_0, x = v_155_cast_fp16)[name = tensor("op_34804_cast_fp16")]; + tensor var_34808_begin_0 = const()[name = tensor("op_34808_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_34808_end_0 = const()[name = tensor("op_34808_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_34808_end_mask_0 = const()[name = tensor("op_34808_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34808_cast_fp16 = slice_by_index(begin = var_34808_begin_0, end = var_34808_end_0, end_mask = var_34808_end_mask_0, x = v_155_cast_fp16)[name = tensor("op_34808_cast_fp16")]; + tensor var_34812_begin_0 = const()[name = tensor("op_34812_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_34812_end_0 = const()[name = tensor("op_34812_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_34812_end_mask_0 = const()[name = tensor("op_34812_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34812_cast_fp16 = slice_by_index(begin = var_34812_begin_0, end = var_34812_end_0, end_mask = var_34812_end_mask_0, x = v_155_cast_fp16)[name = tensor("op_34812_cast_fp16")]; + tensor var_34816_begin_0 = const()[name = tensor("op_34816_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_34816_end_0 = const()[name = tensor("op_34816_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_34816_end_mask_0 = const()[name = tensor("op_34816_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34816_cast_fp16 = slice_by_index(begin = var_34816_begin_0, end = var_34816_end_0, end_mask = var_34816_end_mask_0, x = v_155_cast_fp16)[name = tensor("op_34816_cast_fp16")]; + tensor var_34820_begin_0 = const()[name = tensor("op_34820_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_34820_end_0 = const()[name = tensor("op_34820_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_34820_end_mask_0 = const()[name = tensor("op_34820_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34820_cast_fp16 = slice_by_index(begin = var_34820_begin_0, end = var_34820_end_0, end_mask = var_34820_end_mask_0, x = v_155_cast_fp16)[name = tensor("op_34820_cast_fp16")]; + tensor var_34824_equation_0 = const()[name = tensor("op_34824_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34824_cast_fp16 = einsum(equation = var_34824_equation_0, values = (var_34666_cast_fp16, var_34583_cast_fp16))[name = tensor("op_34824_cast_fp16")]; + tensor var_34825_to_fp16 = const()[name = tensor("op_34825_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2921_cast_fp16 = mul(x = var_34824_cast_fp16, y = var_34825_to_fp16)[name = tensor("aw_2921_cast_fp16")]; + tensor var_34828_equation_0 = const()[name = tensor("op_34828_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34828_cast_fp16 = einsum(equation = var_34828_equation_0, values = (var_34670_cast_fp16, var_34587_cast_fp16))[name = tensor("op_34828_cast_fp16")]; + tensor var_34829_to_fp16 = const()[name = tensor("op_34829_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2923_cast_fp16 = mul(x = var_34828_cast_fp16, y = var_34829_to_fp16)[name = tensor("aw_2923_cast_fp16")]; + tensor var_34832_equation_0 = const()[name = tensor("op_34832_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34832_cast_fp16 = einsum(equation = var_34832_equation_0, values = (var_34674_cast_fp16, var_34591_cast_fp16))[name = tensor("op_34832_cast_fp16")]; + tensor var_34833_to_fp16 = const()[name = tensor("op_34833_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2925_cast_fp16 = mul(x = var_34832_cast_fp16, y = var_34833_to_fp16)[name = tensor("aw_2925_cast_fp16")]; + tensor var_34836_equation_0 = const()[name = tensor("op_34836_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34836_cast_fp16 = einsum(equation = var_34836_equation_0, values = (var_34678_cast_fp16, var_34595_cast_fp16))[name = tensor("op_34836_cast_fp16")]; + tensor var_34837_to_fp16 = const()[name = tensor("op_34837_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2927_cast_fp16 = mul(x = var_34836_cast_fp16, y = var_34837_to_fp16)[name = tensor("aw_2927_cast_fp16")]; + tensor var_34840_equation_0 = const()[name = tensor("op_34840_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34840_cast_fp16 = einsum(equation = var_34840_equation_0, values = (var_34682_cast_fp16, var_34599_cast_fp16))[name = tensor("op_34840_cast_fp16")]; + tensor var_34841_to_fp16 = const()[name = tensor("op_34841_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2929_cast_fp16 = mul(x = var_34840_cast_fp16, y = var_34841_to_fp16)[name = tensor("aw_2929_cast_fp16")]; + tensor var_34844_equation_0 = const()[name = tensor("op_34844_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34844_cast_fp16 = einsum(equation = var_34844_equation_0, values = (var_34686_cast_fp16, var_34603_cast_fp16))[name = tensor("op_34844_cast_fp16")]; + tensor var_34845_to_fp16 = const()[name = tensor("op_34845_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2931_cast_fp16 = mul(x = var_34844_cast_fp16, y = var_34845_to_fp16)[name = tensor("aw_2931_cast_fp16")]; + tensor var_34848_equation_0 = const()[name = tensor("op_34848_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34848_cast_fp16 = einsum(equation = var_34848_equation_0, values = (var_34690_cast_fp16, var_34607_cast_fp16))[name = tensor("op_34848_cast_fp16")]; + tensor var_34849_to_fp16 = const()[name = tensor("op_34849_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2933_cast_fp16 = mul(x = var_34848_cast_fp16, y = var_34849_to_fp16)[name = tensor("aw_2933_cast_fp16")]; + tensor var_34852_equation_0 = const()[name = tensor("op_34852_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34852_cast_fp16 = einsum(equation = var_34852_equation_0, values = (var_34694_cast_fp16, var_34611_cast_fp16))[name = tensor("op_34852_cast_fp16")]; + tensor var_34853_to_fp16 = const()[name = tensor("op_34853_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2935_cast_fp16 = mul(x = var_34852_cast_fp16, y = var_34853_to_fp16)[name = tensor("aw_2935_cast_fp16")]; + tensor var_34856_equation_0 = const()[name = tensor("op_34856_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34856_cast_fp16 = einsum(equation = var_34856_equation_0, values = (var_34698_cast_fp16, var_34615_cast_fp16))[name = tensor("op_34856_cast_fp16")]; + tensor var_34857_to_fp16 = const()[name = tensor("op_34857_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2937_cast_fp16 = mul(x = var_34856_cast_fp16, y = var_34857_to_fp16)[name = tensor("aw_2937_cast_fp16")]; + tensor var_34860_equation_0 = const()[name = tensor("op_34860_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34860_cast_fp16 = einsum(equation = var_34860_equation_0, values = (var_34702_cast_fp16, var_34619_cast_fp16))[name = tensor("op_34860_cast_fp16")]; + tensor var_34861_to_fp16 = const()[name = tensor("op_34861_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2939_cast_fp16 = mul(x = var_34860_cast_fp16, y = var_34861_to_fp16)[name = tensor("aw_2939_cast_fp16")]; + tensor var_34864_equation_0 = const()[name = tensor("op_34864_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34864_cast_fp16 = einsum(equation = var_34864_equation_0, values = (var_34706_cast_fp16, var_34623_cast_fp16))[name = tensor("op_34864_cast_fp16")]; + tensor var_34865_to_fp16 = const()[name = tensor("op_34865_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2941_cast_fp16 = mul(x = var_34864_cast_fp16, y = var_34865_to_fp16)[name = tensor("aw_2941_cast_fp16")]; + tensor var_34868_equation_0 = const()[name = tensor("op_34868_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34868_cast_fp16 = einsum(equation = var_34868_equation_0, values = (var_34710_cast_fp16, var_34627_cast_fp16))[name = tensor("op_34868_cast_fp16")]; + tensor var_34869_to_fp16 = const()[name = tensor("op_34869_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2943_cast_fp16 = mul(x = var_34868_cast_fp16, y = var_34869_to_fp16)[name = tensor("aw_2943_cast_fp16")]; + tensor var_34872_equation_0 = const()[name = tensor("op_34872_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34872_cast_fp16 = einsum(equation = var_34872_equation_0, values = (var_34714_cast_fp16, var_34631_cast_fp16))[name = tensor("op_34872_cast_fp16")]; + tensor var_34873_to_fp16 = const()[name = tensor("op_34873_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2945_cast_fp16 = mul(x = var_34872_cast_fp16, y = var_34873_to_fp16)[name = tensor("aw_2945_cast_fp16")]; + tensor var_34876_equation_0 = const()[name = tensor("op_34876_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34876_cast_fp16 = einsum(equation = var_34876_equation_0, values = (var_34718_cast_fp16, var_34635_cast_fp16))[name = tensor("op_34876_cast_fp16")]; + tensor var_34877_to_fp16 = const()[name = tensor("op_34877_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2947_cast_fp16 = mul(x = var_34876_cast_fp16, y = var_34877_to_fp16)[name = tensor("aw_2947_cast_fp16")]; + tensor var_34880_equation_0 = const()[name = tensor("op_34880_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34880_cast_fp16 = einsum(equation = var_34880_equation_0, values = (var_34722_cast_fp16, var_34639_cast_fp16))[name = tensor("op_34880_cast_fp16")]; + tensor var_34881_to_fp16 = const()[name = tensor("op_34881_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2949_cast_fp16 = mul(x = var_34880_cast_fp16, y = var_34881_to_fp16)[name = tensor("aw_2949_cast_fp16")]; + tensor var_34884_equation_0 = const()[name = tensor("op_34884_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34884_cast_fp16 = einsum(equation = var_34884_equation_0, values = (var_34726_cast_fp16, var_34643_cast_fp16))[name = tensor("op_34884_cast_fp16")]; + tensor var_34885_to_fp16 = const()[name = tensor("op_34885_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2951_cast_fp16 = mul(x = var_34884_cast_fp16, y = var_34885_to_fp16)[name = tensor("aw_2951_cast_fp16")]; + tensor var_34888_equation_0 = const()[name = tensor("op_34888_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34888_cast_fp16 = einsum(equation = var_34888_equation_0, values = (var_34730_cast_fp16, var_34647_cast_fp16))[name = tensor("op_34888_cast_fp16")]; + tensor var_34889_to_fp16 = const()[name = tensor("op_34889_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2953_cast_fp16 = mul(x = var_34888_cast_fp16, y = var_34889_to_fp16)[name = tensor("aw_2953_cast_fp16")]; + tensor var_34892_equation_0 = const()[name = tensor("op_34892_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34892_cast_fp16 = einsum(equation = var_34892_equation_0, values = (var_34734_cast_fp16, var_34651_cast_fp16))[name = tensor("op_34892_cast_fp16")]; + tensor var_34893_to_fp16 = const()[name = tensor("op_34893_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2955_cast_fp16 = mul(x = var_34892_cast_fp16, y = var_34893_to_fp16)[name = tensor("aw_2955_cast_fp16")]; + tensor var_34896_equation_0 = const()[name = tensor("op_34896_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34896_cast_fp16 = einsum(equation = var_34896_equation_0, values = (var_34738_cast_fp16, var_34655_cast_fp16))[name = tensor("op_34896_cast_fp16")]; + tensor var_34897_to_fp16 = const()[name = tensor("op_34897_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2957_cast_fp16 = mul(x = var_34896_cast_fp16, y = var_34897_to_fp16)[name = tensor("aw_2957_cast_fp16")]; + tensor var_34900_equation_0 = const()[name = tensor("op_34900_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34900_cast_fp16 = einsum(equation = var_34900_equation_0, values = (var_34742_cast_fp16, var_34659_cast_fp16))[name = tensor("op_34900_cast_fp16")]; + tensor var_34901_to_fp16 = const()[name = tensor("op_34901_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2959_cast_fp16 = mul(x = var_34900_cast_fp16, y = var_34901_to_fp16)[name = tensor("aw_2959_cast_fp16")]; + tensor var_34903_cast_fp16 = softmax(axis = var_30355, x = aw_2921_cast_fp16)[name = tensor("op_34903_cast_fp16")]; + tensor var_34904_cast_fp16 = softmax(axis = var_30355, x = aw_2923_cast_fp16)[name = tensor("op_34904_cast_fp16")]; + tensor var_34905_cast_fp16 = softmax(axis = var_30355, x = aw_2925_cast_fp16)[name = tensor("op_34905_cast_fp16")]; + tensor var_34906_cast_fp16 = softmax(axis = var_30355, x = aw_2927_cast_fp16)[name = tensor("op_34906_cast_fp16")]; + tensor var_34907_cast_fp16 = softmax(axis = var_30355, x = aw_2929_cast_fp16)[name = tensor("op_34907_cast_fp16")]; + tensor var_34908_cast_fp16 = softmax(axis = var_30355, x = aw_2931_cast_fp16)[name = tensor("op_34908_cast_fp16")]; + tensor var_34909_cast_fp16 = softmax(axis = var_30355, x = aw_2933_cast_fp16)[name = tensor("op_34909_cast_fp16")]; + tensor var_34910_cast_fp16 = softmax(axis = var_30355, x = aw_2935_cast_fp16)[name = tensor("op_34910_cast_fp16")]; + tensor var_34911_cast_fp16 = softmax(axis = var_30355, x = aw_2937_cast_fp16)[name = tensor("op_34911_cast_fp16")]; + tensor var_34912_cast_fp16 = softmax(axis = var_30355, x = aw_2939_cast_fp16)[name = tensor("op_34912_cast_fp16")]; + tensor var_34913_cast_fp16 = softmax(axis = var_30355, x = aw_2941_cast_fp16)[name = tensor("op_34913_cast_fp16")]; + tensor var_34914_cast_fp16 = softmax(axis = var_30355, x = aw_2943_cast_fp16)[name = tensor("op_34914_cast_fp16")]; + tensor var_34915_cast_fp16 = softmax(axis = var_30355, x = aw_2945_cast_fp16)[name = tensor("op_34915_cast_fp16")]; + tensor var_34916_cast_fp16 = softmax(axis = var_30355, x = aw_2947_cast_fp16)[name = tensor("op_34916_cast_fp16")]; + tensor var_34917_cast_fp16 = softmax(axis = var_30355, x = aw_2949_cast_fp16)[name = tensor("op_34917_cast_fp16")]; + tensor var_34918_cast_fp16 = softmax(axis = var_30355, x = aw_2951_cast_fp16)[name = tensor("op_34918_cast_fp16")]; + tensor var_34919_cast_fp16 = softmax(axis = var_30355, x = aw_2953_cast_fp16)[name = tensor("op_34919_cast_fp16")]; + tensor var_34920_cast_fp16 = softmax(axis = var_30355, x = aw_2955_cast_fp16)[name = tensor("op_34920_cast_fp16")]; + tensor var_34921_cast_fp16 = softmax(axis = var_30355, x = aw_2957_cast_fp16)[name = tensor("op_34921_cast_fp16")]; + tensor var_34922_cast_fp16 = softmax(axis = var_30355, x = aw_2959_cast_fp16)[name = tensor("op_34922_cast_fp16")]; + tensor var_34924_equation_0 = const()[name = tensor("op_34924_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34924_cast_fp16 = einsum(equation = var_34924_equation_0, values = (var_34744_cast_fp16, var_34903_cast_fp16))[name = tensor("op_34924_cast_fp16")]; + tensor var_34926_equation_0 = const()[name = tensor("op_34926_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34926_cast_fp16 = einsum(equation = var_34926_equation_0, values = (var_34748_cast_fp16, var_34904_cast_fp16))[name = tensor("op_34926_cast_fp16")]; + tensor var_34928_equation_0 = const()[name = tensor("op_34928_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34928_cast_fp16 = einsum(equation = var_34928_equation_0, values = (var_34752_cast_fp16, var_34905_cast_fp16))[name = tensor("op_34928_cast_fp16")]; + tensor var_34930_equation_0 = const()[name = tensor("op_34930_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34930_cast_fp16 = einsum(equation = var_34930_equation_0, values = (var_34756_cast_fp16, var_34906_cast_fp16))[name = tensor("op_34930_cast_fp16")]; + tensor var_34932_equation_0 = const()[name = tensor("op_34932_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34932_cast_fp16 = einsum(equation = var_34932_equation_0, values = (var_34760_cast_fp16, var_34907_cast_fp16))[name = tensor("op_34932_cast_fp16")]; + tensor var_34934_equation_0 = const()[name = tensor("op_34934_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34934_cast_fp16 = einsum(equation = var_34934_equation_0, values = (var_34764_cast_fp16, var_34908_cast_fp16))[name = tensor("op_34934_cast_fp16")]; + tensor var_34936_equation_0 = const()[name = tensor("op_34936_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34936_cast_fp16 = einsum(equation = var_34936_equation_0, values = (var_34768_cast_fp16, var_34909_cast_fp16))[name = tensor("op_34936_cast_fp16")]; + tensor var_34938_equation_0 = const()[name = tensor("op_34938_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34938_cast_fp16 = einsum(equation = var_34938_equation_0, values = (var_34772_cast_fp16, var_34910_cast_fp16))[name = tensor("op_34938_cast_fp16")]; + tensor var_34940_equation_0 = const()[name = tensor("op_34940_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34940_cast_fp16 = einsum(equation = var_34940_equation_0, values = (var_34776_cast_fp16, var_34911_cast_fp16))[name = tensor("op_34940_cast_fp16")]; + tensor var_34942_equation_0 = const()[name = tensor("op_34942_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34942_cast_fp16 = einsum(equation = var_34942_equation_0, values = (var_34780_cast_fp16, var_34912_cast_fp16))[name = tensor("op_34942_cast_fp16")]; + tensor var_34944_equation_0 = const()[name = tensor("op_34944_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34944_cast_fp16 = einsum(equation = var_34944_equation_0, values = (var_34784_cast_fp16, var_34913_cast_fp16))[name = tensor("op_34944_cast_fp16")]; + tensor var_34946_equation_0 = const()[name = tensor("op_34946_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34946_cast_fp16 = einsum(equation = var_34946_equation_0, values = (var_34788_cast_fp16, var_34914_cast_fp16))[name = tensor("op_34946_cast_fp16")]; + tensor var_34948_equation_0 = const()[name = tensor("op_34948_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34948_cast_fp16 = einsum(equation = var_34948_equation_0, values = (var_34792_cast_fp16, var_34915_cast_fp16))[name = tensor("op_34948_cast_fp16")]; + tensor var_34950_equation_0 = const()[name = tensor("op_34950_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34950_cast_fp16 = einsum(equation = var_34950_equation_0, values = (var_34796_cast_fp16, var_34916_cast_fp16))[name = tensor("op_34950_cast_fp16")]; + tensor var_34952_equation_0 = const()[name = tensor("op_34952_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34952_cast_fp16 = einsum(equation = var_34952_equation_0, values = (var_34800_cast_fp16, var_34917_cast_fp16))[name = tensor("op_34952_cast_fp16")]; + tensor var_34954_equation_0 = const()[name = tensor("op_34954_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34954_cast_fp16 = einsum(equation = var_34954_equation_0, values = (var_34804_cast_fp16, var_34918_cast_fp16))[name = tensor("op_34954_cast_fp16")]; + tensor var_34956_equation_0 = const()[name = tensor("op_34956_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34956_cast_fp16 = einsum(equation = var_34956_equation_0, values = (var_34808_cast_fp16, var_34919_cast_fp16))[name = tensor("op_34956_cast_fp16")]; + tensor var_34958_equation_0 = const()[name = tensor("op_34958_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34958_cast_fp16 = einsum(equation = var_34958_equation_0, values = (var_34812_cast_fp16, var_34920_cast_fp16))[name = tensor("op_34958_cast_fp16")]; + tensor var_34960_equation_0 = const()[name = tensor("op_34960_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34960_cast_fp16 = einsum(equation = var_34960_equation_0, values = (var_34816_cast_fp16, var_34921_cast_fp16))[name = tensor("op_34960_cast_fp16")]; + tensor var_34962_equation_0 = const()[name = tensor("op_34962_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34962_cast_fp16 = einsum(equation = var_34962_equation_0, values = (var_34820_cast_fp16, var_34922_cast_fp16))[name = tensor("op_34962_cast_fp16")]; + tensor input_473_interleave_0 = const()[name = tensor("input_473_interleave_0"), val = tensor(false)]; + tensor input_473_cast_fp16 = concat(axis = var_30355, interleave = input_473_interleave_0, values = (var_34924_cast_fp16, var_34926_cast_fp16, var_34928_cast_fp16, var_34930_cast_fp16, var_34932_cast_fp16, var_34934_cast_fp16, var_34936_cast_fp16, var_34938_cast_fp16, var_34940_cast_fp16, var_34942_cast_fp16, var_34944_cast_fp16, var_34946_cast_fp16, var_34948_cast_fp16, var_34950_cast_fp16, var_34952_cast_fp16, var_34954_cast_fp16, var_34956_cast_fp16, var_34958_cast_fp16, var_34960_cast_fp16, var_34962_cast_fp16))[name = tensor("input_473_cast_fp16")]; + tensor var_34968 = const()[name = tensor("op_34968"), val = tensor([1, 1])]; + tensor var_34970 = const()[name = tensor("op_34970"), val = tensor([1, 1])]; + tensor var_34972_pad_type_0 = const()[name = tensor("op_34972_pad_type_0"), val = tensor("custom")]; + tensor var_34972_pad_0 = const()[name = tensor("op_34972_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_4_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_4_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2893356544)))]; + tensor up_blocks_0_attentions_0_transformer_blocks_4_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_4_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2896633408)))]; + tensor var_34972_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_4_attn2_to_out_0_bias_to_fp16, dilations = var_34970, groups = var_30355, pad = var_34972_pad_0, pad_type = var_34972_pad_type_0, strides = var_34968, weight = up_blocks_0_attentions_0_transformer_blocks_4_attn2_to_out_0_weight_to_fp16, x = input_473_cast_fp16)[name = tensor("op_34972_cast_fp16")]; + tensor inputs_233_cast_fp16 = add(x = var_34972_cast_fp16, y = inputs_231_cast_fp16)[name = tensor("inputs_233_cast_fp16")]; + tensor input_475_axes_0 = const()[name = tensor("input_475_axes_0"), val = tensor([1])]; + tensor input_475_gamma_0_to_fp16 = const()[name = tensor("input_475_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2896636032)))]; + tensor input_475_beta_0_to_fp16 = const()[name = tensor("input_475_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2896638656)))]; + tensor var_34982_to_fp16 = const()[name = tensor("op_34982_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_475_cast_fp16 = layer_norm(axes = input_475_axes_0, beta = input_475_beta_0_to_fp16, epsilon = var_34982_to_fp16, gamma = input_475_gamma_0_to_fp16, x = inputs_233_cast_fp16)[name = tensor("input_475_cast_fp16")]; + tensor var_34998 = const()[name = tensor("op_34998"), val = tensor([1, 1])]; + tensor var_35000 = const()[name = tensor("op_35000"), val = tensor([1, 1])]; + tensor var_35002_pad_type_0 = const()[name = tensor("op_35002_pad_type_0"), val = tensor("custom")]; + tensor var_35002_pad_0 = const()[name = tensor("op_35002_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_4_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_4_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2896641280)))]; + tensor up_blocks_0_attentions_0_transformer_blocks_4_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_4_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2922855744)))]; + tensor var_35002_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_4_ff_net_0_proj_bias_to_fp16, dilations = var_35000, groups = var_30355, pad = var_35002_pad_0, pad_type = var_35002_pad_type_0, strides = var_34998, weight = up_blocks_0_attentions_0_transformer_blocks_4_ff_net_0_proj_weight_to_fp16, x = input_475_cast_fp16)[name = tensor("op_35002_cast_fp16")]; + tensor var_35003_split_sizes_0 = const()[name = tensor("op_35003_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_35003_axis_0 = const()[name = tensor("op_35003_axis_0"), val = tensor(1)]; + tensor var_35003_cast_fp16_0, tensor var_35003_cast_fp16_1 = split(axis = var_35003_axis_0, split_sizes = var_35003_split_sizes_0, x = var_35002_cast_fp16)[name = tensor("op_35003_cast_fp16")]; + tensor var_35005_mode_0 = const()[name = tensor("op_35005_mode_0"), val = tensor("EXACT")]; + tensor var_35005_cast_fp16 = gelu(mode = var_35005_mode_0, x = var_35003_cast_fp16_1)[name = tensor("op_35005_cast_fp16")]; + tensor input_477_cast_fp16 = mul(x = var_35003_cast_fp16_0, y = var_35005_cast_fp16)[name = tensor("input_477_cast_fp16")]; + tensor var_35009 = const()[name = tensor("op_35009"), val = tensor([1, 1])]; + tensor var_35011 = const()[name = tensor("op_35011"), val = tensor([1, 1])]; + tensor var_35013_pad_type_0 = const()[name = tensor("op_35013_pad_type_0"), val = tensor("custom")]; + tensor var_35013_pad_0 = const()[name = tensor("op_35013_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_4_ff_net_2_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_4_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2922876288)))]; + tensor up_blocks_0_attentions_0_transformer_blocks_4_ff_net_2_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_4_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2935983552)))]; + tensor var_35013_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_4_ff_net_2_bias_to_fp16, dilations = var_35011, groups = var_30355, pad = var_35013_pad_0, pad_type = var_35013_pad_type_0, strides = var_35009, weight = up_blocks_0_attentions_0_transformer_blocks_4_ff_net_2_weight_to_fp16, x = input_477_cast_fp16)[name = tensor("op_35013_cast_fp16")]; + tensor inputs_235_cast_fp16 = add(x = var_35013_cast_fp16, y = inputs_233_cast_fp16)[name = tensor("inputs_235_cast_fp16")]; + tensor hidden_states_317_axes_0 = const()[name = tensor("hidden_states_317_axes_0"), val = tensor([1])]; + tensor hidden_states_317_gamma_0_to_fp16 = const()[name = tensor("hidden_states_317_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2935986176)))]; + tensor hidden_states_317_beta_0_to_fp16 = const()[name = tensor("hidden_states_317_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2935988800)))]; + tensor var_35029_to_fp16 = const()[name = tensor("op_35029_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_317_cast_fp16 = layer_norm(axes = hidden_states_317_axes_0, beta = hidden_states_317_beta_0_to_fp16, epsilon = var_35029_to_fp16, gamma = hidden_states_317_gamma_0_to_fp16, x = inputs_235_cast_fp16)[name = tensor("hidden_states_317_cast_fp16")]; + tensor var_35044 = const()[name = tensor("op_35044"), val = tensor([1, 1])]; + tensor var_35046 = const()[name = tensor("op_35046"), val = tensor([1, 1])]; + tensor q_157_pad_type_0 = const()[name = tensor("q_157_pad_type_0"), val = tensor("custom")]; + tensor q_157_pad_0 = const()[name = tensor("q_157_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_5_attn1_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_5_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2935991424)))]; + tensor q_157_cast_fp16 = conv(dilations = var_35046, groups = var_30355, pad = q_157_pad_0, pad_type = q_157_pad_type_0, strides = var_35044, weight = up_blocks_0_attentions_0_transformer_blocks_5_attn1_to_q_weight_to_fp16, x = hidden_states_317_cast_fp16)[name = tensor("q_157_cast_fp16")]; + tensor var_35050 = const()[name = tensor("op_35050"), val = tensor([1, 1])]; + tensor var_35052 = const()[name = tensor("op_35052"), val = tensor([1, 1])]; + tensor k_313_pad_type_0 = const()[name = tensor("k_313_pad_type_0"), val = tensor("custom")]; + tensor k_313_pad_0 = const()[name = tensor("k_313_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_5_attn1_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_5_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2939268288)))]; + tensor k_313_cast_fp16 = conv(dilations = var_35052, groups = var_30355, pad = k_313_pad_0, pad_type = k_313_pad_type_0, strides = var_35050, weight = up_blocks_0_attentions_0_transformer_blocks_5_attn1_to_k_weight_to_fp16, x = hidden_states_317_cast_fp16)[name = tensor("k_313_cast_fp16")]; + tensor var_35056 = const()[name = tensor("op_35056"), val = tensor([1, 1])]; + tensor var_35058 = const()[name = tensor("op_35058"), val = tensor([1, 1])]; + tensor v_157_pad_type_0 = const()[name = tensor("v_157_pad_type_0"), val = tensor("custom")]; + tensor v_157_pad_0 = const()[name = tensor("v_157_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_5_attn1_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_5_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2942545152)))]; + tensor v_157_cast_fp16 = conv(dilations = var_35058, groups = var_30355, pad = v_157_pad_0, pad_type = v_157_pad_type_0, strides = var_35056, weight = up_blocks_0_attentions_0_transformer_blocks_5_attn1_to_v_weight_to_fp16, x = hidden_states_317_cast_fp16)[name = tensor("v_157_cast_fp16")]; + tensor var_35062_begin_0 = const()[name = tensor("op_35062_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_35062_end_0 = const()[name = tensor("op_35062_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_35062_end_mask_0 = const()[name = tensor("op_35062_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35062_cast_fp16 = slice_by_index(begin = var_35062_begin_0, end = var_35062_end_0, end_mask = var_35062_end_mask_0, x = q_157_cast_fp16)[name = tensor("op_35062_cast_fp16")]; + tensor var_35066_begin_0 = const()[name = tensor("op_35066_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_35066_end_0 = const()[name = tensor("op_35066_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_35066_end_mask_0 = const()[name = tensor("op_35066_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35066_cast_fp16 = slice_by_index(begin = var_35066_begin_0, end = var_35066_end_0, end_mask = var_35066_end_mask_0, x = q_157_cast_fp16)[name = tensor("op_35066_cast_fp16")]; + tensor var_35070_begin_0 = const()[name = tensor("op_35070_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_35070_end_0 = const()[name = tensor("op_35070_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_35070_end_mask_0 = const()[name = tensor("op_35070_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35070_cast_fp16 = slice_by_index(begin = var_35070_begin_0, end = var_35070_end_0, end_mask = var_35070_end_mask_0, x = q_157_cast_fp16)[name = tensor("op_35070_cast_fp16")]; + tensor var_35074_begin_0 = const()[name = tensor("op_35074_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_35074_end_0 = const()[name = tensor("op_35074_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_35074_end_mask_0 = const()[name = tensor("op_35074_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35074_cast_fp16 = slice_by_index(begin = var_35074_begin_0, end = var_35074_end_0, end_mask = var_35074_end_mask_0, x = q_157_cast_fp16)[name = tensor("op_35074_cast_fp16")]; + tensor var_35078_begin_0 = const()[name = tensor("op_35078_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_35078_end_0 = const()[name = tensor("op_35078_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_35078_end_mask_0 = const()[name = tensor("op_35078_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35078_cast_fp16 = slice_by_index(begin = var_35078_begin_0, end = var_35078_end_0, end_mask = var_35078_end_mask_0, x = q_157_cast_fp16)[name = tensor("op_35078_cast_fp16")]; + tensor var_35082_begin_0 = const()[name = tensor("op_35082_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_35082_end_0 = const()[name = tensor("op_35082_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_35082_end_mask_0 = const()[name = tensor("op_35082_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35082_cast_fp16 = slice_by_index(begin = var_35082_begin_0, end = var_35082_end_0, end_mask = var_35082_end_mask_0, x = q_157_cast_fp16)[name = tensor("op_35082_cast_fp16")]; + tensor var_35086_begin_0 = const()[name = tensor("op_35086_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_35086_end_0 = const()[name = tensor("op_35086_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_35086_end_mask_0 = const()[name = tensor("op_35086_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35086_cast_fp16 = slice_by_index(begin = var_35086_begin_0, end = var_35086_end_0, end_mask = var_35086_end_mask_0, x = q_157_cast_fp16)[name = tensor("op_35086_cast_fp16")]; + tensor var_35090_begin_0 = const()[name = tensor("op_35090_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_35090_end_0 = const()[name = tensor("op_35090_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_35090_end_mask_0 = const()[name = tensor("op_35090_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35090_cast_fp16 = slice_by_index(begin = var_35090_begin_0, end = var_35090_end_0, end_mask = var_35090_end_mask_0, x = q_157_cast_fp16)[name = tensor("op_35090_cast_fp16")]; + tensor var_35094_begin_0 = const()[name = tensor("op_35094_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_35094_end_0 = const()[name = tensor("op_35094_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_35094_end_mask_0 = const()[name = tensor("op_35094_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35094_cast_fp16 = slice_by_index(begin = var_35094_begin_0, end = var_35094_end_0, end_mask = var_35094_end_mask_0, x = q_157_cast_fp16)[name = tensor("op_35094_cast_fp16")]; + tensor var_35098_begin_0 = const()[name = tensor("op_35098_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_35098_end_0 = const()[name = tensor("op_35098_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_35098_end_mask_0 = const()[name = tensor("op_35098_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35098_cast_fp16 = slice_by_index(begin = var_35098_begin_0, end = var_35098_end_0, end_mask = var_35098_end_mask_0, x = q_157_cast_fp16)[name = tensor("op_35098_cast_fp16")]; + tensor var_35102_begin_0 = const()[name = tensor("op_35102_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_35102_end_0 = const()[name = tensor("op_35102_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_35102_end_mask_0 = const()[name = tensor("op_35102_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35102_cast_fp16 = slice_by_index(begin = var_35102_begin_0, end = var_35102_end_0, end_mask = var_35102_end_mask_0, x = q_157_cast_fp16)[name = tensor("op_35102_cast_fp16")]; + tensor var_35106_begin_0 = const()[name = tensor("op_35106_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_35106_end_0 = const()[name = tensor("op_35106_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_35106_end_mask_0 = const()[name = tensor("op_35106_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35106_cast_fp16 = slice_by_index(begin = var_35106_begin_0, end = var_35106_end_0, end_mask = var_35106_end_mask_0, x = q_157_cast_fp16)[name = tensor("op_35106_cast_fp16")]; + tensor var_35110_begin_0 = const()[name = tensor("op_35110_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_35110_end_0 = const()[name = tensor("op_35110_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_35110_end_mask_0 = const()[name = tensor("op_35110_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35110_cast_fp16 = slice_by_index(begin = var_35110_begin_0, end = var_35110_end_0, end_mask = var_35110_end_mask_0, x = q_157_cast_fp16)[name = tensor("op_35110_cast_fp16")]; + tensor var_35114_begin_0 = const()[name = tensor("op_35114_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_35114_end_0 = const()[name = tensor("op_35114_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_35114_end_mask_0 = const()[name = tensor("op_35114_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35114_cast_fp16 = slice_by_index(begin = var_35114_begin_0, end = var_35114_end_0, end_mask = var_35114_end_mask_0, x = q_157_cast_fp16)[name = tensor("op_35114_cast_fp16")]; + tensor var_35118_begin_0 = const()[name = tensor("op_35118_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_35118_end_0 = const()[name = tensor("op_35118_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_35118_end_mask_0 = const()[name = tensor("op_35118_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35118_cast_fp16 = slice_by_index(begin = var_35118_begin_0, end = var_35118_end_0, end_mask = var_35118_end_mask_0, x = q_157_cast_fp16)[name = tensor("op_35118_cast_fp16")]; + tensor var_35122_begin_0 = const()[name = tensor("op_35122_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_35122_end_0 = const()[name = tensor("op_35122_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_35122_end_mask_0 = const()[name = tensor("op_35122_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35122_cast_fp16 = slice_by_index(begin = var_35122_begin_0, end = var_35122_end_0, end_mask = var_35122_end_mask_0, x = q_157_cast_fp16)[name = tensor("op_35122_cast_fp16")]; + tensor var_35126_begin_0 = const()[name = tensor("op_35126_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_35126_end_0 = const()[name = tensor("op_35126_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_35126_end_mask_0 = const()[name = tensor("op_35126_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35126_cast_fp16 = slice_by_index(begin = var_35126_begin_0, end = var_35126_end_0, end_mask = var_35126_end_mask_0, x = q_157_cast_fp16)[name = tensor("op_35126_cast_fp16")]; + tensor var_35130_begin_0 = const()[name = tensor("op_35130_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_35130_end_0 = const()[name = tensor("op_35130_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_35130_end_mask_0 = const()[name = tensor("op_35130_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35130_cast_fp16 = slice_by_index(begin = var_35130_begin_0, end = var_35130_end_0, end_mask = var_35130_end_mask_0, x = q_157_cast_fp16)[name = tensor("op_35130_cast_fp16")]; + tensor var_35134_begin_0 = const()[name = tensor("op_35134_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_35134_end_0 = const()[name = tensor("op_35134_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_35134_end_mask_0 = const()[name = tensor("op_35134_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35134_cast_fp16 = slice_by_index(begin = var_35134_begin_0, end = var_35134_end_0, end_mask = var_35134_end_mask_0, x = q_157_cast_fp16)[name = tensor("op_35134_cast_fp16")]; + tensor var_35138_begin_0 = const()[name = tensor("op_35138_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_35138_end_0 = const()[name = tensor("op_35138_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_35138_end_mask_0 = const()[name = tensor("op_35138_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35138_cast_fp16 = slice_by_index(begin = var_35138_begin_0, end = var_35138_end_0, end_mask = var_35138_end_mask_0, x = q_157_cast_fp16)[name = tensor("op_35138_cast_fp16")]; + tensor k_315_perm_0 = const()[name = tensor("k_315_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_35145_begin_0 = const()[name = tensor("op_35145_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_35145_end_0 = const()[name = tensor("op_35145_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_35145_end_mask_0 = const()[name = tensor("op_35145_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_315_cast_fp16 = transpose(perm = k_315_perm_0, x = k_313_cast_fp16)[name = tensor("transpose_61")]; + tensor var_35145_cast_fp16 = slice_by_index(begin = var_35145_begin_0, end = var_35145_end_0, end_mask = var_35145_end_mask_0, x = k_315_cast_fp16)[name = tensor("op_35145_cast_fp16")]; + tensor var_35149_begin_0 = const()[name = tensor("op_35149_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_35149_end_0 = const()[name = tensor("op_35149_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_35149_end_mask_0 = const()[name = tensor("op_35149_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35149_cast_fp16 = slice_by_index(begin = var_35149_begin_0, end = var_35149_end_0, end_mask = var_35149_end_mask_0, x = k_315_cast_fp16)[name = tensor("op_35149_cast_fp16")]; + tensor var_35153_begin_0 = const()[name = tensor("op_35153_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_35153_end_0 = const()[name = tensor("op_35153_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_35153_end_mask_0 = const()[name = tensor("op_35153_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35153_cast_fp16 = slice_by_index(begin = var_35153_begin_0, end = var_35153_end_0, end_mask = var_35153_end_mask_0, x = k_315_cast_fp16)[name = tensor("op_35153_cast_fp16")]; + tensor var_35157_begin_0 = const()[name = tensor("op_35157_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_35157_end_0 = const()[name = tensor("op_35157_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_35157_end_mask_0 = const()[name = tensor("op_35157_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35157_cast_fp16 = slice_by_index(begin = var_35157_begin_0, end = var_35157_end_0, end_mask = var_35157_end_mask_0, x = k_315_cast_fp16)[name = tensor("op_35157_cast_fp16")]; + tensor var_35161_begin_0 = const()[name = tensor("op_35161_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_35161_end_0 = const()[name = tensor("op_35161_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_35161_end_mask_0 = const()[name = tensor("op_35161_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35161_cast_fp16 = slice_by_index(begin = var_35161_begin_0, end = var_35161_end_0, end_mask = var_35161_end_mask_0, x = k_315_cast_fp16)[name = tensor("op_35161_cast_fp16")]; + tensor var_35165_begin_0 = const()[name = tensor("op_35165_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_35165_end_0 = const()[name = tensor("op_35165_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_35165_end_mask_0 = const()[name = tensor("op_35165_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35165_cast_fp16 = slice_by_index(begin = var_35165_begin_0, end = var_35165_end_0, end_mask = var_35165_end_mask_0, x = k_315_cast_fp16)[name = tensor("op_35165_cast_fp16")]; + tensor var_35169_begin_0 = const()[name = tensor("op_35169_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_35169_end_0 = const()[name = tensor("op_35169_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_35169_end_mask_0 = const()[name = tensor("op_35169_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35169_cast_fp16 = slice_by_index(begin = var_35169_begin_0, end = var_35169_end_0, end_mask = var_35169_end_mask_0, x = k_315_cast_fp16)[name = tensor("op_35169_cast_fp16")]; + tensor var_35173_begin_0 = const()[name = tensor("op_35173_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_35173_end_0 = const()[name = tensor("op_35173_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_35173_end_mask_0 = const()[name = tensor("op_35173_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35173_cast_fp16 = slice_by_index(begin = var_35173_begin_0, end = var_35173_end_0, end_mask = var_35173_end_mask_0, x = k_315_cast_fp16)[name = tensor("op_35173_cast_fp16")]; + tensor var_35177_begin_0 = const()[name = tensor("op_35177_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_35177_end_0 = const()[name = tensor("op_35177_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_35177_end_mask_0 = const()[name = tensor("op_35177_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35177_cast_fp16 = slice_by_index(begin = var_35177_begin_0, end = var_35177_end_0, end_mask = var_35177_end_mask_0, x = k_315_cast_fp16)[name = tensor("op_35177_cast_fp16")]; + tensor var_35181_begin_0 = const()[name = tensor("op_35181_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_35181_end_0 = const()[name = tensor("op_35181_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_35181_end_mask_0 = const()[name = tensor("op_35181_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35181_cast_fp16 = slice_by_index(begin = var_35181_begin_0, end = var_35181_end_0, end_mask = var_35181_end_mask_0, x = k_315_cast_fp16)[name = tensor("op_35181_cast_fp16")]; + tensor var_35185_begin_0 = const()[name = tensor("op_35185_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_35185_end_0 = const()[name = tensor("op_35185_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_35185_end_mask_0 = const()[name = tensor("op_35185_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35185_cast_fp16 = slice_by_index(begin = var_35185_begin_0, end = var_35185_end_0, end_mask = var_35185_end_mask_0, x = k_315_cast_fp16)[name = tensor("op_35185_cast_fp16")]; + tensor var_35189_begin_0 = const()[name = tensor("op_35189_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_35189_end_0 = const()[name = tensor("op_35189_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_35189_end_mask_0 = const()[name = tensor("op_35189_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35189_cast_fp16 = slice_by_index(begin = var_35189_begin_0, end = var_35189_end_0, end_mask = var_35189_end_mask_0, x = k_315_cast_fp16)[name = tensor("op_35189_cast_fp16")]; + tensor var_35193_begin_0 = const()[name = tensor("op_35193_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_35193_end_0 = const()[name = tensor("op_35193_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_35193_end_mask_0 = const()[name = tensor("op_35193_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35193_cast_fp16 = slice_by_index(begin = var_35193_begin_0, end = var_35193_end_0, end_mask = var_35193_end_mask_0, x = k_315_cast_fp16)[name = tensor("op_35193_cast_fp16")]; + tensor var_35197_begin_0 = const()[name = tensor("op_35197_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_35197_end_0 = const()[name = tensor("op_35197_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_35197_end_mask_0 = const()[name = tensor("op_35197_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35197_cast_fp16 = slice_by_index(begin = var_35197_begin_0, end = var_35197_end_0, end_mask = var_35197_end_mask_0, x = k_315_cast_fp16)[name = tensor("op_35197_cast_fp16")]; + tensor var_35201_begin_0 = const()[name = tensor("op_35201_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_35201_end_0 = const()[name = tensor("op_35201_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_35201_end_mask_0 = const()[name = tensor("op_35201_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35201_cast_fp16 = slice_by_index(begin = var_35201_begin_0, end = var_35201_end_0, end_mask = var_35201_end_mask_0, x = k_315_cast_fp16)[name = tensor("op_35201_cast_fp16")]; + tensor var_35205_begin_0 = const()[name = tensor("op_35205_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_35205_end_0 = const()[name = tensor("op_35205_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_35205_end_mask_0 = const()[name = tensor("op_35205_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35205_cast_fp16 = slice_by_index(begin = var_35205_begin_0, end = var_35205_end_0, end_mask = var_35205_end_mask_0, x = k_315_cast_fp16)[name = tensor("op_35205_cast_fp16")]; + tensor var_35209_begin_0 = const()[name = tensor("op_35209_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_35209_end_0 = const()[name = tensor("op_35209_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_35209_end_mask_0 = const()[name = tensor("op_35209_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35209_cast_fp16 = slice_by_index(begin = var_35209_begin_0, end = var_35209_end_0, end_mask = var_35209_end_mask_0, x = k_315_cast_fp16)[name = tensor("op_35209_cast_fp16")]; + tensor var_35213_begin_0 = const()[name = tensor("op_35213_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_35213_end_0 = const()[name = tensor("op_35213_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_35213_end_mask_0 = const()[name = tensor("op_35213_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35213_cast_fp16 = slice_by_index(begin = var_35213_begin_0, end = var_35213_end_0, end_mask = var_35213_end_mask_0, x = k_315_cast_fp16)[name = tensor("op_35213_cast_fp16")]; + tensor var_35217_begin_0 = const()[name = tensor("op_35217_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_35217_end_0 = const()[name = tensor("op_35217_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_35217_end_mask_0 = const()[name = tensor("op_35217_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35217_cast_fp16 = slice_by_index(begin = var_35217_begin_0, end = var_35217_end_0, end_mask = var_35217_end_mask_0, x = k_315_cast_fp16)[name = tensor("op_35217_cast_fp16")]; + tensor var_35221_begin_0 = const()[name = tensor("op_35221_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_35221_end_0 = const()[name = tensor("op_35221_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_35221_end_mask_0 = const()[name = tensor("op_35221_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35221_cast_fp16 = slice_by_index(begin = var_35221_begin_0, end = var_35221_end_0, end_mask = var_35221_end_mask_0, x = k_315_cast_fp16)[name = tensor("op_35221_cast_fp16")]; + tensor var_35223_begin_0 = const()[name = tensor("op_35223_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_35223_end_0 = const()[name = tensor("op_35223_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_35223_end_mask_0 = const()[name = tensor("op_35223_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35223_cast_fp16 = slice_by_index(begin = var_35223_begin_0, end = var_35223_end_0, end_mask = var_35223_end_mask_0, x = v_157_cast_fp16)[name = tensor("op_35223_cast_fp16")]; + tensor var_35227_begin_0 = const()[name = tensor("op_35227_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_35227_end_0 = const()[name = tensor("op_35227_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_35227_end_mask_0 = const()[name = tensor("op_35227_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35227_cast_fp16 = slice_by_index(begin = var_35227_begin_0, end = var_35227_end_0, end_mask = var_35227_end_mask_0, x = v_157_cast_fp16)[name = tensor("op_35227_cast_fp16")]; + tensor var_35231_begin_0 = const()[name = tensor("op_35231_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_35231_end_0 = const()[name = tensor("op_35231_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_35231_end_mask_0 = const()[name = tensor("op_35231_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35231_cast_fp16 = slice_by_index(begin = var_35231_begin_0, end = var_35231_end_0, end_mask = var_35231_end_mask_0, x = v_157_cast_fp16)[name = tensor("op_35231_cast_fp16")]; + tensor var_35235_begin_0 = const()[name = tensor("op_35235_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_35235_end_0 = const()[name = tensor("op_35235_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_35235_end_mask_0 = const()[name = tensor("op_35235_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35235_cast_fp16 = slice_by_index(begin = var_35235_begin_0, end = var_35235_end_0, end_mask = var_35235_end_mask_0, x = v_157_cast_fp16)[name = tensor("op_35235_cast_fp16")]; + tensor var_35239_begin_0 = const()[name = tensor("op_35239_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_35239_end_0 = const()[name = tensor("op_35239_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_35239_end_mask_0 = const()[name = tensor("op_35239_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35239_cast_fp16 = slice_by_index(begin = var_35239_begin_0, end = var_35239_end_0, end_mask = var_35239_end_mask_0, x = v_157_cast_fp16)[name = tensor("op_35239_cast_fp16")]; + tensor var_35243_begin_0 = const()[name = tensor("op_35243_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_35243_end_0 = const()[name = tensor("op_35243_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_35243_end_mask_0 = const()[name = tensor("op_35243_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35243_cast_fp16 = slice_by_index(begin = var_35243_begin_0, end = var_35243_end_0, end_mask = var_35243_end_mask_0, x = v_157_cast_fp16)[name = tensor("op_35243_cast_fp16")]; + tensor var_35247_begin_0 = const()[name = tensor("op_35247_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_35247_end_0 = const()[name = tensor("op_35247_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_35247_end_mask_0 = const()[name = tensor("op_35247_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35247_cast_fp16 = slice_by_index(begin = var_35247_begin_0, end = var_35247_end_0, end_mask = var_35247_end_mask_0, x = v_157_cast_fp16)[name = tensor("op_35247_cast_fp16")]; + tensor var_35251_begin_0 = const()[name = tensor("op_35251_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_35251_end_0 = const()[name = tensor("op_35251_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_35251_end_mask_0 = const()[name = tensor("op_35251_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35251_cast_fp16 = slice_by_index(begin = var_35251_begin_0, end = var_35251_end_0, end_mask = var_35251_end_mask_0, x = v_157_cast_fp16)[name = tensor("op_35251_cast_fp16")]; + tensor var_35255_begin_0 = const()[name = tensor("op_35255_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_35255_end_0 = const()[name = tensor("op_35255_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_35255_end_mask_0 = const()[name = tensor("op_35255_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35255_cast_fp16 = slice_by_index(begin = var_35255_begin_0, end = var_35255_end_0, end_mask = var_35255_end_mask_0, x = v_157_cast_fp16)[name = tensor("op_35255_cast_fp16")]; + tensor var_35259_begin_0 = const()[name = tensor("op_35259_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_35259_end_0 = const()[name = tensor("op_35259_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_35259_end_mask_0 = const()[name = tensor("op_35259_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35259_cast_fp16 = slice_by_index(begin = var_35259_begin_0, end = var_35259_end_0, end_mask = var_35259_end_mask_0, x = v_157_cast_fp16)[name = tensor("op_35259_cast_fp16")]; + tensor var_35263_begin_0 = const()[name = tensor("op_35263_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_35263_end_0 = const()[name = tensor("op_35263_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_35263_end_mask_0 = const()[name = tensor("op_35263_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35263_cast_fp16 = slice_by_index(begin = var_35263_begin_0, end = var_35263_end_0, end_mask = var_35263_end_mask_0, x = v_157_cast_fp16)[name = tensor("op_35263_cast_fp16")]; + tensor var_35267_begin_0 = const()[name = tensor("op_35267_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_35267_end_0 = const()[name = tensor("op_35267_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_35267_end_mask_0 = const()[name = tensor("op_35267_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35267_cast_fp16 = slice_by_index(begin = var_35267_begin_0, end = var_35267_end_0, end_mask = var_35267_end_mask_0, x = v_157_cast_fp16)[name = tensor("op_35267_cast_fp16")]; + tensor var_35271_begin_0 = const()[name = tensor("op_35271_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_35271_end_0 = const()[name = tensor("op_35271_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_35271_end_mask_0 = const()[name = tensor("op_35271_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35271_cast_fp16 = slice_by_index(begin = var_35271_begin_0, end = var_35271_end_0, end_mask = var_35271_end_mask_0, x = v_157_cast_fp16)[name = tensor("op_35271_cast_fp16")]; + tensor var_35275_begin_0 = const()[name = tensor("op_35275_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_35275_end_0 = const()[name = tensor("op_35275_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_35275_end_mask_0 = const()[name = tensor("op_35275_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35275_cast_fp16 = slice_by_index(begin = var_35275_begin_0, end = var_35275_end_0, end_mask = var_35275_end_mask_0, x = v_157_cast_fp16)[name = tensor("op_35275_cast_fp16")]; + tensor var_35279_begin_0 = const()[name = tensor("op_35279_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_35279_end_0 = const()[name = tensor("op_35279_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_35279_end_mask_0 = const()[name = tensor("op_35279_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35279_cast_fp16 = slice_by_index(begin = var_35279_begin_0, end = var_35279_end_0, end_mask = var_35279_end_mask_0, x = v_157_cast_fp16)[name = tensor("op_35279_cast_fp16")]; + tensor var_35283_begin_0 = const()[name = tensor("op_35283_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_35283_end_0 = const()[name = tensor("op_35283_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_35283_end_mask_0 = const()[name = tensor("op_35283_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35283_cast_fp16 = slice_by_index(begin = var_35283_begin_0, end = var_35283_end_0, end_mask = var_35283_end_mask_0, x = v_157_cast_fp16)[name = tensor("op_35283_cast_fp16")]; + tensor var_35287_begin_0 = const()[name = tensor("op_35287_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_35287_end_0 = const()[name = tensor("op_35287_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_35287_end_mask_0 = const()[name = tensor("op_35287_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35287_cast_fp16 = slice_by_index(begin = var_35287_begin_0, end = var_35287_end_0, end_mask = var_35287_end_mask_0, x = v_157_cast_fp16)[name = tensor("op_35287_cast_fp16")]; + tensor var_35291_begin_0 = const()[name = tensor("op_35291_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_35291_end_0 = const()[name = tensor("op_35291_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_35291_end_mask_0 = const()[name = tensor("op_35291_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35291_cast_fp16 = slice_by_index(begin = var_35291_begin_0, end = var_35291_end_0, end_mask = var_35291_end_mask_0, x = v_157_cast_fp16)[name = tensor("op_35291_cast_fp16")]; + tensor var_35295_begin_0 = const()[name = tensor("op_35295_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_35295_end_0 = const()[name = tensor("op_35295_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_35295_end_mask_0 = const()[name = tensor("op_35295_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35295_cast_fp16 = slice_by_index(begin = var_35295_begin_0, end = var_35295_end_0, end_mask = var_35295_end_mask_0, x = v_157_cast_fp16)[name = tensor("op_35295_cast_fp16")]; + tensor var_35299_begin_0 = const()[name = tensor("op_35299_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_35299_end_0 = const()[name = tensor("op_35299_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_35299_end_mask_0 = const()[name = tensor("op_35299_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35299_cast_fp16 = slice_by_index(begin = var_35299_begin_0, end = var_35299_end_0, end_mask = var_35299_end_mask_0, x = v_157_cast_fp16)[name = tensor("op_35299_cast_fp16")]; + tensor var_35303_equation_0 = const()[name = tensor("op_35303_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35303_cast_fp16 = einsum(equation = var_35303_equation_0, values = (var_35145_cast_fp16, var_35062_cast_fp16))[name = tensor("op_35303_cast_fp16")]; + tensor var_35304_to_fp16 = const()[name = tensor("op_35304_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2961_cast_fp16 = mul(x = var_35303_cast_fp16, y = var_35304_to_fp16)[name = tensor("aw_2961_cast_fp16")]; + tensor var_35307_equation_0 = const()[name = tensor("op_35307_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35307_cast_fp16 = einsum(equation = var_35307_equation_0, values = (var_35149_cast_fp16, var_35066_cast_fp16))[name = tensor("op_35307_cast_fp16")]; + tensor var_35308_to_fp16 = const()[name = tensor("op_35308_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2963_cast_fp16 = mul(x = var_35307_cast_fp16, y = var_35308_to_fp16)[name = tensor("aw_2963_cast_fp16")]; + tensor var_35311_equation_0 = const()[name = tensor("op_35311_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35311_cast_fp16 = einsum(equation = var_35311_equation_0, values = (var_35153_cast_fp16, var_35070_cast_fp16))[name = tensor("op_35311_cast_fp16")]; + tensor var_35312_to_fp16 = const()[name = tensor("op_35312_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2965_cast_fp16 = mul(x = var_35311_cast_fp16, y = var_35312_to_fp16)[name = tensor("aw_2965_cast_fp16")]; + tensor var_35315_equation_0 = const()[name = tensor("op_35315_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35315_cast_fp16 = einsum(equation = var_35315_equation_0, values = (var_35157_cast_fp16, var_35074_cast_fp16))[name = tensor("op_35315_cast_fp16")]; + tensor var_35316_to_fp16 = const()[name = tensor("op_35316_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2967_cast_fp16 = mul(x = var_35315_cast_fp16, y = var_35316_to_fp16)[name = tensor("aw_2967_cast_fp16")]; + tensor var_35319_equation_0 = const()[name = tensor("op_35319_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35319_cast_fp16 = einsum(equation = var_35319_equation_0, values = (var_35161_cast_fp16, var_35078_cast_fp16))[name = tensor("op_35319_cast_fp16")]; + tensor var_35320_to_fp16 = const()[name = tensor("op_35320_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2969_cast_fp16 = mul(x = var_35319_cast_fp16, y = var_35320_to_fp16)[name = tensor("aw_2969_cast_fp16")]; + tensor var_35323_equation_0 = const()[name = tensor("op_35323_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35323_cast_fp16 = einsum(equation = var_35323_equation_0, values = (var_35165_cast_fp16, var_35082_cast_fp16))[name = tensor("op_35323_cast_fp16")]; + tensor var_35324_to_fp16 = const()[name = tensor("op_35324_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2971_cast_fp16 = mul(x = var_35323_cast_fp16, y = var_35324_to_fp16)[name = tensor("aw_2971_cast_fp16")]; + tensor var_35327_equation_0 = const()[name = tensor("op_35327_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35327_cast_fp16 = einsum(equation = var_35327_equation_0, values = (var_35169_cast_fp16, var_35086_cast_fp16))[name = tensor("op_35327_cast_fp16")]; + tensor var_35328_to_fp16 = const()[name = tensor("op_35328_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2973_cast_fp16 = mul(x = var_35327_cast_fp16, y = var_35328_to_fp16)[name = tensor("aw_2973_cast_fp16")]; + tensor var_35331_equation_0 = const()[name = tensor("op_35331_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35331_cast_fp16 = einsum(equation = var_35331_equation_0, values = (var_35173_cast_fp16, var_35090_cast_fp16))[name = tensor("op_35331_cast_fp16")]; + tensor var_35332_to_fp16 = const()[name = tensor("op_35332_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2975_cast_fp16 = mul(x = var_35331_cast_fp16, y = var_35332_to_fp16)[name = tensor("aw_2975_cast_fp16")]; + tensor var_35335_equation_0 = const()[name = tensor("op_35335_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35335_cast_fp16 = einsum(equation = var_35335_equation_0, values = (var_35177_cast_fp16, var_35094_cast_fp16))[name = tensor("op_35335_cast_fp16")]; + tensor var_35336_to_fp16 = const()[name = tensor("op_35336_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2977_cast_fp16 = mul(x = var_35335_cast_fp16, y = var_35336_to_fp16)[name = tensor("aw_2977_cast_fp16")]; + tensor var_35339_equation_0 = const()[name = tensor("op_35339_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35339_cast_fp16 = einsum(equation = var_35339_equation_0, values = (var_35181_cast_fp16, var_35098_cast_fp16))[name = tensor("op_35339_cast_fp16")]; + tensor var_35340_to_fp16 = const()[name = tensor("op_35340_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2979_cast_fp16 = mul(x = var_35339_cast_fp16, y = var_35340_to_fp16)[name = tensor("aw_2979_cast_fp16")]; + tensor var_35343_equation_0 = const()[name = tensor("op_35343_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35343_cast_fp16 = einsum(equation = var_35343_equation_0, values = (var_35185_cast_fp16, var_35102_cast_fp16))[name = tensor("op_35343_cast_fp16")]; + tensor var_35344_to_fp16 = const()[name = tensor("op_35344_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2981_cast_fp16 = mul(x = var_35343_cast_fp16, y = var_35344_to_fp16)[name = tensor("aw_2981_cast_fp16")]; + tensor var_35347_equation_0 = const()[name = tensor("op_35347_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35347_cast_fp16 = einsum(equation = var_35347_equation_0, values = (var_35189_cast_fp16, var_35106_cast_fp16))[name = tensor("op_35347_cast_fp16")]; + tensor var_35348_to_fp16 = const()[name = tensor("op_35348_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2983_cast_fp16 = mul(x = var_35347_cast_fp16, y = var_35348_to_fp16)[name = tensor("aw_2983_cast_fp16")]; + tensor var_35351_equation_0 = const()[name = tensor("op_35351_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35351_cast_fp16 = einsum(equation = var_35351_equation_0, values = (var_35193_cast_fp16, var_35110_cast_fp16))[name = tensor("op_35351_cast_fp16")]; + tensor var_35352_to_fp16 = const()[name = tensor("op_35352_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2985_cast_fp16 = mul(x = var_35351_cast_fp16, y = var_35352_to_fp16)[name = tensor("aw_2985_cast_fp16")]; + tensor var_35355_equation_0 = const()[name = tensor("op_35355_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35355_cast_fp16 = einsum(equation = var_35355_equation_0, values = (var_35197_cast_fp16, var_35114_cast_fp16))[name = tensor("op_35355_cast_fp16")]; + tensor var_35356_to_fp16 = const()[name = tensor("op_35356_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2987_cast_fp16 = mul(x = var_35355_cast_fp16, y = var_35356_to_fp16)[name = tensor("aw_2987_cast_fp16")]; + tensor var_35359_equation_0 = const()[name = tensor("op_35359_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35359_cast_fp16 = einsum(equation = var_35359_equation_0, values = (var_35201_cast_fp16, var_35118_cast_fp16))[name = tensor("op_35359_cast_fp16")]; + tensor var_35360_to_fp16 = const()[name = tensor("op_35360_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2989_cast_fp16 = mul(x = var_35359_cast_fp16, y = var_35360_to_fp16)[name = tensor("aw_2989_cast_fp16")]; + tensor var_35363_equation_0 = const()[name = tensor("op_35363_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35363_cast_fp16 = einsum(equation = var_35363_equation_0, values = (var_35205_cast_fp16, var_35122_cast_fp16))[name = tensor("op_35363_cast_fp16")]; + tensor var_35364_to_fp16 = const()[name = tensor("op_35364_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2991_cast_fp16 = mul(x = var_35363_cast_fp16, y = var_35364_to_fp16)[name = tensor("aw_2991_cast_fp16")]; + tensor var_35367_equation_0 = const()[name = tensor("op_35367_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35367_cast_fp16 = einsum(equation = var_35367_equation_0, values = (var_35209_cast_fp16, var_35126_cast_fp16))[name = tensor("op_35367_cast_fp16")]; + tensor var_35368_to_fp16 = const()[name = tensor("op_35368_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2993_cast_fp16 = mul(x = var_35367_cast_fp16, y = var_35368_to_fp16)[name = tensor("aw_2993_cast_fp16")]; + tensor var_35371_equation_0 = const()[name = tensor("op_35371_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35371_cast_fp16 = einsum(equation = var_35371_equation_0, values = (var_35213_cast_fp16, var_35130_cast_fp16))[name = tensor("op_35371_cast_fp16")]; + tensor var_35372_to_fp16 = const()[name = tensor("op_35372_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2995_cast_fp16 = mul(x = var_35371_cast_fp16, y = var_35372_to_fp16)[name = tensor("aw_2995_cast_fp16")]; + tensor var_35375_equation_0 = const()[name = tensor("op_35375_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35375_cast_fp16 = einsum(equation = var_35375_equation_0, values = (var_35217_cast_fp16, var_35134_cast_fp16))[name = tensor("op_35375_cast_fp16")]; + tensor var_35376_to_fp16 = const()[name = tensor("op_35376_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2997_cast_fp16 = mul(x = var_35375_cast_fp16, y = var_35376_to_fp16)[name = tensor("aw_2997_cast_fp16")]; + tensor var_35379_equation_0 = const()[name = tensor("op_35379_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35379_cast_fp16 = einsum(equation = var_35379_equation_0, values = (var_35221_cast_fp16, var_35138_cast_fp16))[name = tensor("op_35379_cast_fp16")]; + tensor var_35380_to_fp16 = const()[name = tensor("op_35380_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_2999_cast_fp16 = mul(x = var_35379_cast_fp16, y = var_35380_to_fp16)[name = tensor("aw_2999_cast_fp16")]; + tensor var_35382_cast_fp16 = softmax(axis = var_30355, x = aw_2961_cast_fp16)[name = tensor("op_35382_cast_fp16")]; + tensor var_35383_cast_fp16 = softmax(axis = var_30355, x = aw_2963_cast_fp16)[name = tensor("op_35383_cast_fp16")]; + tensor var_35384_cast_fp16 = softmax(axis = var_30355, x = aw_2965_cast_fp16)[name = tensor("op_35384_cast_fp16")]; + tensor var_35385_cast_fp16 = softmax(axis = var_30355, x = aw_2967_cast_fp16)[name = tensor("op_35385_cast_fp16")]; + tensor var_35386_cast_fp16 = softmax(axis = var_30355, x = aw_2969_cast_fp16)[name = tensor("op_35386_cast_fp16")]; + tensor var_35387_cast_fp16 = softmax(axis = var_30355, x = aw_2971_cast_fp16)[name = tensor("op_35387_cast_fp16")]; + tensor var_35388_cast_fp16 = softmax(axis = var_30355, x = aw_2973_cast_fp16)[name = tensor("op_35388_cast_fp16")]; + tensor var_35389_cast_fp16 = softmax(axis = var_30355, x = aw_2975_cast_fp16)[name = tensor("op_35389_cast_fp16")]; + tensor var_35390_cast_fp16 = softmax(axis = var_30355, x = aw_2977_cast_fp16)[name = tensor("op_35390_cast_fp16")]; + tensor var_35391_cast_fp16 = softmax(axis = var_30355, x = aw_2979_cast_fp16)[name = tensor("op_35391_cast_fp16")]; + tensor var_35392_cast_fp16 = softmax(axis = var_30355, x = aw_2981_cast_fp16)[name = tensor("op_35392_cast_fp16")]; + tensor var_35393_cast_fp16 = softmax(axis = var_30355, x = aw_2983_cast_fp16)[name = tensor("op_35393_cast_fp16")]; + tensor var_35394_cast_fp16 = softmax(axis = var_30355, x = aw_2985_cast_fp16)[name = tensor("op_35394_cast_fp16")]; + tensor var_35395_cast_fp16 = softmax(axis = var_30355, x = aw_2987_cast_fp16)[name = tensor("op_35395_cast_fp16")]; + tensor var_35396_cast_fp16 = softmax(axis = var_30355, x = aw_2989_cast_fp16)[name = tensor("op_35396_cast_fp16")]; + tensor var_35397_cast_fp16 = softmax(axis = var_30355, x = aw_2991_cast_fp16)[name = tensor("op_35397_cast_fp16")]; + tensor var_35398_cast_fp16 = softmax(axis = var_30355, x = aw_2993_cast_fp16)[name = tensor("op_35398_cast_fp16")]; + tensor var_35399_cast_fp16 = softmax(axis = var_30355, x = aw_2995_cast_fp16)[name = tensor("op_35399_cast_fp16")]; + tensor var_35400_cast_fp16 = softmax(axis = var_30355, x = aw_2997_cast_fp16)[name = tensor("op_35400_cast_fp16")]; + tensor var_35401_cast_fp16 = softmax(axis = var_30355, x = aw_2999_cast_fp16)[name = tensor("op_35401_cast_fp16")]; + tensor var_35403_equation_0 = const()[name = tensor("op_35403_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35403_cast_fp16 = einsum(equation = var_35403_equation_0, values = (var_35223_cast_fp16, var_35382_cast_fp16))[name = tensor("op_35403_cast_fp16")]; + tensor var_35405_equation_0 = const()[name = tensor("op_35405_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35405_cast_fp16 = einsum(equation = var_35405_equation_0, values = (var_35227_cast_fp16, var_35383_cast_fp16))[name = tensor("op_35405_cast_fp16")]; + tensor var_35407_equation_0 = const()[name = tensor("op_35407_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35407_cast_fp16 = einsum(equation = var_35407_equation_0, values = (var_35231_cast_fp16, var_35384_cast_fp16))[name = tensor("op_35407_cast_fp16")]; + tensor var_35409_equation_0 = const()[name = tensor("op_35409_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35409_cast_fp16 = einsum(equation = var_35409_equation_0, values = (var_35235_cast_fp16, var_35385_cast_fp16))[name = tensor("op_35409_cast_fp16")]; + tensor var_35411_equation_0 = const()[name = tensor("op_35411_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35411_cast_fp16 = einsum(equation = var_35411_equation_0, values = (var_35239_cast_fp16, var_35386_cast_fp16))[name = tensor("op_35411_cast_fp16")]; + tensor var_35413_equation_0 = const()[name = tensor("op_35413_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35413_cast_fp16 = einsum(equation = var_35413_equation_0, values = (var_35243_cast_fp16, var_35387_cast_fp16))[name = tensor("op_35413_cast_fp16")]; + tensor var_35415_equation_0 = const()[name = tensor("op_35415_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35415_cast_fp16 = einsum(equation = var_35415_equation_0, values = (var_35247_cast_fp16, var_35388_cast_fp16))[name = tensor("op_35415_cast_fp16")]; + tensor var_35417_equation_0 = const()[name = tensor("op_35417_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35417_cast_fp16 = einsum(equation = var_35417_equation_0, values = (var_35251_cast_fp16, var_35389_cast_fp16))[name = tensor("op_35417_cast_fp16")]; + tensor var_35419_equation_0 = const()[name = tensor("op_35419_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35419_cast_fp16 = einsum(equation = var_35419_equation_0, values = (var_35255_cast_fp16, var_35390_cast_fp16))[name = tensor("op_35419_cast_fp16")]; + tensor var_35421_equation_0 = const()[name = tensor("op_35421_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35421_cast_fp16 = einsum(equation = var_35421_equation_0, values = (var_35259_cast_fp16, var_35391_cast_fp16))[name = tensor("op_35421_cast_fp16")]; + tensor var_35423_equation_0 = const()[name = tensor("op_35423_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35423_cast_fp16 = einsum(equation = var_35423_equation_0, values = (var_35263_cast_fp16, var_35392_cast_fp16))[name = tensor("op_35423_cast_fp16")]; + tensor var_35425_equation_0 = const()[name = tensor("op_35425_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35425_cast_fp16 = einsum(equation = var_35425_equation_0, values = (var_35267_cast_fp16, var_35393_cast_fp16))[name = tensor("op_35425_cast_fp16")]; + tensor var_35427_equation_0 = const()[name = tensor("op_35427_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35427_cast_fp16 = einsum(equation = var_35427_equation_0, values = (var_35271_cast_fp16, var_35394_cast_fp16))[name = tensor("op_35427_cast_fp16")]; + tensor var_35429_equation_0 = const()[name = tensor("op_35429_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35429_cast_fp16 = einsum(equation = var_35429_equation_0, values = (var_35275_cast_fp16, var_35395_cast_fp16))[name = tensor("op_35429_cast_fp16")]; + tensor var_35431_equation_0 = const()[name = tensor("op_35431_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35431_cast_fp16 = einsum(equation = var_35431_equation_0, values = (var_35279_cast_fp16, var_35396_cast_fp16))[name = tensor("op_35431_cast_fp16")]; + tensor var_35433_equation_0 = const()[name = tensor("op_35433_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35433_cast_fp16 = einsum(equation = var_35433_equation_0, values = (var_35283_cast_fp16, var_35397_cast_fp16))[name = tensor("op_35433_cast_fp16")]; + tensor var_35435_equation_0 = const()[name = tensor("op_35435_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35435_cast_fp16 = einsum(equation = var_35435_equation_0, values = (var_35287_cast_fp16, var_35398_cast_fp16))[name = tensor("op_35435_cast_fp16")]; + tensor var_35437_equation_0 = const()[name = tensor("op_35437_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35437_cast_fp16 = einsum(equation = var_35437_equation_0, values = (var_35291_cast_fp16, var_35399_cast_fp16))[name = tensor("op_35437_cast_fp16")]; + tensor var_35439_equation_0 = const()[name = tensor("op_35439_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35439_cast_fp16 = einsum(equation = var_35439_equation_0, values = (var_35295_cast_fp16, var_35400_cast_fp16))[name = tensor("op_35439_cast_fp16")]; + tensor var_35441_equation_0 = const()[name = tensor("op_35441_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35441_cast_fp16 = einsum(equation = var_35441_equation_0, values = (var_35299_cast_fp16, var_35401_cast_fp16))[name = tensor("op_35441_cast_fp16")]; + tensor input_479_interleave_0 = const()[name = tensor("input_479_interleave_0"), val = tensor(false)]; + tensor input_479_cast_fp16 = concat(axis = var_30355, interleave = input_479_interleave_0, values = (var_35403_cast_fp16, var_35405_cast_fp16, var_35407_cast_fp16, var_35409_cast_fp16, var_35411_cast_fp16, var_35413_cast_fp16, var_35415_cast_fp16, var_35417_cast_fp16, var_35419_cast_fp16, var_35421_cast_fp16, var_35423_cast_fp16, var_35425_cast_fp16, var_35427_cast_fp16, var_35429_cast_fp16, var_35431_cast_fp16, var_35433_cast_fp16, var_35435_cast_fp16, var_35437_cast_fp16, var_35439_cast_fp16, var_35441_cast_fp16))[name = tensor("input_479_cast_fp16")]; + tensor var_35447 = const()[name = tensor("op_35447"), val = tensor([1, 1])]; + tensor var_35449 = const()[name = tensor("op_35449"), val = tensor([1, 1])]; + tensor var_35451_pad_type_0 = const()[name = tensor("op_35451_pad_type_0"), val = tensor("custom")]; + tensor var_35451_pad_0 = const()[name = tensor("op_35451_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_5_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_5_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2945822016)))]; + tensor up_blocks_0_attentions_0_transformer_blocks_5_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_5_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2949098880)))]; + tensor var_35451_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_5_attn1_to_out_0_bias_to_fp16, dilations = var_35449, groups = var_30355, pad = var_35451_pad_0, pad_type = var_35451_pad_type_0, strides = var_35447, weight = up_blocks_0_attentions_0_transformer_blocks_5_attn1_to_out_0_weight_to_fp16, x = input_479_cast_fp16)[name = tensor("op_35451_cast_fp16")]; + tensor inputs_237_cast_fp16 = add(x = var_35451_cast_fp16, y = inputs_235_cast_fp16)[name = tensor("inputs_237_cast_fp16")]; + tensor hidden_states_319_axes_0 = const()[name = tensor("hidden_states_319_axes_0"), val = tensor([1])]; + tensor hidden_states_319_gamma_0_to_fp16 = const()[name = tensor("hidden_states_319_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2949101504)))]; + tensor hidden_states_319_beta_0_to_fp16 = const()[name = tensor("hidden_states_319_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2949104128)))]; + tensor var_35461_to_fp16 = const()[name = tensor("op_35461_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_319_cast_fp16 = layer_norm(axes = hidden_states_319_axes_0, beta = hidden_states_319_beta_0_to_fp16, epsilon = var_35461_to_fp16, gamma = hidden_states_319_gamma_0_to_fp16, x = inputs_237_cast_fp16)[name = tensor("hidden_states_319_cast_fp16")]; + tensor var_35476 = const()[name = tensor("op_35476"), val = tensor([1, 1])]; + tensor var_35478 = const()[name = tensor("op_35478"), val = tensor([1, 1])]; + tensor q_159_pad_type_0 = const()[name = tensor("q_159_pad_type_0"), val = tensor("custom")]; + tensor q_159_pad_0 = const()[name = tensor("q_159_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_5_attn2_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_5_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2949106752)))]; + tensor q_159_cast_fp16 = conv(dilations = var_35478, groups = var_30355, pad = q_159_pad_0, pad_type = q_159_pad_type_0, strides = var_35476, weight = up_blocks_0_attentions_0_transformer_blocks_5_attn2_to_q_weight_to_fp16, x = hidden_states_319_cast_fp16)[name = tensor("q_159_cast_fp16")]; + tensor var_35482 = const()[name = tensor("op_35482"), val = tensor([1, 1])]; + tensor var_35484 = const()[name = tensor("op_35484"), val = tensor([1, 1])]; + tensor k_317_pad_type_0 = const()[name = tensor("k_317_pad_type_0"), val = tensor("custom")]; + tensor k_317_pad_0 = const()[name = tensor("k_317_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_5_attn2_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_5_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2952383616)))]; + tensor k_317_cast_fp16 = conv(dilations = var_35484, groups = var_30355, pad = k_317_pad_0, pad_type = k_317_pad_type_0, strides = var_35482, weight = up_blocks_0_attentions_0_transformer_blocks_5_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_317_cast_fp16")]; + tensor var_35488 = const()[name = tensor("op_35488"), val = tensor([1, 1])]; + tensor var_35490 = const()[name = tensor("op_35490"), val = tensor([1, 1])]; + tensor v_159_pad_type_0 = const()[name = tensor("v_159_pad_type_0"), val = tensor("custom")]; + tensor v_159_pad_0 = const()[name = tensor("v_159_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_5_attn2_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_5_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2957626560)))]; + tensor v_159_cast_fp16 = conv(dilations = var_35490, groups = var_30355, pad = v_159_pad_0, pad_type = v_159_pad_type_0, strides = var_35488, weight = up_blocks_0_attentions_0_transformer_blocks_5_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_159_cast_fp16")]; + tensor var_35494_begin_0 = const()[name = tensor("op_35494_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_35494_end_0 = const()[name = tensor("op_35494_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_35494_end_mask_0 = const()[name = tensor("op_35494_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35494_cast_fp16 = slice_by_index(begin = var_35494_begin_0, end = var_35494_end_0, end_mask = var_35494_end_mask_0, x = q_159_cast_fp16)[name = tensor("op_35494_cast_fp16")]; + tensor var_35498_begin_0 = const()[name = tensor("op_35498_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_35498_end_0 = const()[name = tensor("op_35498_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_35498_end_mask_0 = const()[name = tensor("op_35498_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35498_cast_fp16 = slice_by_index(begin = var_35498_begin_0, end = var_35498_end_0, end_mask = var_35498_end_mask_0, x = q_159_cast_fp16)[name = tensor("op_35498_cast_fp16")]; + tensor var_35502_begin_0 = const()[name = tensor("op_35502_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_35502_end_0 = const()[name = tensor("op_35502_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_35502_end_mask_0 = const()[name = tensor("op_35502_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35502_cast_fp16 = slice_by_index(begin = var_35502_begin_0, end = var_35502_end_0, end_mask = var_35502_end_mask_0, x = q_159_cast_fp16)[name = tensor("op_35502_cast_fp16")]; + tensor var_35506_begin_0 = const()[name = tensor("op_35506_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_35506_end_0 = const()[name = tensor("op_35506_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_35506_end_mask_0 = const()[name = tensor("op_35506_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35506_cast_fp16 = slice_by_index(begin = var_35506_begin_0, end = var_35506_end_0, end_mask = var_35506_end_mask_0, x = q_159_cast_fp16)[name = tensor("op_35506_cast_fp16")]; + tensor var_35510_begin_0 = const()[name = tensor("op_35510_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_35510_end_0 = const()[name = tensor("op_35510_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_35510_end_mask_0 = const()[name = tensor("op_35510_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35510_cast_fp16 = slice_by_index(begin = var_35510_begin_0, end = var_35510_end_0, end_mask = var_35510_end_mask_0, x = q_159_cast_fp16)[name = tensor("op_35510_cast_fp16")]; + tensor var_35514_begin_0 = const()[name = tensor("op_35514_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_35514_end_0 = const()[name = tensor("op_35514_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_35514_end_mask_0 = const()[name = tensor("op_35514_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35514_cast_fp16 = slice_by_index(begin = var_35514_begin_0, end = var_35514_end_0, end_mask = var_35514_end_mask_0, x = q_159_cast_fp16)[name = tensor("op_35514_cast_fp16")]; + tensor var_35518_begin_0 = const()[name = tensor("op_35518_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_35518_end_0 = const()[name = tensor("op_35518_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_35518_end_mask_0 = const()[name = tensor("op_35518_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35518_cast_fp16 = slice_by_index(begin = var_35518_begin_0, end = var_35518_end_0, end_mask = var_35518_end_mask_0, x = q_159_cast_fp16)[name = tensor("op_35518_cast_fp16")]; + tensor var_35522_begin_0 = const()[name = tensor("op_35522_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_35522_end_0 = const()[name = tensor("op_35522_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_35522_end_mask_0 = const()[name = tensor("op_35522_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35522_cast_fp16 = slice_by_index(begin = var_35522_begin_0, end = var_35522_end_0, end_mask = var_35522_end_mask_0, x = q_159_cast_fp16)[name = tensor("op_35522_cast_fp16")]; + tensor var_35526_begin_0 = const()[name = tensor("op_35526_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_35526_end_0 = const()[name = tensor("op_35526_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_35526_end_mask_0 = const()[name = tensor("op_35526_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35526_cast_fp16 = slice_by_index(begin = var_35526_begin_0, end = var_35526_end_0, end_mask = var_35526_end_mask_0, x = q_159_cast_fp16)[name = tensor("op_35526_cast_fp16")]; + tensor var_35530_begin_0 = const()[name = tensor("op_35530_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_35530_end_0 = const()[name = tensor("op_35530_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_35530_end_mask_0 = const()[name = tensor("op_35530_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35530_cast_fp16 = slice_by_index(begin = var_35530_begin_0, end = var_35530_end_0, end_mask = var_35530_end_mask_0, x = q_159_cast_fp16)[name = tensor("op_35530_cast_fp16")]; + tensor var_35534_begin_0 = const()[name = tensor("op_35534_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_35534_end_0 = const()[name = tensor("op_35534_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_35534_end_mask_0 = const()[name = tensor("op_35534_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35534_cast_fp16 = slice_by_index(begin = var_35534_begin_0, end = var_35534_end_0, end_mask = var_35534_end_mask_0, x = q_159_cast_fp16)[name = tensor("op_35534_cast_fp16")]; + tensor var_35538_begin_0 = const()[name = tensor("op_35538_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_35538_end_0 = const()[name = tensor("op_35538_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_35538_end_mask_0 = const()[name = tensor("op_35538_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35538_cast_fp16 = slice_by_index(begin = var_35538_begin_0, end = var_35538_end_0, end_mask = var_35538_end_mask_0, x = q_159_cast_fp16)[name = tensor("op_35538_cast_fp16")]; + tensor var_35542_begin_0 = const()[name = tensor("op_35542_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_35542_end_0 = const()[name = tensor("op_35542_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_35542_end_mask_0 = const()[name = tensor("op_35542_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35542_cast_fp16 = slice_by_index(begin = var_35542_begin_0, end = var_35542_end_0, end_mask = var_35542_end_mask_0, x = q_159_cast_fp16)[name = tensor("op_35542_cast_fp16")]; + tensor var_35546_begin_0 = const()[name = tensor("op_35546_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_35546_end_0 = const()[name = tensor("op_35546_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_35546_end_mask_0 = const()[name = tensor("op_35546_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35546_cast_fp16 = slice_by_index(begin = var_35546_begin_0, end = var_35546_end_0, end_mask = var_35546_end_mask_0, x = q_159_cast_fp16)[name = tensor("op_35546_cast_fp16")]; + tensor var_35550_begin_0 = const()[name = tensor("op_35550_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_35550_end_0 = const()[name = tensor("op_35550_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_35550_end_mask_0 = const()[name = tensor("op_35550_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35550_cast_fp16 = slice_by_index(begin = var_35550_begin_0, end = var_35550_end_0, end_mask = var_35550_end_mask_0, x = q_159_cast_fp16)[name = tensor("op_35550_cast_fp16")]; + tensor var_35554_begin_0 = const()[name = tensor("op_35554_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_35554_end_0 = const()[name = tensor("op_35554_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_35554_end_mask_0 = const()[name = tensor("op_35554_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35554_cast_fp16 = slice_by_index(begin = var_35554_begin_0, end = var_35554_end_0, end_mask = var_35554_end_mask_0, x = q_159_cast_fp16)[name = tensor("op_35554_cast_fp16")]; + tensor var_35558_begin_0 = const()[name = tensor("op_35558_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_35558_end_0 = const()[name = tensor("op_35558_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_35558_end_mask_0 = const()[name = tensor("op_35558_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35558_cast_fp16 = slice_by_index(begin = var_35558_begin_0, end = var_35558_end_0, end_mask = var_35558_end_mask_0, x = q_159_cast_fp16)[name = tensor("op_35558_cast_fp16")]; + tensor var_35562_begin_0 = const()[name = tensor("op_35562_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_35562_end_0 = const()[name = tensor("op_35562_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_35562_end_mask_0 = const()[name = tensor("op_35562_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35562_cast_fp16 = slice_by_index(begin = var_35562_begin_0, end = var_35562_end_0, end_mask = var_35562_end_mask_0, x = q_159_cast_fp16)[name = tensor("op_35562_cast_fp16")]; + tensor var_35566_begin_0 = const()[name = tensor("op_35566_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_35566_end_0 = const()[name = tensor("op_35566_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_35566_end_mask_0 = const()[name = tensor("op_35566_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35566_cast_fp16 = slice_by_index(begin = var_35566_begin_0, end = var_35566_end_0, end_mask = var_35566_end_mask_0, x = q_159_cast_fp16)[name = tensor("op_35566_cast_fp16")]; + tensor var_35570_begin_0 = const()[name = tensor("op_35570_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_35570_end_0 = const()[name = tensor("op_35570_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_35570_end_mask_0 = const()[name = tensor("op_35570_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35570_cast_fp16 = slice_by_index(begin = var_35570_begin_0, end = var_35570_end_0, end_mask = var_35570_end_mask_0, x = q_159_cast_fp16)[name = tensor("op_35570_cast_fp16")]; + tensor k_319_perm_0 = const()[name = tensor("k_319_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_35577_begin_0 = const()[name = tensor("op_35577_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_35577_end_0 = const()[name = tensor("op_35577_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_35577_end_mask_0 = const()[name = tensor("op_35577_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_319_cast_fp16 = transpose(perm = k_319_perm_0, x = k_317_cast_fp16)[name = tensor("transpose_60")]; + tensor var_35577_cast_fp16 = slice_by_index(begin = var_35577_begin_0, end = var_35577_end_0, end_mask = var_35577_end_mask_0, x = k_319_cast_fp16)[name = tensor("op_35577_cast_fp16")]; + tensor var_35581_begin_0 = const()[name = tensor("op_35581_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_35581_end_0 = const()[name = tensor("op_35581_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_35581_end_mask_0 = const()[name = tensor("op_35581_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35581_cast_fp16 = slice_by_index(begin = var_35581_begin_0, end = var_35581_end_0, end_mask = var_35581_end_mask_0, x = k_319_cast_fp16)[name = tensor("op_35581_cast_fp16")]; + tensor var_35585_begin_0 = const()[name = tensor("op_35585_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_35585_end_0 = const()[name = tensor("op_35585_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_35585_end_mask_0 = const()[name = tensor("op_35585_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35585_cast_fp16 = slice_by_index(begin = var_35585_begin_0, end = var_35585_end_0, end_mask = var_35585_end_mask_0, x = k_319_cast_fp16)[name = tensor("op_35585_cast_fp16")]; + tensor var_35589_begin_0 = const()[name = tensor("op_35589_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_35589_end_0 = const()[name = tensor("op_35589_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_35589_end_mask_0 = const()[name = tensor("op_35589_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35589_cast_fp16 = slice_by_index(begin = var_35589_begin_0, end = var_35589_end_0, end_mask = var_35589_end_mask_0, x = k_319_cast_fp16)[name = tensor("op_35589_cast_fp16")]; + tensor var_35593_begin_0 = const()[name = tensor("op_35593_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_35593_end_0 = const()[name = tensor("op_35593_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_35593_end_mask_0 = const()[name = tensor("op_35593_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35593_cast_fp16 = slice_by_index(begin = var_35593_begin_0, end = var_35593_end_0, end_mask = var_35593_end_mask_0, x = k_319_cast_fp16)[name = tensor("op_35593_cast_fp16")]; + tensor var_35597_begin_0 = const()[name = tensor("op_35597_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_35597_end_0 = const()[name = tensor("op_35597_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_35597_end_mask_0 = const()[name = tensor("op_35597_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35597_cast_fp16 = slice_by_index(begin = var_35597_begin_0, end = var_35597_end_0, end_mask = var_35597_end_mask_0, x = k_319_cast_fp16)[name = tensor("op_35597_cast_fp16")]; + tensor var_35601_begin_0 = const()[name = tensor("op_35601_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_35601_end_0 = const()[name = tensor("op_35601_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_35601_end_mask_0 = const()[name = tensor("op_35601_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35601_cast_fp16 = slice_by_index(begin = var_35601_begin_0, end = var_35601_end_0, end_mask = var_35601_end_mask_0, x = k_319_cast_fp16)[name = tensor("op_35601_cast_fp16")]; + tensor var_35605_begin_0 = const()[name = tensor("op_35605_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_35605_end_0 = const()[name = tensor("op_35605_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_35605_end_mask_0 = const()[name = tensor("op_35605_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35605_cast_fp16 = slice_by_index(begin = var_35605_begin_0, end = var_35605_end_0, end_mask = var_35605_end_mask_0, x = k_319_cast_fp16)[name = tensor("op_35605_cast_fp16")]; + tensor var_35609_begin_0 = const()[name = tensor("op_35609_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_35609_end_0 = const()[name = tensor("op_35609_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_35609_end_mask_0 = const()[name = tensor("op_35609_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35609_cast_fp16 = slice_by_index(begin = var_35609_begin_0, end = var_35609_end_0, end_mask = var_35609_end_mask_0, x = k_319_cast_fp16)[name = tensor("op_35609_cast_fp16")]; + tensor var_35613_begin_0 = const()[name = tensor("op_35613_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_35613_end_0 = const()[name = tensor("op_35613_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_35613_end_mask_0 = const()[name = tensor("op_35613_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35613_cast_fp16 = slice_by_index(begin = var_35613_begin_0, end = var_35613_end_0, end_mask = var_35613_end_mask_0, x = k_319_cast_fp16)[name = tensor("op_35613_cast_fp16")]; + tensor var_35617_begin_0 = const()[name = tensor("op_35617_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_35617_end_0 = const()[name = tensor("op_35617_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_35617_end_mask_0 = const()[name = tensor("op_35617_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35617_cast_fp16 = slice_by_index(begin = var_35617_begin_0, end = var_35617_end_0, end_mask = var_35617_end_mask_0, x = k_319_cast_fp16)[name = tensor("op_35617_cast_fp16")]; + tensor var_35621_begin_0 = const()[name = tensor("op_35621_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_35621_end_0 = const()[name = tensor("op_35621_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_35621_end_mask_0 = const()[name = tensor("op_35621_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35621_cast_fp16 = slice_by_index(begin = var_35621_begin_0, end = var_35621_end_0, end_mask = var_35621_end_mask_0, x = k_319_cast_fp16)[name = tensor("op_35621_cast_fp16")]; + tensor var_35625_begin_0 = const()[name = tensor("op_35625_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_35625_end_0 = const()[name = tensor("op_35625_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_35625_end_mask_0 = const()[name = tensor("op_35625_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35625_cast_fp16 = slice_by_index(begin = var_35625_begin_0, end = var_35625_end_0, end_mask = var_35625_end_mask_0, x = k_319_cast_fp16)[name = tensor("op_35625_cast_fp16")]; + tensor var_35629_begin_0 = const()[name = tensor("op_35629_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_35629_end_0 = const()[name = tensor("op_35629_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_35629_end_mask_0 = const()[name = tensor("op_35629_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35629_cast_fp16 = slice_by_index(begin = var_35629_begin_0, end = var_35629_end_0, end_mask = var_35629_end_mask_0, x = k_319_cast_fp16)[name = tensor("op_35629_cast_fp16")]; + tensor var_35633_begin_0 = const()[name = tensor("op_35633_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_35633_end_0 = const()[name = tensor("op_35633_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_35633_end_mask_0 = const()[name = tensor("op_35633_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35633_cast_fp16 = slice_by_index(begin = var_35633_begin_0, end = var_35633_end_0, end_mask = var_35633_end_mask_0, x = k_319_cast_fp16)[name = tensor("op_35633_cast_fp16")]; + tensor var_35637_begin_0 = const()[name = tensor("op_35637_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_35637_end_0 = const()[name = tensor("op_35637_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_35637_end_mask_0 = const()[name = tensor("op_35637_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35637_cast_fp16 = slice_by_index(begin = var_35637_begin_0, end = var_35637_end_0, end_mask = var_35637_end_mask_0, x = k_319_cast_fp16)[name = tensor("op_35637_cast_fp16")]; + tensor var_35641_begin_0 = const()[name = tensor("op_35641_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_35641_end_0 = const()[name = tensor("op_35641_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_35641_end_mask_0 = const()[name = tensor("op_35641_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35641_cast_fp16 = slice_by_index(begin = var_35641_begin_0, end = var_35641_end_0, end_mask = var_35641_end_mask_0, x = k_319_cast_fp16)[name = tensor("op_35641_cast_fp16")]; + tensor var_35645_begin_0 = const()[name = tensor("op_35645_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_35645_end_0 = const()[name = tensor("op_35645_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_35645_end_mask_0 = const()[name = tensor("op_35645_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35645_cast_fp16 = slice_by_index(begin = var_35645_begin_0, end = var_35645_end_0, end_mask = var_35645_end_mask_0, x = k_319_cast_fp16)[name = tensor("op_35645_cast_fp16")]; + tensor var_35649_begin_0 = const()[name = tensor("op_35649_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_35649_end_0 = const()[name = tensor("op_35649_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_35649_end_mask_0 = const()[name = tensor("op_35649_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35649_cast_fp16 = slice_by_index(begin = var_35649_begin_0, end = var_35649_end_0, end_mask = var_35649_end_mask_0, x = k_319_cast_fp16)[name = tensor("op_35649_cast_fp16")]; + tensor var_35653_begin_0 = const()[name = tensor("op_35653_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_35653_end_0 = const()[name = tensor("op_35653_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_35653_end_mask_0 = const()[name = tensor("op_35653_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35653_cast_fp16 = slice_by_index(begin = var_35653_begin_0, end = var_35653_end_0, end_mask = var_35653_end_mask_0, x = k_319_cast_fp16)[name = tensor("op_35653_cast_fp16")]; + tensor var_35655_begin_0 = const()[name = tensor("op_35655_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_35655_end_0 = const()[name = tensor("op_35655_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_35655_end_mask_0 = const()[name = tensor("op_35655_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35655_cast_fp16 = slice_by_index(begin = var_35655_begin_0, end = var_35655_end_0, end_mask = var_35655_end_mask_0, x = v_159_cast_fp16)[name = tensor("op_35655_cast_fp16")]; + tensor var_35659_begin_0 = const()[name = tensor("op_35659_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_35659_end_0 = const()[name = tensor("op_35659_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_35659_end_mask_0 = const()[name = tensor("op_35659_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35659_cast_fp16 = slice_by_index(begin = var_35659_begin_0, end = var_35659_end_0, end_mask = var_35659_end_mask_0, x = v_159_cast_fp16)[name = tensor("op_35659_cast_fp16")]; + tensor var_35663_begin_0 = const()[name = tensor("op_35663_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_35663_end_0 = const()[name = tensor("op_35663_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_35663_end_mask_0 = const()[name = tensor("op_35663_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35663_cast_fp16 = slice_by_index(begin = var_35663_begin_0, end = var_35663_end_0, end_mask = var_35663_end_mask_0, x = v_159_cast_fp16)[name = tensor("op_35663_cast_fp16")]; + tensor var_35667_begin_0 = const()[name = tensor("op_35667_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_35667_end_0 = const()[name = tensor("op_35667_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_35667_end_mask_0 = const()[name = tensor("op_35667_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35667_cast_fp16 = slice_by_index(begin = var_35667_begin_0, end = var_35667_end_0, end_mask = var_35667_end_mask_0, x = v_159_cast_fp16)[name = tensor("op_35667_cast_fp16")]; + tensor var_35671_begin_0 = const()[name = tensor("op_35671_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_35671_end_0 = const()[name = tensor("op_35671_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_35671_end_mask_0 = const()[name = tensor("op_35671_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35671_cast_fp16 = slice_by_index(begin = var_35671_begin_0, end = var_35671_end_0, end_mask = var_35671_end_mask_0, x = v_159_cast_fp16)[name = tensor("op_35671_cast_fp16")]; + tensor var_35675_begin_0 = const()[name = tensor("op_35675_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_35675_end_0 = const()[name = tensor("op_35675_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_35675_end_mask_0 = const()[name = tensor("op_35675_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35675_cast_fp16 = slice_by_index(begin = var_35675_begin_0, end = var_35675_end_0, end_mask = var_35675_end_mask_0, x = v_159_cast_fp16)[name = tensor("op_35675_cast_fp16")]; + tensor var_35679_begin_0 = const()[name = tensor("op_35679_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_35679_end_0 = const()[name = tensor("op_35679_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_35679_end_mask_0 = const()[name = tensor("op_35679_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35679_cast_fp16 = slice_by_index(begin = var_35679_begin_0, end = var_35679_end_0, end_mask = var_35679_end_mask_0, x = v_159_cast_fp16)[name = tensor("op_35679_cast_fp16")]; + tensor var_35683_begin_0 = const()[name = tensor("op_35683_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_35683_end_0 = const()[name = tensor("op_35683_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_35683_end_mask_0 = const()[name = tensor("op_35683_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35683_cast_fp16 = slice_by_index(begin = var_35683_begin_0, end = var_35683_end_0, end_mask = var_35683_end_mask_0, x = v_159_cast_fp16)[name = tensor("op_35683_cast_fp16")]; + tensor var_35687_begin_0 = const()[name = tensor("op_35687_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_35687_end_0 = const()[name = tensor("op_35687_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_35687_end_mask_0 = const()[name = tensor("op_35687_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35687_cast_fp16 = slice_by_index(begin = var_35687_begin_0, end = var_35687_end_0, end_mask = var_35687_end_mask_0, x = v_159_cast_fp16)[name = tensor("op_35687_cast_fp16")]; + tensor var_35691_begin_0 = const()[name = tensor("op_35691_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_35691_end_0 = const()[name = tensor("op_35691_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_35691_end_mask_0 = const()[name = tensor("op_35691_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35691_cast_fp16 = slice_by_index(begin = var_35691_begin_0, end = var_35691_end_0, end_mask = var_35691_end_mask_0, x = v_159_cast_fp16)[name = tensor("op_35691_cast_fp16")]; + tensor var_35695_begin_0 = const()[name = tensor("op_35695_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_35695_end_0 = const()[name = tensor("op_35695_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_35695_end_mask_0 = const()[name = tensor("op_35695_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35695_cast_fp16 = slice_by_index(begin = var_35695_begin_0, end = var_35695_end_0, end_mask = var_35695_end_mask_0, x = v_159_cast_fp16)[name = tensor("op_35695_cast_fp16")]; + tensor var_35699_begin_0 = const()[name = tensor("op_35699_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_35699_end_0 = const()[name = tensor("op_35699_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_35699_end_mask_0 = const()[name = tensor("op_35699_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35699_cast_fp16 = slice_by_index(begin = var_35699_begin_0, end = var_35699_end_0, end_mask = var_35699_end_mask_0, x = v_159_cast_fp16)[name = tensor("op_35699_cast_fp16")]; + tensor var_35703_begin_0 = const()[name = tensor("op_35703_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_35703_end_0 = const()[name = tensor("op_35703_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_35703_end_mask_0 = const()[name = tensor("op_35703_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35703_cast_fp16 = slice_by_index(begin = var_35703_begin_0, end = var_35703_end_0, end_mask = var_35703_end_mask_0, x = v_159_cast_fp16)[name = tensor("op_35703_cast_fp16")]; + tensor var_35707_begin_0 = const()[name = tensor("op_35707_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_35707_end_0 = const()[name = tensor("op_35707_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_35707_end_mask_0 = const()[name = tensor("op_35707_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35707_cast_fp16 = slice_by_index(begin = var_35707_begin_0, end = var_35707_end_0, end_mask = var_35707_end_mask_0, x = v_159_cast_fp16)[name = tensor("op_35707_cast_fp16")]; + tensor var_35711_begin_0 = const()[name = tensor("op_35711_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_35711_end_0 = const()[name = tensor("op_35711_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_35711_end_mask_0 = const()[name = tensor("op_35711_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35711_cast_fp16 = slice_by_index(begin = var_35711_begin_0, end = var_35711_end_0, end_mask = var_35711_end_mask_0, x = v_159_cast_fp16)[name = tensor("op_35711_cast_fp16")]; + tensor var_35715_begin_0 = const()[name = tensor("op_35715_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_35715_end_0 = const()[name = tensor("op_35715_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_35715_end_mask_0 = const()[name = tensor("op_35715_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35715_cast_fp16 = slice_by_index(begin = var_35715_begin_0, end = var_35715_end_0, end_mask = var_35715_end_mask_0, x = v_159_cast_fp16)[name = tensor("op_35715_cast_fp16")]; + tensor var_35719_begin_0 = const()[name = tensor("op_35719_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_35719_end_0 = const()[name = tensor("op_35719_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_35719_end_mask_0 = const()[name = tensor("op_35719_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35719_cast_fp16 = slice_by_index(begin = var_35719_begin_0, end = var_35719_end_0, end_mask = var_35719_end_mask_0, x = v_159_cast_fp16)[name = tensor("op_35719_cast_fp16")]; + tensor var_35723_begin_0 = const()[name = tensor("op_35723_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_35723_end_0 = const()[name = tensor("op_35723_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_35723_end_mask_0 = const()[name = tensor("op_35723_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35723_cast_fp16 = slice_by_index(begin = var_35723_begin_0, end = var_35723_end_0, end_mask = var_35723_end_mask_0, x = v_159_cast_fp16)[name = tensor("op_35723_cast_fp16")]; + tensor var_35727_begin_0 = const()[name = tensor("op_35727_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_35727_end_0 = const()[name = tensor("op_35727_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_35727_end_mask_0 = const()[name = tensor("op_35727_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35727_cast_fp16 = slice_by_index(begin = var_35727_begin_0, end = var_35727_end_0, end_mask = var_35727_end_mask_0, x = v_159_cast_fp16)[name = tensor("op_35727_cast_fp16")]; + tensor var_35731_begin_0 = const()[name = tensor("op_35731_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_35731_end_0 = const()[name = tensor("op_35731_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_35731_end_mask_0 = const()[name = tensor("op_35731_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35731_cast_fp16 = slice_by_index(begin = var_35731_begin_0, end = var_35731_end_0, end_mask = var_35731_end_mask_0, x = v_159_cast_fp16)[name = tensor("op_35731_cast_fp16")]; + tensor var_35735_equation_0 = const()[name = tensor("op_35735_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35735_cast_fp16 = einsum(equation = var_35735_equation_0, values = (var_35577_cast_fp16, var_35494_cast_fp16))[name = tensor("op_35735_cast_fp16")]; + tensor var_35736_to_fp16 = const()[name = tensor("op_35736_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3001_cast_fp16 = mul(x = var_35735_cast_fp16, y = var_35736_to_fp16)[name = tensor("aw_3001_cast_fp16")]; + tensor var_35739_equation_0 = const()[name = tensor("op_35739_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35739_cast_fp16 = einsum(equation = var_35739_equation_0, values = (var_35581_cast_fp16, var_35498_cast_fp16))[name = tensor("op_35739_cast_fp16")]; + tensor var_35740_to_fp16 = const()[name = tensor("op_35740_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3003_cast_fp16 = mul(x = var_35739_cast_fp16, y = var_35740_to_fp16)[name = tensor("aw_3003_cast_fp16")]; + tensor var_35743_equation_0 = const()[name = tensor("op_35743_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35743_cast_fp16 = einsum(equation = var_35743_equation_0, values = (var_35585_cast_fp16, var_35502_cast_fp16))[name = tensor("op_35743_cast_fp16")]; + tensor var_35744_to_fp16 = const()[name = tensor("op_35744_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3005_cast_fp16 = mul(x = var_35743_cast_fp16, y = var_35744_to_fp16)[name = tensor("aw_3005_cast_fp16")]; + tensor var_35747_equation_0 = const()[name = tensor("op_35747_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35747_cast_fp16 = einsum(equation = var_35747_equation_0, values = (var_35589_cast_fp16, var_35506_cast_fp16))[name = tensor("op_35747_cast_fp16")]; + tensor var_35748_to_fp16 = const()[name = tensor("op_35748_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3007_cast_fp16 = mul(x = var_35747_cast_fp16, y = var_35748_to_fp16)[name = tensor("aw_3007_cast_fp16")]; + tensor var_35751_equation_0 = const()[name = tensor("op_35751_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35751_cast_fp16 = einsum(equation = var_35751_equation_0, values = (var_35593_cast_fp16, var_35510_cast_fp16))[name = tensor("op_35751_cast_fp16")]; + tensor var_35752_to_fp16 = const()[name = tensor("op_35752_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3009_cast_fp16 = mul(x = var_35751_cast_fp16, y = var_35752_to_fp16)[name = tensor("aw_3009_cast_fp16")]; + tensor var_35755_equation_0 = const()[name = tensor("op_35755_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35755_cast_fp16 = einsum(equation = var_35755_equation_0, values = (var_35597_cast_fp16, var_35514_cast_fp16))[name = tensor("op_35755_cast_fp16")]; + tensor var_35756_to_fp16 = const()[name = tensor("op_35756_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3011_cast_fp16 = mul(x = var_35755_cast_fp16, y = var_35756_to_fp16)[name = tensor("aw_3011_cast_fp16")]; + tensor var_35759_equation_0 = const()[name = tensor("op_35759_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35759_cast_fp16 = einsum(equation = var_35759_equation_0, values = (var_35601_cast_fp16, var_35518_cast_fp16))[name = tensor("op_35759_cast_fp16")]; + tensor var_35760_to_fp16 = const()[name = tensor("op_35760_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3013_cast_fp16 = mul(x = var_35759_cast_fp16, y = var_35760_to_fp16)[name = tensor("aw_3013_cast_fp16")]; + tensor var_35763_equation_0 = const()[name = tensor("op_35763_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35763_cast_fp16 = einsum(equation = var_35763_equation_0, values = (var_35605_cast_fp16, var_35522_cast_fp16))[name = tensor("op_35763_cast_fp16")]; + tensor var_35764_to_fp16 = const()[name = tensor("op_35764_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3015_cast_fp16 = mul(x = var_35763_cast_fp16, y = var_35764_to_fp16)[name = tensor("aw_3015_cast_fp16")]; + tensor var_35767_equation_0 = const()[name = tensor("op_35767_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35767_cast_fp16 = einsum(equation = var_35767_equation_0, values = (var_35609_cast_fp16, var_35526_cast_fp16))[name = tensor("op_35767_cast_fp16")]; + tensor var_35768_to_fp16 = const()[name = tensor("op_35768_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3017_cast_fp16 = mul(x = var_35767_cast_fp16, y = var_35768_to_fp16)[name = tensor("aw_3017_cast_fp16")]; + tensor var_35771_equation_0 = const()[name = tensor("op_35771_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35771_cast_fp16 = einsum(equation = var_35771_equation_0, values = (var_35613_cast_fp16, var_35530_cast_fp16))[name = tensor("op_35771_cast_fp16")]; + tensor var_35772_to_fp16 = const()[name = tensor("op_35772_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3019_cast_fp16 = mul(x = var_35771_cast_fp16, y = var_35772_to_fp16)[name = tensor("aw_3019_cast_fp16")]; + tensor var_35775_equation_0 = const()[name = tensor("op_35775_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35775_cast_fp16 = einsum(equation = var_35775_equation_0, values = (var_35617_cast_fp16, var_35534_cast_fp16))[name = tensor("op_35775_cast_fp16")]; + tensor var_35776_to_fp16 = const()[name = tensor("op_35776_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3021_cast_fp16 = mul(x = var_35775_cast_fp16, y = var_35776_to_fp16)[name = tensor("aw_3021_cast_fp16")]; + tensor var_35779_equation_0 = const()[name = tensor("op_35779_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35779_cast_fp16 = einsum(equation = var_35779_equation_0, values = (var_35621_cast_fp16, var_35538_cast_fp16))[name = tensor("op_35779_cast_fp16")]; + tensor var_35780_to_fp16 = const()[name = tensor("op_35780_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3023_cast_fp16 = mul(x = var_35779_cast_fp16, y = var_35780_to_fp16)[name = tensor("aw_3023_cast_fp16")]; + tensor var_35783_equation_0 = const()[name = tensor("op_35783_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35783_cast_fp16 = einsum(equation = var_35783_equation_0, values = (var_35625_cast_fp16, var_35542_cast_fp16))[name = tensor("op_35783_cast_fp16")]; + tensor var_35784_to_fp16 = const()[name = tensor("op_35784_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3025_cast_fp16 = mul(x = var_35783_cast_fp16, y = var_35784_to_fp16)[name = tensor("aw_3025_cast_fp16")]; + tensor var_35787_equation_0 = const()[name = tensor("op_35787_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35787_cast_fp16 = einsum(equation = var_35787_equation_0, values = (var_35629_cast_fp16, var_35546_cast_fp16))[name = tensor("op_35787_cast_fp16")]; + tensor var_35788_to_fp16 = const()[name = tensor("op_35788_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3027_cast_fp16 = mul(x = var_35787_cast_fp16, y = var_35788_to_fp16)[name = tensor("aw_3027_cast_fp16")]; + tensor var_35791_equation_0 = const()[name = tensor("op_35791_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35791_cast_fp16 = einsum(equation = var_35791_equation_0, values = (var_35633_cast_fp16, var_35550_cast_fp16))[name = tensor("op_35791_cast_fp16")]; + tensor var_35792_to_fp16 = const()[name = tensor("op_35792_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3029_cast_fp16 = mul(x = var_35791_cast_fp16, y = var_35792_to_fp16)[name = tensor("aw_3029_cast_fp16")]; + tensor var_35795_equation_0 = const()[name = tensor("op_35795_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35795_cast_fp16 = einsum(equation = var_35795_equation_0, values = (var_35637_cast_fp16, var_35554_cast_fp16))[name = tensor("op_35795_cast_fp16")]; + tensor var_35796_to_fp16 = const()[name = tensor("op_35796_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3031_cast_fp16 = mul(x = var_35795_cast_fp16, y = var_35796_to_fp16)[name = tensor("aw_3031_cast_fp16")]; + tensor var_35799_equation_0 = const()[name = tensor("op_35799_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35799_cast_fp16 = einsum(equation = var_35799_equation_0, values = (var_35641_cast_fp16, var_35558_cast_fp16))[name = tensor("op_35799_cast_fp16")]; + tensor var_35800_to_fp16 = const()[name = tensor("op_35800_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3033_cast_fp16 = mul(x = var_35799_cast_fp16, y = var_35800_to_fp16)[name = tensor("aw_3033_cast_fp16")]; + tensor var_35803_equation_0 = const()[name = tensor("op_35803_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35803_cast_fp16 = einsum(equation = var_35803_equation_0, values = (var_35645_cast_fp16, var_35562_cast_fp16))[name = tensor("op_35803_cast_fp16")]; + tensor var_35804_to_fp16 = const()[name = tensor("op_35804_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3035_cast_fp16 = mul(x = var_35803_cast_fp16, y = var_35804_to_fp16)[name = tensor("aw_3035_cast_fp16")]; + tensor var_35807_equation_0 = const()[name = tensor("op_35807_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35807_cast_fp16 = einsum(equation = var_35807_equation_0, values = (var_35649_cast_fp16, var_35566_cast_fp16))[name = tensor("op_35807_cast_fp16")]; + tensor var_35808_to_fp16 = const()[name = tensor("op_35808_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3037_cast_fp16 = mul(x = var_35807_cast_fp16, y = var_35808_to_fp16)[name = tensor("aw_3037_cast_fp16")]; + tensor var_35811_equation_0 = const()[name = tensor("op_35811_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35811_cast_fp16 = einsum(equation = var_35811_equation_0, values = (var_35653_cast_fp16, var_35570_cast_fp16))[name = tensor("op_35811_cast_fp16")]; + tensor var_35812_to_fp16 = const()[name = tensor("op_35812_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3039_cast_fp16 = mul(x = var_35811_cast_fp16, y = var_35812_to_fp16)[name = tensor("aw_3039_cast_fp16")]; + tensor var_35814_cast_fp16 = softmax(axis = var_30355, x = aw_3001_cast_fp16)[name = tensor("op_35814_cast_fp16")]; + tensor var_35815_cast_fp16 = softmax(axis = var_30355, x = aw_3003_cast_fp16)[name = tensor("op_35815_cast_fp16")]; + tensor var_35816_cast_fp16 = softmax(axis = var_30355, x = aw_3005_cast_fp16)[name = tensor("op_35816_cast_fp16")]; + tensor var_35817_cast_fp16 = softmax(axis = var_30355, x = aw_3007_cast_fp16)[name = tensor("op_35817_cast_fp16")]; + tensor var_35818_cast_fp16 = softmax(axis = var_30355, x = aw_3009_cast_fp16)[name = tensor("op_35818_cast_fp16")]; + tensor var_35819_cast_fp16 = softmax(axis = var_30355, x = aw_3011_cast_fp16)[name = tensor("op_35819_cast_fp16")]; + tensor var_35820_cast_fp16 = softmax(axis = var_30355, x = aw_3013_cast_fp16)[name = tensor("op_35820_cast_fp16")]; + tensor var_35821_cast_fp16 = softmax(axis = var_30355, x = aw_3015_cast_fp16)[name = tensor("op_35821_cast_fp16")]; + tensor var_35822_cast_fp16 = softmax(axis = var_30355, x = aw_3017_cast_fp16)[name = tensor("op_35822_cast_fp16")]; + tensor var_35823_cast_fp16 = softmax(axis = var_30355, x = aw_3019_cast_fp16)[name = tensor("op_35823_cast_fp16")]; + tensor var_35824_cast_fp16 = softmax(axis = var_30355, x = aw_3021_cast_fp16)[name = tensor("op_35824_cast_fp16")]; + tensor var_35825_cast_fp16 = softmax(axis = var_30355, x = aw_3023_cast_fp16)[name = tensor("op_35825_cast_fp16")]; + tensor var_35826_cast_fp16 = softmax(axis = var_30355, x = aw_3025_cast_fp16)[name = tensor("op_35826_cast_fp16")]; + tensor var_35827_cast_fp16 = softmax(axis = var_30355, x = aw_3027_cast_fp16)[name = tensor("op_35827_cast_fp16")]; + tensor var_35828_cast_fp16 = softmax(axis = var_30355, x = aw_3029_cast_fp16)[name = tensor("op_35828_cast_fp16")]; + tensor var_35829_cast_fp16 = softmax(axis = var_30355, x = aw_3031_cast_fp16)[name = tensor("op_35829_cast_fp16")]; + tensor var_35830_cast_fp16 = softmax(axis = var_30355, x = aw_3033_cast_fp16)[name = tensor("op_35830_cast_fp16")]; + tensor var_35831_cast_fp16 = softmax(axis = var_30355, x = aw_3035_cast_fp16)[name = tensor("op_35831_cast_fp16")]; + tensor var_35832_cast_fp16 = softmax(axis = var_30355, x = aw_3037_cast_fp16)[name = tensor("op_35832_cast_fp16")]; + tensor var_35833_cast_fp16 = softmax(axis = var_30355, x = aw_3039_cast_fp16)[name = tensor("op_35833_cast_fp16")]; + tensor var_35835_equation_0 = const()[name = tensor("op_35835_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35835_cast_fp16 = einsum(equation = var_35835_equation_0, values = (var_35655_cast_fp16, var_35814_cast_fp16))[name = tensor("op_35835_cast_fp16")]; + tensor var_35837_equation_0 = const()[name = tensor("op_35837_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35837_cast_fp16 = einsum(equation = var_35837_equation_0, values = (var_35659_cast_fp16, var_35815_cast_fp16))[name = tensor("op_35837_cast_fp16")]; + tensor var_35839_equation_0 = const()[name = tensor("op_35839_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35839_cast_fp16 = einsum(equation = var_35839_equation_0, values = (var_35663_cast_fp16, var_35816_cast_fp16))[name = tensor("op_35839_cast_fp16")]; + tensor var_35841_equation_0 = const()[name = tensor("op_35841_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35841_cast_fp16 = einsum(equation = var_35841_equation_0, values = (var_35667_cast_fp16, var_35817_cast_fp16))[name = tensor("op_35841_cast_fp16")]; + tensor var_35843_equation_0 = const()[name = tensor("op_35843_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35843_cast_fp16 = einsum(equation = var_35843_equation_0, values = (var_35671_cast_fp16, var_35818_cast_fp16))[name = tensor("op_35843_cast_fp16")]; + tensor var_35845_equation_0 = const()[name = tensor("op_35845_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35845_cast_fp16 = einsum(equation = var_35845_equation_0, values = (var_35675_cast_fp16, var_35819_cast_fp16))[name = tensor("op_35845_cast_fp16")]; + tensor var_35847_equation_0 = const()[name = tensor("op_35847_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35847_cast_fp16 = einsum(equation = var_35847_equation_0, values = (var_35679_cast_fp16, var_35820_cast_fp16))[name = tensor("op_35847_cast_fp16")]; + tensor var_35849_equation_0 = const()[name = tensor("op_35849_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35849_cast_fp16 = einsum(equation = var_35849_equation_0, values = (var_35683_cast_fp16, var_35821_cast_fp16))[name = tensor("op_35849_cast_fp16")]; + tensor var_35851_equation_0 = const()[name = tensor("op_35851_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35851_cast_fp16 = einsum(equation = var_35851_equation_0, values = (var_35687_cast_fp16, var_35822_cast_fp16))[name = tensor("op_35851_cast_fp16")]; + tensor var_35853_equation_0 = const()[name = tensor("op_35853_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35853_cast_fp16 = einsum(equation = var_35853_equation_0, values = (var_35691_cast_fp16, var_35823_cast_fp16))[name = tensor("op_35853_cast_fp16")]; + tensor var_35855_equation_0 = const()[name = tensor("op_35855_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35855_cast_fp16 = einsum(equation = var_35855_equation_0, values = (var_35695_cast_fp16, var_35824_cast_fp16))[name = tensor("op_35855_cast_fp16")]; + tensor var_35857_equation_0 = const()[name = tensor("op_35857_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35857_cast_fp16 = einsum(equation = var_35857_equation_0, values = (var_35699_cast_fp16, var_35825_cast_fp16))[name = tensor("op_35857_cast_fp16")]; + tensor var_35859_equation_0 = const()[name = tensor("op_35859_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35859_cast_fp16 = einsum(equation = var_35859_equation_0, values = (var_35703_cast_fp16, var_35826_cast_fp16))[name = tensor("op_35859_cast_fp16")]; + tensor var_35861_equation_0 = const()[name = tensor("op_35861_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35861_cast_fp16 = einsum(equation = var_35861_equation_0, values = (var_35707_cast_fp16, var_35827_cast_fp16))[name = tensor("op_35861_cast_fp16")]; + tensor var_35863_equation_0 = const()[name = tensor("op_35863_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35863_cast_fp16 = einsum(equation = var_35863_equation_0, values = (var_35711_cast_fp16, var_35828_cast_fp16))[name = tensor("op_35863_cast_fp16")]; + tensor var_35865_equation_0 = const()[name = tensor("op_35865_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35865_cast_fp16 = einsum(equation = var_35865_equation_0, values = (var_35715_cast_fp16, var_35829_cast_fp16))[name = tensor("op_35865_cast_fp16")]; + tensor var_35867_equation_0 = const()[name = tensor("op_35867_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35867_cast_fp16 = einsum(equation = var_35867_equation_0, values = (var_35719_cast_fp16, var_35830_cast_fp16))[name = tensor("op_35867_cast_fp16")]; + tensor var_35869_equation_0 = const()[name = tensor("op_35869_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35869_cast_fp16 = einsum(equation = var_35869_equation_0, values = (var_35723_cast_fp16, var_35831_cast_fp16))[name = tensor("op_35869_cast_fp16")]; + tensor var_35871_equation_0 = const()[name = tensor("op_35871_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35871_cast_fp16 = einsum(equation = var_35871_equation_0, values = (var_35727_cast_fp16, var_35832_cast_fp16))[name = tensor("op_35871_cast_fp16")]; + tensor var_35873_equation_0 = const()[name = tensor("op_35873_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35873_cast_fp16 = einsum(equation = var_35873_equation_0, values = (var_35731_cast_fp16, var_35833_cast_fp16))[name = tensor("op_35873_cast_fp16")]; + tensor input_481_interleave_0 = const()[name = tensor("input_481_interleave_0"), val = tensor(false)]; + tensor input_481_cast_fp16 = concat(axis = var_30355, interleave = input_481_interleave_0, values = (var_35835_cast_fp16, var_35837_cast_fp16, var_35839_cast_fp16, var_35841_cast_fp16, var_35843_cast_fp16, var_35845_cast_fp16, var_35847_cast_fp16, var_35849_cast_fp16, var_35851_cast_fp16, var_35853_cast_fp16, var_35855_cast_fp16, var_35857_cast_fp16, var_35859_cast_fp16, var_35861_cast_fp16, var_35863_cast_fp16, var_35865_cast_fp16, var_35867_cast_fp16, var_35869_cast_fp16, var_35871_cast_fp16, var_35873_cast_fp16))[name = tensor("input_481_cast_fp16")]; + tensor var_35879 = const()[name = tensor("op_35879"), val = tensor([1, 1])]; + tensor var_35881 = const()[name = tensor("op_35881"), val = tensor([1, 1])]; + tensor var_35883_pad_type_0 = const()[name = tensor("op_35883_pad_type_0"), val = tensor("custom")]; + tensor var_35883_pad_0 = const()[name = tensor("op_35883_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_5_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_5_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2962869504)))]; + tensor up_blocks_0_attentions_0_transformer_blocks_5_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_5_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2966146368)))]; + tensor var_35883_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_5_attn2_to_out_0_bias_to_fp16, dilations = var_35881, groups = var_30355, pad = var_35883_pad_0, pad_type = var_35883_pad_type_0, strides = var_35879, weight = up_blocks_0_attentions_0_transformer_blocks_5_attn2_to_out_0_weight_to_fp16, x = input_481_cast_fp16)[name = tensor("op_35883_cast_fp16")]; + tensor inputs_239_cast_fp16 = add(x = var_35883_cast_fp16, y = inputs_237_cast_fp16)[name = tensor("inputs_239_cast_fp16")]; + tensor input_483_axes_0 = const()[name = tensor("input_483_axes_0"), val = tensor([1])]; + tensor input_483_gamma_0_to_fp16 = const()[name = tensor("input_483_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2966148992)))]; + tensor input_483_beta_0_to_fp16 = const()[name = tensor("input_483_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2966151616)))]; + tensor var_35893_to_fp16 = const()[name = tensor("op_35893_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_483_cast_fp16 = layer_norm(axes = input_483_axes_0, beta = input_483_beta_0_to_fp16, epsilon = var_35893_to_fp16, gamma = input_483_gamma_0_to_fp16, x = inputs_239_cast_fp16)[name = tensor("input_483_cast_fp16")]; + tensor var_35909 = const()[name = tensor("op_35909"), val = tensor([1, 1])]; + tensor var_35911 = const()[name = tensor("op_35911"), val = tensor([1, 1])]; + tensor var_35913_pad_type_0 = const()[name = tensor("op_35913_pad_type_0"), val = tensor("custom")]; + tensor var_35913_pad_0 = const()[name = tensor("op_35913_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_5_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_5_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2966154240)))]; + tensor up_blocks_0_attentions_0_transformer_blocks_5_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_5_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2992368704)))]; + tensor var_35913_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_5_ff_net_0_proj_bias_to_fp16, dilations = var_35911, groups = var_30355, pad = var_35913_pad_0, pad_type = var_35913_pad_type_0, strides = var_35909, weight = up_blocks_0_attentions_0_transformer_blocks_5_ff_net_0_proj_weight_to_fp16, x = input_483_cast_fp16)[name = tensor("op_35913_cast_fp16")]; + tensor var_35914_split_sizes_0 = const()[name = tensor("op_35914_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_35914_axis_0 = const()[name = tensor("op_35914_axis_0"), val = tensor(1)]; + tensor var_35914_cast_fp16_0, tensor var_35914_cast_fp16_1 = split(axis = var_35914_axis_0, split_sizes = var_35914_split_sizes_0, x = var_35913_cast_fp16)[name = tensor("op_35914_cast_fp16")]; + tensor var_35916_mode_0 = const()[name = tensor("op_35916_mode_0"), val = tensor("EXACT")]; + tensor var_35916_cast_fp16 = gelu(mode = var_35916_mode_0, x = var_35914_cast_fp16_1)[name = tensor("op_35916_cast_fp16")]; + tensor input_485_cast_fp16 = mul(x = var_35914_cast_fp16_0, y = var_35916_cast_fp16)[name = tensor("input_485_cast_fp16")]; + tensor var_35920 = const()[name = tensor("op_35920"), val = tensor([1, 1])]; + tensor var_35922 = const()[name = tensor("op_35922"), val = tensor([1, 1])]; + tensor var_35924_pad_type_0 = const()[name = tensor("op_35924_pad_type_0"), val = tensor("custom")]; + tensor var_35924_pad_0 = const()[name = tensor("op_35924_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_5_ff_net_2_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_5_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2992389248)))]; + tensor up_blocks_0_attentions_0_transformer_blocks_5_ff_net_2_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_5_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3005496512)))]; + tensor var_35924_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_5_ff_net_2_bias_to_fp16, dilations = var_35922, groups = var_30355, pad = var_35924_pad_0, pad_type = var_35924_pad_type_0, strides = var_35920, weight = up_blocks_0_attentions_0_transformer_blocks_5_ff_net_2_weight_to_fp16, x = input_485_cast_fp16)[name = tensor("op_35924_cast_fp16")]; + tensor inputs_241_cast_fp16 = add(x = var_35924_cast_fp16, y = inputs_239_cast_fp16)[name = tensor("inputs_241_cast_fp16")]; + tensor hidden_states_323_axes_0 = const()[name = tensor("hidden_states_323_axes_0"), val = tensor([1])]; + tensor hidden_states_323_gamma_0_to_fp16 = const()[name = tensor("hidden_states_323_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3005499136)))]; + tensor hidden_states_323_beta_0_to_fp16 = const()[name = tensor("hidden_states_323_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3005501760)))]; + tensor var_35940_to_fp16 = const()[name = tensor("op_35940_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_323_cast_fp16 = layer_norm(axes = hidden_states_323_axes_0, beta = hidden_states_323_beta_0_to_fp16, epsilon = var_35940_to_fp16, gamma = hidden_states_323_gamma_0_to_fp16, x = inputs_241_cast_fp16)[name = tensor("hidden_states_323_cast_fp16")]; + tensor var_35955 = const()[name = tensor("op_35955"), val = tensor([1, 1])]; + tensor var_35957 = const()[name = tensor("op_35957"), val = tensor([1, 1])]; + tensor q_161_pad_type_0 = const()[name = tensor("q_161_pad_type_0"), val = tensor("custom")]; + tensor q_161_pad_0 = const()[name = tensor("q_161_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_6_attn1_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_6_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3005504384)))]; + tensor q_161_cast_fp16 = conv(dilations = var_35957, groups = var_30355, pad = q_161_pad_0, pad_type = q_161_pad_type_0, strides = var_35955, weight = up_blocks_0_attentions_0_transformer_blocks_6_attn1_to_q_weight_to_fp16, x = hidden_states_323_cast_fp16)[name = tensor("q_161_cast_fp16")]; + tensor var_35961 = const()[name = tensor("op_35961"), val = tensor([1, 1])]; + tensor var_35963 = const()[name = tensor("op_35963"), val = tensor([1, 1])]; + tensor k_321_pad_type_0 = const()[name = tensor("k_321_pad_type_0"), val = tensor("custom")]; + tensor k_321_pad_0 = const()[name = tensor("k_321_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_6_attn1_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_6_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3008781248)))]; + tensor k_321_cast_fp16 = conv(dilations = var_35963, groups = var_30355, pad = k_321_pad_0, pad_type = k_321_pad_type_0, strides = var_35961, weight = up_blocks_0_attentions_0_transformer_blocks_6_attn1_to_k_weight_to_fp16, x = hidden_states_323_cast_fp16)[name = tensor("k_321_cast_fp16")]; + tensor var_35967 = const()[name = tensor("op_35967"), val = tensor([1, 1])]; + tensor var_35969 = const()[name = tensor("op_35969"), val = tensor([1, 1])]; + tensor v_161_pad_type_0 = const()[name = tensor("v_161_pad_type_0"), val = tensor("custom")]; + tensor v_161_pad_0 = const()[name = tensor("v_161_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_6_attn1_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_6_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3012058112)))]; + tensor v_161_cast_fp16 = conv(dilations = var_35969, groups = var_30355, pad = v_161_pad_0, pad_type = v_161_pad_type_0, strides = var_35967, weight = up_blocks_0_attentions_0_transformer_blocks_6_attn1_to_v_weight_to_fp16, x = hidden_states_323_cast_fp16)[name = tensor("v_161_cast_fp16")]; + tensor var_35973_begin_0 = const()[name = tensor("op_35973_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_35973_end_0 = const()[name = tensor("op_35973_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_35973_end_mask_0 = const()[name = tensor("op_35973_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35973_cast_fp16 = slice_by_index(begin = var_35973_begin_0, end = var_35973_end_0, end_mask = var_35973_end_mask_0, x = q_161_cast_fp16)[name = tensor("op_35973_cast_fp16")]; + tensor var_35977_begin_0 = const()[name = tensor("op_35977_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_35977_end_0 = const()[name = tensor("op_35977_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_35977_end_mask_0 = const()[name = tensor("op_35977_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35977_cast_fp16 = slice_by_index(begin = var_35977_begin_0, end = var_35977_end_0, end_mask = var_35977_end_mask_0, x = q_161_cast_fp16)[name = tensor("op_35977_cast_fp16")]; + tensor var_35981_begin_0 = const()[name = tensor("op_35981_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_35981_end_0 = const()[name = tensor("op_35981_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_35981_end_mask_0 = const()[name = tensor("op_35981_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35981_cast_fp16 = slice_by_index(begin = var_35981_begin_0, end = var_35981_end_0, end_mask = var_35981_end_mask_0, x = q_161_cast_fp16)[name = tensor("op_35981_cast_fp16")]; + tensor var_35985_begin_0 = const()[name = tensor("op_35985_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_35985_end_0 = const()[name = tensor("op_35985_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_35985_end_mask_0 = const()[name = tensor("op_35985_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35985_cast_fp16 = slice_by_index(begin = var_35985_begin_0, end = var_35985_end_0, end_mask = var_35985_end_mask_0, x = q_161_cast_fp16)[name = tensor("op_35985_cast_fp16")]; + tensor var_35989_begin_0 = const()[name = tensor("op_35989_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_35989_end_0 = const()[name = tensor("op_35989_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_35989_end_mask_0 = const()[name = tensor("op_35989_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35989_cast_fp16 = slice_by_index(begin = var_35989_begin_0, end = var_35989_end_0, end_mask = var_35989_end_mask_0, x = q_161_cast_fp16)[name = tensor("op_35989_cast_fp16")]; + tensor var_35993_begin_0 = const()[name = tensor("op_35993_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_35993_end_0 = const()[name = tensor("op_35993_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_35993_end_mask_0 = const()[name = tensor("op_35993_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35993_cast_fp16 = slice_by_index(begin = var_35993_begin_0, end = var_35993_end_0, end_mask = var_35993_end_mask_0, x = q_161_cast_fp16)[name = tensor("op_35993_cast_fp16")]; + tensor var_35997_begin_0 = const()[name = tensor("op_35997_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_35997_end_0 = const()[name = tensor("op_35997_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_35997_end_mask_0 = const()[name = tensor("op_35997_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35997_cast_fp16 = slice_by_index(begin = var_35997_begin_0, end = var_35997_end_0, end_mask = var_35997_end_mask_0, x = q_161_cast_fp16)[name = tensor("op_35997_cast_fp16")]; + tensor var_36001_begin_0 = const()[name = tensor("op_36001_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_36001_end_0 = const()[name = tensor("op_36001_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_36001_end_mask_0 = const()[name = tensor("op_36001_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36001_cast_fp16 = slice_by_index(begin = var_36001_begin_0, end = var_36001_end_0, end_mask = var_36001_end_mask_0, x = q_161_cast_fp16)[name = tensor("op_36001_cast_fp16")]; + tensor var_36005_begin_0 = const()[name = tensor("op_36005_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_36005_end_0 = const()[name = tensor("op_36005_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_36005_end_mask_0 = const()[name = tensor("op_36005_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36005_cast_fp16 = slice_by_index(begin = var_36005_begin_0, end = var_36005_end_0, end_mask = var_36005_end_mask_0, x = q_161_cast_fp16)[name = tensor("op_36005_cast_fp16")]; + tensor var_36009_begin_0 = const()[name = tensor("op_36009_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_36009_end_0 = const()[name = tensor("op_36009_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_36009_end_mask_0 = const()[name = tensor("op_36009_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36009_cast_fp16 = slice_by_index(begin = var_36009_begin_0, end = var_36009_end_0, end_mask = var_36009_end_mask_0, x = q_161_cast_fp16)[name = tensor("op_36009_cast_fp16")]; + tensor var_36013_begin_0 = const()[name = tensor("op_36013_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_36013_end_0 = const()[name = tensor("op_36013_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_36013_end_mask_0 = const()[name = tensor("op_36013_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36013_cast_fp16 = slice_by_index(begin = var_36013_begin_0, end = var_36013_end_0, end_mask = var_36013_end_mask_0, x = q_161_cast_fp16)[name = tensor("op_36013_cast_fp16")]; + tensor var_36017_begin_0 = const()[name = tensor("op_36017_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_36017_end_0 = const()[name = tensor("op_36017_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_36017_end_mask_0 = const()[name = tensor("op_36017_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36017_cast_fp16 = slice_by_index(begin = var_36017_begin_0, end = var_36017_end_0, end_mask = var_36017_end_mask_0, x = q_161_cast_fp16)[name = tensor("op_36017_cast_fp16")]; + tensor var_36021_begin_0 = const()[name = tensor("op_36021_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_36021_end_0 = const()[name = tensor("op_36021_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_36021_end_mask_0 = const()[name = tensor("op_36021_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36021_cast_fp16 = slice_by_index(begin = var_36021_begin_0, end = var_36021_end_0, end_mask = var_36021_end_mask_0, x = q_161_cast_fp16)[name = tensor("op_36021_cast_fp16")]; + tensor var_36025_begin_0 = const()[name = tensor("op_36025_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_36025_end_0 = const()[name = tensor("op_36025_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_36025_end_mask_0 = const()[name = tensor("op_36025_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36025_cast_fp16 = slice_by_index(begin = var_36025_begin_0, end = var_36025_end_0, end_mask = var_36025_end_mask_0, x = q_161_cast_fp16)[name = tensor("op_36025_cast_fp16")]; + tensor var_36029_begin_0 = const()[name = tensor("op_36029_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_36029_end_0 = const()[name = tensor("op_36029_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_36029_end_mask_0 = const()[name = tensor("op_36029_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36029_cast_fp16 = slice_by_index(begin = var_36029_begin_0, end = var_36029_end_0, end_mask = var_36029_end_mask_0, x = q_161_cast_fp16)[name = tensor("op_36029_cast_fp16")]; + tensor var_36033_begin_0 = const()[name = tensor("op_36033_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_36033_end_0 = const()[name = tensor("op_36033_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_36033_end_mask_0 = const()[name = tensor("op_36033_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36033_cast_fp16 = slice_by_index(begin = var_36033_begin_0, end = var_36033_end_0, end_mask = var_36033_end_mask_0, x = q_161_cast_fp16)[name = tensor("op_36033_cast_fp16")]; + tensor var_36037_begin_0 = const()[name = tensor("op_36037_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_36037_end_0 = const()[name = tensor("op_36037_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_36037_end_mask_0 = const()[name = tensor("op_36037_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36037_cast_fp16 = slice_by_index(begin = var_36037_begin_0, end = var_36037_end_0, end_mask = var_36037_end_mask_0, x = q_161_cast_fp16)[name = tensor("op_36037_cast_fp16")]; + tensor var_36041_begin_0 = const()[name = tensor("op_36041_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_36041_end_0 = const()[name = tensor("op_36041_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_36041_end_mask_0 = const()[name = tensor("op_36041_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36041_cast_fp16 = slice_by_index(begin = var_36041_begin_0, end = var_36041_end_0, end_mask = var_36041_end_mask_0, x = q_161_cast_fp16)[name = tensor("op_36041_cast_fp16")]; + tensor var_36045_begin_0 = const()[name = tensor("op_36045_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_36045_end_0 = const()[name = tensor("op_36045_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_36045_end_mask_0 = const()[name = tensor("op_36045_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36045_cast_fp16 = slice_by_index(begin = var_36045_begin_0, end = var_36045_end_0, end_mask = var_36045_end_mask_0, x = q_161_cast_fp16)[name = tensor("op_36045_cast_fp16")]; + tensor var_36049_begin_0 = const()[name = tensor("op_36049_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_36049_end_0 = const()[name = tensor("op_36049_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_36049_end_mask_0 = const()[name = tensor("op_36049_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36049_cast_fp16 = slice_by_index(begin = var_36049_begin_0, end = var_36049_end_0, end_mask = var_36049_end_mask_0, x = q_161_cast_fp16)[name = tensor("op_36049_cast_fp16")]; + tensor k_323_perm_0 = const()[name = tensor("k_323_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_36056_begin_0 = const()[name = tensor("op_36056_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_36056_end_0 = const()[name = tensor("op_36056_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_36056_end_mask_0 = const()[name = tensor("op_36056_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_323_cast_fp16 = transpose(perm = k_323_perm_0, x = k_321_cast_fp16)[name = tensor("transpose_59")]; + tensor var_36056_cast_fp16 = slice_by_index(begin = var_36056_begin_0, end = var_36056_end_0, end_mask = var_36056_end_mask_0, x = k_323_cast_fp16)[name = tensor("op_36056_cast_fp16")]; + tensor var_36060_begin_0 = const()[name = tensor("op_36060_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_36060_end_0 = const()[name = tensor("op_36060_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_36060_end_mask_0 = const()[name = tensor("op_36060_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36060_cast_fp16 = slice_by_index(begin = var_36060_begin_0, end = var_36060_end_0, end_mask = var_36060_end_mask_0, x = k_323_cast_fp16)[name = tensor("op_36060_cast_fp16")]; + tensor var_36064_begin_0 = const()[name = tensor("op_36064_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_36064_end_0 = const()[name = tensor("op_36064_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_36064_end_mask_0 = const()[name = tensor("op_36064_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36064_cast_fp16 = slice_by_index(begin = var_36064_begin_0, end = var_36064_end_0, end_mask = var_36064_end_mask_0, x = k_323_cast_fp16)[name = tensor("op_36064_cast_fp16")]; + tensor var_36068_begin_0 = const()[name = tensor("op_36068_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_36068_end_0 = const()[name = tensor("op_36068_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_36068_end_mask_0 = const()[name = tensor("op_36068_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36068_cast_fp16 = slice_by_index(begin = var_36068_begin_0, end = var_36068_end_0, end_mask = var_36068_end_mask_0, x = k_323_cast_fp16)[name = tensor("op_36068_cast_fp16")]; + tensor var_36072_begin_0 = const()[name = tensor("op_36072_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_36072_end_0 = const()[name = tensor("op_36072_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_36072_end_mask_0 = const()[name = tensor("op_36072_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36072_cast_fp16 = slice_by_index(begin = var_36072_begin_0, end = var_36072_end_0, end_mask = var_36072_end_mask_0, x = k_323_cast_fp16)[name = tensor("op_36072_cast_fp16")]; + tensor var_36076_begin_0 = const()[name = tensor("op_36076_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_36076_end_0 = const()[name = tensor("op_36076_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_36076_end_mask_0 = const()[name = tensor("op_36076_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36076_cast_fp16 = slice_by_index(begin = var_36076_begin_0, end = var_36076_end_0, end_mask = var_36076_end_mask_0, x = k_323_cast_fp16)[name = tensor("op_36076_cast_fp16")]; + tensor var_36080_begin_0 = const()[name = tensor("op_36080_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_36080_end_0 = const()[name = tensor("op_36080_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_36080_end_mask_0 = const()[name = tensor("op_36080_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36080_cast_fp16 = slice_by_index(begin = var_36080_begin_0, end = var_36080_end_0, end_mask = var_36080_end_mask_0, x = k_323_cast_fp16)[name = tensor("op_36080_cast_fp16")]; + tensor var_36084_begin_0 = const()[name = tensor("op_36084_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_36084_end_0 = const()[name = tensor("op_36084_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_36084_end_mask_0 = const()[name = tensor("op_36084_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36084_cast_fp16 = slice_by_index(begin = var_36084_begin_0, end = var_36084_end_0, end_mask = var_36084_end_mask_0, x = k_323_cast_fp16)[name = tensor("op_36084_cast_fp16")]; + tensor var_36088_begin_0 = const()[name = tensor("op_36088_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_36088_end_0 = const()[name = tensor("op_36088_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_36088_end_mask_0 = const()[name = tensor("op_36088_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36088_cast_fp16 = slice_by_index(begin = var_36088_begin_0, end = var_36088_end_0, end_mask = var_36088_end_mask_0, x = k_323_cast_fp16)[name = tensor("op_36088_cast_fp16")]; + tensor var_36092_begin_0 = const()[name = tensor("op_36092_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_36092_end_0 = const()[name = tensor("op_36092_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_36092_end_mask_0 = const()[name = tensor("op_36092_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36092_cast_fp16 = slice_by_index(begin = var_36092_begin_0, end = var_36092_end_0, end_mask = var_36092_end_mask_0, x = k_323_cast_fp16)[name = tensor("op_36092_cast_fp16")]; + tensor var_36096_begin_0 = const()[name = tensor("op_36096_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_36096_end_0 = const()[name = tensor("op_36096_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_36096_end_mask_0 = const()[name = tensor("op_36096_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36096_cast_fp16 = slice_by_index(begin = var_36096_begin_0, end = var_36096_end_0, end_mask = var_36096_end_mask_0, x = k_323_cast_fp16)[name = tensor("op_36096_cast_fp16")]; + tensor var_36100_begin_0 = const()[name = tensor("op_36100_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_36100_end_0 = const()[name = tensor("op_36100_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_36100_end_mask_0 = const()[name = tensor("op_36100_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36100_cast_fp16 = slice_by_index(begin = var_36100_begin_0, end = var_36100_end_0, end_mask = var_36100_end_mask_0, x = k_323_cast_fp16)[name = tensor("op_36100_cast_fp16")]; + tensor var_36104_begin_0 = const()[name = tensor("op_36104_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_36104_end_0 = const()[name = tensor("op_36104_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_36104_end_mask_0 = const()[name = tensor("op_36104_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36104_cast_fp16 = slice_by_index(begin = var_36104_begin_0, end = var_36104_end_0, end_mask = var_36104_end_mask_0, x = k_323_cast_fp16)[name = tensor("op_36104_cast_fp16")]; + tensor var_36108_begin_0 = const()[name = tensor("op_36108_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_36108_end_0 = const()[name = tensor("op_36108_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_36108_end_mask_0 = const()[name = tensor("op_36108_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36108_cast_fp16 = slice_by_index(begin = var_36108_begin_0, end = var_36108_end_0, end_mask = var_36108_end_mask_0, x = k_323_cast_fp16)[name = tensor("op_36108_cast_fp16")]; + tensor var_36112_begin_0 = const()[name = tensor("op_36112_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_36112_end_0 = const()[name = tensor("op_36112_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_36112_end_mask_0 = const()[name = tensor("op_36112_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36112_cast_fp16 = slice_by_index(begin = var_36112_begin_0, end = var_36112_end_0, end_mask = var_36112_end_mask_0, x = k_323_cast_fp16)[name = tensor("op_36112_cast_fp16")]; + tensor var_36116_begin_0 = const()[name = tensor("op_36116_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_36116_end_0 = const()[name = tensor("op_36116_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_36116_end_mask_0 = const()[name = tensor("op_36116_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36116_cast_fp16 = slice_by_index(begin = var_36116_begin_0, end = var_36116_end_0, end_mask = var_36116_end_mask_0, x = k_323_cast_fp16)[name = tensor("op_36116_cast_fp16")]; + tensor var_36120_begin_0 = const()[name = tensor("op_36120_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_36120_end_0 = const()[name = tensor("op_36120_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_36120_end_mask_0 = const()[name = tensor("op_36120_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36120_cast_fp16 = slice_by_index(begin = var_36120_begin_0, end = var_36120_end_0, end_mask = var_36120_end_mask_0, x = k_323_cast_fp16)[name = tensor("op_36120_cast_fp16")]; + tensor var_36124_begin_0 = const()[name = tensor("op_36124_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_36124_end_0 = const()[name = tensor("op_36124_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_36124_end_mask_0 = const()[name = tensor("op_36124_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36124_cast_fp16 = slice_by_index(begin = var_36124_begin_0, end = var_36124_end_0, end_mask = var_36124_end_mask_0, x = k_323_cast_fp16)[name = tensor("op_36124_cast_fp16")]; + tensor var_36128_begin_0 = const()[name = tensor("op_36128_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_36128_end_0 = const()[name = tensor("op_36128_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_36128_end_mask_0 = const()[name = tensor("op_36128_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36128_cast_fp16 = slice_by_index(begin = var_36128_begin_0, end = var_36128_end_0, end_mask = var_36128_end_mask_0, x = k_323_cast_fp16)[name = tensor("op_36128_cast_fp16")]; + tensor var_36132_begin_0 = const()[name = tensor("op_36132_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_36132_end_0 = const()[name = tensor("op_36132_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_36132_end_mask_0 = const()[name = tensor("op_36132_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36132_cast_fp16 = slice_by_index(begin = var_36132_begin_0, end = var_36132_end_0, end_mask = var_36132_end_mask_0, x = k_323_cast_fp16)[name = tensor("op_36132_cast_fp16")]; + tensor var_36134_begin_0 = const()[name = tensor("op_36134_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_36134_end_0 = const()[name = tensor("op_36134_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_36134_end_mask_0 = const()[name = tensor("op_36134_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36134_cast_fp16 = slice_by_index(begin = var_36134_begin_0, end = var_36134_end_0, end_mask = var_36134_end_mask_0, x = v_161_cast_fp16)[name = tensor("op_36134_cast_fp16")]; + tensor var_36138_begin_0 = const()[name = tensor("op_36138_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_36138_end_0 = const()[name = tensor("op_36138_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_36138_end_mask_0 = const()[name = tensor("op_36138_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36138_cast_fp16 = slice_by_index(begin = var_36138_begin_0, end = var_36138_end_0, end_mask = var_36138_end_mask_0, x = v_161_cast_fp16)[name = tensor("op_36138_cast_fp16")]; + tensor var_36142_begin_0 = const()[name = tensor("op_36142_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_36142_end_0 = const()[name = tensor("op_36142_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_36142_end_mask_0 = const()[name = tensor("op_36142_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36142_cast_fp16 = slice_by_index(begin = var_36142_begin_0, end = var_36142_end_0, end_mask = var_36142_end_mask_0, x = v_161_cast_fp16)[name = tensor("op_36142_cast_fp16")]; + tensor var_36146_begin_0 = const()[name = tensor("op_36146_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_36146_end_0 = const()[name = tensor("op_36146_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_36146_end_mask_0 = const()[name = tensor("op_36146_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36146_cast_fp16 = slice_by_index(begin = var_36146_begin_0, end = var_36146_end_0, end_mask = var_36146_end_mask_0, x = v_161_cast_fp16)[name = tensor("op_36146_cast_fp16")]; + tensor var_36150_begin_0 = const()[name = tensor("op_36150_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_36150_end_0 = const()[name = tensor("op_36150_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_36150_end_mask_0 = const()[name = tensor("op_36150_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36150_cast_fp16 = slice_by_index(begin = var_36150_begin_0, end = var_36150_end_0, end_mask = var_36150_end_mask_0, x = v_161_cast_fp16)[name = tensor("op_36150_cast_fp16")]; + tensor var_36154_begin_0 = const()[name = tensor("op_36154_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_36154_end_0 = const()[name = tensor("op_36154_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_36154_end_mask_0 = const()[name = tensor("op_36154_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36154_cast_fp16 = slice_by_index(begin = var_36154_begin_0, end = var_36154_end_0, end_mask = var_36154_end_mask_0, x = v_161_cast_fp16)[name = tensor("op_36154_cast_fp16")]; + tensor var_36158_begin_0 = const()[name = tensor("op_36158_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_36158_end_0 = const()[name = tensor("op_36158_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_36158_end_mask_0 = const()[name = tensor("op_36158_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36158_cast_fp16 = slice_by_index(begin = var_36158_begin_0, end = var_36158_end_0, end_mask = var_36158_end_mask_0, x = v_161_cast_fp16)[name = tensor("op_36158_cast_fp16")]; + tensor var_36162_begin_0 = const()[name = tensor("op_36162_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_36162_end_0 = const()[name = tensor("op_36162_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_36162_end_mask_0 = const()[name = tensor("op_36162_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36162_cast_fp16 = slice_by_index(begin = var_36162_begin_0, end = var_36162_end_0, end_mask = var_36162_end_mask_0, x = v_161_cast_fp16)[name = tensor("op_36162_cast_fp16")]; + tensor var_36166_begin_0 = const()[name = tensor("op_36166_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_36166_end_0 = const()[name = tensor("op_36166_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_36166_end_mask_0 = const()[name = tensor("op_36166_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36166_cast_fp16 = slice_by_index(begin = var_36166_begin_0, end = var_36166_end_0, end_mask = var_36166_end_mask_0, x = v_161_cast_fp16)[name = tensor("op_36166_cast_fp16")]; + tensor var_36170_begin_0 = const()[name = tensor("op_36170_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_36170_end_0 = const()[name = tensor("op_36170_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_36170_end_mask_0 = const()[name = tensor("op_36170_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36170_cast_fp16 = slice_by_index(begin = var_36170_begin_0, end = var_36170_end_0, end_mask = var_36170_end_mask_0, x = v_161_cast_fp16)[name = tensor("op_36170_cast_fp16")]; + tensor var_36174_begin_0 = const()[name = tensor("op_36174_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_36174_end_0 = const()[name = tensor("op_36174_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_36174_end_mask_0 = const()[name = tensor("op_36174_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36174_cast_fp16 = slice_by_index(begin = var_36174_begin_0, end = var_36174_end_0, end_mask = var_36174_end_mask_0, x = v_161_cast_fp16)[name = tensor("op_36174_cast_fp16")]; + tensor var_36178_begin_0 = const()[name = tensor("op_36178_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_36178_end_0 = const()[name = tensor("op_36178_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_36178_end_mask_0 = const()[name = tensor("op_36178_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36178_cast_fp16 = slice_by_index(begin = var_36178_begin_0, end = var_36178_end_0, end_mask = var_36178_end_mask_0, x = v_161_cast_fp16)[name = tensor("op_36178_cast_fp16")]; + tensor var_36182_begin_0 = const()[name = tensor("op_36182_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_36182_end_0 = const()[name = tensor("op_36182_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_36182_end_mask_0 = const()[name = tensor("op_36182_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36182_cast_fp16 = slice_by_index(begin = var_36182_begin_0, end = var_36182_end_0, end_mask = var_36182_end_mask_0, x = v_161_cast_fp16)[name = tensor("op_36182_cast_fp16")]; + tensor var_36186_begin_0 = const()[name = tensor("op_36186_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_36186_end_0 = const()[name = tensor("op_36186_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_36186_end_mask_0 = const()[name = tensor("op_36186_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36186_cast_fp16 = slice_by_index(begin = var_36186_begin_0, end = var_36186_end_0, end_mask = var_36186_end_mask_0, x = v_161_cast_fp16)[name = tensor("op_36186_cast_fp16")]; + tensor var_36190_begin_0 = const()[name = tensor("op_36190_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_36190_end_0 = const()[name = tensor("op_36190_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_36190_end_mask_0 = const()[name = tensor("op_36190_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36190_cast_fp16 = slice_by_index(begin = var_36190_begin_0, end = var_36190_end_0, end_mask = var_36190_end_mask_0, x = v_161_cast_fp16)[name = tensor("op_36190_cast_fp16")]; + tensor var_36194_begin_0 = const()[name = tensor("op_36194_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_36194_end_0 = const()[name = tensor("op_36194_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_36194_end_mask_0 = const()[name = tensor("op_36194_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36194_cast_fp16 = slice_by_index(begin = var_36194_begin_0, end = var_36194_end_0, end_mask = var_36194_end_mask_0, x = v_161_cast_fp16)[name = tensor("op_36194_cast_fp16")]; + tensor var_36198_begin_0 = const()[name = tensor("op_36198_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_36198_end_0 = const()[name = tensor("op_36198_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_36198_end_mask_0 = const()[name = tensor("op_36198_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36198_cast_fp16 = slice_by_index(begin = var_36198_begin_0, end = var_36198_end_0, end_mask = var_36198_end_mask_0, x = v_161_cast_fp16)[name = tensor("op_36198_cast_fp16")]; + tensor var_36202_begin_0 = const()[name = tensor("op_36202_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_36202_end_0 = const()[name = tensor("op_36202_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_36202_end_mask_0 = const()[name = tensor("op_36202_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36202_cast_fp16 = slice_by_index(begin = var_36202_begin_0, end = var_36202_end_0, end_mask = var_36202_end_mask_0, x = v_161_cast_fp16)[name = tensor("op_36202_cast_fp16")]; + tensor var_36206_begin_0 = const()[name = tensor("op_36206_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_36206_end_0 = const()[name = tensor("op_36206_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_36206_end_mask_0 = const()[name = tensor("op_36206_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36206_cast_fp16 = slice_by_index(begin = var_36206_begin_0, end = var_36206_end_0, end_mask = var_36206_end_mask_0, x = v_161_cast_fp16)[name = tensor("op_36206_cast_fp16")]; + tensor var_36210_begin_0 = const()[name = tensor("op_36210_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_36210_end_0 = const()[name = tensor("op_36210_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_36210_end_mask_0 = const()[name = tensor("op_36210_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36210_cast_fp16 = slice_by_index(begin = var_36210_begin_0, end = var_36210_end_0, end_mask = var_36210_end_mask_0, x = v_161_cast_fp16)[name = tensor("op_36210_cast_fp16")]; + tensor var_36214_equation_0 = const()[name = tensor("op_36214_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_36214_cast_fp16 = einsum(equation = var_36214_equation_0, values = (var_36056_cast_fp16, var_35973_cast_fp16))[name = tensor("op_36214_cast_fp16")]; + tensor var_36215_to_fp16 = const()[name = tensor("op_36215_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3041_cast_fp16 = mul(x = var_36214_cast_fp16, y = var_36215_to_fp16)[name = tensor("aw_3041_cast_fp16")]; + tensor var_36218_equation_0 = const()[name = tensor("op_36218_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_36218_cast_fp16 = einsum(equation = var_36218_equation_0, values = (var_36060_cast_fp16, var_35977_cast_fp16))[name = tensor("op_36218_cast_fp16")]; + tensor var_36219_to_fp16 = const()[name = tensor("op_36219_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3043_cast_fp16 = mul(x = var_36218_cast_fp16, y = var_36219_to_fp16)[name = tensor("aw_3043_cast_fp16")]; + tensor var_36222_equation_0 = const()[name = tensor("op_36222_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_36222_cast_fp16 = einsum(equation = var_36222_equation_0, values = (var_36064_cast_fp16, var_35981_cast_fp16))[name = tensor("op_36222_cast_fp16")]; + tensor var_36223_to_fp16 = const()[name = tensor("op_36223_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3045_cast_fp16 = mul(x = var_36222_cast_fp16, y = var_36223_to_fp16)[name = tensor("aw_3045_cast_fp16")]; + tensor var_36226_equation_0 = const()[name = tensor("op_36226_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_36226_cast_fp16 = einsum(equation = var_36226_equation_0, values = (var_36068_cast_fp16, var_35985_cast_fp16))[name = tensor("op_36226_cast_fp16")]; + tensor var_36227_to_fp16 = const()[name = tensor("op_36227_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3047_cast_fp16 = mul(x = var_36226_cast_fp16, y = var_36227_to_fp16)[name = tensor("aw_3047_cast_fp16")]; + tensor var_36230_equation_0 = const()[name = tensor("op_36230_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_36230_cast_fp16 = einsum(equation = var_36230_equation_0, values = (var_36072_cast_fp16, var_35989_cast_fp16))[name = tensor("op_36230_cast_fp16")]; + tensor var_36231_to_fp16 = const()[name = tensor("op_36231_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3049_cast_fp16 = mul(x = var_36230_cast_fp16, y = var_36231_to_fp16)[name = tensor("aw_3049_cast_fp16")]; + tensor var_36234_equation_0 = const()[name = tensor("op_36234_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_36234_cast_fp16 = einsum(equation = var_36234_equation_0, values = (var_36076_cast_fp16, var_35993_cast_fp16))[name = tensor("op_36234_cast_fp16")]; + tensor var_36235_to_fp16 = const()[name = tensor("op_36235_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3051_cast_fp16 = mul(x = var_36234_cast_fp16, y = var_36235_to_fp16)[name = tensor("aw_3051_cast_fp16")]; + tensor var_36238_equation_0 = const()[name = tensor("op_36238_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_36238_cast_fp16 = einsum(equation = var_36238_equation_0, values = (var_36080_cast_fp16, var_35997_cast_fp16))[name = tensor("op_36238_cast_fp16")]; + tensor var_36239_to_fp16 = const()[name = tensor("op_36239_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3053_cast_fp16 = mul(x = var_36238_cast_fp16, y = var_36239_to_fp16)[name = tensor("aw_3053_cast_fp16")]; + tensor var_36242_equation_0 = const()[name = tensor("op_36242_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_36242_cast_fp16 = einsum(equation = var_36242_equation_0, values = (var_36084_cast_fp16, var_36001_cast_fp16))[name = tensor("op_36242_cast_fp16")]; + tensor var_36243_to_fp16 = const()[name = tensor("op_36243_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3055_cast_fp16 = mul(x = var_36242_cast_fp16, y = var_36243_to_fp16)[name = tensor("aw_3055_cast_fp16")]; + tensor var_36246_equation_0 = const()[name = tensor("op_36246_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_36246_cast_fp16 = einsum(equation = var_36246_equation_0, values = (var_36088_cast_fp16, var_36005_cast_fp16))[name = tensor("op_36246_cast_fp16")]; + tensor var_36247_to_fp16 = const()[name = tensor("op_36247_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3057_cast_fp16 = mul(x = var_36246_cast_fp16, y = var_36247_to_fp16)[name = tensor("aw_3057_cast_fp16")]; + tensor var_36250_equation_0 = const()[name = tensor("op_36250_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_36250_cast_fp16 = einsum(equation = var_36250_equation_0, values = (var_36092_cast_fp16, var_36009_cast_fp16))[name = tensor("op_36250_cast_fp16")]; + tensor var_36251_to_fp16 = const()[name = tensor("op_36251_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3059_cast_fp16 = mul(x = var_36250_cast_fp16, y = var_36251_to_fp16)[name = tensor("aw_3059_cast_fp16")]; + tensor var_36254_equation_0 = const()[name = tensor("op_36254_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_36254_cast_fp16 = einsum(equation = var_36254_equation_0, values = (var_36096_cast_fp16, var_36013_cast_fp16))[name = tensor("op_36254_cast_fp16")]; + tensor var_36255_to_fp16 = const()[name = tensor("op_36255_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3061_cast_fp16 = mul(x = var_36254_cast_fp16, y = var_36255_to_fp16)[name = tensor("aw_3061_cast_fp16")]; + tensor var_36258_equation_0 = const()[name = tensor("op_36258_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_36258_cast_fp16 = einsum(equation = var_36258_equation_0, values = (var_36100_cast_fp16, var_36017_cast_fp16))[name = tensor("op_36258_cast_fp16")]; + tensor var_36259_to_fp16 = const()[name = tensor("op_36259_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3063_cast_fp16 = mul(x = var_36258_cast_fp16, y = var_36259_to_fp16)[name = tensor("aw_3063_cast_fp16")]; + tensor var_36262_equation_0 = const()[name = tensor("op_36262_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_36262_cast_fp16 = einsum(equation = var_36262_equation_0, values = (var_36104_cast_fp16, var_36021_cast_fp16))[name = tensor("op_36262_cast_fp16")]; + tensor var_36263_to_fp16 = const()[name = tensor("op_36263_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3065_cast_fp16 = mul(x = var_36262_cast_fp16, y = var_36263_to_fp16)[name = tensor("aw_3065_cast_fp16")]; + tensor var_36266_equation_0 = const()[name = tensor("op_36266_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_36266_cast_fp16 = einsum(equation = var_36266_equation_0, values = (var_36108_cast_fp16, var_36025_cast_fp16))[name = tensor("op_36266_cast_fp16")]; + tensor var_36267_to_fp16 = const()[name = tensor("op_36267_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3067_cast_fp16 = mul(x = var_36266_cast_fp16, y = var_36267_to_fp16)[name = tensor("aw_3067_cast_fp16")]; + tensor var_36270_equation_0 = const()[name = tensor("op_36270_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_36270_cast_fp16 = einsum(equation = var_36270_equation_0, values = (var_36112_cast_fp16, var_36029_cast_fp16))[name = tensor("op_36270_cast_fp16")]; + tensor var_36271_to_fp16 = const()[name = tensor("op_36271_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3069_cast_fp16 = mul(x = var_36270_cast_fp16, y = var_36271_to_fp16)[name = tensor("aw_3069_cast_fp16")]; + tensor var_36274_equation_0 = const()[name = tensor("op_36274_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_36274_cast_fp16 = einsum(equation = var_36274_equation_0, values = (var_36116_cast_fp16, var_36033_cast_fp16))[name = tensor("op_36274_cast_fp16")]; + tensor var_36275_to_fp16 = const()[name = tensor("op_36275_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3071_cast_fp16 = mul(x = var_36274_cast_fp16, y = var_36275_to_fp16)[name = tensor("aw_3071_cast_fp16")]; + tensor var_36278_equation_0 = const()[name = tensor("op_36278_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_36278_cast_fp16 = einsum(equation = var_36278_equation_0, values = (var_36120_cast_fp16, var_36037_cast_fp16))[name = tensor("op_36278_cast_fp16")]; + tensor var_36279_to_fp16 = const()[name = tensor("op_36279_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3073_cast_fp16 = mul(x = var_36278_cast_fp16, y = var_36279_to_fp16)[name = tensor("aw_3073_cast_fp16")]; + tensor var_36282_equation_0 = const()[name = tensor("op_36282_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_36282_cast_fp16 = einsum(equation = var_36282_equation_0, values = (var_36124_cast_fp16, var_36041_cast_fp16))[name = tensor("op_36282_cast_fp16")]; + tensor var_36283_to_fp16 = const()[name = tensor("op_36283_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3075_cast_fp16 = mul(x = var_36282_cast_fp16, y = var_36283_to_fp16)[name = tensor("aw_3075_cast_fp16")]; + tensor var_36286_equation_0 = const()[name = tensor("op_36286_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_36286_cast_fp16 = einsum(equation = var_36286_equation_0, values = (var_36128_cast_fp16, var_36045_cast_fp16))[name = tensor("op_36286_cast_fp16")]; + tensor var_36287_to_fp16 = const()[name = tensor("op_36287_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3077_cast_fp16 = mul(x = var_36286_cast_fp16, y = var_36287_to_fp16)[name = tensor("aw_3077_cast_fp16")]; + tensor var_36290_equation_0 = const()[name = tensor("op_36290_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_36290_cast_fp16 = einsum(equation = var_36290_equation_0, values = (var_36132_cast_fp16, var_36049_cast_fp16))[name = tensor("op_36290_cast_fp16")]; + tensor var_36291_to_fp16 = const()[name = tensor("op_36291_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3079_cast_fp16 = mul(x = var_36290_cast_fp16, y = var_36291_to_fp16)[name = tensor("aw_3079_cast_fp16")]; + tensor var_36293_cast_fp16 = softmax(axis = var_30355, x = aw_3041_cast_fp16)[name = tensor("op_36293_cast_fp16")]; + tensor var_36294_cast_fp16 = softmax(axis = var_30355, x = aw_3043_cast_fp16)[name = tensor("op_36294_cast_fp16")]; + tensor var_36295_cast_fp16 = softmax(axis = var_30355, x = aw_3045_cast_fp16)[name = tensor("op_36295_cast_fp16")]; + tensor var_36296_cast_fp16 = softmax(axis = var_30355, x = aw_3047_cast_fp16)[name = tensor("op_36296_cast_fp16")]; + tensor var_36297_cast_fp16 = softmax(axis = var_30355, x = aw_3049_cast_fp16)[name = tensor("op_36297_cast_fp16")]; + tensor var_36298_cast_fp16 = softmax(axis = var_30355, x = aw_3051_cast_fp16)[name = tensor("op_36298_cast_fp16")]; + tensor var_36299_cast_fp16 = softmax(axis = var_30355, x = aw_3053_cast_fp16)[name = tensor("op_36299_cast_fp16")]; + tensor var_36300_cast_fp16 = softmax(axis = var_30355, x = aw_3055_cast_fp16)[name = tensor("op_36300_cast_fp16")]; + tensor var_36301_cast_fp16 = softmax(axis = var_30355, x = aw_3057_cast_fp16)[name = tensor("op_36301_cast_fp16")]; + tensor var_36302_cast_fp16 = softmax(axis = var_30355, x = aw_3059_cast_fp16)[name = tensor("op_36302_cast_fp16")]; + tensor var_36303_cast_fp16 = softmax(axis = var_30355, x = aw_3061_cast_fp16)[name = tensor("op_36303_cast_fp16")]; + tensor var_36304_cast_fp16 = softmax(axis = var_30355, x = aw_3063_cast_fp16)[name = tensor("op_36304_cast_fp16")]; + tensor var_36305_cast_fp16 = softmax(axis = var_30355, x = aw_3065_cast_fp16)[name = tensor("op_36305_cast_fp16")]; + tensor var_36306_cast_fp16 = softmax(axis = var_30355, x = aw_3067_cast_fp16)[name = tensor("op_36306_cast_fp16")]; + tensor var_36307_cast_fp16 = softmax(axis = var_30355, x = aw_3069_cast_fp16)[name = tensor("op_36307_cast_fp16")]; + tensor var_36308_cast_fp16 = softmax(axis = var_30355, x = aw_3071_cast_fp16)[name = tensor("op_36308_cast_fp16")]; + tensor var_36309_cast_fp16 = softmax(axis = var_30355, x = aw_3073_cast_fp16)[name = tensor("op_36309_cast_fp16")]; + tensor var_36310_cast_fp16 = softmax(axis = var_30355, x = aw_3075_cast_fp16)[name = tensor("op_36310_cast_fp16")]; + tensor var_36311_cast_fp16 = softmax(axis = var_30355, x = aw_3077_cast_fp16)[name = tensor("op_36311_cast_fp16")]; + tensor var_36312_cast_fp16 = softmax(axis = var_30355, x = aw_3079_cast_fp16)[name = tensor("op_36312_cast_fp16")]; + tensor var_36314_equation_0 = const()[name = tensor("op_36314_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_36314_cast_fp16 = einsum(equation = var_36314_equation_0, values = (var_36134_cast_fp16, var_36293_cast_fp16))[name = tensor("op_36314_cast_fp16")]; + tensor var_36316_equation_0 = const()[name = tensor("op_36316_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_36316_cast_fp16 = einsum(equation = var_36316_equation_0, values = (var_36138_cast_fp16, var_36294_cast_fp16))[name = tensor("op_36316_cast_fp16")]; + tensor var_36318_equation_0 = const()[name = tensor("op_36318_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_36318_cast_fp16 = einsum(equation = var_36318_equation_0, values = (var_36142_cast_fp16, var_36295_cast_fp16))[name = tensor("op_36318_cast_fp16")]; + tensor var_36320_equation_0 = const()[name = tensor("op_36320_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_36320_cast_fp16 = einsum(equation = var_36320_equation_0, values = (var_36146_cast_fp16, var_36296_cast_fp16))[name = tensor("op_36320_cast_fp16")]; + tensor var_36322_equation_0 = const()[name = tensor("op_36322_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_36322_cast_fp16 = einsum(equation = var_36322_equation_0, values = (var_36150_cast_fp16, var_36297_cast_fp16))[name = tensor("op_36322_cast_fp16")]; + tensor var_36324_equation_0 = const()[name = tensor("op_36324_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_36324_cast_fp16 = einsum(equation = var_36324_equation_0, values = (var_36154_cast_fp16, var_36298_cast_fp16))[name = tensor("op_36324_cast_fp16")]; + tensor var_36326_equation_0 = const()[name = tensor("op_36326_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_36326_cast_fp16 = einsum(equation = var_36326_equation_0, values = (var_36158_cast_fp16, var_36299_cast_fp16))[name = tensor("op_36326_cast_fp16")]; + tensor var_36328_equation_0 = const()[name = tensor("op_36328_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_36328_cast_fp16 = einsum(equation = var_36328_equation_0, values = (var_36162_cast_fp16, var_36300_cast_fp16))[name = tensor("op_36328_cast_fp16")]; + tensor var_36330_equation_0 = const()[name = tensor("op_36330_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_36330_cast_fp16 = einsum(equation = var_36330_equation_0, values = (var_36166_cast_fp16, var_36301_cast_fp16))[name = tensor("op_36330_cast_fp16")]; + tensor var_36332_equation_0 = const()[name = tensor("op_36332_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_36332_cast_fp16 = einsum(equation = var_36332_equation_0, values = (var_36170_cast_fp16, var_36302_cast_fp16))[name = tensor("op_36332_cast_fp16")]; + tensor var_36334_equation_0 = const()[name = tensor("op_36334_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_36334_cast_fp16 = einsum(equation = var_36334_equation_0, values = (var_36174_cast_fp16, var_36303_cast_fp16))[name = tensor("op_36334_cast_fp16")]; + tensor var_36336_equation_0 = const()[name = tensor("op_36336_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_36336_cast_fp16 = einsum(equation = var_36336_equation_0, values = (var_36178_cast_fp16, var_36304_cast_fp16))[name = tensor("op_36336_cast_fp16")]; + tensor var_36338_equation_0 = const()[name = tensor("op_36338_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_36338_cast_fp16 = einsum(equation = var_36338_equation_0, values = (var_36182_cast_fp16, var_36305_cast_fp16))[name = tensor("op_36338_cast_fp16")]; + tensor var_36340_equation_0 = const()[name = tensor("op_36340_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_36340_cast_fp16 = einsum(equation = var_36340_equation_0, values = (var_36186_cast_fp16, var_36306_cast_fp16))[name = tensor("op_36340_cast_fp16")]; + tensor var_36342_equation_0 = const()[name = tensor("op_36342_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_36342_cast_fp16 = einsum(equation = var_36342_equation_0, values = (var_36190_cast_fp16, var_36307_cast_fp16))[name = tensor("op_36342_cast_fp16")]; + tensor var_36344_equation_0 = const()[name = tensor("op_36344_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_36344_cast_fp16 = einsum(equation = var_36344_equation_0, values = (var_36194_cast_fp16, var_36308_cast_fp16))[name = tensor("op_36344_cast_fp16")]; + tensor var_36346_equation_0 = const()[name = tensor("op_36346_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_36346_cast_fp16 = einsum(equation = var_36346_equation_0, values = (var_36198_cast_fp16, var_36309_cast_fp16))[name = tensor("op_36346_cast_fp16")]; + tensor var_36348_equation_0 = const()[name = tensor("op_36348_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_36348_cast_fp16 = einsum(equation = var_36348_equation_0, values = (var_36202_cast_fp16, var_36310_cast_fp16))[name = tensor("op_36348_cast_fp16")]; + tensor var_36350_equation_0 = const()[name = tensor("op_36350_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_36350_cast_fp16 = einsum(equation = var_36350_equation_0, values = (var_36206_cast_fp16, var_36311_cast_fp16))[name = tensor("op_36350_cast_fp16")]; + tensor var_36352_equation_0 = const()[name = tensor("op_36352_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_36352_cast_fp16 = einsum(equation = var_36352_equation_0, values = (var_36210_cast_fp16, var_36312_cast_fp16))[name = tensor("op_36352_cast_fp16")]; + tensor input_487_interleave_0 = const()[name = tensor("input_487_interleave_0"), val = tensor(false)]; + tensor input_487_cast_fp16 = concat(axis = var_30355, interleave = input_487_interleave_0, values = (var_36314_cast_fp16, var_36316_cast_fp16, var_36318_cast_fp16, var_36320_cast_fp16, var_36322_cast_fp16, var_36324_cast_fp16, var_36326_cast_fp16, var_36328_cast_fp16, var_36330_cast_fp16, var_36332_cast_fp16, var_36334_cast_fp16, var_36336_cast_fp16, var_36338_cast_fp16, var_36340_cast_fp16, var_36342_cast_fp16, var_36344_cast_fp16, var_36346_cast_fp16, var_36348_cast_fp16, var_36350_cast_fp16, var_36352_cast_fp16))[name = tensor("input_487_cast_fp16")]; + tensor var_36358 = const()[name = tensor("op_36358"), val = tensor([1, 1])]; + tensor var_36360 = const()[name = tensor("op_36360"), val = tensor([1, 1])]; + tensor var_36362_pad_type_0 = const()[name = tensor("op_36362_pad_type_0"), val = tensor("custom")]; + tensor var_36362_pad_0 = const()[name = tensor("op_36362_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_6_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_6_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3015334976)))]; + tensor up_blocks_0_attentions_0_transformer_blocks_6_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_6_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3018611840)))]; + tensor var_36362_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_6_attn1_to_out_0_bias_to_fp16, dilations = var_36360, groups = var_30355, pad = var_36362_pad_0, pad_type = var_36362_pad_type_0, strides = var_36358, weight = up_blocks_0_attentions_0_transformer_blocks_6_attn1_to_out_0_weight_to_fp16, x = input_487_cast_fp16)[name = tensor("op_36362_cast_fp16")]; + tensor inputs_243_cast_fp16 = add(x = var_36362_cast_fp16, y = inputs_241_cast_fp16)[name = tensor("inputs_243_cast_fp16")]; + tensor hidden_states_325_axes_0 = const()[name = tensor("hidden_states_325_axes_0"), val = tensor([1])]; + tensor hidden_states_325_gamma_0_to_fp16 = const()[name = tensor("hidden_states_325_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3018614464)))]; + tensor hidden_states_325_beta_0_to_fp16 = const()[name = tensor("hidden_states_325_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3018617088)))]; + tensor var_36372_to_fp16 = const()[name = tensor("op_36372_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_325_cast_fp16 = layer_norm(axes = hidden_states_325_axes_0, beta = hidden_states_325_beta_0_to_fp16, epsilon = var_36372_to_fp16, gamma = hidden_states_325_gamma_0_to_fp16, x = inputs_243_cast_fp16)[name = tensor("hidden_states_325_cast_fp16")]; + tensor var_36387 = const()[name = tensor("op_36387"), val = tensor([1, 1])]; + tensor var_36389 = const()[name = tensor("op_36389"), val = tensor([1, 1])]; + tensor q_163_pad_type_0 = const()[name = tensor("q_163_pad_type_0"), val = tensor("custom")]; + tensor q_163_pad_0 = const()[name = tensor("q_163_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_6_attn2_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_6_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3018619712)))]; + tensor q_163_cast_fp16 = conv(dilations = var_36389, groups = var_30355, pad = q_163_pad_0, pad_type = q_163_pad_type_0, strides = var_36387, weight = up_blocks_0_attentions_0_transformer_blocks_6_attn2_to_q_weight_to_fp16, x = hidden_states_325_cast_fp16)[name = tensor("q_163_cast_fp16")]; + tensor var_36393 = const()[name = tensor("op_36393"), val = tensor([1, 1])]; + tensor var_36395 = const()[name = tensor("op_36395"), val = tensor([1, 1])]; + tensor k_325_pad_type_0 = const()[name = tensor("k_325_pad_type_0"), val = tensor("custom")]; + tensor k_325_pad_0 = const()[name = tensor("k_325_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_6_attn2_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_6_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3021896576)))]; + tensor k_325_cast_fp16 = conv(dilations = var_36395, groups = var_30355, pad = k_325_pad_0, pad_type = k_325_pad_type_0, strides = var_36393, weight = up_blocks_0_attentions_0_transformer_blocks_6_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_325_cast_fp16")]; + tensor var_36399 = const()[name = tensor("op_36399"), val = tensor([1, 1])]; + tensor var_36401 = const()[name = tensor("op_36401"), val = tensor([1, 1])]; + tensor v_163_pad_type_0 = const()[name = tensor("v_163_pad_type_0"), val = tensor("custom")]; + tensor v_163_pad_0 = const()[name = tensor("v_163_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_6_attn2_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_6_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3027139520)))]; + tensor v_163_cast_fp16 = conv(dilations = var_36401, groups = var_30355, pad = v_163_pad_0, pad_type = v_163_pad_type_0, strides = var_36399, weight = up_blocks_0_attentions_0_transformer_blocks_6_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_163_cast_fp16")]; + tensor var_36405_begin_0 = const()[name = tensor("op_36405_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_36405_end_0 = const()[name = tensor("op_36405_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_36405_end_mask_0 = const()[name = tensor("op_36405_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36405_cast_fp16 = slice_by_index(begin = var_36405_begin_0, end = var_36405_end_0, end_mask = var_36405_end_mask_0, x = q_163_cast_fp16)[name = tensor("op_36405_cast_fp16")]; + tensor var_36409_begin_0 = const()[name = tensor("op_36409_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_36409_end_0 = const()[name = tensor("op_36409_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_36409_end_mask_0 = const()[name = tensor("op_36409_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36409_cast_fp16 = slice_by_index(begin = var_36409_begin_0, end = var_36409_end_0, end_mask = var_36409_end_mask_0, x = q_163_cast_fp16)[name = tensor("op_36409_cast_fp16")]; + tensor var_36413_begin_0 = const()[name = tensor("op_36413_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_36413_end_0 = const()[name = tensor("op_36413_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_36413_end_mask_0 = const()[name = tensor("op_36413_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36413_cast_fp16 = slice_by_index(begin = var_36413_begin_0, end = var_36413_end_0, end_mask = var_36413_end_mask_0, x = q_163_cast_fp16)[name = tensor("op_36413_cast_fp16")]; + tensor var_36417_begin_0 = const()[name = tensor("op_36417_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_36417_end_0 = const()[name = tensor("op_36417_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_36417_end_mask_0 = const()[name = tensor("op_36417_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36417_cast_fp16 = slice_by_index(begin = var_36417_begin_0, end = var_36417_end_0, end_mask = var_36417_end_mask_0, x = q_163_cast_fp16)[name = tensor("op_36417_cast_fp16")]; + tensor var_36421_begin_0 = const()[name = tensor("op_36421_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_36421_end_0 = const()[name = tensor("op_36421_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_36421_end_mask_0 = const()[name = tensor("op_36421_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36421_cast_fp16 = slice_by_index(begin = var_36421_begin_0, end = var_36421_end_0, end_mask = var_36421_end_mask_0, x = q_163_cast_fp16)[name = tensor("op_36421_cast_fp16")]; + tensor var_36425_begin_0 = const()[name = tensor("op_36425_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_36425_end_0 = const()[name = tensor("op_36425_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_36425_end_mask_0 = const()[name = tensor("op_36425_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36425_cast_fp16 = slice_by_index(begin = var_36425_begin_0, end = var_36425_end_0, end_mask = var_36425_end_mask_0, x = q_163_cast_fp16)[name = tensor("op_36425_cast_fp16")]; + tensor var_36429_begin_0 = const()[name = tensor("op_36429_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_36429_end_0 = const()[name = tensor("op_36429_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_36429_end_mask_0 = const()[name = tensor("op_36429_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36429_cast_fp16 = slice_by_index(begin = var_36429_begin_0, end = var_36429_end_0, end_mask = var_36429_end_mask_0, x = q_163_cast_fp16)[name = tensor("op_36429_cast_fp16")]; + tensor var_36433_begin_0 = const()[name = tensor("op_36433_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_36433_end_0 = const()[name = tensor("op_36433_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_36433_end_mask_0 = const()[name = tensor("op_36433_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36433_cast_fp16 = slice_by_index(begin = var_36433_begin_0, end = var_36433_end_0, end_mask = var_36433_end_mask_0, x = q_163_cast_fp16)[name = tensor("op_36433_cast_fp16")]; + tensor var_36437_begin_0 = const()[name = tensor("op_36437_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_36437_end_0 = const()[name = tensor("op_36437_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_36437_end_mask_0 = const()[name = tensor("op_36437_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36437_cast_fp16 = slice_by_index(begin = var_36437_begin_0, end = var_36437_end_0, end_mask = var_36437_end_mask_0, x = q_163_cast_fp16)[name = tensor("op_36437_cast_fp16")]; + tensor var_36441_begin_0 = const()[name = tensor("op_36441_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_36441_end_0 = const()[name = tensor("op_36441_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_36441_end_mask_0 = const()[name = tensor("op_36441_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36441_cast_fp16 = slice_by_index(begin = var_36441_begin_0, end = var_36441_end_0, end_mask = var_36441_end_mask_0, x = q_163_cast_fp16)[name = tensor("op_36441_cast_fp16")]; + tensor var_36445_begin_0 = const()[name = tensor("op_36445_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_36445_end_0 = const()[name = tensor("op_36445_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_36445_end_mask_0 = const()[name = tensor("op_36445_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36445_cast_fp16 = slice_by_index(begin = var_36445_begin_0, end = var_36445_end_0, end_mask = var_36445_end_mask_0, x = q_163_cast_fp16)[name = tensor("op_36445_cast_fp16")]; + tensor var_36449_begin_0 = const()[name = tensor("op_36449_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_36449_end_0 = const()[name = tensor("op_36449_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_36449_end_mask_0 = const()[name = tensor("op_36449_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36449_cast_fp16 = slice_by_index(begin = var_36449_begin_0, end = var_36449_end_0, end_mask = var_36449_end_mask_0, x = q_163_cast_fp16)[name = tensor("op_36449_cast_fp16")]; + tensor var_36453_begin_0 = const()[name = tensor("op_36453_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_36453_end_0 = const()[name = tensor("op_36453_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_36453_end_mask_0 = const()[name = tensor("op_36453_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36453_cast_fp16 = slice_by_index(begin = var_36453_begin_0, end = var_36453_end_0, end_mask = var_36453_end_mask_0, x = q_163_cast_fp16)[name = tensor("op_36453_cast_fp16")]; + tensor var_36457_begin_0 = const()[name = tensor("op_36457_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_36457_end_0 = const()[name = tensor("op_36457_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_36457_end_mask_0 = const()[name = tensor("op_36457_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36457_cast_fp16 = slice_by_index(begin = var_36457_begin_0, end = var_36457_end_0, end_mask = var_36457_end_mask_0, x = q_163_cast_fp16)[name = tensor("op_36457_cast_fp16")]; + tensor var_36461_begin_0 = const()[name = tensor("op_36461_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_36461_end_0 = const()[name = tensor("op_36461_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_36461_end_mask_0 = const()[name = tensor("op_36461_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36461_cast_fp16 = slice_by_index(begin = var_36461_begin_0, end = var_36461_end_0, end_mask = var_36461_end_mask_0, x = q_163_cast_fp16)[name = tensor("op_36461_cast_fp16")]; + tensor var_36465_begin_0 = const()[name = tensor("op_36465_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_36465_end_0 = const()[name = tensor("op_36465_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_36465_end_mask_0 = const()[name = tensor("op_36465_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36465_cast_fp16 = slice_by_index(begin = var_36465_begin_0, end = var_36465_end_0, end_mask = var_36465_end_mask_0, x = q_163_cast_fp16)[name = tensor("op_36465_cast_fp16")]; + tensor var_36469_begin_0 = const()[name = tensor("op_36469_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_36469_end_0 = const()[name = tensor("op_36469_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_36469_end_mask_0 = const()[name = tensor("op_36469_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36469_cast_fp16 = slice_by_index(begin = var_36469_begin_0, end = var_36469_end_0, end_mask = var_36469_end_mask_0, x = q_163_cast_fp16)[name = tensor("op_36469_cast_fp16")]; + tensor var_36473_begin_0 = const()[name = tensor("op_36473_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_36473_end_0 = const()[name = tensor("op_36473_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_36473_end_mask_0 = const()[name = tensor("op_36473_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36473_cast_fp16 = slice_by_index(begin = var_36473_begin_0, end = var_36473_end_0, end_mask = var_36473_end_mask_0, x = q_163_cast_fp16)[name = tensor("op_36473_cast_fp16")]; + tensor var_36477_begin_0 = const()[name = tensor("op_36477_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_36477_end_0 = const()[name = tensor("op_36477_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_36477_end_mask_0 = const()[name = tensor("op_36477_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36477_cast_fp16 = slice_by_index(begin = var_36477_begin_0, end = var_36477_end_0, end_mask = var_36477_end_mask_0, x = q_163_cast_fp16)[name = tensor("op_36477_cast_fp16")]; + tensor var_36481_begin_0 = const()[name = tensor("op_36481_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_36481_end_0 = const()[name = tensor("op_36481_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_36481_end_mask_0 = const()[name = tensor("op_36481_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36481_cast_fp16 = slice_by_index(begin = var_36481_begin_0, end = var_36481_end_0, end_mask = var_36481_end_mask_0, x = q_163_cast_fp16)[name = tensor("op_36481_cast_fp16")]; + tensor k_327_perm_0 = const()[name = tensor("k_327_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_36488_begin_0 = const()[name = tensor("op_36488_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_36488_end_0 = const()[name = tensor("op_36488_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_36488_end_mask_0 = const()[name = tensor("op_36488_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_327_cast_fp16 = transpose(perm = k_327_perm_0, x = k_325_cast_fp16)[name = tensor("transpose_58")]; + tensor var_36488_cast_fp16 = slice_by_index(begin = var_36488_begin_0, end = var_36488_end_0, end_mask = var_36488_end_mask_0, x = k_327_cast_fp16)[name = tensor("op_36488_cast_fp16")]; + tensor var_36492_begin_0 = const()[name = tensor("op_36492_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_36492_end_0 = const()[name = tensor("op_36492_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_36492_end_mask_0 = const()[name = tensor("op_36492_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36492_cast_fp16 = slice_by_index(begin = var_36492_begin_0, end = var_36492_end_0, end_mask = var_36492_end_mask_0, x = k_327_cast_fp16)[name = tensor("op_36492_cast_fp16")]; + tensor var_36496_begin_0 = const()[name = tensor("op_36496_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_36496_end_0 = const()[name = tensor("op_36496_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_36496_end_mask_0 = const()[name = tensor("op_36496_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36496_cast_fp16 = slice_by_index(begin = var_36496_begin_0, end = var_36496_end_0, end_mask = var_36496_end_mask_0, x = k_327_cast_fp16)[name = tensor("op_36496_cast_fp16")]; + tensor var_36500_begin_0 = const()[name = tensor("op_36500_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_36500_end_0 = const()[name = tensor("op_36500_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_36500_end_mask_0 = const()[name = tensor("op_36500_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36500_cast_fp16 = slice_by_index(begin = var_36500_begin_0, end = var_36500_end_0, end_mask = var_36500_end_mask_0, x = k_327_cast_fp16)[name = tensor("op_36500_cast_fp16")]; + tensor var_36504_begin_0 = const()[name = tensor("op_36504_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_36504_end_0 = const()[name = tensor("op_36504_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_36504_end_mask_0 = const()[name = tensor("op_36504_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36504_cast_fp16 = slice_by_index(begin = var_36504_begin_0, end = var_36504_end_0, end_mask = var_36504_end_mask_0, x = k_327_cast_fp16)[name = tensor("op_36504_cast_fp16")]; + tensor var_36508_begin_0 = const()[name = tensor("op_36508_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_36508_end_0 = const()[name = tensor("op_36508_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_36508_end_mask_0 = const()[name = tensor("op_36508_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36508_cast_fp16 = slice_by_index(begin = var_36508_begin_0, end = var_36508_end_0, end_mask = var_36508_end_mask_0, x = k_327_cast_fp16)[name = tensor("op_36508_cast_fp16")]; + tensor var_36512_begin_0 = const()[name = tensor("op_36512_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_36512_end_0 = const()[name = tensor("op_36512_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_36512_end_mask_0 = const()[name = tensor("op_36512_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36512_cast_fp16 = slice_by_index(begin = var_36512_begin_0, end = var_36512_end_0, end_mask = var_36512_end_mask_0, x = k_327_cast_fp16)[name = tensor("op_36512_cast_fp16")]; + tensor var_36516_begin_0 = const()[name = tensor("op_36516_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_36516_end_0 = const()[name = tensor("op_36516_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_36516_end_mask_0 = const()[name = tensor("op_36516_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36516_cast_fp16 = slice_by_index(begin = var_36516_begin_0, end = var_36516_end_0, end_mask = var_36516_end_mask_0, x = k_327_cast_fp16)[name = tensor("op_36516_cast_fp16")]; + tensor var_36520_begin_0 = const()[name = tensor("op_36520_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_36520_end_0 = const()[name = tensor("op_36520_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_36520_end_mask_0 = const()[name = tensor("op_36520_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36520_cast_fp16 = slice_by_index(begin = var_36520_begin_0, end = var_36520_end_0, end_mask = var_36520_end_mask_0, x = k_327_cast_fp16)[name = tensor("op_36520_cast_fp16")]; + tensor var_36524_begin_0 = const()[name = tensor("op_36524_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_36524_end_0 = const()[name = tensor("op_36524_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_36524_end_mask_0 = const()[name = tensor("op_36524_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36524_cast_fp16 = slice_by_index(begin = var_36524_begin_0, end = var_36524_end_0, end_mask = var_36524_end_mask_0, x = k_327_cast_fp16)[name = tensor("op_36524_cast_fp16")]; + tensor var_36528_begin_0 = const()[name = tensor("op_36528_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_36528_end_0 = const()[name = tensor("op_36528_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_36528_end_mask_0 = const()[name = tensor("op_36528_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36528_cast_fp16 = slice_by_index(begin = var_36528_begin_0, end = var_36528_end_0, end_mask = var_36528_end_mask_0, x = k_327_cast_fp16)[name = tensor("op_36528_cast_fp16")]; + tensor var_36532_begin_0 = const()[name = tensor("op_36532_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_36532_end_0 = const()[name = tensor("op_36532_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_36532_end_mask_0 = const()[name = tensor("op_36532_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36532_cast_fp16 = slice_by_index(begin = var_36532_begin_0, end = var_36532_end_0, end_mask = var_36532_end_mask_0, x = k_327_cast_fp16)[name = tensor("op_36532_cast_fp16")]; + tensor var_36536_begin_0 = const()[name = tensor("op_36536_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_36536_end_0 = const()[name = tensor("op_36536_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_36536_end_mask_0 = const()[name = tensor("op_36536_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36536_cast_fp16 = slice_by_index(begin = var_36536_begin_0, end = var_36536_end_0, end_mask = var_36536_end_mask_0, x = k_327_cast_fp16)[name = tensor("op_36536_cast_fp16")]; + tensor var_36540_begin_0 = const()[name = tensor("op_36540_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_36540_end_0 = const()[name = tensor("op_36540_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_36540_end_mask_0 = const()[name = tensor("op_36540_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36540_cast_fp16 = slice_by_index(begin = var_36540_begin_0, end = var_36540_end_0, end_mask = var_36540_end_mask_0, x = k_327_cast_fp16)[name = tensor("op_36540_cast_fp16")]; + tensor var_36544_begin_0 = const()[name = tensor("op_36544_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_36544_end_0 = const()[name = tensor("op_36544_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_36544_end_mask_0 = const()[name = tensor("op_36544_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36544_cast_fp16 = slice_by_index(begin = var_36544_begin_0, end = var_36544_end_0, end_mask = var_36544_end_mask_0, x = k_327_cast_fp16)[name = tensor("op_36544_cast_fp16")]; + tensor var_36548_begin_0 = const()[name = tensor("op_36548_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_36548_end_0 = const()[name = tensor("op_36548_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_36548_end_mask_0 = const()[name = tensor("op_36548_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36548_cast_fp16 = slice_by_index(begin = var_36548_begin_0, end = var_36548_end_0, end_mask = var_36548_end_mask_0, x = k_327_cast_fp16)[name = tensor("op_36548_cast_fp16")]; + tensor var_36552_begin_0 = const()[name = tensor("op_36552_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_36552_end_0 = const()[name = tensor("op_36552_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_36552_end_mask_0 = const()[name = tensor("op_36552_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36552_cast_fp16 = slice_by_index(begin = var_36552_begin_0, end = var_36552_end_0, end_mask = var_36552_end_mask_0, x = k_327_cast_fp16)[name = tensor("op_36552_cast_fp16")]; + tensor var_36556_begin_0 = const()[name = tensor("op_36556_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_36556_end_0 = const()[name = tensor("op_36556_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_36556_end_mask_0 = const()[name = tensor("op_36556_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36556_cast_fp16 = slice_by_index(begin = var_36556_begin_0, end = var_36556_end_0, end_mask = var_36556_end_mask_0, x = k_327_cast_fp16)[name = tensor("op_36556_cast_fp16")]; + tensor var_36560_begin_0 = const()[name = tensor("op_36560_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_36560_end_0 = const()[name = tensor("op_36560_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_36560_end_mask_0 = const()[name = tensor("op_36560_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36560_cast_fp16 = slice_by_index(begin = var_36560_begin_0, end = var_36560_end_0, end_mask = var_36560_end_mask_0, x = k_327_cast_fp16)[name = tensor("op_36560_cast_fp16")]; + tensor var_36564_begin_0 = const()[name = tensor("op_36564_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_36564_end_0 = const()[name = tensor("op_36564_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_36564_end_mask_0 = const()[name = tensor("op_36564_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36564_cast_fp16 = slice_by_index(begin = var_36564_begin_0, end = var_36564_end_0, end_mask = var_36564_end_mask_0, x = k_327_cast_fp16)[name = tensor("op_36564_cast_fp16")]; + tensor var_36566_begin_0 = const()[name = tensor("op_36566_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_36566_end_0 = const()[name = tensor("op_36566_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_36566_end_mask_0 = const()[name = tensor("op_36566_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36566_cast_fp16 = slice_by_index(begin = var_36566_begin_0, end = var_36566_end_0, end_mask = var_36566_end_mask_0, x = v_163_cast_fp16)[name = tensor("op_36566_cast_fp16")]; + tensor var_36570_begin_0 = const()[name = tensor("op_36570_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_36570_end_0 = const()[name = tensor("op_36570_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_36570_end_mask_0 = const()[name = tensor("op_36570_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36570_cast_fp16 = slice_by_index(begin = var_36570_begin_0, end = var_36570_end_0, end_mask = var_36570_end_mask_0, x = v_163_cast_fp16)[name = tensor("op_36570_cast_fp16")]; + tensor var_36574_begin_0 = const()[name = tensor("op_36574_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_36574_end_0 = const()[name = tensor("op_36574_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_36574_end_mask_0 = const()[name = tensor("op_36574_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36574_cast_fp16 = slice_by_index(begin = var_36574_begin_0, end = var_36574_end_0, end_mask = var_36574_end_mask_0, x = v_163_cast_fp16)[name = tensor("op_36574_cast_fp16")]; + tensor var_36578_begin_0 = const()[name = tensor("op_36578_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_36578_end_0 = const()[name = tensor("op_36578_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_36578_end_mask_0 = const()[name = tensor("op_36578_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36578_cast_fp16 = slice_by_index(begin = var_36578_begin_0, end = var_36578_end_0, end_mask = var_36578_end_mask_0, x = v_163_cast_fp16)[name = tensor("op_36578_cast_fp16")]; + tensor var_36582_begin_0 = const()[name = tensor("op_36582_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_36582_end_0 = const()[name = tensor("op_36582_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_36582_end_mask_0 = const()[name = tensor("op_36582_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36582_cast_fp16 = slice_by_index(begin = var_36582_begin_0, end = var_36582_end_0, end_mask = var_36582_end_mask_0, x = v_163_cast_fp16)[name = tensor("op_36582_cast_fp16")]; + tensor var_36586_begin_0 = const()[name = tensor("op_36586_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_36586_end_0 = const()[name = tensor("op_36586_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_36586_end_mask_0 = const()[name = tensor("op_36586_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36586_cast_fp16 = slice_by_index(begin = var_36586_begin_0, end = var_36586_end_0, end_mask = var_36586_end_mask_0, x = v_163_cast_fp16)[name = tensor("op_36586_cast_fp16")]; + tensor var_36590_begin_0 = const()[name = tensor("op_36590_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_36590_end_0 = const()[name = tensor("op_36590_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_36590_end_mask_0 = const()[name = tensor("op_36590_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36590_cast_fp16 = slice_by_index(begin = var_36590_begin_0, end = var_36590_end_0, end_mask = var_36590_end_mask_0, x = v_163_cast_fp16)[name = tensor("op_36590_cast_fp16")]; + tensor var_36594_begin_0 = const()[name = tensor("op_36594_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_36594_end_0 = const()[name = tensor("op_36594_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_36594_end_mask_0 = const()[name = tensor("op_36594_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36594_cast_fp16 = slice_by_index(begin = var_36594_begin_0, end = var_36594_end_0, end_mask = var_36594_end_mask_0, x = v_163_cast_fp16)[name = tensor("op_36594_cast_fp16")]; + tensor var_36598_begin_0 = const()[name = tensor("op_36598_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_36598_end_0 = const()[name = tensor("op_36598_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_36598_end_mask_0 = const()[name = tensor("op_36598_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36598_cast_fp16 = slice_by_index(begin = var_36598_begin_0, end = var_36598_end_0, end_mask = var_36598_end_mask_0, x = v_163_cast_fp16)[name = tensor("op_36598_cast_fp16")]; + tensor var_36602_begin_0 = const()[name = tensor("op_36602_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_36602_end_0 = const()[name = tensor("op_36602_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_36602_end_mask_0 = const()[name = tensor("op_36602_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36602_cast_fp16 = slice_by_index(begin = var_36602_begin_0, end = var_36602_end_0, end_mask = var_36602_end_mask_0, x = v_163_cast_fp16)[name = tensor("op_36602_cast_fp16")]; + tensor var_36606_begin_0 = const()[name = tensor("op_36606_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_36606_end_0 = const()[name = tensor("op_36606_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_36606_end_mask_0 = const()[name = tensor("op_36606_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36606_cast_fp16 = slice_by_index(begin = var_36606_begin_0, end = var_36606_end_0, end_mask = var_36606_end_mask_0, x = v_163_cast_fp16)[name = tensor("op_36606_cast_fp16")]; + tensor var_36610_begin_0 = const()[name = tensor("op_36610_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_36610_end_0 = const()[name = tensor("op_36610_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_36610_end_mask_0 = const()[name = tensor("op_36610_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36610_cast_fp16 = slice_by_index(begin = var_36610_begin_0, end = var_36610_end_0, end_mask = var_36610_end_mask_0, x = v_163_cast_fp16)[name = tensor("op_36610_cast_fp16")]; + tensor var_36614_begin_0 = const()[name = tensor("op_36614_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_36614_end_0 = const()[name = tensor("op_36614_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_36614_end_mask_0 = const()[name = tensor("op_36614_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36614_cast_fp16 = slice_by_index(begin = var_36614_begin_0, end = var_36614_end_0, end_mask = var_36614_end_mask_0, x = v_163_cast_fp16)[name = tensor("op_36614_cast_fp16")]; + tensor var_36618_begin_0 = const()[name = tensor("op_36618_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_36618_end_0 = const()[name = tensor("op_36618_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_36618_end_mask_0 = const()[name = tensor("op_36618_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36618_cast_fp16 = slice_by_index(begin = var_36618_begin_0, end = var_36618_end_0, end_mask = var_36618_end_mask_0, x = v_163_cast_fp16)[name = tensor("op_36618_cast_fp16")]; + tensor var_36622_begin_0 = const()[name = tensor("op_36622_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_36622_end_0 = const()[name = tensor("op_36622_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_36622_end_mask_0 = const()[name = tensor("op_36622_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36622_cast_fp16 = slice_by_index(begin = var_36622_begin_0, end = var_36622_end_0, end_mask = var_36622_end_mask_0, x = v_163_cast_fp16)[name = tensor("op_36622_cast_fp16")]; + tensor var_36626_begin_0 = const()[name = tensor("op_36626_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_36626_end_0 = const()[name = tensor("op_36626_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_36626_end_mask_0 = const()[name = tensor("op_36626_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36626_cast_fp16 = slice_by_index(begin = var_36626_begin_0, end = var_36626_end_0, end_mask = var_36626_end_mask_0, x = v_163_cast_fp16)[name = tensor("op_36626_cast_fp16")]; + tensor var_36630_begin_0 = const()[name = tensor("op_36630_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_36630_end_0 = const()[name = tensor("op_36630_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_36630_end_mask_0 = const()[name = tensor("op_36630_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36630_cast_fp16 = slice_by_index(begin = var_36630_begin_0, end = var_36630_end_0, end_mask = var_36630_end_mask_0, x = v_163_cast_fp16)[name = tensor("op_36630_cast_fp16")]; + tensor var_36634_begin_0 = const()[name = tensor("op_36634_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_36634_end_0 = const()[name = tensor("op_36634_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_36634_end_mask_0 = const()[name = tensor("op_36634_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36634_cast_fp16 = slice_by_index(begin = var_36634_begin_0, end = var_36634_end_0, end_mask = var_36634_end_mask_0, x = v_163_cast_fp16)[name = tensor("op_36634_cast_fp16")]; + tensor var_36638_begin_0 = const()[name = tensor("op_36638_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_36638_end_0 = const()[name = tensor("op_36638_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_36638_end_mask_0 = const()[name = tensor("op_36638_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36638_cast_fp16 = slice_by_index(begin = var_36638_begin_0, end = var_36638_end_0, end_mask = var_36638_end_mask_0, x = v_163_cast_fp16)[name = tensor("op_36638_cast_fp16")]; + tensor var_36642_begin_0 = const()[name = tensor("op_36642_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_36642_end_0 = const()[name = tensor("op_36642_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_36642_end_mask_0 = const()[name = tensor("op_36642_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36642_cast_fp16 = slice_by_index(begin = var_36642_begin_0, end = var_36642_end_0, end_mask = var_36642_end_mask_0, x = v_163_cast_fp16)[name = tensor("op_36642_cast_fp16")]; + tensor var_36646_equation_0 = const()[name = tensor("op_36646_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_36646_cast_fp16 = einsum(equation = var_36646_equation_0, values = (var_36488_cast_fp16, var_36405_cast_fp16))[name = tensor("op_36646_cast_fp16")]; + tensor var_36647_to_fp16 = const()[name = tensor("op_36647_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3081_cast_fp16 = mul(x = var_36646_cast_fp16, y = var_36647_to_fp16)[name = tensor("aw_3081_cast_fp16")]; + tensor var_36650_equation_0 = const()[name = tensor("op_36650_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_36650_cast_fp16 = einsum(equation = var_36650_equation_0, values = (var_36492_cast_fp16, var_36409_cast_fp16))[name = tensor("op_36650_cast_fp16")]; + tensor var_36651_to_fp16 = const()[name = tensor("op_36651_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3083_cast_fp16 = mul(x = var_36650_cast_fp16, y = var_36651_to_fp16)[name = tensor("aw_3083_cast_fp16")]; + tensor var_36654_equation_0 = const()[name = tensor("op_36654_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_36654_cast_fp16 = einsum(equation = var_36654_equation_0, values = (var_36496_cast_fp16, var_36413_cast_fp16))[name = tensor("op_36654_cast_fp16")]; + tensor var_36655_to_fp16 = const()[name = tensor("op_36655_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3085_cast_fp16 = mul(x = var_36654_cast_fp16, y = var_36655_to_fp16)[name = tensor("aw_3085_cast_fp16")]; + tensor var_36658_equation_0 = const()[name = tensor("op_36658_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_36658_cast_fp16 = einsum(equation = var_36658_equation_0, values = (var_36500_cast_fp16, var_36417_cast_fp16))[name = tensor("op_36658_cast_fp16")]; + tensor var_36659_to_fp16 = const()[name = tensor("op_36659_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3087_cast_fp16 = mul(x = var_36658_cast_fp16, y = var_36659_to_fp16)[name = tensor("aw_3087_cast_fp16")]; + tensor var_36662_equation_0 = const()[name = tensor("op_36662_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_36662_cast_fp16 = einsum(equation = var_36662_equation_0, values = (var_36504_cast_fp16, var_36421_cast_fp16))[name = tensor("op_36662_cast_fp16")]; + tensor var_36663_to_fp16 = const()[name = tensor("op_36663_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3089_cast_fp16 = mul(x = var_36662_cast_fp16, y = var_36663_to_fp16)[name = tensor("aw_3089_cast_fp16")]; + tensor var_36666_equation_0 = const()[name = tensor("op_36666_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_36666_cast_fp16 = einsum(equation = var_36666_equation_0, values = (var_36508_cast_fp16, var_36425_cast_fp16))[name = tensor("op_36666_cast_fp16")]; + tensor var_36667_to_fp16 = const()[name = tensor("op_36667_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3091_cast_fp16 = mul(x = var_36666_cast_fp16, y = var_36667_to_fp16)[name = tensor("aw_3091_cast_fp16")]; + tensor var_36670_equation_0 = const()[name = tensor("op_36670_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_36670_cast_fp16 = einsum(equation = var_36670_equation_0, values = (var_36512_cast_fp16, var_36429_cast_fp16))[name = tensor("op_36670_cast_fp16")]; + tensor var_36671_to_fp16 = const()[name = tensor("op_36671_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3093_cast_fp16 = mul(x = var_36670_cast_fp16, y = var_36671_to_fp16)[name = tensor("aw_3093_cast_fp16")]; + tensor var_36674_equation_0 = const()[name = tensor("op_36674_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_36674_cast_fp16 = einsum(equation = var_36674_equation_0, values = (var_36516_cast_fp16, var_36433_cast_fp16))[name = tensor("op_36674_cast_fp16")]; + tensor var_36675_to_fp16 = const()[name = tensor("op_36675_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3095_cast_fp16 = mul(x = var_36674_cast_fp16, y = var_36675_to_fp16)[name = tensor("aw_3095_cast_fp16")]; + tensor var_36678_equation_0 = const()[name = tensor("op_36678_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_36678_cast_fp16 = einsum(equation = var_36678_equation_0, values = (var_36520_cast_fp16, var_36437_cast_fp16))[name = tensor("op_36678_cast_fp16")]; + tensor var_36679_to_fp16 = const()[name = tensor("op_36679_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3097_cast_fp16 = mul(x = var_36678_cast_fp16, y = var_36679_to_fp16)[name = tensor("aw_3097_cast_fp16")]; + tensor var_36682_equation_0 = const()[name = tensor("op_36682_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_36682_cast_fp16 = einsum(equation = var_36682_equation_0, values = (var_36524_cast_fp16, var_36441_cast_fp16))[name = tensor("op_36682_cast_fp16")]; + tensor var_36683_to_fp16 = const()[name = tensor("op_36683_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3099_cast_fp16 = mul(x = var_36682_cast_fp16, y = var_36683_to_fp16)[name = tensor("aw_3099_cast_fp16")]; + tensor var_36686_equation_0 = const()[name = tensor("op_36686_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_36686_cast_fp16 = einsum(equation = var_36686_equation_0, values = (var_36528_cast_fp16, var_36445_cast_fp16))[name = tensor("op_36686_cast_fp16")]; + tensor var_36687_to_fp16 = const()[name = tensor("op_36687_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3101_cast_fp16 = mul(x = var_36686_cast_fp16, y = var_36687_to_fp16)[name = tensor("aw_3101_cast_fp16")]; + tensor var_36690_equation_0 = const()[name = tensor("op_36690_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_36690_cast_fp16 = einsum(equation = var_36690_equation_0, values = (var_36532_cast_fp16, var_36449_cast_fp16))[name = tensor("op_36690_cast_fp16")]; + tensor var_36691_to_fp16 = const()[name = tensor("op_36691_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3103_cast_fp16 = mul(x = var_36690_cast_fp16, y = var_36691_to_fp16)[name = tensor("aw_3103_cast_fp16")]; + tensor var_36694_equation_0 = const()[name = tensor("op_36694_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_36694_cast_fp16 = einsum(equation = var_36694_equation_0, values = (var_36536_cast_fp16, var_36453_cast_fp16))[name = tensor("op_36694_cast_fp16")]; + tensor var_36695_to_fp16 = const()[name = tensor("op_36695_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3105_cast_fp16 = mul(x = var_36694_cast_fp16, y = var_36695_to_fp16)[name = tensor("aw_3105_cast_fp16")]; + tensor var_36698_equation_0 = const()[name = tensor("op_36698_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_36698_cast_fp16 = einsum(equation = var_36698_equation_0, values = (var_36540_cast_fp16, var_36457_cast_fp16))[name = tensor("op_36698_cast_fp16")]; + tensor var_36699_to_fp16 = const()[name = tensor("op_36699_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3107_cast_fp16 = mul(x = var_36698_cast_fp16, y = var_36699_to_fp16)[name = tensor("aw_3107_cast_fp16")]; + tensor var_36702_equation_0 = const()[name = tensor("op_36702_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_36702_cast_fp16 = einsum(equation = var_36702_equation_0, values = (var_36544_cast_fp16, var_36461_cast_fp16))[name = tensor("op_36702_cast_fp16")]; + tensor var_36703_to_fp16 = const()[name = tensor("op_36703_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3109_cast_fp16 = mul(x = var_36702_cast_fp16, y = var_36703_to_fp16)[name = tensor("aw_3109_cast_fp16")]; + tensor var_36706_equation_0 = const()[name = tensor("op_36706_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_36706_cast_fp16 = einsum(equation = var_36706_equation_0, values = (var_36548_cast_fp16, var_36465_cast_fp16))[name = tensor("op_36706_cast_fp16")]; + tensor var_36707_to_fp16 = const()[name = tensor("op_36707_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3111_cast_fp16 = mul(x = var_36706_cast_fp16, y = var_36707_to_fp16)[name = tensor("aw_3111_cast_fp16")]; + tensor var_36710_equation_0 = const()[name = tensor("op_36710_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_36710_cast_fp16 = einsum(equation = var_36710_equation_0, values = (var_36552_cast_fp16, var_36469_cast_fp16))[name = tensor("op_36710_cast_fp16")]; + tensor var_36711_to_fp16 = const()[name = tensor("op_36711_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3113_cast_fp16 = mul(x = var_36710_cast_fp16, y = var_36711_to_fp16)[name = tensor("aw_3113_cast_fp16")]; + tensor var_36714_equation_0 = const()[name = tensor("op_36714_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_36714_cast_fp16 = einsum(equation = var_36714_equation_0, values = (var_36556_cast_fp16, var_36473_cast_fp16))[name = tensor("op_36714_cast_fp16")]; + tensor var_36715_to_fp16 = const()[name = tensor("op_36715_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3115_cast_fp16 = mul(x = var_36714_cast_fp16, y = var_36715_to_fp16)[name = tensor("aw_3115_cast_fp16")]; + tensor var_36718_equation_0 = const()[name = tensor("op_36718_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_36718_cast_fp16 = einsum(equation = var_36718_equation_0, values = (var_36560_cast_fp16, var_36477_cast_fp16))[name = tensor("op_36718_cast_fp16")]; + tensor var_36719_to_fp16 = const()[name = tensor("op_36719_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3117_cast_fp16 = mul(x = var_36718_cast_fp16, y = var_36719_to_fp16)[name = tensor("aw_3117_cast_fp16")]; + tensor var_36722_equation_0 = const()[name = tensor("op_36722_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_36722_cast_fp16 = einsum(equation = var_36722_equation_0, values = (var_36564_cast_fp16, var_36481_cast_fp16))[name = tensor("op_36722_cast_fp16")]; + tensor var_36723_to_fp16 = const()[name = tensor("op_36723_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3119_cast_fp16 = mul(x = var_36722_cast_fp16, y = var_36723_to_fp16)[name = tensor("aw_3119_cast_fp16")]; + tensor var_36725_cast_fp16 = softmax(axis = var_30355, x = aw_3081_cast_fp16)[name = tensor("op_36725_cast_fp16")]; + tensor var_36726_cast_fp16 = softmax(axis = var_30355, x = aw_3083_cast_fp16)[name = tensor("op_36726_cast_fp16")]; + tensor var_36727_cast_fp16 = softmax(axis = var_30355, x = aw_3085_cast_fp16)[name = tensor("op_36727_cast_fp16")]; + tensor var_36728_cast_fp16 = softmax(axis = var_30355, x = aw_3087_cast_fp16)[name = tensor("op_36728_cast_fp16")]; + tensor var_36729_cast_fp16 = softmax(axis = var_30355, x = aw_3089_cast_fp16)[name = tensor("op_36729_cast_fp16")]; + tensor var_36730_cast_fp16 = softmax(axis = var_30355, x = aw_3091_cast_fp16)[name = tensor("op_36730_cast_fp16")]; + tensor var_36731_cast_fp16 = softmax(axis = var_30355, x = aw_3093_cast_fp16)[name = tensor("op_36731_cast_fp16")]; + tensor var_36732_cast_fp16 = softmax(axis = var_30355, x = aw_3095_cast_fp16)[name = tensor("op_36732_cast_fp16")]; + tensor var_36733_cast_fp16 = softmax(axis = var_30355, x = aw_3097_cast_fp16)[name = tensor("op_36733_cast_fp16")]; + tensor var_36734_cast_fp16 = softmax(axis = var_30355, x = aw_3099_cast_fp16)[name = tensor("op_36734_cast_fp16")]; + tensor var_36735_cast_fp16 = softmax(axis = var_30355, x = aw_3101_cast_fp16)[name = tensor("op_36735_cast_fp16")]; + tensor var_36736_cast_fp16 = softmax(axis = var_30355, x = aw_3103_cast_fp16)[name = tensor("op_36736_cast_fp16")]; + tensor var_36737_cast_fp16 = softmax(axis = var_30355, x = aw_3105_cast_fp16)[name = tensor("op_36737_cast_fp16")]; + tensor var_36738_cast_fp16 = softmax(axis = var_30355, x = aw_3107_cast_fp16)[name = tensor("op_36738_cast_fp16")]; + tensor var_36739_cast_fp16 = softmax(axis = var_30355, x = aw_3109_cast_fp16)[name = tensor("op_36739_cast_fp16")]; + tensor var_36740_cast_fp16 = softmax(axis = var_30355, x = aw_3111_cast_fp16)[name = tensor("op_36740_cast_fp16")]; + tensor var_36741_cast_fp16 = softmax(axis = var_30355, x = aw_3113_cast_fp16)[name = tensor("op_36741_cast_fp16")]; + tensor var_36742_cast_fp16 = softmax(axis = var_30355, x = aw_3115_cast_fp16)[name = tensor("op_36742_cast_fp16")]; + tensor var_36743_cast_fp16 = softmax(axis = var_30355, x = aw_3117_cast_fp16)[name = tensor("op_36743_cast_fp16")]; + tensor var_36744_cast_fp16 = softmax(axis = var_30355, x = aw_3119_cast_fp16)[name = tensor("op_36744_cast_fp16")]; + tensor var_36746_equation_0 = const()[name = tensor("op_36746_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_36746_cast_fp16 = einsum(equation = var_36746_equation_0, values = (var_36566_cast_fp16, var_36725_cast_fp16))[name = tensor("op_36746_cast_fp16")]; + tensor var_36748_equation_0 = const()[name = tensor("op_36748_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_36748_cast_fp16 = einsum(equation = var_36748_equation_0, values = (var_36570_cast_fp16, var_36726_cast_fp16))[name = tensor("op_36748_cast_fp16")]; + tensor var_36750_equation_0 = const()[name = tensor("op_36750_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_36750_cast_fp16 = einsum(equation = var_36750_equation_0, values = (var_36574_cast_fp16, var_36727_cast_fp16))[name = tensor("op_36750_cast_fp16")]; + tensor var_36752_equation_0 = const()[name = tensor("op_36752_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_36752_cast_fp16 = einsum(equation = var_36752_equation_0, values = (var_36578_cast_fp16, var_36728_cast_fp16))[name = tensor("op_36752_cast_fp16")]; + tensor var_36754_equation_0 = const()[name = tensor("op_36754_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_36754_cast_fp16 = einsum(equation = var_36754_equation_0, values = (var_36582_cast_fp16, var_36729_cast_fp16))[name = tensor("op_36754_cast_fp16")]; + tensor var_36756_equation_0 = const()[name = tensor("op_36756_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_36756_cast_fp16 = einsum(equation = var_36756_equation_0, values = (var_36586_cast_fp16, var_36730_cast_fp16))[name = tensor("op_36756_cast_fp16")]; + tensor var_36758_equation_0 = const()[name = tensor("op_36758_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_36758_cast_fp16 = einsum(equation = var_36758_equation_0, values = (var_36590_cast_fp16, var_36731_cast_fp16))[name = tensor("op_36758_cast_fp16")]; + tensor var_36760_equation_0 = const()[name = tensor("op_36760_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_36760_cast_fp16 = einsum(equation = var_36760_equation_0, values = (var_36594_cast_fp16, var_36732_cast_fp16))[name = tensor("op_36760_cast_fp16")]; + tensor var_36762_equation_0 = const()[name = tensor("op_36762_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_36762_cast_fp16 = einsum(equation = var_36762_equation_0, values = (var_36598_cast_fp16, var_36733_cast_fp16))[name = tensor("op_36762_cast_fp16")]; + tensor var_36764_equation_0 = const()[name = tensor("op_36764_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_36764_cast_fp16 = einsum(equation = var_36764_equation_0, values = (var_36602_cast_fp16, var_36734_cast_fp16))[name = tensor("op_36764_cast_fp16")]; + tensor var_36766_equation_0 = const()[name = tensor("op_36766_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_36766_cast_fp16 = einsum(equation = var_36766_equation_0, values = (var_36606_cast_fp16, var_36735_cast_fp16))[name = tensor("op_36766_cast_fp16")]; + tensor var_36768_equation_0 = const()[name = tensor("op_36768_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_36768_cast_fp16 = einsum(equation = var_36768_equation_0, values = (var_36610_cast_fp16, var_36736_cast_fp16))[name = tensor("op_36768_cast_fp16")]; + tensor var_36770_equation_0 = const()[name = tensor("op_36770_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_36770_cast_fp16 = einsum(equation = var_36770_equation_0, values = (var_36614_cast_fp16, var_36737_cast_fp16))[name = tensor("op_36770_cast_fp16")]; + tensor var_36772_equation_0 = const()[name = tensor("op_36772_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_36772_cast_fp16 = einsum(equation = var_36772_equation_0, values = (var_36618_cast_fp16, var_36738_cast_fp16))[name = tensor("op_36772_cast_fp16")]; + tensor var_36774_equation_0 = const()[name = tensor("op_36774_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_36774_cast_fp16 = einsum(equation = var_36774_equation_0, values = (var_36622_cast_fp16, var_36739_cast_fp16))[name = tensor("op_36774_cast_fp16")]; + tensor var_36776_equation_0 = const()[name = tensor("op_36776_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_36776_cast_fp16 = einsum(equation = var_36776_equation_0, values = (var_36626_cast_fp16, var_36740_cast_fp16))[name = tensor("op_36776_cast_fp16")]; + tensor var_36778_equation_0 = const()[name = tensor("op_36778_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_36778_cast_fp16 = einsum(equation = var_36778_equation_0, values = (var_36630_cast_fp16, var_36741_cast_fp16))[name = tensor("op_36778_cast_fp16")]; + tensor var_36780_equation_0 = const()[name = tensor("op_36780_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_36780_cast_fp16 = einsum(equation = var_36780_equation_0, values = (var_36634_cast_fp16, var_36742_cast_fp16))[name = tensor("op_36780_cast_fp16")]; + tensor var_36782_equation_0 = const()[name = tensor("op_36782_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_36782_cast_fp16 = einsum(equation = var_36782_equation_0, values = (var_36638_cast_fp16, var_36743_cast_fp16))[name = tensor("op_36782_cast_fp16")]; + tensor var_36784_equation_0 = const()[name = tensor("op_36784_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_36784_cast_fp16 = einsum(equation = var_36784_equation_0, values = (var_36642_cast_fp16, var_36744_cast_fp16))[name = tensor("op_36784_cast_fp16")]; + tensor input_489_interleave_0 = const()[name = tensor("input_489_interleave_0"), val = tensor(false)]; + tensor input_489_cast_fp16 = concat(axis = var_30355, interleave = input_489_interleave_0, values = (var_36746_cast_fp16, var_36748_cast_fp16, var_36750_cast_fp16, var_36752_cast_fp16, var_36754_cast_fp16, var_36756_cast_fp16, var_36758_cast_fp16, var_36760_cast_fp16, var_36762_cast_fp16, var_36764_cast_fp16, var_36766_cast_fp16, var_36768_cast_fp16, var_36770_cast_fp16, var_36772_cast_fp16, var_36774_cast_fp16, var_36776_cast_fp16, var_36778_cast_fp16, var_36780_cast_fp16, var_36782_cast_fp16, var_36784_cast_fp16))[name = tensor("input_489_cast_fp16")]; + tensor var_36790 = const()[name = tensor("op_36790"), val = tensor([1, 1])]; + tensor var_36792 = const()[name = tensor("op_36792"), val = tensor([1, 1])]; + tensor var_36794_pad_type_0 = const()[name = tensor("op_36794_pad_type_0"), val = tensor("custom")]; + tensor var_36794_pad_0 = const()[name = tensor("op_36794_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_6_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_6_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3032382464)))]; + tensor up_blocks_0_attentions_0_transformer_blocks_6_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_6_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3035659328)))]; + tensor var_36794_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_6_attn2_to_out_0_bias_to_fp16, dilations = var_36792, groups = var_30355, pad = var_36794_pad_0, pad_type = var_36794_pad_type_0, strides = var_36790, weight = up_blocks_0_attentions_0_transformer_blocks_6_attn2_to_out_0_weight_to_fp16, x = input_489_cast_fp16)[name = tensor("op_36794_cast_fp16")]; + tensor inputs_245_cast_fp16 = add(x = var_36794_cast_fp16, y = inputs_243_cast_fp16)[name = tensor("inputs_245_cast_fp16")]; + tensor input_491_axes_0 = const()[name = tensor("input_491_axes_0"), val = tensor([1])]; + tensor input_491_gamma_0_to_fp16 = const()[name = tensor("input_491_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3035661952)))]; + tensor input_491_beta_0_to_fp16 = const()[name = tensor("input_491_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3035664576)))]; + tensor var_36804_to_fp16 = const()[name = tensor("op_36804_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_491_cast_fp16 = layer_norm(axes = input_491_axes_0, beta = input_491_beta_0_to_fp16, epsilon = var_36804_to_fp16, gamma = input_491_gamma_0_to_fp16, x = inputs_245_cast_fp16)[name = tensor("input_491_cast_fp16")]; + tensor var_36820 = const()[name = tensor("op_36820"), val = tensor([1, 1])]; + tensor var_36822 = const()[name = tensor("op_36822"), val = tensor([1, 1])]; + tensor var_36824_pad_type_0 = const()[name = tensor("op_36824_pad_type_0"), val = tensor("custom")]; + tensor var_36824_pad_0 = const()[name = tensor("op_36824_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_6_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_6_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3035667200)))]; + tensor up_blocks_0_attentions_0_transformer_blocks_6_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_6_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3061881664)))]; + tensor var_36824_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_6_ff_net_0_proj_bias_to_fp16, dilations = var_36822, groups = var_30355, pad = var_36824_pad_0, pad_type = var_36824_pad_type_0, strides = var_36820, weight = up_blocks_0_attentions_0_transformer_blocks_6_ff_net_0_proj_weight_to_fp16, x = input_491_cast_fp16)[name = tensor("op_36824_cast_fp16")]; + tensor var_36825_split_sizes_0 = const()[name = tensor("op_36825_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_36825_axis_0 = const()[name = tensor("op_36825_axis_0"), val = tensor(1)]; + tensor var_36825_cast_fp16_0, tensor var_36825_cast_fp16_1 = split(axis = var_36825_axis_0, split_sizes = var_36825_split_sizes_0, x = var_36824_cast_fp16)[name = tensor("op_36825_cast_fp16")]; + tensor var_36827_mode_0 = const()[name = tensor("op_36827_mode_0"), val = tensor("EXACT")]; + tensor var_36827_cast_fp16 = gelu(mode = var_36827_mode_0, x = var_36825_cast_fp16_1)[name = tensor("op_36827_cast_fp16")]; + tensor input_493_cast_fp16 = mul(x = var_36825_cast_fp16_0, y = var_36827_cast_fp16)[name = tensor("input_493_cast_fp16")]; + tensor var_36831 = const()[name = tensor("op_36831"), val = tensor([1, 1])]; + tensor var_36833 = const()[name = tensor("op_36833"), val = tensor([1, 1])]; + tensor var_36835_pad_type_0 = const()[name = tensor("op_36835_pad_type_0"), val = tensor("custom")]; + tensor var_36835_pad_0 = const()[name = tensor("op_36835_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_6_ff_net_2_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_6_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3061902208)))]; + tensor up_blocks_0_attentions_0_transformer_blocks_6_ff_net_2_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_6_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3075009472)))]; + tensor var_36835_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_6_ff_net_2_bias_to_fp16, dilations = var_36833, groups = var_30355, pad = var_36835_pad_0, pad_type = var_36835_pad_type_0, strides = var_36831, weight = up_blocks_0_attentions_0_transformer_blocks_6_ff_net_2_weight_to_fp16, x = input_493_cast_fp16)[name = tensor("op_36835_cast_fp16")]; + tensor inputs_247_cast_fp16 = add(x = var_36835_cast_fp16, y = inputs_245_cast_fp16)[name = tensor("inputs_247_cast_fp16")]; + tensor hidden_states_329_axes_0 = const()[name = tensor("hidden_states_329_axes_0"), val = tensor([1])]; + tensor hidden_states_329_gamma_0_to_fp16 = const()[name = tensor("hidden_states_329_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3075012096)))]; + tensor hidden_states_329_beta_0_to_fp16 = const()[name = tensor("hidden_states_329_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3075014720)))]; + tensor var_36851_to_fp16 = const()[name = tensor("op_36851_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_329_cast_fp16 = layer_norm(axes = hidden_states_329_axes_0, beta = hidden_states_329_beta_0_to_fp16, epsilon = var_36851_to_fp16, gamma = hidden_states_329_gamma_0_to_fp16, x = inputs_247_cast_fp16)[name = tensor("hidden_states_329_cast_fp16")]; + tensor var_36866 = const()[name = tensor("op_36866"), val = tensor([1, 1])]; + tensor var_36868 = const()[name = tensor("op_36868"), val = tensor([1, 1])]; + tensor q_165_pad_type_0 = const()[name = tensor("q_165_pad_type_0"), val = tensor("custom")]; + tensor q_165_pad_0 = const()[name = tensor("q_165_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_7_attn1_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_7_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3075017344)))]; + tensor q_165_cast_fp16 = conv(dilations = var_36868, groups = var_30355, pad = q_165_pad_0, pad_type = q_165_pad_type_0, strides = var_36866, weight = up_blocks_0_attentions_0_transformer_blocks_7_attn1_to_q_weight_to_fp16, x = hidden_states_329_cast_fp16)[name = tensor("q_165_cast_fp16")]; + tensor var_36872 = const()[name = tensor("op_36872"), val = tensor([1, 1])]; + tensor var_36874 = const()[name = tensor("op_36874"), val = tensor([1, 1])]; + tensor k_329_pad_type_0 = const()[name = tensor("k_329_pad_type_0"), val = tensor("custom")]; + tensor k_329_pad_0 = const()[name = tensor("k_329_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_7_attn1_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_7_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3078294208)))]; + tensor k_329_cast_fp16 = conv(dilations = var_36874, groups = var_30355, pad = k_329_pad_0, pad_type = k_329_pad_type_0, strides = var_36872, weight = up_blocks_0_attentions_0_transformer_blocks_7_attn1_to_k_weight_to_fp16, x = hidden_states_329_cast_fp16)[name = tensor("k_329_cast_fp16")]; + tensor var_36878 = const()[name = tensor("op_36878"), val = tensor([1, 1])]; + tensor var_36880 = const()[name = tensor("op_36880"), val = tensor([1, 1])]; + tensor v_165_pad_type_0 = const()[name = tensor("v_165_pad_type_0"), val = tensor("custom")]; + tensor v_165_pad_0 = const()[name = tensor("v_165_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_7_attn1_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_7_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3081571072)))]; + tensor v_165_cast_fp16 = conv(dilations = var_36880, groups = var_30355, pad = v_165_pad_0, pad_type = v_165_pad_type_0, strides = var_36878, weight = up_blocks_0_attentions_0_transformer_blocks_7_attn1_to_v_weight_to_fp16, x = hidden_states_329_cast_fp16)[name = tensor("v_165_cast_fp16")]; + tensor var_36884_begin_0 = const()[name = tensor("op_36884_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_36884_end_0 = const()[name = tensor("op_36884_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_36884_end_mask_0 = const()[name = tensor("op_36884_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36884_cast_fp16 = slice_by_index(begin = var_36884_begin_0, end = var_36884_end_0, end_mask = var_36884_end_mask_0, x = q_165_cast_fp16)[name = tensor("op_36884_cast_fp16")]; + tensor var_36888_begin_0 = const()[name = tensor("op_36888_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_36888_end_0 = const()[name = tensor("op_36888_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_36888_end_mask_0 = const()[name = tensor("op_36888_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36888_cast_fp16 = slice_by_index(begin = var_36888_begin_0, end = var_36888_end_0, end_mask = var_36888_end_mask_0, x = q_165_cast_fp16)[name = tensor("op_36888_cast_fp16")]; + tensor var_36892_begin_0 = const()[name = tensor("op_36892_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_36892_end_0 = const()[name = tensor("op_36892_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_36892_end_mask_0 = const()[name = tensor("op_36892_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36892_cast_fp16 = slice_by_index(begin = var_36892_begin_0, end = var_36892_end_0, end_mask = var_36892_end_mask_0, x = q_165_cast_fp16)[name = tensor("op_36892_cast_fp16")]; + tensor var_36896_begin_0 = const()[name = tensor("op_36896_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_36896_end_0 = const()[name = tensor("op_36896_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_36896_end_mask_0 = const()[name = tensor("op_36896_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36896_cast_fp16 = slice_by_index(begin = var_36896_begin_0, end = var_36896_end_0, end_mask = var_36896_end_mask_0, x = q_165_cast_fp16)[name = tensor("op_36896_cast_fp16")]; + tensor var_36900_begin_0 = const()[name = tensor("op_36900_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_36900_end_0 = const()[name = tensor("op_36900_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_36900_end_mask_0 = const()[name = tensor("op_36900_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36900_cast_fp16 = slice_by_index(begin = var_36900_begin_0, end = var_36900_end_0, end_mask = var_36900_end_mask_0, x = q_165_cast_fp16)[name = tensor("op_36900_cast_fp16")]; + tensor var_36904_begin_0 = const()[name = tensor("op_36904_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_36904_end_0 = const()[name = tensor("op_36904_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_36904_end_mask_0 = const()[name = tensor("op_36904_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36904_cast_fp16 = slice_by_index(begin = var_36904_begin_0, end = var_36904_end_0, end_mask = var_36904_end_mask_0, x = q_165_cast_fp16)[name = tensor("op_36904_cast_fp16")]; + tensor var_36908_begin_0 = const()[name = tensor("op_36908_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_36908_end_0 = const()[name = tensor("op_36908_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_36908_end_mask_0 = const()[name = tensor("op_36908_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36908_cast_fp16 = slice_by_index(begin = var_36908_begin_0, end = var_36908_end_0, end_mask = var_36908_end_mask_0, x = q_165_cast_fp16)[name = tensor("op_36908_cast_fp16")]; + tensor var_36912_begin_0 = const()[name = tensor("op_36912_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_36912_end_0 = const()[name = tensor("op_36912_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_36912_end_mask_0 = const()[name = tensor("op_36912_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36912_cast_fp16 = slice_by_index(begin = var_36912_begin_0, end = var_36912_end_0, end_mask = var_36912_end_mask_0, x = q_165_cast_fp16)[name = tensor("op_36912_cast_fp16")]; + tensor var_36916_begin_0 = const()[name = tensor("op_36916_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_36916_end_0 = const()[name = tensor("op_36916_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_36916_end_mask_0 = const()[name = tensor("op_36916_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36916_cast_fp16 = slice_by_index(begin = var_36916_begin_0, end = var_36916_end_0, end_mask = var_36916_end_mask_0, x = q_165_cast_fp16)[name = tensor("op_36916_cast_fp16")]; + tensor var_36920_begin_0 = const()[name = tensor("op_36920_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_36920_end_0 = const()[name = tensor("op_36920_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_36920_end_mask_0 = const()[name = tensor("op_36920_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36920_cast_fp16 = slice_by_index(begin = var_36920_begin_0, end = var_36920_end_0, end_mask = var_36920_end_mask_0, x = q_165_cast_fp16)[name = tensor("op_36920_cast_fp16")]; + tensor var_36924_begin_0 = const()[name = tensor("op_36924_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_36924_end_0 = const()[name = tensor("op_36924_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_36924_end_mask_0 = const()[name = tensor("op_36924_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36924_cast_fp16 = slice_by_index(begin = var_36924_begin_0, end = var_36924_end_0, end_mask = var_36924_end_mask_0, x = q_165_cast_fp16)[name = tensor("op_36924_cast_fp16")]; + tensor var_36928_begin_0 = const()[name = tensor("op_36928_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_36928_end_0 = const()[name = tensor("op_36928_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_36928_end_mask_0 = const()[name = tensor("op_36928_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36928_cast_fp16 = slice_by_index(begin = var_36928_begin_0, end = var_36928_end_0, end_mask = var_36928_end_mask_0, x = q_165_cast_fp16)[name = tensor("op_36928_cast_fp16")]; + tensor var_36932_begin_0 = const()[name = tensor("op_36932_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_36932_end_0 = const()[name = tensor("op_36932_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_36932_end_mask_0 = const()[name = tensor("op_36932_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36932_cast_fp16 = slice_by_index(begin = var_36932_begin_0, end = var_36932_end_0, end_mask = var_36932_end_mask_0, x = q_165_cast_fp16)[name = tensor("op_36932_cast_fp16")]; + tensor var_36936_begin_0 = const()[name = tensor("op_36936_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_36936_end_0 = const()[name = tensor("op_36936_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_36936_end_mask_0 = const()[name = tensor("op_36936_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36936_cast_fp16 = slice_by_index(begin = var_36936_begin_0, end = var_36936_end_0, end_mask = var_36936_end_mask_0, x = q_165_cast_fp16)[name = tensor("op_36936_cast_fp16")]; + tensor var_36940_begin_0 = const()[name = tensor("op_36940_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_36940_end_0 = const()[name = tensor("op_36940_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_36940_end_mask_0 = const()[name = tensor("op_36940_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36940_cast_fp16 = slice_by_index(begin = var_36940_begin_0, end = var_36940_end_0, end_mask = var_36940_end_mask_0, x = q_165_cast_fp16)[name = tensor("op_36940_cast_fp16")]; + tensor var_36944_begin_0 = const()[name = tensor("op_36944_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_36944_end_0 = const()[name = tensor("op_36944_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_36944_end_mask_0 = const()[name = tensor("op_36944_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36944_cast_fp16 = slice_by_index(begin = var_36944_begin_0, end = var_36944_end_0, end_mask = var_36944_end_mask_0, x = q_165_cast_fp16)[name = tensor("op_36944_cast_fp16")]; + tensor var_36948_begin_0 = const()[name = tensor("op_36948_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_36948_end_0 = const()[name = tensor("op_36948_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_36948_end_mask_0 = const()[name = tensor("op_36948_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36948_cast_fp16 = slice_by_index(begin = var_36948_begin_0, end = var_36948_end_0, end_mask = var_36948_end_mask_0, x = q_165_cast_fp16)[name = tensor("op_36948_cast_fp16")]; + tensor var_36952_begin_0 = const()[name = tensor("op_36952_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_36952_end_0 = const()[name = tensor("op_36952_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_36952_end_mask_0 = const()[name = tensor("op_36952_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36952_cast_fp16 = slice_by_index(begin = var_36952_begin_0, end = var_36952_end_0, end_mask = var_36952_end_mask_0, x = q_165_cast_fp16)[name = tensor("op_36952_cast_fp16")]; + tensor var_36956_begin_0 = const()[name = tensor("op_36956_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_36956_end_0 = const()[name = tensor("op_36956_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_36956_end_mask_0 = const()[name = tensor("op_36956_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36956_cast_fp16 = slice_by_index(begin = var_36956_begin_0, end = var_36956_end_0, end_mask = var_36956_end_mask_0, x = q_165_cast_fp16)[name = tensor("op_36956_cast_fp16")]; + tensor var_36960_begin_0 = const()[name = tensor("op_36960_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_36960_end_0 = const()[name = tensor("op_36960_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_36960_end_mask_0 = const()[name = tensor("op_36960_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36960_cast_fp16 = slice_by_index(begin = var_36960_begin_0, end = var_36960_end_0, end_mask = var_36960_end_mask_0, x = q_165_cast_fp16)[name = tensor("op_36960_cast_fp16")]; + tensor k_331_perm_0 = const()[name = tensor("k_331_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_36967_begin_0 = const()[name = tensor("op_36967_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_36967_end_0 = const()[name = tensor("op_36967_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_36967_end_mask_0 = const()[name = tensor("op_36967_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_331_cast_fp16 = transpose(perm = k_331_perm_0, x = k_329_cast_fp16)[name = tensor("transpose_57")]; + tensor var_36967_cast_fp16 = slice_by_index(begin = var_36967_begin_0, end = var_36967_end_0, end_mask = var_36967_end_mask_0, x = k_331_cast_fp16)[name = tensor("op_36967_cast_fp16")]; + tensor var_36971_begin_0 = const()[name = tensor("op_36971_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_36971_end_0 = const()[name = tensor("op_36971_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_36971_end_mask_0 = const()[name = tensor("op_36971_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36971_cast_fp16 = slice_by_index(begin = var_36971_begin_0, end = var_36971_end_0, end_mask = var_36971_end_mask_0, x = k_331_cast_fp16)[name = tensor("op_36971_cast_fp16")]; + tensor var_36975_begin_0 = const()[name = tensor("op_36975_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_36975_end_0 = const()[name = tensor("op_36975_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_36975_end_mask_0 = const()[name = tensor("op_36975_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36975_cast_fp16 = slice_by_index(begin = var_36975_begin_0, end = var_36975_end_0, end_mask = var_36975_end_mask_0, x = k_331_cast_fp16)[name = tensor("op_36975_cast_fp16")]; + tensor var_36979_begin_0 = const()[name = tensor("op_36979_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_36979_end_0 = const()[name = tensor("op_36979_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_36979_end_mask_0 = const()[name = tensor("op_36979_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36979_cast_fp16 = slice_by_index(begin = var_36979_begin_0, end = var_36979_end_0, end_mask = var_36979_end_mask_0, x = k_331_cast_fp16)[name = tensor("op_36979_cast_fp16")]; + tensor var_36983_begin_0 = const()[name = tensor("op_36983_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_36983_end_0 = const()[name = tensor("op_36983_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_36983_end_mask_0 = const()[name = tensor("op_36983_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36983_cast_fp16 = slice_by_index(begin = var_36983_begin_0, end = var_36983_end_0, end_mask = var_36983_end_mask_0, x = k_331_cast_fp16)[name = tensor("op_36983_cast_fp16")]; + tensor var_36987_begin_0 = const()[name = tensor("op_36987_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_36987_end_0 = const()[name = tensor("op_36987_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_36987_end_mask_0 = const()[name = tensor("op_36987_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36987_cast_fp16 = slice_by_index(begin = var_36987_begin_0, end = var_36987_end_0, end_mask = var_36987_end_mask_0, x = k_331_cast_fp16)[name = tensor("op_36987_cast_fp16")]; + tensor var_36991_begin_0 = const()[name = tensor("op_36991_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_36991_end_0 = const()[name = tensor("op_36991_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_36991_end_mask_0 = const()[name = tensor("op_36991_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36991_cast_fp16 = slice_by_index(begin = var_36991_begin_0, end = var_36991_end_0, end_mask = var_36991_end_mask_0, x = k_331_cast_fp16)[name = tensor("op_36991_cast_fp16")]; + tensor var_36995_begin_0 = const()[name = tensor("op_36995_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_36995_end_0 = const()[name = tensor("op_36995_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_36995_end_mask_0 = const()[name = tensor("op_36995_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36995_cast_fp16 = slice_by_index(begin = var_36995_begin_0, end = var_36995_end_0, end_mask = var_36995_end_mask_0, x = k_331_cast_fp16)[name = tensor("op_36995_cast_fp16")]; + tensor var_36999_begin_0 = const()[name = tensor("op_36999_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_36999_end_0 = const()[name = tensor("op_36999_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_36999_end_mask_0 = const()[name = tensor("op_36999_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36999_cast_fp16 = slice_by_index(begin = var_36999_begin_0, end = var_36999_end_0, end_mask = var_36999_end_mask_0, x = k_331_cast_fp16)[name = tensor("op_36999_cast_fp16")]; + tensor var_37003_begin_0 = const()[name = tensor("op_37003_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_37003_end_0 = const()[name = tensor("op_37003_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_37003_end_mask_0 = const()[name = tensor("op_37003_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37003_cast_fp16 = slice_by_index(begin = var_37003_begin_0, end = var_37003_end_0, end_mask = var_37003_end_mask_0, x = k_331_cast_fp16)[name = tensor("op_37003_cast_fp16")]; + tensor var_37007_begin_0 = const()[name = tensor("op_37007_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_37007_end_0 = const()[name = tensor("op_37007_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_37007_end_mask_0 = const()[name = tensor("op_37007_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37007_cast_fp16 = slice_by_index(begin = var_37007_begin_0, end = var_37007_end_0, end_mask = var_37007_end_mask_0, x = k_331_cast_fp16)[name = tensor("op_37007_cast_fp16")]; + tensor var_37011_begin_0 = const()[name = tensor("op_37011_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_37011_end_0 = const()[name = tensor("op_37011_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_37011_end_mask_0 = const()[name = tensor("op_37011_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37011_cast_fp16 = slice_by_index(begin = var_37011_begin_0, end = var_37011_end_0, end_mask = var_37011_end_mask_0, x = k_331_cast_fp16)[name = tensor("op_37011_cast_fp16")]; + tensor var_37015_begin_0 = const()[name = tensor("op_37015_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_37015_end_0 = const()[name = tensor("op_37015_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_37015_end_mask_0 = const()[name = tensor("op_37015_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37015_cast_fp16 = slice_by_index(begin = var_37015_begin_0, end = var_37015_end_0, end_mask = var_37015_end_mask_0, x = k_331_cast_fp16)[name = tensor("op_37015_cast_fp16")]; + tensor var_37019_begin_0 = const()[name = tensor("op_37019_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_37019_end_0 = const()[name = tensor("op_37019_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_37019_end_mask_0 = const()[name = tensor("op_37019_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37019_cast_fp16 = slice_by_index(begin = var_37019_begin_0, end = var_37019_end_0, end_mask = var_37019_end_mask_0, x = k_331_cast_fp16)[name = tensor("op_37019_cast_fp16")]; + tensor var_37023_begin_0 = const()[name = tensor("op_37023_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_37023_end_0 = const()[name = tensor("op_37023_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_37023_end_mask_0 = const()[name = tensor("op_37023_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37023_cast_fp16 = slice_by_index(begin = var_37023_begin_0, end = var_37023_end_0, end_mask = var_37023_end_mask_0, x = k_331_cast_fp16)[name = tensor("op_37023_cast_fp16")]; + tensor var_37027_begin_0 = const()[name = tensor("op_37027_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_37027_end_0 = const()[name = tensor("op_37027_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_37027_end_mask_0 = const()[name = tensor("op_37027_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37027_cast_fp16 = slice_by_index(begin = var_37027_begin_0, end = var_37027_end_0, end_mask = var_37027_end_mask_0, x = k_331_cast_fp16)[name = tensor("op_37027_cast_fp16")]; + tensor var_37031_begin_0 = const()[name = tensor("op_37031_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_37031_end_0 = const()[name = tensor("op_37031_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_37031_end_mask_0 = const()[name = tensor("op_37031_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37031_cast_fp16 = slice_by_index(begin = var_37031_begin_0, end = var_37031_end_0, end_mask = var_37031_end_mask_0, x = k_331_cast_fp16)[name = tensor("op_37031_cast_fp16")]; + tensor var_37035_begin_0 = const()[name = tensor("op_37035_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_37035_end_0 = const()[name = tensor("op_37035_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_37035_end_mask_0 = const()[name = tensor("op_37035_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37035_cast_fp16 = slice_by_index(begin = var_37035_begin_0, end = var_37035_end_0, end_mask = var_37035_end_mask_0, x = k_331_cast_fp16)[name = tensor("op_37035_cast_fp16")]; + tensor var_37039_begin_0 = const()[name = tensor("op_37039_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_37039_end_0 = const()[name = tensor("op_37039_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_37039_end_mask_0 = const()[name = tensor("op_37039_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37039_cast_fp16 = slice_by_index(begin = var_37039_begin_0, end = var_37039_end_0, end_mask = var_37039_end_mask_0, x = k_331_cast_fp16)[name = tensor("op_37039_cast_fp16")]; + tensor var_37043_begin_0 = const()[name = tensor("op_37043_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_37043_end_0 = const()[name = tensor("op_37043_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_37043_end_mask_0 = const()[name = tensor("op_37043_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37043_cast_fp16 = slice_by_index(begin = var_37043_begin_0, end = var_37043_end_0, end_mask = var_37043_end_mask_0, x = k_331_cast_fp16)[name = tensor("op_37043_cast_fp16")]; + tensor var_37045_begin_0 = const()[name = tensor("op_37045_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_37045_end_0 = const()[name = tensor("op_37045_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_37045_end_mask_0 = const()[name = tensor("op_37045_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37045_cast_fp16 = slice_by_index(begin = var_37045_begin_0, end = var_37045_end_0, end_mask = var_37045_end_mask_0, x = v_165_cast_fp16)[name = tensor("op_37045_cast_fp16")]; + tensor var_37049_begin_0 = const()[name = tensor("op_37049_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_37049_end_0 = const()[name = tensor("op_37049_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_37049_end_mask_0 = const()[name = tensor("op_37049_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37049_cast_fp16 = slice_by_index(begin = var_37049_begin_0, end = var_37049_end_0, end_mask = var_37049_end_mask_0, x = v_165_cast_fp16)[name = tensor("op_37049_cast_fp16")]; + tensor var_37053_begin_0 = const()[name = tensor("op_37053_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_37053_end_0 = const()[name = tensor("op_37053_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_37053_end_mask_0 = const()[name = tensor("op_37053_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37053_cast_fp16 = slice_by_index(begin = var_37053_begin_0, end = var_37053_end_0, end_mask = var_37053_end_mask_0, x = v_165_cast_fp16)[name = tensor("op_37053_cast_fp16")]; + tensor var_37057_begin_0 = const()[name = tensor("op_37057_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_37057_end_0 = const()[name = tensor("op_37057_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_37057_end_mask_0 = const()[name = tensor("op_37057_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37057_cast_fp16 = slice_by_index(begin = var_37057_begin_0, end = var_37057_end_0, end_mask = var_37057_end_mask_0, x = v_165_cast_fp16)[name = tensor("op_37057_cast_fp16")]; + tensor var_37061_begin_0 = const()[name = tensor("op_37061_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_37061_end_0 = const()[name = tensor("op_37061_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_37061_end_mask_0 = const()[name = tensor("op_37061_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37061_cast_fp16 = slice_by_index(begin = var_37061_begin_0, end = var_37061_end_0, end_mask = var_37061_end_mask_0, x = v_165_cast_fp16)[name = tensor("op_37061_cast_fp16")]; + tensor var_37065_begin_0 = const()[name = tensor("op_37065_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_37065_end_0 = const()[name = tensor("op_37065_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_37065_end_mask_0 = const()[name = tensor("op_37065_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37065_cast_fp16 = slice_by_index(begin = var_37065_begin_0, end = var_37065_end_0, end_mask = var_37065_end_mask_0, x = v_165_cast_fp16)[name = tensor("op_37065_cast_fp16")]; + tensor var_37069_begin_0 = const()[name = tensor("op_37069_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_37069_end_0 = const()[name = tensor("op_37069_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_37069_end_mask_0 = const()[name = tensor("op_37069_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37069_cast_fp16 = slice_by_index(begin = var_37069_begin_0, end = var_37069_end_0, end_mask = var_37069_end_mask_0, x = v_165_cast_fp16)[name = tensor("op_37069_cast_fp16")]; + tensor var_37073_begin_0 = const()[name = tensor("op_37073_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_37073_end_0 = const()[name = tensor("op_37073_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_37073_end_mask_0 = const()[name = tensor("op_37073_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37073_cast_fp16 = slice_by_index(begin = var_37073_begin_0, end = var_37073_end_0, end_mask = var_37073_end_mask_0, x = v_165_cast_fp16)[name = tensor("op_37073_cast_fp16")]; + tensor var_37077_begin_0 = const()[name = tensor("op_37077_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_37077_end_0 = const()[name = tensor("op_37077_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_37077_end_mask_0 = const()[name = tensor("op_37077_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37077_cast_fp16 = slice_by_index(begin = var_37077_begin_0, end = var_37077_end_0, end_mask = var_37077_end_mask_0, x = v_165_cast_fp16)[name = tensor("op_37077_cast_fp16")]; + tensor var_37081_begin_0 = const()[name = tensor("op_37081_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_37081_end_0 = const()[name = tensor("op_37081_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_37081_end_mask_0 = const()[name = tensor("op_37081_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37081_cast_fp16 = slice_by_index(begin = var_37081_begin_0, end = var_37081_end_0, end_mask = var_37081_end_mask_0, x = v_165_cast_fp16)[name = tensor("op_37081_cast_fp16")]; + tensor var_37085_begin_0 = const()[name = tensor("op_37085_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_37085_end_0 = const()[name = tensor("op_37085_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_37085_end_mask_0 = const()[name = tensor("op_37085_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37085_cast_fp16 = slice_by_index(begin = var_37085_begin_0, end = var_37085_end_0, end_mask = var_37085_end_mask_0, x = v_165_cast_fp16)[name = tensor("op_37085_cast_fp16")]; + tensor var_37089_begin_0 = const()[name = tensor("op_37089_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_37089_end_0 = const()[name = tensor("op_37089_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_37089_end_mask_0 = const()[name = tensor("op_37089_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37089_cast_fp16 = slice_by_index(begin = var_37089_begin_0, end = var_37089_end_0, end_mask = var_37089_end_mask_0, x = v_165_cast_fp16)[name = tensor("op_37089_cast_fp16")]; + tensor var_37093_begin_0 = const()[name = tensor("op_37093_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_37093_end_0 = const()[name = tensor("op_37093_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_37093_end_mask_0 = const()[name = tensor("op_37093_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37093_cast_fp16 = slice_by_index(begin = var_37093_begin_0, end = var_37093_end_0, end_mask = var_37093_end_mask_0, x = v_165_cast_fp16)[name = tensor("op_37093_cast_fp16")]; + tensor var_37097_begin_0 = const()[name = tensor("op_37097_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_37097_end_0 = const()[name = tensor("op_37097_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_37097_end_mask_0 = const()[name = tensor("op_37097_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37097_cast_fp16 = slice_by_index(begin = var_37097_begin_0, end = var_37097_end_0, end_mask = var_37097_end_mask_0, x = v_165_cast_fp16)[name = tensor("op_37097_cast_fp16")]; + tensor var_37101_begin_0 = const()[name = tensor("op_37101_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_37101_end_0 = const()[name = tensor("op_37101_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_37101_end_mask_0 = const()[name = tensor("op_37101_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37101_cast_fp16 = slice_by_index(begin = var_37101_begin_0, end = var_37101_end_0, end_mask = var_37101_end_mask_0, x = v_165_cast_fp16)[name = tensor("op_37101_cast_fp16")]; + tensor var_37105_begin_0 = const()[name = tensor("op_37105_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_37105_end_0 = const()[name = tensor("op_37105_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_37105_end_mask_0 = const()[name = tensor("op_37105_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37105_cast_fp16 = slice_by_index(begin = var_37105_begin_0, end = var_37105_end_0, end_mask = var_37105_end_mask_0, x = v_165_cast_fp16)[name = tensor("op_37105_cast_fp16")]; + tensor var_37109_begin_0 = const()[name = tensor("op_37109_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_37109_end_0 = const()[name = tensor("op_37109_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_37109_end_mask_0 = const()[name = tensor("op_37109_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37109_cast_fp16 = slice_by_index(begin = var_37109_begin_0, end = var_37109_end_0, end_mask = var_37109_end_mask_0, x = v_165_cast_fp16)[name = tensor("op_37109_cast_fp16")]; + tensor var_37113_begin_0 = const()[name = tensor("op_37113_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_37113_end_0 = const()[name = tensor("op_37113_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_37113_end_mask_0 = const()[name = tensor("op_37113_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37113_cast_fp16 = slice_by_index(begin = var_37113_begin_0, end = var_37113_end_0, end_mask = var_37113_end_mask_0, x = v_165_cast_fp16)[name = tensor("op_37113_cast_fp16")]; + tensor var_37117_begin_0 = const()[name = tensor("op_37117_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_37117_end_0 = const()[name = tensor("op_37117_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_37117_end_mask_0 = const()[name = tensor("op_37117_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37117_cast_fp16 = slice_by_index(begin = var_37117_begin_0, end = var_37117_end_0, end_mask = var_37117_end_mask_0, x = v_165_cast_fp16)[name = tensor("op_37117_cast_fp16")]; + tensor var_37121_begin_0 = const()[name = tensor("op_37121_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_37121_end_0 = const()[name = tensor("op_37121_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_37121_end_mask_0 = const()[name = tensor("op_37121_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37121_cast_fp16 = slice_by_index(begin = var_37121_begin_0, end = var_37121_end_0, end_mask = var_37121_end_mask_0, x = v_165_cast_fp16)[name = tensor("op_37121_cast_fp16")]; + tensor var_37125_equation_0 = const()[name = tensor("op_37125_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37125_cast_fp16 = einsum(equation = var_37125_equation_0, values = (var_36967_cast_fp16, var_36884_cast_fp16))[name = tensor("op_37125_cast_fp16")]; + tensor var_37126_to_fp16 = const()[name = tensor("op_37126_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3121_cast_fp16 = mul(x = var_37125_cast_fp16, y = var_37126_to_fp16)[name = tensor("aw_3121_cast_fp16")]; + tensor var_37129_equation_0 = const()[name = tensor("op_37129_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37129_cast_fp16 = einsum(equation = var_37129_equation_0, values = (var_36971_cast_fp16, var_36888_cast_fp16))[name = tensor("op_37129_cast_fp16")]; + tensor var_37130_to_fp16 = const()[name = tensor("op_37130_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3123_cast_fp16 = mul(x = var_37129_cast_fp16, y = var_37130_to_fp16)[name = tensor("aw_3123_cast_fp16")]; + tensor var_37133_equation_0 = const()[name = tensor("op_37133_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37133_cast_fp16 = einsum(equation = var_37133_equation_0, values = (var_36975_cast_fp16, var_36892_cast_fp16))[name = tensor("op_37133_cast_fp16")]; + tensor var_37134_to_fp16 = const()[name = tensor("op_37134_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3125_cast_fp16 = mul(x = var_37133_cast_fp16, y = var_37134_to_fp16)[name = tensor("aw_3125_cast_fp16")]; + tensor var_37137_equation_0 = const()[name = tensor("op_37137_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37137_cast_fp16 = einsum(equation = var_37137_equation_0, values = (var_36979_cast_fp16, var_36896_cast_fp16))[name = tensor("op_37137_cast_fp16")]; + tensor var_37138_to_fp16 = const()[name = tensor("op_37138_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3127_cast_fp16 = mul(x = var_37137_cast_fp16, y = var_37138_to_fp16)[name = tensor("aw_3127_cast_fp16")]; + tensor var_37141_equation_0 = const()[name = tensor("op_37141_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37141_cast_fp16 = einsum(equation = var_37141_equation_0, values = (var_36983_cast_fp16, var_36900_cast_fp16))[name = tensor("op_37141_cast_fp16")]; + tensor var_37142_to_fp16 = const()[name = tensor("op_37142_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3129_cast_fp16 = mul(x = var_37141_cast_fp16, y = var_37142_to_fp16)[name = tensor("aw_3129_cast_fp16")]; + tensor var_37145_equation_0 = const()[name = tensor("op_37145_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37145_cast_fp16 = einsum(equation = var_37145_equation_0, values = (var_36987_cast_fp16, var_36904_cast_fp16))[name = tensor("op_37145_cast_fp16")]; + tensor var_37146_to_fp16 = const()[name = tensor("op_37146_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3131_cast_fp16 = mul(x = var_37145_cast_fp16, y = var_37146_to_fp16)[name = tensor("aw_3131_cast_fp16")]; + tensor var_37149_equation_0 = const()[name = tensor("op_37149_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37149_cast_fp16 = einsum(equation = var_37149_equation_0, values = (var_36991_cast_fp16, var_36908_cast_fp16))[name = tensor("op_37149_cast_fp16")]; + tensor var_37150_to_fp16 = const()[name = tensor("op_37150_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3133_cast_fp16 = mul(x = var_37149_cast_fp16, y = var_37150_to_fp16)[name = tensor("aw_3133_cast_fp16")]; + tensor var_37153_equation_0 = const()[name = tensor("op_37153_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37153_cast_fp16 = einsum(equation = var_37153_equation_0, values = (var_36995_cast_fp16, var_36912_cast_fp16))[name = tensor("op_37153_cast_fp16")]; + tensor var_37154_to_fp16 = const()[name = tensor("op_37154_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3135_cast_fp16 = mul(x = var_37153_cast_fp16, y = var_37154_to_fp16)[name = tensor("aw_3135_cast_fp16")]; + tensor var_37157_equation_0 = const()[name = tensor("op_37157_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37157_cast_fp16 = einsum(equation = var_37157_equation_0, values = (var_36999_cast_fp16, var_36916_cast_fp16))[name = tensor("op_37157_cast_fp16")]; + tensor var_37158_to_fp16 = const()[name = tensor("op_37158_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3137_cast_fp16 = mul(x = var_37157_cast_fp16, y = var_37158_to_fp16)[name = tensor("aw_3137_cast_fp16")]; + tensor var_37161_equation_0 = const()[name = tensor("op_37161_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37161_cast_fp16 = einsum(equation = var_37161_equation_0, values = (var_37003_cast_fp16, var_36920_cast_fp16))[name = tensor("op_37161_cast_fp16")]; + tensor var_37162_to_fp16 = const()[name = tensor("op_37162_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3139_cast_fp16 = mul(x = var_37161_cast_fp16, y = var_37162_to_fp16)[name = tensor("aw_3139_cast_fp16")]; + tensor var_37165_equation_0 = const()[name = tensor("op_37165_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37165_cast_fp16 = einsum(equation = var_37165_equation_0, values = (var_37007_cast_fp16, var_36924_cast_fp16))[name = tensor("op_37165_cast_fp16")]; + tensor var_37166_to_fp16 = const()[name = tensor("op_37166_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3141_cast_fp16 = mul(x = var_37165_cast_fp16, y = var_37166_to_fp16)[name = tensor("aw_3141_cast_fp16")]; + tensor var_37169_equation_0 = const()[name = tensor("op_37169_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37169_cast_fp16 = einsum(equation = var_37169_equation_0, values = (var_37011_cast_fp16, var_36928_cast_fp16))[name = tensor("op_37169_cast_fp16")]; + tensor var_37170_to_fp16 = const()[name = tensor("op_37170_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3143_cast_fp16 = mul(x = var_37169_cast_fp16, y = var_37170_to_fp16)[name = tensor("aw_3143_cast_fp16")]; + tensor var_37173_equation_0 = const()[name = tensor("op_37173_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37173_cast_fp16 = einsum(equation = var_37173_equation_0, values = (var_37015_cast_fp16, var_36932_cast_fp16))[name = tensor("op_37173_cast_fp16")]; + tensor var_37174_to_fp16 = const()[name = tensor("op_37174_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3145_cast_fp16 = mul(x = var_37173_cast_fp16, y = var_37174_to_fp16)[name = tensor("aw_3145_cast_fp16")]; + tensor var_37177_equation_0 = const()[name = tensor("op_37177_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37177_cast_fp16 = einsum(equation = var_37177_equation_0, values = (var_37019_cast_fp16, var_36936_cast_fp16))[name = tensor("op_37177_cast_fp16")]; + tensor var_37178_to_fp16 = const()[name = tensor("op_37178_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3147_cast_fp16 = mul(x = var_37177_cast_fp16, y = var_37178_to_fp16)[name = tensor("aw_3147_cast_fp16")]; + tensor var_37181_equation_0 = const()[name = tensor("op_37181_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37181_cast_fp16 = einsum(equation = var_37181_equation_0, values = (var_37023_cast_fp16, var_36940_cast_fp16))[name = tensor("op_37181_cast_fp16")]; + tensor var_37182_to_fp16 = const()[name = tensor("op_37182_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3149_cast_fp16 = mul(x = var_37181_cast_fp16, y = var_37182_to_fp16)[name = tensor("aw_3149_cast_fp16")]; + tensor var_37185_equation_0 = const()[name = tensor("op_37185_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37185_cast_fp16 = einsum(equation = var_37185_equation_0, values = (var_37027_cast_fp16, var_36944_cast_fp16))[name = tensor("op_37185_cast_fp16")]; + tensor var_37186_to_fp16 = const()[name = tensor("op_37186_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3151_cast_fp16 = mul(x = var_37185_cast_fp16, y = var_37186_to_fp16)[name = tensor("aw_3151_cast_fp16")]; + tensor var_37189_equation_0 = const()[name = tensor("op_37189_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37189_cast_fp16 = einsum(equation = var_37189_equation_0, values = (var_37031_cast_fp16, var_36948_cast_fp16))[name = tensor("op_37189_cast_fp16")]; + tensor var_37190_to_fp16 = const()[name = tensor("op_37190_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3153_cast_fp16 = mul(x = var_37189_cast_fp16, y = var_37190_to_fp16)[name = tensor("aw_3153_cast_fp16")]; + tensor var_37193_equation_0 = const()[name = tensor("op_37193_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37193_cast_fp16 = einsum(equation = var_37193_equation_0, values = (var_37035_cast_fp16, var_36952_cast_fp16))[name = tensor("op_37193_cast_fp16")]; + tensor var_37194_to_fp16 = const()[name = tensor("op_37194_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3155_cast_fp16 = mul(x = var_37193_cast_fp16, y = var_37194_to_fp16)[name = tensor("aw_3155_cast_fp16")]; + tensor var_37197_equation_0 = const()[name = tensor("op_37197_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37197_cast_fp16 = einsum(equation = var_37197_equation_0, values = (var_37039_cast_fp16, var_36956_cast_fp16))[name = tensor("op_37197_cast_fp16")]; + tensor var_37198_to_fp16 = const()[name = tensor("op_37198_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3157_cast_fp16 = mul(x = var_37197_cast_fp16, y = var_37198_to_fp16)[name = tensor("aw_3157_cast_fp16")]; + tensor var_37201_equation_0 = const()[name = tensor("op_37201_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37201_cast_fp16 = einsum(equation = var_37201_equation_0, values = (var_37043_cast_fp16, var_36960_cast_fp16))[name = tensor("op_37201_cast_fp16")]; + tensor var_37202_to_fp16 = const()[name = tensor("op_37202_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3159_cast_fp16 = mul(x = var_37201_cast_fp16, y = var_37202_to_fp16)[name = tensor("aw_3159_cast_fp16")]; + tensor var_37204_cast_fp16 = softmax(axis = var_30355, x = aw_3121_cast_fp16)[name = tensor("op_37204_cast_fp16")]; + tensor var_37205_cast_fp16 = softmax(axis = var_30355, x = aw_3123_cast_fp16)[name = tensor("op_37205_cast_fp16")]; + tensor var_37206_cast_fp16 = softmax(axis = var_30355, x = aw_3125_cast_fp16)[name = tensor("op_37206_cast_fp16")]; + tensor var_37207_cast_fp16 = softmax(axis = var_30355, x = aw_3127_cast_fp16)[name = tensor("op_37207_cast_fp16")]; + tensor var_37208_cast_fp16 = softmax(axis = var_30355, x = aw_3129_cast_fp16)[name = tensor("op_37208_cast_fp16")]; + tensor var_37209_cast_fp16 = softmax(axis = var_30355, x = aw_3131_cast_fp16)[name = tensor("op_37209_cast_fp16")]; + tensor var_37210_cast_fp16 = softmax(axis = var_30355, x = aw_3133_cast_fp16)[name = tensor("op_37210_cast_fp16")]; + tensor var_37211_cast_fp16 = softmax(axis = var_30355, x = aw_3135_cast_fp16)[name = tensor("op_37211_cast_fp16")]; + tensor var_37212_cast_fp16 = softmax(axis = var_30355, x = aw_3137_cast_fp16)[name = tensor("op_37212_cast_fp16")]; + tensor var_37213_cast_fp16 = softmax(axis = var_30355, x = aw_3139_cast_fp16)[name = tensor("op_37213_cast_fp16")]; + tensor var_37214_cast_fp16 = softmax(axis = var_30355, x = aw_3141_cast_fp16)[name = tensor("op_37214_cast_fp16")]; + tensor var_37215_cast_fp16 = softmax(axis = var_30355, x = aw_3143_cast_fp16)[name = tensor("op_37215_cast_fp16")]; + tensor var_37216_cast_fp16 = softmax(axis = var_30355, x = aw_3145_cast_fp16)[name = tensor("op_37216_cast_fp16")]; + tensor var_37217_cast_fp16 = softmax(axis = var_30355, x = aw_3147_cast_fp16)[name = tensor("op_37217_cast_fp16")]; + tensor var_37218_cast_fp16 = softmax(axis = var_30355, x = aw_3149_cast_fp16)[name = tensor("op_37218_cast_fp16")]; + tensor var_37219_cast_fp16 = softmax(axis = var_30355, x = aw_3151_cast_fp16)[name = tensor("op_37219_cast_fp16")]; + tensor var_37220_cast_fp16 = softmax(axis = var_30355, x = aw_3153_cast_fp16)[name = tensor("op_37220_cast_fp16")]; + tensor var_37221_cast_fp16 = softmax(axis = var_30355, x = aw_3155_cast_fp16)[name = tensor("op_37221_cast_fp16")]; + tensor var_37222_cast_fp16 = softmax(axis = var_30355, x = aw_3157_cast_fp16)[name = tensor("op_37222_cast_fp16")]; + tensor var_37223_cast_fp16 = softmax(axis = var_30355, x = aw_3159_cast_fp16)[name = tensor("op_37223_cast_fp16")]; + tensor var_37225_equation_0 = const()[name = tensor("op_37225_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37225_cast_fp16 = einsum(equation = var_37225_equation_0, values = (var_37045_cast_fp16, var_37204_cast_fp16))[name = tensor("op_37225_cast_fp16")]; + tensor var_37227_equation_0 = const()[name = tensor("op_37227_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37227_cast_fp16 = einsum(equation = var_37227_equation_0, values = (var_37049_cast_fp16, var_37205_cast_fp16))[name = tensor("op_37227_cast_fp16")]; + tensor var_37229_equation_0 = const()[name = tensor("op_37229_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37229_cast_fp16 = einsum(equation = var_37229_equation_0, values = (var_37053_cast_fp16, var_37206_cast_fp16))[name = tensor("op_37229_cast_fp16")]; + tensor var_37231_equation_0 = const()[name = tensor("op_37231_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37231_cast_fp16 = einsum(equation = var_37231_equation_0, values = (var_37057_cast_fp16, var_37207_cast_fp16))[name = tensor("op_37231_cast_fp16")]; + tensor var_37233_equation_0 = const()[name = tensor("op_37233_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37233_cast_fp16 = einsum(equation = var_37233_equation_0, values = (var_37061_cast_fp16, var_37208_cast_fp16))[name = tensor("op_37233_cast_fp16")]; + tensor var_37235_equation_0 = const()[name = tensor("op_37235_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37235_cast_fp16 = einsum(equation = var_37235_equation_0, values = (var_37065_cast_fp16, var_37209_cast_fp16))[name = tensor("op_37235_cast_fp16")]; + tensor var_37237_equation_0 = const()[name = tensor("op_37237_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37237_cast_fp16 = einsum(equation = var_37237_equation_0, values = (var_37069_cast_fp16, var_37210_cast_fp16))[name = tensor("op_37237_cast_fp16")]; + tensor var_37239_equation_0 = const()[name = tensor("op_37239_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37239_cast_fp16 = einsum(equation = var_37239_equation_0, values = (var_37073_cast_fp16, var_37211_cast_fp16))[name = tensor("op_37239_cast_fp16")]; + tensor var_37241_equation_0 = const()[name = tensor("op_37241_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37241_cast_fp16 = einsum(equation = var_37241_equation_0, values = (var_37077_cast_fp16, var_37212_cast_fp16))[name = tensor("op_37241_cast_fp16")]; + tensor var_37243_equation_0 = const()[name = tensor("op_37243_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37243_cast_fp16 = einsum(equation = var_37243_equation_0, values = (var_37081_cast_fp16, var_37213_cast_fp16))[name = tensor("op_37243_cast_fp16")]; + tensor var_37245_equation_0 = const()[name = tensor("op_37245_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37245_cast_fp16 = einsum(equation = var_37245_equation_0, values = (var_37085_cast_fp16, var_37214_cast_fp16))[name = tensor("op_37245_cast_fp16")]; + tensor var_37247_equation_0 = const()[name = tensor("op_37247_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37247_cast_fp16 = einsum(equation = var_37247_equation_0, values = (var_37089_cast_fp16, var_37215_cast_fp16))[name = tensor("op_37247_cast_fp16")]; + tensor var_37249_equation_0 = const()[name = tensor("op_37249_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37249_cast_fp16 = einsum(equation = var_37249_equation_0, values = (var_37093_cast_fp16, var_37216_cast_fp16))[name = tensor("op_37249_cast_fp16")]; + tensor var_37251_equation_0 = const()[name = tensor("op_37251_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37251_cast_fp16 = einsum(equation = var_37251_equation_0, values = (var_37097_cast_fp16, var_37217_cast_fp16))[name = tensor("op_37251_cast_fp16")]; + tensor var_37253_equation_0 = const()[name = tensor("op_37253_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37253_cast_fp16 = einsum(equation = var_37253_equation_0, values = (var_37101_cast_fp16, var_37218_cast_fp16))[name = tensor("op_37253_cast_fp16")]; + tensor var_37255_equation_0 = const()[name = tensor("op_37255_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37255_cast_fp16 = einsum(equation = var_37255_equation_0, values = (var_37105_cast_fp16, var_37219_cast_fp16))[name = tensor("op_37255_cast_fp16")]; + tensor var_37257_equation_0 = const()[name = tensor("op_37257_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37257_cast_fp16 = einsum(equation = var_37257_equation_0, values = (var_37109_cast_fp16, var_37220_cast_fp16))[name = tensor("op_37257_cast_fp16")]; + tensor var_37259_equation_0 = const()[name = tensor("op_37259_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37259_cast_fp16 = einsum(equation = var_37259_equation_0, values = (var_37113_cast_fp16, var_37221_cast_fp16))[name = tensor("op_37259_cast_fp16")]; + tensor var_37261_equation_0 = const()[name = tensor("op_37261_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37261_cast_fp16 = einsum(equation = var_37261_equation_0, values = (var_37117_cast_fp16, var_37222_cast_fp16))[name = tensor("op_37261_cast_fp16")]; + tensor var_37263_equation_0 = const()[name = tensor("op_37263_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37263_cast_fp16 = einsum(equation = var_37263_equation_0, values = (var_37121_cast_fp16, var_37223_cast_fp16))[name = tensor("op_37263_cast_fp16")]; + tensor input_495_interleave_0 = const()[name = tensor("input_495_interleave_0"), val = tensor(false)]; + tensor input_495_cast_fp16 = concat(axis = var_30355, interleave = input_495_interleave_0, values = (var_37225_cast_fp16, var_37227_cast_fp16, var_37229_cast_fp16, var_37231_cast_fp16, var_37233_cast_fp16, var_37235_cast_fp16, var_37237_cast_fp16, var_37239_cast_fp16, var_37241_cast_fp16, var_37243_cast_fp16, var_37245_cast_fp16, var_37247_cast_fp16, var_37249_cast_fp16, var_37251_cast_fp16, var_37253_cast_fp16, var_37255_cast_fp16, var_37257_cast_fp16, var_37259_cast_fp16, var_37261_cast_fp16, var_37263_cast_fp16))[name = tensor("input_495_cast_fp16")]; + tensor var_37269 = const()[name = tensor("op_37269"), val = tensor([1, 1])]; + tensor var_37271 = const()[name = tensor("op_37271"), val = tensor([1, 1])]; + tensor var_37273_pad_type_0 = const()[name = tensor("op_37273_pad_type_0"), val = tensor("custom")]; + tensor var_37273_pad_0 = const()[name = tensor("op_37273_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_7_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_7_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3084847936)))]; + tensor up_blocks_0_attentions_0_transformer_blocks_7_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_7_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3088124800)))]; + tensor var_37273_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_7_attn1_to_out_0_bias_to_fp16, dilations = var_37271, groups = var_30355, pad = var_37273_pad_0, pad_type = var_37273_pad_type_0, strides = var_37269, weight = up_blocks_0_attentions_0_transformer_blocks_7_attn1_to_out_0_weight_to_fp16, x = input_495_cast_fp16)[name = tensor("op_37273_cast_fp16")]; + tensor inputs_249_cast_fp16 = add(x = var_37273_cast_fp16, y = inputs_247_cast_fp16)[name = tensor("inputs_249_cast_fp16")]; + tensor hidden_states_331_axes_0 = const()[name = tensor("hidden_states_331_axes_0"), val = tensor([1])]; + tensor hidden_states_331_gamma_0_to_fp16 = const()[name = tensor("hidden_states_331_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3088127424)))]; + tensor hidden_states_331_beta_0_to_fp16 = const()[name = tensor("hidden_states_331_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3088130048)))]; + tensor var_37283_to_fp16 = const()[name = tensor("op_37283_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_331_cast_fp16 = layer_norm(axes = hidden_states_331_axes_0, beta = hidden_states_331_beta_0_to_fp16, epsilon = var_37283_to_fp16, gamma = hidden_states_331_gamma_0_to_fp16, x = inputs_249_cast_fp16)[name = tensor("hidden_states_331_cast_fp16")]; + tensor var_37298 = const()[name = tensor("op_37298"), val = tensor([1, 1])]; + tensor var_37300 = const()[name = tensor("op_37300"), val = tensor([1, 1])]; + tensor q_167_pad_type_0 = const()[name = tensor("q_167_pad_type_0"), val = tensor("custom")]; + tensor q_167_pad_0 = const()[name = tensor("q_167_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_7_attn2_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_7_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3088132672)))]; + tensor q_167_cast_fp16 = conv(dilations = var_37300, groups = var_30355, pad = q_167_pad_0, pad_type = q_167_pad_type_0, strides = var_37298, weight = up_blocks_0_attentions_0_transformer_blocks_7_attn2_to_q_weight_to_fp16, x = hidden_states_331_cast_fp16)[name = tensor("q_167_cast_fp16")]; + tensor var_37304 = const()[name = tensor("op_37304"), val = tensor([1, 1])]; + tensor var_37306 = const()[name = tensor("op_37306"), val = tensor([1, 1])]; + tensor k_333_pad_type_0 = const()[name = tensor("k_333_pad_type_0"), val = tensor("custom")]; + tensor k_333_pad_0 = const()[name = tensor("k_333_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_7_attn2_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_7_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3091409536)))]; + tensor k_333_cast_fp16 = conv(dilations = var_37306, groups = var_30355, pad = k_333_pad_0, pad_type = k_333_pad_type_0, strides = var_37304, weight = up_blocks_0_attentions_0_transformer_blocks_7_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_333_cast_fp16")]; + tensor var_37310 = const()[name = tensor("op_37310"), val = tensor([1, 1])]; + tensor var_37312 = const()[name = tensor("op_37312"), val = tensor([1, 1])]; + tensor v_167_pad_type_0 = const()[name = tensor("v_167_pad_type_0"), val = tensor("custom")]; + tensor v_167_pad_0 = const()[name = tensor("v_167_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_7_attn2_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_7_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3096652480)))]; + tensor v_167_cast_fp16 = conv(dilations = var_37312, groups = var_30355, pad = v_167_pad_0, pad_type = v_167_pad_type_0, strides = var_37310, weight = up_blocks_0_attentions_0_transformer_blocks_7_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_167_cast_fp16")]; + tensor var_37316_begin_0 = const()[name = tensor("op_37316_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_37316_end_0 = const()[name = tensor("op_37316_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_37316_end_mask_0 = const()[name = tensor("op_37316_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37316_cast_fp16 = slice_by_index(begin = var_37316_begin_0, end = var_37316_end_0, end_mask = var_37316_end_mask_0, x = q_167_cast_fp16)[name = tensor("op_37316_cast_fp16")]; + tensor var_37320_begin_0 = const()[name = tensor("op_37320_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_37320_end_0 = const()[name = tensor("op_37320_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_37320_end_mask_0 = const()[name = tensor("op_37320_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37320_cast_fp16 = slice_by_index(begin = var_37320_begin_0, end = var_37320_end_0, end_mask = var_37320_end_mask_0, x = q_167_cast_fp16)[name = tensor("op_37320_cast_fp16")]; + tensor var_37324_begin_0 = const()[name = tensor("op_37324_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_37324_end_0 = const()[name = tensor("op_37324_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_37324_end_mask_0 = const()[name = tensor("op_37324_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37324_cast_fp16 = slice_by_index(begin = var_37324_begin_0, end = var_37324_end_0, end_mask = var_37324_end_mask_0, x = q_167_cast_fp16)[name = tensor("op_37324_cast_fp16")]; + tensor var_37328_begin_0 = const()[name = tensor("op_37328_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_37328_end_0 = const()[name = tensor("op_37328_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_37328_end_mask_0 = const()[name = tensor("op_37328_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37328_cast_fp16 = slice_by_index(begin = var_37328_begin_0, end = var_37328_end_0, end_mask = var_37328_end_mask_0, x = q_167_cast_fp16)[name = tensor("op_37328_cast_fp16")]; + tensor var_37332_begin_0 = const()[name = tensor("op_37332_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_37332_end_0 = const()[name = tensor("op_37332_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_37332_end_mask_0 = const()[name = tensor("op_37332_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37332_cast_fp16 = slice_by_index(begin = var_37332_begin_0, end = var_37332_end_0, end_mask = var_37332_end_mask_0, x = q_167_cast_fp16)[name = tensor("op_37332_cast_fp16")]; + tensor var_37336_begin_0 = const()[name = tensor("op_37336_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_37336_end_0 = const()[name = tensor("op_37336_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_37336_end_mask_0 = const()[name = tensor("op_37336_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37336_cast_fp16 = slice_by_index(begin = var_37336_begin_0, end = var_37336_end_0, end_mask = var_37336_end_mask_0, x = q_167_cast_fp16)[name = tensor("op_37336_cast_fp16")]; + tensor var_37340_begin_0 = const()[name = tensor("op_37340_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_37340_end_0 = const()[name = tensor("op_37340_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_37340_end_mask_0 = const()[name = tensor("op_37340_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37340_cast_fp16 = slice_by_index(begin = var_37340_begin_0, end = var_37340_end_0, end_mask = var_37340_end_mask_0, x = q_167_cast_fp16)[name = tensor("op_37340_cast_fp16")]; + tensor var_37344_begin_0 = const()[name = tensor("op_37344_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_37344_end_0 = const()[name = tensor("op_37344_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_37344_end_mask_0 = const()[name = tensor("op_37344_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37344_cast_fp16 = slice_by_index(begin = var_37344_begin_0, end = var_37344_end_0, end_mask = var_37344_end_mask_0, x = q_167_cast_fp16)[name = tensor("op_37344_cast_fp16")]; + tensor var_37348_begin_0 = const()[name = tensor("op_37348_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_37348_end_0 = const()[name = tensor("op_37348_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_37348_end_mask_0 = const()[name = tensor("op_37348_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37348_cast_fp16 = slice_by_index(begin = var_37348_begin_0, end = var_37348_end_0, end_mask = var_37348_end_mask_0, x = q_167_cast_fp16)[name = tensor("op_37348_cast_fp16")]; + tensor var_37352_begin_0 = const()[name = tensor("op_37352_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_37352_end_0 = const()[name = tensor("op_37352_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_37352_end_mask_0 = const()[name = tensor("op_37352_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37352_cast_fp16 = slice_by_index(begin = var_37352_begin_0, end = var_37352_end_0, end_mask = var_37352_end_mask_0, x = q_167_cast_fp16)[name = tensor("op_37352_cast_fp16")]; + tensor var_37356_begin_0 = const()[name = tensor("op_37356_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_37356_end_0 = const()[name = tensor("op_37356_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_37356_end_mask_0 = const()[name = tensor("op_37356_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37356_cast_fp16 = slice_by_index(begin = var_37356_begin_0, end = var_37356_end_0, end_mask = var_37356_end_mask_0, x = q_167_cast_fp16)[name = tensor("op_37356_cast_fp16")]; + tensor var_37360_begin_0 = const()[name = tensor("op_37360_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_37360_end_0 = const()[name = tensor("op_37360_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_37360_end_mask_0 = const()[name = tensor("op_37360_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37360_cast_fp16 = slice_by_index(begin = var_37360_begin_0, end = var_37360_end_0, end_mask = var_37360_end_mask_0, x = q_167_cast_fp16)[name = tensor("op_37360_cast_fp16")]; + tensor var_37364_begin_0 = const()[name = tensor("op_37364_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_37364_end_0 = const()[name = tensor("op_37364_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_37364_end_mask_0 = const()[name = tensor("op_37364_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37364_cast_fp16 = slice_by_index(begin = var_37364_begin_0, end = var_37364_end_0, end_mask = var_37364_end_mask_0, x = q_167_cast_fp16)[name = tensor("op_37364_cast_fp16")]; + tensor var_37368_begin_0 = const()[name = tensor("op_37368_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_37368_end_0 = const()[name = tensor("op_37368_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_37368_end_mask_0 = const()[name = tensor("op_37368_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37368_cast_fp16 = slice_by_index(begin = var_37368_begin_0, end = var_37368_end_0, end_mask = var_37368_end_mask_0, x = q_167_cast_fp16)[name = tensor("op_37368_cast_fp16")]; + tensor var_37372_begin_0 = const()[name = tensor("op_37372_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_37372_end_0 = const()[name = tensor("op_37372_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_37372_end_mask_0 = const()[name = tensor("op_37372_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37372_cast_fp16 = slice_by_index(begin = var_37372_begin_0, end = var_37372_end_0, end_mask = var_37372_end_mask_0, x = q_167_cast_fp16)[name = tensor("op_37372_cast_fp16")]; + tensor var_37376_begin_0 = const()[name = tensor("op_37376_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_37376_end_0 = const()[name = tensor("op_37376_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_37376_end_mask_0 = const()[name = tensor("op_37376_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37376_cast_fp16 = slice_by_index(begin = var_37376_begin_0, end = var_37376_end_0, end_mask = var_37376_end_mask_0, x = q_167_cast_fp16)[name = tensor("op_37376_cast_fp16")]; + tensor var_37380_begin_0 = const()[name = tensor("op_37380_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_37380_end_0 = const()[name = tensor("op_37380_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_37380_end_mask_0 = const()[name = tensor("op_37380_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37380_cast_fp16 = slice_by_index(begin = var_37380_begin_0, end = var_37380_end_0, end_mask = var_37380_end_mask_0, x = q_167_cast_fp16)[name = tensor("op_37380_cast_fp16")]; + tensor var_37384_begin_0 = const()[name = tensor("op_37384_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_37384_end_0 = const()[name = tensor("op_37384_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_37384_end_mask_0 = const()[name = tensor("op_37384_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37384_cast_fp16 = slice_by_index(begin = var_37384_begin_0, end = var_37384_end_0, end_mask = var_37384_end_mask_0, x = q_167_cast_fp16)[name = tensor("op_37384_cast_fp16")]; + tensor var_37388_begin_0 = const()[name = tensor("op_37388_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_37388_end_0 = const()[name = tensor("op_37388_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_37388_end_mask_0 = const()[name = tensor("op_37388_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37388_cast_fp16 = slice_by_index(begin = var_37388_begin_0, end = var_37388_end_0, end_mask = var_37388_end_mask_0, x = q_167_cast_fp16)[name = tensor("op_37388_cast_fp16")]; + tensor var_37392_begin_0 = const()[name = tensor("op_37392_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_37392_end_0 = const()[name = tensor("op_37392_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_37392_end_mask_0 = const()[name = tensor("op_37392_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37392_cast_fp16 = slice_by_index(begin = var_37392_begin_0, end = var_37392_end_0, end_mask = var_37392_end_mask_0, x = q_167_cast_fp16)[name = tensor("op_37392_cast_fp16")]; + tensor k_335_perm_0 = const()[name = tensor("k_335_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_37399_begin_0 = const()[name = tensor("op_37399_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_37399_end_0 = const()[name = tensor("op_37399_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_37399_end_mask_0 = const()[name = tensor("op_37399_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_335_cast_fp16 = transpose(perm = k_335_perm_0, x = k_333_cast_fp16)[name = tensor("transpose_56")]; + tensor var_37399_cast_fp16 = slice_by_index(begin = var_37399_begin_0, end = var_37399_end_0, end_mask = var_37399_end_mask_0, x = k_335_cast_fp16)[name = tensor("op_37399_cast_fp16")]; + tensor var_37403_begin_0 = const()[name = tensor("op_37403_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_37403_end_0 = const()[name = tensor("op_37403_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_37403_end_mask_0 = const()[name = tensor("op_37403_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37403_cast_fp16 = slice_by_index(begin = var_37403_begin_0, end = var_37403_end_0, end_mask = var_37403_end_mask_0, x = k_335_cast_fp16)[name = tensor("op_37403_cast_fp16")]; + tensor var_37407_begin_0 = const()[name = tensor("op_37407_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_37407_end_0 = const()[name = tensor("op_37407_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_37407_end_mask_0 = const()[name = tensor("op_37407_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37407_cast_fp16 = slice_by_index(begin = var_37407_begin_0, end = var_37407_end_0, end_mask = var_37407_end_mask_0, x = k_335_cast_fp16)[name = tensor("op_37407_cast_fp16")]; + tensor var_37411_begin_0 = const()[name = tensor("op_37411_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_37411_end_0 = const()[name = tensor("op_37411_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_37411_end_mask_0 = const()[name = tensor("op_37411_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37411_cast_fp16 = slice_by_index(begin = var_37411_begin_0, end = var_37411_end_0, end_mask = var_37411_end_mask_0, x = k_335_cast_fp16)[name = tensor("op_37411_cast_fp16")]; + tensor var_37415_begin_0 = const()[name = tensor("op_37415_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_37415_end_0 = const()[name = tensor("op_37415_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_37415_end_mask_0 = const()[name = tensor("op_37415_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37415_cast_fp16 = slice_by_index(begin = var_37415_begin_0, end = var_37415_end_0, end_mask = var_37415_end_mask_0, x = k_335_cast_fp16)[name = tensor("op_37415_cast_fp16")]; + tensor var_37419_begin_0 = const()[name = tensor("op_37419_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_37419_end_0 = const()[name = tensor("op_37419_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_37419_end_mask_0 = const()[name = tensor("op_37419_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37419_cast_fp16 = slice_by_index(begin = var_37419_begin_0, end = var_37419_end_0, end_mask = var_37419_end_mask_0, x = k_335_cast_fp16)[name = tensor("op_37419_cast_fp16")]; + tensor var_37423_begin_0 = const()[name = tensor("op_37423_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_37423_end_0 = const()[name = tensor("op_37423_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_37423_end_mask_0 = const()[name = tensor("op_37423_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37423_cast_fp16 = slice_by_index(begin = var_37423_begin_0, end = var_37423_end_0, end_mask = var_37423_end_mask_0, x = k_335_cast_fp16)[name = tensor("op_37423_cast_fp16")]; + tensor var_37427_begin_0 = const()[name = tensor("op_37427_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_37427_end_0 = const()[name = tensor("op_37427_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_37427_end_mask_0 = const()[name = tensor("op_37427_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37427_cast_fp16 = slice_by_index(begin = var_37427_begin_0, end = var_37427_end_0, end_mask = var_37427_end_mask_0, x = k_335_cast_fp16)[name = tensor("op_37427_cast_fp16")]; + tensor var_37431_begin_0 = const()[name = tensor("op_37431_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_37431_end_0 = const()[name = tensor("op_37431_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_37431_end_mask_0 = const()[name = tensor("op_37431_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37431_cast_fp16 = slice_by_index(begin = var_37431_begin_0, end = var_37431_end_0, end_mask = var_37431_end_mask_0, x = k_335_cast_fp16)[name = tensor("op_37431_cast_fp16")]; + tensor var_37435_begin_0 = const()[name = tensor("op_37435_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_37435_end_0 = const()[name = tensor("op_37435_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_37435_end_mask_0 = const()[name = tensor("op_37435_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37435_cast_fp16 = slice_by_index(begin = var_37435_begin_0, end = var_37435_end_0, end_mask = var_37435_end_mask_0, x = k_335_cast_fp16)[name = tensor("op_37435_cast_fp16")]; + tensor var_37439_begin_0 = const()[name = tensor("op_37439_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_37439_end_0 = const()[name = tensor("op_37439_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_37439_end_mask_0 = const()[name = tensor("op_37439_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37439_cast_fp16 = slice_by_index(begin = var_37439_begin_0, end = var_37439_end_0, end_mask = var_37439_end_mask_0, x = k_335_cast_fp16)[name = tensor("op_37439_cast_fp16")]; + tensor var_37443_begin_0 = const()[name = tensor("op_37443_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_37443_end_0 = const()[name = tensor("op_37443_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_37443_end_mask_0 = const()[name = tensor("op_37443_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37443_cast_fp16 = slice_by_index(begin = var_37443_begin_0, end = var_37443_end_0, end_mask = var_37443_end_mask_0, x = k_335_cast_fp16)[name = tensor("op_37443_cast_fp16")]; + tensor var_37447_begin_0 = const()[name = tensor("op_37447_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_37447_end_0 = const()[name = tensor("op_37447_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_37447_end_mask_0 = const()[name = tensor("op_37447_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37447_cast_fp16 = slice_by_index(begin = var_37447_begin_0, end = var_37447_end_0, end_mask = var_37447_end_mask_0, x = k_335_cast_fp16)[name = tensor("op_37447_cast_fp16")]; + tensor var_37451_begin_0 = const()[name = tensor("op_37451_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_37451_end_0 = const()[name = tensor("op_37451_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_37451_end_mask_0 = const()[name = tensor("op_37451_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37451_cast_fp16 = slice_by_index(begin = var_37451_begin_0, end = var_37451_end_0, end_mask = var_37451_end_mask_0, x = k_335_cast_fp16)[name = tensor("op_37451_cast_fp16")]; + tensor var_37455_begin_0 = const()[name = tensor("op_37455_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_37455_end_0 = const()[name = tensor("op_37455_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_37455_end_mask_0 = const()[name = tensor("op_37455_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37455_cast_fp16 = slice_by_index(begin = var_37455_begin_0, end = var_37455_end_0, end_mask = var_37455_end_mask_0, x = k_335_cast_fp16)[name = tensor("op_37455_cast_fp16")]; + tensor var_37459_begin_0 = const()[name = tensor("op_37459_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_37459_end_0 = const()[name = tensor("op_37459_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_37459_end_mask_0 = const()[name = tensor("op_37459_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37459_cast_fp16 = slice_by_index(begin = var_37459_begin_0, end = var_37459_end_0, end_mask = var_37459_end_mask_0, x = k_335_cast_fp16)[name = tensor("op_37459_cast_fp16")]; + tensor var_37463_begin_0 = const()[name = tensor("op_37463_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_37463_end_0 = const()[name = tensor("op_37463_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_37463_end_mask_0 = const()[name = tensor("op_37463_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37463_cast_fp16 = slice_by_index(begin = var_37463_begin_0, end = var_37463_end_0, end_mask = var_37463_end_mask_0, x = k_335_cast_fp16)[name = tensor("op_37463_cast_fp16")]; + tensor var_37467_begin_0 = const()[name = tensor("op_37467_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_37467_end_0 = const()[name = tensor("op_37467_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_37467_end_mask_0 = const()[name = tensor("op_37467_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37467_cast_fp16 = slice_by_index(begin = var_37467_begin_0, end = var_37467_end_0, end_mask = var_37467_end_mask_0, x = k_335_cast_fp16)[name = tensor("op_37467_cast_fp16")]; + tensor var_37471_begin_0 = const()[name = tensor("op_37471_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_37471_end_0 = const()[name = tensor("op_37471_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_37471_end_mask_0 = const()[name = tensor("op_37471_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37471_cast_fp16 = slice_by_index(begin = var_37471_begin_0, end = var_37471_end_0, end_mask = var_37471_end_mask_0, x = k_335_cast_fp16)[name = tensor("op_37471_cast_fp16")]; + tensor var_37475_begin_0 = const()[name = tensor("op_37475_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_37475_end_0 = const()[name = tensor("op_37475_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_37475_end_mask_0 = const()[name = tensor("op_37475_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37475_cast_fp16 = slice_by_index(begin = var_37475_begin_0, end = var_37475_end_0, end_mask = var_37475_end_mask_0, x = k_335_cast_fp16)[name = tensor("op_37475_cast_fp16")]; + tensor var_37477_begin_0 = const()[name = tensor("op_37477_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_37477_end_0 = const()[name = tensor("op_37477_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_37477_end_mask_0 = const()[name = tensor("op_37477_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37477_cast_fp16 = slice_by_index(begin = var_37477_begin_0, end = var_37477_end_0, end_mask = var_37477_end_mask_0, x = v_167_cast_fp16)[name = tensor("op_37477_cast_fp16")]; + tensor var_37481_begin_0 = const()[name = tensor("op_37481_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_37481_end_0 = const()[name = tensor("op_37481_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_37481_end_mask_0 = const()[name = tensor("op_37481_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37481_cast_fp16 = slice_by_index(begin = var_37481_begin_0, end = var_37481_end_0, end_mask = var_37481_end_mask_0, x = v_167_cast_fp16)[name = tensor("op_37481_cast_fp16")]; + tensor var_37485_begin_0 = const()[name = tensor("op_37485_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_37485_end_0 = const()[name = tensor("op_37485_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_37485_end_mask_0 = const()[name = tensor("op_37485_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37485_cast_fp16 = slice_by_index(begin = var_37485_begin_0, end = var_37485_end_0, end_mask = var_37485_end_mask_0, x = v_167_cast_fp16)[name = tensor("op_37485_cast_fp16")]; + tensor var_37489_begin_0 = const()[name = tensor("op_37489_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_37489_end_0 = const()[name = tensor("op_37489_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_37489_end_mask_0 = const()[name = tensor("op_37489_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37489_cast_fp16 = slice_by_index(begin = var_37489_begin_0, end = var_37489_end_0, end_mask = var_37489_end_mask_0, x = v_167_cast_fp16)[name = tensor("op_37489_cast_fp16")]; + tensor var_37493_begin_0 = const()[name = tensor("op_37493_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_37493_end_0 = const()[name = tensor("op_37493_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_37493_end_mask_0 = const()[name = tensor("op_37493_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37493_cast_fp16 = slice_by_index(begin = var_37493_begin_0, end = var_37493_end_0, end_mask = var_37493_end_mask_0, x = v_167_cast_fp16)[name = tensor("op_37493_cast_fp16")]; + tensor var_37497_begin_0 = const()[name = tensor("op_37497_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_37497_end_0 = const()[name = tensor("op_37497_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_37497_end_mask_0 = const()[name = tensor("op_37497_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37497_cast_fp16 = slice_by_index(begin = var_37497_begin_0, end = var_37497_end_0, end_mask = var_37497_end_mask_0, x = v_167_cast_fp16)[name = tensor("op_37497_cast_fp16")]; + tensor var_37501_begin_0 = const()[name = tensor("op_37501_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_37501_end_0 = const()[name = tensor("op_37501_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_37501_end_mask_0 = const()[name = tensor("op_37501_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37501_cast_fp16 = slice_by_index(begin = var_37501_begin_0, end = var_37501_end_0, end_mask = var_37501_end_mask_0, x = v_167_cast_fp16)[name = tensor("op_37501_cast_fp16")]; + tensor var_37505_begin_0 = const()[name = tensor("op_37505_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_37505_end_0 = const()[name = tensor("op_37505_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_37505_end_mask_0 = const()[name = tensor("op_37505_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37505_cast_fp16 = slice_by_index(begin = var_37505_begin_0, end = var_37505_end_0, end_mask = var_37505_end_mask_0, x = v_167_cast_fp16)[name = tensor("op_37505_cast_fp16")]; + tensor var_37509_begin_0 = const()[name = tensor("op_37509_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_37509_end_0 = const()[name = tensor("op_37509_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_37509_end_mask_0 = const()[name = tensor("op_37509_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37509_cast_fp16 = slice_by_index(begin = var_37509_begin_0, end = var_37509_end_0, end_mask = var_37509_end_mask_0, x = v_167_cast_fp16)[name = tensor("op_37509_cast_fp16")]; + tensor var_37513_begin_0 = const()[name = tensor("op_37513_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_37513_end_0 = const()[name = tensor("op_37513_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_37513_end_mask_0 = const()[name = tensor("op_37513_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37513_cast_fp16 = slice_by_index(begin = var_37513_begin_0, end = var_37513_end_0, end_mask = var_37513_end_mask_0, x = v_167_cast_fp16)[name = tensor("op_37513_cast_fp16")]; + tensor var_37517_begin_0 = const()[name = tensor("op_37517_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_37517_end_0 = const()[name = tensor("op_37517_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_37517_end_mask_0 = const()[name = tensor("op_37517_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37517_cast_fp16 = slice_by_index(begin = var_37517_begin_0, end = var_37517_end_0, end_mask = var_37517_end_mask_0, x = v_167_cast_fp16)[name = tensor("op_37517_cast_fp16")]; + tensor var_37521_begin_0 = const()[name = tensor("op_37521_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_37521_end_0 = const()[name = tensor("op_37521_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_37521_end_mask_0 = const()[name = tensor("op_37521_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37521_cast_fp16 = slice_by_index(begin = var_37521_begin_0, end = var_37521_end_0, end_mask = var_37521_end_mask_0, x = v_167_cast_fp16)[name = tensor("op_37521_cast_fp16")]; + tensor var_37525_begin_0 = const()[name = tensor("op_37525_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_37525_end_0 = const()[name = tensor("op_37525_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_37525_end_mask_0 = const()[name = tensor("op_37525_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37525_cast_fp16 = slice_by_index(begin = var_37525_begin_0, end = var_37525_end_0, end_mask = var_37525_end_mask_0, x = v_167_cast_fp16)[name = tensor("op_37525_cast_fp16")]; + tensor var_37529_begin_0 = const()[name = tensor("op_37529_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_37529_end_0 = const()[name = tensor("op_37529_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_37529_end_mask_0 = const()[name = tensor("op_37529_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37529_cast_fp16 = slice_by_index(begin = var_37529_begin_0, end = var_37529_end_0, end_mask = var_37529_end_mask_0, x = v_167_cast_fp16)[name = tensor("op_37529_cast_fp16")]; + tensor var_37533_begin_0 = const()[name = tensor("op_37533_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_37533_end_0 = const()[name = tensor("op_37533_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_37533_end_mask_0 = const()[name = tensor("op_37533_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37533_cast_fp16 = slice_by_index(begin = var_37533_begin_0, end = var_37533_end_0, end_mask = var_37533_end_mask_0, x = v_167_cast_fp16)[name = tensor("op_37533_cast_fp16")]; + tensor var_37537_begin_0 = const()[name = tensor("op_37537_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_37537_end_0 = const()[name = tensor("op_37537_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_37537_end_mask_0 = const()[name = tensor("op_37537_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37537_cast_fp16 = slice_by_index(begin = var_37537_begin_0, end = var_37537_end_0, end_mask = var_37537_end_mask_0, x = v_167_cast_fp16)[name = tensor("op_37537_cast_fp16")]; + tensor var_37541_begin_0 = const()[name = tensor("op_37541_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_37541_end_0 = const()[name = tensor("op_37541_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_37541_end_mask_0 = const()[name = tensor("op_37541_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37541_cast_fp16 = slice_by_index(begin = var_37541_begin_0, end = var_37541_end_0, end_mask = var_37541_end_mask_0, x = v_167_cast_fp16)[name = tensor("op_37541_cast_fp16")]; + tensor var_37545_begin_0 = const()[name = tensor("op_37545_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_37545_end_0 = const()[name = tensor("op_37545_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_37545_end_mask_0 = const()[name = tensor("op_37545_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37545_cast_fp16 = slice_by_index(begin = var_37545_begin_0, end = var_37545_end_0, end_mask = var_37545_end_mask_0, x = v_167_cast_fp16)[name = tensor("op_37545_cast_fp16")]; + tensor var_37549_begin_0 = const()[name = tensor("op_37549_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_37549_end_0 = const()[name = tensor("op_37549_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_37549_end_mask_0 = const()[name = tensor("op_37549_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37549_cast_fp16 = slice_by_index(begin = var_37549_begin_0, end = var_37549_end_0, end_mask = var_37549_end_mask_0, x = v_167_cast_fp16)[name = tensor("op_37549_cast_fp16")]; + tensor var_37553_begin_0 = const()[name = tensor("op_37553_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_37553_end_0 = const()[name = tensor("op_37553_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_37553_end_mask_0 = const()[name = tensor("op_37553_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37553_cast_fp16 = slice_by_index(begin = var_37553_begin_0, end = var_37553_end_0, end_mask = var_37553_end_mask_0, x = v_167_cast_fp16)[name = tensor("op_37553_cast_fp16")]; + tensor var_37557_equation_0 = const()[name = tensor("op_37557_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37557_cast_fp16 = einsum(equation = var_37557_equation_0, values = (var_37399_cast_fp16, var_37316_cast_fp16))[name = tensor("op_37557_cast_fp16")]; + tensor var_37558_to_fp16 = const()[name = tensor("op_37558_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3161_cast_fp16 = mul(x = var_37557_cast_fp16, y = var_37558_to_fp16)[name = tensor("aw_3161_cast_fp16")]; + tensor var_37561_equation_0 = const()[name = tensor("op_37561_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37561_cast_fp16 = einsum(equation = var_37561_equation_0, values = (var_37403_cast_fp16, var_37320_cast_fp16))[name = tensor("op_37561_cast_fp16")]; + tensor var_37562_to_fp16 = const()[name = tensor("op_37562_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3163_cast_fp16 = mul(x = var_37561_cast_fp16, y = var_37562_to_fp16)[name = tensor("aw_3163_cast_fp16")]; + tensor var_37565_equation_0 = const()[name = tensor("op_37565_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37565_cast_fp16 = einsum(equation = var_37565_equation_0, values = (var_37407_cast_fp16, var_37324_cast_fp16))[name = tensor("op_37565_cast_fp16")]; + tensor var_37566_to_fp16 = const()[name = tensor("op_37566_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3165_cast_fp16 = mul(x = var_37565_cast_fp16, y = var_37566_to_fp16)[name = tensor("aw_3165_cast_fp16")]; + tensor var_37569_equation_0 = const()[name = tensor("op_37569_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37569_cast_fp16 = einsum(equation = var_37569_equation_0, values = (var_37411_cast_fp16, var_37328_cast_fp16))[name = tensor("op_37569_cast_fp16")]; + tensor var_37570_to_fp16 = const()[name = tensor("op_37570_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3167_cast_fp16 = mul(x = var_37569_cast_fp16, y = var_37570_to_fp16)[name = tensor("aw_3167_cast_fp16")]; + tensor var_37573_equation_0 = const()[name = tensor("op_37573_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37573_cast_fp16 = einsum(equation = var_37573_equation_0, values = (var_37415_cast_fp16, var_37332_cast_fp16))[name = tensor("op_37573_cast_fp16")]; + tensor var_37574_to_fp16 = const()[name = tensor("op_37574_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3169_cast_fp16 = mul(x = var_37573_cast_fp16, y = var_37574_to_fp16)[name = tensor("aw_3169_cast_fp16")]; + tensor var_37577_equation_0 = const()[name = tensor("op_37577_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37577_cast_fp16 = einsum(equation = var_37577_equation_0, values = (var_37419_cast_fp16, var_37336_cast_fp16))[name = tensor("op_37577_cast_fp16")]; + tensor var_37578_to_fp16 = const()[name = tensor("op_37578_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3171_cast_fp16 = mul(x = var_37577_cast_fp16, y = var_37578_to_fp16)[name = tensor("aw_3171_cast_fp16")]; + tensor var_37581_equation_0 = const()[name = tensor("op_37581_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37581_cast_fp16 = einsum(equation = var_37581_equation_0, values = (var_37423_cast_fp16, var_37340_cast_fp16))[name = tensor("op_37581_cast_fp16")]; + tensor var_37582_to_fp16 = const()[name = tensor("op_37582_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3173_cast_fp16 = mul(x = var_37581_cast_fp16, y = var_37582_to_fp16)[name = tensor("aw_3173_cast_fp16")]; + tensor var_37585_equation_0 = const()[name = tensor("op_37585_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37585_cast_fp16 = einsum(equation = var_37585_equation_0, values = (var_37427_cast_fp16, var_37344_cast_fp16))[name = tensor("op_37585_cast_fp16")]; + tensor var_37586_to_fp16 = const()[name = tensor("op_37586_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3175_cast_fp16 = mul(x = var_37585_cast_fp16, y = var_37586_to_fp16)[name = tensor("aw_3175_cast_fp16")]; + tensor var_37589_equation_0 = const()[name = tensor("op_37589_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37589_cast_fp16 = einsum(equation = var_37589_equation_0, values = (var_37431_cast_fp16, var_37348_cast_fp16))[name = tensor("op_37589_cast_fp16")]; + tensor var_37590_to_fp16 = const()[name = tensor("op_37590_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3177_cast_fp16 = mul(x = var_37589_cast_fp16, y = var_37590_to_fp16)[name = tensor("aw_3177_cast_fp16")]; + tensor var_37593_equation_0 = const()[name = tensor("op_37593_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37593_cast_fp16 = einsum(equation = var_37593_equation_0, values = (var_37435_cast_fp16, var_37352_cast_fp16))[name = tensor("op_37593_cast_fp16")]; + tensor var_37594_to_fp16 = const()[name = tensor("op_37594_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3179_cast_fp16 = mul(x = var_37593_cast_fp16, y = var_37594_to_fp16)[name = tensor("aw_3179_cast_fp16")]; + tensor var_37597_equation_0 = const()[name = tensor("op_37597_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37597_cast_fp16 = einsum(equation = var_37597_equation_0, values = (var_37439_cast_fp16, var_37356_cast_fp16))[name = tensor("op_37597_cast_fp16")]; + tensor var_37598_to_fp16 = const()[name = tensor("op_37598_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3181_cast_fp16 = mul(x = var_37597_cast_fp16, y = var_37598_to_fp16)[name = tensor("aw_3181_cast_fp16")]; + tensor var_37601_equation_0 = const()[name = tensor("op_37601_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37601_cast_fp16 = einsum(equation = var_37601_equation_0, values = (var_37443_cast_fp16, var_37360_cast_fp16))[name = tensor("op_37601_cast_fp16")]; + tensor var_37602_to_fp16 = const()[name = tensor("op_37602_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3183_cast_fp16 = mul(x = var_37601_cast_fp16, y = var_37602_to_fp16)[name = tensor("aw_3183_cast_fp16")]; + tensor var_37605_equation_0 = const()[name = tensor("op_37605_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37605_cast_fp16 = einsum(equation = var_37605_equation_0, values = (var_37447_cast_fp16, var_37364_cast_fp16))[name = tensor("op_37605_cast_fp16")]; + tensor var_37606_to_fp16 = const()[name = tensor("op_37606_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3185_cast_fp16 = mul(x = var_37605_cast_fp16, y = var_37606_to_fp16)[name = tensor("aw_3185_cast_fp16")]; + tensor var_37609_equation_0 = const()[name = tensor("op_37609_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37609_cast_fp16 = einsum(equation = var_37609_equation_0, values = (var_37451_cast_fp16, var_37368_cast_fp16))[name = tensor("op_37609_cast_fp16")]; + tensor var_37610_to_fp16 = const()[name = tensor("op_37610_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3187_cast_fp16 = mul(x = var_37609_cast_fp16, y = var_37610_to_fp16)[name = tensor("aw_3187_cast_fp16")]; + tensor var_37613_equation_0 = const()[name = tensor("op_37613_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37613_cast_fp16 = einsum(equation = var_37613_equation_0, values = (var_37455_cast_fp16, var_37372_cast_fp16))[name = tensor("op_37613_cast_fp16")]; + tensor var_37614_to_fp16 = const()[name = tensor("op_37614_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3189_cast_fp16 = mul(x = var_37613_cast_fp16, y = var_37614_to_fp16)[name = tensor("aw_3189_cast_fp16")]; + tensor var_37617_equation_0 = const()[name = tensor("op_37617_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37617_cast_fp16 = einsum(equation = var_37617_equation_0, values = (var_37459_cast_fp16, var_37376_cast_fp16))[name = tensor("op_37617_cast_fp16")]; + tensor var_37618_to_fp16 = const()[name = tensor("op_37618_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3191_cast_fp16 = mul(x = var_37617_cast_fp16, y = var_37618_to_fp16)[name = tensor("aw_3191_cast_fp16")]; + tensor var_37621_equation_0 = const()[name = tensor("op_37621_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37621_cast_fp16 = einsum(equation = var_37621_equation_0, values = (var_37463_cast_fp16, var_37380_cast_fp16))[name = tensor("op_37621_cast_fp16")]; + tensor var_37622_to_fp16 = const()[name = tensor("op_37622_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3193_cast_fp16 = mul(x = var_37621_cast_fp16, y = var_37622_to_fp16)[name = tensor("aw_3193_cast_fp16")]; + tensor var_37625_equation_0 = const()[name = tensor("op_37625_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37625_cast_fp16 = einsum(equation = var_37625_equation_0, values = (var_37467_cast_fp16, var_37384_cast_fp16))[name = tensor("op_37625_cast_fp16")]; + tensor var_37626_to_fp16 = const()[name = tensor("op_37626_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3195_cast_fp16 = mul(x = var_37625_cast_fp16, y = var_37626_to_fp16)[name = tensor("aw_3195_cast_fp16")]; + tensor var_37629_equation_0 = const()[name = tensor("op_37629_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37629_cast_fp16 = einsum(equation = var_37629_equation_0, values = (var_37471_cast_fp16, var_37388_cast_fp16))[name = tensor("op_37629_cast_fp16")]; + tensor var_37630_to_fp16 = const()[name = tensor("op_37630_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3197_cast_fp16 = mul(x = var_37629_cast_fp16, y = var_37630_to_fp16)[name = tensor("aw_3197_cast_fp16")]; + tensor var_37633_equation_0 = const()[name = tensor("op_37633_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37633_cast_fp16 = einsum(equation = var_37633_equation_0, values = (var_37475_cast_fp16, var_37392_cast_fp16))[name = tensor("op_37633_cast_fp16")]; + tensor var_37634_to_fp16 = const()[name = tensor("op_37634_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3199_cast_fp16 = mul(x = var_37633_cast_fp16, y = var_37634_to_fp16)[name = tensor("aw_3199_cast_fp16")]; + tensor var_37636_cast_fp16 = softmax(axis = var_30355, x = aw_3161_cast_fp16)[name = tensor("op_37636_cast_fp16")]; + tensor var_37637_cast_fp16 = softmax(axis = var_30355, x = aw_3163_cast_fp16)[name = tensor("op_37637_cast_fp16")]; + tensor var_37638_cast_fp16 = softmax(axis = var_30355, x = aw_3165_cast_fp16)[name = tensor("op_37638_cast_fp16")]; + tensor var_37639_cast_fp16 = softmax(axis = var_30355, x = aw_3167_cast_fp16)[name = tensor("op_37639_cast_fp16")]; + tensor var_37640_cast_fp16 = softmax(axis = var_30355, x = aw_3169_cast_fp16)[name = tensor("op_37640_cast_fp16")]; + tensor var_37641_cast_fp16 = softmax(axis = var_30355, x = aw_3171_cast_fp16)[name = tensor("op_37641_cast_fp16")]; + tensor var_37642_cast_fp16 = softmax(axis = var_30355, x = aw_3173_cast_fp16)[name = tensor("op_37642_cast_fp16")]; + tensor var_37643_cast_fp16 = softmax(axis = var_30355, x = aw_3175_cast_fp16)[name = tensor("op_37643_cast_fp16")]; + tensor var_37644_cast_fp16 = softmax(axis = var_30355, x = aw_3177_cast_fp16)[name = tensor("op_37644_cast_fp16")]; + tensor var_37645_cast_fp16 = softmax(axis = var_30355, x = aw_3179_cast_fp16)[name = tensor("op_37645_cast_fp16")]; + tensor var_37646_cast_fp16 = softmax(axis = var_30355, x = aw_3181_cast_fp16)[name = tensor("op_37646_cast_fp16")]; + tensor var_37647_cast_fp16 = softmax(axis = var_30355, x = aw_3183_cast_fp16)[name = tensor("op_37647_cast_fp16")]; + tensor var_37648_cast_fp16 = softmax(axis = var_30355, x = aw_3185_cast_fp16)[name = tensor("op_37648_cast_fp16")]; + tensor var_37649_cast_fp16 = softmax(axis = var_30355, x = aw_3187_cast_fp16)[name = tensor("op_37649_cast_fp16")]; + tensor var_37650_cast_fp16 = softmax(axis = var_30355, x = aw_3189_cast_fp16)[name = tensor("op_37650_cast_fp16")]; + tensor var_37651_cast_fp16 = softmax(axis = var_30355, x = aw_3191_cast_fp16)[name = tensor("op_37651_cast_fp16")]; + tensor var_37652_cast_fp16 = softmax(axis = var_30355, x = aw_3193_cast_fp16)[name = tensor("op_37652_cast_fp16")]; + tensor var_37653_cast_fp16 = softmax(axis = var_30355, x = aw_3195_cast_fp16)[name = tensor("op_37653_cast_fp16")]; + tensor var_37654_cast_fp16 = softmax(axis = var_30355, x = aw_3197_cast_fp16)[name = tensor("op_37654_cast_fp16")]; + tensor var_37655_cast_fp16 = softmax(axis = var_30355, x = aw_3199_cast_fp16)[name = tensor("op_37655_cast_fp16")]; + tensor var_37657_equation_0 = const()[name = tensor("op_37657_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37657_cast_fp16 = einsum(equation = var_37657_equation_0, values = (var_37477_cast_fp16, var_37636_cast_fp16))[name = tensor("op_37657_cast_fp16")]; + tensor var_37659_equation_0 = const()[name = tensor("op_37659_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37659_cast_fp16 = einsum(equation = var_37659_equation_0, values = (var_37481_cast_fp16, var_37637_cast_fp16))[name = tensor("op_37659_cast_fp16")]; + tensor var_37661_equation_0 = const()[name = tensor("op_37661_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37661_cast_fp16 = einsum(equation = var_37661_equation_0, values = (var_37485_cast_fp16, var_37638_cast_fp16))[name = tensor("op_37661_cast_fp16")]; + tensor var_37663_equation_0 = const()[name = tensor("op_37663_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37663_cast_fp16 = einsum(equation = var_37663_equation_0, values = (var_37489_cast_fp16, var_37639_cast_fp16))[name = tensor("op_37663_cast_fp16")]; + tensor var_37665_equation_0 = const()[name = tensor("op_37665_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37665_cast_fp16 = einsum(equation = var_37665_equation_0, values = (var_37493_cast_fp16, var_37640_cast_fp16))[name = tensor("op_37665_cast_fp16")]; + tensor var_37667_equation_0 = const()[name = tensor("op_37667_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37667_cast_fp16 = einsum(equation = var_37667_equation_0, values = (var_37497_cast_fp16, var_37641_cast_fp16))[name = tensor("op_37667_cast_fp16")]; + tensor var_37669_equation_0 = const()[name = tensor("op_37669_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37669_cast_fp16 = einsum(equation = var_37669_equation_0, values = (var_37501_cast_fp16, var_37642_cast_fp16))[name = tensor("op_37669_cast_fp16")]; + tensor var_37671_equation_0 = const()[name = tensor("op_37671_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37671_cast_fp16 = einsum(equation = var_37671_equation_0, values = (var_37505_cast_fp16, var_37643_cast_fp16))[name = tensor("op_37671_cast_fp16")]; + tensor var_37673_equation_0 = const()[name = tensor("op_37673_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37673_cast_fp16 = einsum(equation = var_37673_equation_0, values = (var_37509_cast_fp16, var_37644_cast_fp16))[name = tensor("op_37673_cast_fp16")]; + tensor var_37675_equation_0 = const()[name = tensor("op_37675_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37675_cast_fp16 = einsum(equation = var_37675_equation_0, values = (var_37513_cast_fp16, var_37645_cast_fp16))[name = tensor("op_37675_cast_fp16")]; + tensor var_37677_equation_0 = const()[name = tensor("op_37677_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37677_cast_fp16 = einsum(equation = var_37677_equation_0, values = (var_37517_cast_fp16, var_37646_cast_fp16))[name = tensor("op_37677_cast_fp16")]; + tensor var_37679_equation_0 = const()[name = tensor("op_37679_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37679_cast_fp16 = einsum(equation = var_37679_equation_0, values = (var_37521_cast_fp16, var_37647_cast_fp16))[name = tensor("op_37679_cast_fp16")]; + tensor var_37681_equation_0 = const()[name = tensor("op_37681_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37681_cast_fp16 = einsum(equation = var_37681_equation_0, values = (var_37525_cast_fp16, var_37648_cast_fp16))[name = tensor("op_37681_cast_fp16")]; + tensor var_37683_equation_0 = const()[name = tensor("op_37683_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37683_cast_fp16 = einsum(equation = var_37683_equation_0, values = (var_37529_cast_fp16, var_37649_cast_fp16))[name = tensor("op_37683_cast_fp16")]; + tensor var_37685_equation_0 = const()[name = tensor("op_37685_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37685_cast_fp16 = einsum(equation = var_37685_equation_0, values = (var_37533_cast_fp16, var_37650_cast_fp16))[name = tensor("op_37685_cast_fp16")]; + tensor var_37687_equation_0 = const()[name = tensor("op_37687_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37687_cast_fp16 = einsum(equation = var_37687_equation_0, values = (var_37537_cast_fp16, var_37651_cast_fp16))[name = tensor("op_37687_cast_fp16")]; + tensor var_37689_equation_0 = const()[name = tensor("op_37689_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37689_cast_fp16 = einsum(equation = var_37689_equation_0, values = (var_37541_cast_fp16, var_37652_cast_fp16))[name = tensor("op_37689_cast_fp16")]; + tensor var_37691_equation_0 = const()[name = tensor("op_37691_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37691_cast_fp16 = einsum(equation = var_37691_equation_0, values = (var_37545_cast_fp16, var_37653_cast_fp16))[name = tensor("op_37691_cast_fp16")]; + tensor var_37693_equation_0 = const()[name = tensor("op_37693_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37693_cast_fp16 = einsum(equation = var_37693_equation_0, values = (var_37549_cast_fp16, var_37654_cast_fp16))[name = tensor("op_37693_cast_fp16")]; + tensor var_37695_equation_0 = const()[name = tensor("op_37695_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37695_cast_fp16 = einsum(equation = var_37695_equation_0, values = (var_37553_cast_fp16, var_37655_cast_fp16))[name = tensor("op_37695_cast_fp16")]; + tensor input_497_interleave_0 = const()[name = tensor("input_497_interleave_0"), val = tensor(false)]; + tensor input_497_cast_fp16 = concat(axis = var_30355, interleave = input_497_interleave_0, values = (var_37657_cast_fp16, var_37659_cast_fp16, var_37661_cast_fp16, var_37663_cast_fp16, var_37665_cast_fp16, var_37667_cast_fp16, var_37669_cast_fp16, var_37671_cast_fp16, var_37673_cast_fp16, var_37675_cast_fp16, var_37677_cast_fp16, var_37679_cast_fp16, var_37681_cast_fp16, var_37683_cast_fp16, var_37685_cast_fp16, var_37687_cast_fp16, var_37689_cast_fp16, var_37691_cast_fp16, var_37693_cast_fp16, var_37695_cast_fp16))[name = tensor("input_497_cast_fp16")]; + tensor var_37701 = const()[name = tensor("op_37701"), val = tensor([1, 1])]; + tensor var_37703 = const()[name = tensor("op_37703"), val = tensor([1, 1])]; + tensor var_37705_pad_type_0 = const()[name = tensor("op_37705_pad_type_0"), val = tensor("custom")]; + tensor var_37705_pad_0 = const()[name = tensor("op_37705_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_7_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_7_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3101895424)))]; + tensor up_blocks_0_attentions_0_transformer_blocks_7_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_7_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3105172288)))]; + tensor var_37705_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_7_attn2_to_out_0_bias_to_fp16, dilations = var_37703, groups = var_30355, pad = var_37705_pad_0, pad_type = var_37705_pad_type_0, strides = var_37701, weight = up_blocks_0_attentions_0_transformer_blocks_7_attn2_to_out_0_weight_to_fp16, x = input_497_cast_fp16)[name = tensor("op_37705_cast_fp16")]; + tensor inputs_251_cast_fp16 = add(x = var_37705_cast_fp16, y = inputs_249_cast_fp16)[name = tensor("inputs_251_cast_fp16")]; + tensor input_499_axes_0 = const()[name = tensor("input_499_axes_0"), val = tensor([1])]; + tensor input_499_gamma_0_to_fp16 = const()[name = tensor("input_499_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3105174912)))]; + tensor input_499_beta_0_to_fp16 = const()[name = tensor("input_499_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3105177536)))]; + tensor var_37715_to_fp16 = const()[name = tensor("op_37715_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_499_cast_fp16 = layer_norm(axes = input_499_axes_0, beta = input_499_beta_0_to_fp16, epsilon = var_37715_to_fp16, gamma = input_499_gamma_0_to_fp16, x = inputs_251_cast_fp16)[name = tensor("input_499_cast_fp16")]; + tensor var_37731 = const()[name = tensor("op_37731"), val = tensor([1, 1])]; + tensor var_37733 = const()[name = tensor("op_37733"), val = tensor([1, 1])]; + tensor var_37735_pad_type_0 = const()[name = tensor("op_37735_pad_type_0"), val = tensor("custom")]; + tensor var_37735_pad_0 = const()[name = tensor("op_37735_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_7_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_7_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3105180160)))]; + tensor up_blocks_0_attentions_0_transformer_blocks_7_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_7_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3131394624)))]; + tensor var_37735_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_7_ff_net_0_proj_bias_to_fp16, dilations = var_37733, groups = var_30355, pad = var_37735_pad_0, pad_type = var_37735_pad_type_0, strides = var_37731, weight = up_blocks_0_attentions_0_transformer_blocks_7_ff_net_0_proj_weight_to_fp16, x = input_499_cast_fp16)[name = tensor("op_37735_cast_fp16")]; + tensor var_37736_split_sizes_0 = const()[name = tensor("op_37736_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_37736_axis_0 = const()[name = tensor("op_37736_axis_0"), val = tensor(1)]; + tensor var_37736_cast_fp16_0, tensor var_37736_cast_fp16_1 = split(axis = var_37736_axis_0, split_sizes = var_37736_split_sizes_0, x = var_37735_cast_fp16)[name = tensor("op_37736_cast_fp16")]; + tensor var_37738_mode_0 = const()[name = tensor("op_37738_mode_0"), val = tensor("EXACT")]; + tensor var_37738_cast_fp16 = gelu(mode = var_37738_mode_0, x = var_37736_cast_fp16_1)[name = tensor("op_37738_cast_fp16")]; + tensor input_501_cast_fp16 = mul(x = var_37736_cast_fp16_0, y = var_37738_cast_fp16)[name = tensor("input_501_cast_fp16")]; + tensor var_37742 = const()[name = tensor("op_37742"), val = tensor([1, 1])]; + tensor var_37744 = const()[name = tensor("op_37744"), val = tensor([1, 1])]; + tensor var_37746_pad_type_0 = const()[name = tensor("op_37746_pad_type_0"), val = tensor("custom")]; + tensor var_37746_pad_0 = const()[name = tensor("op_37746_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_7_ff_net_2_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_7_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3131415168)))]; + tensor up_blocks_0_attentions_0_transformer_blocks_7_ff_net_2_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_7_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3144522432)))]; + tensor var_37746_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_7_ff_net_2_bias_to_fp16, dilations = var_37744, groups = var_30355, pad = var_37746_pad_0, pad_type = var_37746_pad_type_0, strides = var_37742, weight = up_blocks_0_attentions_0_transformer_blocks_7_ff_net_2_weight_to_fp16, x = input_501_cast_fp16)[name = tensor("op_37746_cast_fp16")]; + tensor inputs_253_cast_fp16 = add(x = var_37746_cast_fp16, y = inputs_251_cast_fp16)[name = tensor("inputs_253_cast_fp16")]; + tensor hidden_states_335_axes_0 = const()[name = tensor("hidden_states_335_axes_0"), val = tensor([1])]; + tensor hidden_states_335_gamma_0_to_fp16 = const()[name = tensor("hidden_states_335_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3144525056)))]; + tensor hidden_states_335_beta_0_to_fp16 = const()[name = tensor("hidden_states_335_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3144527680)))]; + tensor var_37762_to_fp16 = const()[name = tensor("op_37762_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_335_cast_fp16 = layer_norm(axes = hidden_states_335_axes_0, beta = hidden_states_335_beta_0_to_fp16, epsilon = var_37762_to_fp16, gamma = hidden_states_335_gamma_0_to_fp16, x = inputs_253_cast_fp16)[name = tensor("hidden_states_335_cast_fp16")]; + tensor var_37777 = const()[name = tensor("op_37777"), val = tensor([1, 1])]; + tensor var_37779 = const()[name = tensor("op_37779"), val = tensor([1, 1])]; + tensor q_169_pad_type_0 = const()[name = tensor("q_169_pad_type_0"), val = tensor("custom")]; + tensor q_169_pad_0 = const()[name = tensor("q_169_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_8_attn1_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_8_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3144530304)))]; + tensor q_169_cast_fp16 = conv(dilations = var_37779, groups = var_30355, pad = q_169_pad_0, pad_type = q_169_pad_type_0, strides = var_37777, weight = up_blocks_0_attentions_0_transformer_blocks_8_attn1_to_q_weight_to_fp16, x = hidden_states_335_cast_fp16)[name = tensor("q_169_cast_fp16")]; + tensor var_37783 = const()[name = tensor("op_37783"), val = tensor([1, 1])]; + tensor var_37785 = const()[name = tensor("op_37785"), val = tensor([1, 1])]; + tensor k_337_pad_type_0 = const()[name = tensor("k_337_pad_type_0"), val = tensor("custom")]; + tensor k_337_pad_0 = const()[name = tensor("k_337_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_8_attn1_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_8_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3147807168)))]; + tensor k_337_cast_fp16 = conv(dilations = var_37785, groups = var_30355, pad = k_337_pad_0, pad_type = k_337_pad_type_0, strides = var_37783, weight = up_blocks_0_attentions_0_transformer_blocks_8_attn1_to_k_weight_to_fp16, x = hidden_states_335_cast_fp16)[name = tensor("k_337_cast_fp16")]; + tensor var_37789 = const()[name = tensor("op_37789"), val = tensor([1, 1])]; + tensor var_37791 = const()[name = tensor("op_37791"), val = tensor([1, 1])]; + tensor v_169_pad_type_0 = const()[name = tensor("v_169_pad_type_0"), val = tensor("custom")]; + tensor v_169_pad_0 = const()[name = tensor("v_169_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_8_attn1_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_8_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3151084032)))]; + tensor v_169_cast_fp16 = conv(dilations = var_37791, groups = var_30355, pad = v_169_pad_0, pad_type = v_169_pad_type_0, strides = var_37789, weight = up_blocks_0_attentions_0_transformer_blocks_8_attn1_to_v_weight_to_fp16, x = hidden_states_335_cast_fp16)[name = tensor("v_169_cast_fp16")]; + tensor var_37795_begin_0 = const()[name = tensor("op_37795_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_37795_end_0 = const()[name = tensor("op_37795_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_37795_end_mask_0 = const()[name = tensor("op_37795_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37795_cast_fp16 = slice_by_index(begin = var_37795_begin_0, end = var_37795_end_0, end_mask = var_37795_end_mask_0, x = q_169_cast_fp16)[name = tensor("op_37795_cast_fp16")]; + tensor var_37799_begin_0 = const()[name = tensor("op_37799_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_37799_end_0 = const()[name = tensor("op_37799_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_37799_end_mask_0 = const()[name = tensor("op_37799_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37799_cast_fp16 = slice_by_index(begin = var_37799_begin_0, end = var_37799_end_0, end_mask = var_37799_end_mask_0, x = q_169_cast_fp16)[name = tensor("op_37799_cast_fp16")]; + tensor var_37803_begin_0 = const()[name = tensor("op_37803_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_37803_end_0 = const()[name = tensor("op_37803_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_37803_end_mask_0 = const()[name = tensor("op_37803_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37803_cast_fp16 = slice_by_index(begin = var_37803_begin_0, end = var_37803_end_0, end_mask = var_37803_end_mask_0, x = q_169_cast_fp16)[name = tensor("op_37803_cast_fp16")]; + tensor var_37807_begin_0 = const()[name = tensor("op_37807_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_37807_end_0 = const()[name = tensor("op_37807_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_37807_end_mask_0 = const()[name = tensor("op_37807_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37807_cast_fp16 = slice_by_index(begin = var_37807_begin_0, end = var_37807_end_0, end_mask = var_37807_end_mask_0, x = q_169_cast_fp16)[name = tensor("op_37807_cast_fp16")]; + tensor var_37811_begin_0 = const()[name = tensor("op_37811_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_37811_end_0 = const()[name = tensor("op_37811_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_37811_end_mask_0 = const()[name = tensor("op_37811_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37811_cast_fp16 = slice_by_index(begin = var_37811_begin_0, end = var_37811_end_0, end_mask = var_37811_end_mask_0, x = q_169_cast_fp16)[name = tensor("op_37811_cast_fp16")]; + tensor var_37815_begin_0 = const()[name = tensor("op_37815_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_37815_end_0 = const()[name = tensor("op_37815_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_37815_end_mask_0 = const()[name = tensor("op_37815_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37815_cast_fp16 = slice_by_index(begin = var_37815_begin_0, end = var_37815_end_0, end_mask = var_37815_end_mask_0, x = q_169_cast_fp16)[name = tensor("op_37815_cast_fp16")]; + tensor var_37819_begin_0 = const()[name = tensor("op_37819_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_37819_end_0 = const()[name = tensor("op_37819_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_37819_end_mask_0 = const()[name = tensor("op_37819_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37819_cast_fp16 = slice_by_index(begin = var_37819_begin_0, end = var_37819_end_0, end_mask = var_37819_end_mask_0, x = q_169_cast_fp16)[name = tensor("op_37819_cast_fp16")]; + tensor var_37823_begin_0 = const()[name = tensor("op_37823_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_37823_end_0 = const()[name = tensor("op_37823_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_37823_end_mask_0 = const()[name = tensor("op_37823_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37823_cast_fp16 = slice_by_index(begin = var_37823_begin_0, end = var_37823_end_0, end_mask = var_37823_end_mask_0, x = q_169_cast_fp16)[name = tensor("op_37823_cast_fp16")]; + tensor var_37827_begin_0 = const()[name = tensor("op_37827_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_37827_end_0 = const()[name = tensor("op_37827_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_37827_end_mask_0 = const()[name = tensor("op_37827_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37827_cast_fp16 = slice_by_index(begin = var_37827_begin_0, end = var_37827_end_0, end_mask = var_37827_end_mask_0, x = q_169_cast_fp16)[name = tensor("op_37827_cast_fp16")]; + tensor var_37831_begin_0 = const()[name = tensor("op_37831_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_37831_end_0 = const()[name = tensor("op_37831_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_37831_end_mask_0 = const()[name = tensor("op_37831_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37831_cast_fp16 = slice_by_index(begin = var_37831_begin_0, end = var_37831_end_0, end_mask = var_37831_end_mask_0, x = q_169_cast_fp16)[name = tensor("op_37831_cast_fp16")]; + tensor var_37835_begin_0 = const()[name = tensor("op_37835_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_37835_end_0 = const()[name = tensor("op_37835_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_37835_end_mask_0 = const()[name = tensor("op_37835_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37835_cast_fp16 = slice_by_index(begin = var_37835_begin_0, end = var_37835_end_0, end_mask = var_37835_end_mask_0, x = q_169_cast_fp16)[name = tensor("op_37835_cast_fp16")]; + tensor var_37839_begin_0 = const()[name = tensor("op_37839_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_37839_end_0 = const()[name = tensor("op_37839_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_37839_end_mask_0 = const()[name = tensor("op_37839_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37839_cast_fp16 = slice_by_index(begin = var_37839_begin_0, end = var_37839_end_0, end_mask = var_37839_end_mask_0, x = q_169_cast_fp16)[name = tensor("op_37839_cast_fp16")]; + tensor var_37843_begin_0 = const()[name = tensor("op_37843_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_37843_end_0 = const()[name = tensor("op_37843_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_37843_end_mask_0 = const()[name = tensor("op_37843_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37843_cast_fp16 = slice_by_index(begin = var_37843_begin_0, end = var_37843_end_0, end_mask = var_37843_end_mask_0, x = q_169_cast_fp16)[name = tensor("op_37843_cast_fp16")]; + tensor var_37847_begin_0 = const()[name = tensor("op_37847_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_37847_end_0 = const()[name = tensor("op_37847_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_37847_end_mask_0 = const()[name = tensor("op_37847_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37847_cast_fp16 = slice_by_index(begin = var_37847_begin_0, end = var_37847_end_0, end_mask = var_37847_end_mask_0, x = q_169_cast_fp16)[name = tensor("op_37847_cast_fp16")]; + tensor var_37851_begin_0 = const()[name = tensor("op_37851_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_37851_end_0 = const()[name = tensor("op_37851_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_37851_end_mask_0 = const()[name = tensor("op_37851_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37851_cast_fp16 = slice_by_index(begin = var_37851_begin_0, end = var_37851_end_0, end_mask = var_37851_end_mask_0, x = q_169_cast_fp16)[name = tensor("op_37851_cast_fp16")]; + tensor var_37855_begin_0 = const()[name = tensor("op_37855_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_37855_end_0 = const()[name = tensor("op_37855_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_37855_end_mask_0 = const()[name = tensor("op_37855_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37855_cast_fp16 = slice_by_index(begin = var_37855_begin_0, end = var_37855_end_0, end_mask = var_37855_end_mask_0, x = q_169_cast_fp16)[name = tensor("op_37855_cast_fp16")]; + tensor var_37859_begin_0 = const()[name = tensor("op_37859_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_37859_end_0 = const()[name = tensor("op_37859_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_37859_end_mask_0 = const()[name = tensor("op_37859_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37859_cast_fp16 = slice_by_index(begin = var_37859_begin_0, end = var_37859_end_0, end_mask = var_37859_end_mask_0, x = q_169_cast_fp16)[name = tensor("op_37859_cast_fp16")]; + tensor var_37863_begin_0 = const()[name = tensor("op_37863_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_37863_end_0 = const()[name = tensor("op_37863_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_37863_end_mask_0 = const()[name = tensor("op_37863_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37863_cast_fp16 = slice_by_index(begin = var_37863_begin_0, end = var_37863_end_0, end_mask = var_37863_end_mask_0, x = q_169_cast_fp16)[name = tensor("op_37863_cast_fp16")]; + tensor var_37867_begin_0 = const()[name = tensor("op_37867_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_37867_end_0 = const()[name = tensor("op_37867_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_37867_end_mask_0 = const()[name = tensor("op_37867_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37867_cast_fp16 = slice_by_index(begin = var_37867_begin_0, end = var_37867_end_0, end_mask = var_37867_end_mask_0, x = q_169_cast_fp16)[name = tensor("op_37867_cast_fp16")]; + tensor var_37871_begin_0 = const()[name = tensor("op_37871_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_37871_end_0 = const()[name = tensor("op_37871_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_37871_end_mask_0 = const()[name = tensor("op_37871_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37871_cast_fp16 = slice_by_index(begin = var_37871_begin_0, end = var_37871_end_0, end_mask = var_37871_end_mask_0, x = q_169_cast_fp16)[name = tensor("op_37871_cast_fp16")]; + tensor k_339_perm_0 = const()[name = tensor("k_339_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_37878_begin_0 = const()[name = tensor("op_37878_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_37878_end_0 = const()[name = tensor("op_37878_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_37878_end_mask_0 = const()[name = tensor("op_37878_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_339_cast_fp16 = transpose(perm = k_339_perm_0, x = k_337_cast_fp16)[name = tensor("transpose_55")]; + tensor var_37878_cast_fp16 = slice_by_index(begin = var_37878_begin_0, end = var_37878_end_0, end_mask = var_37878_end_mask_0, x = k_339_cast_fp16)[name = tensor("op_37878_cast_fp16")]; + tensor var_37882_begin_0 = const()[name = tensor("op_37882_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_37882_end_0 = const()[name = tensor("op_37882_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_37882_end_mask_0 = const()[name = tensor("op_37882_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37882_cast_fp16 = slice_by_index(begin = var_37882_begin_0, end = var_37882_end_0, end_mask = var_37882_end_mask_0, x = k_339_cast_fp16)[name = tensor("op_37882_cast_fp16")]; + tensor var_37886_begin_0 = const()[name = tensor("op_37886_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_37886_end_0 = const()[name = tensor("op_37886_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_37886_end_mask_0 = const()[name = tensor("op_37886_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37886_cast_fp16 = slice_by_index(begin = var_37886_begin_0, end = var_37886_end_0, end_mask = var_37886_end_mask_0, x = k_339_cast_fp16)[name = tensor("op_37886_cast_fp16")]; + tensor var_37890_begin_0 = const()[name = tensor("op_37890_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_37890_end_0 = const()[name = tensor("op_37890_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_37890_end_mask_0 = const()[name = tensor("op_37890_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37890_cast_fp16 = slice_by_index(begin = var_37890_begin_0, end = var_37890_end_0, end_mask = var_37890_end_mask_0, x = k_339_cast_fp16)[name = tensor("op_37890_cast_fp16")]; + tensor var_37894_begin_0 = const()[name = tensor("op_37894_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_37894_end_0 = const()[name = tensor("op_37894_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_37894_end_mask_0 = const()[name = tensor("op_37894_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37894_cast_fp16 = slice_by_index(begin = var_37894_begin_0, end = var_37894_end_0, end_mask = var_37894_end_mask_0, x = k_339_cast_fp16)[name = tensor("op_37894_cast_fp16")]; + tensor var_37898_begin_0 = const()[name = tensor("op_37898_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_37898_end_0 = const()[name = tensor("op_37898_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_37898_end_mask_0 = const()[name = tensor("op_37898_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37898_cast_fp16 = slice_by_index(begin = var_37898_begin_0, end = var_37898_end_0, end_mask = var_37898_end_mask_0, x = k_339_cast_fp16)[name = tensor("op_37898_cast_fp16")]; + tensor var_37902_begin_0 = const()[name = tensor("op_37902_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_37902_end_0 = const()[name = tensor("op_37902_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_37902_end_mask_0 = const()[name = tensor("op_37902_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37902_cast_fp16 = slice_by_index(begin = var_37902_begin_0, end = var_37902_end_0, end_mask = var_37902_end_mask_0, x = k_339_cast_fp16)[name = tensor("op_37902_cast_fp16")]; + tensor var_37906_begin_0 = const()[name = tensor("op_37906_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_37906_end_0 = const()[name = tensor("op_37906_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_37906_end_mask_0 = const()[name = tensor("op_37906_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37906_cast_fp16 = slice_by_index(begin = var_37906_begin_0, end = var_37906_end_0, end_mask = var_37906_end_mask_0, x = k_339_cast_fp16)[name = tensor("op_37906_cast_fp16")]; + tensor var_37910_begin_0 = const()[name = tensor("op_37910_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_37910_end_0 = const()[name = tensor("op_37910_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_37910_end_mask_0 = const()[name = tensor("op_37910_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37910_cast_fp16 = slice_by_index(begin = var_37910_begin_0, end = var_37910_end_0, end_mask = var_37910_end_mask_0, x = k_339_cast_fp16)[name = tensor("op_37910_cast_fp16")]; + tensor var_37914_begin_0 = const()[name = tensor("op_37914_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_37914_end_0 = const()[name = tensor("op_37914_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_37914_end_mask_0 = const()[name = tensor("op_37914_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37914_cast_fp16 = slice_by_index(begin = var_37914_begin_0, end = var_37914_end_0, end_mask = var_37914_end_mask_0, x = k_339_cast_fp16)[name = tensor("op_37914_cast_fp16")]; + tensor var_37918_begin_0 = const()[name = tensor("op_37918_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_37918_end_0 = const()[name = tensor("op_37918_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_37918_end_mask_0 = const()[name = tensor("op_37918_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37918_cast_fp16 = slice_by_index(begin = var_37918_begin_0, end = var_37918_end_0, end_mask = var_37918_end_mask_0, x = k_339_cast_fp16)[name = tensor("op_37918_cast_fp16")]; + tensor var_37922_begin_0 = const()[name = tensor("op_37922_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_37922_end_0 = const()[name = tensor("op_37922_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_37922_end_mask_0 = const()[name = tensor("op_37922_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37922_cast_fp16 = slice_by_index(begin = var_37922_begin_0, end = var_37922_end_0, end_mask = var_37922_end_mask_0, x = k_339_cast_fp16)[name = tensor("op_37922_cast_fp16")]; + tensor var_37926_begin_0 = const()[name = tensor("op_37926_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_37926_end_0 = const()[name = tensor("op_37926_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_37926_end_mask_0 = const()[name = tensor("op_37926_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37926_cast_fp16 = slice_by_index(begin = var_37926_begin_0, end = var_37926_end_0, end_mask = var_37926_end_mask_0, x = k_339_cast_fp16)[name = tensor("op_37926_cast_fp16")]; + tensor var_37930_begin_0 = const()[name = tensor("op_37930_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_37930_end_0 = const()[name = tensor("op_37930_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_37930_end_mask_0 = const()[name = tensor("op_37930_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37930_cast_fp16 = slice_by_index(begin = var_37930_begin_0, end = var_37930_end_0, end_mask = var_37930_end_mask_0, x = k_339_cast_fp16)[name = tensor("op_37930_cast_fp16")]; + tensor var_37934_begin_0 = const()[name = tensor("op_37934_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_37934_end_0 = const()[name = tensor("op_37934_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_37934_end_mask_0 = const()[name = tensor("op_37934_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37934_cast_fp16 = slice_by_index(begin = var_37934_begin_0, end = var_37934_end_0, end_mask = var_37934_end_mask_0, x = k_339_cast_fp16)[name = tensor("op_37934_cast_fp16")]; + tensor var_37938_begin_0 = const()[name = tensor("op_37938_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_37938_end_0 = const()[name = tensor("op_37938_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_37938_end_mask_0 = const()[name = tensor("op_37938_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37938_cast_fp16 = slice_by_index(begin = var_37938_begin_0, end = var_37938_end_0, end_mask = var_37938_end_mask_0, x = k_339_cast_fp16)[name = tensor("op_37938_cast_fp16")]; + tensor var_37942_begin_0 = const()[name = tensor("op_37942_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_37942_end_0 = const()[name = tensor("op_37942_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_37942_end_mask_0 = const()[name = tensor("op_37942_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37942_cast_fp16 = slice_by_index(begin = var_37942_begin_0, end = var_37942_end_0, end_mask = var_37942_end_mask_0, x = k_339_cast_fp16)[name = tensor("op_37942_cast_fp16")]; + tensor var_37946_begin_0 = const()[name = tensor("op_37946_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_37946_end_0 = const()[name = tensor("op_37946_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_37946_end_mask_0 = const()[name = tensor("op_37946_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37946_cast_fp16 = slice_by_index(begin = var_37946_begin_0, end = var_37946_end_0, end_mask = var_37946_end_mask_0, x = k_339_cast_fp16)[name = tensor("op_37946_cast_fp16")]; + tensor var_37950_begin_0 = const()[name = tensor("op_37950_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_37950_end_0 = const()[name = tensor("op_37950_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_37950_end_mask_0 = const()[name = tensor("op_37950_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37950_cast_fp16 = slice_by_index(begin = var_37950_begin_0, end = var_37950_end_0, end_mask = var_37950_end_mask_0, x = k_339_cast_fp16)[name = tensor("op_37950_cast_fp16")]; + tensor var_37954_begin_0 = const()[name = tensor("op_37954_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_37954_end_0 = const()[name = tensor("op_37954_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_37954_end_mask_0 = const()[name = tensor("op_37954_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37954_cast_fp16 = slice_by_index(begin = var_37954_begin_0, end = var_37954_end_0, end_mask = var_37954_end_mask_0, x = k_339_cast_fp16)[name = tensor("op_37954_cast_fp16")]; + tensor var_37956_begin_0 = const()[name = tensor("op_37956_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_37956_end_0 = const()[name = tensor("op_37956_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_37956_end_mask_0 = const()[name = tensor("op_37956_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37956_cast_fp16 = slice_by_index(begin = var_37956_begin_0, end = var_37956_end_0, end_mask = var_37956_end_mask_0, x = v_169_cast_fp16)[name = tensor("op_37956_cast_fp16")]; + tensor var_37960_begin_0 = const()[name = tensor("op_37960_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_37960_end_0 = const()[name = tensor("op_37960_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_37960_end_mask_0 = const()[name = tensor("op_37960_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37960_cast_fp16 = slice_by_index(begin = var_37960_begin_0, end = var_37960_end_0, end_mask = var_37960_end_mask_0, x = v_169_cast_fp16)[name = tensor("op_37960_cast_fp16")]; + tensor var_37964_begin_0 = const()[name = tensor("op_37964_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_37964_end_0 = const()[name = tensor("op_37964_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_37964_end_mask_0 = const()[name = tensor("op_37964_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37964_cast_fp16 = slice_by_index(begin = var_37964_begin_0, end = var_37964_end_0, end_mask = var_37964_end_mask_0, x = v_169_cast_fp16)[name = tensor("op_37964_cast_fp16")]; + tensor var_37968_begin_0 = const()[name = tensor("op_37968_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_37968_end_0 = const()[name = tensor("op_37968_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_37968_end_mask_0 = const()[name = tensor("op_37968_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37968_cast_fp16 = slice_by_index(begin = var_37968_begin_0, end = var_37968_end_0, end_mask = var_37968_end_mask_0, x = v_169_cast_fp16)[name = tensor("op_37968_cast_fp16")]; + tensor var_37972_begin_0 = const()[name = tensor("op_37972_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_37972_end_0 = const()[name = tensor("op_37972_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_37972_end_mask_0 = const()[name = tensor("op_37972_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37972_cast_fp16 = slice_by_index(begin = var_37972_begin_0, end = var_37972_end_0, end_mask = var_37972_end_mask_0, x = v_169_cast_fp16)[name = tensor("op_37972_cast_fp16")]; + tensor var_37976_begin_0 = const()[name = tensor("op_37976_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_37976_end_0 = const()[name = tensor("op_37976_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_37976_end_mask_0 = const()[name = tensor("op_37976_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37976_cast_fp16 = slice_by_index(begin = var_37976_begin_0, end = var_37976_end_0, end_mask = var_37976_end_mask_0, x = v_169_cast_fp16)[name = tensor("op_37976_cast_fp16")]; + tensor var_37980_begin_0 = const()[name = tensor("op_37980_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_37980_end_0 = const()[name = tensor("op_37980_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_37980_end_mask_0 = const()[name = tensor("op_37980_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37980_cast_fp16 = slice_by_index(begin = var_37980_begin_0, end = var_37980_end_0, end_mask = var_37980_end_mask_0, x = v_169_cast_fp16)[name = tensor("op_37980_cast_fp16")]; + tensor var_37984_begin_0 = const()[name = tensor("op_37984_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_37984_end_0 = const()[name = tensor("op_37984_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_37984_end_mask_0 = const()[name = tensor("op_37984_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37984_cast_fp16 = slice_by_index(begin = var_37984_begin_0, end = var_37984_end_0, end_mask = var_37984_end_mask_0, x = v_169_cast_fp16)[name = tensor("op_37984_cast_fp16")]; + tensor var_37988_begin_0 = const()[name = tensor("op_37988_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_37988_end_0 = const()[name = tensor("op_37988_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_37988_end_mask_0 = const()[name = tensor("op_37988_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37988_cast_fp16 = slice_by_index(begin = var_37988_begin_0, end = var_37988_end_0, end_mask = var_37988_end_mask_0, x = v_169_cast_fp16)[name = tensor("op_37988_cast_fp16")]; + tensor var_37992_begin_0 = const()[name = tensor("op_37992_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_37992_end_0 = const()[name = tensor("op_37992_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_37992_end_mask_0 = const()[name = tensor("op_37992_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37992_cast_fp16 = slice_by_index(begin = var_37992_begin_0, end = var_37992_end_0, end_mask = var_37992_end_mask_0, x = v_169_cast_fp16)[name = tensor("op_37992_cast_fp16")]; + tensor var_37996_begin_0 = const()[name = tensor("op_37996_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_37996_end_0 = const()[name = tensor("op_37996_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_37996_end_mask_0 = const()[name = tensor("op_37996_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37996_cast_fp16 = slice_by_index(begin = var_37996_begin_0, end = var_37996_end_0, end_mask = var_37996_end_mask_0, x = v_169_cast_fp16)[name = tensor("op_37996_cast_fp16")]; + tensor var_38000_begin_0 = const()[name = tensor("op_38000_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_38000_end_0 = const()[name = tensor("op_38000_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_38000_end_mask_0 = const()[name = tensor("op_38000_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38000_cast_fp16 = slice_by_index(begin = var_38000_begin_0, end = var_38000_end_0, end_mask = var_38000_end_mask_0, x = v_169_cast_fp16)[name = tensor("op_38000_cast_fp16")]; + tensor var_38004_begin_0 = const()[name = tensor("op_38004_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_38004_end_0 = const()[name = tensor("op_38004_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_38004_end_mask_0 = const()[name = tensor("op_38004_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38004_cast_fp16 = slice_by_index(begin = var_38004_begin_0, end = var_38004_end_0, end_mask = var_38004_end_mask_0, x = v_169_cast_fp16)[name = tensor("op_38004_cast_fp16")]; + tensor var_38008_begin_0 = const()[name = tensor("op_38008_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_38008_end_0 = const()[name = tensor("op_38008_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_38008_end_mask_0 = const()[name = tensor("op_38008_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38008_cast_fp16 = slice_by_index(begin = var_38008_begin_0, end = var_38008_end_0, end_mask = var_38008_end_mask_0, x = v_169_cast_fp16)[name = tensor("op_38008_cast_fp16")]; + tensor var_38012_begin_0 = const()[name = tensor("op_38012_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_38012_end_0 = const()[name = tensor("op_38012_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_38012_end_mask_0 = const()[name = tensor("op_38012_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38012_cast_fp16 = slice_by_index(begin = var_38012_begin_0, end = var_38012_end_0, end_mask = var_38012_end_mask_0, x = v_169_cast_fp16)[name = tensor("op_38012_cast_fp16")]; + tensor var_38016_begin_0 = const()[name = tensor("op_38016_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_38016_end_0 = const()[name = tensor("op_38016_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_38016_end_mask_0 = const()[name = tensor("op_38016_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38016_cast_fp16 = slice_by_index(begin = var_38016_begin_0, end = var_38016_end_0, end_mask = var_38016_end_mask_0, x = v_169_cast_fp16)[name = tensor("op_38016_cast_fp16")]; + tensor var_38020_begin_0 = const()[name = tensor("op_38020_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_38020_end_0 = const()[name = tensor("op_38020_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_38020_end_mask_0 = const()[name = tensor("op_38020_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38020_cast_fp16 = slice_by_index(begin = var_38020_begin_0, end = var_38020_end_0, end_mask = var_38020_end_mask_0, x = v_169_cast_fp16)[name = tensor("op_38020_cast_fp16")]; + tensor var_38024_begin_0 = const()[name = tensor("op_38024_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_38024_end_0 = const()[name = tensor("op_38024_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_38024_end_mask_0 = const()[name = tensor("op_38024_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38024_cast_fp16 = slice_by_index(begin = var_38024_begin_0, end = var_38024_end_0, end_mask = var_38024_end_mask_0, x = v_169_cast_fp16)[name = tensor("op_38024_cast_fp16")]; + tensor var_38028_begin_0 = const()[name = tensor("op_38028_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_38028_end_0 = const()[name = tensor("op_38028_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_38028_end_mask_0 = const()[name = tensor("op_38028_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38028_cast_fp16 = slice_by_index(begin = var_38028_begin_0, end = var_38028_end_0, end_mask = var_38028_end_mask_0, x = v_169_cast_fp16)[name = tensor("op_38028_cast_fp16")]; + tensor var_38032_begin_0 = const()[name = tensor("op_38032_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_38032_end_0 = const()[name = tensor("op_38032_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_38032_end_mask_0 = const()[name = tensor("op_38032_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38032_cast_fp16 = slice_by_index(begin = var_38032_begin_0, end = var_38032_end_0, end_mask = var_38032_end_mask_0, x = v_169_cast_fp16)[name = tensor("op_38032_cast_fp16")]; + tensor var_38036_equation_0 = const()[name = tensor("op_38036_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38036_cast_fp16 = einsum(equation = var_38036_equation_0, values = (var_37878_cast_fp16, var_37795_cast_fp16))[name = tensor("op_38036_cast_fp16")]; + tensor var_38037_to_fp16 = const()[name = tensor("op_38037_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3201_cast_fp16 = mul(x = var_38036_cast_fp16, y = var_38037_to_fp16)[name = tensor("aw_3201_cast_fp16")]; + tensor var_38040_equation_0 = const()[name = tensor("op_38040_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38040_cast_fp16 = einsum(equation = var_38040_equation_0, values = (var_37882_cast_fp16, var_37799_cast_fp16))[name = tensor("op_38040_cast_fp16")]; + tensor var_38041_to_fp16 = const()[name = tensor("op_38041_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3203_cast_fp16 = mul(x = var_38040_cast_fp16, y = var_38041_to_fp16)[name = tensor("aw_3203_cast_fp16")]; + tensor var_38044_equation_0 = const()[name = tensor("op_38044_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38044_cast_fp16 = einsum(equation = var_38044_equation_0, values = (var_37886_cast_fp16, var_37803_cast_fp16))[name = tensor("op_38044_cast_fp16")]; + tensor var_38045_to_fp16 = const()[name = tensor("op_38045_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3205_cast_fp16 = mul(x = var_38044_cast_fp16, y = var_38045_to_fp16)[name = tensor("aw_3205_cast_fp16")]; + tensor var_38048_equation_0 = const()[name = tensor("op_38048_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38048_cast_fp16 = einsum(equation = var_38048_equation_0, values = (var_37890_cast_fp16, var_37807_cast_fp16))[name = tensor("op_38048_cast_fp16")]; + tensor var_38049_to_fp16 = const()[name = tensor("op_38049_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3207_cast_fp16 = mul(x = var_38048_cast_fp16, y = var_38049_to_fp16)[name = tensor("aw_3207_cast_fp16")]; + tensor var_38052_equation_0 = const()[name = tensor("op_38052_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38052_cast_fp16 = einsum(equation = var_38052_equation_0, values = (var_37894_cast_fp16, var_37811_cast_fp16))[name = tensor("op_38052_cast_fp16")]; + tensor var_38053_to_fp16 = const()[name = tensor("op_38053_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3209_cast_fp16 = mul(x = var_38052_cast_fp16, y = var_38053_to_fp16)[name = tensor("aw_3209_cast_fp16")]; + tensor var_38056_equation_0 = const()[name = tensor("op_38056_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38056_cast_fp16 = einsum(equation = var_38056_equation_0, values = (var_37898_cast_fp16, var_37815_cast_fp16))[name = tensor("op_38056_cast_fp16")]; + tensor var_38057_to_fp16 = const()[name = tensor("op_38057_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3211_cast_fp16 = mul(x = var_38056_cast_fp16, y = var_38057_to_fp16)[name = tensor("aw_3211_cast_fp16")]; + tensor var_38060_equation_0 = const()[name = tensor("op_38060_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38060_cast_fp16 = einsum(equation = var_38060_equation_0, values = (var_37902_cast_fp16, var_37819_cast_fp16))[name = tensor("op_38060_cast_fp16")]; + tensor var_38061_to_fp16 = const()[name = tensor("op_38061_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3213_cast_fp16 = mul(x = var_38060_cast_fp16, y = var_38061_to_fp16)[name = tensor("aw_3213_cast_fp16")]; + tensor var_38064_equation_0 = const()[name = tensor("op_38064_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38064_cast_fp16 = einsum(equation = var_38064_equation_0, values = (var_37906_cast_fp16, var_37823_cast_fp16))[name = tensor("op_38064_cast_fp16")]; + tensor var_38065_to_fp16 = const()[name = tensor("op_38065_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3215_cast_fp16 = mul(x = var_38064_cast_fp16, y = var_38065_to_fp16)[name = tensor("aw_3215_cast_fp16")]; + tensor var_38068_equation_0 = const()[name = tensor("op_38068_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38068_cast_fp16 = einsum(equation = var_38068_equation_0, values = (var_37910_cast_fp16, var_37827_cast_fp16))[name = tensor("op_38068_cast_fp16")]; + tensor var_38069_to_fp16 = const()[name = tensor("op_38069_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3217_cast_fp16 = mul(x = var_38068_cast_fp16, y = var_38069_to_fp16)[name = tensor("aw_3217_cast_fp16")]; + tensor var_38072_equation_0 = const()[name = tensor("op_38072_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38072_cast_fp16 = einsum(equation = var_38072_equation_0, values = (var_37914_cast_fp16, var_37831_cast_fp16))[name = tensor("op_38072_cast_fp16")]; + tensor var_38073_to_fp16 = const()[name = tensor("op_38073_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3219_cast_fp16 = mul(x = var_38072_cast_fp16, y = var_38073_to_fp16)[name = tensor("aw_3219_cast_fp16")]; + tensor var_38076_equation_0 = const()[name = tensor("op_38076_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38076_cast_fp16 = einsum(equation = var_38076_equation_0, values = (var_37918_cast_fp16, var_37835_cast_fp16))[name = tensor("op_38076_cast_fp16")]; + tensor var_38077_to_fp16 = const()[name = tensor("op_38077_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3221_cast_fp16 = mul(x = var_38076_cast_fp16, y = var_38077_to_fp16)[name = tensor("aw_3221_cast_fp16")]; + tensor var_38080_equation_0 = const()[name = tensor("op_38080_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38080_cast_fp16 = einsum(equation = var_38080_equation_0, values = (var_37922_cast_fp16, var_37839_cast_fp16))[name = tensor("op_38080_cast_fp16")]; + tensor var_38081_to_fp16 = const()[name = tensor("op_38081_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3223_cast_fp16 = mul(x = var_38080_cast_fp16, y = var_38081_to_fp16)[name = tensor("aw_3223_cast_fp16")]; + tensor var_38084_equation_0 = const()[name = tensor("op_38084_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38084_cast_fp16 = einsum(equation = var_38084_equation_0, values = (var_37926_cast_fp16, var_37843_cast_fp16))[name = tensor("op_38084_cast_fp16")]; + tensor var_38085_to_fp16 = const()[name = tensor("op_38085_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3225_cast_fp16 = mul(x = var_38084_cast_fp16, y = var_38085_to_fp16)[name = tensor("aw_3225_cast_fp16")]; + tensor var_38088_equation_0 = const()[name = tensor("op_38088_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38088_cast_fp16 = einsum(equation = var_38088_equation_0, values = (var_37930_cast_fp16, var_37847_cast_fp16))[name = tensor("op_38088_cast_fp16")]; + tensor var_38089_to_fp16 = const()[name = tensor("op_38089_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3227_cast_fp16 = mul(x = var_38088_cast_fp16, y = var_38089_to_fp16)[name = tensor("aw_3227_cast_fp16")]; + tensor var_38092_equation_0 = const()[name = tensor("op_38092_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38092_cast_fp16 = einsum(equation = var_38092_equation_0, values = (var_37934_cast_fp16, var_37851_cast_fp16))[name = tensor("op_38092_cast_fp16")]; + tensor var_38093_to_fp16 = const()[name = tensor("op_38093_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3229_cast_fp16 = mul(x = var_38092_cast_fp16, y = var_38093_to_fp16)[name = tensor("aw_3229_cast_fp16")]; + tensor var_38096_equation_0 = const()[name = tensor("op_38096_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38096_cast_fp16 = einsum(equation = var_38096_equation_0, values = (var_37938_cast_fp16, var_37855_cast_fp16))[name = tensor("op_38096_cast_fp16")]; + tensor var_38097_to_fp16 = const()[name = tensor("op_38097_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3231_cast_fp16 = mul(x = var_38096_cast_fp16, y = var_38097_to_fp16)[name = tensor("aw_3231_cast_fp16")]; + tensor var_38100_equation_0 = const()[name = tensor("op_38100_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38100_cast_fp16 = einsum(equation = var_38100_equation_0, values = (var_37942_cast_fp16, var_37859_cast_fp16))[name = tensor("op_38100_cast_fp16")]; + tensor var_38101_to_fp16 = const()[name = tensor("op_38101_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3233_cast_fp16 = mul(x = var_38100_cast_fp16, y = var_38101_to_fp16)[name = tensor("aw_3233_cast_fp16")]; + tensor var_38104_equation_0 = const()[name = tensor("op_38104_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38104_cast_fp16 = einsum(equation = var_38104_equation_0, values = (var_37946_cast_fp16, var_37863_cast_fp16))[name = tensor("op_38104_cast_fp16")]; + tensor var_38105_to_fp16 = const()[name = tensor("op_38105_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3235_cast_fp16 = mul(x = var_38104_cast_fp16, y = var_38105_to_fp16)[name = tensor("aw_3235_cast_fp16")]; + tensor var_38108_equation_0 = const()[name = tensor("op_38108_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38108_cast_fp16 = einsum(equation = var_38108_equation_0, values = (var_37950_cast_fp16, var_37867_cast_fp16))[name = tensor("op_38108_cast_fp16")]; + tensor var_38109_to_fp16 = const()[name = tensor("op_38109_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3237_cast_fp16 = mul(x = var_38108_cast_fp16, y = var_38109_to_fp16)[name = tensor("aw_3237_cast_fp16")]; + tensor var_38112_equation_0 = const()[name = tensor("op_38112_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38112_cast_fp16 = einsum(equation = var_38112_equation_0, values = (var_37954_cast_fp16, var_37871_cast_fp16))[name = tensor("op_38112_cast_fp16")]; + tensor var_38113_to_fp16 = const()[name = tensor("op_38113_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3239_cast_fp16 = mul(x = var_38112_cast_fp16, y = var_38113_to_fp16)[name = tensor("aw_3239_cast_fp16")]; + tensor var_38115_cast_fp16 = softmax(axis = var_30355, x = aw_3201_cast_fp16)[name = tensor("op_38115_cast_fp16")]; + tensor var_38116_cast_fp16 = softmax(axis = var_30355, x = aw_3203_cast_fp16)[name = tensor("op_38116_cast_fp16")]; + tensor var_38117_cast_fp16 = softmax(axis = var_30355, x = aw_3205_cast_fp16)[name = tensor("op_38117_cast_fp16")]; + tensor var_38118_cast_fp16 = softmax(axis = var_30355, x = aw_3207_cast_fp16)[name = tensor("op_38118_cast_fp16")]; + tensor var_38119_cast_fp16 = softmax(axis = var_30355, x = aw_3209_cast_fp16)[name = tensor("op_38119_cast_fp16")]; + tensor var_38120_cast_fp16 = softmax(axis = var_30355, x = aw_3211_cast_fp16)[name = tensor("op_38120_cast_fp16")]; + tensor var_38121_cast_fp16 = softmax(axis = var_30355, x = aw_3213_cast_fp16)[name = tensor("op_38121_cast_fp16")]; + tensor var_38122_cast_fp16 = softmax(axis = var_30355, x = aw_3215_cast_fp16)[name = tensor("op_38122_cast_fp16")]; + tensor var_38123_cast_fp16 = softmax(axis = var_30355, x = aw_3217_cast_fp16)[name = tensor("op_38123_cast_fp16")]; + tensor var_38124_cast_fp16 = softmax(axis = var_30355, x = aw_3219_cast_fp16)[name = tensor("op_38124_cast_fp16")]; + tensor var_38125_cast_fp16 = softmax(axis = var_30355, x = aw_3221_cast_fp16)[name = tensor("op_38125_cast_fp16")]; + tensor var_38126_cast_fp16 = softmax(axis = var_30355, x = aw_3223_cast_fp16)[name = tensor("op_38126_cast_fp16")]; + tensor var_38127_cast_fp16 = softmax(axis = var_30355, x = aw_3225_cast_fp16)[name = tensor("op_38127_cast_fp16")]; + tensor var_38128_cast_fp16 = softmax(axis = var_30355, x = aw_3227_cast_fp16)[name = tensor("op_38128_cast_fp16")]; + tensor var_38129_cast_fp16 = softmax(axis = var_30355, x = aw_3229_cast_fp16)[name = tensor("op_38129_cast_fp16")]; + tensor var_38130_cast_fp16 = softmax(axis = var_30355, x = aw_3231_cast_fp16)[name = tensor("op_38130_cast_fp16")]; + tensor var_38131_cast_fp16 = softmax(axis = var_30355, x = aw_3233_cast_fp16)[name = tensor("op_38131_cast_fp16")]; + tensor var_38132_cast_fp16 = softmax(axis = var_30355, x = aw_3235_cast_fp16)[name = tensor("op_38132_cast_fp16")]; + tensor var_38133_cast_fp16 = softmax(axis = var_30355, x = aw_3237_cast_fp16)[name = tensor("op_38133_cast_fp16")]; + tensor var_38134_cast_fp16 = softmax(axis = var_30355, x = aw_3239_cast_fp16)[name = tensor("op_38134_cast_fp16")]; + tensor var_38136_equation_0 = const()[name = tensor("op_38136_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_38136_cast_fp16 = einsum(equation = var_38136_equation_0, values = (var_37956_cast_fp16, var_38115_cast_fp16))[name = tensor("op_38136_cast_fp16")]; + tensor var_38138_equation_0 = const()[name = tensor("op_38138_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_38138_cast_fp16 = einsum(equation = var_38138_equation_0, values = (var_37960_cast_fp16, var_38116_cast_fp16))[name = tensor("op_38138_cast_fp16")]; + tensor var_38140_equation_0 = const()[name = tensor("op_38140_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_38140_cast_fp16 = einsum(equation = var_38140_equation_0, values = (var_37964_cast_fp16, var_38117_cast_fp16))[name = tensor("op_38140_cast_fp16")]; + tensor var_38142_equation_0 = const()[name = tensor("op_38142_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_38142_cast_fp16 = einsum(equation = var_38142_equation_0, values = (var_37968_cast_fp16, var_38118_cast_fp16))[name = tensor("op_38142_cast_fp16")]; + tensor var_38144_equation_0 = const()[name = tensor("op_38144_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_38144_cast_fp16 = einsum(equation = var_38144_equation_0, values = (var_37972_cast_fp16, var_38119_cast_fp16))[name = tensor("op_38144_cast_fp16")]; + tensor var_38146_equation_0 = const()[name = tensor("op_38146_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_38146_cast_fp16 = einsum(equation = var_38146_equation_0, values = (var_37976_cast_fp16, var_38120_cast_fp16))[name = tensor("op_38146_cast_fp16")]; + tensor var_38148_equation_0 = const()[name = tensor("op_38148_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_38148_cast_fp16 = einsum(equation = var_38148_equation_0, values = (var_37980_cast_fp16, var_38121_cast_fp16))[name = tensor("op_38148_cast_fp16")]; + tensor var_38150_equation_0 = const()[name = tensor("op_38150_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_38150_cast_fp16 = einsum(equation = var_38150_equation_0, values = (var_37984_cast_fp16, var_38122_cast_fp16))[name = tensor("op_38150_cast_fp16")]; + tensor var_38152_equation_0 = const()[name = tensor("op_38152_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_38152_cast_fp16 = einsum(equation = var_38152_equation_0, values = (var_37988_cast_fp16, var_38123_cast_fp16))[name = tensor("op_38152_cast_fp16")]; + tensor var_38154_equation_0 = const()[name = tensor("op_38154_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_38154_cast_fp16 = einsum(equation = var_38154_equation_0, values = (var_37992_cast_fp16, var_38124_cast_fp16))[name = tensor("op_38154_cast_fp16")]; + tensor var_38156_equation_0 = const()[name = tensor("op_38156_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_38156_cast_fp16 = einsum(equation = var_38156_equation_0, values = (var_37996_cast_fp16, var_38125_cast_fp16))[name = tensor("op_38156_cast_fp16")]; + tensor var_38158_equation_0 = const()[name = tensor("op_38158_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_38158_cast_fp16 = einsum(equation = var_38158_equation_0, values = (var_38000_cast_fp16, var_38126_cast_fp16))[name = tensor("op_38158_cast_fp16")]; + tensor var_38160_equation_0 = const()[name = tensor("op_38160_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_38160_cast_fp16 = einsum(equation = var_38160_equation_0, values = (var_38004_cast_fp16, var_38127_cast_fp16))[name = tensor("op_38160_cast_fp16")]; + tensor var_38162_equation_0 = const()[name = tensor("op_38162_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_38162_cast_fp16 = einsum(equation = var_38162_equation_0, values = (var_38008_cast_fp16, var_38128_cast_fp16))[name = tensor("op_38162_cast_fp16")]; + tensor var_38164_equation_0 = const()[name = tensor("op_38164_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_38164_cast_fp16 = einsum(equation = var_38164_equation_0, values = (var_38012_cast_fp16, var_38129_cast_fp16))[name = tensor("op_38164_cast_fp16")]; + tensor var_38166_equation_0 = const()[name = tensor("op_38166_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_38166_cast_fp16 = einsum(equation = var_38166_equation_0, values = (var_38016_cast_fp16, var_38130_cast_fp16))[name = tensor("op_38166_cast_fp16")]; + tensor var_38168_equation_0 = const()[name = tensor("op_38168_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_38168_cast_fp16 = einsum(equation = var_38168_equation_0, values = (var_38020_cast_fp16, var_38131_cast_fp16))[name = tensor("op_38168_cast_fp16")]; + tensor var_38170_equation_0 = const()[name = tensor("op_38170_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_38170_cast_fp16 = einsum(equation = var_38170_equation_0, values = (var_38024_cast_fp16, var_38132_cast_fp16))[name = tensor("op_38170_cast_fp16")]; + tensor var_38172_equation_0 = const()[name = tensor("op_38172_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_38172_cast_fp16 = einsum(equation = var_38172_equation_0, values = (var_38028_cast_fp16, var_38133_cast_fp16))[name = tensor("op_38172_cast_fp16")]; + tensor var_38174_equation_0 = const()[name = tensor("op_38174_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_38174_cast_fp16 = einsum(equation = var_38174_equation_0, values = (var_38032_cast_fp16, var_38134_cast_fp16))[name = tensor("op_38174_cast_fp16")]; + tensor input_503_interleave_0 = const()[name = tensor("input_503_interleave_0"), val = tensor(false)]; + tensor input_503_cast_fp16 = concat(axis = var_30355, interleave = input_503_interleave_0, values = (var_38136_cast_fp16, var_38138_cast_fp16, var_38140_cast_fp16, var_38142_cast_fp16, var_38144_cast_fp16, var_38146_cast_fp16, var_38148_cast_fp16, var_38150_cast_fp16, var_38152_cast_fp16, var_38154_cast_fp16, var_38156_cast_fp16, var_38158_cast_fp16, var_38160_cast_fp16, var_38162_cast_fp16, var_38164_cast_fp16, var_38166_cast_fp16, var_38168_cast_fp16, var_38170_cast_fp16, var_38172_cast_fp16, var_38174_cast_fp16))[name = tensor("input_503_cast_fp16")]; + tensor var_38180 = const()[name = tensor("op_38180"), val = tensor([1, 1])]; + tensor var_38182 = const()[name = tensor("op_38182"), val = tensor([1, 1])]; + tensor var_38184_pad_type_0 = const()[name = tensor("op_38184_pad_type_0"), val = tensor("custom")]; + tensor var_38184_pad_0 = const()[name = tensor("op_38184_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_8_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_8_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3154360896)))]; + tensor up_blocks_0_attentions_0_transformer_blocks_8_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_8_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3157637760)))]; + tensor var_38184_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_8_attn1_to_out_0_bias_to_fp16, dilations = var_38182, groups = var_30355, pad = var_38184_pad_0, pad_type = var_38184_pad_type_0, strides = var_38180, weight = up_blocks_0_attentions_0_transformer_blocks_8_attn1_to_out_0_weight_to_fp16, x = input_503_cast_fp16)[name = tensor("op_38184_cast_fp16")]; + tensor inputs_255_cast_fp16 = add(x = var_38184_cast_fp16, y = inputs_253_cast_fp16)[name = tensor("inputs_255_cast_fp16")]; + tensor hidden_states_337_axes_0 = const()[name = tensor("hidden_states_337_axes_0"), val = tensor([1])]; + tensor hidden_states_337_gamma_0_to_fp16 = const()[name = tensor("hidden_states_337_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3157640384)))]; + tensor hidden_states_337_beta_0_to_fp16 = const()[name = tensor("hidden_states_337_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3157643008)))]; + tensor var_38194_to_fp16 = const()[name = tensor("op_38194_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_337_cast_fp16 = layer_norm(axes = hidden_states_337_axes_0, beta = hidden_states_337_beta_0_to_fp16, epsilon = var_38194_to_fp16, gamma = hidden_states_337_gamma_0_to_fp16, x = inputs_255_cast_fp16)[name = tensor("hidden_states_337_cast_fp16")]; + tensor var_38209 = const()[name = tensor("op_38209"), val = tensor([1, 1])]; + tensor var_38211 = const()[name = tensor("op_38211"), val = tensor([1, 1])]; + tensor q_171_pad_type_0 = const()[name = tensor("q_171_pad_type_0"), val = tensor("custom")]; + tensor q_171_pad_0 = const()[name = tensor("q_171_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_8_attn2_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_8_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3157645632)))]; + tensor q_171_cast_fp16 = conv(dilations = var_38211, groups = var_30355, pad = q_171_pad_0, pad_type = q_171_pad_type_0, strides = var_38209, weight = up_blocks_0_attentions_0_transformer_blocks_8_attn2_to_q_weight_to_fp16, x = hidden_states_337_cast_fp16)[name = tensor("q_171_cast_fp16")]; + tensor var_38215 = const()[name = tensor("op_38215"), val = tensor([1, 1])]; + tensor var_38217 = const()[name = tensor("op_38217"), val = tensor([1, 1])]; + tensor k_341_pad_type_0 = const()[name = tensor("k_341_pad_type_0"), val = tensor("custom")]; + tensor k_341_pad_0 = const()[name = tensor("k_341_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_8_attn2_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_8_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3160922496)))]; + tensor k_341_cast_fp16 = conv(dilations = var_38217, groups = var_30355, pad = k_341_pad_0, pad_type = k_341_pad_type_0, strides = var_38215, weight = up_blocks_0_attentions_0_transformer_blocks_8_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_341_cast_fp16")]; + tensor var_38221 = const()[name = tensor("op_38221"), val = tensor([1, 1])]; + tensor var_38223 = const()[name = tensor("op_38223"), val = tensor([1, 1])]; + tensor v_171_pad_type_0 = const()[name = tensor("v_171_pad_type_0"), val = tensor("custom")]; + tensor v_171_pad_0 = const()[name = tensor("v_171_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_8_attn2_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_8_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3166165440)))]; + tensor v_171_cast_fp16 = conv(dilations = var_38223, groups = var_30355, pad = v_171_pad_0, pad_type = v_171_pad_type_0, strides = var_38221, weight = up_blocks_0_attentions_0_transformer_blocks_8_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_171_cast_fp16")]; + tensor var_38227_begin_0 = const()[name = tensor("op_38227_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_38227_end_0 = const()[name = tensor("op_38227_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_38227_end_mask_0 = const()[name = tensor("op_38227_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38227_cast_fp16 = slice_by_index(begin = var_38227_begin_0, end = var_38227_end_0, end_mask = var_38227_end_mask_0, x = q_171_cast_fp16)[name = tensor("op_38227_cast_fp16")]; + tensor var_38231_begin_0 = const()[name = tensor("op_38231_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_38231_end_0 = const()[name = tensor("op_38231_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_38231_end_mask_0 = const()[name = tensor("op_38231_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38231_cast_fp16 = slice_by_index(begin = var_38231_begin_0, end = var_38231_end_0, end_mask = var_38231_end_mask_0, x = q_171_cast_fp16)[name = tensor("op_38231_cast_fp16")]; + tensor var_38235_begin_0 = const()[name = tensor("op_38235_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_38235_end_0 = const()[name = tensor("op_38235_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_38235_end_mask_0 = const()[name = tensor("op_38235_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38235_cast_fp16 = slice_by_index(begin = var_38235_begin_0, end = var_38235_end_0, end_mask = var_38235_end_mask_0, x = q_171_cast_fp16)[name = tensor("op_38235_cast_fp16")]; + tensor var_38239_begin_0 = const()[name = tensor("op_38239_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_38239_end_0 = const()[name = tensor("op_38239_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_38239_end_mask_0 = const()[name = tensor("op_38239_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38239_cast_fp16 = slice_by_index(begin = var_38239_begin_0, end = var_38239_end_0, end_mask = var_38239_end_mask_0, x = q_171_cast_fp16)[name = tensor("op_38239_cast_fp16")]; + tensor var_38243_begin_0 = const()[name = tensor("op_38243_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_38243_end_0 = const()[name = tensor("op_38243_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_38243_end_mask_0 = const()[name = tensor("op_38243_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38243_cast_fp16 = slice_by_index(begin = var_38243_begin_0, end = var_38243_end_0, end_mask = var_38243_end_mask_0, x = q_171_cast_fp16)[name = tensor("op_38243_cast_fp16")]; + tensor var_38247_begin_0 = const()[name = tensor("op_38247_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_38247_end_0 = const()[name = tensor("op_38247_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_38247_end_mask_0 = const()[name = tensor("op_38247_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38247_cast_fp16 = slice_by_index(begin = var_38247_begin_0, end = var_38247_end_0, end_mask = var_38247_end_mask_0, x = q_171_cast_fp16)[name = tensor("op_38247_cast_fp16")]; + tensor var_38251_begin_0 = const()[name = tensor("op_38251_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_38251_end_0 = const()[name = tensor("op_38251_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_38251_end_mask_0 = const()[name = tensor("op_38251_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38251_cast_fp16 = slice_by_index(begin = var_38251_begin_0, end = var_38251_end_0, end_mask = var_38251_end_mask_0, x = q_171_cast_fp16)[name = tensor("op_38251_cast_fp16")]; + tensor var_38255_begin_0 = const()[name = tensor("op_38255_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_38255_end_0 = const()[name = tensor("op_38255_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_38255_end_mask_0 = const()[name = tensor("op_38255_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38255_cast_fp16 = slice_by_index(begin = var_38255_begin_0, end = var_38255_end_0, end_mask = var_38255_end_mask_0, x = q_171_cast_fp16)[name = tensor("op_38255_cast_fp16")]; + tensor var_38259_begin_0 = const()[name = tensor("op_38259_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_38259_end_0 = const()[name = tensor("op_38259_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_38259_end_mask_0 = const()[name = tensor("op_38259_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38259_cast_fp16 = slice_by_index(begin = var_38259_begin_0, end = var_38259_end_0, end_mask = var_38259_end_mask_0, x = q_171_cast_fp16)[name = tensor("op_38259_cast_fp16")]; + tensor var_38263_begin_0 = const()[name = tensor("op_38263_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_38263_end_0 = const()[name = tensor("op_38263_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_38263_end_mask_0 = const()[name = tensor("op_38263_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38263_cast_fp16 = slice_by_index(begin = var_38263_begin_0, end = var_38263_end_0, end_mask = var_38263_end_mask_0, x = q_171_cast_fp16)[name = tensor("op_38263_cast_fp16")]; + tensor var_38267_begin_0 = const()[name = tensor("op_38267_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_38267_end_0 = const()[name = tensor("op_38267_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_38267_end_mask_0 = const()[name = tensor("op_38267_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38267_cast_fp16 = slice_by_index(begin = var_38267_begin_0, end = var_38267_end_0, end_mask = var_38267_end_mask_0, x = q_171_cast_fp16)[name = tensor("op_38267_cast_fp16")]; + tensor var_38271_begin_0 = const()[name = tensor("op_38271_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_38271_end_0 = const()[name = tensor("op_38271_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_38271_end_mask_0 = const()[name = tensor("op_38271_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38271_cast_fp16 = slice_by_index(begin = var_38271_begin_0, end = var_38271_end_0, end_mask = var_38271_end_mask_0, x = q_171_cast_fp16)[name = tensor("op_38271_cast_fp16")]; + tensor var_38275_begin_0 = const()[name = tensor("op_38275_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_38275_end_0 = const()[name = tensor("op_38275_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_38275_end_mask_0 = const()[name = tensor("op_38275_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38275_cast_fp16 = slice_by_index(begin = var_38275_begin_0, end = var_38275_end_0, end_mask = var_38275_end_mask_0, x = q_171_cast_fp16)[name = tensor("op_38275_cast_fp16")]; + tensor var_38279_begin_0 = const()[name = tensor("op_38279_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_38279_end_0 = const()[name = tensor("op_38279_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_38279_end_mask_0 = const()[name = tensor("op_38279_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38279_cast_fp16 = slice_by_index(begin = var_38279_begin_0, end = var_38279_end_0, end_mask = var_38279_end_mask_0, x = q_171_cast_fp16)[name = tensor("op_38279_cast_fp16")]; + tensor var_38283_begin_0 = const()[name = tensor("op_38283_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_38283_end_0 = const()[name = tensor("op_38283_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_38283_end_mask_0 = const()[name = tensor("op_38283_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38283_cast_fp16 = slice_by_index(begin = var_38283_begin_0, end = var_38283_end_0, end_mask = var_38283_end_mask_0, x = q_171_cast_fp16)[name = tensor("op_38283_cast_fp16")]; + tensor var_38287_begin_0 = const()[name = tensor("op_38287_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_38287_end_0 = const()[name = tensor("op_38287_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_38287_end_mask_0 = const()[name = tensor("op_38287_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38287_cast_fp16 = slice_by_index(begin = var_38287_begin_0, end = var_38287_end_0, end_mask = var_38287_end_mask_0, x = q_171_cast_fp16)[name = tensor("op_38287_cast_fp16")]; + tensor var_38291_begin_0 = const()[name = tensor("op_38291_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_38291_end_0 = const()[name = tensor("op_38291_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_38291_end_mask_0 = const()[name = tensor("op_38291_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38291_cast_fp16 = slice_by_index(begin = var_38291_begin_0, end = var_38291_end_0, end_mask = var_38291_end_mask_0, x = q_171_cast_fp16)[name = tensor("op_38291_cast_fp16")]; + tensor var_38295_begin_0 = const()[name = tensor("op_38295_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_38295_end_0 = const()[name = tensor("op_38295_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_38295_end_mask_0 = const()[name = tensor("op_38295_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38295_cast_fp16 = slice_by_index(begin = var_38295_begin_0, end = var_38295_end_0, end_mask = var_38295_end_mask_0, x = q_171_cast_fp16)[name = tensor("op_38295_cast_fp16")]; + tensor var_38299_begin_0 = const()[name = tensor("op_38299_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_38299_end_0 = const()[name = tensor("op_38299_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_38299_end_mask_0 = const()[name = tensor("op_38299_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38299_cast_fp16 = slice_by_index(begin = var_38299_begin_0, end = var_38299_end_0, end_mask = var_38299_end_mask_0, x = q_171_cast_fp16)[name = tensor("op_38299_cast_fp16")]; + tensor var_38303_begin_0 = const()[name = tensor("op_38303_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_38303_end_0 = const()[name = tensor("op_38303_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_38303_end_mask_0 = const()[name = tensor("op_38303_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38303_cast_fp16 = slice_by_index(begin = var_38303_begin_0, end = var_38303_end_0, end_mask = var_38303_end_mask_0, x = q_171_cast_fp16)[name = tensor("op_38303_cast_fp16")]; + tensor k_343_perm_0 = const()[name = tensor("k_343_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_38310_begin_0 = const()[name = tensor("op_38310_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_38310_end_0 = const()[name = tensor("op_38310_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_38310_end_mask_0 = const()[name = tensor("op_38310_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_343_cast_fp16 = transpose(perm = k_343_perm_0, x = k_341_cast_fp16)[name = tensor("transpose_54")]; + tensor var_38310_cast_fp16 = slice_by_index(begin = var_38310_begin_0, end = var_38310_end_0, end_mask = var_38310_end_mask_0, x = k_343_cast_fp16)[name = tensor("op_38310_cast_fp16")]; + tensor var_38314_begin_0 = const()[name = tensor("op_38314_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_38314_end_0 = const()[name = tensor("op_38314_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_38314_end_mask_0 = const()[name = tensor("op_38314_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38314_cast_fp16 = slice_by_index(begin = var_38314_begin_0, end = var_38314_end_0, end_mask = var_38314_end_mask_0, x = k_343_cast_fp16)[name = tensor("op_38314_cast_fp16")]; + tensor var_38318_begin_0 = const()[name = tensor("op_38318_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_38318_end_0 = const()[name = tensor("op_38318_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_38318_end_mask_0 = const()[name = tensor("op_38318_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38318_cast_fp16 = slice_by_index(begin = var_38318_begin_0, end = var_38318_end_0, end_mask = var_38318_end_mask_0, x = k_343_cast_fp16)[name = tensor("op_38318_cast_fp16")]; + tensor var_38322_begin_0 = const()[name = tensor("op_38322_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_38322_end_0 = const()[name = tensor("op_38322_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_38322_end_mask_0 = const()[name = tensor("op_38322_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38322_cast_fp16 = slice_by_index(begin = var_38322_begin_0, end = var_38322_end_0, end_mask = var_38322_end_mask_0, x = k_343_cast_fp16)[name = tensor("op_38322_cast_fp16")]; + tensor var_38326_begin_0 = const()[name = tensor("op_38326_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_38326_end_0 = const()[name = tensor("op_38326_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_38326_end_mask_0 = const()[name = tensor("op_38326_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38326_cast_fp16 = slice_by_index(begin = var_38326_begin_0, end = var_38326_end_0, end_mask = var_38326_end_mask_0, x = k_343_cast_fp16)[name = tensor("op_38326_cast_fp16")]; + tensor var_38330_begin_0 = const()[name = tensor("op_38330_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_38330_end_0 = const()[name = tensor("op_38330_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_38330_end_mask_0 = const()[name = tensor("op_38330_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38330_cast_fp16 = slice_by_index(begin = var_38330_begin_0, end = var_38330_end_0, end_mask = var_38330_end_mask_0, x = k_343_cast_fp16)[name = tensor("op_38330_cast_fp16")]; + tensor var_38334_begin_0 = const()[name = tensor("op_38334_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_38334_end_0 = const()[name = tensor("op_38334_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_38334_end_mask_0 = const()[name = tensor("op_38334_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38334_cast_fp16 = slice_by_index(begin = var_38334_begin_0, end = var_38334_end_0, end_mask = var_38334_end_mask_0, x = k_343_cast_fp16)[name = tensor("op_38334_cast_fp16")]; + tensor var_38338_begin_0 = const()[name = tensor("op_38338_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_38338_end_0 = const()[name = tensor("op_38338_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_38338_end_mask_0 = const()[name = tensor("op_38338_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38338_cast_fp16 = slice_by_index(begin = var_38338_begin_0, end = var_38338_end_0, end_mask = var_38338_end_mask_0, x = k_343_cast_fp16)[name = tensor("op_38338_cast_fp16")]; + tensor var_38342_begin_0 = const()[name = tensor("op_38342_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_38342_end_0 = const()[name = tensor("op_38342_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_38342_end_mask_0 = const()[name = tensor("op_38342_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38342_cast_fp16 = slice_by_index(begin = var_38342_begin_0, end = var_38342_end_0, end_mask = var_38342_end_mask_0, x = k_343_cast_fp16)[name = tensor("op_38342_cast_fp16")]; + tensor var_38346_begin_0 = const()[name = tensor("op_38346_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_38346_end_0 = const()[name = tensor("op_38346_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_38346_end_mask_0 = const()[name = tensor("op_38346_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38346_cast_fp16 = slice_by_index(begin = var_38346_begin_0, end = var_38346_end_0, end_mask = var_38346_end_mask_0, x = k_343_cast_fp16)[name = tensor("op_38346_cast_fp16")]; + tensor var_38350_begin_0 = const()[name = tensor("op_38350_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_38350_end_0 = const()[name = tensor("op_38350_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_38350_end_mask_0 = const()[name = tensor("op_38350_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38350_cast_fp16 = slice_by_index(begin = var_38350_begin_0, end = var_38350_end_0, end_mask = var_38350_end_mask_0, x = k_343_cast_fp16)[name = tensor("op_38350_cast_fp16")]; + tensor var_38354_begin_0 = const()[name = tensor("op_38354_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_38354_end_0 = const()[name = tensor("op_38354_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_38354_end_mask_0 = const()[name = tensor("op_38354_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38354_cast_fp16 = slice_by_index(begin = var_38354_begin_0, end = var_38354_end_0, end_mask = var_38354_end_mask_0, x = k_343_cast_fp16)[name = tensor("op_38354_cast_fp16")]; + tensor var_38358_begin_0 = const()[name = tensor("op_38358_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_38358_end_0 = const()[name = tensor("op_38358_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_38358_end_mask_0 = const()[name = tensor("op_38358_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38358_cast_fp16 = slice_by_index(begin = var_38358_begin_0, end = var_38358_end_0, end_mask = var_38358_end_mask_0, x = k_343_cast_fp16)[name = tensor("op_38358_cast_fp16")]; + tensor var_38362_begin_0 = const()[name = tensor("op_38362_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_38362_end_0 = const()[name = tensor("op_38362_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_38362_end_mask_0 = const()[name = tensor("op_38362_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38362_cast_fp16 = slice_by_index(begin = var_38362_begin_0, end = var_38362_end_0, end_mask = var_38362_end_mask_0, x = k_343_cast_fp16)[name = tensor("op_38362_cast_fp16")]; + tensor var_38366_begin_0 = const()[name = tensor("op_38366_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_38366_end_0 = const()[name = tensor("op_38366_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_38366_end_mask_0 = const()[name = tensor("op_38366_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38366_cast_fp16 = slice_by_index(begin = var_38366_begin_0, end = var_38366_end_0, end_mask = var_38366_end_mask_0, x = k_343_cast_fp16)[name = tensor("op_38366_cast_fp16")]; + tensor var_38370_begin_0 = const()[name = tensor("op_38370_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_38370_end_0 = const()[name = tensor("op_38370_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_38370_end_mask_0 = const()[name = tensor("op_38370_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38370_cast_fp16 = slice_by_index(begin = var_38370_begin_0, end = var_38370_end_0, end_mask = var_38370_end_mask_0, x = k_343_cast_fp16)[name = tensor("op_38370_cast_fp16")]; + tensor var_38374_begin_0 = const()[name = tensor("op_38374_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_38374_end_0 = const()[name = tensor("op_38374_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_38374_end_mask_0 = const()[name = tensor("op_38374_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38374_cast_fp16 = slice_by_index(begin = var_38374_begin_0, end = var_38374_end_0, end_mask = var_38374_end_mask_0, x = k_343_cast_fp16)[name = tensor("op_38374_cast_fp16")]; + tensor var_38378_begin_0 = const()[name = tensor("op_38378_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_38378_end_0 = const()[name = tensor("op_38378_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_38378_end_mask_0 = const()[name = tensor("op_38378_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38378_cast_fp16 = slice_by_index(begin = var_38378_begin_0, end = var_38378_end_0, end_mask = var_38378_end_mask_0, x = k_343_cast_fp16)[name = tensor("op_38378_cast_fp16")]; + tensor var_38382_begin_0 = const()[name = tensor("op_38382_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_38382_end_0 = const()[name = tensor("op_38382_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_38382_end_mask_0 = const()[name = tensor("op_38382_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38382_cast_fp16 = slice_by_index(begin = var_38382_begin_0, end = var_38382_end_0, end_mask = var_38382_end_mask_0, x = k_343_cast_fp16)[name = tensor("op_38382_cast_fp16")]; + tensor var_38386_begin_0 = const()[name = tensor("op_38386_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_38386_end_0 = const()[name = tensor("op_38386_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_38386_end_mask_0 = const()[name = tensor("op_38386_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38386_cast_fp16 = slice_by_index(begin = var_38386_begin_0, end = var_38386_end_0, end_mask = var_38386_end_mask_0, x = k_343_cast_fp16)[name = tensor("op_38386_cast_fp16")]; + tensor var_38388_begin_0 = const()[name = tensor("op_38388_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_38388_end_0 = const()[name = tensor("op_38388_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_38388_end_mask_0 = const()[name = tensor("op_38388_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38388_cast_fp16 = slice_by_index(begin = var_38388_begin_0, end = var_38388_end_0, end_mask = var_38388_end_mask_0, x = v_171_cast_fp16)[name = tensor("op_38388_cast_fp16")]; + tensor var_38392_begin_0 = const()[name = tensor("op_38392_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_38392_end_0 = const()[name = tensor("op_38392_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_38392_end_mask_0 = const()[name = tensor("op_38392_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38392_cast_fp16 = slice_by_index(begin = var_38392_begin_0, end = var_38392_end_0, end_mask = var_38392_end_mask_0, x = v_171_cast_fp16)[name = tensor("op_38392_cast_fp16")]; + tensor var_38396_begin_0 = const()[name = tensor("op_38396_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_38396_end_0 = const()[name = tensor("op_38396_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_38396_end_mask_0 = const()[name = tensor("op_38396_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38396_cast_fp16 = slice_by_index(begin = var_38396_begin_0, end = var_38396_end_0, end_mask = var_38396_end_mask_0, x = v_171_cast_fp16)[name = tensor("op_38396_cast_fp16")]; + tensor var_38400_begin_0 = const()[name = tensor("op_38400_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_38400_end_0 = const()[name = tensor("op_38400_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_38400_end_mask_0 = const()[name = tensor("op_38400_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38400_cast_fp16 = slice_by_index(begin = var_38400_begin_0, end = var_38400_end_0, end_mask = var_38400_end_mask_0, x = v_171_cast_fp16)[name = tensor("op_38400_cast_fp16")]; + tensor var_38404_begin_0 = const()[name = tensor("op_38404_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_38404_end_0 = const()[name = tensor("op_38404_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_38404_end_mask_0 = const()[name = tensor("op_38404_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38404_cast_fp16 = slice_by_index(begin = var_38404_begin_0, end = var_38404_end_0, end_mask = var_38404_end_mask_0, x = v_171_cast_fp16)[name = tensor("op_38404_cast_fp16")]; + tensor var_38408_begin_0 = const()[name = tensor("op_38408_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_38408_end_0 = const()[name = tensor("op_38408_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_38408_end_mask_0 = const()[name = tensor("op_38408_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38408_cast_fp16 = slice_by_index(begin = var_38408_begin_0, end = var_38408_end_0, end_mask = var_38408_end_mask_0, x = v_171_cast_fp16)[name = tensor("op_38408_cast_fp16")]; + tensor var_38412_begin_0 = const()[name = tensor("op_38412_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_38412_end_0 = const()[name = tensor("op_38412_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_38412_end_mask_0 = const()[name = tensor("op_38412_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38412_cast_fp16 = slice_by_index(begin = var_38412_begin_0, end = var_38412_end_0, end_mask = var_38412_end_mask_0, x = v_171_cast_fp16)[name = tensor("op_38412_cast_fp16")]; + tensor var_38416_begin_0 = const()[name = tensor("op_38416_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_38416_end_0 = const()[name = tensor("op_38416_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_38416_end_mask_0 = const()[name = tensor("op_38416_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38416_cast_fp16 = slice_by_index(begin = var_38416_begin_0, end = var_38416_end_0, end_mask = var_38416_end_mask_0, x = v_171_cast_fp16)[name = tensor("op_38416_cast_fp16")]; + tensor var_38420_begin_0 = const()[name = tensor("op_38420_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_38420_end_0 = const()[name = tensor("op_38420_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_38420_end_mask_0 = const()[name = tensor("op_38420_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38420_cast_fp16 = slice_by_index(begin = var_38420_begin_0, end = var_38420_end_0, end_mask = var_38420_end_mask_0, x = v_171_cast_fp16)[name = tensor("op_38420_cast_fp16")]; + tensor var_38424_begin_0 = const()[name = tensor("op_38424_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_38424_end_0 = const()[name = tensor("op_38424_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_38424_end_mask_0 = const()[name = tensor("op_38424_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38424_cast_fp16 = slice_by_index(begin = var_38424_begin_0, end = var_38424_end_0, end_mask = var_38424_end_mask_0, x = v_171_cast_fp16)[name = tensor("op_38424_cast_fp16")]; + tensor var_38428_begin_0 = const()[name = tensor("op_38428_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_38428_end_0 = const()[name = tensor("op_38428_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_38428_end_mask_0 = const()[name = tensor("op_38428_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38428_cast_fp16 = slice_by_index(begin = var_38428_begin_0, end = var_38428_end_0, end_mask = var_38428_end_mask_0, x = v_171_cast_fp16)[name = tensor("op_38428_cast_fp16")]; + tensor var_38432_begin_0 = const()[name = tensor("op_38432_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_38432_end_0 = const()[name = tensor("op_38432_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_38432_end_mask_0 = const()[name = tensor("op_38432_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38432_cast_fp16 = slice_by_index(begin = var_38432_begin_0, end = var_38432_end_0, end_mask = var_38432_end_mask_0, x = v_171_cast_fp16)[name = tensor("op_38432_cast_fp16")]; + tensor var_38436_begin_0 = const()[name = tensor("op_38436_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_38436_end_0 = const()[name = tensor("op_38436_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_38436_end_mask_0 = const()[name = tensor("op_38436_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38436_cast_fp16 = slice_by_index(begin = var_38436_begin_0, end = var_38436_end_0, end_mask = var_38436_end_mask_0, x = v_171_cast_fp16)[name = tensor("op_38436_cast_fp16")]; + tensor var_38440_begin_0 = const()[name = tensor("op_38440_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_38440_end_0 = const()[name = tensor("op_38440_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_38440_end_mask_0 = const()[name = tensor("op_38440_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38440_cast_fp16 = slice_by_index(begin = var_38440_begin_0, end = var_38440_end_0, end_mask = var_38440_end_mask_0, x = v_171_cast_fp16)[name = tensor("op_38440_cast_fp16")]; + tensor var_38444_begin_0 = const()[name = tensor("op_38444_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_38444_end_0 = const()[name = tensor("op_38444_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_38444_end_mask_0 = const()[name = tensor("op_38444_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38444_cast_fp16 = slice_by_index(begin = var_38444_begin_0, end = var_38444_end_0, end_mask = var_38444_end_mask_0, x = v_171_cast_fp16)[name = tensor("op_38444_cast_fp16")]; + tensor var_38448_begin_0 = const()[name = tensor("op_38448_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_38448_end_0 = const()[name = tensor("op_38448_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_38448_end_mask_0 = const()[name = tensor("op_38448_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38448_cast_fp16 = slice_by_index(begin = var_38448_begin_0, end = var_38448_end_0, end_mask = var_38448_end_mask_0, x = v_171_cast_fp16)[name = tensor("op_38448_cast_fp16")]; + tensor var_38452_begin_0 = const()[name = tensor("op_38452_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_38452_end_0 = const()[name = tensor("op_38452_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_38452_end_mask_0 = const()[name = tensor("op_38452_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38452_cast_fp16 = slice_by_index(begin = var_38452_begin_0, end = var_38452_end_0, end_mask = var_38452_end_mask_0, x = v_171_cast_fp16)[name = tensor("op_38452_cast_fp16")]; + tensor var_38456_begin_0 = const()[name = tensor("op_38456_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_38456_end_0 = const()[name = tensor("op_38456_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_38456_end_mask_0 = const()[name = tensor("op_38456_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38456_cast_fp16 = slice_by_index(begin = var_38456_begin_0, end = var_38456_end_0, end_mask = var_38456_end_mask_0, x = v_171_cast_fp16)[name = tensor("op_38456_cast_fp16")]; + tensor var_38460_begin_0 = const()[name = tensor("op_38460_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_38460_end_0 = const()[name = tensor("op_38460_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_38460_end_mask_0 = const()[name = tensor("op_38460_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38460_cast_fp16 = slice_by_index(begin = var_38460_begin_0, end = var_38460_end_0, end_mask = var_38460_end_mask_0, x = v_171_cast_fp16)[name = tensor("op_38460_cast_fp16")]; + tensor var_38464_begin_0 = const()[name = tensor("op_38464_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_38464_end_0 = const()[name = tensor("op_38464_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_38464_end_mask_0 = const()[name = tensor("op_38464_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38464_cast_fp16 = slice_by_index(begin = var_38464_begin_0, end = var_38464_end_0, end_mask = var_38464_end_mask_0, x = v_171_cast_fp16)[name = tensor("op_38464_cast_fp16")]; + tensor var_38468_equation_0 = const()[name = tensor("op_38468_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38468_cast_fp16 = einsum(equation = var_38468_equation_0, values = (var_38310_cast_fp16, var_38227_cast_fp16))[name = tensor("op_38468_cast_fp16")]; + tensor var_38469_to_fp16 = const()[name = tensor("op_38469_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3241_cast_fp16 = mul(x = var_38468_cast_fp16, y = var_38469_to_fp16)[name = tensor("aw_3241_cast_fp16")]; + tensor var_38472_equation_0 = const()[name = tensor("op_38472_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38472_cast_fp16 = einsum(equation = var_38472_equation_0, values = (var_38314_cast_fp16, var_38231_cast_fp16))[name = tensor("op_38472_cast_fp16")]; + tensor var_38473_to_fp16 = const()[name = tensor("op_38473_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3243_cast_fp16 = mul(x = var_38472_cast_fp16, y = var_38473_to_fp16)[name = tensor("aw_3243_cast_fp16")]; + tensor var_38476_equation_0 = const()[name = tensor("op_38476_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38476_cast_fp16 = einsum(equation = var_38476_equation_0, values = (var_38318_cast_fp16, var_38235_cast_fp16))[name = tensor("op_38476_cast_fp16")]; + tensor var_38477_to_fp16 = const()[name = tensor("op_38477_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3245_cast_fp16 = mul(x = var_38476_cast_fp16, y = var_38477_to_fp16)[name = tensor("aw_3245_cast_fp16")]; + tensor var_38480_equation_0 = const()[name = tensor("op_38480_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38480_cast_fp16 = einsum(equation = var_38480_equation_0, values = (var_38322_cast_fp16, var_38239_cast_fp16))[name = tensor("op_38480_cast_fp16")]; + tensor var_38481_to_fp16 = const()[name = tensor("op_38481_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3247_cast_fp16 = mul(x = var_38480_cast_fp16, y = var_38481_to_fp16)[name = tensor("aw_3247_cast_fp16")]; + tensor var_38484_equation_0 = const()[name = tensor("op_38484_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38484_cast_fp16 = einsum(equation = var_38484_equation_0, values = (var_38326_cast_fp16, var_38243_cast_fp16))[name = tensor("op_38484_cast_fp16")]; + tensor var_38485_to_fp16 = const()[name = tensor("op_38485_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3249_cast_fp16 = mul(x = var_38484_cast_fp16, y = var_38485_to_fp16)[name = tensor("aw_3249_cast_fp16")]; + tensor var_38488_equation_0 = const()[name = tensor("op_38488_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38488_cast_fp16 = einsum(equation = var_38488_equation_0, values = (var_38330_cast_fp16, var_38247_cast_fp16))[name = tensor("op_38488_cast_fp16")]; + tensor var_38489_to_fp16 = const()[name = tensor("op_38489_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3251_cast_fp16 = mul(x = var_38488_cast_fp16, y = var_38489_to_fp16)[name = tensor("aw_3251_cast_fp16")]; + tensor var_38492_equation_0 = const()[name = tensor("op_38492_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38492_cast_fp16 = einsum(equation = var_38492_equation_0, values = (var_38334_cast_fp16, var_38251_cast_fp16))[name = tensor("op_38492_cast_fp16")]; + tensor var_38493_to_fp16 = const()[name = tensor("op_38493_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3253_cast_fp16 = mul(x = var_38492_cast_fp16, y = var_38493_to_fp16)[name = tensor("aw_3253_cast_fp16")]; + tensor var_38496_equation_0 = const()[name = tensor("op_38496_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38496_cast_fp16 = einsum(equation = var_38496_equation_0, values = (var_38338_cast_fp16, var_38255_cast_fp16))[name = tensor("op_38496_cast_fp16")]; + tensor var_38497_to_fp16 = const()[name = tensor("op_38497_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3255_cast_fp16 = mul(x = var_38496_cast_fp16, y = var_38497_to_fp16)[name = tensor("aw_3255_cast_fp16")]; + tensor var_38500_equation_0 = const()[name = tensor("op_38500_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38500_cast_fp16 = einsum(equation = var_38500_equation_0, values = (var_38342_cast_fp16, var_38259_cast_fp16))[name = tensor("op_38500_cast_fp16")]; + tensor var_38501_to_fp16 = const()[name = tensor("op_38501_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3257_cast_fp16 = mul(x = var_38500_cast_fp16, y = var_38501_to_fp16)[name = tensor("aw_3257_cast_fp16")]; + tensor var_38504_equation_0 = const()[name = tensor("op_38504_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38504_cast_fp16 = einsum(equation = var_38504_equation_0, values = (var_38346_cast_fp16, var_38263_cast_fp16))[name = tensor("op_38504_cast_fp16")]; + tensor var_38505_to_fp16 = const()[name = tensor("op_38505_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3259_cast_fp16 = mul(x = var_38504_cast_fp16, y = var_38505_to_fp16)[name = tensor("aw_3259_cast_fp16")]; + tensor var_38508_equation_0 = const()[name = tensor("op_38508_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38508_cast_fp16 = einsum(equation = var_38508_equation_0, values = (var_38350_cast_fp16, var_38267_cast_fp16))[name = tensor("op_38508_cast_fp16")]; + tensor var_38509_to_fp16 = const()[name = tensor("op_38509_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3261_cast_fp16 = mul(x = var_38508_cast_fp16, y = var_38509_to_fp16)[name = tensor("aw_3261_cast_fp16")]; + tensor var_38512_equation_0 = const()[name = tensor("op_38512_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38512_cast_fp16 = einsum(equation = var_38512_equation_0, values = (var_38354_cast_fp16, var_38271_cast_fp16))[name = tensor("op_38512_cast_fp16")]; + tensor var_38513_to_fp16 = const()[name = tensor("op_38513_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3263_cast_fp16 = mul(x = var_38512_cast_fp16, y = var_38513_to_fp16)[name = tensor("aw_3263_cast_fp16")]; + tensor var_38516_equation_0 = const()[name = tensor("op_38516_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38516_cast_fp16 = einsum(equation = var_38516_equation_0, values = (var_38358_cast_fp16, var_38275_cast_fp16))[name = tensor("op_38516_cast_fp16")]; + tensor var_38517_to_fp16 = const()[name = tensor("op_38517_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3265_cast_fp16 = mul(x = var_38516_cast_fp16, y = var_38517_to_fp16)[name = tensor("aw_3265_cast_fp16")]; + tensor var_38520_equation_0 = const()[name = tensor("op_38520_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38520_cast_fp16 = einsum(equation = var_38520_equation_0, values = (var_38362_cast_fp16, var_38279_cast_fp16))[name = tensor("op_38520_cast_fp16")]; + tensor var_38521_to_fp16 = const()[name = tensor("op_38521_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3267_cast_fp16 = mul(x = var_38520_cast_fp16, y = var_38521_to_fp16)[name = tensor("aw_3267_cast_fp16")]; + tensor var_38524_equation_0 = const()[name = tensor("op_38524_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38524_cast_fp16 = einsum(equation = var_38524_equation_0, values = (var_38366_cast_fp16, var_38283_cast_fp16))[name = tensor("op_38524_cast_fp16")]; + tensor var_38525_to_fp16 = const()[name = tensor("op_38525_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3269_cast_fp16 = mul(x = var_38524_cast_fp16, y = var_38525_to_fp16)[name = tensor("aw_3269_cast_fp16")]; + tensor var_38528_equation_0 = const()[name = tensor("op_38528_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38528_cast_fp16 = einsum(equation = var_38528_equation_0, values = (var_38370_cast_fp16, var_38287_cast_fp16))[name = tensor("op_38528_cast_fp16")]; + tensor var_38529_to_fp16 = const()[name = tensor("op_38529_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3271_cast_fp16 = mul(x = var_38528_cast_fp16, y = var_38529_to_fp16)[name = tensor("aw_3271_cast_fp16")]; + tensor var_38532_equation_0 = const()[name = tensor("op_38532_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38532_cast_fp16 = einsum(equation = var_38532_equation_0, values = (var_38374_cast_fp16, var_38291_cast_fp16))[name = tensor("op_38532_cast_fp16")]; + tensor var_38533_to_fp16 = const()[name = tensor("op_38533_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3273_cast_fp16 = mul(x = var_38532_cast_fp16, y = var_38533_to_fp16)[name = tensor("aw_3273_cast_fp16")]; + tensor var_38536_equation_0 = const()[name = tensor("op_38536_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38536_cast_fp16 = einsum(equation = var_38536_equation_0, values = (var_38378_cast_fp16, var_38295_cast_fp16))[name = tensor("op_38536_cast_fp16")]; + tensor var_38537_to_fp16 = const()[name = tensor("op_38537_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3275_cast_fp16 = mul(x = var_38536_cast_fp16, y = var_38537_to_fp16)[name = tensor("aw_3275_cast_fp16")]; + tensor var_38540_equation_0 = const()[name = tensor("op_38540_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38540_cast_fp16 = einsum(equation = var_38540_equation_0, values = (var_38382_cast_fp16, var_38299_cast_fp16))[name = tensor("op_38540_cast_fp16")]; + tensor var_38541_to_fp16 = const()[name = tensor("op_38541_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3277_cast_fp16 = mul(x = var_38540_cast_fp16, y = var_38541_to_fp16)[name = tensor("aw_3277_cast_fp16")]; + tensor var_38544_equation_0 = const()[name = tensor("op_38544_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38544_cast_fp16 = einsum(equation = var_38544_equation_0, values = (var_38386_cast_fp16, var_38303_cast_fp16))[name = tensor("op_38544_cast_fp16")]; + tensor var_38545_to_fp16 = const()[name = tensor("op_38545_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3279_cast_fp16 = mul(x = var_38544_cast_fp16, y = var_38545_to_fp16)[name = tensor("aw_3279_cast_fp16")]; + tensor var_38547_cast_fp16 = softmax(axis = var_30355, x = aw_3241_cast_fp16)[name = tensor("op_38547_cast_fp16")]; + tensor var_38548_cast_fp16 = softmax(axis = var_30355, x = aw_3243_cast_fp16)[name = tensor("op_38548_cast_fp16")]; + tensor var_38549_cast_fp16 = softmax(axis = var_30355, x = aw_3245_cast_fp16)[name = tensor("op_38549_cast_fp16")]; + tensor var_38550_cast_fp16 = softmax(axis = var_30355, x = aw_3247_cast_fp16)[name = tensor("op_38550_cast_fp16")]; + tensor var_38551_cast_fp16 = softmax(axis = var_30355, x = aw_3249_cast_fp16)[name = tensor("op_38551_cast_fp16")]; + tensor var_38552_cast_fp16 = softmax(axis = var_30355, x = aw_3251_cast_fp16)[name = tensor("op_38552_cast_fp16")]; + tensor var_38553_cast_fp16 = softmax(axis = var_30355, x = aw_3253_cast_fp16)[name = tensor("op_38553_cast_fp16")]; + tensor var_38554_cast_fp16 = softmax(axis = var_30355, x = aw_3255_cast_fp16)[name = tensor("op_38554_cast_fp16")]; + tensor var_38555_cast_fp16 = softmax(axis = var_30355, x = aw_3257_cast_fp16)[name = tensor("op_38555_cast_fp16")]; + tensor var_38556_cast_fp16 = softmax(axis = var_30355, x = aw_3259_cast_fp16)[name = tensor("op_38556_cast_fp16")]; + tensor var_38557_cast_fp16 = softmax(axis = var_30355, x = aw_3261_cast_fp16)[name = tensor("op_38557_cast_fp16")]; + tensor var_38558_cast_fp16 = softmax(axis = var_30355, x = aw_3263_cast_fp16)[name = tensor("op_38558_cast_fp16")]; + tensor var_38559_cast_fp16 = softmax(axis = var_30355, x = aw_3265_cast_fp16)[name = tensor("op_38559_cast_fp16")]; + tensor var_38560_cast_fp16 = softmax(axis = var_30355, x = aw_3267_cast_fp16)[name = tensor("op_38560_cast_fp16")]; + tensor var_38561_cast_fp16 = softmax(axis = var_30355, x = aw_3269_cast_fp16)[name = tensor("op_38561_cast_fp16")]; + tensor var_38562_cast_fp16 = softmax(axis = var_30355, x = aw_3271_cast_fp16)[name = tensor("op_38562_cast_fp16")]; + tensor var_38563_cast_fp16 = softmax(axis = var_30355, x = aw_3273_cast_fp16)[name = tensor("op_38563_cast_fp16")]; + tensor var_38564_cast_fp16 = softmax(axis = var_30355, x = aw_3275_cast_fp16)[name = tensor("op_38564_cast_fp16")]; + tensor var_38565_cast_fp16 = softmax(axis = var_30355, x = aw_3277_cast_fp16)[name = tensor("op_38565_cast_fp16")]; + tensor var_38566_cast_fp16 = softmax(axis = var_30355, x = aw_3279_cast_fp16)[name = tensor("op_38566_cast_fp16")]; + tensor var_38568_equation_0 = const()[name = tensor("op_38568_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_38568_cast_fp16 = einsum(equation = var_38568_equation_0, values = (var_38388_cast_fp16, var_38547_cast_fp16))[name = tensor("op_38568_cast_fp16")]; + tensor var_38570_equation_0 = const()[name = tensor("op_38570_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_38570_cast_fp16 = einsum(equation = var_38570_equation_0, values = (var_38392_cast_fp16, var_38548_cast_fp16))[name = tensor("op_38570_cast_fp16")]; + tensor var_38572_equation_0 = const()[name = tensor("op_38572_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_38572_cast_fp16 = einsum(equation = var_38572_equation_0, values = (var_38396_cast_fp16, var_38549_cast_fp16))[name = tensor("op_38572_cast_fp16")]; + tensor var_38574_equation_0 = const()[name = tensor("op_38574_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_38574_cast_fp16 = einsum(equation = var_38574_equation_0, values = (var_38400_cast_fp16, var_38550_cast_fp16))[name = tensor("op_38574_cast_fp16")]; + tensor var_38576_equation_0 = const()[name = tensor("op_38576_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_38576_cast_fp16 = einsum(equation = var_38576_equation_0, values = (var_38404_cast_fp16, var_38551_cast_fp16))[name = tensor("op_38576_cast_fp16")]; + tensor var_38578_equation_0 = const()[name = tensor("op_38578_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_38578_cast_fp16 = einsum(equation = var_38578_equation_0, values = (var_38408_cast_fp16, var_38552_cast_fp16))[name = tensor("op_38578_cast_fp16")]; + tensor var_38580_equation_0 = const()[name = tensor("op_38580_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_38580_cast_fp16 = einsum(equation = var_38580_equation_0, values = (var_38412_cast_fp16, var_38553_cast_fp16))[name = tensor("op_38580_cast_fp16")]; + tensor var_38582_equation_0 = const()[name = tensor("op_38582_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_38582_cast_fp16 = einsum(equation = var_38582_equation_0, values = (var_38416_cast_fp16, var_38554_cast_fp16))[name = tensor("op_38582_cast_fp16")]; + tensor var_38584_equation_0 = const()[name = tensor("op_38584_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_38584_cast_fp16 = einsum(equation = var_38584_equation_0, values = (var_38420_cast_fp16, var_38555_cast_fp16))[name = tensor("op_38584_cast_fp16")]; + tensor var_38586_equation_0 = const()[name = tensor("op_38586_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_38586_cast_fp16 = einsum(equation = var_38586_equation_0, values = (var_38424_cast_fp16, var_38556_cast_fp16))[name = tensor("op_38586_cast_fp16")]; + tensor var_38588_equation_0 = const()[name = tensor("op_38588_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_38588_cast_fp16 = einsum(equation = var_38588_equation_0, values = (var_38428_cast_fp16, var_38557_cast_fp16))[name = tensor("op_38588_cast_fp16")]; + tensor var_38590_equation_0 = const()[name = tensor("op_38590_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_38590_cast_fp16 = einsum(equation = var_38590_equation_0, values = (var_38432_cast_fp16, var_38558_cast_fp16))[name = tensor("op_38590_cast_fp16")]; + tensor var_38592_equation_0 = const()[name = tensor("op_38592_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_38592_cast_fp16 = einsum(equation = var_38592_equation_0, values = (var_38436_cast_fp16, var_38559_cast_fp16))[name = tensor("op_38592_cast_fp16")]; + tensor var_38594_equation_0 = const()[name = tensor("op_38594_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_38594_cast_fp16 = einsum(equation = var_38594_equation_0, values = (var_38440_cast_fp16, var_38560_cast_fp16))[name = tensor("op_38594_cast_fp16")]; + tensor var_38596_equation_0 = const()[name = tensor("op_38596_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_38596_cast_fp16 = einsum(equation = var_38596_equation_0, values = (var_38444_cast_fp16, var_38561_cast_fp16))[name = tensor("op_38596_cast_fp16")]; + tensor var_38598_equation_0 = const()[name = tensor("op_38598_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_38598_cast_fp16 = einsum(equation = var_38598_equation_0, values = (var_38448_cast_fp16, var_38562_cast_fp16))[name = tensor("op_38598_cast_fp16")]; + tensor var_38600_equation_0 = const()[name = tensor("op_38600_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_38600_cast_fp16 = einsum(equation = var_38600_equation_0, values = (var_38452_cast_fp16, var_38563_cast_fp16))[name = tensor("op_38600_cast_fp16")]; + tensor var_38602_equation_0 = const()[name = tensor("op_38602_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_38602_cast_fp16 = einsum(equation = var_38602_equation_0, values = (var_38456_cast_fp16, var_38564_cast_fp16))[name = tensor("op_38602_cast_fp16")]; + tensor var_38604_equation_0 = const()[name = tensor("op_38604_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_38604_cast_fp16 = einsum(equation = var_38604_equation_0, values = (var_38460_cast_fp16, var_38565_cast_fp16))[name = tensor("op_38604_cast_fp16")]; + tensor var_38606_equation_0 = const()[name = tensor("op_38606_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_38606_cast_fp16 = einsum(equation = var_38606_equation_0, values = (var_38464_cast_fp16, var_38566_cast_fp16))[name = tensor("op_38606_cast_fp16")]; + tensor input_505_interleave_0 = const()[name = tensor("input_505_interleave_0"), val = tensor(false)]; + tensor input_505_cast_fp16 = concat(axis = var_30355, interleave = input_505_interleave_0, values = (var_38568_cast_fp16, var_38570_cast_fp16, var_38572_cast_fp16, var_38574_cast_fp16, var_38576_cast_fp16, var_38578_cast_fp16, var_38580_cast_fp16, var_38582_cast_fp16, var_38584_cast_fp16, var_38586_cast_fp16, var_38588_cast_fp16, var_38590_cast_fp16, var_38592_cast_fp16, var_38594_cast_fp16, var_38596_cast_fp16, var_38598_cast_fp16, var_38600_cast_fp16, var_38602_cast_fp16, var_38604_cast_fp16, var_38606_cast_fp16))[name = tensor("input_505_cast_fp16")]; + tensor var_38612 = const()[name = tensor("op_38612"), val = tensor([1, 1])]; + tensor var_38614 = const()[name = tensor("op_38614"), val = tensor([1, 1])]; + tensor var_38616_pad_type_0 = const()[name = tensor("op_38616_pad_type_0"), val = tensor("custom")]; + tensor var_38616_pad_0 = const()[name = tensor("op_38616_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_8_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_8_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3171408384)))]; + tensor up_blocks_0_attentions_0_transformer_blocks_8_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_8_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3174685248)))]; + tensor var_38616_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_8_attn2_to_out_0_bias_to_fp16, dilations = var_38614, groups = var_30355, pad = var_38616_pad_0, pad_type = var_38616_pad_type_0, strides = var_38612, weight = up_blocks_0_attentions_0_transformer_blocks_8_attn2_to_out_0_weight_to_fp16, x = input_505_cast_fp16)[name = tensor("op_38616_cast_fp16")]; + tensor inputs_257_cast_fp16 = add(x = var_38616_cast_fp16, y = inputs_255_cast_fp16)[name = tensor("inputs_257_cast_fp16")]; + tensor input_507_axes_0 = const()[name = tensor("input_507_axes_0"), val = tensor([1])]; + tensor input_507_gamma_0_to_fp16 = const()[name = tensor("input_507_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3174687872)))]; + tensor input_507_beta_0_to_fp16 = const()[name = tensor("input_507_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3174690496)))]; + tensor var_38626_to_fp16 = const()[name = tensor("op_38626_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_507_cast_fp16 = layer_norm(axes = input_507_axes_0, beta = input_507_beta_0_to_fp16, epsilon = var_38626_to_fp16, gamma = input_507_gamma_0_to_fp16, x = inputs_257_cast_fp16)[name = tensor("input_507_cast_fp16")]; + tensor var_38642 = const()[name = tensor("op_38642"), val = tensor([1, 1])]; + tensor var_38644 = const()[name = tensor("op_38644"), val = tensor([1, 1])]; + tensor var_38646_pad_type_0 = const()[name = tensor("op_38646_pad_type_0"), val = tensor("custom")]; + tensor var_38646_pad_0 = const()[name = tensor("op_38646_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_8_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_8_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3174693120)))]; + tensor up_blocks_0_attentions_0_transformer_blocks_8_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_8_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3200907584)))]; + tensor var_38646_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_8_ff_net_0_proj_bias_to_fp16, dilations = var_38644, groups = var_30355, pad = var_38646_pad_0, pad_type = var_38646_pad_type_0, strides = var_38642, weight = up_blocks_0_attentions_0_transformer_blocks_8_ff_net_0_proj_weight_to_fp16, x = input_507_cast_fp16)[name = tensor("op_38646_cast_fp16")]; + tensor var_38647_split_sizes_0 = const()[name = tensor("op_38647_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_38647_axis_0 = const()[name = tensor("op_38647_axis_0"), val = tensor(1)]; + tensor var_38647_cast_fp16_0, tensor var_38647_cast_fp16_1 = split(axis = var_38647_axis_0, split_sizes = var_38647_split_sizes_0, x = var_38646_cast_fp16)[name = tensor("op_38647_cast_fp16")]; + tensor var_38649_mode_0 = const()[name = tensor("op_38649_mode_0"), val = tensor("EXACT")]; + tensor var_38649_cast_fp16 = gelu(mode = var_38649_mode_0, x = var_38647_cast_fp16_1)[name = tensor("op_38649_cast_fp16")]; + tensor input_509_cast_fp16 = mul(x = var_38647_cast_fp16_0, y = var_38649_cast_fp16)[name = tensor("input_509_cast_fp16")]; + tensor var_38653 = const()[name = tensor("op_38653"), val = tensor([1, 1])]; + tensor var_38655 = const()[name = tensor("op_38655"), val = tensor([1, 1])]; + tensor var_38657_pad_type_0 = const()[name = tensor("op_38657_pad_type_0"), val = tensor("custom")]; + tensor var_38657_pad_0 = const()[name = tensor("op_38657_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_8_ff_net_2_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_8_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3200928128)))]; + tensor up_blocks_0_attentions_0_transformer_blocks_8_ff_net_2_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_8_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3214035392)))]; + tensor var_38657_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_8_ff_net_2_bias_to_fp16, dilations = var_38655, groups = var_30355, pad = var_38657_pad_0, pad_type = var_38657_pad_type_0, strides = var_38653, weight = up_blocks_0_attentions_0_transformer_blocks_8_ff_net_2_weight_to_fp16, x = input_509_cast_fp16)[name = tensor("op_38657_cast_fp16")]; + tensor inputs_259_cast_fp16 = add(x = var_38657_cast_fp16, y = inputs_257_cast_fp16)[name = tensor("inputs_259_cast_fp16")]; + tensor hidden_states_341_axes_0 = const()[name = tensor("hidden_states_341_axes_0"), val = tensor([1])]; + tensor hidden_states_341_gamma_0_to_fp16 = const()[name = tensor("hidden_states_341_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3214038016)))]; + tensor hidden_states_341_beta_0_to_fp16 = const()[name = tensor("hidden_states_341_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3214040640)))]; + tensor var_38673_to_fp16 = const()[name = tensor("op_38673_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_341_cast_fp16 = layer_norm(axes = hidden_states_341_axes_0, beta = hidden_states_341_beta_0_to_fp16, epsilon = var_38673_to_fp16, gamma = hidden_states_341_gamma_0_to_fp16, x = inputs_259_cast_fp16)[name = tensor("hidden_states_341_cast_fp16")]; + tensor var_38688 = const()[name = tensor("op_38688"), val = tensor([1, 1])]; + tensor var_38690 = const()[name = tensor("op_38690"), val = tensor([1, 1])]; + tensor q_173_pad_type_0 = const()[name = tensor("q_173_pad_type_0"), val = tensor("custom")]; + tensor q_173_pad_0 = const()[name = tensor("q_173_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_9_attn1_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_9_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3214043264)))]; + tensor q_173_cast_fp16 = conv(dilations = var_38690, groups = var_30355, pad = q_173_pad_0, pad_type = q_173_pad_type_0, strides = var_38688, weight = up_blocks_0_attentions_0_transformer_blocks_9_attn1_to_q_weight_to_fp16, x = hidden_states_341_cast_fp16)[name = tensor("q_173_cast_fp16")]; + tensor var_38694 = const()[name = tensor("op_38694"), val = tensor([1, 1])]; + tensor var_38696 = const()[name = tensor("op_38696"), val = tensor([1, 1])]; + tensor k_345_pad_type_0 = const()[name = tensor("k_345_pad_type_0"), val = tensor("custom")]; + tensor k_345_pad_0 = const()[name = tensor("k_345_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_9_attn1_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_9_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3217320128)))]; + tensor k_345_cast_fp16 = conv(dilations = var_38696, groups = var_30355, pad = k_345_pad_0, pad_type = k_345_pad_type_0, strides = var_38694, weight = up_blocks_0_attentions_0_transformer_blocks_9_attn1_to_k_weight_to_fp16, x = hidden_states_341_cast_fp16)[name = tensor("k_345_cast_fp16")]; + tensor var_38700 = const()[name = tensor("op_38700"), val = tensor([1, 1])]; + tensor var_38702 = const()[name = tensor("op_38702"), val = tensor([1, 1])]; + tensor v_173_pad_type_0 = const()[name = tensor("v_173_pad_type_0"), val = tensor("custom")]; + tensor v_173_pad_0 = const()[name = tensor("v_173_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_9_attn1_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_9_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3220596992)))]; + tensor v_173_cast_fp16 = conv(dilations = var_38702, groups = var_30355, pad = v_173_pad_0, pad_type = v_173_pad_type_0, strides = var_38700, weight = up_blocks_0_attentions_0_transformer_blocks_9_attn1_to_v_weight_to_fp16, x = hidden_states_341_cast_fp16)[name = tensor("v_173_cast_fp16")]; + tensor var_38706_begin_0 = const()[name = tensor("op_38706_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_38706_end_0 = const()[name = tensor("op_38706_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_38706_end_mask_0 = const()[name = tensor("op_38706_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38706_cast_fp16 = slice_by_index(begin = var_38706_begin_0, end = var_38706_end_0, end_mask = var_38706_end_mask_0, x = q_173_cast_fp16)[name = tensor("op_38706_cast_fp16")]; + tensor var_38710_begin_0 = const()[name = tensor("op_38710_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_38710_end_0 = const()[name = tensor("op_38710_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_38710_end_mask_0 = const()[name = tensor("op_38710_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38710_cast_fp16 = slice_by_index(begin = var_38710_begin_0, end = var_38710_end_0, end_mask = var_38710_end_mask_0, x = q_173_cast_fp16)[name = tensor("op_38710_cast_fp16")]; + tensor var_38714_begin_0 = const()[name = tensor("op_38714_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_38714_end_0 = const()[name = tensor("op_38714_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_38714_end_mask_0 = const()[name = tensor("op_38714_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38714_cast_fp16 = slice_by_index(begin = var_38714_begin_0, end = var_38714_end_0, end_mask = var_38714_end_mask_0, x = q_173_cast_fp16)[name = tensor("op_38714_cast_fp16")]; + tensor var_38718_begin_0 = const()[name = tensor("op_38718_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_38718_end_0 = const()[name = tensor("op_38718_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_38718_end_mask_0 = const()[name = tensor("op_38718_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38718_cast_fp16 = slice_by_index(begin = var_38718_begin_0, end = var_38718_end_0, end_mask = var_38718_end_mask_0, x = q_173_cast_fp16)[name = tensor("op_38718_cast_fp16")]; + tensor var_38722_begin_0 = const()[name = tensor("op_38722_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_38722_end_0 = const()[name = tensor("op_38722_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_38722_end_mask_0 = const()[name = tensor("op_38722_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38722_cast_fp16 = slice_by_index(begin = var_38722_begin_0, end = var_38722_end_0, end_mask = var_38722_end_mask_0, x = q_173_cast_fp16)[name = tensor("op_38722_cast_fp16")]; + tensor var_38726_begin_0 = const()[name = tensor("op_38726_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_38726_end_0 = const()[name = tensor("op_38726_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_38726_end_mask_0 = const()[name = tensor("op_38726_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38726_cast_fp16 = slice_by_index(begin = var_38726_begin_0, end = var_38726_end_0, end_mask = var_38726_end_mask_0, x = q_173_cast_fp16)[name = tensor("op_38726_cast_fp16")]; + tensor var_38730_begin_0 = const()[name = tensor("op_38730_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_38730_end_0 = const()[name = tensor("op_38730_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_38730_end_mask_0 = const()[name = tensor("op_38730_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38730_cast_fp16 = slice_by_index(begin = var_38730_begin_0, end = var_38730_end_0, end_mask = var_38730_end_mask_0, x = q_173_cast_fp16)[name = tensor("op_38730_cast_fp16")]; + tensor var_38734_begin_0 = const()[name = tensor("op_38734_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_38734_end_0 = const()[name = tensor("op_38734_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_38734_end_mask_0 = const()[name = tensor("op_38734_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38734_cast_fp16 = slice_by_index(begin = var_38734_begin_0, end = var_38734_end_0, end_mask = var_38734_end_mask_0, x = q_173_cast_fp16)[name = tensor("op_38734_cast_fp16")]; + tensor var_38738_begin_0 = const()[name = tensor("op_38738_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_38738_end_0 = const()[name = tensor("op_38738_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_38738_end_mask_0 = const()[name = tensor("op_38738_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38738_cast_fp16 = slice_by_index(begin = var_38738_begin_0, end = var_38738_end_0, end_mask = var_38738_end_mask_0, x = q_173_cast_fp16)[name = tensor("op_38738_cast_fp16")]; + tensor var_38742_begin_0 = const()[name = tensor("op_38742_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_38742_end_0 = const()[name = tensor("op_38742_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_38742_end_mask_0 = const()[name = tensor("op_38742_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38742_cast_fp16 = slice_by_index(begin = var_38742_begin_0, end = var_38742_end_0, end_mask = var_38742_end_mask_0, x = q_173_cast_fp16)[name = tensor("op_38742_cast_fp16")]; + tensor var_38746_begin_0 = const()[name = tensor("op_38746_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_38746_end_0 = const()[name = tensor("op_38746_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_38746_end_mask_0 = const()[name = tensor("op_38746_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38746_cast_fp16 = slice_by_index(begin = var_38746_begin_0, end = var_38746_end_0, end_mask = var_38746_end_mask_0, x = q_173_cast_fp16)[name = tensor("op_38746_cast_fp16")]; + tensor var_38750_begin_0 = const()[name = tensor("op_38750_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_38750_end_0 = const()[name = tensor("op_38750_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_38750_end_mask_0 = const()[name = tensor("op_38750_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38750_cast_fp16 = slice_by_index(begin = var_38750_begin_0, end = var_38750_end_0, end_mask = var_38750_end_mask_0, x = q_173_cast_fp16)[name = tensor("op_38750_cast_fp16")]; + tensor var_38754_begin_0 = const()[name = tensor("op_38754_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_38754_end_0 = const()[name = tensor("op_38754_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_38754_end_mask_0 = const()[name = tensor("op_38754_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38754_cast_fp16 = slice_by_index(begin = var_38754_begin_0, end = var_38754_end_0, end_mask = var_38754_end_mask_0, x = q_173_cast_fp16)[name = tensor("op_38754_cast_fp16")]; + tensor var_38758_begin_0 = const()[name = tensor("op_38758_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_38758_end_0 = const()[name = tensor("op_38758_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_38758_end_mask_0 = const()[name = tensor("op_38758_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38758_cast_fp16 = slice_by_index(begin = var_38758_begin_0, end = var_38758_end_0, end_mask = var_38758_end_mask_0, x = q_173_cast_fp16)[name = tensor("op_38758_cast_fp16")]; + tensor var_38762_begin_0 = const()[name = tensor("op_38762_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_38762_end_0 = const()[name = tensor("op_38762_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_38762_end_mask_0 = const()[name = tensor("op_38762_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38762_cast_fp16 = slice_by_index(begin = var_38762_begin_0, end = var_38762_end_0, end_mask = var_38762_end_mask_0, x = q_173_cast_fp16)[name = tensor("op_38762_cast_fp16")]; + tensor var_38766_begin_0 = const()[name = tensor("op_38766_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_38766_end_0 = const()[name = tensor("op_38766_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_38766_end_mask_0 = const()[name = tensor("op_38766_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38766_cast_fp16 = slice_by_index(begin = var_38766_begin_0, end = var_38766_end_0, end_mask = var_38766_end_mask_0, x = q_173_cast_fp16)[name = tensor("op_38766_cast_fp16")]; + tensor var_38770_begin_0 = const()[name = tensor("op_38770_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_38770_end_0 = const()[name = tensor("op_38770_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_38770_end_mask_0 = const()[name = tensor("op_38770_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38770_cast_fp16 = slice_by_index(begin = var_38770_begin_0, end = var_38770_end_0, end_mask = var_38770_end_mask_0, x = q_173_cast_fp16)[name = tensor("op_38770_cast_fp16")]; + tensor var_38774_begin_0 = const()[name = tensor("op_38774_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_38774_end_0 = const()[name = tensor("op_38774_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_38774_end_mask_0 = const()[name = tensor("op_38774_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38774_cast_fp16 = slice_by_index(begin = var_38774_begin_0, end = var_38774_end_0, end_mask = var_38774_end_mask_0, x = q_173_cast_fp16)[name = tensor("op_38774_cast_fp16")]; + tensor var_38778_begin_0 = const()[name = tensor("op_38778_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_38778_end_0 = const()[name = tensor("op_38778_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_38778_end_mask_0 = const()[name = tensor("op_38778_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38778_cast_fp16 = slice_by_index(begin = var_38778_begin_0, end = var_38778_end_0, end_mask = var_38778_end_mask_0, x = q_173_cast_fp16)[name = tensor("op_38778_cast_fp16")]; + tensor var_38782_begin_0 = const()[name = tensor("op_38782_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_38782_end_0 = const()[name = tensor("op_38782_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_38782_end_mask_0 = const()[name = tensor("op_38782_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38782_cast_fp16 = slice_by_index(begin = var_38782_begin_0, end = var_38782_end_0, end_mask = var_38782_end_mask_0, x = q_173_cast_fp16)[name = tensor("op_38782_cast_fp16")]; + tensor k_347_perm_0 = const()[name = tensor("k_347_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_38789_begin_0 = const()[name = tensor("op_38789_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_38789_end_0 = const()[name = tensor("op_38789_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_38789_end_mask_0 = const()[name = tensor("op_38789_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_347_cast_fp16 = transpose(perm = k_347_perm_0, x = k_345_cast_fp16)[name = tensor("transpose_53")]; + tensor var_38789_cast_fp16 = slice_by_index(begin = var_38789_begin_0, end = var_38789_end_0, end_mask = var_38789_end_mask_0, x = k_347_cast_fp16)[name = tensor("op_38789_cast_fp16")]; + tensor var_38793_begin_0 = const()[name = tensor("op_38793_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_38793_end_0 = const()[name = tensor("op_38793_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_38793_end_mask_0 = const()[name = tensor("op_38793_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38793_cast_fp16 = slice_by_index(begin = var_38793_begin_0, end = var_38793_end_0, end_mask = var_38793_end_mask_0, x = k_347_cast_fp16)[name = tensor("op_38793_cast_fp16")]; + tensor var_38797_begin_0 = const()[name = tensor("op_38797_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_38797_end_0 = const()[name = tensor("op_38797_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_38797_end_mask_0 = const()[name = tensor("op_38797_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38797_cast_fp16 = slice_by_index(begin = var_38797_begin_0, end = var_38797_end_0, end_mask = var_38797_end_mask_0, x = k_347_cast_fp16)[name = tensor("op_38797_cast_fp16")]; + tensor var_38801_begin_0 = const()[name = tensor("op_38801_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_38801_end_0 = const()[name = tensor("op_38801_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_38801_end_mask_0 = const()[name = tensor("op_38801_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38801_cast_fp16 = slice_by_index(begin = var_38801_begin_0, end = var_38801_end_0, end_mask = var_38801_end_mask_0, x = k_347_cast_fp16)[name = tensor("op_38801_cast_fp16")]; + tensor var_38805_begin_0 = const()[name = tensor("op_38805_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_38805_end_0 = const()[name = tensor("op_38805_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_38805_end_mask_0 = const()[name = tensor("op_38805_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38805_cast_fp16 = slice_by_index(begin = var_38805_begin_0, end = var_38805_end_0, end_mask = var_38805_end_mask_0, x = k_347_cast_fp16)[name = tensor("op_38805_cast_fp16")]; + tensor var_38809_begin_0 = const()[name = tensor("op_38809_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_38809_end_0 = const()[name = tensor("op_38809_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_38809_end_mask_0 = const()[name = tensor("op_38809_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38809_cast_fp16 = slice_by_index(begin = var_38809_begin_0, end = var_38809_end_0, end_mask = var_38809_end_mask_0, x = k_347_cast_fp16)[name = tensor("op_38809_cast_fp16")]; + tensor var_38813_begin_0 = const()[name = tensor("op_38813_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_38813_end_0 = const()[name = tensor("op_38813_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_38813_end_mask_0 = const()[name = tensor("op_38813_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38813_cast_fp16 = slice_by_index(begin = var_38813_begin_0, end = var_38813_end_0, end_mask = var_38813_end_mask_0, x = k_347_cast_fp16)[name = tensor("op_38813_cast_fp16")]; + tensor var_38817_begin_0 = const()[name = tensor("op_38817_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_38817_end_0 = const()[name = tensor("op_38817_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_38817_end_mask_0 = const()[name = tensor("op_38817_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38817_cast_fp16 = slice_by_index(begin = var_38817_begin_0, end = var_38817_end_0, end_mask = var_38817_end_mask_0, x = k_347_cast_fp16)[name = tensor("op_38817_cast_fp16")]; + tensor var_38821_begin_0 = const()[name = tensor("op_38821_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_38821_end_0 = const()[name = tensor("op_38821_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_38821_end_mask_0 = const()[name = tensor("op_38821_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38821_cast_fp16 = slice_by_index(begin = var_38821_begin_0, end = var_38821_end_0, end_mask = var_38821_end_mask_0, x = k_347_cast_fp16)[name = tensor("op_38821_cast_fp16")]; + tensor var_38825_begin_0 = const()[name = tensor("op_38825_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_38825_end_0 = const()[name = tensor("op_38825_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_38825_end_mask_0 = const()[name = tensor("op_38825_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38825_cast_fp16 = slice_by_index(begin = var_38825_begin_0, end = var_38825_end_0, end_mask = var_38825_end_mask_0, x = k_347_cast_fp16)[name = tensor("op_38825_cast_fp16")]; + tensor var_38829_begin_0 = const()[name = tensor("op_38829_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_38829_end_0 = const()[name = tensor("op_38829_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_38829_end_mask_0 = const()[name = tensor("op_38829_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38829_cast_fp16 = slice_by_index(begin = var_38829_begin_0, end = var_38829_end_0, end_mask = var_38829_end_mask_0, x = k_347_cast_fp16)[name = tensor("op_38829_cast_fp16")]; + tensor var_38833_begin_0 = const()[name = tensor("op_38833_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_38833_end_0 = const()[name = tensor("op_38833_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_38833_end_mask_0 = const()[name = tensor("op_38833_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38833_cast_fp16 = slice_by_index(begin = var_38833_begin_0, end = var_38833_end_0, end_mask = var_38833_end_mask_0, x = k_347_cast_fp16)[name = tensor("op_38833_cast_fp16")]; + tensor var_38837_begin_0 = const()[name = tensor("op_38837_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_38837_end_0 = const()[name = tensor("op_38837_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_38837_end_mask_0 = const()[name = tensor("op_38837_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38837_cast_fp16 = slice_by_index(begin = var_38837_begin_0, end = var_38837_end_0, end_mask = var_38837_end_mask_0, x = k_347_cast_fp16)[name = tensor("op_38837_cast_fp16")]; + tensor var_38841_begin_0 = const()[name = tensor("op_38841_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_38841_end_0 = const()[name = tensor("op_38841_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_38841_end_mask_0 = const()[name = tensor("op_38841_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38841_cast_fp16 = slice_by_index(begin = var_38841_begin_0, end = var_38841_end_0, end_mask = var_38841_end_mask_0, x = k_347_cast_fp16)[name = tensor("op_38841_cast_fp16")]; + tensor var_38845_begin_0 = const()[name = tensor("op_38845_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_38845_end_0 = const()[name = tensor("op_38845_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_38845_end_mask_0 = const()[name = tensor("op_38845_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38845_cast_fp16 = slice_by_index(begin = var_38845_begin_0, end = var_38845_end_0, end_mask = var_38845_end_mask_0, x = k_347_cast_fp16)[name = tensor("op_38845_cast_fp16")]; + tensor var_38849_begin_0 = const()[name = tensor("op_38849_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_38849_end_0 = const()[name = tensor("op_38849_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_38849_end_mask_0 = const()[name = tensor("op_38849_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38849_cast_fp16 = slice_by_index(begin = var_38849_begin_0, end = var_38849_end_0, end_mask = var_38849_end_mask_0, x = k_347_cast_fp16)[name = tensor("op_38849_cast_fp16")]; + tensor var_38853_begin_0 = const()[name = tensor("op_38853_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_38853_end_0 = const()[name = tensor("op_38853_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_38853_end_mask_0 = const()[name = tensor("op_38853_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38853_cast_fp16 = slice_by_index(begin = var_38853_begin_0, end = var_38853_end_0, end_mask = var_38853_end_mask_0, x = k_347_cast_fp16)[name = tensor("op_38853_cast_fp16")]; + tensor var_38857_begin_0 = const()[name = tensor("op_38857_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_38857_end_0 = const()[name = tensor("op_38857_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_38857_end_mask_0 = const()[name = tensor("op_38857_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38857_cast_fp16 = slice_by_index(begin = var_38857_begin_0, end = var_38857_end_0, end_mask = var_38857_end_mask_0, x = k_347_cast_fp16)[name = tensor("op_38857_cast_fp16")]; + tensor var_38861_begin_0 = const()[name = tensor("op_38861_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_38861_end_0 = const()[name = tensor("op_38861_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_38861_end_mask_0 = const()[name = tensor("op_38861_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38861_cast_fp16 = slice_by_index(begin = var_38861_begin_0, end = var_38861_end_0, end_mask = var_38861_end_mask_0, x = k_347_cast_fp16)[name = tensor("op_38861_cast_fp16")]; + tensor var_38865_begin_0 = const()[name = tensor("op_38865_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_38865_end_0 = const()[name = tensor("op_38865_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_38865_end_mask_0 = const()[name = tensor("op_38865_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38865_cast_fp16 = slice_by_index(begin = var_38865_begin_0, end = var_38865_end_0, end_mask = var_38865_end_mask_0, x = k_347_cast_fp16)[name = tensor("op_38865_cast_fp16")]; + tensor var_38867_begin_0 = const()[name = tensor("op_38867_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_38867_end_0 = const()[name = tensor("op_38867_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_38867_end_mask_0 = const()[name = tensor("op_38867_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38867_cast_fp16 = slice_by_index(begin = var_38867_begin_0, end = var_38867_end_0, end_mask = var_38867_end_mask_0, x = v_173_cast_fp16)[name = tensor("op_38867_cast_fp16")]; + tensor var_38871_begin_0 = const()[name = tensor("op_38871_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_38871_end_0 = const()[name = tensor("op_38871_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_38871_end_mask_0 = const()[name = tensor("op_38871_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38871_cast_fp16 = slice_by_index(begin = var_38871_begin_0, end = var_38871_end_0, end_mask = var_38871_end_mask_0, x = v_173_cast_fp16)[name = tensor("op_38871_cast_fp16")]; + tensor var_38875_begin_0 = const()[name = tensor("op_38875_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_38875_end_0 = const()[name = tensor("op_38875_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_38875_end_mask_0 = const()[name = tensor("op_38875_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38875_cast_fp16 = slice_by_index(begin = var_38875_begin_0, end = var_38875_end_0, end_mask = var_38875_end_mask_0, x = v_173_cast_fp16)[name = tensor("op_38875_cast_fp16")]; + tensor var_38879_begin_0 = const()[name = tensor("op_38879_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_38879_end_0 = const()[name = tensor("op_38879_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_38879_end_mask_0 = const()[name = tensor("op_38879_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38879_cast_fp16 = slice_by_index(begin = var_38879_begin_0, end = var_38879_end_0, end_mask = var_38879_end_mask_0, x = v_173_cast_fp16)[name = tensor("op_38879_cast_fp16")]; + tensor var_38883_begin_0 = const()[name = tensor("op_38883_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_38883_end_0 = const()[name = tensor("op_38883_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_38883_end_mask_0 = const()[name = tensor("op_38883_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38883_cast_fp16 = slice_by_index(begin = var_38883_begin_0, end = var_38883_end_0, end_mask = var_38883_end_mask_0, x = v_173_cast_fp16)[name = tensor("op_38883_cast_fp16")]; + tensor var_38887_begin_0 = const()[name = tensor("op_38887_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_38887_end_0 = const()[name = tensor("op_38887_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_38887_end_mask_0 = const()[name = tensor("op_38887_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38887_cast_fp16 = slice_by_index(begin = var_38887_begin_0, end = var_38887_end_0, end_mask = var_38887_end_mask_0, x = v_173_cast_fp16)[name = tensor("op_38887_cast_fp16")]; + tensor var_38891_begin_0 = const()[name = tensor("op_38891_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_38891_end_0 = const()[name = tensor("op_38891_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_38891_end_mask_0 = const()[name = tensor("op_38891_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38891_cast_fp16 = slice_by_index(begin = var_38891_begin_0, end = var_38891_end_0, end_mask = var_38891_end_mask_0, x = v_173_cast_fp16)[name = tensor("op_38891_cast_fp16")]; + tensor var_38895_begin_0 = const()[name = tensor("op_38895_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_38895_end_0 = const()[name = tensor("op_38895_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_38895_end_mask_0 = const()[name = tensor("op_38895_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38895_cast_fp16 = slice_by_index(begin = var_38895_begin_0, end = var_38895_end_0, end_mask = var_38895_end_mask_0, x = v_173_cast_fp16)[name = tensor("op_38895_cast_fp16")]; + tensor var_38899_begin_0 = const()[name = tensor("op_38899_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_38899_end_0 = const()[name = tensor("op_38899_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_38899_end_mask_0 = const()[name = tensor("op_38899_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38899_cast_fp16 = slice_by_index(begin = var_38899_begin_0, end = var_38899_end_0, end_mask = var_38899_end_mask_0, x = v_173_cast_fp16)[name = tensor("op_38899_cast_fp16")]; + tensor var_38903_begin_0 = const()[name = tensor("op_38903_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_38903_end_0 = const()[name = tensor("op_38903_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_38903_end_mask_0 = const()[name = tensor("op_38903_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38903_cast_fp16 = slice_by_index(begin = var_38903_begin_0, end = var_38903_end_0, end_mask = var_38903_end_mask_0, x = v_173_cast_fp16)[name = tensor("op_38903_cast_fp16")]; + tensor var_38907_begin_0 = const()[name = tensor("op_38907_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_38907_end_0 = const()[name = tensor("op_38907_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_38907_end_mask_0 = const()[name = tensor("op_38907_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38907_cast_fp16 = slice_by_index(begin = var_38907_begin_0, end = var_38907_end_0, end_mask = var_38907_end_mask_0, x = v_173_cast_fp16)[name = tensor("op_38907_cast_fp16")]; + tensor var_38911_begin_0 = const()[name = tensor("op_38911_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_38911_end_0 = const()[name = tensor("op_38911_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_38911_end_mask_0 = const()[name = tensor("op_38911_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38911_cast_fp16 = slice_by_index(begin = var_38911_begin_0, end = var_38911_end_0, end_mask = var_38911_end_mask_0, x = v_173_cast_fp16)[name = tensor("op_38911_cast_fp16")]; + tensor var_38915_begin_0 = const()[name = tensor("op_38915_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_38915_end_0 = const()[name = tensor("op_38915_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_38915_end_mask_0 = const()[name = tensor("op_38915_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38915_cast_fp16 = slice_by_index(begin = var_38915_begin_0, end = var_38915_end_0, end_mask = var_38915_end_mask_0, x = v_173_cast_fp16)[name = tensor("op_38915_cast_fp16")]; + tensor var_38919_begin_0 = const()[name = tensor("op_38919_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_38919_end_0 = const()[name = tensor("op_38919_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_38919_end_mask_0 = const()[name = tensor("op_38919_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38919_cast_fp16 = slice_by_index(begin = var_38919_begin_0, end = var_38919_end_0, end_mask = var_38919_end_mask_0, x = v_173_cast_fp16)[name = tensor("op_38919_cast_fp16")]; + tensor var_38923_begin_0 = const()[name = tensor("op_38923_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_38923_end_0 = const()[name = tensor("op_38923_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_38923_end_mask_0 = const()[name = tensor("op_38923_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38923_cast_fp16 = slice_by_index(begin = var_38923_begin_0, end = var_38923_end_0, end_mask = var_38923_end_mask_0, x = v_173_cast_fp16)[name = tensor("op_38923_cast_fp16")]; + tensor var_38927_begin_0 = const()[name = tensor("op_38927_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_38927_end_0 = const()[name = tensor("op_38927_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_38927_end_mask_0 = const()[name = tensor("op_38927_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38927_cast_fp16 = slice_by_index(begin = var_38927_begin_0, end = var_38927_end_0, end_mask = var_38927_end_mask_0, x = v_173_cast_fp16)[name = tensor("op_38927_cast_fp16")]; + tensor var_38931_begin_0 = const()[name = tensor("op_38931_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_38931_end_0 = const()[name = tensor("op_38931_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_38931_end_mask_0 = const()[name = tensor("op_38931_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38931_cast_fp16 = slice_by_index(begin = var_38931_begin_0, end = var_38931_end_0, end_mask = var_38931_end_mask_0, x = v_173_cast_fp16)[name = tensor("op_38931_cast_fp16")]; + tensor var_38935_begin_0 = const()[name = tensor("op_38935_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_38935_end_0 = const()[name = tensor("op_38935_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_38935_end_mask_0 = const()[name = tensor("op_38935_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38935_cast_fp16 = slice_by_index(begin = var_38935_begin_0, end = var_38935_end_0, end_mask = var_38935_end_mask_0, x = v_173_cast_fp16)[name = tensor("op_38935_cast_fp16")]; + tensor var_38939_begin_0 = const()[name = tensor("op_38939_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_38939_end_0 = const()[name = tensor("op_38939_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_38939_end_mask_0 = const()[name = tensor("op_38939_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38939_cast_fp16 = slice_by_index(begin = var_38939_begin_0, end = var_38939_end_0, end_mask = var_38939_end_mask_0, x = v_173_cast_fp16)[name = tensor("op_38939_cast_fp16")]; + tensor var_38943_begin_0 = const()[name = tensor("op_38943_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_38943_end_0 = const()[name = tensor("op_38943_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_38943_end_mask_0 = const()[name = tensor("op_38943_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38943_cast_fp16 = slice_by_index(begin = var_38943_begin_0, end = var_38943_end_0, end_mask = var_38943_end_mask_0, x = v_173_cast_fp16)[name = tensor("op_38943_cast_fp16")]; + tensor var_38947_equation_0 = const()[name = tensor("op_38947_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38947_cast_fp16 = einsum(equation = var_38947_equation_0, values = (var_38789_cast_fp16, var_38706_cast_fp16))[name = tensor("op_38947_cast_fp16")]; + tensor var_38948_to_fp16 = const()[name = tensor("op_38948_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3281_cast_fp16 = mul(x = var_38947_cast_fp16, y = var_38948_to_fp16)[name = tensor("aw_3281_cast_fp16")]; + tensor var_38951_equation_0 = const()[name = tensor("op_38951_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38951_cast_fp16 = einsum(equation = var_38951_equation_0, values = (var_38793_cast_fp16, var_38710_cast_fp16))[name = tensor("op_38951_cast_fp16")]; + tensor var_38952_to_fp16 = const()[name = tensor("op_38952_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3283_cast_fp16 = mul(x = var_38951_cast_fp16, y = var_38952_to_fp16)[name = tensor("aw_3283_cast_fp16")]; + tensor var_38955_equation_0 = const()[name = tensor("op_38955_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38955_cast_fp16 = einsum(equation = var_38955_equation_0, values = (var_38797_cast_fp16, var_38714_cast_fp16))[name = tensor("op_38955_cast_fp16")]; + tensor var_38956_to_fp16 = const()[name = tensor("op_38956_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3285_cast_fp16 = mul(x = var_38955_cast_fp16, y = var_38956_to_fp16)[name = tensor("aw_3285_cast_fp16")]; + tensor var_38959_equation_0 = const()[name = tensor("op_38959_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38959_cast_fp16 = einsum(equation = var_38959_equation_0, values = (var_38801_cast_fp16, var_38718_cast_fp16))[name = tensor("op_38959_cast_fp16")]; + tensor var_38960_to_fp16 = const()[name = tensor("op_38960_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3287_cast_fp16 = mul(x = var_38959_cast_fp16, y = var_38960_to_fp16)[name = tensor("aw_3287_cast_fp16")]; + tensor var_38963_equation_0 = const()[name = tensor("op_38963_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38963_cast_fp16 = einsum(equation = var_38963_equation_0, values = (var_38805_cast_fp16, var_38722_cast_fp16))[name = tensor("op_38963_cast_fp16")]; + tensor var_38964_to_fp16 = const()[name = tensor("op_38964_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3289_cast_fp16 = mul(x = var_38963_cast_fp16, y = var_38964_to_fp16)[name = tensor("aw_3289_cast_fp16")]; + tensor var_38967_equation_0 = const()[name = tensor("op_38967_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38967_cast_fp16 = einsum(equation = var_38967_equation_0, values = (var_38809_cast_fp16, var_38726_cast_fp16))[name = tensor("op_38967_cast_fp16")]; + tensor var_38968_to_fp16 = const()[name = tensor("op_38968_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3291_cast_fp16 = mul(x = var_38967_cast_fp16, y = var_38968_to_fp16)[name = tensor("aw_3291_cast_fp16")]; + tensor var_38971_equation_0 = const()[name = tensor("op_38971_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38971_cast_fp16 = einsum(equation = var_38971_equation_0, values = (var_38813_cast_fp16, var_38730_cast_fp16))[name = tensor("op_38971_cast_fp16")]; + tensor var_38972_to_fp16 = const()[name = tensor("op_38972_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3293_cast_fp16 = mul(x = var_38971_cast_fp16, y = var_38972_to_fp16)[name = tensor("aw_3293_cast_fp16")]; + tensor var_38975_equation_0 = const()[name = tensor("op_38975_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38975_cast_fp16 = einsum(equation = var_38975_equation_0, values = (var_38817_cast_fp16, var_38734_cast_fp16))[name = tensor("op_38975_cast_fp16")]; + tensor var_38976_to_fp16 = const()[name = tensor("op_38976_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3295_cast_fp16 = mul(x = var_38975_cast_fp16, y = var_38976_to_fp16)[name = tensor("aw_3295_cast_fp16")]; + tensor var_38979_equation_0 = const()[name = tensor("op_38979_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38979_cast_fp16 = einsum(equation = var_38979_equation_0, values = (var_38821_cast_fp16, var_38738_cast_fp16))[name = tensor("op_38979_cast_fp16")]; + tensor var_38980_to_fp16 = const()[name = tensor("op_38980_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3297_cast_fp16 = mul(x = var_38979_cast_fp16, y = var_38980_to_fp16)[name = tensor("aw_3297_cast_fp16")]; + tensor var_38983_equation_0 = const()[name = tensor("op_38983_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38983_cast_fp16 = einsum(equation = var_38983_equation_0, values = (var_38825_cast_fp16, var_38742_cast_fp16))[name = tensor("op_38983_cast_fp16")]; + tensor var_38984_to_fp16 = const()[name = tensor("op_38984_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3299_cast_fp16 = mul(x = var_38983_cast_fp16, y = var_38984_to_fp16)[name = tensor("aw_3299_cast_fp16")]; + tensor var_38987_equation_0 = const()[name = tensor("op_38987_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38987_cast_fp16 = einsum(equation = var_38987_equation_0, values = (var_38829_cast_fp16, var_38746_cast_fp16))[name = tensor("op_38987_cast_fp16")]; + tensor var_38988_to_fp16 = const()[name = tensor("op_38988_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3301_cast_fp16 = mul(x = var_38987_cast_fp16, y = var_38988_to_fp16)[name = tensor("aw_3301_cast_fp16")]; + tensor var_38991_equation_0 = const()[name = tensor("op_38991_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38991_cast_fp16 = einsum(equation = var_38991_equation_0, values = (var_38833_cast_fp16, var_38750_cast_fp16))[name = tensor("op_38991_cast_fp16")]; + tensor var_38992_to_fp16 = const()[name = tensor("op_38992_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3303_cast_fp16 = mul(x = var_38991_cast_fp16, y = var_38992_to_fp16)[name = tensor("aw_3303_cast_fp16")]; + tensor var_38995_equation_0 = const()[name = tensor("op_38995_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38995_cast_fp16 = einsum(equation = var_38995_equation_0, values = (var_38837_cast_fp16, var_38754_cast_fp16))[name = tensor("op_38995_cast_fp16")]; + tensor var_38996_to_fp16 = const()[name = tensor("op_38996_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3305_cast_fp16 = mul(x = var_38995_cast_fp16, y = var_38996_to_fp16)[name = tensor("aw_3305_cast_fp16")]; + tensor var_38999_equation_0 = const()[name = tensor("op_38999_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38999_cast_fp16 = einsum(equation = var_38999_equation_0, values = (var_38841_cast_fp16, var_38758_cast_fp16))[name = tensor("op_38999_cast_fp16")]; + tensor var_39000_to_fp16 = const()[name = tensor("op_39000_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3307_cast_fp16 = mul(x = var_38999_cast_fp16, y = var_39000_to_fp16)[name = tensor("aw_3307_cast_fp16")]; + tensor var_39003_equation_0 = const()[name = tensor("op_39003_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_39003_cast_fp16 = einsum(equation = var_39003_equation_0, values = (var_38845_cast_fp16, var_38762_cast_fp16))[name = tensor("op_39003_cast_fp16")]; + tensor var_39004_to_fp16 = const()[name = tensor("op_39004_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3309_cast_fp16 = mul(x = var_39003_cast_fp16, y = var_39004_to_fp16)[name = tensor("aw_3309_cast_fp16")]; + tensor var_39007_equation_0 = const()[name = tensor("op_39007_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_39007_cast_fp16 = einsum(equation = var_39007_equation_0, values = (var_38849_cast_fp16, var_38766_cast_fp16))[name = tensor("op_39007_cast_fp16")]; + tensor var_39008_to_fp16 = const()[name = tensor("op_39008_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3311_cast_fp16 = mul(x = var_39007_cast_fp16, y = var_39008_to_fp16)[name = tensor("aw_3311_cast_fp16")]; + tensor var_39011_equation_0 = const()[name = tensor("op_39011_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_39011_cast_fp16 = einsum(equation = var_39011_equation_0, values = (var_38853_cast_fp16, var_38770_cast_fp16))[name = tensor("op_39011_cast_fp16")]; + tensor var_39012_to_fp16 = const()[name = tensor("op_39012_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3313_cast_fp16 = mul(x = var_39011_cast_fp16, y = var_39012_to_fp16)[name = tensor("aw_3313_cast_fp16")]; + tensor var_39015_equation_0 = const()[name = tensor("op_39015_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_39015_cast_fp16 = einsum(equation = var_39015_equation_0, values = (var_38857_cast_fp16, var_38774_cast_fp16))[name = tensor("op_39015_cast_fp16")]; + tensor var_39016_to_fp16 = const()[name = tensor("op_39016_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3315_cast_fp16 = mul(x = var_39015_cast_fp16, y = var_39016_to_fp16)[name = tensor("aw_3315_cast_fp16")]; + tensor var_39019_equation_0 = const()[name = tensor("op_39019_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_39019_cast_fp16 = einsum(equation = var_39019_equation_0, values = (var_38861_cast_fp16, var_38778_cast_fp16))[name = tensor("op_39019_cast_fp16")]; + tensor var_39020_to_fp16 = const()[name = tensor("op_39020_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3317_cast_fp16 = mul(x = var_39019_cast_fp16, y = var_39020_to_fp16)[name = tensor("aw_3317_cast_fp16")]; + tensor var_39023_equation_0 = const()[name = tensor("op_39023_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_39023_cast_fp16 = einsum(equation = var_39023_equation_0, values = (var_38865_cast_fp16, var_38782_cast_fp16))[name = tensor("op_39023_cast_fp16")]; + tensor var_39024_to_fp16 = const()[name = tensor("op_39024_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3319_cast_fp16 = mul(x = var_39023_cast_fp16, y = var_39024_to_fp16)[name = tensor("aw_3319_cast_fp16")]; + tensor var_39026_cast_fp16 = softmax(axis = var_30355, x = aw_3281_cast_fp16)[name = tensor("op_39026_cast_fp16")]; + tensor var_39027_cast_fp16 = softmax(axis = var_30355, x = aw_3283_cast_fp16)[name = tensor("op_39027_cast_fp16")]; + tensor var_39028_cast_fp16 = softmax(axis = var_30355, x = aw_3285_cast_fp16)[name = tensor("op_39028_cast_fp16")]; + tensor var_39029_cast_fp16 = softmax(axis = var_30355, x = aw_3287_cast_fp16)[name = tensor("op_39029_cast_fp16")]; + tensor var_39030_cast_fp16 = softmax(axis = var_30355, x = aw_3289_cast_fp16)[name = tensor("op_39030_cast_fp16")]; + tensor var_39031_cast_fp16 = softmax(axis = var_30355, x = aw_3291_cast_fp16)[name = tensor("op_39031_cast_fp16")]; + tensor var_39032_cast_fp16 = softmax(axis = var_30355, x = aw_3293_cast_fp16)[name = tensor("op_39032_cast_fp16")]; + tensor var_39033_cast_fp16 = softmax(axis = var_30355, x = aw_3295_cast_fp16)[name = tensor("op_39033_cast_fp16")]; + tensor var_39034_cast_fp16 = softmax(axis = var_30355, x = aw_3297_cast_fp16)[name = tensor("op_39034_cast_fp16")]; + tensor var_39035_cast_fp16 = softmax(axis = var_30355, x = aw_3299_cast_fp16)[name = tensor("op_39035_cast_fp16")]; + tensor var_39036_cast_fp16 = softmax(axis = var_30355, x = aw_3301_cast_fp16)[name = tensor("op_39036_cast_fp16")]; + tensor var_39037_cast_fp16 = softmax(axis = var_30355, x = aw_3303_cast_fp16)[name = tensor("op_39037_cast_fp16")]; + tensor var_39038_cast_fp16 = softmax(axis = var_30355, x = aw_3305_cast_fp16)[name = tensor("op_39038_cast_fp16")]; + tensor var_39039_cast_fp16 = softmax(axis = var_30355, x = aw_3307_cast_fp16)[name = tensor("op_39039_cast_fp16")]; + tensor var_39040_cast_fp16 = softmax(axis = var_30355, x = aw_3309_cast_fp16)[name = tensor("op_39040_cast_fp16")]; + tensor var_39041_cast_fp16 = softmax(axis = var_30355, x = aw_3311_cast_fp16)[name = tensor("op_39041_cast_fp16")]; + tensor var_39042_cast_fp16 = softmax(axis = var_30355, x = aw_3313_cast_fp16)[name = tensor("op_39042_cast_fp16")]; + tensor var_39043_cast_fp16 = softmax(axis = var_30355, x = aw_3315_cast_fp16)[name = tensor("op_39043_cast_fp16")]; + tensor var_39044_cast_fp16 = softmax(axis = var_30355, x = aw_3317_cast_fp16)[name = tensor("op_39044_cast_fp16")]; + tensor var_39045_cast_fp16 = softmax(axis = var_30355, x = aw_3319_cast_fp16)[name = tensor("op_39045_cast_fp16")]; + tensor var_39047_equation_0 = const()[name = tensor("op_39047_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39047_cast_fp16 = einsum(equation = var_39047_equation_0, values = (var_38867_cast_fp16, var_39026_cast_fp16))[name = tensor("op_39047_cast_fp16")]; + tensor var_39049_equation_0 = const()[name = tensor("op_39049_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39049_cast_fp16 = einsum(equation = var_39049_equation_0, values = (var_38871_cast_fp16, var_39027_cast_fp16))[name = tensor("op_39049_cast_fp16")]; + tensor var_39051_equation_0 = const()[name = tensor("op_39051_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39051_cast_fp16 = einsum(equation = var_39051_equation_0, values = (var_38875_cast_fp16, var_39028_cast_fp16))[name = tensor("op_39051_cast_fp16")]; + tensor var_39053_equation_0 = const()[name = tensor("op_39053_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39053_cast_fp16 = einsum(equation = var_39053_equation_0, values = (var_38879_cast_fp16, var_39029_cast_fp16))[name = tensor("op_39053_cast_fp16")]; + tensor var_39055_equation_0 = const()[name = tensor("op_39055_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39055_cast_fp16 = einsum(equation = var_39055_equation_0, values = (var_38883_cast_fp16, var_39030_cast_fp16))[name = tensor("op_39055_cast_fp16")]; + tensor var_39057_equation_0 = const()[name = tensor("op_39057_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39057_cast_fp16 = einsum(equation = var_39057_equation_0, values = (var_38887_cast_fp16, var_39031_cast_fp16))[name = tensor("op_39057_cast_fp16")]; + tensor var_39059_equation_0 = const()[name = tensor("op_39059_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39059_cast_fp16 = einsum(equation = var_39059_equation_0, values = (var_38891_cast_fp16, var_39032_cast_fp16))[name = tensor("op_39059_cast_fp16")]; + tensor var_39061_equation_0 = const()[name = tensor("op_39061_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39061_cast_fp16 = einsum(equation = var_39061_equation_0, values = (var_38895_cast_fp16, var_39033_cast_fp16))[name = tensor("op_39061_cast_fp16")]; + tensor var_39063_equation_0 = const()[name = tensor("op_39063_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39063_cast_fp16 = einsum(equation = var_39063_equation_0, values = (var_38899_cast_fp16, var_39034_cast_fp16))[name = tensor("op_39063_cast_fp16")]; + tensor var_39065_equation_0 = const()[name = tensor("op_39065_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39065_cast_fp16 = einsum(equation = var_39065_equation_0, values = (var_38903_cast_fp16, var_39035_cast_fp16))[name = tensor("op_39065_cast_fp16")]; + tensor var_39067_equation_0 = const()[name = tensor("op_39067_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39067_cast_fp16 = einsum(equation = var_39067_equation_0, values = (var_38907_cast_fp16, var_39036_cast_fp16))[name = tensor("op_39067_cast_fp16")]; + tensor var_39069_equation_0 = const()[name = tensor("op_39069_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39069_cast_fp16 = einsum(equation = var_39069_equation_0, values = (var_38911_cast_fp16, var_39037_cast_fp16))[name = tensor("op_39069_cast_fp16")]; + tensor var_39071_equation_0 = const()[name = tensor("op_39071_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39071_cast_fp16 = einsum(equation = var_39071_equation_0, values = (var_38915_cast_fp16, var_39038_cast_fp16))[name = tensor("op_39071_cast_fp16")]; + tensor var_39073_equation_0 = const()[name = tensor("op_39073_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39073_cast_fp16 = einsum(equation = var_39073_equation_0, values = (var_38919_cast_fp16, var_39039_cast_fp16))[name = tensor("op_39073_cast_fp16")]; + tensor var_39075_equation_0 = const()[name = tensor("op_39075_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39075_cast_fp16 = einsum(equation = var_39075_equation_0, values = (var_38923_cast_fp16, var_39040_cast_fp16))[name = tensor("op_39075_cast_fp16")]; + tensor var_39077_equation_0 = const()[name = tensor("op_39077_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39077_cast_fp16 = einsum(equation = var_39077_equation_0, values = (var_38927_cast_fp16, var_39041_cast_fp16))[name = tensor("op_39077_cast_fp16")]; + tensor var_39079_equation_0 = const()[name = tensor("op_39079_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39079_cast_fp16 = einsum(equation = var_39079_equation_0, values = (var_38931_cast_fp16, var_39042_cast_fp16))[name = tensor("op_39079_cast_fp16")]; + tensor var_39081_equation_0 = const()[name = tensor("op_39081_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39081_cast_fp16 = einsum(equation = var_39081_equation_0, values = (var_38935_cast_fp16, var_39043_cast_fp16))[name = tensor("op_39081_cast_fp16")]; + tensor var_39083_equation_0 = const()[name = tensor("op_39083_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39083_cast_fp16 = einsum(equation = var_39083_equation_0, values = (var_38939_cast_fp16, var_39044_cast_fp16))[name = tensor("op_39083_cast_fp16")]; + tensor var_39085_equation_0 = const()[name = tensor("op_39085_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39085_cast_fp16 = einsum(equation = var_39085_equation_0, values = (var_38943_cast_fp16, var_39045_cast_fp16))[name = tensor("op_39085_cast_fp16")]; + tensor input_511_interleave_0 = const()[name = tensor("input_511_interleave_0"), val = tensor(false)]; + tensor input_511_cast_fp16 = concat(axis = var_30355, interleave = input_511_interleave_0, values = (var_39047_cast_fp16, var_39049_cast_fp16, var_39051_cast_fp16, var_39053_cast_fp16, var_39055_cast_fp16, var_39057_cast_fp16, var_39059_cast_fp16, var_39061_cast_fp16, var_39063_cast_fp16, var_39065_cast_fp16, var_39067_cast_fp16, var_39069_cast_fp16, var_39071_cast_fp16, var_39073_cast_fp16, var_39075_cast_fp16, var_39077_cast_fp16, var_39079_cast_fp16, var_39081_cast_fp16, var_39083_cast_fp16, var_39085_cast_fp16))[name = tensor("input_511_cast_fp16")]; + tensor var_39091 = const()[name = tensor("op_39091"), val = tensor([1, 1])]; + tensor var_39093 = const()[name = tensor("op_39093"), val = tensor([1, 1])]; + tensor var_39095_pad_type_0 = const()[name = tensor("op_39095_pad_type_0"), val = tensor("custom")]; + tensor var_39095_pad_0 = const()[name = tensor("op_39095_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_9_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_9_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3223873856)))]; + tensor up_blocks_0_attentions_0_transformer_blocks_9_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_9_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3227150720)))]; + tensor var_39095_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_9_attn1_to_out_0_bias_to_fp16, dilations = var_39093, groups = var_30355, pad = var_39095_pad_0, pad_type = var_39095_pad_type_0, strides = var_39091, weight = up_blocks_0_attentions_0_transformer_blocks_9_attn1_to_out_0_weight_to_fp16, x = input_511_cast_fp16)[name = tensor("op_39095_cast_fp16")]; + tensor inputs_261_cast_fp16 = add(x = var_39095_cast_fp16, y = inputs_259_cast_fp16)[name = tensor("inputs_261_cast_fp16")]; + tensor hidden_states_343_axes_0 = const()[name = tensor("hidden_states_343_axes_0"), val = tensor([1])]; + tensor hidden_states_343_gamma_0_to_fp16 = const()[name = tensor("hidden_states_343_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3227153344)))]; + tensor hidden_states_343_beta_0_to_fp16 = const()[name = tensor("hidden_states_343_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3227155968)))]; + tensor var_39105_to_fp16 = const()[name = tensor("op_39105_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_343_cast_fp16 = layer_norm(axes = hidden_states_343_axes_0, beta = hidden_states_343_beta_0_to_fp16, epsilon = var_39105_to_fp16, gamma = hidden_states_343_gamma_0_to_fp16, x = inputs_261_cast_fp16)[name = tensor("hidden_states_343_cast_fp16")]; + tensor var_39120 = const()[name = tensor("op_39120"), val = tensor([1, 1])]; + tensor var_39122 = const()[name = tensor("op_39122"), val = tensor([1, 1])]; + tensor q_175_pad_type_0 = const()[name = tensor("q_175_pad_type_0"), val = tensor("custom")]; + tensor q_175_pad_0 = const()[name = tensor("q_175_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_9_attn2_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_9_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3227158592)))]; + tensor q_175_cast_fp16 = conv(dilations = var_39122, groups = var_30355, pad = q_175_pad_0, pad_type = q_175_pad_type_0, strides = var_39120, weight = up_blocks_0_attentions_0_transformer_blocks_9_attn2_to_q_weight_to_fp16, x = hidden_states_343_cast_fp16)[name = tensor("q_175_cast_fp16")]; + tensor var_39126 = const()[name = tensor("op_39126"), val = tensor([1, 1])]; + tensor var_39128 = const()[name = tensor("op_39128"), val = tensor([1, 1])]; + tensor k_349_pad_type_0 = const()[name = tensor("k_349_pad_type_0"), val = tensor("custom")]; + tensor k_349_pad_0 = const()[name = tensor("k_349_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_9_attn2_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_9_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3230435456)))]; + tensor k_349_cast_fp16 = conv(dilations = var_39128, groups = var_30355, pad = k_349_pad_0, pad_type = k_349_pad_type_0, strides = var_39126, weight = up_blocks_0_attentions_0_transformer_blocks_9_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_349_cast_fp16")]; + tensor var_39132 = const()[name = tensor("op_39132"), val = tensor([1, 1])]; + tensor var_39134 = const()[name = tensor("op_39134"), val = tensor([1, 1])]; + tensor v_175_pad_type_0 = const()[name = tensor("v_175_pad_type_0"), val = tensor("custom")]; + tensor v_175_pad_0 = const()[name = tensor("v_175_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_9_attn2_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_9_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3235678400)))]; + tensor v_175_cast_fp16 = conv(dilations = var_39134, groups = var_30355, pad = v_175_pad_0, pad_type = v_175_pad_type_0, strides = var_39132, weight = up_blocks_0_attentions_0_transformer_blocks_9_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_175_cast_fp16")]; + tensor var_39138_begin_0 = const()[name = tensor("op_39138_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_39138_end_0 = const()[name = tensor("op_39138_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_39138_end_mask_0 = const()[name = tensor("op_39138_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39138_cast_fp16 = slice_by_index(begin = var_39138_begin_0, end = var_39138_end_0, end_mask = var_39138_end_mask_0, x = q_175_cast_fp16)[name = tensor("op_39138_cast_fp16")]; + tensor var_39142_begin_0 = const()[name = tensor("op_39142_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_39142_end_0 = const()[name = tensor("op_39142_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_39142_end_mask_0 = const()[name = tensor("op_39142_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39142_cast_fp16 = slice_by_index(begin = var_39142_begin_0, end = var_39142_end_0, end_mask = var_39142_end_mask_0, x = q_175_cast_fp16)[name = tensor("op_39142_cast_fp16")]; + tensor var_39146_begin_0 = const()[name = tensor("op_39146_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_39146_end_0 = const()[name = tensor("op_39146_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_39146_end_mask_0 = const()[name = tensor("op_39146_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39146_cast_fp16 = slice_by_index(begin = var_39146_begin_0, end = var_39146_end_0, end_mask = var_39146_end_mask_0, x = q_175_cast_fp16)[name = tensor("op_39146_cast_fp16")]; + tensor var_39150_begin_0 = const()[name = tensor("op_39150_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_39150_end_0 = const()[name = tensor("op_39150_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_39150_end_mask_0 = const()[name = tensor("op_39150_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39150_cast_fp16 = slice_by_index(begin = var_39150_begin_0, end = var_39150_end_0, end_mask = var_39150_end_mask_0, x = q_175_cast_fp16)[name = tensor("op_39150_cast_fp16")]; + tensor var_39154_begin_0 = const()[name = tensor("op_39154_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_39154_end_0 = const()[name = tensor("op_39154_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_39154_end_mask_0 = const()[name = tensor("op_39154_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39154_cast_fp16 = slice_by_index(begin = var_39154_begin_0, end = var_39154_end_0, end_mask = var_39154_end_mask_0, x = q_175_cast_fp16)[name = tensor("op_39154_cast_fp16")]; + tensor var_39158_begin_0 = const()[name = tensor("op_39158_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_39158_end_0 = const()[name = tensor("op_39158_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_39158_end_mask_0 = const()[name = tensor("op_39158_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39158_cast_fp16 = slice_by_index(begin = var_39158_begin_0, end = var_39158_end_0, end_mask = var_39158_end_mask_0, x = q_175_cast_fp16)[name = tensor("op_39158_cast_fp16")]; + tensor var_39162_begin_0 = const()[name = tensor("op_39162_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_39162_end_0 = const()[name = tensor("op_39162_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_39162_end_mask_0 = const()[name = tensor("op_39162_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39162_cast_fp16 = slice_by_index(begin = var_39162_begin_0, end = var_39162_end_0, end_mask = var_39162_end_mask_0, x = q_175_cast_fp16)[name = tensor("op_39162_cast_fp16")]; + tensor var_39166_begin_0 = const()[name = tensor("op_39166_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_39166_end_0 = const()[name = tensor("op_39166_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_39166_end_mask_0 = const()[name = tensor("op_39166_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39166_cast_fp16 = slice_by_index(begin = var_39166_begin_0, end = var_39166_end_0, end_mask = var_39166_end_mask_0, x = q_175_cast_fp16)[name = tensor("op_39166_cast_fp16")]; + tensor var_39170_begin_0 = const()[name = tensor("op_39170_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_39170_end_0 = const()[name = tensor("op_39170_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_39170_end_mask_0 = const()[name = tensor("op_39170_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39170_cast_fp16 = slice_by_index(begin = var_39170_begin_0, end = var_39170_end_0, end_mask = var_39170_end_mask_0, x = q_175_cast_fp16)[name = tensor("op_39170_cast_fp16")]; + tensor var_39174_begin_0 = const()[name = tensor("op_39174_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_39174_end_0 = const()[name = tensor("op_39174_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_39174_end_mask_0 = const()[name = tensor("op_39174_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39174_cast_fp16 = slice_by_index(begin = var_39174_begin_0, end = var_39174_end_0, end_mask = var_39174_end_mask_0, x = q_175_cast_fp16)[name = tensor("op_39174_cast_fp16")]; + tensor var_39178_begin_0 = const()[name = tensor("op_39178_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_39178_end_0 = const()[name = tensor("op_39178_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_39178_end_mask_0 = const()[name = tensor("op_39178_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39178_cast_fp16 = slice_by_index(begin = var_39178_begin_0, end = var_39178_end_0, end_mask = var_39178_end_mask_0, x = q_175_cast_fp16)[name = tensor("op_39178_cast_fp16")]; + tensor var_39182_begin_0 = const()[name = tensor("op_39182_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_39182_end_0 = const()[name = tensor("op_39182_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_39182_end_mask_0 = const()[name = tensor("op_39182_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39182_cast_fp16 = slice_by_index(begin = var_39182_begin_0, end = var_39182_end_0, end_mask = var_39182_end_mask_0, x = q_175_cast_fp16)[name = tensor("op_39182_cast_fp16")]; + tensor var_39186_begin_0 = const()[name = tensor("op_39186_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_39186_end_0 = const()[name = tensor("op_39186_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_39186_end_mask_0 = const()[name = tensor("op_39186_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39186_cast_fp16 = slice_by_index(begin = var_39186_begin_0, end = var_39186_end_0, end_mask = var_39186_end_mask_0, x = q_175_cast_fp16)[name = tensor("op_39186_cast_fp16")]; + tensor var_39190_begin_0 = const()[name = tensor("op_39190_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_39190_end_0 = const()[name = tensor("op_39190_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_39190_end_mask_0 = const()[name = tensor("op_39190_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39190_cast_fp16 = slice_by_index(begin = var_39190_begin_0, end = var_39190_end_0, end_mask = var_39190_end_mask_0, x = q_175_cast_fp16)[name = tensor("op_39190_cast_fp16")]; + tensor var_39194_begin_0 = const()[name = tensor("op_39194_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_39194_end_0 = const()[name = tensor("op_39194_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_39194_end_mask_0 = const()[name = tensor("op_39194_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39194_cast_fp16 = slice_by_index(begin = var_39194_begin_0, end = var_39194_end_0, end_mask = var_39194_end_mask_0, x = q_175_cast_fp16)[name = tensor("op_39194_cast_fp16")]; + tensor var_39198_begin_0 = const()[name = tensor("op_39198_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_39198_end_0 = const()[name = tensor("op_39198_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_39198_end_mask_0 = const()[name = tensor("op_39198_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39198_cast_fp16 = slice_by_index(begin = var_39198_begin_0, end = var_39198_end_0, end_mask = var_39198_end_mask_0, x = q_175_cast_fp16)[name = tensor("op_39198_cast_fp16")]; + tensor var_39202_begin_0 = const()[name = tensor("op_39202_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_39202_end_0 = const()[name = tensor("op_39202_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_39202_end_mask_0 = const()[name = tensor("op_39202_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39202_cast_fp16 = slice_by_index(begin = var_39202_begin_0, end = var_39202_end_0, end_mask = var_39202_end_mask_0, x = q_175_cast_fp16)[name = tensor("op_39202_cast_fp16")]; + tensor var_39206_begin_0 = const()[name = tensor("op_39206_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_39206_end_0 = const()[name = tensor("op_39206_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_39206_end_mask_0 = const()[name = tensor("op_39206_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39206_cast_fp16 = slice_by_index(begin = var_39206_begin_0, end = var_39206_end_0, end_mask = var_39206_end_mask_0, x = q_175_cast_fp16)[name = tensor("op_39206_cast_fp16")]; + tensor var_39210_begin_0 = const()[name = tensor("op_39210_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_39210_end_0 = const()[name = tensor("op_39210_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_39210_end_mask_0 = const()[name = tensor("op_39210_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39210_cast_fp16 = slice_by_index(begin = var_39210_begin_0, end = var_39210_end_0, end_mask = var_39210_end_mask_0, x = q_175_cast_fp16)[name = tensor("op_39210_cast_fp16")]; + tensor var_39214_begin_0 = const()[name = tensor("op_39214_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_39214_end_0 = const()[name = tensor("op_39214_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_39214_end_mask_0 = const()[name = tensor("op_39214_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39214_cast_fp16 = slice_by_index(begin = var_39214_begin_0, end = var_39214_end_0, end_mask = var_39214_end_mask_0, x = q_175_cast_fp16)[name = tensor("op_39214_cast_fp16")]; + tensor k_351_perm_0 = const()[name = tensor("k_351_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_39221_begin_0 = const()[name = tensor("op_39221_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_39221_end_0 = const()[name = tensor("op_39221_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_39221_end_mask_0 = const()[name = tensor("op_39221_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_351_cast_fp16 = transpose(perm = k_351_perm_0, x = k_349_cast_fp16)[name = tensor("transpose_52")]; + tensor var_39221_cast_fp16 = slice_by_index(begin = var_39221_begin_0, end = var_39221_end_0, end_mask = var_39221_end_mask_0, x = k_351_cast_fp16)[name = tensor("op_39221_cast_fp16")]; + tensor var_39225_begin_0 = const()[name = tensor("op_39225_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_39225_end_0 = const()[name = tensor("op_39225_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_39225_end_mask_0 = const()[name = tensor("op_39225_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39225_cast_fp16 = slice_by_index(begin = var_39225_begin_0, end = var_39225_end_0, end_mask = var_39225_end_mask_0, x = k_351_cast_fp16)[name = tensor("op_39225_cast_fp16")]; + tensor var_39229_begin_0 = const()[name = tensor("op_39229_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_39229_end_0 = const()[name = tensor("op_39229_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_39229_end_mask_0 = const()[name = tensor("op_39229_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39229_cast_fp16 = slice_by_index(begin = var_39229_begin_0, end = var_39229_end_0, end_mask = var_39229_end_mask_0, x = k_351_cast_fp16)[name = tensor("op_39229_cast_fp16")]; + tensor var_39233_begin_0 = const()[name = tensor("op_39233_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_39233_end_0 = const()[name = tensor("op_39233_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_39233_end_mask_0 = const()[name = tensor("op_39233_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39233_cast_fp16 = slice_by_index(begin = var_39233_begin_0, end = var_39233_end_0, end_mask = var_39233_end_mask_0, x = k_351_cast_fp16)[name = tensor("op_39233_cast_fp16")]; + tensor var_39237_begin_0 = const()[name = tensor("op_39237_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_39237_end_0 = const()[name = tensor("op_39237_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_39237_end_mask_0 = const()[name = tensor("op_39237_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39237_cast_fp16 = slice_by_index(begin = var_39237_begin_0, end = var_39237_end_0, end_mask = var_39237_end_mask_0, x = k_351_cast_fp16)[name = tensor("op_39237_cast_fp16")]; + tensor var_39241_begin_0 = const()[name = tensor("op_39241_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_39241_end_0 = const()[name = tensor("op_39241_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_39241_end_mask_0 = const()[name = tensor("op_39241_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39241_cast_fp16 = slice_by_index(begin = var_39241_begin_0, end = var_39241_end_0, end_mask = var_39241_end_mask_0, x = k_351_cast_fp16)[name = tensor("op_39241_cast_fp16")]; + tensor var_39245_begin_0 = const()[name = tensor("op_39245_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_39245_end_0 = const()[name = tensor("op_39245_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_39245_end_mask_0 = const()[name = tensor("op_39245_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39245_cast_fp16 = slice_by_index(begin = var_39245_begin_0, end = var_39245_end_0, end_mask = var_39245_end_mask_0, x = k_351_cast_fp16)[name = tensor("op_39245_cast_fp16")]; + tensor var_39249_begin_0 = const()[name = tensor("op_39249_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_39249_end_0 = const()[name = tensor("op_39249_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_39249_end_mask_0 = const()[name = tensor("op_39249_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39249_cast_fp16 = slice_by_index(begin = var_39249_begin_0, end = var_39249_end_0, end_mask = var_39249_end_mask_0, x = k_351_cast_fp16)[name = tensor("op_39249_cast_fp16")]; + tensor var_39253_begin_0 = const()[name = tensor("op_39253_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_39253_end_0 = const()[name = tensor("op_39253_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_39253_end_mask_0 = const()[name = tensor("op_39253_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39253_cast_fp16 = slice_by_index(begin = var_39253_begin_0, end = var_39253_end_0, end_mask = var_39253_end_mask_0, x = k_351_cast_fp16)[name = tensor("op_39253_cast_fp16")]; + tensor var_39257_begin_0 = const()[name = tensor("op_39257_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_39257_end_0 = const()[name = tensor("op_39257_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_39257_end_mask_0 = const()[name = tensor("op_39257_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39257_cast_fp16 = slice_by_index(begin = var_39257_begin_0, end = var_39257_end_0, end_mask = var_39257_end_mask_0, x = k_351_cast_fp16)[name = tensor("op_39257_cast_fp16")]; + tensor var_39261_begin_0 = const()[name = tensor("op_39261_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_39261_end_0 = const()[name = tensor("op_39261_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_39261_end_mask_0 = const()[name = tensor("op_39261_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39261_cast_fp16 = slice_by_index(begin = var_39261_begin_0, end = var_39261_end_0, end_mask = var_39261_end_mask_0, x = k_351_cast_fp16)[name = tensor("op_39261_cast_fp16")]; + tensor var_39265_begin_0 = const()[name = tensor("op_39265_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_39265_end_0 = const()[name = tensor("op_39265_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_39265_end_mask_0 = const()[name = tensor("op_39265_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39265_cast_fp16 = slice_by_index(begin = var_39265_begin_0, end = var_39265_end_0, end_mask = var_39265_end_mask_0, x = k_351_cast_fp16)[name = tensor("op_39265_cast_fp16")]; + tensor var_39269_begin_0 = const()[name = tensor("op_39269_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_39269_end_0 = const()[name = tensor("op_39269_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_39269_end_mask_0 = const()[name = tensor("op_39269_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39269_cast_fp16 = slice_by_index(begin = var_39269_begin_0, end = var_39269_end_0, end_mask = var_39269_end_mask_0, x = k_351_cast_fp16)[name = tensor("op_39269_cast_fp16")]; + tensor var_39273_begin_0 = const()[name = tensor("op_39273_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_39273_end_0 = const()[name = tensor("op_39273_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_39273_end_mask_0 = const()[name = tensor("op_39273_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39273_cast_fp16 = slice_by_index(begin = var_39273_begin_0, end = var_39273_end_0, end_mask = var_39273_end_mask_0, x = k_351_cast_fp16)[name = tensor("op_39273_cast_fp16")]; + tensor var_39277_begin_0 = const()[name = tensor("op_39277_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_39277_end_0 = const()[name = tensor("op_39277_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_39277_end_mask_0 = const()[name = tensor("op_39277_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39277_cast_fp16 = slice_by_index(begin = var_39277_begin_0, end = var_39277_end_0, end_mask = var_39277_end_mask_0, x = k_351_cast_fp16)[name = tensor("op_39277_cast_fp16")]; + tensor var_39281_begin_0 = const()[name = tensor("op_39281_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_39281_end_0 = const()[name = tensor("op_39281_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_39281_end_mask_0 = const()[name = tensor("op_39281_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39281_cast_fp16 = slice_by_index(begin = var_39281_begin_0, end = var_39281_end_0, end_mask = var_39281_end_mask_0, x = k_351_cast_fp16)[name = tensor("op_39281_cast_fp16")]; + tensor var_39285_begin_0 = const()[name = tensor("op_39285_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_39285_end_0 = const()[name = tensor("op_39285_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_39285_end_mask_0 = const()[name = tensor("op_39285_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39285_cast_fp16 = slice_by_index(begin = var_39285_begin_0, end = var_39285_end_0, end_mask = var_39285_end_mask_0, x = k_351_cast_fp16)[name = tensor("op_39285_cast_fp16")]; + tensor var_39289_begin_0 = const()[name = tensor("op_39289_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_39289_end_0 = const()[name = tensor("op_39289_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_39289_end_mask_0 = const()[name = tensor("op_39289_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39289_cast_fp16 = slice_by_index(begin = var_39289_begin_0, end = var_39289_end_0, end_mask = var_39289_end_mask_0, x = k_351_cast_fp16)[name = tensor("op_39289_cast_fp16")]; + tensor var_39293_begin_0 = const()[name = tensor("op_39293_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_39293_end_0 = const()[name = tensor("op_39293_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_39293_end_mask_0 = const()[name = tensor("op_39293_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39293_cast_fp16 = slice_by_index(begin = var_39293_begin_0, end = var_39293_end_0, end_mask = var_39293_end_mask_0, x = k_351_cast_fp16)[name = tensor("op_39293_cast_fp16")]; + tensor var_39297_begin_0 = const()[name = tensor("op_39297_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_39297_end_0 = const()[name = tensor("op_39297_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_39297_end_mask_0 = const()[name = tensor("op_39297_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39297_cast_fp16 = slice_by_index(begin = var_39297_begin_0, end = var_39297_end_0, end_mask = var_39297_end_mask_0, x = k_351_cast_fp16)[name = tensor("op_39297_cast_fp16")]; + tensor var_39299_begin_0 = const()[name = tensor("op_39299_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_39299_end_0 = const()[name = tensor("op_39299_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_39299_end_mask_0 = const()[name = tensor("op_39299_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39299_cast_fp16 = slice_by_index(begin = var_39299_begin_0, end = var_39299_end_0, end_mask = var_39299_end_mask_0, x = v_175_cast_fp16)[name = tensor("op_39299_cast_fp16")]; + tensor var_39303_begin_0 = const()[name = tensor("op_39303_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_39303_end_0 = const()[name = tensor("op_39303_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_39303_end_mask_0 = const()[name = tensor("op_39303_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39303_cast_fp16 = slice_by_index(begin = var_39303_begin_0, end = var_39303_end_0, end_mask = var_39303_end_mask_0, x = v_175_cast_fp16)[name = tensor("op_39303_cast_fp16")]; + tensor var_39307_begin_0 = const()[name = tensor("op_39307_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_39307_end_0 = const()[name = tensor("op_39307_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_39307_end_mask_0 = const()[name = tensor("op_39307_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39307_cast_fp16 = slice_by_index(begin = var_39307_begin_0, end = var_39307_end_0, end_mask = var_39307_end_mask_0, x = v_175_cast_fp16)[name = tensor("op_39307_cast_fp16")]; + tensor var_39311_begin_0 = const()[name = tensor("op_39311_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_39311_end_0 = const()[name = tensor("op_39311_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_39311_end_mask_0 = const()[name = tensor("op_39311_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39311_cast_fp16 = slice_by_index(begin = var_39311_begin_0, end = var_39311_end_0, end_mask = var_39311_end_mask_0, x = v_175_cast_fp16)[name = tensor("op_39311_cast_fp16")]; + tensor var_39315_begin_0 = const()[name = tensor("op_39315_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_39315_end_0 = const()[name = tensor("op_39315_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_39315_end_mask_0 = const()[name = tensor("op_39315_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39315_cast_fp16 = slice_by_index(begin = var_39315_begin_0, end = var_39315_end_0, end_mask = var_39315_end_mask_0, x = v_175_cast_fp16)[name = tensor("op_39315_cast_fp16")]; + tensor var_39319_begin_0 = const()[name = tensor("op_39319_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_39319_end_0 = const()[name = tensor("op_39319_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_39319_end_mask_0 = const()[name = tensor("op_39319_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39319_cast_fp16 = slice_by_index(begin = var_39319_begin_0, end = var_39319_end_0, end_mask = var_39319_end_mask_0, x = v_175_cast_fp16)[name = tensor("op_39319_cast_fp16")]; + tensor var_39323_begin_0 = const()[name = tensor("op_39323_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_39323_end_0 = const()[name = tensor("op_39323_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_39323_end_mask_0 = const()[name = tensor("op_39323_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39323_cast_fp16 = slice_by_index(begin = var_39323_begin_0, end = var_39323_end_0, end_mask = var_39323_end_mask_0, x = v_175_cast_fp16)[name = tensor("op_39323_cast_fp16")]; + tensor var_39327_begin_0 = const()[name = tensor("op_39327_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_39327_end_0 = const()[name = tensor("op_39327_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_39327_end_mask_0 = const()[name = tensor("op_39327_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39327_cast_fp16 = slice_by_index(begin = var_39327_begin_0, end = var_39327_end_0, end_mask = var_39327_end_mask_0, x = v_175_cast_fp16)[name = tensor("op_39327_cast_fp16")]; + tensor var_39331_begin_0 = const()[name = tensor("op_39331_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_39331_end_0 = const()[name = tensor("op_39331_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_39331_end_mask_0 = const()[name = tensor("op_39331_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39331_cast_fp16 = slice_by_index(begin = var_39331_begin_0, end = var_39331_end_0, end_mask = var_39331_end_mask_0, x = v_175_cast_fp16)[name = tensor("op_39331_cast_fp16")]; + tensor var_39335_begin_0 = const()[name = tensor("op_39335_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_39335_end_0 = const()[name = tensor("op_39335_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_39335_end_mask_0 = const()[name = tensor("op_39335_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39335_cast_fp16 = slice_by_index(begin = var_39335_begin_0, end = var_39335_end_0, end_mask = var_39335_end_mask_0, x = v_175_cast_fp16)[name = tensor("op_39335_cast_fp16")]; + tensor var_39339_begin_0 = const()[name = tensor("op_39339_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_39339_end_0 = const()[name = tensor("op_39339_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_39339_end_mask_0 = const()[name = tensor("op_39339_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39339_cast_fp16 = slice_by_index(begin = var_39339_begin_0, end = var_39339_end_0, end_mask = var_39339_end_mask_0, x = v_175_cast_fp16)[name = tensor("op_39339_cast_fp16")]; + tensor var_39343_begin_0 = const()[name = tensor("op_39343_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_39343_end_0 = const()[name = tensor("op_39343_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_39343_end_mask_0 = const()[name = tensor("op_39343_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39343_cast_fp16 = slice_by_index(begin = var_39343_begin_0, end = var_39343_end_0, end_mask = var_39343_end_mask_0, x = v_175_cast_fp16)[name = tensor("op_39343_cast_fp16")]; + tensor var_39347_begin_0 = const()[name = tensor("op_39347_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_39347_end_0 = const()[name = tensor("op_39347_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_39347_end_mask_0 = const()[name = tensor("op_39347_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39347_cast_fp16 = slice_by_index(begin = var_39347_begin_0, end = var_39347_end_0, end_mask = var_39347_end_mask_0, x = v_175_cast_fp16)[name = tensor("op_39347_cast_fp16")]; + tensor var_39351_begin_0 = const()[name = tensor("op_39351_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_39351_end_0 = const()[name = tensor("op_39351_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_39351_end_mask_0 = const()[name = tensor("op_39351_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39351_cast_fp16 = slice_by_index(begin = var_39351_begin_0, end = var_39351_end_0, end_mask = var_39351_end_mask_0, x = v_175_cast_fp16)[name = tensor("op_39351_cast_fp16")]; + tensor var_39355_begin_0 = const()[name = tensor("op_39355_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_39355_end_0 = const()[name = tensor("op_39355_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_39355_end_mask_0 = const()[name = tensor("op_39355_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39355_cast_fp16 = slice_by_index(begin = var_39355_begin_0, end = var_39355_end_0, end_mask = var_39355_end_mask_0, x = v_175_cast_fp16)[name = tensor("op_39355_cast_fp16")]; + tensor var_39359_begin_0 = const()[name = tensor("op_39359_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_39359_end_0 = const()[name = tensor("op_39359_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_39359_end_mask_0 = const()[name = tensor("op_39359_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39359_cast_fp16 = slice_by_index(begin = var_39359_begin_0, end = var_39359_end_0, end_mask = var_39359_end_mask_0, x = v_175_cast_fp16)[name = tensor("op_39359_cast_fp16")]; + tensor var_39363_begin_0 = const()[name = tensor("op_39363_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_39363_end_0 = const()[name = tensor("op_39363_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_39363_end_mask_0 = const()[name = tensor("op_39363_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39363_cast_fp16 = slice_by_index(begin = var_39363_begin_0, end = var_39363_end_0, end_mask = var_39363_end_mask_0, x = v_175_cast_fp16)[name = tensor("op_39363_cast_fp16")]; + tensor var_39367_begin_0 = const()[name = tensor("op_39367_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_39367_end_0 = const()[name = tensor("op_39367_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_39367_end_mask_0 = const()[name = tensor("op_39367_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39367_cast_fp16 = slice_by_index(begin = var_39367_begin_0, end = var_39367_end_0, end_mask = var_39367_end_mask_0, x = v_175_cast_fp16)[name = tensor("op_39367_cast_fp16")]; + tensor var_39371_begin_0 = const()[name = tensor("op_39371_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_39371_end_0 = const()[name = tensor("op_39371_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_39371_end_mask_0 = const()[name = tensor("op_39371_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39371_cast_fp16 = slice_by_index(begin = var_39371_begin_0, end = var_39371_end_0, end_mask = var_39371_end_mask_0, x = v_175_cast_fp16)[name = tensor("op_39371_cast_fp16")]; + tensor var_39375_begin_0 = const()[name = tensor("op_39375_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_39375_end_0 = const()[name = tensor("op_39375_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_39375_end_mask_0 = const()[name = tensor("op_39375_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39375_cast_fp16 = slice_by_index(begin = var_39375_begin_0, end = var_39375_end_0, end_mask = var_39375_end_mask_0, x = v_175_cast_fp16)[name = tensor("op_39375_cast_fp16")]; + tensor var_39379_equation_0 = const()[name = tensor("op_39379_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_39379_cast_fp16 = einsum(equation = var_39379_equation_0, values = (var_39221_cast_fp16, var_39138_cast_fp16))[name = tensor("op_39379_cast_fp16")]; + tensor var_39380_to_fp16 = const()[name = tensor("op_39380_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3321_cast_fp16 = mul(x = var_39379_cast_fp16, y = var_39380_to_fp16)[name = tensor("aw_3321_cast_fp16")]; + tensor var_39383_equation_0 = const()[name = tensor("op_39383_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_39383_cast_fp16 = einsum(equation = var_39383_equation_0, values = (var_39225_cast_fp16, var_39142_cast_fp16))[name = tensor("op_39383_cast_fp16")]; + tensor var_39384_to_fp16 = const()[name = tensor("op_39384_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3323_cast_fp16 = mul(x = var_39383_cast_fp16, y = var_39384_to_fp16)[name = tensor("aw_3323_cast_fp16")]; + tensor var_39387_equation_0 = const()[name = tensor("op_39387_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_39387_cast_fp16 = einsum(equation = var_39387_equation_0, values = (var_39229_cast_fp16, var_39146_cast_fp16))[name = tensor("op_39387_cast_fp16")]; + tensor var_39388_to_fp16 = const()[name = tensor("op_39388_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3325_cast_fp16 = mul(x = var_39387_cast_fp16, y = var_39388_to_fp16)[name = tensor("aw_3325_cast_fp16")]; + tensor var_39391_equation_0 = const()[name = tensor("op_39391_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_39391_cast_fp16 = einsum(equation = var_39391_equation_0, values = (var_39233_cast_fp16, var_39150_cast_fp16))[name = tensor("op_39391_cast_fp16")]; + tensor var_39392_to_fp16 = const()[name = tensor("op_39392_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3327_cast_fp16 = mul(x = var_39391_cast_fp16, y = var_39392_to_fp16)[name = tensor("aw_3327_cast_fp16")]; + tensor var_39395_equation_0 = const()[name = tensor("op_39395_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_39395_cast_fp16 = einsum(equation = var_39395_equation_0, values = (var_39237_cast_fp16, var_39154_cast_fp16))[name = tensor("op_39395_cast_fp16")]; + tensor var_39396_to_fp16 = const()[name = tensor("op_39396_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3329_cast_fp16 = mul(x = var_39395_cast_fp16, y = var_39396_to_fp16)[name = tensor("aw_3329_cast_fp16")]; + tensor var_39399_equation_0 = const()[name = tensor("op_39399_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_39399_cast_fp16 = einsum(equation = var_39399_equation_0, values = (var_39241_cast_fp16, var_39158_cast_fp16))[name = tensor("op_39399_cast_fp16")]; + tensor var_39400_to_fp16 = const()[name = tensor("op_39400_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3331_cast_fp16 = mul(x = var_39399_cast_fp16, y = var_39400_to_fp16)[name = tensor("aw_3331_cast_fp16")]; + tensor var_39403_equation_0 = const()[name = tensor("op_39403_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_39403_cast_fp16 = einsum(equation = var_39403_equation_0, values = (var_39245_cast_fp16, var_39162_cast_fp16))[name = tensor("op_39403_cast_fp16")]; + tensor var_39404_to_fp16 = const()[name = tensor("op_39404_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3333_cast_fp16 = mul(x = var_39403_cast_fp16, y = var_39404_to_fp16)[name = tensor("aw_3333_cast_fp16")]; + tensor var_39407_equation_0 = const()[name = tensor("op_39407_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_39407_cast_fp16 = einsum(equation = var_39407_equation_0, values = (var_39249_cast_fp16, var_39166_cast_fp16))[name = tensor("op_39407_cast_fp16")]; + tensor var_39408_to_fp16 = const()[name = tensor("op_39408_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3335_cast_fp16 = mul(x = var_39407_cast_fp16, y = var_39408_to_fp16)[name = tensor("aw_3335_cast_fp16")]; + tensor var_39411_equation_0 = const()[name = tensor("op_39411_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_39411_cast_fp16 = einsum(equation = var_39411_equation_0, values = (var_39253_cast_fp16, var_39170_cast_fp16))[name = tensor("op_39411_cast_fp16")]; + tensor var_39412_to_fp16 = const()[name = tensor("op_39412_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3337_cast_fp16 = mul(x = var_39411_cast_fp16, y = var_39412_to_fp16)[name = tensor("aw_3337_cast_fp16")]; + tensor var_39415_equation_0 = const()[name = tensor("op_39415_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_39415_cast_fp16 = einsum(equation = var_39415_equation_0, values = (var_39257_cast_fp16, var_39174_cast_fp16))[name = tensor("op_39415_cast_fp16")]; + tensor var_39416_to_fp16 = const()[name = tensor("op_39416_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3339_cast_fp16 = mul(x = var_39415_cast_fp16, y = var_39416_to_fp16)[name = tensor("aw_3339_cast_fp16")]; + tensor var_39419_equation_0 = const()[name = tensor("op_39419_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_39419_cast_fp16 = einsum(equation = var_39419_equation_0, values = (var_39261_cast_fp16, var_39178_cast_fp16))[name = tensor("op_39419_cast_fp16")]; + tensor var_39420_to_fp16 = const()[name = tensor("op_39420_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3341_cast_fp16 = mul(x = var_39419_cast_fp16, y = var_39420_to_fp16)[name = tensor("aw_3341_cast_fp16")]; + tensor var_39423_equation_0 = const()[name = tensor("op_39423_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_39423_cast_fp16 = einsum(equation = var_39423_equation_0, values = (var_39265_cast_fp16, var_39182_cast_fp16))[name = tensor("op_39423_cast_fp16")]; + tensor var_39424_to_fp16 = const()[name = tensor("op_39424_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3343_cast_fp16 = mul(x = var_39423_cast_fp16, y = var_39424_to_fp16)[name = tensor("aw_3343_cast_fp16")]; + tensor var_39427_equation_0 = const()[name = tensor("op_39427_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_39427_cast_fp16 = einsum(equation = var_39427_equation_0, values = (var_39269_cast_fp16, var_39186_cast_fp16))[name = tensor("op_39427_cast_fp16")]; + tensor var_39428_to_fp16 = const()[name = tensor("op_39428_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3345_cast_fp16 = mul(x = var_39427_cast_fp16, y = var_39428_to_fp16)[name = tensor("aw_3345_cast_fp16")]; + tensor var_39431_equation_0 = const()[name = tensor("op_39431_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_39431_cast_fp16 = einsum(equation = var_39431_equation_0, values = (var_39273_cast_fp16, var_39190_cast_fp16))[name = tensor("op_39431_cast_fp16")]; + tensor var_39432_to_fp16 = const()[name = tensor("op_39432_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3347_cast_fp16 = mul(x = var_39431_cast_fp16, y = var_39432_to_fp16)[name = tensor("aw_3347_cast_fp16")]; + tensor var_39435_equation_0 = const()[name = tensor("op_39435_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_39435_cast_fp16 = einsum(equation = var_39435_equation_0, values = (var_39277_cast_fp16, var_39194_cast_fp16))[name = tensor("op_39435_cast_fp16")]; + tensor var_39436_to_fp16 = const()[name = tensor("op_39436_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3349_cast_fp16 = mul(x = var_39435_cast_fp16, y = var_39436_to_fp16)[name = tensor("aw_3349_cast_fp16")]; + tensor var_39439_equation_0 = const()[name = tensor("op_39439_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_39439_cast_fp16 = einsum(equation = var_39439_equation_0, values = (var_39281_cast_fp16, var_39198_cast_fp16))[name = tensor("op_39439_cast_fp16")]; + tensor var_39440_to_fp16 = const()[name = tensor("op_39440_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3351_cast_fp16 = mul(x = var_39439_cast_fp16, y = var_39440_to_fp16)[name = tensor("aw_3351_cast_fp16")]; + tensor var_39443_equation_0 = const()[name = tensor("op_39443_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_39443_cast_fp16 = einsum(equation = var_39443_equation_0, values = (var_39285_cast_fp16, var_39202_cast_fp16))[name = tensor("op_39443_cast_fp16")]; + tensor var_39444_to_fp16 = const()[name = tensor("op_39444_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3353_cast_fp16 = mul(x = var_39443_cast_fp16, y = var_39444_to_fp16)[name = tensor("aw_3353_cast_fp16")]; + tensor var_39447_equation_0 = const()[name = tensor("op_39447_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_39447_cast_fp16 = einsum(equation = var_39447_equation_0, values = (var_39289_cast_fp16, var_39206_cast_fp16))[name = tensor("op_39447_cast_fp16")]; + tensor var_39448_to_fp16 = const()[name = tensor("op_39448_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3355_cast_fp16 = mul(x = var_39447_cast_fp16, y = var_39448_to_fp16)[name = tensor("aw_3355_cast_fp16")]; + tensor var_39451_equation_0 = const()[name = tensor("op_39451_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_39451_cast_fp16 = einsum(equation = var_39451_equation_0, values = (var_39293_cast_fp16, var_39210_cast_fp16))[name = tensor("op_39451_cast_fp16")]; + tensor var_39452_to_fp16 = const()[name = tensor("op_39452_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3357_cast_fp16 = mul(x = var_39451_cast_fp16, y = var_39452_to_fp16)[name = tensor("aw_3357_cast_fp16")]; + tensor var_39455_equation_0 = const()[name = tensor("op_39455_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_39455_cast_fp16 = einsum(equation = var_39455_equation_0, values = (var_39297_cast_fp16, var_39214_cast_fp16))[name = tensor("op_39455_cast_fp16")]; + tensor var_39456_to_fp16 = const()[name = tensor("op_39456_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3359_cast_fp16 = mul(x = var_39455_cast_fp16, y = var_39456_to_fp16)[name = tensor("aw_3359_cast_fp16")]; + tensor var_39458_cast_fp16 = softmax(axis = var_30355, x = aw_3321_cast_fp16)[name = tensor("op_39458_cast_fp16")]; + tensor var_39459_cast_fp16 = softmax(axis = var_30355, x = aw_3323_cast_fp16)[name = tensor("op_39459_cast_fp16")]; + tensor var_39460_cast_fp16 = softmax(axis = var_30355, x = aw_3325_cast_fp16)[name = tensor("op_39460_cast_fp16")]; + tensor var_39461_cast_fp16 = softmax(axis = var_30355, x = aw_3327_cast_fp16)[name = tensor("op_39461_cast_fp16")]; + tensor var_39462_cast_fp16 = softmax(axis = var_30355, x = aw_3329_cast_fp16)[name = tensor("op_39462_cast_fp16")]; + tensor var_39463_cast_fp16 = softmax(axis = var_30355, x = aw_3331_cast_fp16)[name = tensor("op_39463_cast_fp16")]; + tensor var_39464_cast_fp16 = softmax(axis = var_30355, x = aw_3333_cast_fp16)[name = tensor("op_39464_cast_fp16")]; + tensor var_39465_cast_fp16 = softmax(axis = var_30355, x = aw_3335_cast_fp16)[name = tensor("op_39465_cast_fp16")]; + tensor var_39466_cast_fp16 = softmax(axis = var_30355, x = aw_3337_cast_fp16)[name = tensor("op_39466_cast_fp16")]; + tensor var_39467_cast_fp16 = softmax(axis = var_30355, x = aw_3339_cast_fp16)[name = tensor("op_39467_cast_fp16")]; + tensor var_39468_cast_fp16 = softmax(axis = var_30355, x = aw_3341_cast_fp16)[name = tensor("op_39468_cast_fp16")]; + tensor var_39469_cast_fp16 = softmax(axis = var_30355, x = aw_3343_cast_fp16)[name = tensor("op_39469_cast_fp16")]; + tensor var_39470_cast_fp16 = softmax(axis = var_30355, x = aw_3345_cast_fp16)[name = tensor("op_39470_cast_fp16")]; + tensor var_39471_cast_fp16 = softmax(axis = var_30355, x = aw_3347_cast_fp16)[name = tensor("op_39471_cast_fp16")]; + tensor var_39472_cast_fp16 = softmax(axis = var_30355, x = aw_3349_cast_fp16)[name = tensor("op_39472_cast_fp16")]; + tensor var_39473_cast_fp16 = softmax(axis = var_30355, x = aw_3351_cast_fp16)[name = tensor("op_39473_cast_fp16")]; + tensor var_39474_cast_fp16 = softmax(axis = var_30355, x = aw_3353_cast_fp16)[name = tensor("op_39474_cast_fp16")]; + tensor var_39475_cast_fp16 = softmax(axis = var_30355, x = aw_3355_cast_fp16)[name = tensor("op_39475_cast_fp16")]; + tensor var_39476_cast_fp16 = softmax(axis = var_30355, x = aw_3357_cast_fp16)[name = tensor("op_39476_cast_fp16")]; + tensor var_39477_cast_fp16 = softmax(axis = var_30355, x = aw_3359_cast_fp16)[name = tensor("op_39477_cast_fp16")]; + tensor var_39479_equation_0 = const()[name = tensor("op_39479_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39479_cast_fp16 = einsum(equation = var_39479_equation_0, values = (var_39299_cast_fp16, var_39458_cast_fp16))[name = tensor("op_39479_cast_fp16")]; + tensor var_39481_equation_0 = const()[name = tensor("op_39481_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39481_cast_fp16 = einsum(equation = var_39481_equation_0, values = (var_39303_cast_fp16, var_39459_cast_fp16))[name = tensor("op_39481_cast_fp16")]; + tensor var_39483_equation_0 = const()[name = tensor("op_39483_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39483_cast_fp16 = einsum(equation = var_39483_equation_0, values = (var_39307_cast_fp16, var_39460_cast_fp16))[name = tensor("op_39483_cast_fp16")]; + tensor var_39485_equation_0 = const()[name = tensor("op_39485_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39485_cast_fp16 = einsum(equation = var_39485_equation_0, values = (var_39311_cast_fp16, var_39461_cast_fp16))[name = tensor("op_39485_cast_fp16")]; + tensor var_39487_equation_0 = const()[name = tensor("op_39487_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39487_cast_fp16 = einsum(equation = var_39487_equation_0, values = (var_39315_cast_fp16, var_39462_cast_fp16))[name = tensor("op_39487_cast_fp16")]; + tensor var_39489_equation_0 = const()[name = tensor("op_39489_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39489_cast_fp16 = einsum(equation = var_39489_equation_0, values = (var_39319_cast_fp16, var_39463_cast_fp16))[name = tensor("op_39489_cast_fp16")]; + tensor var_39491_equation_0 = const()[name = tensor("op_39491_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39491_cast_fp16 = einsum(equation = var_39491_equation_0, values = (var_39323_cast_fp16, var_39464_cast_fp16))[name = tensor("op_39491_cast_fp16")]; + tensor var_39493_equation_0 = const()[name = tensor("op_39493_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39493_cast_fp16 = einsum(equation = var_39493_equation_0, values = (var_39327_cast_fp16, var_39465_cast_fp16))[name = tensor("op_39493_cast_fp16")]; + tensor var_39495_equation_0 = const()[name = tensor("op_39495_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39495_cast_fp16 = einsum(equation = var_39495_equation_0, values = (var_39331_cast_fp16, var_39466_cast_fp16))[name = tensor("op_39495_cast_fp16")]; + tensor var_39497_equation_0 = const()[name = tensor("op_39497_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39497_cast_fp16 = einsum(equation = var_39497_equation_0, values = (var_39335_cast_fp16, var_39467_cast_fp16))[name = tensor("op_39497_cast_fp16")]; + tensor var_39499_equation_0 = const()[name = tensor("op_39499_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39499_cast_fp16 = einsum(equation = var_39499_equation_0, values = (var_39339_cast_fp16, var_39468_cast_fp16))[name = tensor("op_39499_cast_fp16")]; + tensor var_39501_equation_0 = const()[name = tensor("op_39501_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39501_cast_fp16 = einsum(equation = var_39501_equation_0, values = (var_39343_cast_fp16, var_39469_cast_fp16))[name = tensor("op_39501_cast_fp16")]; + tensor var_39503_equation_0 = const()[name = tensor("op_39503_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39503_cast_fp16 = einsum(equation = var_39503_equation_0, values = (var_39347_cast_fp16, var_39470_cast_fp16))[name = tensor("op_39503_cast_fp16")]; + tensor var_39505_equation_0 = const()[name = tensor("op_39505_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39505_cast_fp16 = einsum(equation = var_39505_equation_0, values = (var_39351_cast_fp16, var_39471_cast_fp16))[name = tensor("op_39505_cast_fp16")]; + tensor var_39507_equation_0 = const()[name = tensor("op_39507_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39507_cast_fp16 = einsum(equation = var_39507_equation_0, values = (var_39355_cast_fp16, var_39472_cast_fp16))[name = tensor("op_39507_cast_fp16")]; + tensor var_39509_equation_0 = const()[name = tensor("op_39509_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39509_cast_fp16 = einsum(equation = var_39509_equation_0, values = (var_39359_cast_fp16, var_39473_cast_fp16))[name = tensor("op_39509_cast_fp16")]; + tensor var_39511_equation_0 = const()[name = tensor("op_39511_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39511_cast_fp16 = einsum(equation = var_39511_equation_0, values = (var_39363_cast_fp16, var_39474_cast_fp16))[name = tensor("op_39511_cast_fp16")]; + tensor var_39513_equation_0 = const()[name = tensor("op_39513_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39513_cast_fp16 = einsum(equation = var_39513_equation_0, values = (var_39367_cast_fp16, var_39475_cast_fp16))[name = tensor("op_39513_cast_fp16")]; + tensor var_39515_equation_0 = const()[name = tensor("op_39515_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39515_cast_fp16 = einsum(equation = var_39515_equation_0, values = (var_39371_cast_fp16, var_39476_cast_fp16))[name = tensor("op_39515_cast_fp16")]; + tensor var_39517_equation_0 = const()[name = tensor("op_39517_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39517_cast_fp16 = einsum(equation = var_39517_equation_0, values = (var_39375_cast_fp16, var_39477_cast_fp16))[name = tensor("op_39517_cast_fp16")]; + tensor input_513_interleave_0 = const()[name = tensor("input_513_interleave_0"), val = tensor(false)]; + tensor input_513_cast_fp16 = concat(axis = var_30355, interleave = input_513_interleave_0, values = (var_39479_cast_fp16, var_39481_cast_fp16, var_39483_cast_fp16, var_39485_cast_fp16, var_39487_cast_fp16, var_39489_cast_fp16, var_39491_cast_fp16, var_39493_cast_fp16, var_39495_cast_fp16, var_39497_cast_fp16, var_39499_cast_fp16, var_39501_cast_fp16, var_39503_cast_fp16, var_39505_cast_fp16, var_39507_cast_fp16, var_39509_cast_fp16, var_39511_cast_fp16, var_39513_cast_fp16, var_39515_cast_fp16, var_39517_cast_fp16))[name = tensor("input_513_cast_fp16")]; + tensor var_39523 = const()[name = tensor("op_39523"), val = tensor([1, 1])]; + tensor var_39525 = const()[name = tensor("op_39525"), val = tensor([1, 1])]; + tensor var_39527_pad_type_0 = const()[name = tensor("op_39527_pad_type_0"), val = tensor("custom")]; + tensor var_39527_pad_0 = const()[name = tensor("op_39527_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_9_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_9_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3240921344)))]; + tensor up_blocks_0_attentions_0_transformer_blocks_9_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_9_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3244198208)))]; + tensor var_39527_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_9_attn2_to_out_0_bias_to_fp16, dilations = var_39525, groups = var_30355, pad = var_39527_pad_0, pad_type = var_39527_pad_type_0, strides = var_39523, weight = up_blocks_0_attentions_0_transformer_blocks_9_attn2_to_out_0_weight_to_fp16, x = input_513_cast_fp16)[name = tensor("op_39527_cast_fp16")]; + tensor inputs_263_cast_fp16 = add(x = var_39527_cast_fp16, y = inputs_261_cast_fp16)[name = tensor("inputs_263_cast_fp16")]; + tensor input_515_axes_0 = const()[name = tensor("input_515_axes_0"), val = tensor([1])]; + tensor input_515_gamma_0_to_fp16 = const()[name = tensor("input_515_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3244200832)))]; + tensor input_515_beta_0_to_fp16 = const()[name = tensor("input_515_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3244203456)))]; + tensor var_39537_to_fp16 = const()[name = tensor("op_39537_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_515_cast_fp16 = layer_norm(axes = input_515_axes_0, beta = input_515_beta_0_to_fp16, epsilon = var_39537_to_fp16, gamma = input_515_gamma_0_to_fp16, x = inputs_263_cast_fp16)[name = tensor("input_515_cast_fp16")]; + tensor var_39553 = const()[name = tensor("op_39553"), val = tensor([1, 1])]; + tensor var_39555 = const()[name = tensor("op_39555"), val = tensor([1, 1])]; + tensor var_39557_pad_type_0 = const()[name = tensor("op_39557_pad_type_0"), val = tensor("custom")]; + tensor var_39557_pad_0 = const()[name = tensor("op_39557_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_9_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_9_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3244206080)))]; + tensor up_blocks_0_attentions_0_transformer_blocks_9_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_9_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3270420544)))]; + tensor var_39557_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_9_ff_net_0_proj_bias_to_fp16, dilations = var_39555, groups = var_30355, pad = var_39557_pad_0, pad_type = var_39557_pad_type_0, strides = var_39553, weight = up_blocks_0_attentions_0_transformer_blocks_9_ff_net_0_proj_weight_to_fp16, x = input_515_cast_fp16)[name = tensor("op_39557_cast_fp16")]; + tensor var_39558_split_sizes_0 = const()[name = tensor("op_39558_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_39558_axis_0 = const()[name = tensor("op_39558_axis_0"), val = tensor(1)]; + tensor var_39558_cast_fp16_0, tensor var_39558_cast_fp16_1 = split(axis = var_39558_axis_0, split_sizes = var_39558_split_sizes_0, x = var_39557_cast_fp16)[name = tensor("op_39558_cast_fp16")]; + tensor var_39560_mode_0 = const()[name = tensor("op_39560_mode_0"), val = tensor("EXACT")]; + tensor var_39560_cast_fp16 = gelu(mode = var_39560_mode_0, x = var_39558_cast_fp16_1)[name = tensor("op_39560_cast_fp16")]; + tensor input_517_cast_fp16 = mul(x = var_39558_cast_fp16_0, y = var_39560_cast_fp16)[name = tensor("input_517_cast_fp16")]; + tensor var_39564 = const()[name = tensor("op_39564"), val = tensor([1, 1])]; + tensor var_39566 = const()[name = tensor("op_39566"), val = tensor([1, 1])]; + tensor var_39568_pad_type_0 = const()[name = tensor("op_39568_pad_type_0"), val = tensor("custom")]; + tensor var_39568_pad_0 = const()[name = tensor("op_39568_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_transformer_blocks_9_ff_net_2_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_9_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3270441088)))]; + tensor up_blocks_0_attentions_0_transformer_blocks_9_ff_net_2_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_transformer_blocks_9_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3283548352)))]; + tensor var_39568_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_9_ff_net_2_bias_to_fp16, dilations = var_39566, groups = var_30355, pad = var_39568_pad_0, pad_type = var_39568_pad_type_0, strides = var_39564, weight = up_blocks_0_attentions_0_transformer_blocks_9_ff_net_2_weight_to_fp16, x = input_517_cast_fp16)[name = tensor("op_39568_cast_fp16")]; + tensor hidden_states_347_cast_fp16 = add(x = var_39568_cast_fp16, y = inputs_263_cast_fp16)[name = tensor("hidden_states_347_cast_fp16")]; + tensor var_39570 = const()[name = tensor("op_39570"), val = tensor([2, 1280, 32, 32])]; + tensor input_519_cast_fp16 = reshape(shape = var_39570, x = hidden_states_347_cast_fp16)[name = tensor("input_519_cast_fp16")]; + tensor var_39574 = const()[name = tensor("op_39574"), val = tensor([1, 1])]; + tensor var_39576 = const()[name = tensor("op_39576"), val = tensor([1, 1])]; + tensor hidden_states_349_pad_type_0 = const()[name = tensor("hidden_states_349_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_349_pad_0 = const()[name = tensor("hidden_states_349_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_0_proj_out_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_proj_out_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3283550976)))]; + tensor up_blocks_0_attentions_0_proj_out_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_proj_out_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3286827840)))]; + tensor hidden_states_349_cast_fp16 = conv(bias = up_blocks_0_attentions_0_proj_out_bias_to_fp16, dilations = var_39576, groups = var_30355, pad = hidden_states_349_pad_0, pad_type = hidden_states_349_pad_type_0, strides = var_39574, weight = up_blocks_0_attentions_0_proj_out_weight_to_fp16, x = input_519_cast_fp16)[name = tensor("hidden_states_349_cast_fp16")]; + tensor hidden_states_351_cast_fp16 = add(x = hidden_states_349_cast_fp16, y = hidden_states_283_cast_fp16)[name = tensor("hidden_states_351_cast_fp16")]; + tensor input_521_interleave_0 = const()[name = tensor("input_521_interleave_0"), val = tensor(false)]; + tensor input_521_cast_fp16 = concat(axis = var_30355, interleave = input_521_interleave_0, values = (hidden_states_351_cast_fp16, input_213_cast_fp16))[name = tensor("input_521_cast_fp16")]; + tensor reshape_96_shape_0 = const()[name = tensor("reshape_96_shape_0"), val = tensor([2, 32, 80, 32, 32])]; + tensor reshape_96_cast_fp16 = reshape(shape = reshape_96_shape_0, x = input_521_cast_fp16)[name = tensor("reshape_96_cast_fp16")]; + tensor reduce_mean_72_axes_0 = const()[name = tensor("reduce_mean_72_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_72_keep_dims_0 = const()[name = tensor("reduce_mean_72_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_72_cast_fp16 = reduce_mean(axes = reduce_mean_72_axes_0, keep_dims = reduce_mean_72_keep_dims_0, x = reshape_96_cast_fp16)[name = tensor("reduce_mean_72_cast_fp16")]; + tensor sub_48_cast_fp16 = sub(x = reshape_96_cast_fp16, y = reduce_mean_72_cast_fp16)[name = tensor("sub_48_cast_fp16")]; + tensor square_24_cast_fp16 = square(x = sub_48_cast_fp16)[name = tensor("square_24_cast_fp16")]; + tensor reduce_mean_74_axes_0 = const()[name = tensor("reduce_mean_74_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_74_keep_dims_0 = const()[name = tensor("reduce_mean_74_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_74_cast_fp16 = reduce_mean(axes = reduce_mean_74_axes_0, keep_dims = reduce_mean_74_keep_dims_0, x = square_24_cast_fp16)[name = tensor("reduce_mean_74_cast_fp16")]; + tensor add_48_y_0_to_fp16 = const()[name = tensor("add_48_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_48_cast_fp16 = add(x = reduce_mean_74_cast_fp16, y = add_48_y_0_to_fp16)[name = tensor("add_48_cast_fp16")]; + tensor sqrt_24_cast_fp16 = sqrt(x = add_48_cast_fp16)[name = tensor("sqrt_24_cast_fp16")]; + tensor real_div_24_cast_fp16 = real_div(x = sub_48_cast_fp16, y = sqrt_24_cast_fp16)[name = tensor("real_div_24_cast_fp16")]; + tensor reshape_97_shape_0 = const()[name = tensor("reshape_97_shape_0"), val = tensor([2, 2560, 32, 32])]; + tensor reshape_97_cast_fp16 = reshape(shape = reshape_97_shape_0, x = real_div_24_cast_fp16)[name = tensor("reshape_97_cast_fp16")]; + tensor add_49_gamma_0_to_fp16 = const()[name = tensor("add_49_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3286830464)))]; + tensor add_49_beta_0_to_fp16 = const()[name = tensor("add_49_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3286835648)))]; + tensor add_49_epsilon_0_to_fp16 = const()[name = tensor("add_49_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_49_cast_fp16 = batch_norm(beta = add_49_beta_0_to_fp16, epsilon = add_49_epsilon_0_to_fp16, gamma = add_49_gamma_0_to_fp16, mean = add_43_mean_0_to_fp16, variance = add_43_variance_0_to_fp16, x = reshape_97_cast_fp16)[name = tensor("add_49_cast_fp16")]; + tensor input_525_cast_fp16 = silu(x = add_49_cast_fp16)[name = tensor("input_525_cast_fp16")]; + tensor var_39594 = const()[name = tensor("op_39594"), val = tensor([1, 1])]; + tensor var_39596 = const()[name = tensor("op_39596"), val = tensor([1, 1])]; + tensor hidden_states_353_pad_type_0 = const()[name = tensor("hidden_states_353_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_353_pad_0 = const()[name = tensor("hidden_states_353_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor up_blocks_0_resnets_1_conv1_weight_to_fp16 = const()[name = tensor("up_blocks_0_resnets_1_conv1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3286840832)))]; + tensor up_blocks_0_resnets_1_conv1_bias_to_fp16 = const()[name = tensor("up_blocks_0_resnets_1_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3345823296)))]; + tensor hidden_states_353_cast_fp16 = conv(bias = up_blocks_0_resnets_1_conv1_bias_to_fp16, dilations = var_39596, groups = var_30355, pad = hidden_states_353_pad_0, pad_type = hidden_states_353_pad_type_0, strides = var_39594, weight = up_blocks_0_resnets_1_conv1_weight_to_fp16, x = input_525_cast_fp16)[name = tensor("hidden_states_353_cast_fp16")]; + tensor var_39602 = const()[name = tensor("op_39602"), val = tensor([1, 1])]; + tensor var_39604 = const()[name = tensor("op_39604"), val = tensor([1, 1])]; + tensor temb_19_pad_type_0 = const()[name = tensor("temb_19_pad_type_0"), val = tensor("custom")]; + tensor temb_19_pad_0 = const()[name = tensor("temb_19_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_resnets_1_time_emb_proj_weight_to_fp16 = const()[name = tensor("up_blocks_0_resnets_1_time_emb_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3345825920)))]; + tensor up_blocks_0_resnets_1_time_emb_proj_bias_to_fp16 = const()[name = tensor("up_blocks_0_resnets_1_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3349102784)))]; + tensor temb_19_cast_fp16 = conv(bias = up_blocks_0_resnets_1_time_emb_proj_bias_to_fp16, dilations = var_39604, groups = var_30355, pad = temb_19_pad_0, pad_type = temb_19_pad_type_0, strides = var_39602, weight = up_blocks_0_resnets_1_time_emb_proj_weight_to_fp16, x = input_21_cast_fp16)[name = tensor("temb_19_cast_fp16")]; + tensor input_529_cast_fp16 = add(x = hidden_states_353_cast_fp16, y = temb_19_cast_fp16)[name = tensor("input_529_cast_fp16")]; + tensor reshape_100_shape_0 = const()[name = tensor("reshape_100_shape_0"), val = tensor([2, 32, 40, 32, 32])]; + tensor reshape_100_cast_fp16 = reshape(shape = reshape_100_shape_0, x = input_529_cast_fp16)[name = tensor("reshape_100_cast_fp16")]; + tensor reduce_mean_75_axes_0 = const()[name = tensor("reduce_mean_75_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_75_keep_dims_0 = const()[name = tensor("reduce_mean_75_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_75_cast_fp16 = reduce_mean(axes = reduce_mean_75_axes_0, keep_dims = reduce_mean_75_keep_dims_0, x = reshape_100_cast_fp16)[name = tensor("reduce_mean_75_cast_fp16")]; + tensor sub_50_cast_fp16 = sub(x = reshape_100_cast_fp16, y = reduce_mean_75_cast_fp16)[name = tensor("sub_50_cast_fp16")]; + tensor square_25_cast_fp16 = square(x = sub_50_cast_fp16)[name = tensor("square_25_cast_fp16")]; + tensor reduce_mean_77_axes_0 = const()[name = tensor("reduce_mean_77_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_77_keep_dims_0 = const()[name = tensor("reduce_mean_77_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_77_cast_fp16 = reduce_mean(axes = reduce_mean_77_axes_0, keep_dims = reduce_mean_77_keep_dims_0, x = square_25_cast_fp16)[name = tensor("reduce_mean_77_cast_fp16")]; + tensor add_50_y_0_to_fp16 = const()[name = tensor("add_50_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_50_cast_fp16 = add(x = reduce_mean_77_cast_fp16, y = add_50_y_0_to_fp16)[name = tensor("add_50_cast_fp16")]; + tensor sqrt_25_cast_fp16 = sqrt(x = add_50_cast_fp16)[name = tensor("sqrt_25_cast_fp16")]; + tensor real_div_25_cast_fp16 = real_div(x = sub_50_cast_fp16, y = sqrt_25_cast_fp16)[name = tensor("real_div_25_cast_fp16")]; + tensor reshape_101_shape_0 = const()[name = tensor("reshape_101_shape_0"), val = tensor([2, 1280, 32, 32])]; + tensor reshape_101_cast_fp16 = reshape(shape = reshape_101_shape_0, x = real_div_25_cast_fp16)[name = tensor("reshape_101_cast_fp16")]; + tensor add_51_gamma_0_to_fp16 = const()[name = tensor("add_51_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3349105408)))]; + tensor add_51_beta_0_to_fp16 = const()[name = tensor("add_51_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3349108032)))]; + tensor add_51_epsilon_0_to_fp16 = const()[name = tensor("add_51_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_51_cast_fp16 = batch_norm(beta = add_51_beta_0_to_fp16, epsilon = add_51_epsilon_0_to_fp16, gamma = add_51_gamma_0_to_fp16, mean = add_23_mean_0_to_fp16, variance = add_23_variance_0_to_fp16, x = reshape_101_cast_fp16)[name = tensor("add_51_cast_fp16")]; + tensor input_533_cast_fp16 = silu(x = add_51_cast_fp16)[name = tensor("input_533_cast_fp16")]; + tensor var_39614 = const()[name = tensor("op_39614"), val = tensor([1, 1])]; + tensor var_39616 = const()[name = tensor("op_39616"), val = tensor([1, 1])]; + tensor hidden_states_355_pad_type_0 = const()[name = tensor("hidden_states_355_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_355_pad_0 = const()[name = tensor("hidden_states_355_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor up_blocks_0_resnets_1_conv2_weight_to_fp16 = const()[name = tensor("up_blocks_0_resnets_1_conv2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3349110656)))]; + tensor up_blocks_0_resnets_1_conv2_bias_to_fp16 = const()[name = tensor("up_blocks_0_resnets_1_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3378601920)))]; + tensor hidden_states_355_cast_fp16 = conv(bias = up_blocks_0_resnets_1_conv2_bias_to_fp16, dilations = var_39616, groups = var_30355, pad = hidden_states_355_pad_0, pad_type = hidden_states_355_pad_type_0, strides = var_39614, weight = up_blocks_0_resnets_1_conv2_weight_to_fp16, x = input_533_cast_fp16)[name = tensor("hidden_states_355_cast_fp16")]; + tensor var_39621 = const()[name = tensor("op_39621"), val = tensor([1, 1])]; + tensor var_39623 = const()[name = tensor("op_39623"), val = tensor([1, 1])]; + tensor x_7_pad_type_0 = const()[name = tensor("x_7_pad_type_0"), val = tensor("custom")]; + tensor x_7_pad_0 = const()[name = tensor("x_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_resnets_1_conv_shortcut_weight_to_fp16 = const()[name = tensor("up_blocks_0_resnets_1_conv_shortcut_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3378604544)))]; + tensor up_blocks_0_resnets_1_conv_shortcut_bias_to_fp16 = const()[name = tensor("up_blocks_0_resnets_1_conv_shortcut_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3385158208)))]; + tensor x_7_cast_fp16 = conv(bias = up_blocks_0_resnets_1_conv_shortcut_bias_to_fp16, dilations = var_39623, groups = var_30355, pad = x_7_pad_0, pad_type = x_7_pad_type_0, strides = var_39621, weight = up_blocks_0_resnets_1_conv_shortcut_weight_to_fp16, x = input_521_cast_fp16)[name = tensor("x_7_cast_fp16")]; + tensor hidden_states_357_cast_fp16 = add(x = x_7_cast_fp16, y = hidden_states_355_cast_fp16)[name = tensor("hidden_states_357_cast_fp16")]; + tensor reshape_104_shape_0 = const()[name = tensor("reshape_104_shape_0"), val = tensor([2, 32, 40, 32, 32])]; + tensor reshape_104_cast_fp16 = reshape(shape = reshape_104_shape_0, x = hidden_states_357_cast_fp16)[name = tensor("reshape_104_cast_fp16")]; + tensor reduce_mean_78_axes_0 = const()[name = tensor("reduce_mean_78_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_78_keep_dims_0 = const()[name = tensor("reduce_mean_78_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_78_cast_fp16 = reduce_mean(axes = reduce_mean_78_axes_0, keep_dims = reduce_mean_78_keep_dims_0, x = reshape_104_cast_fp16)[name = tensor("reduce_mean_78_cast_fp16")]; + tensor sub_52_cast_fp16 = sub(x = reshape_104_cast_fp16, y = reduce_mean_78_cast_fp16)[name = tensor("sub_52_cast_fp16")]; + tensor square_26_cast_fp16 = square(x = sub_52_cast_fp16)[name = tensor("square_26_cast_fp16")]; + tensor reduce_mean_80_axes_0 = const()[name = tensor("reduce_mean_80_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_80_keep_dims_0 = const()[name = tensor("reduce_mean_80_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_80_cast_fp16 = reduce_mean(axes = reduce_mean_80_axes_0, keep_dims = reduce_mean_80_keep_dims_0, x = square_26_cast_fp16)[name = tensor("reduce_mean_80_cast_fp16")]; + tensor add_52_y_0_to_fp16 = const()[name = tensor("add_52_y_0_to_fp16"), val = tensor(0x1.1p-20)]; + tensor add_52_cast_fp16 = add(x = reduce_mean_80_cast_fp16, y = add_52_y_0_to_fp16)[name = tensor("add_52_cast_fp16")]; + tensor sqrt_26_cast_fp16 = sqrt(x = add_52_cast_fp16)[name = tensor("sqrt_26_cast_fp16")]; + tensor real_div_26_cast_fp16 = real_div(x = sub_52_cast_fp16, y = sqrt_26_cast_fp16)[name = tensor("real_div_26_cast_fp16")]; + tensor reshape_105_shape_0 = const()[name = tensor("reshape_105_shape_0"), val = tensor([2, 1280, 32, 32])]; + tensor reshape_105_cast_fp16 = reshape(shape = reshape_105_shape_0, x = real_div_26_cast_fp16)[name = tensor("reshape_105_cast_fp16")]; + tensor add_53_gamma_0_to_fp16 = const()[name = tensor("add_53_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3385160832)))]; + tensor add_53_beta_0_to_fp16 = const()[name = tensor("add_53_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3385163456)))]; + tensor add_53_epsilon_0_to_fp16 = const()[name = tensor("add_53_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_53_cast_fp16 = batch_norm(beta = add_53_beta_0_to_fp16, epsilon = add_53_epsilon_0_to_fp16, gamma = add_53_gamma_0_to_fp16, mean = add_23_mean_0_to_fp16, variance = add_23_variance_0_to_fp16, x = reshape_105_cast_fp16)[name = tensor("add_53_cast_fp16")]; + tensor var_39661 = const()[name = tensor("op_39661"), val = tensor([1, 1])]; + tensor var_39663 = const()[name = tensor("op_39663"), val = tensor([1, 1])]; + tensor hidden_states_359_pad_type_0 = const()[name = tensor("hidden_states_359_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_359_pad_0 = const()[name = tensor("hidden_states_359_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_proj_in_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_proj_in_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3385166080)))]; + tensor up_blocks_0_attentions_1_proj_in_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_proj_in_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3388442944)))]; + tensor hidden_states_359_cast_fp16 = conv(bias = up_blocks_0_attentions_1_proj_in_bias_to_fp16, dilations = var_39663, groups = var_30355, pad = hidden_states_359_pad_0, pad_type = hidden_states_359_pad_type_0, strides = var_39661, weight = up_blocks_0_attentions_1_proj_in_weight_to_fp16, x = add_53_cast_fp16)[name = tensor("hidden_states_359_cast_fp16")]; + tensor var_39668 = const()[name = tensor("op_39668"), val = tensor([2, 1280, 1, 1024])]; + tensor inputs_265_cast_fp16 = reshape(shape = var_39668, x = hidden_states_359_cast_fp16)[name = tensor("inputs_265_cast_fp16")]; + tensor hidden_states_361_axes_0 = const()[name = tensor("hidden_states_361_axes_0"), val = tensor([1])]; + tensor hidden_states_361_gamma_0_to_fp16 = const()[name = tensor("hidden_states_361_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3388445568)))]; + tensor hidden_states_361_beta_0_to_fp16 = const()[name = tensor("hidden_states_361_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3388448192)))]; + tensor var_39684_to_fp16 = const()[name = tensor("op_39684_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_361_cast_fp16 = layer_norm(axes = hidden_states_361_axes_0, beta = hidden_states_361_beta_0_to_fp16, epsilon = var_39684_to_fp16, gamma = hidden_states_361_gamma_0_to_fp16, x = inputs_265_cast_fp16)[name = tensor("hidden_states_361_cast_fp16")]; + tensor var_39699 = const()[name = tensor("op_39699"), val = tensor([1, 1])]; + tensor var_39701 = const()[name = tensor("op_39701"), val = tensor([1, 1])]; + tensor q_177_pad_type_0 = const()[name = tensor("q_177_pad_type_0"), val = tensor("custom")]; + tensor q_177_pad_0 = const()[name = tensor("q_177_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3388450816)))]; + tensor q_177_cast_fp16 = conv(dilations = var_39701, groups = var_30355, pad = q_177_pad_0, pad_type = q_177_pad_type_0, strides = var_39699, weight = up_blocks_0_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16, x = hidden_states_361_cast_fp16)[name = tensor("q_177_cast_fp16")]; + tensor var_39705 = const()[name = tensor("op_39705"), val = tensor([1, 1])]; + tensor var_39707 = const()[name = tensor("op_39707"), val = tensor([1, 1])]; + tensor k_353_pad_type_0 = const()[name = tensor("k_353_pad_type_0"), val = tensor("custom")]; + tensor k_353_pad_0 = const()[name = tensor("k_353_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3391727680)))]; + tensor k_353_cast_fp16 = conv(dilations = var_39707, groups = var_30355, pad = k_353_pad_0, pad_type = k_353_pad_type_0, strides = var_39705, weight = up_blocks_0_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16, x = hidden_states_361_cast_fp16)[name = tensor("k_353_cast_fp16")]; + tensor var_39711 = const()[name = tensor("op_39711"), val = tensor([1, 1])]; + tensor var_39713 = const()[name = tensor("op_39713"), val = tensor([1, 1])]; + tensor v_177_pad_type_0 = const()[name = tensor("v_177_pad_type_0"), val = tensor("custom")]; + tensor v_177_pad_0 = const()[name = tensor("v_177_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3395004544)))]; + tensor v_177_cast_fp16 = conv(dilations = var_39713, groups = var_30355, pad = v_177_pad_0, pad_type = v_177_pad_type_0, strides = var_39711, weight = up_blocks_0_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16, x = hidden_states_361_cast_fp16)[name = tensor("v_177_cast_fp16")]; + tensor var_39717_begin_0 = const()[name = tensor("op_39717_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_39717_end_0 = const()[name = tensor("op_39717_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_39717_end_mask_0 = const()[name = tensor("op_39717_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39717_cast_fp16 = slice_by_index(begin = var_39717_begin_0, end = var_39717_end_0, end_mask = var_39717_end_mask_0, x = q_177_cast_fp16)[name = tensor("op_39717_cast_fp16")]; + tensor var_39721_begin_0 = const()[name = tensor("op_39721_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_39721_end_0 = const()[name = tensor("op_39721_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_39721_end_mask_0 = const()[name = tensor("op_39721_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39721_cast_fp16 = slice_by_index(begin = var_39721_begin_0, end = var_39721_end_0, end_mask = var_39721_end_mask_0, x = q_177_cast_fp16)[name = tensor("op_39721_cast_fp16")]; + tensor var_39725_begin_0 = const()[name = tensor("op_39725_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_39725_end_0 = const()[name = tensor("op_39725_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_39725_end_mask_0 = const()[name = tensor("op_39725_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39725_cast_fp16 = slice_by_index(begin = var_39725_begin_0, end = var_39725_end_0, end_mask = var_39725_end_mask_0, x = q_177_cast_fp16)[name = tensor("op_39725_cast_fp16")]; + tensor var_39729_begin_0 = const()[name = tensor("op_39729_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_39729_end_0 = const()[name = tensor("op_39729_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_39729_end_mask_0 = const()[name = tensor("op_39729_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39729_cast_fp16 = slice_by_index(begin = var_39729_begin_0, end = var_39729_end_0, end_mask = var_39729_end_mask_0, x = q_177_cast_fp16)[name = tensor("op_39729_cast_fp16")]; + tensor var_39733_begin_0 = const()[name = tensor("op_39733_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_39733_end_0 = const()[name = tensor("op_39733_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_39733_end_mask_0 = const()[name = tensor("op_39733_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39733_cast_fp16 = slice_by_index(begin = var_39733_begin_0, end = var_39733_end_0, end_mask = var_39733_end_mask_0, x = q_177_cast_fp16)[name = tensor("op_39733_cast_fp16")]; + tensor var_39737_begin_0 = const()[name = tensor("op_39737_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_39737_end_0 = const()[name = tensor("op_39737_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_39737_end_mask_0 = const()[name = tensor("op_39737_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39737_cast_fp16 = slice_by_index(begin = var_39737_begin_0, end = var_39737_end_0, end_mask = var_39737_end_mask_0, x = q_177_cast_fp16)[name = tensor("op_39737_cast_fp16")]; + tensor var_39741_begin_0 = const()[name = tensor("op_39741_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_39741_end_0 = const()[name = tensor("op_39741_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_39741_end_mask_0 = const()[name = tensor("op_39741_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39741_cast_fp16 = slice_by_index(begin = var_39741_begin_0, end = var_39741_end_0, end_mask = var_39741_end_mask_0, x = q_177_cast_fp16)[name = tensor("op_39741_cast_fp16")]; + tensor var_39745_begin_0 = const()[name = tensor("op_39745_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_39745_end_0 = const()[name = tensor("op_39745_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_39745_end_mask_0 = const()[name = tensor("op_39745_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39745_cast_fp16 = slice_by_index(begin = var_39745_begin_0, end = var_39745_end_0, end_mask = var_39745_end_mask_0, x = q_177_cast_fp16)[name = tensor("op_39745_cast_fp16")]; + tensor var_39749_begin_0 = const()[name = tensor("op_39749_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_39749_end_0 = const()[name = tensor("op_39749_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_39749_end_mask_0 = const()[name = tensor("op_39749_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39749_cast_fp16 = slice_by_index(begin = var_39749_begin_0, end = var_39749_end_0, end_mask = var_39749_end_mask_0, x = q_177_cast_fp16)[name = tensor("op_39749_cast_fp16")]; + tensor var_39753_begin_0 = const()[name = tensor("op_39753_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_39753_end_0 = const()[name = tensor("op_39753_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_39753_end_mask_0 = const()[name = tensor("op_39753_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39753_cast_fp16 = slice_by_index(begin = var_39753_begin_0, end = var_39753_end_0, end_mask = var_39753_end_mask_0, x = q_177_cast_fp16)[name = tensor("op_39753_cast_fp16")]; + tensor var_39757_begin_0 = const()[name = tensor("op_39757_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_39757_end_0 = const()[name = tensor("op_39757_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_39757_end_mask_0 = const()[name = tensor("op_39757_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39757_cast_fp16 = slice_by_index(begin = var_39757_begin_0, end = var_39757_end_0, end_mask = var_39757_end_mask_0, x = q_177_cast_fp16)[name = tensor("op_39757_cast_fp16")]; + tensor var_39761_begin_0 = const()[name = tensor("op_39761_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_39761_end_0 = const()[name = tensor("op_39761_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_39761_end_mask_0 = const()[name = tensor("op_39761_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39761_cast_fp16 = slice_by_index(begin = var_39761_begin_0, end = var_39761_end_0, end_mask = var_39761_end_mask_0, x = q_177_cast_fp16)[name = tensor("op_39761_cast_fp16")]; + tensor var_39765_begin_0 = const()[name = tensor("op_39765_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_39765_end_0 = const()[name = tensor("op_39765_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_39765_end_mask_0 = const()[name = tensor("op_39765_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39765_cast_fp16 = slice_by_index(begin = var_39765_begin_0, end = var_39765_end_0, end_mask = var_39765_end_mask_0, x = q_177_cast_fp16)[name = tensor("op_39765_cast_fp16")]; + tensor var_39769_begin_0 = const()[name = tensor("op_39769_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_39769_end_0 = const()[name = tensor("op_39769_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_39769_end_mask_0 = const()[name = tensor("op_39769_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39769_cast_fp16 = slice_by_index(begin = var_39769_begin_0, end = var_39769_end_0, end_mask = var_39769_end_mask_0, x = q_177_cast_fp16)[name = tensor("op_39769_cast_fp16")]; + tensor var_39773_begin_0 = const()[name = tensor("op_39773_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_39773_end_0 = const()[name = tensor("op_39773_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_39773_end_mask_0 = const()[name = tensor("op_39773_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39773_cast_fp16 = slice_by_index(begin = var_39773_begin_0, end = var_39773_end_0, end_mask = var_39773_end_mask_0, x = q_177_cast_fp16)[name = tensor("op_39773_cast_fp16")]; + tensor var_39777_begin_0 = const()[name = tensor("op_39777_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_39777_end_0 = const()[name = tensor("op_39777_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_39777_end_mask_0 = const()[name = tensor("op_39777_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39777_cast_fp16 = slice_by_index(begin = var_39777_begin_0, end = var_39777_end_0, end_mask = var_39777_end_mask_0, x = q_177_cast_fp16)[name = tensor("op_39777_cast_fp16")]; + tensor var_39781_begin_0 = const()[name = tensor("op_39781_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_39781_end_0 = const()[name = tensor("op_39781_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_39781_end_mask_0 = const()[name = tensor("op_39781_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39781_cast_fp16 = slice_by_index(begin = var_39781_begin_0, end = var_39781_end_0, end_mask = var_39781_end_mask_0, x = q_177_cast_fp16)[name = tensor("op_39781_cast_fp16")]; + tensor var_39785_begin_0 = const()[name = tensor("op_39785_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_39785_end_0 = const()[name = tensor("op_39785_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_39785_end_mask_0 = const()[name = tensor("op_39785_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39785_cast_fp16 = slice_by_index(begin = var_39785_begin_0, end = var_39785_end_0, end_mask = var_39785_end_mask_0, x = q_177_cast_fp16)[name = tensor("op_39785_cast_fp16")]; + tensor var_39789_begin_0 = const()[name = tensor("op_39789_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_39789_end_0 = const()[name = tensor("op_39789_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_39789_end_mask_0 = const()[name = tensor("op_39789_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39789_cast_fp16 = slice_by_index(begin = var_39789_begin_0, end = var_39789_end_0, end_mask = var_39789_end_mask_0, x = q_177_cast_fp16)[name = tensor("op_39789_cast_fp16")]; + tensor var_39793_begin_0 = const()[name = tensor("op_39793_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_39793_end_0 = const()[name = tensor("op_39793_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_39793_end_mask_0 = const()[name = tensor("op_39793_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39793_cast_fp16 = slice_by_index(begin = var_39793_begin_0, end = var_39793_end_0, end_mask = var_39793_end_mask_0, x = q_177_cast_fp16)[name = tensor("op_39793_cast_fp16")]; + tensor k_355_perm_0 = const()[name = tensor("k_355_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_39800_begin_0 = const()[name = tensor("op_39800_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_39800_end_0 = const()[name = tensor("op_39800_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_39800_end_mask_0 = const()[name = tensor("op_39800_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_355_cast_fp16 = transpose(perm = k_355_perm_0, x = k_353_cast_fp16)[name = tensor("transpose_51")]; + tensor var_39800_cast_fp16 = slice_by_index(begin = var_39800_begin_0, end = var_39800_end_0, end_mask = var_39800_end_mask_0, x = k_355_cast_fp16)[name = tensor("op_39800_cast_fp16")]; + tensor var_39804_begin_0 = const()[name = tensor("op_39804_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_39804_end_0 = const()[name = tensor("op_39804_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_39804_end_mask_0 = const()[name = tensor("op_39804_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39804_cast_fp16 = slice_by_index(begin = var_39804_begin_0, end = var_39804_end_0, end_mask = var_39804_end_mask_0, x = k_355_cast_fp16)[name = tensor("op_39804_cast_fp16")]; + tensor var_39808_begin_0 = const()[name = tensor("op_39808_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_39808_end_0 = const()[name = tensor("op_39808_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_39808_end_mask_0 = const()[name = tensor("op_39808_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39808_cast_fp16 = slice_by_index(begin = var_39808_begin_0, end = var_39808_end_0, end_mask = var_39808_end_mask_0, x = k_355_cast_fp16)[name = tensor("op_39808_cast_fp16")]; + tensor var_39812_begin_0 = const()[name = tensor("op_39812_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_39812_end_0 = const()[name = tensor("op_39812_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_39812_end_mask_0 = const()[name = tensor("op_39812_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39812_cast_fp16 = slice_by_index(begin = var_39812_begin_0, end = var_39812_end_0, end_mask = var_39812_end_mask_0, x = k_355_cast_fp16)[name = tensor("op_39812_cast_fp16")]; + tensor var_39816_begin_0 = const()[name = tensor("op_39816_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_39816_end_0 = const()[name = tensor("op_39816_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_39816_end_mask_0 = const()[name = tensor("op_39816_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39816_cast_fp16 = slice_by_index(begin = var_39816_begin_0, end = var_39816_end_0, end_mask = var_39816_end_mask_0, x = k_355_cast_fp16)[name = tensor("op_39816_cast_fp16")]; + tensor var_39820_begin_0 = const()[name = tensor("op_39820_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_39820_end_0 = const()[name = tensor("op_39820_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_39820_end_mask_0 = const()[name = tensor("op_39820_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39820_cast_fp16 = slice_by_index(begin = var_39820_begin_0, end = var_39820_end_0, end_mask = var_39820_end_mask_0, x = k_355_cast_fp16)[name = tensor("op_39820_cast_fp16")]; + tensor var_39824_begin_0 = const()[name = tensor("op_39824_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_39824_end_0 = const()[name = tensor("op_39824_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_39824_end_mask_0 = const()[name = tensor("op_39824_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39824_cast_fp16 = slice_by_index(begin = var_39824_begin_0, end = var_39824_end_0, end_mask = var_39824_end_mask_0, x = k_355_cast_fp16)[name = tensor("op_39824_cast_fp16")]; + tensor var_39828_begin_0 = const()[name = tensor("op_39828_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_39828_end_0 = const()[name = tensor("op_39828_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_39828_end_mask_0 = const()[name = tensor("op_39828_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39828_cast_fp16 = slice_by_index(begin = var_39828_begin_0, end = var_39828_end_0, end_mask = var_39828_end_mask_0, x = k_355_cast_fp16)[name = tensor("op_39828_cast_fp16")]; + tensor var_39832_begin_0 = const()[name = tensor("op_39832_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_39832_end_0 = const()[name = tensor("op_39832_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_39832_end_mask_0 = const()[name = tensor("op_39832_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39832_cast_fp16 = slice_by_index(begin = var_39832_begin_0, end = var_39832_end_0, end_mask = var_39832_end_mask_0, x = k_355_cast_fp16)[name = tensor("op_39832_cast_fp16")]; + tensor var_39836_begin_0 = const()[name = tensor("op_39836_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_39836_end_0 = const()[name = tensor("op_39836_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_39836_end_mask_0 = const()[name = tensor("op_39836_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39836_cast_fp16 = slice_by_index(begin = var_39836_begin_0, end = var_39836_end_0, end_mask = var_39836_end_mask_0, x = k_355_cast_fp16)[name = tensor("op_39836_cast_fp16")]; + tensor var_39840_begin_0 = const()[name = tensor("op_39840_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_39840_end_0 = const()[name = tensor("op_39840_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_39840_end_mask_0 = const()[name = tensor("op_39840_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39840_cast_fp16 = slice_by_index(begin = var_39840_begin_0, end = var_39840_end_0, end_mask = var_39840_end_mask_0, x = k_355_cast_fp16)[name = tensor("op_39840_cast_fp16")]; + tensor var_39844_begin_0 = const()[name = tensor("op_39844_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_39844_end_0 = const()[name = tensor("op_39844_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_39844_end_mask_0 = const()[name = tensor("op_39844_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39844_cast_fp16 = slice_by_index(begin = var_39844_begin_0, end = var_39844_end_0, end_mask = var_39844_end_mask_0, x = k_355_cast_fp16)[name = tensor("op_39844_cast_fp16")]; + tensor var_39848_begin_0 = const()[name = tensor("op_39848_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_39848_end_0 = const()[name = tensor("op_39848_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_39848_end_mask_0 = const()[name = tensor("op_39848_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39848_cast_fp16 = slice_by_index(begin = var_39848_begin_0, end = var_39848_end_0, end_mask = var_39848_end_mask_0, x = k_355_cast_fp16)[name = tensor("op_39848_cast_fp16")]; + tensor var_39852_begin_0 = const()[name = tensor("op_39852_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_39852_end_0 = const()[name = tensor("op_39852_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_39852_end_mask_0 = const()[name = tensor("op_39852_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39852_cast_fp16 = slice_by_index(begin = var_39852_begin_0, end = var_39852_end_0, end_mask = var_39852_end_mask_0, x = k_355_cast_fp16)[name = tensor("op_39852_cast_fp16")]; + tensor var_39856_begin_0 = const()[name = tensor("op_39856_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_39856_end_0 = const()[name = tensor("op_39856_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_39856_end_mask_0 = const()[name = tensor("op_39856_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39856_cast_fp16 = slice_by_index(begin = var_39856_begin_0, end = var_39856_end_0, end_mask = var_39856_end_mask_0, x = k_355_cast_fp16)[name = tensor("op_39856_cast_fp16")]; + tensor var_39860_begin_0 = const()[name = tensor("op_39860_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_39860_end_0 = const()[name = tensor("op_39860_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_39860_end_mask_0 = const()[name = tensor("op_39860_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39860_cast_fp16 = slice_by_index(begin = var_39860_begin_0, end = var_39860_end_0, end_mask = var_39860_end_mask_0, x = k_355_cast_fp16)[name = tensor("op_39860_cast_fp16")]; + tensor var_39864_begin_0 = const()[name = tensor("op_39864_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_39864_end_0 = const()[name = tensor("op_39864_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_39864_end_mask_0 = const()[name = tensor("op_39864_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39864_cast_fp16 = slice_by_index(begin = var_39864_begin_0, end = var_39864_end_0, end_mask = var_39864_end_mask_0, x = k_355_cast_fp16)[name = tensor("op_39864_cast_fp16")]; + tensor var_39868_begin_0 = const()[name = tensor("op_39868_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_39868_end_0 = const()[name = tensor("op_39868_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_39868_end_mask_0 = const()[name = tensor("op_39868_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39868_cast_fp16 = slice_by_index(begin = var_39868_begin_0, end = var_39868_end_0, end_mask = var_39868_end_mask_0, x = k_355_cast_fp16)[name = tensor("op_39868_cast_fp16")]; + tensor var_39872_begin_0 = const()[name = tensor("op_39872_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_39872_end_0 = const()[name = tensor("op_39872_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_39872_end_mask_0 = const()[name = tensor("op_39872_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39872_cast_fp16 = slice_by_index(begin = var_39872_begin_0, end = var_39872_end_0, end_mask = var_39872_end_mask_0, x = k_355_cast_fp16)[name = tensor("op_39872_cast_fp16")]; + tensor var_39876_begin_0 = const()[name = tensor("op_39876_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_39876_end_0 = const()[name = tensor("op_39876_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_39876_end_mask_0 = const()[name = tensor("op_39876_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39876_cast_fp16 = slice_by_index(begin = var_39876_begin_0, end = var_39876_end_0, end_mask = var_39876_end_mask_0, x = k_355_cast_fp16)[name = tensor("op_39876_cast_fp16")]; + tensor var_39878_begin_0 = const()[name = tensor("op_39878_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_39878_end_0 = const()[name = tensor("op_39878_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_39878_end_mask_0 = const()[name = tensor("op_39878_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39878_cast_fp16 = slice_by_index(begin = var_39878_begin_0, end = var_39878_end_0, end_mask = var_39878_end_mask_0, x = v_177_cast_fp16)[name = tensor("op_39878_cast_fp16")]; + tensor var_39882_begin_0 = const()[name = tensor("op_39882_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_39882_end_0 = const()[name = tensor("op_39882_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_39882_end_mask_0 = const()[name = tensor("op_39882_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39882_cast_fp16 = slice_by_index(begin = var_39882_begin_0, end = var_39882_end_0, end_mask = var_39882_end_mask_0, x = v_177_cast_fp16)[name = tensor("op_39882_cast_fp16")]; + tensor var_39886_begin_0 = const()[name = tensor("op_39886_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_39886_end_0 = const()[name = tensor("op_39886_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_39886_end_mask_0 = const()[name = tensor("op_39886_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39886_cast_fp16 = slice_by_index(begin = var_39886_begin_0, end = var_39886_end_0, end_mask = var_39886_end_mask_0, x = v_177_cast_fp16)[name = tensor("op_39886_cast_fp16")]; + tensor var_39890_begin_0 = const()[name = tensor("op_39890_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_39890_end_0 = const()[name = tensor("op_39890_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_39890_end_mask_0 = const()[name = tensor("op_39890_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39890_cast_fp16 = slice_by_index(begin = var_39890_begin_0, end = var_39890_end_0, end_mask = var_39890_end_mask_0, x = v_177_cast_fp16)[name = tensor("op_39890_cast_fp16")]; + tensor var_39894_begin_0 = const()[name = tensor("op_39894_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_39894_end_0 = const()[name = tensor("op_39894_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_39894_end_mask_0 = const()[name = tensor("op_39894_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39894_cast_fp16 = slice_by_index(begin = var_39894_begin_0, end = var_39894_end_0, end_mask = var_39894_end_mask_0, x = v_177_cast_fp16)[name = tensor("op_39894_cast_fp16")]; + tensor var_39898_begin_0 = const()[name = tensor("op_39898_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_39898_end_0 = const()[name = tensor("op_39898_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_39898_end_mask_0 = const()[name = tensor("op_39898_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39898_cast_fp16 = slice_by_index(begin = var_39898_begin_0, end = var_39898_end_0, end_mask = var_39898_end_mask_0, x = v_177_cast_fp16)[name = tensor("op_39898_cast_fp16")]; + tensor var_39902_begin_0 = const()[name = tensor("op_39902_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_39902_end_0 = const()[name = tensor("op_39902_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_39902_end_mask_0 = const()[name = tensor("op_39902_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39902_cast_fp16 = slice_by_index(begin = var_39902_begin_0, end = var_39902_end_0, end_mask = var_39902_end_mask_0, x = v_177_cast_fp16)[name = tensor("op_39902_cast_fp16")]; + tensor var_39906_begin_0 = const()[name = tensor("op_39906_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_39906_end_0 = const()[name = tensor("op_39906_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_39906_end_mask_0 = const()[name = tensor("op_39906_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39906_cast_fp16 = slice_by_index(begin = var_39906_begin_0, end = var_39906_end_0, end_mask = var_39906_end_mask_0, x = v_177_cast_fp16)[name = tensor("op_39906_cast_fp16")]; + tensor var_39910_begin_0 = const()[name = tensor("op_39910_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_39910_end_0 = const()[name = tensor("op_39910_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_39910_end_mask_0 = const()[name = tensor("op_39910_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39910_cast_fp16 = slice_by_index(begin = var_39910_begin_0, end = var_39910_end_0, end_mask = var_39910_end_mask_0, x = v_177_cast_fp16)[name = tensor("op_39910_cast_fp16")]; + tensor var_39914_begin_0 = const()[name = tensor("op_39914_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_39914_end_0 = const()[name = tensor("op_39914_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_39914_end_mask_0 = const()[name = tensor("op_39914_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39914_cast_fp16 = slice_by_index(begin = var_39914_begin_0, end = var_39914_end_0, end_mask = var_39914_end_mask_0, x = v_177_cast_fp16)[name = tensor("op_39914_cast_fp16")]; + tensor var_39918_begin_0 = const()[name = tensor("op_39918_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_39918_end_0 = const()[name = tensor("op_39918_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_39918_end_mask_0 = const()[name = tensor("op_39918_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39918_cast_fp16 = slice_by_index(begin = var_39918_begin_0, end = var_39918_end_0, end_mask = var_39918_end_mask_0, x = v_177_cast_fp16)[name = tensor("op_39918_cast_fp16")]; + tensor var_39922_begin_0 = const()[name = tensor("op_39922_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_39922_end_0 = const()[name = tensor("op_39922_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_39922_end_mask_0 = const()[name = tensor("op_39922_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39922_cast_fp16 = slice_by_index(begin = var_39922_begin_0, end = var_39922_end_0, end_mask = var_39922_end_mask_0, x = v_177_cast_fp16)[name = tensor("op_39922_cast_fp16")]; + tensor var_39926_begin_0 = const()[name = tensor("op_39926_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_39926_end_0 = const()[name = tensor("op_39926_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_39926_end_mask_0 = const()[name = tensor("op_39926_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39926_cast_fp16 = slice_by_index(begin = var_39926_begin_0, end = var_39926_end_0, end_mask = var_39926_end_mask_0, x = v_177_cast_fp16)[name = tensor("op_39926_cast_fp16")]; + tensor var_39930_begin_0 = const()[name = tensor("op_39930_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_39930_end_0 = const()[name = tensor("op_39930_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_39930_end_mask_0 = const()[name = tensor("op_39930_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39930_cast_fp16 = slice_by_index(begin = var_39930_begin_0, end = var_39930_end_0, end_mask = var_39930_end_mask_0, x = v_177_cast_fp16)[name = tensor("op_39930_cast_fp16")]; + tensor var_39934_begin_0 = const()[name = tensor("op_39934_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_39934_end_0 = const()[name = tensor("op_39934_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_39934_end_mask_0 = const()[name = tensor("op_39934_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39934_cast_fp16 = slice_by_index(begin = var_39934_begin_0, end = var_39934_end_0, end_mask = var_39934_end_mask_0, x = v_177_cast_fp16)[name = tensor("op_39934_cast_fp16")]; + tensor var_39938_begin_0 = const()[name = tensor("op_39938_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_39938_end_0 = const()[name = tensor("op_39938_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_39938_end_mask_0 = const()[name = tensor("op_39938_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39938_cast_fp16 = slice_by_index(begin = var_39938_begin_0, end = var_39938_end_0, end_mask = var_39938_end_mask_0, x = v_177_cast_fp16)[name = tensor("op_39938_cast_fp16")]; + tensor var_39942_begin_0 = const()[name = tensor("op_39942_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_39942_end_0 = const()[name = tensor("op_39942_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_39942_end_mask_0 = const()[name = tensor("op_39942_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39942_cast_fp16 = slice_by_index(begin = var_39942_begin_0, end = var_39942_end_0, end_mask = var_39942_end_mask_0, x = v_177_cast_fp16)[name = tensor("op_39942_cast_fp16")]; + tensor var_39946_begin_0 = const()[name = tensor("op_39946_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_39946_end_0 = const()[name = tensor("op_39946_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_39946_end_mask_0 = const()[name = tensor("op_39946_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39946_cast_fp16 = slice_by_index(begin = var_39946_begin_0, end = var_39946_end_0, end_mask = var_39946_end_mask_0, x = v_177_cast_fp16)[name = tensor("op_39946_cast_fp16")]; + tensor var_39950_begin_0 = const()[name = tensor("op_39950_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_39950_end_0 = const()[name = tensor("op_39950_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_39950_end_mask_0 = const()[name = tensor("op_39950_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39950_cast_fp16 = slice_by_index(begin = var_39950_begin_0, end = var_39950_end_0, end_mask = var_39950_end_mask_0, x = v_177_cast_fp16)[name = tensor("op_39950_cast_fp16")]; + tensor var_39954_begin_0 = const()[name = tensor("op_39954_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_39954_end_0 = const()[name = tensor("op_39954_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_39954_end_mask_0 = const()[name = tensor("op_39954_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39954_cast_fp16 = slice_by_index(begin = var_39954_begin_0, end = var_39954_end_0, end_mask = var_39954_end_mask_0, x = v_177_cast_fp16)[name = tensor("op_39954_cast_fp16")]; + tensor var_39958_equation_0 = const()[name = tensor("op_39958_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_39958_cast_fp16 = einsum(equation = var_39958_equation_0, values = (var_39800_cast_fp16, var_39717_cast_fp16))[name = tensor("op_39958_cast_fp16")]; + tensor var_39959_to_fp16 = const()[name = tensor("op_39959_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3361_cast_fp16 = mul(x = var_39958_cast_fp16, y = var_39959_to_fp16)[name = tensor("aw_3361_cast_fp16")]; + tensor var_39962_equation_0 = const()[name = tensor("op_39962_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_39962_cast_fp16 = einsum(equation = var_39962_equation_0, values = (var_39804_cast_fp16, var_39721_cast_fp16))[name = tensor("op_39962_cast_fp16")]; + tensor var_39963_to_fp16 = const()[name = tensor("op_39963_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3363_cast_fp16 = mul(x = var_39962_cast_fp16, y = var_39963_to_fp16)[name = tensor("aw_3363_cast_fp16")]; + tensor var_39966_equation_0 = const()[name = tensor("op_39966_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_39966_cast_fp16 = einsum(equation = var_39966_equation_0, values = (var_39808_cast_fp16, var_39725_cast_fp16))[name = tensor("op_39966_cast_fp16")]; + tensor var_39967_to_fp16 = const()[name = tensor("op_39967_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3365_cast_fp16 = mul(x = var_39966_cast_fp16, y = var_39967_to_fp16)[name = tensor("aw_3365_cast_fp16")]; + tensor var_39970_equation_0 = const()[name = tensor("op_39970_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_39970_cast_fp16 = einsum(equation = var_39970_equation_0, values = (var_39812_cast_fp16, var_39729_cast_fp16))[name = tensor("op_39970_cast_fp16")]; + tensor var_39971_to_fp16 = const()[name = tensor("op_39971_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3367_cast_fp16 = mul(x = var_39970_cast_fp16, y = var_39971_to_fp16)[name = tensor("aw_3367_cast_fp16")]; + tensor var_39974_equation_0 = const()[name = tensor("op_39974_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_39974_cast_fp16 = einsum(equation = var_39974_equation_0, values = (var_39816_cast_fp16, var_39733_cast_fp16))[name = tensor("op_39974_cast_fp16")]; + tensor var_39975_to_fp16 = const()[name = tensor("op_39975_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3369_cast_fp16 = mul(x = var_39974_cast_fp16, y = var_39975_to_fp16)[name = tensor("aw_3369_cast_fp16")]; + tensor var_39978_equation_0 = const()[name = tensor("op_39978_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_39978_cast_fp16 = einsum(equation = var_39978_equation_0, values = (var_39820_cast_fp16, var_39737_cast_fp16))[name = tensor("op_39978_cast_fp16")]; + tensor var_39979_to_fp16 = const()[name = tensor("op_39979_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3371_cast_fp16 = mul(x = var_39978_cast_fp16, y = var_39979_to_fp16)[name = tensor("aw_3371_cast_fp16")]; + tensor var_39982_equation_0 = const()[name = tensor("op_39982_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_39982_cast_fp16 = einsum(equation = var_39982_equation_0, values = (var_39824_cast_fp16, var_39741_cast_fp16))[name = tensor("op_39982_cast_fp16")]; + tensor var_39983_to_fp16 = const()[name = tensor("op_39983_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3373_cast_fp16 = mul(x = var_39982_cast_fp16, y = var_39983_to_fp16)[name = tensor("aw_3373_cast_fp16")]; + tensor var_39986_equation_0 = const()[name = tensor("op_39986_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_39986_cast_fp16 = einsum(equation = var_39986_equation_0, values = (var_39828_cast_fp16, var_39745_cast_fp16))[name = tensor("op_39986_cast_fp16")]; + tensor var_39987_to_fp16 = const()[name = tensor("op_39987_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3375_cast_fp16 = mul(x = var_39986_cast_fp16, y = var_39987_to_fp16)[name = tensor("aw_3375_cast_fp16")]; + tensor var_39990_equation_0 = const()[name = tensor("op_39990_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_39990_cast_fp16 = einsum(equation = var_39990_equation_0, values = (var_39832_cast_fp16, var_39749_cast_fp16))[name = tensor("op_39990_cast_fp16")]; + tensor var_39991_to_fp16 = const()[name = tensor("op_39991_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3377_cast_fp16 = mul(x = var_39990_cast_fp16, y = var_39991_to_fp16)[name = tensor("aw_3377_cast_fp16")]; + tensor var_39994_equation_0 = const()[name = tensor("op_39994_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_39994_cast_fp16 = einsum(equation = var_39994_equation_0, values = (var_39836_cast_fp16, var_39753_cast_fp16))[name = tensor("op_39994_cast_fp16")]; + tensor var_39995_to_fp16 = const()[name = tensor("op_39995_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3379_cast_fp16 = mul(x = var_39994_cast_fp16, y = var_39995_to_fp16)[name = tensor("aw_3379_cast_fp16")]; + tensor var_39998_equation_0 = const()[name = tensor("op_39998_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_39998_cast_fp16 = einsum(equation = var_39998_equation_0, values = (var_39840_cast_fp16, var_39757_cast_fp16))[name = tensor("op_39998_cast_fp16")]; + tensor var_39999_to_fp16 = const()[name = tensor("op_39999_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3381_cast_fp16 = mul(x = var_39998_cast_fp16, y = var_39999_to_fp16)[name = tensor("aw_3381_cast_fp16")]; + tensor var_40002_equation_0 = const()[name = tensor("op_40002_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40002_cast_fp16 = einsum(equation = var_40002_equation_0, values = (var_39844_cast_fp16, var_39761_cast_fp16))[name = tensor("op_40002_cast_fp16")]; + tensor var_40003_to_fp16 = const()[name = tensor("op_40003_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3383_cast_fp16 = mul(x = var_40002_cast_fp16, y = var_40003_to_fp16)[name = tensor("aw_3383_cast_fp16")]; + tensor var_40006_equation_0 = const()[name = tensor("op_40006_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40006_cast_fp16 = einsum(equation = var_40006_equation_0, values = (var_39848_cast_fp16, var_39765_cast_fp16))[name = tensor("op_40006_cast_fp16")]; + tensor var_40007_to_fp16 = const()[name = tensor("op_40007_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3385_cast_fp16 = mul(x = var_40006_cast_fp16, y = var_40007_to_fp16)[name = tensor("aw_3385_cast_fp16")]; + tensor var_40010_equation_0 = const()[name = tensor("op_40010_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40010_cast_fp16 = einsum(equation = var_40010_equation_0, values = (var_39852_cast_fp16, var_39769_cast_fp16))[name = tensor("op_40010_cast_fp16")]; + tensor var_40011_to_fp16 = const()[name = tensor("op_40011_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3387_cast_fp16 = mul(x = var_40010_cast_fp16, y = var_40011_to_fp16)[name = tensor("aw_3387_cast_fp16")]; + tensor var_40014_equation_0 = const()[name = tensor("op_40014_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40014_cast_fp16 = einsum(equation = var_40014_equation_0, values = (var_39856_cast_fp16, var_39773_cast_fp16))[name = tensor("op_40014_cast_fp16")]; + tensor var_40015_to_fp16 = const()[name = tensor("op_40015_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3389_cast_fp16 = mul(x = var_40014_cast_fp16, y = var_40015_to_fp16)[name = tensor("aw_3389_cast_fp16")]; + tensor var_40018_equation_0 = const()[name = tensor("op_40018_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40018_cast_fp16 = einsum(equation = var_40018_equation_0, values = (var_39860_cast_fp16, var_39777_cast_fp16))[name = tensor("op_40018_cast_fp16")]; + tensor var_40019_to_fp16 = const()[name = tensor("op_40019_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3391_cast_fp16 = mul(x = var_40018_cast_fp16, y = var_40019_to_fp16)[name = tensor("aw_3391_cast_fp16")]; + tensor var_40022_equation_0 = const()[name = tensor("op_40022_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40022_cast_fp16 = einsum(equation = var_40022_equation_0, values = (var_39864_cast_fp16, var_39781_cast_fp16))[name = tensor("op_40022_cast_fp16")]; + tensor var_40023_to_fp16 = const()[name = tensor("op_40023_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3393_cast_fp16 = mul(x = var_40022_cast_fp16, y = var_40023_to_fp16)[name = tensor("aw_3393_cast_fp16")]; + tensor var_40026_equation_0 = const()[name = tensor("op_40026_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40026_cast_fp16 = einsum(equation = var_40026_equation_0, values = (var_39868_cast_fp16, var_39785_cast_fp16))[name = tensor("op_40026_cast_fp16")]; + tensor var_40027_to_fp16 = const()[name = tensor("op_40027_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3395_cast_fp16 = mul(x = var_40026_cast_fp16, y = var_40027_to_fp16)[name = tensor("aw_3395_cast_fp16")]; + tensor var_40030_equation_0 = const()[name = tensor("op_40030_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40030_cast_fp16 = einsum(equation = var_40030_equation_0, values = (var_39872_cast_fp16, var_39789_cast_fp16))[name = tensor("op_40030_cast_fp16")]; + tensor var_40031_to_fp16 = const()[name = tensor("op_40031_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3397_cast_fp16 = mul(x = var_40030_cast_fp16, y = var_40031_to_fp16)[name = tensor("aw_3397_cast_fp16")]; + tensor var_40034_equation_0 = const()[name = tensor("op_40034_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40034_cast_fp16 = einsum(equation = var_40034_equation_0, values = (var_39876_cast_fp16, var_39793_cast_fp16))[name = tensor("op_40034_cast_fp16")]; + tensor var_40035_to_fp16 = const()[name = tensor("op_40035_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3399_cast_fp16 = mul(x = var_40034_cast_fp16, y = var_40035_to_fp16)[name = tensor("aw_3399_cast_fp16")]; + tensor var_40037_cast_fp16 = softmax(axis = var_30355, x = aw_3361_cast_fp16)[name = tensor("op_40037_cast_fp16")]; + tensor var_40038_cast_fp16 = softmax(axis = var_30355, x = aw_3363_cast_fp16)[name = tensor("op_40038_cast_fp16")]; + tensor var_40039_cast_fp16 = softmax(axis = var_30355, x = aw_3365_cast_fp16)[name = tensor("op_40039_cast_fp16")]; + tensor var_40040_cast_fp16 = softmax(axis = var_30355, x = aw_3367_cast_fp16)[name = tensor("op_40040_cast_fp16")]; + tensor var_40041_cast_fp16 = softmax(axis = var_30355, x = aw_3369_cast_fp16)[name = tensor("op_40041_cast_fp16")]; + tensor var_40042_cast_fp16 = softmax(axis = var_30355, x = aw_3371_cast_fp16)[name = tensor("op_40042_cast_fp16")]; + tensor var_40043_cast_fp16 = softmax(axis = var_30355, x = aw_3373_cast_fp16)[name = tensor("op_40043_cast_fp16")]; + tensor var_40044_cast_fp16 = softmax(axis = var_30355, x = aw_3375_cast_fp16)[name = tensor("op_40044_cast_fp16")]; + tensor var_40045_cast_fp16 = softmax(axis = var_30355, x = aw_3377_cast_fp16)[name = tensor("op_40045_cast_fp16")]; + tensor var_40046_cast_fp16 = softmax(axis = var_30355, x = aw_3379_cast_fp16)[name = tensor("op_40046_cast_fp16")]; + tensor var_40047_cast_fp16 = softmax(axis = var_30355, x = aw_3381_cast_fp16)[name = tensor("op_40047_cast_fp16")]; + tensor var_40048_cast_fp16 = softmax(axis = var_30355, x = aw_3383_cast_fp16)[name = tensor("op_40048_cast_fp16")]; + tensor var_40049_cast_fp16 = softmax(axis = var_30355, x = aw_3385_cast_fp16)[name = tensor("op_40049_cast_fp16")]; + tensor var_40050_cast_fp16 = softmax(axis = var_30355, x = aw_3387_cast_fp16)[name = tensor("op_40050_cast_fp16")]; + tensor var_40051_cast_fp16 = softmax(axis = var_30355, x = aw_3389_cast_fp16)[name = tensor("op_40051_cast_fp16")]; + tensor var_40052_cast_fp16 = softmax(axis = var_30355, x = aw_3391_cast_fp16)[name = tensor("op_40052_cast_fp16")]; + tensor var_40053_cast_fp16 = softmax(axis = var_30355, x = aw_3393_cast_fp16)[name = tensor("op_40053_cast_fp16")]; + tensor var_40054_cast_fp16 = softmax(axis = var_30355, x = aw_3395_cast_fp16)[name = tensor("op_40054_cast_fp16")]; + tensor var_40055_cast_fp16 = softmax(axis = var_30355, x = aw_3397_cast_fp16)[name = tensor("op_40055_cast_fp16")]; + tensor var_40056_cast_fp16 = softmax(axis = var_30355, x = aw_3399_cast_fp16)[name = tensor("op_40056_cast_fp16")]; + tensor var_40058_equation_0 = const()[name = tensor("op_40058_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40058_cast_fp16 = einsum(equation = var_40058_equation_0, values = (var_39878_cast_fp16, var_40037_cast_fp16))[name = tensor("op_40058_cast_fp16")]; + tensor var_40060_equation_0 = const()[name = tensor("op_40060_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40060_cast_fp16 = einsum(equation = var_40060_equation_0, values = (var_39882_cast_fp16, var_40038_cast_fp16))[name = tensor("op_40060_cast_fp16")]; + tensor var_40062_equation_0 = const()[name = tensor("op_40062_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40062_cast_fp16 = einsum(equation = var_40062_equation_0, values = (var_39886_cast_fp16, var_40039_cast_fp16))[name = tensor("op_40062_cast_fp16")]; + tensor var_40064_equation_0 = const()[name = tensor("op_40064_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40064_cast_fp16 = einsum(equation = var_40064_equation_0, values = (var_39890_cast_fp16, var_40040_cast_fp16))[name = tensor("op_40064_cast_fp16")]; + tensor var_40066_equation_0 = const()[name = tensor("op_40066_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40066_cast_fp16 = einsum(equation = var_40066_equation_0, values = (var_39894_cast_fp16, var_40041_cast_fp16))[name = tensor("op_40066_cast_fp16")]; + tensor var_40068_equation_0 = const()[name = tensor("op_40068_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40068_cast_fp16 = einsum(equation = var_40068_equation_0, values = (var_39898_cast_fp16, var_40042_cast_fp16))[name = tensor("op_40068_cast_fp16")]; + tensor var_40070_equation_0 = const()[name = tensor("op_40070_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40070_cast_fp16 = einsum(equation = var_40070_equation_0, values = (var_39902_cast_fp16, var_40043_cast_fp16))[name = tensor("op_40070_cast_fp16")]; + tensor var_40072_equation_0 = const()[name = tensor("op_40072_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40072_cast_fp16 = einsum(equation = var_40072_equation_0, values = (var_39906_cast_fp16, var_40044_cast_fp16))[name = tensor("op_40072_cast_fp16")]; + tensor var_40074_equation_0 = const()[name = tensor("op_40074_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40074_cast_fp16 = einsum(equation = var_40074_equation_0, values = (var_39910_cast_fp16, var_40045_cast_fp16))[name = tensor("op_40074_cast_fp16")]; + tensor var_40076_equation_0 = const()[name = tensor("op_40076_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40076_cast_fp16 = einsum(equation = var_40076_equation_0, values = (var_39914_cast_fp16, var_40046_cast_fp16))[name = tensor("op_40076_cast_fp16")]; + tensor var_40078_equation_0 = const()[name = tensor("op_40078_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40078_cast_fp16 = einsum(equation = var_40078_equation_0, values = (var_39918_cast_fp16, var_40047_cast_fp16))[name = tensor("op_40078_cast_fp16")]; + tensor var_40080_equation_0 = const()[name = tensor("op_40080_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40080_cast_fp16 = einsum(equation = var_40080_equation_0, values = (var_39922_cast_fp16, var_40048_cast_fp16))[name = tensor("op_40080_cast_fp16")]; + tensor var_40082_equation_0 = const()[name = tensor("op_40082_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40082_cast_fp16 = einsum(equation = var_40082_equation_0, values = (var_39926_cast_fp16, var_40049_cast_fp16))[name = tensor("op_40082_cast_fp16")]; + tensor var_40084_equation_0 = const()[name = tensor("op_40084_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40084_cast_fp16 = einsum(equation = var_40084_equation_0, values = (var_39930_cast_fp16, var_40050_cast_fp16))[name = tensor("op_40084_cast_fp16")]; + tensor var_40086_equation_0 = const()[name = tensor("op_40086_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40086_cast_fp16 = einsum(equation = var_40086_equation_0, values = (var_39934_cast_fp16, var_40051_cast_fp16))[name = tensor("op_40086_cast_fp16")]; + tensor var_40088_equation_0 = const()[name = tensor("op_40088_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40088_cast_fp16 = einsum(equation = var_40088_equation_0, values = (var_39938_cast_fp16, var_40052_cast_fp16))[name = tensor("op_40088_cast_fp16")]; + tensor var_40090_equation_0 = const()[name = tensor("op_40090_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40090_cast_fp16 = einsum(equation = var_40090_equation_0, values = (var_39942_cast_fp16, var_40053_cast_fp16))[name = tensor("op_40090_cast_fp16")]; + tensor var_40092_equation_0 = const()[name = tensor("op_40092_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40092_cast_fp16 = einsum(equation = var_40092_equation_0, values = (var_39946_cast_fp16, var_40054_cast_fp16))[name = tensor("op_40092_cast_fp16")]; + tensor var_40094_equation_0 = const()[name = tensor("op_40094_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40094_cast_fp16 = einsum(equation = var_40094_equation_0, values = (var_39950_cast_fp16, var_40055_cast_fp16))[name = tensor("op_40094_cast_fp16")]; + tensor var_40096_equation_0 = const()[name = tensor("op_40096_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40096_cast_fp16 = einsum(equation = var_40096_equation_0, values = (var_39954_cast_fp16, var_40056_cast_fp16))[name = tensor("op_40096_cast_fp16")]; + tensor input_537_interleave_0 = const()[name = tensor("input_537_interleave_0"), val = tensor(false)]; + tensor input_537_cast_fp16 = concat(axis = var_30355, interleave = input_537_interleave_0, values = (var_40058_cast_fp16, var_40060_cast_fp16, var_40062_cast_fp16, var_40064_cast_fp16, var_40066_cast_fp16, var_40068_cast_fp16, var_40070_cast_fp16, var_40072_cast_fp16, var_40074_cast_fp16, var_40076_cast_fp16, var_40078_cast_fp16, var_40080_cast_fp16, var_40082_cast_fp16, var_40084_cast_fp16, var_40086_cast_fp16, var_40088_cast_fp16, var_40090_cast_fp16, var_40092_cast_fp16, var_40094_cast_fp16, var_40096_cast_fp16))[name = tensor("input_537_cast_fp16")]; + tensor var_40102 = const()[name = tensor("op_40102"), val = tensor([1, 1])]; + tensor var_40104 = const()[name = tensor("op_40104"), val = tensor([1, 1])]; + tensor var_40106_pad_type_0 = const()[name = tensor("op_40106_pad_type_0"), val = tensor("custom")]; + tensor var_40106_pad_0 = const()[name = tensor("op_40106_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3398281408)))]; + tensor up_blocks_0_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3401558272)))]; + tensor var_40106_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_40104, groups = var_30355, pad = var_40106_pad_0, pad_type = var_40106_pad_type_0, strides = var_40102, weight = up_blocks_0_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16, x = input_537_cast_fp16)[name = tensor("op_40106_cast_fp16")]; + tensor inputs_267_cast_fp16 = add(x = var_40106_cast_fp16, y = inputs_265_cast_fp16)[name = tensor("inputs_267_cast_fp16")]; + tensor hidden_states_363_axes_0 = const()[name = tensor("hidden_states_363_axes_0"), val = tensor([1])]; + tensor hidden_states_363_gamma_0_to_fp16 = const()[name = tensor("hidden_states_363_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3401560896)))]; + tensor hidden_states_363_beta_0_to_fp16 = const()[name = tensor("hidden_states_363_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3401563520)))]; + tensor var_40116_to_fp16 = const()[name = tensor("op_40116_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_363_cast_fp16 = layer_norm(axes = hidden_states_363_axes_0, beta = hidden_states_363_beta_0_to_fp16, epsilon = var_40116_to_fp16, gamma = hidden_states_363_gamma_0_to_fp16, x = inputs_267_cast_fp16)[name = tensor("hidden_states_363_cast_fp16")]; + tensor var_40131 = const()[name = tensor("op_40131"), val = tensor([1, 1])]; + tensor var_40133 = const()[name = tensor("op_40133"), val = tensor([1, 1])]; + tensor q_179_pad_type_0 = const()[name = tensor("q_179_pad_type_0"), val = tensor("custom")]; + tensor q_179_pad_0 = const()[name = tensor("q_179_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3401566144)))]; + tensor q_179_cast_fp16 = conv(dilations = var_40133, groups = var_30355, pad = q_179_pad_0, pad_type = q_179_pad_type_0, strides = var_40131, weight = up_blocks_0_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16, x = hidden_states_363_cast_fp16)[name = tensor("q_179_cast_fp16")]; + tensor var_40137 = const()[name = tensor("op_40137"), val = tensor([1, 1])]; + tensor var_40139 = const()[name = tensor("op_40139"), val = tensor([1, 1])]; + tensor k_357_pad_type_0 = const()[name = tensor("k_357_pad_type_0"), val = tensor("custom")]; + tensor k_357_pad_0 = const()[name = tensor("k_357_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3404843008)))]; + tensor k_357_cast_fp16 = conv(dilations = var_40139, groups = var_30355, pad = k_357_pad_0, pad_type = k_357_pad_type_0, strides = var_40137, weight = up_blocks_0_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_357_cast_fp16")]; + tensor var_40143 = const()[name = tensor("op_40143"), val = tensor([1, 1])]; + tensor var_40145 = const()[name = tensor("op_40145"), val = tensor([1, 1])]; + tensor v_179_pad_type_0 = const()[name = tensor("v_179_pad_type_0"), val = tensor("custom")]; + tensor v_179_pad_0 = const()[name = tensor("v_179_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3410085952)))]; + tensor v_179_cast_fp16 = conv(dilations = var_40145, groups = var_30355, pad = v_179_pad_0, pad_type = v_179_pad_type_0, strides = var_40143, weight = up_blocks_0_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_179_cast_fp16")]; + tensor var_40149_begin_0 = const()[name = tensor("op_40149_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_40149_end_0 = const()[name = tensor("op_40149_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_40149_end_mask_0 = const()[name = tensor("op_40149_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40149_cast_fp16 = slice_by_index(begin = var_40149_begin_0, end = var_40149_end_0, end_mask = var_40149_end_mask_0, x = q_179_cast_fp16)[name = tensor("op_40149_cast_fp16")]; + tensor var_40153_begin_0 = const()[name = tensor("op_40153_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_40153_end_0 = const()[name = tensor("op_40153_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_40153_end_mask_0 = const()[name = tensor("op_40153_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40153_cast_fp16 = slice_by_index(begin = var_40153_begin_0, end = var_40153_end_0, end_mask = var_40153_end_mask_0, x = q_179_cast_fp16)[name = tensor("op_40153_cast_fp16")]; + tensor var_40157_begin_0 = const()[name = tensor("op_40157_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_40157_end_0 = const()[name = tensor("op_40157_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_40157_end_mask_0 = const()[name = tensor("op_40157_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40157_cast_fp16 = slice_by_index(begin = var_40157_begin_0, end = var_40157_end_0, end_mask = var_40157_end_mask_0, x = q_179_cast_fp16)[name = tensor("op_40157_cast_fp16")]; + tensor var_40161_begin_0 = const()[name = tensor("op_40161_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_40161_end_0 = const()[name = tensor("op_40161_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_40161_end_mask_0 = const()[name = tensor("op_40161_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40161_cast_fp16 = slice_by_index(begin = var_40161_begin_0, end = var_40161_end_0, end_mask = var_40161_end_mask_0, x = q_179_cast_fp16)[name = tensor("op_40161_cast_fp16")]; + tensor var_40165_begin_0 = const()[name = tensor("op_40165_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_40165_end_0 = const()[name = tensor("op_40165_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_40165_end_mask_0 = const()[name = tensor("op_40165_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40165_cast_fp16 = slice_by_index(begin = var_40165_begin_0, end = var_40165_end_0, end_mask = var_40165_end_mask_0, x = q_179_cast_fp16)[name = tensor("op_40165_cast_fp16")]; + tensor var_40169_begin_0 = const()[name = tensor("op_40169_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_40169_end_0 = const()[name = tensor("op_40169_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_40169_end_mask_0 = const()[name = tensor("op_40169_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40169_cast_fp16 = slice_by_index(begin = var_40169_begin_0, end = var_40169_end_0, end_mask = var_40169_end_mask_0, x = q_179_cast_fp16)[name = tensor("op_40169_cast_fp16")]; + tensor var_40173_begin_0 = const()[name = tensor("op_40173_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_40173_end_0 = const()[name = tensor("op_40173_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_40173_end_mask_0 = const()[name = tensor("op_40173_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40173_cast_fp16 = slice_by_index(begin = var_40173_begin_0, end = var_40173_end_0, end_mask = var_40173_end_mask_0, x = q_179_cast_fp16)[name = tensor("op_40173_cast_fp16")]; + tensor var_40177_begin_0 = const()[name = tensor("op_40177_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_40177_end_0 = const()[name = tensor("op_40177_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_40177_end_mask_0 = const()[name = tensor("op_40177_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40177_cast_fp16 = slice_by_index(begin = var_40177_begin_0, end = var_40177_end_0, end_mask = var_40177_end_mask_0, x = q_179_cast_fp16)[name = tensor("op_40177_cast_fp16")]; + tensor var_40181_begin_0 = const()[name = tensor("op_40181_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_40181_end_0 = const()[name = tensor("op_40181_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_40181_end_mask_0 = const()[name = tensor("op_40181_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40181_cast_fp16 = slice_by_index(begin = var_40181_begin_0, end = var_40181_end_0, end_mask = var_40181_end_mask_0, x = q_179_cast_fp16)[name = tensor("op_40181_cast_fp16")]; + tensor var_40185_begin_0 = const()[name = tensor("op_40185_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_40185_end_0 = const()[name = tensor("op_40185_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_40185_end_mask_0 = const()[name = tensor("op_40185_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40185_cast_fp16 = slice_by_index(begin = var_40185_begin_0, end = var_40185_end_0, end_mask = var_40185_end_mask_0, x = q_179_cast_fp16)[name = tensor("op_40185_cast_fp16")]; + tensor var_40189_begin_0 = const()[name = tensor("op_40189_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_40189_end_0 = const()[name = tensor("op_40189_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_40189_end_mask_0 = const()[name = tensor("op_40189_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40189_cast_fp16 = slice_by_index(begin = var_40189_begin_0, end = var_40189_end_0, end_mask = var_40189_end_mask_0, x = q_179_cast_fp16)[name = tensor("op_40189_cast_fp16")]; + tensor var_40193_begin_0 = const()[name = tensor("op_40193_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_40193_end_0 = const()[name = tensor("op_40193_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_40193_end_mask_0 = const()[name = tensor("op_40193_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40193_cast_fp16 = slice_by_index(begin = var_40193_begin_0, end = var_40193_end_0, end_mask = var_40193_end_mask_0, x = q_179_cast_fp16)[name = tensor("op_40193_cast_fp16")]; + tensor var_40197_begin_0 = const()[name = tensor("op_40197_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_40197_end_0 = const()[name = tensor("op_40197_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_40197_end_mask_0 = const()[name = tensor("op_40197_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40197_cast_fp16 = slice_by_index(begin = var_40197_begin_0, end = var_40197_end_0, end_mask = var_40197_end_mask_0, x = q_179_cast_fp16)[name = tensor("op_40197_cast_fp16")]; + tensor var_40201_begin_0 = const()[name = tensor("op_40201_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_40201_end_0 = const()[name = tensor("op_40201_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_40201_end_mask_0 = const()[name = tensor("op_40201_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40201_cast_fp16 = slice_by_index(begin = var_40201_begin_0, end = var_40201_end_0, end_mask = var_40201_end_mask_0, x = q_179_cast_fp16)[name = tensor("op_40201_cast_fp16")]; + tensor var_40205_begin_0 = const()[name = tensor("op_40205_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_40205_end_0 = const()[name = tensor("op_40205_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_40205_end_mask_0 = const()[name = tensor("op_40205_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40205_cast_fp16 = slice_by_index(begin = var_40205_begin_0, end = var_40205_end_0, end_mask = var_40205_end_mask_0, x = q_179_cast_fp16)[name = tensor("op_40205_cast_fp16")]; + tensor var_40209_begin_0 = const()[name = tensor("op_40209_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_40209_end_0 = const()[name = tensor("op_40209_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_40209_end_mask_0 = const()[name = tensor("op_40209_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40209_cast_fp16 = slice_by_index(begin = var_40209_begin_0, end = var_40209_end_0, end_mask = var_40209_end_mask_0, x = q_179_cast_fp16)[name = tensor("op_40209_cast_fp16")]; + tensor var_40213_begin_0 = const()[name = tensor("op_40213_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_40213_end_0 = const()[name = tensor("op_40213_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_40213_end_mask_0 = const()[name = tensor("op_40213_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40213_cast_fp16 = slice_by_index(begin = var_40213_begin_0, end = var_40213_end_0, end_mask = var_40213_end_mask_0, x = q_179_cast_fp16)[name = tensor("op_40213_cast_fp16")]; + tensor var_40217_begin_0 = const()[name = tensor("op_40217_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_40217_end_0 = const()[name = tensor("op_40217_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_40217_end_mask_0 = const()[name = tensor("op_40217_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40217_cast_fp16 = slice_by_index(begin = var_40217_begin_0, end = var_40217_end_0, end_mask = var_40217_end_mask_0, x = q_179_cast_fp16)[name = tensor("op_40217_cast_fp16")]; + tensor var_40221_begin_0 = const()[name = tensor("op_40221_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_40221_end_0 = const()[name = tensor("op_40221_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_40221_end_mask_0 = const()[name = tensor("op_40221_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40221_cast_fp16 = slice_by_index(begin = var_40221_begin_0, end = var_40221_end_0, end_mask = var_40221_end_mask_0, x = q_179_cast_fp16)[name = tensor("op_40221_cast_fp16")]; + tensor var_40225_begin_0 = const()[name = tensor("op_40225_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_40225_end_0 = const()[name = tensor("op_40225_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_40225_end_mask_0 = const()[name = tensor("op_40225_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40225_cast_fp16 = slice_by_index(begin = var_40225_begin_0, end = var_40225_end_0, end_mask = var_40225_end_mask_0, x = q_179_cast_fp16)[name = tensor("op_40225_cast_fp16")]; + tensor k_359_perm_0 = const()[name = tensor("k_359_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_40232_begin_0 = const()[name = tensor("op_40232_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_40232_end_0 = const()[name = tensor("op_40232_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_40232_end_mask_0 = const()[name = tensor("op_40232_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_359_cast_fp16 = transpose(perm = k_359_perm_0, x = k_357_cast_fp16)[name = tensor("transpose_50")]; + tensor var_40232_cast_fp16 = slice_by_index(begin = var_40232_begin_0, end = var_40232_end_0, end_mask = var_40232_end_mask_0, x = k_359_cast_fp16)[name = tensor("op_40232_cast_fp16")]; + tensor var_40236_begin_0 = const()[name = tensor("op_40236_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_40236_end_0 = const()[name = tensor("op_40236_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_40236_end_mask_0 = const()[name = tensor("op_40236_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40236_cast_fp16 = slice_by_index(begin = var_40236_begin_0, end = var_40236_end_0, end_mask = var_40236_end_mask_0, x = k_359_cast_fp16)[name = tensor("op_40236_cast_fp16")]; + tensor var_40240_begin_0 = const()[name = tensor("op_40240_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_40240_end_0 = const()[name = tensor("op_40240_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_40240_end_mask_0 = const()[name = tensor("op_40240_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40240_cast_fp16 = slice_by_index(begin = var_40240_begin_0, end = var_40240_end_0, end_mask = var_40240_end_mask_0, x = k_359_cast_fp16)[name = tensor("op_40240_cast_fp16")]; + tensor var_40244_begin_0 = const()[name = tensor("op_40244_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_40244_end_0 = const()[name = tensor("op_40244_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_40244_end_mask_0 = const()[name = tensor("op_40244_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40244_cast_fp16 = slice_by_index(begin = var_40244_begin_0, end = var_40244_end_0, end_mask = var_40244_end_mask_0, x = k_359_cast_fp16)[name = tensor("op_40244_cast_fp16")]; + tensor var_40248_begin_0 = const()[name = tensor("op_40248_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_40248_end_0 = const()[name = tensor("op_40248_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_40248_end_mask_0 = const()[name = tensor("op_40248_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40248_cast_fp16 = slice_by_index(begin = var_40248_begin_0, end = var_40248_end_0, end_mask = var_40248_end_mask_0, x = k_359_cast_fp16)[name = tensor("op_40248_cast_fp16")]; + tensor var_40252_begin_0 = const()[name = tensor("op_40252_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_40252_end_0 = const()[name = tensor("op_40252_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_40252_end_mask_0 = const()[name = tensor("op_40252_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40252_cast_fp16 = slice_by_index(begin = var_40252_begin_0, end = var_40252_end_0, end_mask = var_40252_end_mask_0, x = k_359_cast_fp16)[name = tensor("op_40252_cast_fp16")]; + tensor var_40256_begin_0 = const()[name = tensor("op_40256_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_40256_end_0 = const()[name = tensor("op_40256_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_40256_end_mask_0 = const()[name = tensor("op_40256_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40256_cast_fp16 = slice_by_index(begin = var_40256_begin_0, end = var_40256_end_0, end_mask = var_40256_end_mask_0, x = k_359_cast_fp16)[name = tensor("op_40256_cast_fp16")]; + tensor var_40260_begin_0 = const()[name = tensor("op_40260_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_40260_end_0 = const()[name = tensor("op_40260_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_40260_end_mask_0 = const()[name = tensor("op_40260_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40260_cast_fp16 = slice_by_index(begin = var_40260_begin_0, end = var_40260_end_0, end_mask = var_40260_end_mask_0, x = k_359_cast_fp16)[name = tensor("op_40260_cast_fp16")]; + tensor var_40264_begin_0 = const()[name = tensor("op_40264_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_40264_end_0 = const()[name = tensor("op_40264_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_40264_end_mask_0 = const()[name = tensor("op_40264_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40264_cast_fp16 = slice_by_index(begin = var_40264_begin_0, end = var_40264_end_0, end_mask = var_40264_end_mask_0, x = k_359_cast_fp16)[name = tensor("op_40264_cast_fp16")]; + tensor var_40268_begin_0 = const()[name = tensor("op_40268_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_40268_end_0 = const()[name = tensor("op_40268_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_40268_end_mask_0 = const()[name = tensor("op_40268_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40268_cast_fp16 = slice_by_index(begin = var_40268_begin_0, end = var_40268_end_0, end_mask = var_40268_end_mask_0, x = k_359_cast_fp16)[name = tensor("op_40268_cast_fp16")]; + tensor var_40272_begin_0 = const()[name = tensor("op_40272_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_40272_end_0 = const()[name = tensor("op_40272_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_40272_end_mask_0 = const()[name = tensor("op_40272_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40272_cast_fp16 = slice_by_index(begin = var_40272_begin_0, end = var_40272_end_0, end_mask = var_40272_end_mask_0, x = k_359_cast_fp16)[name = tensor("op_40272_cast_fp16")]; + tensor var_40276_begin_0 = const()[name = tensor("op_40276_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_40276_end_0 = const()[name = tensor("op_40276_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_40276_end_mask_0 = const()[name = tensor("op_40276_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40276_cast_fp16 = slice_by_index(begin = var_40276_begin_0, end = var_40276_end_0, end_mask = var_40276_end_mask_0, x = k_359_cast_fp16)[name = tensor("op_40276_cast_fp16")]; + tensor var_40280_begin_0 = const()[name = tensor("op_40280_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_40280_end_0 = const()[name = tensor("op_40280_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_40280_end_mask_0 = const()[name = tensor("op_40280_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40280_cast_fp16 = slice_by_index(begin = var_40280_begin_0, end = var_40280_end_0, end_mask = var_40280_end_mask_0, x = k_359_cast_fp16)[name = tensor("op_40280_cast_fp16")]; + tensor var_40284_begin_0 = const()[name = tensor("op_40284_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_40284_end_0 = const()[name = tensor("op_40284_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_40284_end_mask_0 = const()[name = tensor("op_40284_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40284_cast_fp16 = slice_by_index(begin = var_40284_begin_0, end = var_40284_end_0, end_mask = var_40284_end_mask_0, x = k_359_cast_fp16)[name = tensor("op_40284_cast_fp16")]; + tensor var_40288_begin_0 = const()[name = tensor("op_40288_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_40288_end_0 = const()[name = tensor("op_40288_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_40288_end_mask_0 = const()[name = tensor("op_40288_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40288_cast_fp16 = slice_by_index(begin = var_40288_begin_0, end = var_40288_end_0, end_mask = var_40288_end_mask_0, x = k_359_cast_fp16)[name = tensor("op_40288_cast_fp16")]; + tensor var_40292_begin_0 = const()[name = tensor("op_40292_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_40292_end_0 = const()[name = tensor("op_40292_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_40292_end_mask_0 = const()[name = tensor("op_40292_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40292_cast_fp16 = slice_by_index(begin = var_40292_begin_0, end = var_40292_end_0, end_mask = var_40292_end_mask_0, x = k_359_cast_fp16)[name = tensor("op_40292_cast_fp16")]; + tensor var_40296_begin_0 = const()[name = tensor("op_40296_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_40296_end_0 = const()[name = tensor("op_40296_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_40296_end_mask_0 = const()[name = tensor("op_40296_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40296_cast_fp16 = slice_by_index(begin = var_40296_begin_0, end = var_40296_end_0, end_mask = var_40296_end_mask_0, x = k_359_cast_fp16)[name = tensor("op_40296_cast_fp16")]; + tensor var_40300_begin_0 = const()[name = tensor("op_40300_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_40300_end_0 = const()[name = tensor("op_40300_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_40300_end_mask_0 = const()[name = tensor("op_40300_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40300_cast_fp16 = slice_by_index(begin = var_40300_begin_0, end = var_40300_end_0, end_mask = var_40300_end_mask_0, x = k_359_cast_fp16)[name = tensor("op_40300_cast_fp16")]; + tensor var_40304_begin_0 = const()[name = tensor("op_40304_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_40304_end_0 = const()[name = tensor("op_40304_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_40304_end_mask_0 = const()[name = tensor("op_40304_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40304_cast_fp16 = slice_by_index(begin = var_40304_begin_0, end = var_40304_end_0, end_mask = var_40304_end_mask_0, x = k_359_cast_fp16)[name = tensor("op_40304_cast_fp16")]; + tensor var_40308_begin_0 = const()[name = tensor("op_40308_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_40308_end_0 = const()[name = tensor("op_40308_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_40308_end_mask_0 = const()[name = tensor("op_40308_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40308_cast_fp16 = slice_by_index(begin = var_40308_begin_0, end = var_40308_end_0, end_mask = var_40308_end_mask_0, x = k_359_cast_fp16)[name = tensor("op_40308_cast_fp16")]; + tensor var_40310_begin_0 = const()[name = tensor("op_40310_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_40310_end_0 = const()[name = tensor("op_40310_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_40310_end_mask_0 = const()[name = tensor("op_40310_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40310_cast_fp16 = slice_by_index(begin = var_40310_begin_0, end = var_40310_end_0, end_mask = var_40310_end_mask_0, x = v_179_cast_fp16)[name = tensor("op_40310_cast_fp16")]; + tensor var_40314_begin_0 = const()[name = tensor("op_40314_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_40314_end_0 = const()[name = tensor("op_40314_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_40314_end_mask_0 = const()[name = tensor("op_40314_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40314_cast_fp16 = slice_by_index(begin = var_40314_begin_0, end = var_40314_end_0, end_mask = var_40314_end_mask_0, x = v_179_cast_fp16)[name = tensor("op_40314_cast_fp16")]; + tensor var_40318_begin_0 = const()[name = tensor("op_40318_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_40318_end_0 = const()[name = tensor("op_40318_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_40318_end_mask_0 = const()[name = tensor("op_40318_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40318_cast_fp16 = slice_by_index(begin = var_40318_begin_0, end = var_40318_end_0, end_mask = var_40318_end_mask_0, x = v_179_cast_fp16)[name = tensor("op_40318_cast_fp16")]; + tensor var_40322_begin_0 = const()[name = tensor("op_40322_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_40322_end_0 = const()[name = tensor("op_40322_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_40322_end_mask_0 = const()[name = tensor("op_40322_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40322_cast_fp16 = slice_by_index(begin = var_40322_begin_0, end = var_40322_end_0, end_mask = var_40322_end_mask_0, x = v_179_cast_fp16)[name = tensor("op_40322_cast_fp16")]; + tensor var_40326_begin_0 = const()[name = tensor("op_40326_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_40326_end_0 = const()[name = tensor("op_40326_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_40326_end_mask_0 = const()[name = tensor("op_40326_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40326_cast_fp16 = slice_by_index(begin = var_40326_begin_0, end = var_40326_end_0, end_mask = var_40326_end_mask_0, x = v_179_cast_fp16)[name = tensor("op_40326_cast_fp16")]; + tensor var_40330_begin_0 = const()[name = tensor("op_40330_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_40330_end_0 = const()[name = tensor("op_40330_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_40330_end_mask_0 = const()[name = tensor("op_40330_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40330_cast_fp16 = slice_by_index(begin = var_40330_begin_0, end = var_40330_end_0, end_mask = var_40330_end_mask_0, x = v_179_cast_fp16)[name = tensor("op_40330_cast_fp16")]; + tensor var_40334_begin_0 = const()[name = tensor("op_40334_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_40334_end_0 = const()[name = tensor("op_40334_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_40334_end_mask_0 = const()[name = tensor("op_40334_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40334_cast_fp16 = slice_by_index(begin = var_40334_begin_0, end = var_40334_end_0, end_mask = var_40334_end_mask_0, x = v_179_cast_fp16)[name = tensor("op_40334_cast_fp16")]; + tensor var_40338_begin_0 = const()[name = tensor("op_40338_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_40338_end_0 = const()[name = tensor("op_40338_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_40338_end_mask_0 = const()[name = tensor("op_40338_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40338_cast_fp16 = slice_by_index(begin = var_40338_begin_0, end = var_40338_end_0, end_mask = var_40338_end_mask_0, x = v_179_cast_fp16)[name = tensor("op_40338_cast_fp16")]; + tensor var_40342_begin_0 = const()[name = tensor("op_40342_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_40342_end_0 = const()[name = tensor("op_40342_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_40342_end_mask_0 = const()[name = tensor("op_40342_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40342_cast_fp16 = slice_by_index(begin = var_40342_begin_0, end = var_40342_end_0, end_mask = var_40342_end_mask_0, x = v_179_cast_fp16)[name = tensor("op_40342_cast_fp16")]; + tensor var_40346_begin_0 = const()[name = tensor("op_40346_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_40346_end_0 = const()[name = tensor("op_40346_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_40346_end_mask_0 = const()[name = tensor("op_40346_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40346_cast_fp16 = slice_by_index(begin = var_40346_begin_0, end = var_40346_end_0, end_mask = var_40346_end_mask_0, x = v_179_cast_fp16)[name = tensor("op_40346_cast_fp16")]; + tensor var_40350_begin_0 = const()[name = tensor("op_40350_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_40350_end_0 = const()[name = tensor("op_40350_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_40350_end_mask_0 = const()[name = tensor("op_40350_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40350_cast_fp16 = slice_by_index(begin = var_40350_begin_0, end = var_40350_end_0, end_mask = var_40350_end_mask_0, x = v_179_cast_fp16)[name = tensor("op_40350_cast_fp16")]; + tensor var_40354_begin_0 = const()[name = tensor("op_40354_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_40354_end_0 = const()[name = tensor("op_40354_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_40354_end_mask_0 = const()[name = tensor("op_40354_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40354_cast_fp16 = slice_by_index(begin = var_40354_begin_0, end = var_40354_end_0, end_mask = var_40354_end_mask_0, x = v_179_cast_fp16)[name = tensor("op_40354_cast_fp16")]; + tensor var_40358_begin_0 = const()[name = tensor("op_40358_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_40358_end_0 = const()[name = tensor("op_40358_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_40358_end_mask_0 = const()[name = tensor("op_40358_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40358_cast_fp16 = slice_by_index(begin = var_40358_begin_0, end = var_40358_end_0, end_mask = var_40358_end_mask_0, x = v_179_cast_fp16)[name = tensor("op_40358_cast_fp16")]; + tensor var_40362_begin_0 = const()[name = tensor("op_40362_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_40362_end_0 = const()[name = tensor("op_40362_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_40362_end_mask_0 = const()[name = tensor("op_40362_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40362_cast_fp16 = slice_by_index(begin = var_40362_begin_0, end = var_40362_end_0, end_mask = var_40362_end_mask_0, x = v_179_cast_fp16)[name = tensor("op_40362_cast_fp16")]; + tensor var_40366_begin_0 = const()[name = tensor("op_40366_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_40366_end_0 = const()[name = tensor("op_40366_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_40366_end_mask_0 = const()[name = tensor("op_40366_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40366_cast_fp16 = slice_by_index(begin = var_40366_begin_0, end = var_40366_end_0, end_mask = var_40366_end_mask_0, x = v_179_cast_fp16)[name = tensor("op_40366_cast_fp16")]; + tensor var_40370_begin_0 = const()[name = tensor("op_40370_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_40370_end_0 = const()[name = tensor("op_40370_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_40370_end_mask_0 = const()[name = tensor("op_40370_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40370_cast_fp16 = slice_by_index(begin = var_40370_begin_0, end = var_40370_end_0, end_mask = var_40370_end_mask_0, x = v_179_cast_fp16)[name = tensor("op_40370_cast_fp16")]; + tensor var_40374_begin_0 = const()[name = tensor("op_40374_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_40374_end_0 = const()[name = tensor("op_40374_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_40374_end_mask_0 = const()[name = tensor("op_40374_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40374_cast_fp16 = slice_by_index(begin = var_40374_begin_0, end = var_40374_end_0, end_mask = var_40374_end_mask_0, x = v_179_cast_fp16)[name = tensor("op_40374_cast_fp16")]; + tensor var_40378_begin_0 = const()[name = tensor("op_40378_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_40378_end_0 = const()[name = tensor("op_40378_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_40378_end_mask_0 = const()[name = tensor("op_40378_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40378_cast_fp16 = slice_by_index(begin = var_40378_begin_0, end = var_40378_end_0, end_mask = var_40378_end_mask_0, x = v_179_cast_fp16)[name = tensor("op_40378_cast_fp16")]; + tensor var_40382_begin_0 = const()[name = tensor("op_40382_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_40382_end_0 = const()[name = tensor("op_40382_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_40382_end_mask_0 = const()[name = tensor("op_40382_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40382_cast_fp16 = slice_by_index(begin = var_40382_begin_0, end = var_40382_end_0, end_mask = var_40382_end_mask_0, x = v_179_cast_fp16)[name = tensor("op_40382_cast_fp16")]; + tensor var_40386_begin_0 = const()[name = tensor("op_40386_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_40386_end_0 = const()[name = tensor("op_40386_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_40386_end_mask_0 = const()[name = tensor("op_40386_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40386_cast_fp16 = slice_by_index(begin = var_40386_begin_0, end = var_40386_end_0, end_mask = var_40386_end_mask_0, x = v_179_cast_fp16)[name = tensor("op_40386_cast_fp16")]; + tensor var_40390_equation_0 = const()[name = tensor("op_40390_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40390_cast_fp16 = einsum(equation = var_40390_equation_0, values = (var_40232_cast_fp16, var_40149_cast_fp16))[name = tensor("op_40390_cast_fp16")]; + tensor var_40391_to_fp16 = const()[name = tensor("op_40391_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3401_cast_fp16 = mul(x = var_40390_cast_fp16, y = var_40391_to_fp16)[name = tensor("aw_3401_cast_fp16")]; + tensor var_40394_equation_0 = const()[name = tensor("op_40394_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40394_cast_fp16 = einsum(equation = var_40394_equation_0, values = (var_40236_cast_fp16, var_40153_cast_fp16))[name = tensor("op_40394_cast_fp16")]; + tensor var_40395_to_fp16 = const()[name = tensor("op_40395_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3403_cast_fp16 = mul(x = var_40394_cast_fp16, y = var_40395_to_fp16)[name = tensor("aw_3403_cast_fp16")]; + tensor var_40398_equation_0 = const()[name = tensor("op_40398_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40398_cast_fp16 = einsum(equation = var_40398_equation_0, values = (var_40240_cast_fp16, var_40157_cast_fp16))[name = tensor("op_40398_cast_fp16")]; + tensor var_40399_to_fp16 = const()[name = tensor("op_40399_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3405_cast_fp16 = mul(x = var_40398_cast_fp16, y = var_40399_to_fp16)[name = tensor("aw_3405_cast_fp16")]; + tensor var_40402_equation_0 = const()[name = tensor("op_40402_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40402_cast_fp16 = einsum(equation = var_40402_equation_0, values = (var_40244_cast_fp16, var_40161_cast_fp16))[name = tensor("op_40402_cast_fp16")]; + tensor var_40403_to_fp16 = const()[name = tensor("op_40403_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3407_cast_fp16 = mul(x = var_40402_cast_fp16, y = var_40403_to_fp16)[name = tensor("aw_3407_cast_fp16")]; + tensor var_40406_equation_0 = const()[name = tensor("op_40406_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40406_cast_fp16 = einsum(equation = var_40406_equation_0, values = (var_40248_cast_fp16, var_40165_cast_fp16))[name = tensor("op_40406_cast_fp16")]; + tensor var_40407_to_fp16 = const()[name = tensor("op_40407_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3409_cast_fp16 = mul(x = var_40406_cast_fp16, y = var_40407_to_fp16)[name = tensor("aw_3409_cast_fp16")]; + tensor var_40410_equation_0 = const()[name = tensor("op_40410_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40410_cast_fp16 = einsum(equation = var_40410_equation_0, values = (var_40252_cast_fp16, var_40169_cast_fp16))[name = tensor("op_40410_cast_fp16")]; + tensor var_40411_to_fp16 = const()[name = tensor("op_40411_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3411_cast_fp16 = mul(x = var_40410_cast_fp16, y = var_40411_to_fp16)[name = tensor("aw_3411_cast_fp16")]; + tensor var_40414_equation_0 = const()[name = tensor("op_40414_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40414_cast_fp16 = einsum(equation = var_40414_equation_0, values = (var_40256_cast_fp16, var_40173_cast_fp16))[name = tensor("op_40414_cast_fp16")]; + tensor var_40415_to_fp16 = const()[name = tensor("op_40415_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3413_cast_fp16 = mul(x = var_40414_cast_fp16, y = var_40415_to_fp16)[name = tensor("aw_3413_cast_fp16")]; + tensor var_40418_equation_0 = const()[name = tensor("op_40418_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40418_cast_fp16 = einsum(equation = var_40418_equation_0, values = (var_40260_cast_fp16, var_40177_cast_fp16))[name = tensor("op_40418_cast_fp16")]; + tensor var_40419_to_fp16 = const()[name = tensor("op_40419_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3415_cast_fp16 = mul(x = var_40418_cast_fp16, y = var_40419_to_fp16)[name = tensor("aw_3415_cast_fp16")]; + tensor var_40422_equation_0 = const()[name = tensor("op_40422_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40422_cast_fp16 = einsum(equation = var_40422_equation_0, values = (var_40264_cast_fp16, var_40181_cast_fp16))[name = tensor("op_40422_cast_fp16")]; + tensor var_40423_to_fp16 = const()[name = tensor("op_40423_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3417_cast_fp16 = mul(x = var_40422_cast_fp16, y = var_40423_to_fp16)[name = tensor("aw_3417_cast_fp16")]; + tensor var_40426_equation_0 = const()[name = tensor("op_40426_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40426_cast_fp16 = einsum(equation = var_40426_equation_0, values = (var_40268_cast_fp16, var_40185_cast_fp16))[name = tensor("op_40426_cast_fp16")]; + tensor var_40427_to_fp16 = const()[name = tensor("op_40427_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3419_cast_fp16 = mul(x = var_40426_cast_fp16, y = var_40427_to_fp16)[name = tensor("aw_3419_cast_fp16")]; + tensor var_40430_equation_0 = const()[name = tensor("op_40430_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40430_cast_fp16 = einsum(equation = var_40430_equation_0, values = (var_40272_cast_fp16, var_40189_cast_fp16))[name = tensor("op_40430_cast_fp16")]; + tensor var_40431_to_fp16 = const()[name = tensor("op_40431_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3421_cast_fp16 = mul(x = var_40430_cast_fp16, y = var_40431_to_fp16)[name = tensor("aw_3421_cast_fp16")]; + tensor var_40434_equation_0 = const()[name = tensor("op_40434_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40434_cast_fp16 = einsum(equation = var_40434_equation_0, values = (var_40276_cast_fp16, var_40193_cast_fp16))[name = tensor("op_40434_cast_fp16")]; + tensor var_40435_to_fp16 = const()[name = tensor("op_40435_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3423_cast_fp16 = mul(x = var_40434_cast_fp16, y = var_40435_to_fp16)[name = tensor("aw_3423_cast_fp16")]; + tensor var_40438_equation_0 = const()[name = tensor("op_40438_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40438_cast_fp16 = einsum(equation = var_40438_equation_0, values = (var_40280_cast_fp16, var_40197_cast_fp16))[name = tensor("op_40438_cast_fp16")]; + tensor var_40439_to_fp16 = const()[name = tensor("op_40439_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3425_cast_fp16 = mul(x = var_40438_cast_fp16, y = var_40439_to_fp16)[name = tensor("aw_3425_cast_fp16")]; + tensor var_40442_equation_0 = const()[name = tensor("op_40442_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40442_cast_fp16 = einsum(equation = var_40442_equation_0, values = (var_40284_cast_fp16, var_40201_cast_fp16))[name = tensor("op_40442_cast_fp16")]; + tensor var_40443_to_fp16 = const()[name = tensor("op_40443_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3427_cast_fp16 = mul(x = var_40442_cast_fp16, y = var_40443_to_fp16)[name = tensor("aw_3427_cast_fp16")]; + tensor var_40446_equation_0 = const()[name = tensor("op_40446_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40446_cast_fp16 = einsum(equation = var_40446_equation_0, values = (var_40288_cast_fp16, var_40205_cast_fp16))[name = tensor("op_40446_cast_fp16")]; + tensor var_40447_to_fp16 = const()[name = tensor("op_40447_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3429_cast_fp16 = mul(x = var_40446_cast_fp16, y = var_40447_to_fp16)[name = tensor("aw_3429_cast_fp16")]; + tensor var_40450_equation_0 = const()[name = tensor("op_40450_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40450_cast_fp16 = einsum(equation = var_40450_equation_0, values = (var_40292_cast_fp16, var_40209_cast_fp16))[name = tensor("op_40450_cast_fp16")]; + tensor var_40451_to_fp16 = const()[name = tensor("op_40451_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3431_cast_fp16 = mul(x = var_40450_cast_fp16, y = var_40451_to_fp16)[name = tensor("aw_3431_cast_fp16")]; + tensor var_40454_equation_0 = const()[name = tensor("op_40454_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40454_cast_fp16 = einsum(equation = var_40454_equation_0, values = (var_40296_cast_fp16, var_40213_cast_fp16))[name = tensor("op_40454_cast_fp16")]; + tensor var_40455_to_fp16 = const()[name = tensor("op_40455_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3433_cast_fp16 = mul(x = var_40454_cast_fp16, y = var_40455_to_fp16)[name = tensor("aw_3433_cast_fp16")]; + tensor var_40458_equation_0 = const()[name = tensor("op_40458_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40458_cast_fp16 = einsum(equation = var_40458_equation_0, values = (var_40300_cast_fp16, var_40217_cast_fp16))[name = tensor("op_40458_cast_fp16")]; + tensor var_40459_to_fp16 = const()[name = tensor("op_40459_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3435_cast_fp16 = mul(x = var_40458_cast_fp16, y = var_40459_to_fp16)[name = tensor("aw_3435_cast_fp16")]; + tensor var_40462_equation_0 = const()[name = tensor("op_40462_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40462_cast_fp16 = einsum(equation = var_40462_equation_0, values = (var_40304_cast_fp16, var_40221_cast_fp16))[name = tensor("op_40462_cast_fp16")]; + tensor var_40463_to_fp16 = const()[name = tensor("op_40463_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3437_cast_fp16 = mul(x = var_40462_cast_fp16, y = var_40463_to_fp16)[name = tensor("aw_3437_cast_fp16")]; + tensor var_40466_equation_0 = const()[name = tensor("op_40466_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40466_cast_fp16 = einsum(equation = var_40466_equation_0, values = (var_40308_cast_fp16, var_40225_cast_fp16))[name = tensor("op_40466_cast_fp16")]; + tensor var_40467_to_fp16 = const()[name = tensor("op_40467_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3439_cast_fp16 = mul(x = var_40466_cast_fp16, y = var_40467_to_fp16)[name = tensor("aw_3439_cast_fp16")]; + tensor var_40469_cast_fp16 = softmax(axis = var_30355, x = aw_3401_cast_fp16)[name = tensor("op_40469_cast_fp16")]; + tensor var_40470_cast_fp16 = softmax(axis = var_30355, x = aw_3403_cast_fp16)[name = tensor("op_40470_cast_fp16")]; + tensor var_40471_cast_fp16 = softmax(axis = var_30355, x = aw_3405_cast_fp16)[name = tensor("op_40471_cast_fp16")]; + tensor var_40472_cast_fp16 = softmax(axis = var_30355, x = aw_3407_cast_fp16)[name = tensor("op_40472_cast_fp16")]; + tensor var_40473_cast_fp16 = softmax(axis = var_30355, x = aw_3409_cast_fp16)[name = tensor("op_40473_cast_fp16")]; + tensor var_40474_cast_fp16 = softmax(axis = var_30355, x = aw_3411_cast_fp16)[name = tensor("op_40474_cast_fp16")]; + tensor var_40475_cast_fp16 = softmax(axis = var_30355, x = aw_3413_cast_fp16)[name = tensor("op_40475_cast_fp16")]; + tensor var_40476_cast_fp16 = softmax(axis = var_30355, x = aw_3415_cast_fp16)[name = tensor("op_40476_cast_fp16")]; + tensor var_40477_cast_fp16 = softmax(axis = var_30355, x = aw_3417_cast_fp16)[name = tensor("op_40477_cast_fp16")]; + tensor var_40478_cast_fp16 = softmax(axis = var_30355, x = aw_3419_cast_fp16)[name = tensor("op_40478_cast_fp16")]; + tensor var_40479_cast_fp16 = softmax(axis = var_30355, x = aw_3421_cast_fp16)[name = tensor("op_40479_cast_fp16")]; + tensor var_40480_cast_fp16 = softmax(axis = var_30355, x = aw_3423_cast_fp16)[name = tensor("op_40480_cast_fp16")]; + tensor var_40481_cast_fp16 = softmax(axis = var_30355, x = aw_3425_cast_fp16)[name = tensor("op_40481_cast_fp16")]; + tensor var_40482_cast_fp16 = softmax(axis = var_30355, x = aw_3427_cast_fp16)[name = tensor("op_40482_cast_fp16")]; + tensor var_40483_cast_fp16 = softmax(axis = var_30355, x = aw_3429_cast_fp16)[name = tensor("op_40483_cast_fp16")]; + tensor var_40484_cast_fp16 = softmax(axis = var_30355, x = aw_3431_cast_fp16)[name = tensor("op_40484_cast_fp16")]; + tensor var_40485_cast_fp16 = softmax(axis = var_30355, x = aw_3433_cast_fp16)[name = tensor("op_40485_cast_fp16")]; + tensor var_40486_cast_fp16 = softmax(axis = var_30355, x = aw_3435_cast_fp16)[name = tensor("op_40486_cast_fp16")]; + tensor var_40487_cast_fp16 = softmax(axis = var_30355, x = aw_3437_cast_fp16)[name = tensor("op_40487_cast_fp16")]; + tensor var_40488_cast_fp16 = softmax(axis = var_30355, x = aw_3439_cast_fp16)[name = tensor("op_40488_cast_fp16")]; + tensor var_40490_equation_0 = const()[name = tensor("op_40490_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40490_cast_fp16 = einsum(equation = var_40490_equation_0, values = (var_40310_cast_fp16, var_40469_cast_fp16))[name = tensor("op_40490_cast_fp16")]; + tensor var_40492_equation_0 = const()[name = tensor("op_40492_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40492_cast_fp16 = einsum(equation = var_40492_equation_0, values = (var_40314_cast_fp16, var_40470_cast_fp16))[name = tensor("op_40492_cast_fp16")]; + tensor var_40494_equation_0 = const()[name = tensor("op_40494_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40494_cast_fp16 = einsum(equation = var_40494_equation_0, values = (var_40318_cast_fp16, var_40471_cast_fp16))[name = tensor("op_40494_cast_fp16")]; + tensor var_40496_equation_0 = const()[name = tensor("op_40496_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40496_cast_fp16 = einsum(equation = var_40496_equation_0, values = (var_40322_cast_fp16, var_40472_cast_fp16))[name = tensor("op_40496_cast_fp16")]; + tensor var_40498_equation_0 = const()[name = tensor("op_40498_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40498_cast_fp16 = einsum(equation = var_40498_equation_0, values = (var_40326_cast_fp16, var_40473_cast_fp16))[name = tensor("op_40498_cast_fp16")]; + tensor var_40500_equation_0 = const()[name = tensor("op_40500_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40500_cast_fp16 = einsum(equation = var_40500_equation_0, values = (var_40330_cast_fp16, var_40474_cast_fp16))[name = tensor("op_40500_cast_fp16")]; + tensor var_40502_equation_0 = const()[name = tensor("op_40502_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40502_cast_fp16 = einsum(equation = var_40502_equation_0, values = (var_40334_cast_fp16, var_40475_cast_fp16))[name = tensor("op_40502_cast_fp16")]; + tensor var_40504_equation_0 = const()[name = tensor("op_40504_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40504_cast_fp16 = einsum(equation = var_40504_equation_0, values = (var_40338_cast_fp16, var_40476_cast_fp16))[name = tensor("op_40504_cast_fp16")]; + tensor var_40506_equation_0 = const()[name = tensor("op_40506_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40506_cast_fp16 = einsum(equation = var_40506_equation_0, values = (var_40342_cast_fp16, var_40477_cast_fp16))[name = tensor("op_40506_cast_fp16")]; + tensor var_40508_equation_0 = const()[name = tensor("op_40508_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40508_cast_fp16 = einsum(equation = var_40508_equation_0, values = (var_40346_cast_fp16, var_40478_cast_fp16))[name = tensor("op_40508_cast_fp16")]; + tensor var_40510_equation_0 = const()[name = tensor("op_40510_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40510_cast_fp16 = einsum(equation = var_40510_equation_0, values = (var_40350_cast_fp16, var_40479_cast_fp16))[name = tensor("op_40510_cast_fp16")]; + tensor var_40512_equation_0 = const()[name = tensor("op_40512_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40512_cast_fp16 = einsum(equation = var_40512_equation_0, values = (var_40354_cast_fp16, var_40480_cast_fp16))[name = tensor("op_40512_cast_fp16")]; + tensor var_40514_equation_0 = const()[name = tensor("op_40514_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40514_cast_fp16 = einsum(equation = var_40514_equation_0, values = (var_40358_cast_fp16, var_40481_cast_fp16))[name = tensor("op_40514_cast_fp16")]; + tensor var_40516_equation_0 = const()[name = tensor("op_40516_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40516_cast_fp16 = einsum(equation = var_40516_equation_0, values = (var_40362_cast_fp16, var_40482_cast_fp16))[name = tensor("op_40516_cast_fp16")]; + tensor var_40518_equation_0 = const()[name = tensor("op_40518_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40518_cast_fp16 = einsum(equation = var_40518_equation_0, values = (var_40366_cast_fp16, var_40483_cast_fp16))[name = tensor("op_40518_cast_fp16")]; + tensor var_40520_equation_0 = const()[name = tensor("op_40520_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40520_cast_fp16 = einsum(equation = var_40520_equation_0, values = (var_40370_cast_fp16, var_40484_cast_fp16))[name = tensor("op_40520_cast_fp16")]; + tensor var_40522_equation_0 = const()[name = tensor("op_40522_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40522_cast_fp16 = einsum(equation = var_40522_equation_0, values = (var_40374_cast_fp16, var_40485_cast_fp16))[name = tensor("op_40522_cast_fp16")]; + tensor var_40524_equation_0 = const()[name = tensor("op_40524_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40524_cast_fp16 = einsum(equation = var_40524_equation_0, values = (var_40378_cast_fp16, var_40486_cast_fp16))[name = tensor("op_40524_cast_fp16")]; + tensor var_40526_equation_0 = const()[name = tensor("op_40526_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40526_cast_fp16 = einsum(equation = var_40526_equation_0, values = (var_40382_cast_fp16, var_40487_cast_fp16))[name = tensor("op_40526_cast_fp16")]; + tensor var_40528_equation_0 = const()[name = tensor("op_40528_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40528_cast_fp16 = einsum(equation = var_40528_equation_0, values = (var_40386_cast_fp16, var_40488_cast_fp16))[name = tensor("op_40528_cast_fp16")]; + tensor input_539_interleave_0 = const()[name = tensor("input_539_interleave_0"), val = tensor(false)]; + tensor input_539_cast_fp16 = concat(axis = var_30355, interleave = input_539_interleave_0, values = (var_40490_cast_fp16, var_40492_cast_fp16, var_40494_cast_fp16, var_40496_cast_fp16, var_40498_cast_fp16, var_40500_cast_fp16, var_40502_cast_fp16, var_40504_cast_fp16, var_40506_cast_fp16, var_40508_cast_fp16, var_40510_cast_fp16, var_40512_cast_fp16, var_40514_cast_fp16, var_40516_cast_fp16, var_40518_cast_fp16, var_40520_cast_fp16, var_40522_cast_fp16, var_40524_cast_fp16, var_40526_cast_fp16, var_40528_cast_fp16))[name = tensor("input_539_cast_fp16")]; + tensor var_40534 = const()[name = tensor("op_40534"), val = tensor([1, 1])]; + tensor var_40536 = const()[name = tensor("op_40536"), val = tensor([1, 1])]; + tensor var_40538_pad_type_0 = const()[name = tensor("op_40538_pad_type_0"), val = tensor("custom")]; + tensor var_40538_pad_0 = const()[name = tensor("op_40538_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3415328896)))]; + tensor up_blocks_0_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3418605760)))]; + tensor var_40538_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_40536, groups = var_30355, pad = var_40538_pad_0, pad_type = var_40538_pad_type_0, strides = var_40534, weight = up_blocks_0_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16, x = input_539_cast_fp16)[name = tensor("op_40538_cast_fp16")]; + tensor inputs_269_cast_fp16 = add(x = var_40538_cast_fp16, y = inputs_267_cast_fp16)[name = tensor("inputs_269_cast_fp16")]; + tensor input_541_axes_0 = const()[name = tensor("input_541_axes_0"), val = tensor([1])]; + tensor input_541_gamma_0_to_fp16 = const()[name = tensor("input_541_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3418608384)))]; + tensor input_541_beta_0_to_fp16 = const()[name = tensor("input_541_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3418611008)))]; + tensor var_40548_to_fp16 = const()[name = tensor("op_40548_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_541_cast_fp16 = layer_norm(axes = input_541_axes_0, beta = input_541_beta_0_to_fp16, epsilon = var_40548_to_fp16, gamma = input_541_gamma_0_to_fp16, x = inputs_269_cast_fp16)[name = tensor("input_541_cast_fp16")]; + tensor var_40564 = const()[name = tensor("op_40564"), val = tensor([1, 1])]; + tensor var_40566 = const()[name = tensor("op_40566"), val = tensor([1, 1])]; + tensor var_40568_pad_type_0 = const()[name = tensor("op_40568_pad_type_0"), val = tensor("custom")]; + tensor var_40568_pad_0 = const()[name = tensor("op_40568_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3418613632)))]; + tensor up_blocks_0_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3444828096)))]; + tensor var_40568_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_40566, groups = var_30355, pad = var_40568_pad_0, pad_type = var_40568_pad_type_0, strides = var_40564, weight = up_blocks_0_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16, x = input_541_cast_fp16)[name = tensor("op_40568_cast_fp16")]; + tensor var_40569_split_sizes_0 = const()[name = tensor("op_40569_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_40569_axis_0 = const()[name = tensor("op_40569_axis_0"), val = tensor(1)]; + tensor var_40569_cast_fp16_0, tensor var_40569_cast_fp16_1 = split(axis = var_40569_axis_0, split_sizes = var_40569_split_sizes_0, x = var_40568_cast_fp16)[name = tensor("op_40569_cast_fp16")]; + tensor var_40571_mode_0 = const()[name = tensor("op_40571_mode_0"), val = tensor("EXACT")]; + tensor var_40571_cast_fp16 = gelu(mode = var_40571_mode_0, x = var_40569_cast_fp16_1)[name = tensor("op_40571_cast_fp16")]; + tensor input_543_cast_fp16 = mul(x = var_40569_cast_fp16_0, y = var_40571_cast_fp16)[name = tensor("input_543_cast_fp16")]; + tensor var_40575 = const()[name = tensor("op_40575"), val = tensor([1, 1])]; + tensor var_40577 = const()[name = tensor("op_40577"), val = tensor([1, 1])]; + tensor var_40579_pad_type_0 = const()[name = tensor("op_40579_pad_type_0"), val = tensor("custom")]; + tensor var_40579_pad_0 = const()[name = tensor("op_40579_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3444848640)))]; + tensor up_blocks_0_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3457955904)))]; + tensor var_40579_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_40577, groups = var_30355, pad = var_40579_pad_0, pad_type = var_40579_pad_type_0, strides = var_40575, weight = up_blocks_0_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16, x = input_543_cast_fp16)[name = tensor("op_40579_cast_fp16")]; + tensor inputs_271_cast_fp16 = add(x = var_40579_cast_fp16, y = inputs_269_cast_fp16)[name = tensor("inputs_271_cast_fp16")]; + tensor hidden_states_367_axes_0 = const()[name = tensor("hidden_states_367_axes_0"), val = tensor([1])]; + tensor hidden_states_367_gamma_0_to_fp16 = const()[name = tensor("hidden_states_367_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3457958528)))]; + tensor hidden_states_367_beta_0_to_fp16 = const()[name = tensor("hidden_states_367_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3457961152)))]; + tensor var_40595_to_fp16 = const()[name = tensor("op_40595_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_367_cast_fp16 = layer_norm(axes = hidden_states_367_axes_0, beta = hidden_states_367_beta_0_to_fp16, epsilon = var_40595_to_fp16, gamma = hidden_states_367_gamma_0_to_fp16, x = inputs_271_cast_fp16)[name = tensor("hidden_states_367_cast_fp16")]; + tensor var_40610 = const()[name = tensor("op_40610"), val = tensor([1, 1])]; + tensor var_40612 = const()[name = tensor("op_40612"), val = tensor([1, 1])]; + tensor q_181_pad_type_0 = const()[name = tensor("q_181_pad_type_0"), val = tensor("custom")]; + tensor q_181_pad_0 = const()[name = tensor("q_181_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_1_attn1_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_1_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3457963776)))]; + tensor q_181_cast_fp16 = conv(dilations = var_40612, groups = var_30355, pad = q_181_pad_0, pad_type = q_181_pad_type_0, strides = var_40610, weight = up_blocks_0_attentions_1_transformer_blocks_1_attn1_to_q_weight_to_fp16, x = hidden_states_367_cast_fp16)[name = tensor("q_181_cast_fp16")]; + tensor var_40616 = const()[name = tensor("op_40616"), val = tensor([1, 1])]; + tensor var_40618 = const()[name = tensor("op_40618"), val = tensor([1, 1])]; + tensor k_361_pad_type_0 = const()[name = tensor("k_361_pad_type_0"), val = tensor("custom")]; + tensor k_361_pad_0 = const()[name = tensor("k_361_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_1_attn1_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_1_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3461240640)))]; + tensor k_361_cast_fp16 = conv(dilations = var_40618, groups = var_30355, pad = k_361_pad_0, pad_type = k_361_pad_type_0, strides = var_40616, weight = up_blocks_0_attentions_1_transformer_blocks_1_attn1_to_k_weight_to_fp16, x = hidden_states_367_cast_fp16)[name = tensor("k_361_cast_fp16")]; + tensor var_40622 = const()[name = tensor("op_40622"), val = tensor([1, 1])]; + tensor var_40624 = const()[name = tensor("op_40624"), val = tensor([1, 1])]; + tensor v_181_pad_type_0 = const()[name = tensor("v_181_pad_type_0"), val = tensor("custom")]; + tensor v_181_pad_0 = const()[name = tensor("v_181_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_1_attn1_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_1_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3464517504)))]; + tensor v_181_cast_fp16 = conv(dilations = var_40624, groups = var_30355, pad = v_181_pad_0, pad_type = v_181_pad_type_0, strides = var_40622, weight = up_blocks_0_attentions_1_transformer_blocks_1_attn1_to_v_weight_to_fp16, x = hidden_states_367_cast_fp16)[name = tensor("v_181_cast_fp16")]; + tensor var_40628_begin_0 = const()[name = tensor("op_40628_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_40628_end_0 = const()[name = tensor("op_40628_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_40628_end_mask_0 = const()[name = tensor("op_40628_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40628_cast_fp16 = slice_by_index(begin = var_40628_begin_0, end = var_40628_end_0, end_mask = var_40628_end_mask_0, x = q_181_cast_fp16)[name = tensor("op_40628_cast_fp16")]; + tensor var_40632_begin_0 = const()[name = tensor("op_40632_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_40632_end_0 = const()[name = tensor("op_40632_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_40632_end_mask_0 = const()[name = tensor("op_40632_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40632_cast_fp16 = slice_by_index(begin = var_40632_begin_0, end = var_40632_end_0, end_mask = var_40632_end_mask_0, x = q_181_cast_fp16)[name = tensor("op_40632_cast_fp16")]; + tensor var_40636_begin_0 = const()[name = tensor("op_40636_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_40636_end_0 = const()[name = tensor("op_40636_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_40636_end_mask_0 = const()[name = tensor("op_40636_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40636_cast_fp16 = slice_by_index(begin = var_40636_begin_0, end = var_40636_end_0, end_mask = var_40636_end_mask_0, x = q_181_cast_fp16)[name = tensor("op_40636_cast_fp16")]; + tensor var_40640_begin_0 = const()[name = tensor("op_40640_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_40640_end_0 = const()[name = tensor("op_40640_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_40640_end_mask_0 = const()[name = tensor("op_40640_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40640_cast_fp16 = slice_by_index(begin = var_40640_begin_0, end = var_40640_end_0, end_mask = var_40640_end_mask_0, x = q_181_cast_fp16)[name = tensor("op_40640_cast_fp16")]; + tensor var_40644_begin_0 = const()[name = tensor("op_40644_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_40644_end_0 = const()[name = tensor("op_40644_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_40644_end_mask_0 = const()[name = tensor("op_40644_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40644_cast_fp16 = slice_by_index(begin = var_40644_begin_0, end = var_40644_end_0, end_mask = var_40644_end_mask_0, x = q_181_cast_fp16)[name = tensor("op_40644_cast_fp16")]; + tensor var_40648_begin_0 = const()[name = tensor("op_40648_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_40648_end_0 = const()[name = tensor("op_40648_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_40648_end_mask_0 = const()[name = tensor("op_40648_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40648_cast_fp16 = slice_by_index(begin = var_40648_begin_0, end = var_40648_end_0, end_mask = var_40648_end_mask_0, x = q_181_cast_fp16)[name = tensor("op_40648_cast_fp16")]; + tensor var_40652_begin_0 = const()[name = tensor("op_40652_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_40652_end_0 = const()[name = tensor("op_40652_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_40652_end_mask_0 = const()[name = tensor("op_40652_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40652_cast_fp16 = slice_by_index(begin = var_40652_begin_0, end = var_40652_end_0, end_mask = var_40652_end_mask_0, x = q_181_cast_fp16)[name = tensor("op_40652_cast_fp16")]; + tensor var_40656_begin_0 = const()[name = tensor("op_40656_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_40656_end_0 = const()[name = tensor("op_40656_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_40656_end_mask_0 = const()[name = tensor("op_40656_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40656_cast_fp16 = slice_by_index(begin = var_40656_begin_0, end = var_40656_end_0, end_mask = var_40656_end_mask_0, x = q_181_cast_fp16)[name = tensor("op_40656_cast_fp16")]; + tensor var_40660_begin_0 = const()[name = tensor("op_40660_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_40660_end_0 = const()[name = tensor("op_40660_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_40660_end_mask_0 = const()[name = tensor("op_40660_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40660_cast_fp16 = slice_by_index(begin = var_40660_begin_0, end = var_40660_end_0, end_mask = var_40660_end_mask_0, x = q_181_cast_fp16)[name = tensor("op_40660_cast_fp16")]; + tensor var_40664_begin_0 = const()[name = tensor("op_40664_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_40664_end_0 = const()[name = tensor("op_40664_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_40664_end_mask_0 = const()[name = tensor("op_40664_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40664_cast_fp16 = slice_by_index(begin = var_40664_begin_0, end = var_40664_end_0, end_mask = var_40664_end_mask_0, x = q_181_cast_fp16)[name = tensor("op_40664_cast_fp16")]; + tensor var_40668_begin_0 = const()[name = tensor("op_40668_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_40668_end_0 = const()[name = tensor("op_40668_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_40668_end_mask_0 = const()[name = tensor("op_40668_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40668_cast_fp16 = slice_by_index(begin = var_40668_begin_0, end = var_40668_end_0, end_mask = var_40668_end_mask_0, x = q_181_cast_fp16)[name = tensor("op_40668_cast_fp16")]; + tensor var_40672_begin_0 = const()[name = tensor("op_40672_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_40672_end_0 = const()[name = tensor("op_40672_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_40672_end_mask_0 = const()[name = tensor("op_40672_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40672_cast_fp16 = slice_by_index(begin = var_40672_begin_0, end = var_40672_end_0, end_mask = var_40672_end_mask_0, x = q_181_cast_fp16)[name = tensor("op_40672_cast_fp16")]; + tensor var_40676_begin_0 = const()[name = tensor("op_40676_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_40676_end_0 = const()[name = tensor("op_40676_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_40676_end_mask_0 = const()[name = tensor("op_40676_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40676_cast_fp16 = slice_by_index(begin = var_40676_begin_0, end = var_40676_end_0, end_mask = var_40676_end_mask_0, x = q_181_cast_fp16)[name = tensor("op_40676_cast_fp16")]; + tensor var_40680_begin_0 = const()[name = tensor("op_40680_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_40680_end_0 = const()[name = tensor("op_40680_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_40680_end_mask_0 = const()[name = tensor("op_40680_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40680_cast_fp16 = slice_by_index(begin = var_40680_begin_0, end = var_40680_end_0, end_mask = var_40680_end_mask_0, x = q_181_cast_fp16)[name = tensor("op_40680_cast_fp16")]; + tensor var_40684_begin_0 = const()[name = tensor("op_40684_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_40684_end_0 = const()[name = tensor("op_40684_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_40684_end_mask_0 = const()[name = tensor("op_40684_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40684_cast_fp16 = slice_by_index(begin = var_40684_begin_0, end = var_40684_end_0, end_mask = var_40684_end_mask_0, x = q_181_cast_fp16)[name = tensor("op_40684_cast_fp16")]; + tensor var_40688_begin_0 = const()[name = tensor("op_40688_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_40688_end_0 = const()[name = tensor("op_40688_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_40688_end_mask_0 = const()[name = tensor("op_40688_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40688_cast_fp16 = slice_by_index(begin = var_40688_begin_0, end = var_40688_end_0, end_mask = var_40688_end_mask_0, x = q_181_cast_fp16)[name = tensor("op_40688_cast_fp16")]; + tensor var_40692_begin_0 = const()[name = tensor("op_40692_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_40692_end_0 = const()[name = tensor("op_40692_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_40692_end_mask_0 = const()[name = tensor("op_40692_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40692_cast_fp16 = slice_by_index(begin = var_40692_begin_0, end = var_40692_end_0, end_mask = var_40692_end_mask_0, x = q_181_cast_fp16)[name = tensor("op_40692_cast_fp16")]; + tensor var_40696_begin_0 = const()[name = tensor("op_40696_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_40696_end_0 = const()[name = tensor("op_40696_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_40696_end_mask_0 = const()[name = tensor("op_40696_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40696_cast_fp16 = slice_by_index(begin = var_40696_begin_0, end = var_40696_end_0, end_mask = var_40696_end_mask_0, x = q_181_cast_fp16)[name = tensor("op_40696_cast_fp16")]; + tensor var_40700_begin_0 = const()[name = tensor("op_40700_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_40700_end_0 = const()[name = tensor("op_40700_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_40700_end_mask_0 = const()[name = tensor("op_40700_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40700_cast_fp16 = slice_by_index(begin = var_40700_begin_0, end = var_40700_end_0, end_mask = var_40700_end_mask_0, x = q_181_cast_fp16)[name = tensor("op_40700_cast_fp16")]; + tensor var_40704_begin_0 = const()[name = tensor("op_40704_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_40704_end_0 = const()[name = tensor("op_40704_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_40704_end_mask_0 = const()[name = tensor("op_40704_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40704_cast_fp16 = slice_by_index(begin = var_40704_begin_0, end = var_40704_end_0, end_mask = var_40704_end_mask_0, x = q_181_cast_fp16)[name = tensor("op_40704_cast_fp16")]; + tensor k_363_perm_0 = const()[name = tensor("k_363_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_40711_begin_0 = const()[name = tensor("op_40711_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_40711_end_0 = const()[name = tensor("op_40711_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_40711_end_mask_0 = const()[name = tensor("op_40711_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_363_cast_fp16 = transpose(perm = k_363_perm_0, x = k_361_cast_fp16)[name = tensor("transpose_49")]; + tensor var_40711_cast_fp16 = slice_by_index(begin = var_40711_begin_0, end = var_40711_end_0, end_mask = var_40711_end_mask_0, x = k_363_cast_fp16)[name = tensor("op_40711_cast_fp16")]; + tensor var_40715_begin_0 = const()[name = tensor("op_40715_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_40715_end_0 = const()[name = tensor("op_40715_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_40715_end_mask_0 = const()[name = tensor("op_40715_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40715_cast_fp16 = slice_by_index(begin = var_40715_begin_0, end = var_40715_end_0, end_mask = var_40715_end_mask_0, x = k_363_cast_fp16)[name = tensor("op_40715_cast_fp16")]; + tensor var_40719_begin_0 = const()[name = tensor("op_40719_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_40719_end_0 = const()[name = tensor("op_40719_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_40719_end_mask_0 = const()[name = tensor("op_40719_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40719_cast_fp16 = slice_by_index(begin = var_40719_begin_0, end = var_40719_end_0, end_mask = var_40719_end_mask_0, x = k_363_cast_fp16)[name = tensor("op_40719_cast_fp16")]; + tensor var_40723_begin_0 = const()[name = tensor("op_40723_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_40723_end_0 = const()[name = tensor("op_40723_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_40723_end_mask_0 = const()[name = tensor("op_40723_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40723_cast_fp16 = slice_by_index(begin = var_40723_begin_0, end = var_40723_end_0, end_mask = var_40723_end_mask_0, x = k_363_cast_fp16)[name = tensor("op_40723_cast_fp16")]; + tensor var_40727_begin_0 = const()[name = tensor("op_40727_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_40727_end_0 = const()[name = tensor("op_40727_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_40727_end_mask_0 = const()[name = tensor("op_40727_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40727_cast_fp16 = slice_by_index(begin = var_40727_begin_0, end = var_40727_end_0, end_mask = var_40727_end_mask_0, x = k_363_cast_fp16)[name = tensor("op_40727_cast_fp16")]; + tensor var_40731_begin_0 = const()[name = tensor("op_40731_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_40731_end_0 = const()[name = tensor("op_40731_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_40731_end_mask_0 = const()[name = tensor("op_40731_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40731_cast_fp16 = slice_by_index(begin = var_40731_begin_0, end = var_40731_end_0, end_mask = var_40731_end_mask_0, x = k_363_cast_fp16)[name = tensor("op_40731_cast_fp16")]; + tensor var_40735_begin_0 = const()[name = tensor("op_40735_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_40735_end_0 = const()[name = tensor("op_40735_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_40735_end_mask_0 = const()[name = tensor("op_40735_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40735_cast_fp16 = slice_by_index(begin = var_40735_begin_0, end = var_40735_end_0, end_mask = var_40735_end_mask_0, x = k_363_cast_fp16)[name = tensor("op_40735_cast_fp16")]; + tensor var_40739_begin_0 = const()[name = tensor("op_40739_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_40739_end_0 = const()[name = tensor("op_40739_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_40739_end_mask_0 = const()[name = tensor("op_40739_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40739_cast_fp16 = slice_by_index(begin = var_40739_begin_0, end = var_40739_end_0, end_mask = var_40739_end_mask_0, x = k_363_cast_fp16)[name = tensor("op_40739_cast_fp16")]; + tensor var_40743_begin_0 = const()[name = tensor("op_40743_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_40743_end_0 = const()[name = tensor("op_40743_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_40743_end_mask_0 = const()[name = tensor("op_40743_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40743_cast_fp16 = slice_by_index(begin = var_40743_begin_0, end = var_40743_end_0, end_mask = var_40743_end_mask_0, x = k_363_cast_fp16)[name = tensor("op_40743_cast_fp16")]; + tensor var_40747_begin_0 = const()[name = tensor("op_40747_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_40747_end_0 = const()[name = tensor("op_40747_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_40747_end_mask_0 = const()[name = tensor("op_40747_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40747_cast_fp16 = slice_by_index(begin = var_40747_begin_0, end = var_40747_end_0, end_mask = var_40747_end_mask_0, x = k_363_cast_fp16)[name = tensor("op_40747_cast_fp16")]; + tensor var_40751_begin_0 = const()[name = tensor("op_40751_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_40751_end_0 = const()[name = tensor("op_40751_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_40751_end_mask_0 = const()[name = tensor("op_40751_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40751_cast_fp16 = slice_by_index(begin = var_40751_begin_0, end = var_40751_end_0, end_mask = var_40751_end_mask_0, x = k_363_cast_fp16)[name = tensor("op_40751_cast_fp16")]; + tensor var_40755_begin_0 = const()[name = tensor("op_40755_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_40755_end_0 = const()[name = tensor("op_40755_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_40755_end_mask_0 = const()[name = tensor("op_40755_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40755_cast_fp16 = slice_by_index(begin = var_40755_begin_0, end = var_40755_end_0, end_mask = var_40755_end_mask_0, x = k_363_cast_fp16)[name = tensor("op_40755_cast_fp16")]; + tensor var_40759_begin_0 = const()[name = tensor("op_40759_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_40759_end_0 = const()[name = tensor("op_40759_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_40759_end_mask_0 = const()[name = tensor("op_40759_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40759_cast_fp16 = slice_by_index(begin = var_40759_begin_0, end = var_40759_end_0, end_mask = var_40759_end_mask_0, x = k_363_cast_fp16)[name = tensor("op_40759_cast_fp16")]; + tensor var_40763_begin_0 = const()[name = tensor("op_40763_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_40763_end_0 = const()[name = tensor("op_40763_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_40763_end_mask_0 = const()[name = tensor("op_40763_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40763_cast_fp16 = slice_by_index(begin = var_40763_begin_0, end = var_40763_end_0, end_mask = var_40763_end_mask_0, x = k_363_cast_fp16)[name = tensor("op_40763_cast_fp16")]; + tensor var_40767_begin_0 = const()[name = tensor("op_40767_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_40767_end_0 = const()[name = tensor("op_40767_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_40767_end_mask_0 = const()[name = tensor("op_40767_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40767_cast_fp16 = slice_by_index(begin = var_40767_begin_0, end = var_40767_end_0, end_mask = var_40767_end_mask_0, x = k_363_cast_fp16)[name = tensor("op_40767_cast_fp16")]; + tensor var_40771_begin_0 = const()[name = tensor("op_40771_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_40771_end_0 = const()[name = tensor("op_40771_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_40771_end_mask_0 = const()[name = tensor("op_40771_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40771_cast_fp16 = slice_by_index(begin = var_40771_begin_0, end = var_40771_end_0, end_mask = var_40771_end_mask_0, x = k_363_cast_fp16)[name = tensor("op_40771_cast_fp16")]; + tensor var_40775_begin_0 = const()[name = tensor("op_40775_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_40775_end_0 = const()[name = tensor("op_40775_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_40775_end_mask_0 = const()[name = tensor("op_40775_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40775_cast_fp16 = slice_by_index(begin = var_40775_begin_0, end = var_40775_end_0, end_mask = var_40775_end_mask_0, x = k_363_cast_fp16)[name = tensor("op_40775_cast_fp16")]; + tensor var_40779_begin_0 = const()[name = tensor("op_40779_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_40779_end_0 = const()[name = tensor("op_40779_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_40779_end_mask_0 = const()[name = tensor("op_40779_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40779_cast_fp16 = slice_by_index(begin = var_40779_begin_0, end = var_40779_end_0, end_mask = var_40779_end_mask_0, x = k_363_cast_fp16)[name = tensor("op_40779_cast_fp16")]; + tensor var_40783_begin_0 = const()[name = tensor("op_40783_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_40783_end_0 = const()[name = tensor("op_40783_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_40783_end_mask_0 = const()[name = tensor("op_40783_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40783_cast_fp16 = slice_by_index(begin = var_40783_begin_0, end = var_40783_end_0, end_mask = var_40783_end_mask_0, x = k_363_cast_fp16)[name = tensor("op_40783_cast_fp16")]; + tensor var_40787_begin_0 = const()[name = tensor("op_40787_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_40787_end_0 = const()[name = tensor("op_40787_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_40787_end_mask_0 = const()[name = tensor("op_40787_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40787_cast_fp16 = slice_by_index(begin = var_40787_begin_0, end = var_40787_end_0, end_mask = var_40787_end_mask_0, x = k_363_cast_fp16)[name = tensor("op_40787_cast_fp16")]; + tensor var_40789_begin_0 = const()[name = tensor("op_40789_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_40789_end_0 = const()[name = tensor("op_40789_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_40789_end_mask_0 = const()[name = tensor("op_40789_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40789_cast_fp16 = slice_by_index(begin = var_40789_begin_0, end = var_40789_end_0, end_mask = var_40789_end_mask_0, x = v_181_cast_fp16)[name = tensor("op_40789_cast_fp16")]; + tensor var_40793_begin_0 = const()[name = tensor("op_40793_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_40793_end_0 = const()[name = tensor("op_40793_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_40793_end_mask_0 = const()[name = tensor("op_40793_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40793_cast_fp16 = slice_by_index(begin = var_40793_begin_0, end = var_40793_end_0, end_mask = var_40793_end_mask_0, x = v_181_cast_fp16)[name = tensor("op_40793_cast_fp16")]; + tensor var_40797_begin_0 = const()[name = tensor("op_40797_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_40797_end_0 = const()[name = tensor("op_40797_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_40797_end_mask_0 = const()[name = tensor("op_40797_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40797_cast_fp16 = slice_by_index(begin = var_40797_begin_0, end = var_40797_end_0, end_mask = var_40797_end_mask_0, x = v_181_cast_fp16)[name = tensor("op_40797_cast_fp16")]; + tensor var_40801_begin_0 = const()[name = tensor("op_40801_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_40801_end_0 = const()[name = tensor("op_40801_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_40801_end_mask_0 = const()[name = tensor("op_40801_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40801_cast_fp16 = slice_by_index(begin = var_40801_begin_0, end = var_40801_end_0, end_mask = var_40801_end_mask_0, x = v_181_cast_fp16)[name = tensor("op_40801_cast_fp16")]; + tensor var_40805_begin_0 = const()[name = tensor("op_40805_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_40805_end_0 = const()[name = tensor("op_40805_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_40805_end_mask_0 = const()[name = tensor("op_40805_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40805_cast_fp16 = slice_by_index(begin = var_40805_begin_0, end = var_40805_end_0, end_mask = var_40805_end_mask_0, x = v_181_cast_fp16)[name = tensor("op_40805_cast_fp16")]; + tensor var_40809_begin_0 = const()[name = tensor("op_40809_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_40809_end_0 = const()[name = tensor("op_40809_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_40809_end_mask_0 = const()[name = tensor("op_40809_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40809_cast_fp16 = slice_by_index(begin = var_40809_begin_0, end = var_40809_end_0, end_mask = var_40809_end_mask_0, x = v_181_cast_fp16)[name = tensor("op_40809_cast_fp16")]; + tensor var_40813_begin_0 = const()[name = tensor("op_40813_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_40813_end_0 = const()[name = tensor("op_40813_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_40813_end_mask_0 = const()[name = tensor("op_40813_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40813_cast_fp16 = slice_by_index(begin = var_40813_begin_0, end = var_40813_end_0, end_mask = var_40813_end_mask_0, x = v_181_cast_fp16)[name = tensor("op_40813_cast_fp16")]; + tensor var_40817_begin_0 = const()[name = tensor("op_40817_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_40817_end_0 = const()[name = tensor("op_40817_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_40817_end_mask_0 = const()[name = tensor("op_40817_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40817_cast_fp16 = slice_by_index(begin = var_40817_begin_0, end = var_40817_end_0, end_mask = var_40817_end_mask_0, x = v_181_cast_fp16)[name = tensor("op_40817_cast_fp16")]; + tensor var_40821_begin_0 = const()[name = tensor("op_40821_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_40821_end_0 = const()[name = tensor("op_40821_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_40821_end_mask_0 = const()[name = tensor("op_40821_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40821_cast_fp16 = slice_by_index(begin = var_40821_begin_0, end = var_40821_end_0, end_mask = var_40821_end_mask_0, x = v_181_cast_fp16)[name = tensor("op_40821_cast_fp16")]; + tensor var_40825_begin_0 = const()[name = tensor("op_40825_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_40825_end_0 = const()[name = tensor("op_40825_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_40825_end_mask_0 = const()[name = tensor("op_40825_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40825_cast_fp16 = slice_by_index(begin = var_40825_begin_0, end = var_40825_end_0, end_mask = var_40825_end_mask_0, x = v_181_cast_fp16)[name = tensor("op_40825_cast_fp16")]; + tensor var_40829_begin_0 = const()[name = tensor("op_40829_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_40829_end_0 = const()[name = tensor("op_40829_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_40829_end_mask_0 = const()[name = tensor("op_40829_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40829_cast_fp16 = slice_by_index(begin = var_40829_begin_0, end = var_40829_end_0, end_mask = var_40829_end_mask_0, x = v_181_cast_fp16)[name = tensor("op_40829_cast_fp16")]; + tensor var_40833_begin_0 = const()[name = tensor("op_40833_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_40833_end_0 = const()[name = tensor("op_40833_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_40833_end_mask_0 = const()[name = tensor("op_40833_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40833_cast_fp16 = slice_by_index(begin = var_40833_begin_0, end = var_40833_end_0, end_mask = var_40833_end_mask_0, x = v_181_cast_fp16)[name = tensor("op_40833_cast_fp16")]; + tensor var_40837_begin_0 = const()[name = tensor("op_40837_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_40837_end_0 = const()[name = tensor("op_40837_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_40837_end_mask_0 = const()[name = tensor("op_40837_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40837_cast_fp16 = slice_by_index(begin = var_40837_begin_0, end = var_40837_end_0, end_mask = var_40837_end_mask_0, x = v_181_cast_fp16)[name = tensor("op_40837_cast_fp16")]; + tensor var_40841_begin_0 = const()[name = tensor("op_40841_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_40841_end_0 = const()[name = tensor("op_40841_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_40841_end_mask_0 = const()[name = tensor("op_40841_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40841_cast_fp16 = slice_by_index(begin = var_40841_begin_0, end = var_40841_end_0, end_mask = var_40841_end_mask_0, x = v_181_cast_fp16)[name = tensor("op_40841_cast_fp16")]; + tensor var_40845_begin_0 = const()[name = tensor("op_40845_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_40845_end_0 = const()[name = tensor("op_40845_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_40845_end_mask_0 = const()[name = tensor("op_40845_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40845_cast_fp16 = slice_by_index(begin = var_40845_begin_0, end = var_40845_end_0, end_mask = var_40845_end_mask_0, x = v_181_cast_fp16)[name = tensor("op_40845_cast_fp16")]; + tensor var_40849_begin_0 = const()[name = tensor("op_40849_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_40849_end_0 = const()[name = tensor("op_40849_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_40849_end_mask_0 = const()[name = tensor("op_40849_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40849_cast_fp16 = slice_by_index(begin = var_40849_begin_0, end = var_40849_end_0, end_mask = var_40849_end_mask_0, x = v_181_cast_fp16)[name = tensor("op_40849_cast_fp16")]; + tensor var_40853_begin_0 = const()[name = tensor("op_40853_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_40853_end_0 = const()[name = tensor("op_40853_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_40853_end_mask_0 = const()[name = tensor("op_40853_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40853_cast_fp16 = slice_by_index(begin = var_40853_begin_0, end = var_40853_end_0, end_mask = var_40853_end_mask_0, x = v_181_cast_fp16)[name = tensor("op_40853_cast_fp16")]; + tensor var_40857_begin_0 = const()[name = tensor("op_40857_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_40857_end_0 = const()[name = tensor("op_40857_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_40857_end_mask_0 = const()[name = tensor("op_40857_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40857_cast_fp16 = slice_by_index(begin = var_40857_begin_0, end = var_40857_end_0, end_mask = var_40857_end_mask_0, x = v_181_cast_fp16)[name = tensor("op_40857_cast_fp16")]; + tensor var_40861_begin_0 = const()[name = tensor("op_40861_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_40861_end_0 = const()[name = tensor("op_40861_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_40861_end_mask_0 = const()[name = tensor("op_40861_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40861_cast_fp16 = slice_by_index(begin = var_40861_begin_0, end = var_40861_end_0, end_mask = var_40861_end_mask_0, x = v_181_cast_fp16)[name = tensor("op_40861_cast_fp16")]; + tensor var_40865_begin_0 = const()[name = tensor("op_40865_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_40865_end_0 = const()[name = tensor("op_40865_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_40865_end_mask_0 = const()[name = tensor("op_40865_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40865_cast_fp16 = slice_by_index(begin = var_40865_begin_0, end = var_40865_end_0, end_mask = var_40865_end_mask_0, x = v_181_cast_fp16)[name = tensor("op_40865_cast_fp16")]; + tensor var_40869_equation_0 = const()[name = tensor("op_40869_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40869_cast_fp16 = einsum(equation = var_40869_equation_0, values = (var_40711_cast_fp16, var_40628_cast_fp16))[name = tensor("op_40869_cast_fp16")]; + tensor var_40870_to_fp16 = const()[name = tensor("op_40870_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3441_cast_fp16 = mul(x = var_40869_cast_fp16, y = var_40870_to_fp16)[name = tensor("aw_3441_cast_fp16")]; + tensor var_40873_equation_0 = const()[name = tensor("op_40873_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40873_cast_fp16 = einsum(equation = var_40873_equation_0, values = (var_40715_cast_fp16, var_40632_cast_fp16))[name = tensor("op_40873_cast_fp16")]; + tensor var_40874_to_fp16 = const()[name = tensor("op_40874_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3443_cast_fp16 = mul(x = var_40873_cast_fp16, y = var_40874_to_fp16)[name = tensor("aw_3443_cast_fp16")]; + tensor var_40877_equation_0 = const()[name = tensor("op_40877_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40877_cast_fp16 = einsum(equation = var_40877_equation_0, values = (var_40719_cast_fp16, var_40636_cast_fp16))[name = tensor("op_40877_cast_fp16")]; + tensor var_40878_to_fp16 = const()[name = tensor("op_40878_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3445_cast_fp16 = mul(x = var_40877_cast_fp16, y = var_40878_to_fp16)[name = tensor("aw_3445_cast_fp16")]; + tensor var_40881_equation_0 = const()[name = tensor("op_40881_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40881_cast_fp16 = einsum(equation = var_40881_equation_0, values = (var_40723_cast_fp16, var_40640_cast_fp16))[name = tensor("op_40881_cast_fp16")]; + tensor var_40882_to_fp16 = const()[name = tensor("op_40882_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3447_cast_fp16 = mul(x = var_40881_cast_fp16, y = var_40882_to_fp16)[name = tensor("aw_3447_cast_fp16")]; + tensor var_40885_equation_0 = const()[name = tensor("op_40885_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40885_cast_fp16 = einsum(equation = var_40885_equation_0, values = (var_40727_cast_fp16, var_40644_cast_fp16))[name = tensor("op_40885_cast_fp16")]; + tensor var_40886_to_fp16 = const()[name = tensor("op_40886_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3449_cast_fp16 = mul(x = var_40885_cast_fp16, y = var_40886_to_fp16)[name = tensor("aw_3449_cast_fp16")]; + tensor var_40889_equation_0 = const()[name = tensor("op_40889_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40889_cast_fp16 = einsum(equation = var_40889_equation_0, values = (var_40731_cast_fp16, var_40648_cast_fp16))[name = tensor("op_40889_cast_fp16")]; + tensor var_40890_to_fp16 = const()[name = tensor("op_40890_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3451_cast_fp16 = mul(x = var_40889_cast_fp16, y = var_40890_to_fp16)[name = tensor("aw_3451_cast_fp16")]; + tensor var_40893_equation_0 = const()[name = tensor("op_40893_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40893_cast_fp16 = einsum(equation = var_40893_equation_0, values = (var_40735_cast_fp16, var_40652_cast_fp16))[name = tensor("op_40893_cast_fp16")]; + tensor var_40894_to_fp16 = const()[name = tensor("op_40894_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3453_cast_fp16 = mul(x = var_40893_cast_fp16, y = var_40894_to_fp16)[name = tensor("aw_3453_cast_fp16")]; + tensor var_40897_equation_0 = const()[name = tensor("op_40897_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40897_cast_fp16 = einsum(equation = var_40897_equation_0, values = (var_40739_cast_fp16, var_40656_cast_fp16))[name = tensor("op_40897_cast_fp16")]; + tensor var_40898_to_fp16 = const()[name = tensor("op_40898_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3455_cast_fp16 = mul(x = var_40897_cast_fp16, y = var_40898_to_fp16)[name = tensor("aw_3455_cast_fp16")]; + tensor var_40901_equation_0 = const()[name = tensor("op_40901_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40901_cast_fp16 = einsum(equation = var_40901_equation_0, values = (var_40743_cast_fp16, var_40660_cast_fp16))[name = tensor("op_40901_cast_fp16")]; + tensor var_40902_to_fp16 = const()[name = tensor("op_40902_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3457_cast_fp16 = mul(x = var_40901_cast_fp16, y = var_40902_to_fp16)[name = tensor("aw_3457_cast_fp16")]; + tensor var_40905_equation_0 = const()[name = tensor("op_40905_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40905_cast_fp16 = einsum(equation = var_40905_equation_0, values = (var_40747_cast_fp16, var_40664_cast_fp16))[name = tensor("op_40905_cast_fp16")]; + tensor var_40906_to_fp16 = const()[name = tensor("op_40906_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3459_cast_fp16 = mul(x = var_40905_cast_fp16, y = var_40906_to_fp16)[name = tensor("aw_3459_cast_fp16")]; + tensor var_40909_equation_0 = const()[name = tensor("op_40909_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40909_cast_fp16 = einsum(equation = var_40909_equation_0, values = (var_40751_cast_fp16, var_40668_cast_fp16))[name = tensor("op_40909_cast_fp16")]; + tensor var_40910_to_fp16 = const()[name = tensor("op_40910_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3461_cast_fp16 = mul(x = var_40909_cast_fp16, y = var_40910_to_fp16)[name = tensor("aw_3461_cast_fp16")]; + tensor var_40913_equation_0 = const()[name = tensor("op_40913_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40913_cast_fp16 = einsum(equation = var_40913_equation_0, values = (var_40755_cast_fp16, var_40672_cast_fp16))[name = tensor("op_40913_cast_fp16")]; + tensor var_40914_to_fp16 = const()[name = tensor("op_40914_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3463_cast_fp16 = mul(x = var_40913_cast_fp16, y = var_40914_to_fp16)[name = tensor("aw_3463_cast_fp16")]; + tensor var_40917_equation_0 = const()[name = tensor("op_40917_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40917_cast_fp16 = einsum(equation = var_40917_equation_0, values = (var_40759_cast_fp16, var_40676_cast_fp16))[name = tensor("op_40917_cast_fp16")]; + tensor var_40918_to_fp16 = const()[name = tensor("op_40918_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3465_cast_fp16 = mul(x = var_40917_cast_fp16, y = var_40918_to_fp16)[name = tensor("aw_3465_cast_fp16")]; + tensor var_40921_equation_0 = const()[name = tensor("op_40921_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40921_cast_fp16 = einsum(equation = var_40921_equation_0, values = (var_40763_cast_fp16, var_40680_cast_fp16))[name = tensor("op_40921_cast_fp16")]; + tensor var_40922_to_fp16 = const()[name = tensor("op_40922_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3467_cast_fp16 = mul(x = var_40921_cast_fp16, y = var_40922_to_fp16)[name = tensor("aw_3467_cast_fp16")]; + tensor var_40925_equation_0 = const()[name = tensor("op_40925_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40925_cast_fp16 = einsum(equation = var_40925_equation_0, values = (var_40767_cast_fp16, var_40684_cast_fp16))[name = tensor("op_40925_cast_fp16")]; + tensor var_40926_to_fp16 = const()[name = tensor("op_40926_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3469_cast_fp16 = mul(x = var_40925_cast_fp16, y = var_40926_to_fp16)[name = tensor("aw_3469_cast_fp16")]; + tensor var_40929_equation_0 = const()[name = tensor("op_40929_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40929_cast_fp16 = einsum(equation = var_40929_equation_0, values = (var_40771_cast_fp16, var_40688_cast_fp16))[name = tensor("op_40929_cast_fp16")]; + tensor var_40930_to_fp16 = const()[name = tensor("op_40930_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3471_cast_fp16 = mul(x = var_40929_cast_fp16, y = var_40930_to_fp16)[name = tensor("aw_3471_cast_fp16")]; + tensor var_40933_equation_0 = const()[name = tensor("op_40933_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40933_cast_fp16 = einsum(equation = var_40933_equation_0, values = (var_40775_cast_fp16, var_40692_cast_fp16))[name = tensor("op_40933_cast_fp16")]; + tensor var_40934_to_fp16 = const()[name = tensor("op_40934_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3473_cast_fp16 = mul(x = var_40933_cast_fp16, y = var_40934_to_fp16)[name = tensor("aw_3473_cast_fp16")]; + tensor var_40937_equation_0 = const()[name = tensor("op_40937_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40937_cast_fp16 = einsum(equation = var_40937_equation_0, values = (var_40779_cast_fp16, var_40696_cast_fp16))[name = tensor("op_40937_cast_fp16")]; + tensor var_40938_to_fp16 = const()[name = tensor("op_40938_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3475_cast_fp16 = mul(x = var_40937_cast_fp16, y = var_40938_to_fp16)[name = tensor("aw_3475_cast_fp16")]; + tensor var_40941_equation_0 = const()[name = tensor("op_40941_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40941_cast_fp16 = einsum(equation = var_40941_equation_0, values = (var_40783_cast_fp16, var_40700_cast_fp16))[name = tensor("op_40941_cast_fp16")]; + tensor var_40942_to_fp16 = const()[name = tensor("op_40942_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3477_cast_fp16 = mul(x = var_40941_cast_fp16, y = var_40942_to_fp16)[name = tensor("aw_3477_cast_fp16")]; + tensor var_40945_equation_0 = const()[name = tensor("op_40945_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40945_cast_fp16 = einsum(equation = var_40945_equation_0, values = (var_40787_cast_fp16, var_40704_cast_fp16))[name = tensor("op_40945_cast_fp16")]; + tensor var_40946_to_fp16 = const()[name = tensor("op_40946_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3479_cast_fp16 = mul(x = var_40945_cast_fp16, y = var_40946_to_fp16)[name = tensor("aw_3479_cast_fp16")]; + tensor var_40948_cast_fp16 = softmax(axis = var_30355, x = aw_3441_cast_fp16)[name = tensor("op_40948_cast_fp16")]; + tensor var_40949_cast_fp16 = softmax(axis = var_30355, x = aw_3443_cast_fp16)[name = tensor("op_40949_cast_fp16")]; + tensor var_40950_cast_fp16 = softmax(axis = var_30355, x = aw_3445_cast_fp16)[name = tensor("op_40950_cast_fp16")]; + tensor var_40951_cast_fp16 = softmax(axis = var_30355, x = aw_3447_cast_fp16)[name = tensor("op_40951_cast_fp16")]; + tensor var_40952_cast_fp16 = softmax(axis = var_30355, x = aw_3449_cast_fp16)[name = tensor("op_40952_cast_fp16")]; + tensor var_40953_cast_fp16 = softmax(axis = var_30355, x = aw_3451_cast_fp16)[name = tensor("op_40953_cast_fp16")]; + tensor var_40954_cast_fp16 = softmax(axis = var_30355, x = aw_3453_cast_fp16)[name = tensor("op_40954_cast_fp16")]; + tensor var_40955_cast_fp16 = softmax(axis = var_30355, x = aw_3455_cast_fp16)[name = tensor("op_40955_cast_fp16")]; + tensor var_40956_cast_fp16 = softmax(axis = var_30355, x = aw_3457_cast_fp16)[name = tensor("op_40956_cast_fp16")]; + tensor var_40957_cast_fp16 = softmax(axis = var_30355, x = aw_3459_cast_fp16)[name = tensor("op_40957_cast_fp16")]; + tensor var_40958_cast_fp16 = softmax(axis = var_30355, x = aw_3461_cast_fp16)[name = tensor("op_40958_cast_fp16")]; + tensor var_40959_cast_fp16 = softmax(axis = var_30355, x = aw_3463_cast_fp16)[name = tensor("op_40959_cast_fp16")]; + tensor var_40960_cast_fp16 = softmax(axis = var_30355, x = aw_3465_cast_fp16)[name = tensor("op_40960_cast_fp16")]; + tensor var_40961_cast_fp16 = softmax(axis = var_30355, x = aw_3467_cast_fp16)[name = tensor("op_40961_cast_fp16")]; + tensor var_40962_cast_fp16 = softmax(axis = var_30355, x = aw_3469_cast_fp16)[name = tensor("op_40962_cast_fp16")]; + tensor var_40963_cast_fp16 = softmax(axis = var_30355, x = aw_3471_cast_fp16)[name = tensor("op_40963_cast_fp16")]; + tensor var_40964_cast_fp16 = softmax(axis = var_30355, x = aw_3473_cast_fp16)[name = tensor("op_40964_cast_fp16")]; + tensor var_40965_cast_fp16 = softmax(axis = var_30355, x = aw_3475_cast_fp16)[name = tensor("op_40965_cast_fp16")]; + tensor var_40966_cast_fp16 = softmax(axis = var_30355, x = aw_3477_cast_fp16)[name = tensor("op_40966_cast_fp16")]; + tensor var_40967_cast_fp16 = softmax(axis = var_30355, x = aw_3479_cast_fp16)[name = tensor("op_40967_cast_fp16")]; + tensor var_40969_equation_0 = const()[name = tensor("op_40969_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40969_cast_fp16 = einsum(equation = var_40969_equation_0, values = (var_40789_cast_fp16, var_40948_cast_fp16))[name = tensor("op_40969_cast_fp16")]; + tensor var_40971_equation_0 = const()[name = tensor("op_40971_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40971_cast_fp16 = einsum(equation = var_40971_equation_0, values = (var_40793_cast_fp16, var_40949_cast_fp16))[name = tensor("op_40971_cast_fp16")]; + tensor var_40973_equation_0 = const()[name = tensor("op_40973_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40973_cast_fp16 = einsum(equation = var_40973_equation_0, values = (var_40797_cast_fp16, var_40950_cast_fp16))[name = tensor("op_40973_cast_fp16")]; + tensor var_40975_equation_0 = const()[name = tensor("op_40975_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40975_cast_fp16 = einsum(equation = var_40975_equation_0, values = (var_40801_cast_fp16, var_40951_cast_fp16))[name = tensor("op_40975_cast_fp16")]; + tensor var_40977_equation_0 = const()[name = tensor("op_40977_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40977_cast_fp16 = einsum(equation = var_40977_equation_0, values = (var_40805_cast_fp16, var_40952_cast_fp16))[name = tensor("op_40977_cast_fp16")]; + tensor var_40979_equation_0 = const()[name = tensor("op_40979_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40979_cast_fp16 = einsum(equation = var_40979_equation_0, values = (var_40809_cast_fp16, var_40953_cast_fp16))[name = tensor("op_40979_cast_fp16")]; + tensor var_40981_equation_0 = const()[name = tensor("op_40981_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40981_cast_fp16 = einsum(equation = var_40981_equation_0, values = (var_40813_cast_fp16, var_40954_cast_fp16))[name = tensor("op_40981_cast_fp16")]; + tensor var_40983_equation_0 = const()[name = tensor("op_40983_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40983_cast_fp16 = einsum(equation = var_40983_equation_0, values = (var_40817_cast_fp16, var_40955_cast_fp16))[name = tensor("op_40983_cast_fp16")]; + tensor var_40985_equation_0 = const()[name = tensor("op_40985_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40985_cast_fp16 = einsum(equation = var_40985_equation_0, values = (var_40821_cast_fp16, var_40956_cast_fp16))[name = tensor("op_40985_cast_fp16")]; + tensor var_40987_equation_0 = const()[name = tensor("op_40987_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40987_cast_fp16 = einsum(equation = var_40987_equation_0, values = (var_40825_cast_fp16, var_40957_cast_fp16))[name = tensor("op_40987_cast_fp16")]; + tensor var_40989_equation_0 = const()[name = tensor("op_40989_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40989_cast_fp16 = einsum(equation = var_40989_equation_0, values = (var_40829_cast_fp16, var_40958_cast_fp16))[name = tensor("op_40989_cast_fp16")]; + tensor var_40991_equation_0 = const()[name = tensor("op_40991_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40991_cast_fp16 = einsum(equation = var_40991_equation_0, values = (var_40833_cast_fp16, var_40959_cast_fp16))[name = tensor("op_40991_cast_fp16")]; + tensor var_40993_equation_0 = const()[name = tensor("op_40993_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40993_cast_fp16 = einsum(equation = var_40993_equation_0, values = (var_40837_cast_fp16, var_40960_cast_fp16))[name = tensor("op_40993_cast_fp16")]; + tensor var_40995_equation_0 = const()[name = tensor("op_40995_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40995_cast_fp16 = einsum(equation = var_40995_equation_0, values = (var_40841_cast_fp16, var_40961_cast_fp16))[name = tensor("op_40995_cast_fp16")]; + tensor var_40997_equation_0 = const()[name = tensor("op_40997_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40997_cast_fp16 = einsum(equation = var_40997_equation_0, values = (var_40845_cast_fp16, var_40962_cast_fp16))[name = tensor("op_40997_cast_fp16")]; + tensor var_40999_equation_0 = const()[name = tensor("op_40999_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40999_cast_fp16 = einsum(equation = var_40999_equation_0, values = (var_40849_cast_fp16, var_40963_cast_fp16))[name = tensor("op_40999_cast_fp16")]; + tensor var_41001_equation_0 = const()[name = tensor("op_41001_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_41001_cast_fp16 = einsum(equation = var_41001_equation_0, values = (var_40853_cast_fp16, var_40964_cast_fp16))[name = tensor("op_41001_cast_fp16")]; + tensor var_41003_equation_0 = const()[name = tensor("op_41003_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_41003_cast_fp16 = einsum(equation = var_41003_equation_0, values = (var_40857_cast_fp16, var_40965_cast_fp16))[name = tensor("op_41003_cast_fp16")]; + tensor var_41005_equation_0 = const()[name = tensor("op_41005_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_41005_cast_fp16 = einsum(equation = var_41005_equation_0, values = (var_40861_cast_fp16, var_40966_cast_fp16))[name = tensor("op_41005_cast_fp16")]; + tensor var_41007_equation_0 = const()[name = tensor("op_41007_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_41007_cast_fp16 = einsum(equation = var_41007_equation_0, values = (var_40865_cast_fp16, var_40967_cast_fp16))[name = tensor("op_41007_cast_fp16")]; + tensor input_545_interleave_0 = const()[name = tensor("input_545_interleave_0"), val = tensor(false)]; + tensor input_545_cast_fp16 = concat(axis = var_30355, interleave = input_545_interleave_0, values = (var_40969_cast_fp16, var_40971_cast_fp16, var_40973_cast_fp16, var_40975_cast_fp16, var_40977_cast_fp16, var_40979_cast_fp16, var_40981_cast_fp16, var_40983_cast_fp16, var_40985_cast_fp16, var_40987_cast_fp16, var_40989_cast_fp16, var_40991_cast_fp16, var_40993_cast_fp16, var_40995_cast_fp16, var_40997_cast_fp16, var_40999_cast_fp16, var_41001_cast_fp16, var_41003_cast_fp16, var_41005_cast_fp16, var_41007_cast_fp16))[name = tensor("input_545_cast_fp16")]; + tensor var_41013 = const()[name = tensor("op_41013"), val = tensor([1, 1])]; + tensor var_41015 = const()[name = tensor("op_41015"), val = tensor([1, 1])]; + tensor var_41017_pad_type_0 = const()[name = tensor("op_41017_pad_type_0"), val = tensor("custom")]; + tensor var_41017_pad_0 = const()[name = tensor("op_41017_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_1_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_1_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3467794368)))]; + tensor up_blocks_0_attentions_1_transformer_blocks_1_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_1_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3471071232)))]; + tensor var_41017_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_1_attn1_to_out_0_bias_to_fp16, dilations = var_41015, groups = var_30355, pad = var_41017_pad_0, pad_type = var_41017_pad_type_0, strides = var_41013, weight = up_blocks_0_attentions_1_transformer_blocks_1_attn1_to_out_0_weight_to_fp16, x = input_545_cast_fp16)[name = tensor("op_41017_cast_fp16")]; + tensor inputs_273_cast_fp16 = add(x = var_41017_cast_fp16, y = inputs_271_cast_fp16)[name = tensor("inputs_273_cast_fp16")]; + tensor hidden_states_369_axes_0 = const()[name = tensor("hidden_states_369_axes_0"), val = tensor([1])]; + tensor hidden_states_369_gamma_0_to_fp16 = const()[name = tensor("hidden_states_369_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3471073856)))]; + tensor hidden_states_369_beta_0_to_fp16 = const()[name = tensor("hidden_states_369_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3471076480)))]; + tensor var_41027_to_fp16 = const()[name = tensor("op_41027_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_369_cast_fp16 = layer_norm(axes = hidden_states_369_axes_0, beta = hidden_states_369_beta_0_to_fp16, epsilon = var_41027_to_fp16, gamma = hidden_states_369_gamma_0_to_fp16, x = inputs_273_cast_fp16)[name = tensor("hidden_states_369_cast_fp16")]; + tensor var_41042 = const()[name = tensor("op_41042"), val = tensor([1, 1])]; + tensor var_41044 = const()[name = tensor("op_41044"), val = tensor([1, 1])]; + tensor q_183_pad_type_0 = const()[name = tensor("q_183_pad_type_0"), val = tensor("custom")]; + tensor q_183_pad_0 = const()[name = tensor("q_183_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_1_attn2_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_1_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3471079104)))]; + tensor q_183_cast_fp16 = conv(dilations = var_41044, groups = var_30355, pad = q_183_pad_0, pad_type = q_183_pad_type_0, strides = var_41042, weight = up_blocks_0_attentions_1_transformer_blocks_1_attn2_to_q_weight_to_fp16, x = hidden_states_369_cast_fp16)[name = tensor("q_183_cast_fp16")]; + tensor var_41048 = const()[name = tensor("op_41048"), val = tensor([1, 1])]; + tensor var_41050 = const()[name = tensor("op_41050"), val = tensor([1, 1])]; + tensor k_365_pad_type_0 = const()[name = tensor("k_365_pad_type_0"), val = tensor("custom")]; + tensor k_365_pad_0 = const()[name = tensor("k_365_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_1_attn2_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_1_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3474355968)))]; + tensor k_365_cast_fp16 = conv(dilations = var_41050, groups = var_30355, pad = k_365_pad_0, pad_type = k_365_pad_type_0, strides = var_41048, weight = up_blocks_0_attentions_1_transformer_blocks_1_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_365_cast_fp16")]; + tensor var_41054 = const()[name = tensor("op_41054"), val = tensor([1, 1])]; + tensor var_41056 = const()[name = tensor("op_41056"), val = tensor([1, 1])]; + tensor v_183_pad_type_0 = const()[name = tensor("v_183_pad_type_0"), val = tensor("custom")]; + tensor v_183_pad_0 = const()[name = tensor("v_183_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_1_attn2_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_1_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3479598912)))]; + tensor v_183_cast_fp16 = conv(dilations = var_41056, groups = var_30355, pad = v_183_pad_0, pad_type = v_183_pad_type_0, strides = var_41054, weight = up_blocks_0_attentions_1_transformer_blocks_1_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_183_cast_fp16")]; + tensor var_41060_begin_0 = const()[name = tensor("op_41060_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_41060_end_0 = const()[name = tensor("op_41060_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_41060_end_mask_0 = const()[name = tensor("op_41060_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41060_cast_fp16 = slice_by_index(begin = var_41060_begin_0, end = var_41060_end_0, end_mask = var_41060_end_mask_0, x = q_183_cast_fp16)[name = tensor("op_41060_cast_fp16")]; + tensor var_41064_begin_0 = const()[name = tensor("op_41064_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_41064_end_0 = const()[name = tensor("op_41064_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_41064_end_mask_0 = const()[name = tensor("op_41064_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41064_cast_fp16 = slice_by_index(begin = var_41064_begin_0, end = var_41064_end_0, end_mask = var_41064_end_mask_0, x = q_183_cast_fp16)[name = tensor("op_41064_cast_fp16")]; + tensor var_41068_begin_0 = const()[name = tensor("op_41068_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_41068_end_0 = const()[name = tensor("op_41068_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_41068_end_mask_0 = const()[name = tensor("op_41068_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41068_cast_fp16 = slice_by_index(begin = var_41068_begin_0, end = var_41068_end_0, end_mask = var_41068_end_mask_0, x = q_183_cast_fp16)[name = tensor("op_41068_cast_fp16")]; + tensor var_41072_begin_0 = const()[name = tensor("op_41072_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_41072_end_0 = const()[name = tensor("op_41072_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_41072_end_mask_0 = const()[name = tensor("op_41072_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41072_cast_fp16 = slice_by_index(begin = var_41072_begin_0, end = var_41072_end_0, end_mask = var_41072_end_mask_0, x = q_183_cast_fp16)[name = tensor("op_41072_cast_fp16")]; + tensor var_41076_begin_0 = const()[name = tensor("op_41076_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_41076_end_0 = const()[name = tensor("op_41076_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_41076_end_mask_0 = const()[name = tensor("op_41076_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41076_cast_fp16 = slice_by_index(begin = var_41076_begin_0, end = var_41076_end_0, end_mask = var_41076_end_mask_0, x = q_183_cast_fp16)[name = tensor("op_41076_cast_fp16")]; + tensor var_41080_begin_0 = const()[name = tensor("op_41080_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_41080_end_0 = const()[name = tensor("op_41080_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_41080_end_mask_0 = const()[name = tensor("op_41080_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41080_cast_fp16 = slice_by_index(begin = var_41080_begin_0, end = var_41080_end_0, end_mask = var_41080_end_mask_0, x = q_183_cast_fp16)[name = tensor("op_41080_cast_fp16")]; + tensor var_41084_begin_0 = const()[name = tensor("op_41084_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_41084_end_0 = const()[name = tensor("op_41084_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_41084_end_mask_0 = const()[name = tensor("op_41084_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41084_cast_fp16 = slice_by_index(begin = var_41084_begin_0, end = var_41084_end_0, end_mask = var_41084_end_mask_0, x = q_183_cast_fp16)[name = tensor("op_41084_cast_fp16")]; + tensor var_41088_begin_0 = const()[name = tensor("op_41088_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_41088_end_0 = const()[name = tensor("op_41088_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_41088_end_mask_0 = const()[name = tensor("op_41088_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41088_cast_fp16 = slice_by_index(begin = var_41088_begin_0, end = var_41088_end_0, end_mask = var_41088_end_mask_0, x = q_183_cast_fp16)[name = tensor("op_41088_cast_fp16")]; + tensor var_41092_begin_0 = const()[name = tensor("op_41092_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_41092_end_0 = const()[name = tensor("op_41092_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_41092_end_mask_0 = const()[name = tensor("op_41092_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41092_cast_fp16 = slice_by_index(begin = var_41092_begin_0, end = var_41092_end_0, end_mask = var_41092_end_mask_0, x = q_183_cast_fp16)[name = tensor("op_41092_cast_fp16")]; + tensor var_41096_begin_0 = const()[name = tensor("op_41096_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_41096_end_0 = const()[name = tensor("op_41096_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_41096_end_mask_0 = const()[name = tensor("op_41096_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41096_cast_fp16 = slice_by_index(begin = var_41096_begin_0, end = var_41096_end_0, end_mask = var_41096_end_mask_0, x = q_183_cast_fp16)[name = tensor("op_41096_cast_fp16")]; + tensor var_41100_begin_0 = const()[name = tensor("op_41100_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_41100_end_0 = const()[name = tensor("op_41100_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_41100_end_mask_0 = const()[name = tensor("op_41100_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41100_cast_fp16 = slice_by_index(begin = var_41100_begin_0, end = var_41100_end_0, end_mask = var_41100_end_mask_0, x = q_183_cast_fp16)[name = tensor("op_41100_cast_fp16")]; + tensor var_41104_begin_0 = const()[name = tensor("op_41104_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_41104_end_0 = const()[name = tensor("op_41104_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_41104_end_mask_0 = const()[name = tensor("op_41104_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41104_cast_fp16 = slice_by_index(begin = var_41104_begin_0, end = var_41104_end_0, end_mask = var_41104_end_mask_0, x = q_183_cast_fp16)[name = tensor("op_41104_cast_fp16")]; + tensor var_41108_begin_0 = const()[name = tensor("op_41108_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_41108_end_0 = const()[name = tensor("op_41108_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_41108_end_mask_0 = const()[name = tensor("op_41108_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41108_cast_fp16 = slice_by_index(begin = var_41108_begin_0, end = var_41108_end_0, end_mask = var_41108_end_mask_0, x = q_183_cast_fp16)[name = tensor("op_41108_cast_fp16")]; + tensor var_41112_begin_0 = const()[name = tensor("op_41112_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_41112_end_0 = const()[name = tensor("op_41112_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_41112_end_mask_0 = const()[name = tensor("op_41112_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41112_cast_fp16 = slice_by_index(begin = var_41112_begin_0, end = var_41112_end_0, end_mask = var_41112_end_mask_0, x = q_183_cast_fp16)[name = tensor("op_41112_cast_fp16")]; + tensor var_41116_begin_0 = const()[name = tensor("op_41116_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_41116_end_0 = const()[name = tensor("op_41116_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_41116_end_mask_0 = const()[name = tensor("op_41116_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41116_cast_fp16 = slice_by_index(begin = var_41116_begin_0, end = var_41116_end_0, end_mask = var_41116_end_mask_0, x = q_183_cast_fp16)[name = tensor("op_41116_cast_fp16")]; + tensor var_41120_begin_0 = const()[name = tensor("op_41120_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_41120_end_0 = const()[name = tensor("op_41120_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_41120_end_mask_0 = const()[name = tensor("op_41120_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41120_cast_fp16 = slice_by_index(begin = var_41120_begin_0, end = var_41120_end_0, end_mask = var_41120_end_mask_0, x = q_183_cast_fp16)[name = tensor("op_41120_cast_fp16")]; + tensor var_41124_begin_0 = const()[name = tensor("op_41124_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_41124_end_0 = const()[name = tensor("op_41124_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_41124_end_mask_0 = const()[name = tensor("op_41124_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41124_cast_fp16 = slice_by_index(begin = var_41124_begin_0, end = var_41124_end_0, end_mask = var_41124_end_mask_0, x = q_183_cast_fp16)[name = tensor("op_41124_cast_fp16")]; + tensor var_41128_begin_0 = const()[name = tensor("op_41128_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_41128_end_0 = const()[name = tensor("op_41128_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_41128_end_mask_0 = const()[name = tensor("op_41128_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41128_cast_fp16 = slice_by_index(begin = var_41128_begin_0, end = var_41128_end_0, end_mask = var_41128_end_mask_0, x = q_183_cast_fp16)[name = tensor("op_41128_cast_fp16")]; + tensor var_41132_begin_0 = const()[name = tensor("op_41132_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_41132_end_0 = const()[name = tensor("op_41132_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_41132_end_mask_0 = const()[name = tensor("op_41132_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41132_cast_fp16 = slice_by_index(begin = var_41132_begin_0, end = var_41132_end_0, end_mask = var_41132_end_mask_0, x = q_183_cast_fp16)[name = tensor("op_41132_cast_fp16")]; + tensor var_41136_begin_0 = const()[name = tensor("op_41136_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_41136_end_0 = const()[name = tensor("op_41136_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_41136_end_mask_0 = const()[name = tensor("op_41136_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41136_cast_fp16 = slice_by_index(begin = var_41136_begin_0, end = var_41136_end_0, end_mask = var_41136_end_mask_0, x = q_183_cast_fp16)[name = tensor("op_41136_cast_fp16")]; + tensor k_367_perm_0 = const()[name = tensor("k_367_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_41143_begin_0 = const()[name = tensor("op_41143_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_41143_end_0 = const()[name = tensor("op_41143_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_41143_end_mask_0 = const()[name = tensor("op_41143_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_367_cast_fp16 = transpose(perm = k_367_perm_0, x = k_365_cast_fp16)[name = tensor("transpose_48")]; + tensor var_41143_cast_fp16 = slice_by_index(begin = var_41143_begin_0, end = var_41143_end_0, end_mask = var_41143_end_mask_0, x = k_367_cast_fp16)[name = tensor("op_41143_cast_fp16")]; + tensor var_41147_begin_0 = const()[name = tensor("op_41147_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_41147_end_0 = const()[name = tensor("op_41147_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_41147_end_mask_0 = const()[name = tensor("op_41147_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41147_cast_fp16 = slice_by_index(begin = var_41147_begin_0, end = var_41147_end_0, end_mask = var_41147_end_mask_0, x = k_367_cast_fp16)[name = tensor("op_41147_cast_fp16")]; + tensor var_41151_begin_0 = const()[name = tensor("op_41151_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_41151_end_0 = const()[name = tensor("op_41151_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_41151_end_mask_0 = const()[name = tensor("op_41151_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41151_cast_fp16 = slice_by_index(begin = var_41151_begin_0, end = var_41151_end_0, end_mask = var_41151_end_mask_0, x = k_367_cast_fp16)[name = tensor("op_41151_cast_fp16")]; + tensor var_41155_begin_0 = const()[name = tensor("op_41155_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_41155_end_0 = const()[name = tensor("op_41155_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_41155_end_mask_0 = const()[name = tensor("op_41155_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41155_cast_fp16 = slice_by_index(begin = var_41155_begin_0, end = var_41155_end_0, end_mask = var_41155_end_mask_0, x = k_367_cast_fp16)[name = tensor("op_41155_cast_fp16")]; + tensor var_41159_begin_0 = const()[name = tensor("op_41159_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_41159_end_0 = const()[name = tensor("op_41159_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_41159_end_mask_0 = const()[name = tensor("op_41159_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41159_cast_fp16 = slice_by_index(begin = var_41159_begin_0, end = var_41159_end_0, end_mask = var_41159_end_mask_0, x = k_367_cast_fp16)[name = tensor("op_41159_cast_fp16")]; + tensor var_41163_begin_0 = const()[name = tensor("op_41163_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_41163_end_0 = const()[name = tensor("op_41163_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_41163_end_mask_0 = const()[name = tensor("op_41163_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41163_cast_fp16 = slice_by_index(begin = var_41163_begin_0, end = var_41163_end_0, end_mask = var_41163_end_mask_0, x = k_367_cast_fp16)[name = tensor("op_41163_cast_fp16")]; + tensor var_41167_begin_0 = const()[name = tensor("op_41167_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_41167_end_0 = const()[name = tensor("op_41167_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_41167_end_mask_0 = const()[name = tensor("op_41167_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41167_cast_fp16 = slice_by_index(begin = var_41167_begin_0, end = var_41167_end_0, end_mask = var_41167_end_mask_0, x = k_367_cast_fp16)[name = tensor("op_41167_cast_fp16")]; + tensor var_41171_begin_0 = const()[name = tensor("op_41171_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_41171_end_0 = const()[name = tensor("op_41171_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_41171_end_mask_0 = const()[name = tensor("op_41171_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41171_cast_fp16 = slice_by_index(begin = var_41171_begin_0, end = var_41171_end_0, end_mask = var_41171_end_mask_0, x = k_367_cast_fp16)[name = tensor("op_41171_cast_fp16")]; + tensor var_41175_begin_0 = const()[name = tensor("op_41175_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_41175_end_0 = const()[name = tensor("op_41175_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_41175_end_mask_0 = const()[name = tensor("op_41175_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41175_cast_fp16 = slice_by_index(begin = var_41175_begin_0, end = var_41175_end_0, end_mask = var_41175_end_mask_0, x = k_367_cast_fp16)[name = tensor("op_41175_cast_fp16")]; + tensor var_41179_begin_0 = const()[name = tensor("op_41179_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_41179_end_0 = const()[name = tensor("op_41179_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_41179_end_mask_0 = const()[name = tensor("op_41179_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41179_cast_fp16 = slice_by_index(begin = var_41179_begin_0, end = var_41179_end_0, end_mask = var_41179_end_mask_0, x = k_367_cast_fp16)[name = tensor("op_41179_cast_fp16")]; + tensor var_41183_begin_0 = const()[name = tensor("op_41183_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_41183_end_0 = const()[name = tensor("op_41183_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_41183_end_mask_0 = const()[name = tensor("op_41183_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41183_cast_fp16 = slice_by_index(begin = var_41183_begin_0, end = var_41183_end_0, end_mask = var_41183_end_mask_0, x = k_367_cast_fp16)[name = tensor("op_41183_cast_fp16")]; + tensor var_41187_begin_0 = const()[name = tensor("op_41187_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_41187_end_0 = const()[name = tensor("op_41187_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_41187_end_mask_0 = const()[name = tensor("op_41187_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41187_cast_fp16 = slice_by_index(begin = var_41187_begin_0, end = var_41187_end_0, end_mask = var_41187_end_mask_0, x = k_367_cast_fp16)[name = tensor("op_41187_cast_fp16")]; + tensor var_41191_begin_0 = const()[name = tensor("op_41191_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_41191_end_0 = const()[name = tensor("op_41191_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_41191_end_mask_0 = const()[name = tensor("op_41191_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41191_cast_fp16 = slice_by_index(begin = var_41191_begin_0, end = var_41191_end_0, end_mask = var_41191_end_mask_0, x = k_367_cast_fp16)[name = tensor("op_41191_cast_fp16")]; + tensor var_41195_begin_0 = const()[name = tensor("op_41195_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_41195_end_0 = const()[name = tensor("op_41195_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_41195_end_mask_0 = const()[name = tensor("op_41195_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41195_cast_fp16 = slice_by_index(begin = var_41195_begin_0, end = var_41195_end_0, end_mask = var_41195_end_mask_0, x = k_367_cast_fp16)[name = tensor("op_41195_cast_fp16")]; + tensor var_41199_begin_0 = const()[name = tensor("op_41199_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_41199_end_0 = const()[name = tensor("op_41199_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_41199_end_mask_0 = const()[name = tensor("op_41199_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41199_cast_fp16 = slice_by_index(begin = var_41199_begin_0, end = var_41199_end_0, end_mask = var_41199_end_mask_0, x = k_367_cast_fp16)[name = tensor("op_41199_cast_fp16")]; + tensor var_41203_begin_0 = const()[name = tensor("op_41203_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_41203_end_0 = const()[name = tensor("op_41203_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_41203_end_mask_0 = const()[name = tensor("op_41203_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41203_cast_fp16 = slice_by_index(begin = var_41203_begin_0, end = var_41203_end_0, end_mask = var_41203_end_mask_0, x = k_367_cast_fp16)[name = tensor("op_41203_cast_fp16")]; + tensor var_41207_begin_0 = const()[name = tensor("op_41207_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_41207_end_0 = const()[name = tensor("op_41207_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_41207_end_mask_0 = const()[name = tensor("op_41207_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41207_cast_fp16 = slice_by_index(begin = var_41207_begin_0, end = var_41207_end_0, end_mask = var_41207_end_mask_0, x = k_367_cast_fp16)[name = tensor("op_41207_cast_fp16")]; + tensor var_41211_begin_0 = const()[name = tensor("op_41211_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_41211_end_0 = const()[name = tensor("op_41211_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_41211_end_mask_0 = const()[name = tensor("op_41211_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41211_cast_fp16 = slice_by_index(begin = var_41211_begin_0, end = var_41211_end_0, end_mask = var_41211_end_mask_0, x = k_367_cast_fp16)[name = tensor("op_41211_cast_fp16")]; + tensor var_41215_begin_0 = const()[name = tensor("op_41215_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_41215_end_0 = const()[name = tensor("op_41215_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_41215_end_mask_0 = const()[name = tensor("op_41215_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41215_cast_fp16 = slice_by_index(begin = var_41215_begin_0, end = var_41215_end_0, end_mask = var_41215_end_mask_0, x = k_367_cast_fp16)[name = tensor("op_41215_cast_fp16")]; + tensor var_41219_begin_0 = const()[name = tensor("op_41219_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_41219_end_0 = const()[name = tensor("op_41219_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_41219_end_mask_0 = const()[name = tensor("op_41219_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41219_cast_fp16 = slice_by_index(begin = var_41219_begin_0, end = var_41219_end_0, end_mask = var_41219_end_mask_0, x = k_367_cast_fp16)[name = tensor("op_41219_cast_fp16")]; + tensor var_41221_begin_0 = const()[name = tensor("op_41221_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_41221_end_0 = const()[name = tensor("op_41221_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_41221_end_mask_0 = const()[name = tensor("op_41221_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41221_cast_fp16 = slice_by_index(begin = var_41221_begin_0, end = var_41221_end_0, end_mask = var_41221_end_mask_0, x = v_183_cast_fp16)[name = tensor("op_41221_cast_fp16")]; + tensor var_41225_begin_0 = const()[name = tensor("op_41225_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_41225_end_0 = const()[name = tensor("op_41225_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_41225_end_mask_0 = const()[name = tensor("op_41225_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41225_cast_fp16 = slice_by_index(begin = var_41225_begin_0, end = var_41225_end_0, end_mask = var_41225_end_mask_0, x = v_183_cast_fp16)[name = tensor("op_41225_cast_fp16")]; + tensor var_41229_begin_0 = const()[name = tensor("op_41229_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_41229_end_0 = const()[name = tensor("op_41229_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_41229_end_mask_0 = const()[name = tensor("op_41229_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41229_cast_fp16 = slice_by_index(begin = var_41229_begin_0, end = var_41229_end_0, end_mask = var_41229_end_mask_0, x = v_183_cast_fp16)[name = tensor("op_41229_cast_fp16")]; + tensor var_41233_begin_0 = const()[name = tensor("op_41233_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_41233_end_0 = const()[name = tensor("op_41233_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_41233_end_mask_0 = const()[name = tensor("op_41233_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41233_cast_fp16 = slice_by_index(begin = var_41233_begin_0, end = var_41233_end_0, end_mask = var_41233_end_mask_0, x = v_183_cast_fp16)[name = tensor("op_41233_cast_fp16")]; + tensor var_41237_begin_0 = const()[name = tensor("op_41237_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_41237_end_0 = const()[name = tensor("op_41237_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_41237_end_mask_0 = const()[name = tensor("op_41237_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41237_cast_fp16 = slice_by_index(begin = var_41237_begin_0, end = var_41237_end_0, end_mask = var_41237_end_mask_0, x = v_183_cast_fp16)[name = tensor("op_41237_cast_fp16")]; + tensor var_41241_begin_0 = const()[name = tensor("op_41241_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_41241_end_0 = const()[name = tensor("op_41241_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_41241_end_mask_0 = const()[name = tensor("op_41241_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41241_cast_fp16 = slice_by_index(begin = var_41241_begin_0, end = var_41241_end_0, end_mask = var_41241_end_mask_0, x = v_183_cast_fp16)[name = tensor("op_41241_cast_fp16")]; + tensor var_41245_begin_0 = const()[name = tensor("op_41245_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_41245_end_0 = const()[name = tensor("op_41245_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_41245_end_mask_0 = const()[name = tensor("op_41245_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41245_cast_fp16 = slice_by_index(begin = var_41245_begin_0, end = var_41245_end_0, end_mask = var_41245_end_mask_0, x = v_183_cast_fp16)[name = tensor("op_41245_cast_fp16")]; + tensor var_41249_begin_0 = const()[name = tensor("op_41249_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_41249_end_0 = const()[name = tensor("op_41249_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_41249_end_mask_0 = const()[name = tensor("op_41249_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41249_cast_fp16 = slice_by_index(begin = var_41249_begin_0, end = var_41249_end_0, end_mask = var_41249_end_mask_0, x = v_183_cast_fp16)[name = tensor("op_41249_cast_fp16")]; + tensor var_41253_begin_0 = const()[name = tensor("op_41253_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_41253_end_0 = const()[name = tensor("op_41253_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_41253_end_mask_0 = const()[name = tensor("op_41253_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41253_cast_fp16 = slice_by_index(begin = var_41253_begin_0, end = var_41253_end_0, end_mask = var_41253_end_mask_0, x = v_183_cast_fp16)[name = tensor("op_41253_cast_fp16")]; + tensor var_41257_begin_0 = const()[name = tensor("op_41257_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_41257_end_0 = const()[name = tensor("op_41257_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_41257_end_mask_0 = const()[name = tensor("op_41257_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41257_cast_fp16 = slice_by_index(begin = var_41257_begin_0, end = var_41257_end_0, end_mask = var_41257_end_mask_0, x = v_183_cast_fp16)[name = tensor("op_41257_cast_fp16")]; + tensor var_41261_begin_0 = const()[name = tensor("op_41261_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_41261_end_0 = const()[name = tensor("op_41261_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_41261_end_mask_0 = const()[name = tensor("op_41261_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41261_cast_fp16 = slice_by_index(begin = var_41261_begin_0, end = var_41261_end_0, end_mask = var_41261_end_mask_0, x = v_183_cast_fp16)[name = tensor("op_41261_cast_fp16")]; + tensor var_41265_begin_0 = const()[name = tensor("op_41265_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_41265_end_0 = const()[name = tensor("op_41265_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_41265_end_mask_0 = const()[name = tensor("op_41265_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41265_cast_fp16 = slice_by_index(begin = var_41265_begin_0, end = var_41265_end_0, end_mask = var_41265_end_mask_0, x = v_183_cast_fp16)[name = tensor("op_41265_cast_fp16")]; + tensor var_41269_begin_0 = const()[name = tensor("op_41269_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_41269_end_0 = const()[name = tensor("op_41269_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_41269_end_mask_0 = const()[name = tensor("op_41269_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41269_cast_fp16 = slice_by_index(begin = var_41269_begin_0, end = var_41269_end_0, end_mask = var_41269_end_mask_0, x = v_183_cast_fp16)[name = tensor("op_41269_cast_fp16")]; + tensor var_41273_begin_0 = const()[name = tensor("op_41273_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_41273_end_0 = const()[name = tensor("op_41273_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_41273_end_mask_0 = const()[name = tensor("op_41273_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41273_cast_fp16 = slice_by_index(begin = var_41273_begin_0, end = var_41273_end_0, end_mask = var_41273_end_mask_0, x = v_183_cast_fp16)[name = tensor("op_41273_cast_fp16")]; + tensor var_41277_begin_0 = const()[name = tensor("op_41277_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_41277_end_0 = const()[name = tensor("op_41277_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_41277_end_mask_0 = const()[name = tensor("op_41277_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41277_cast_fp16 = slice_by_index(begin = var_41277_begin_0, end = var_41277_end_0, end_mask = var_41277_end_mask_0, x = v_183_cast_fp16)[name = tensor("op_41277_cast_fp16")]; + tensor var_41281_begin_0 = const()[name = tensor("op_41281_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_41281_end_0 = const()[name = tensor("op_41281_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_41281_end_mask_0 = const()[name = tensor("op_41281_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41281_cast_fp16 = slice_by_index(begin = var_41281_begin_0, end = var_41281_end_0, end_mask = var_41281_end_mask_0, x = v_183_cast_fp16)[name = tensor("op_41281_cast_fp16")]; + tensor var_41285_begin_0 = const()[name = tensor("op_41285_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_41285_end_0 = const()[name = tensor("op_41285_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_41285_end_mask_0 = const()[name = tensor("op_41285_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41285_cast_fp16 = slice_by_index(begin = var_41285_begin_0, end = var_41285_end_0, end_mask = var_41285_end_mask_0, x = v_183_cast_fp16)[name = tensor("op_41285_cast_fp16")]; + tensor var_41289_begin_0 = const()[name = tensor("op_41289_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_41289_end_0 = const()[name = tensor("op_41289_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_41289_end_mask_0 = const()[name = tensor("op_41289_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41289_cast_fp16 = slice_by_index(begin = var_41289_begin_0, end = var_41289_end_0, end_mask = var_41289_end_mask_0, x = v_183_cast_fp16)[name = tensor("op_41289_cast_fp16")]; + tensor var_41293_begin_0 = const()[name = tensor("op_41293_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_41293_end_0 = const()[name = tensor("op_41293_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_41293_end_mask_0 = const()[name = tensor("op_41293_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41293_cast_fp16 = slice_by_index(begin = var_41293_begin_0, end = var_41293_end_0, end_mask = var_41293_end_mask_0, x = v_183_cast_fp16)[name = tensor("op_41293_cast_fp16")]; + tensor var_41297_begin_0 = const()[name = tensor("op_41297_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_41297_end_0 = const()[name = tensor("op_41297_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_41297_end_mask_0 = const()[name = tensor("op_41297_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41297_cast_fp16 = slice_by_index(begin = var_41297_begin_0, end = var_41297_end_0, end_mask = var_41297_end_mask_0, x = v_183_cast_fp16)[name = tensor("op_41297_cast_fp16")]; + tensor var_41301_equation_0 = const()[name = tensor("op_41301_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_41301_cast_fp16 = einsum(equation = var_41301_equation_0, values = (var_41143_cast_fp16, var_41060_cast_fp16))[name = tensor("op_41301_cast_fp16")]; + tensor var_41302_to_fp16 = const()[name = tensor("op_41302_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3481_cast_fp16 = mul(x = var_41301_cast_fp16, y = var_41302_to_fp16)[name = tensor("aw_3481_cast_fp16")]; + tensor var_41305_equation_0 = const()[name = tensor("op_41305_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_41305_cast_fp16 = einsum(equation = var_41305_equation_0, values = (var_41147_cast_fp16, var_41064_cast_fp16))[name = tensor("op_41305_cast_fp16")]; + tensor var_41306_to_fp16 = const()[name = tensor("op_41306_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3483_cast_fp16 = mul(x = var_41305_cast_fp16, y = var_41306_to_fp16)[name = tensor("aw_3483_cast_fp16")]; + tensor var_41309_equation_0 = const()[name = tensor("op_41309_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_41309_cast_fp16 = einsum(equation = var_41309_equation_0, values = (var_41151_cast_fp16, var_41068_cast_fp16))[name = tensor("op_41309_cast_fp16")]; + tensor var_41310_to_fp16 = const()[name = tensor("op_41310_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3485_cast_fp16 = mul(x = var_41309_cast_fp16, y = var_41310_to_fp16)[name = tensor("aw_3485_cast_fp16")]; + tensor var_41313_equation_0 = const()[name = tensor("op_41313_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_41313_cast_fp16 = einsum(equation = var_41313_equation_0, values = (var_41155_cast_fp16, var_41072_cast_fp16))[name = tensor("op_41313_cast_fp16")]; + tensor var_41314_to_fp16 = const()[name = tensor("op_41314_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3487_cast_fp16 = mul(x = var_41313_cast_fp16, y = var_41314_to_fp16)[name = tensor("aw_3487_cast_fp16")]; + tensor var_41317_equation_0 = const()[name = tensor("op_41317_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_41317_cast_fp16 = einsum(equation = var_41317_equation_0, values = (var_41159_cast_fp16, var_41076_cast_fp16))[name = tensor("op_41317_cast_fp16")]; + tensor var_41318_to_fp16 = const()[name = tensor("op_41318_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3489_cast_fp16 = mul(x = var_41317_cast_fp16, y = var_41318_to_fp16)[name = tensor("aw_3489_cast_fp16")]; + tensor var_41321_equation_0 = const()[name = tensor("op_41321_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_41321_cast_fp16 = einsum(equation = var_41321_equation_0, values = (var_41163_cast_fp16, var_41080_cast_fp16))[name = tensor("op_41321_cast_fp16")]; + tensor var_41322_to_fp16 = const()[name = tensor("op_41322_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3491_cast_fp16 = mul(x = var_41321_cast_fp16, y = var_41322_to_fp16)[name = tensor("aw_3491_cast_fp16")]; + tensor var_41325_equation_0 = const()[name = tensor("op_41325_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_41325_cast_fp16 = einsum(equation = var_41325_equation_0, values = (var_41167_cast_fp16, var_41084_cast_fp16))[name = tensor("op_41325_cast_fp16")]; + tensor var_41326_to_fp16 = const()[name = tensor("op_41326_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3493_cast_fp16 = mul(x = var_41325_cast_fp16, y = var_41326_to_fp16)[name = tensor("aw_3493_cast_fp16")]; + tensor var_41329_equation_0 = const()[name = tensor("op_41329_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_41329_cast_fp16 = einsum(equation = var_41329_equation_0, values = (var_41171_cast_fp16, var_41088_cast_fp16))[name = tensor("op_41329_cast_fp16")]; + tensor var_41330_to_fp16 = const()[name = tensor("op_41330_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3495_cast_fp16 = mul(x = var_41329_cast_fp16, y = var_41330_to_fp16)[name = tensor("aw_3495_cast_fp16")]; + tensor var_41333_equation_0 = const()[name = tensor("op_41333_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_41333_cast_fp16 = einsum(equation = var_41333_equation_0, values = (var_41175_cast_fp16, var_41092_cast_fp16))[name = tensor("op_41333_cast_fp16")]; + tensor var_41334_to_fp16 = const()[name = tensor("op_41334_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3497_cast_fp16 = mul(x = var_41333_cast_fp16, y = var_41334_to_fp16)[name = tensor("aw_3497_cast_fp16")]; + tensor var_41337_equation_0 = const()[name = tensor("op_41337_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_41337_cast_fp16 = einsum(equation = var_41337_equation_0, values = (var_41179_cast_fp16, var_41096_cast_fp16))[name = tensor("op_41337_cast_fp16")]; + tensor var_41338_to_fp16 = const()[name = tensor("op_41338_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3499_cast_fp16 = mul(x = var_41337_cast_fp16, y = var_41338_to_fp16)[name = tensor("aw_3499_cast_fp16")]; + tensor var_41341_equation_0 = const()[name = tensor("op_41341_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_41341_cast_fp16 = einsum(equation = var_41341_equation_0, values = (var_41183_cast_fp16, var_41100_cast_fp16))[name = tensor("op_41341_cast_fp16")]; + tensor var_41342_to_fp16 = const()[name = tensor("op_41342_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3501_cast_fp16 = mul(x = var_41341_cast_fp16, y = var_41342_to_fp16)[name = tensor("aw_3501_cast_fp16")]; + tensor var_41345_equation_0 = const()[name = tensor("op_41345_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_41345_cast_fp16 = einsum(equation = var_41345_equation_0, values = (var_41187_cast_fp16, var_41104_cast_fp16))[name = tensor("op_41345_cast_fp16")]; + tensor var_41346_to_fp16 = const()[name = tensor("op_41346_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3503_cast_fp16 = mul(x = var_41345_cast_fp16, y = var_41346_to_fp16)[name = tensor("aw_3503_cast_fp16")]; + tensor var_41349_equation_0 = const()[name = tensor("op_41349_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_41349_cast_fp16 = einsum(equation = var_41349_equation_0, values = (var_41191_cast_fp16, var_41108_cast_fp16))[name = tensor("op_41349_cast_fp16")]; + tensor var_41350_to_fp16 = const()[name = tensor("op_41350_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3505_cast_fp16 = mul(x = var_41349_cast_fp16, y = var_41350_to_fp16)[name = tensor("aw_3505_cast_fp16")]; + tensor var_41353_equation_0 = const()[name = tensor("op_41353_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_41353_cast_fp16 = einsum(equation = var_41353_equation_0, values = (var_41195_cast_fp16, var_41112_cast_fp16))[name = tensor("op_41353_cast_fp16")]; + tensor var_41354_to_fp16 = const()[name = tensor("op_41354_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3507_cast_fp16 = mul(x = var_41353_cast_fp16, y = var_41354_to_fp16)[name = tensor("aw_3507_cast_fp16")]; + tensor var_41357_equation_0 = const()[name = tensor("op_41357_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_41357_cast_fp16 = einsum(equation = var_41357_equation_0, values = (var_41199_cast_fp16, var_41116_cast_fp16))[name = tensor("op_41357_cast_fp16")]; + tensor var_41358_to_fp16 = const()[name = tensor("op_41358_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3509_cast_fp16 = mul(x = var_41357_cast_fp16, y = var_41358_to_fp16)[name = tensor("aw_3509_cast_fp16")]; + tensor var_41361_equation_0 = const()[name = tensor("op_41361_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_41361_cast_fp16 = einsum(equation = var_41361_equation_0, values = (var_41203_cast_fp16, var_41120_cast_fp16))[name = tensor("op_41361_cast_fp16")]; + tensor var_41362_to_fp16 = const()[name = tensor("op_41362_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3511_cast_fp16 = mul(x = var_41361_cast_fp16, y = var_41362_to_fp16)[name = tensor("aw_3511_cast_fp16")]; + tensor var_41365_equation_0 = const()[name = tensor("op_41365_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_41365_cast_fp16 = einsum(equation = var_41365_equation_0, values = (var_41207_cast_fp16, var_41124_cast_fp16))[name = tensor("op_41365_cast_fp16")]; + tensor var_41366_to_fp16 = const()[name = tensor("op_41366_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3513_cast_fp16 = mul(x = var_41365_cast_fp16, y = var_41366_to_fp16)[name = tensor("aw_3513_cast_fp16")]; + tensor var_41369_equation_0 = const()[name = tensor("op_41369_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_41369_cast_fp16 = einsum(equation = var_41369_equation_0, values = (var_41211_cast_fp16, var_41128_cast_fp16))[name = tensor("op_41369_cast_fp16")]; + tensor var_41370_to_fp16 = const()[name = tensor("op_41370_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3515_cast_fp16 = mul(x = var_41369_cast_fp16, y = var_41370_to_fp16)[name = tensor("aw_3515_cast_fp16")]; + tensor var_41373_equation_0 = const()[name = tensor("op_41373_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_41373_cast_fp16 = einsum(equation = var_41373_equation_0, values = (var_41215_cast_fp16, var_41132_cast_fp16))[name = tensor("op_41373_cast_fp16")]; + tensor var_41374_to_fp16 = const()[name = tensor("op_41374_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3517_cast_fp16 = mul(x = var_41373_cast_fp16, y = var_41374_to_fp16)[name = tensor("aw_3517_cast_fp16")]; + tensor var_41377_equation_0 = const()[name = tensor("op_41377_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_41377_cast_fp16 = einsum(equation = var_41377_equation_0, values = (var_41219_cast_fp16, var_41136_cast_fp16))[name = tensor("op_41377_cast_fp16")]; + tensor var_41378_to_fp16 = const()[name = tensor("op_41378_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3519_cast_fp16 = mul(x = var_41377_cast_fp16, y = var_41378_to_fp16)[name = tensor("aw_3519_cast_fp16")]; + tensor var_41380_cast_fp16 = softmax(axis = var_30355, x = aw_3481_cast_fp16)[name = tensor("op_41380_cast_fp16")]; + tensor var_41381_cast_fp16 = softmax(axis = var_30355, x = aw_3483_cast_fp16)[name = tensor("op_41381_cast_fp16")]; + tensor var_41382_cast_fp16 = softmax(axis = var_30355, x = aw_3485_cast_fp16)[name = tensor("op_41382_cast_fp16")]; + tensor var_41383_cast_fp16 = softmax(axis = var_30355, x = aw_3487_cast_fp16)[name = tensor("op_41383_cast_fp16")]; + tensor var_41384_cast_fp16 = softmax(axis = var_30355, x = aw_3489_cast_fp16)[name = tensor("op_41384_cast_fp16")]; + tensor var_41385_cast_fp16 = softmax(axis = var_30355, x = aw_3491_cast_fp16)[name = tensor("op_41385_cast_fp16")]; + tensor var_41386_cast_fp16 = softmax(axis = var_30355, x = aw_3493_cast_fp16)[name = tensor("op_41386_cast_fp16")]; + tensor var_41387_cast_fp16 = softmax(axis = var_30355, x = aw_3495_cast_fp16)[name = tensor("op_41387_cast_fp16")]; + tensor var_41388_cast_fp16 = softmax(axis = var_30355, x = aw_3497_cast_fp16)[name = tensor("op_41388_cast_fp16")]; + tensor var_41389_cast_fp16 = softmax(axis = var_30355, x = aw_3499_cast_fp16)[name = tensor("op_41389_cast_fp16")]; + tensor var_41390_cast_fp16 = softmax(axis = var_30355, x = aw_3501_cast_fp16)[name = tensor("op_41390_cast_fp16")]; + tensor var_41391_cast_fp16 = softmax(axis = var_30355, x = aw_3503_cast_fp16)[name = tensor("op_41391_cast_fp16")]; + tensor var_41392_cast_fp16 = softmax(axis = var_30355, x = aw_3505_cast_fp16)[name = tensor("op_41392_cast_fp16")]; + tensor var_41393_cast_fp16 = softmax(axis = var_30355, x = aw_3507_cast_fp16)[name = tensor("op_41393_cast_fp16")]; + tensor var_41394_cast_fp16 = softmax(axis = var_30355, x = aw_3509_cast_fp16)[name = tensor("op_41394_cast_fp16")]; + tensor var_41395_cast_fp16 = softmax(axis = var_30355, x = aw_3511_cast_fp16)[name = tensor("op_41395_cast_fp16")]; + tensor var_41396_cast_fp16 = softmax(axis = var_30355, x = aw_3513_cast_fp16)[name = tensor("op_41396_cast_fp16")]; + tensor var_41397_cast_fp16 = softmax(axis = var_30355, x = aw_3515_cast_fp16)[name = tensor("op_41397_cast_fp16")]; + tensor var_41398_cast_fp16 = softmax(axis = var_30355, x = aw_3517_cast_fp16)[name = tensor("op_41398_cast_fp16")]; + tensor var_41399_cast_fp16 = softmax(axis = var_30355, x = aw_3519_cast_fp16)[name = tensor("op_41399_cast_fp16")]; + tensor var_41401_equation_0 = const()[name = tensor("op_41401_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_41401_cast_fp16 = einsum(equation = var_41401_equation_0, values = (var_41221_cast_fp16, var_41380_cast_fp16))[name = tensor("op_41401_cast_fp16")]; + tensor var_41403_equation_0 = const()[name = tensor("op_41403_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_41403_cast_fp16 = einsum(equation = var_41403_equation_0, values = (var_41225_cast_fp16, var_41381_cast_fp16))[name = tensor("op_41403_cast_fp16")]; + tensor var_41405_equation_0 = const()[name = tensor("op_41405_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_41405_cast_fp16 = einsum(equation = var_41405_equation_0, values = (var_41229_cast_fp16, var_41382_cast_fp16))[name = tensor("op_41405_cast_fp16")]; + tensor var_41407_equation_0 = const()[name = tensor("op_41407_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_41407_cast_fp16 = einsum(equation = var_41407_equation_0, values = (var_41233_cast_fp16, var_41383_cast_fp16))[name = tensor("op_41407_cast_fp16")]; + tensor var_41409_equation_0 = const()[name = tensor("op_41409_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_41409_cast_fp16 = einsum(equation = var_41409_equation_0, values = (var_41237_cast_fp16, var_41384_cast_fp16))[name = tensor("op_41409_cast_fp16")]; + tensor var_41411_equation_0 = const()[name = tensor("op_41411_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_41411_cast_fp16 = einsum(equation = var_41411_equation_0, values = (var_41241_cast_fp16, var_41385_cast_fp16))[name = tensor("op_41411_cast_fp16")]; + tensor var_41413_equation_0 = const()[name = tensor("op_41413_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_41413_cast_fp16 = einsum(equation = var_41413_equation_0, values = (var_41245_cast_fp16, var_41386_cast_fp16))[name = tensor("op_41413_cast_fp16")]; + tensor var_41415_equation_0 = const()[name = tensor("op_41415_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_41415_cast_fp16 = einsum(equation = var_41415_equation_0, values = (var_41249_cast_fp16, var_41387_cast_fp16))[name = tensor("op_41415_cast_fp16")]; + tensor var_41417_equation_0 = const()[name = tensor("op_41417_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_41417_cast_fp16 = einsum(equation = var_41417_equation_0, values = (var_41253_cast_fp16, var_41388_cast_fp16))[name = tensor("op_41417_cast_fp16")]; + tensor var_41419_equation_0 = const()[name = tensor("op_41419_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_41419_cast_fp16 = einsum(equation = var_41419_equation_0, values = (var_41257_cast_fp16, var_41389_cast_fp16))[name = tensor("op_41419_cast_fp16")]; + tensor var_41421_equation_0 = const()[name = tensor("op_41421_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_41421_cast_fp16 = einsum(equation = var_41421_equation_0, values = (var_41261_cast_fp16, var_41390_cast_fp16))[name = tensor("op_41421_cast_fp16")]; + tensor var_41423_equation_0 = const()[name = tensor("op_41423_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_41423_cast_fp16 = einsum(equation = var_41423_equation_0, values = (var_41265_cast_fp16, var_41391_cast_fp16))[name = tensor("op_41423_cast_fp16")]; + tensor var_41425_equation_0 = const()[name = tensor("op_41425_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_41425_cast_fp16 = einsum(equation = var_41425_equation_0, values = (var_41269_cast_fp16, var_41392_cast_fp16))[name = tensor("op_41425_cast_fp16")]; + tensor var_41427_equation_0 = const()[name = tensor("op_41427_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_41427_cast_fp16 = einsum(equation = var_41427_equation_0, values = (var_41273_cast_fp16, var_41393_cast_fp16))[name = tensor("op_41427_cast_fp16")]; + tensor var_41429_equation_0 = const()[name = tensor("op_41429_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_41429_cast_fp16 = einsum(equation = var_41429_equation_0, values = (var_41277_cast_fp16, var_41394_cast_fp16))[name = tensor("op_41429_cast_fp16")]; + tensor var_41431_equation_0 = const()[name = tensor("op_41431_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_41431_cast_fp16 = einsum(equation = var_41431_equation_0, values = (var_41281_cast_fp16, var_41395_cast_fp16))[name = tensor("op_41431_cast_fp16")]; + tensor var_41433_equation_0 = const()[name = tensor("op_41433_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_41433_cast_fp16 = einsum(equation = var_41433_equation_0, values = (var_41285_cast_fp16, var_41396_cast_fp16))[name = tensor("op_41433_cast_fp16")]; + tensor var_41435_equation_0 = const()[name = tensor("op_41435_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_41435_cast_fp16 = einsum(equation = var_41435_equation_0, values = (var_41289_cast_fp16, var_41397_cast_fp16))[name = tensor("op_41435_cast_fp16")]; + tensor var_41437_equation_0 = const()[name = tensor("op_41437_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_41437_cast_fp16 = einsum(equation = var_41437_equation_0, values = (var_41293_cast_fp16, var_41398_cast_fp16))[name = tensor("op_41437_cast_fp16")]; + tensor var_41439_equation_0 = const()[name = tensor("op_41439_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_41439_cast_fp16 = einsum(equation = var_41439_equation_0, values = (var_41297_cast_fp16, var_41399_cast_fp16))[name = tensor("op_41439_cast_fp16")]; + tensor input_547_interleave_0 = const()[name = tensor("input_547_interleave_0"), val = tensor(false)]; + tensor input_547_cast_fp16 = concat(axis = var_30355, interleave = input_547_interleave_0, values = (var_41401_cast_fp16, var_41403_cast_fp16, var_41405_cast_fp16, var_41407_cast_fp16, var_41409_cast_fp16, var_41411_cast_fp16, var_41413_cast_fp16, var_41415_cast_fp16, var_41417_cast_fp16, var_41419_cast_fp16, var_41421_cast_fp16, var_41423_cast_fp16, var_41425_cast_fp16, var_41427_cast_fp16, var_41429_cast_fp16, var_41431_cast_fp16, var_41433_cast_fp16, var_41435_cast_fp16, var_41437_cast_fp16, var_41439_cast_fp16))[name = tensor("input_547_cast_fp16")]; + tensor var_41445 = const()[name = tensor("op_41445"), val = tensor([1, 1])]; + tensor var_41447 = const()[name = tensor("op_41447"), val = tensor([1, 1])]; + tensor var_41449_pad_type_0 = const()[name = tensor("op_41449_pad_type_0"), val = tensor("custom")]; + tensor var_41449_pad_0 = const()[name = tensor("op_41449_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_1_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_1_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3484841856)))]; + tensor up_blocks_0_attentions_1_transformer_blocks_1_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_1_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3488118720)))]; + tensor var_41449_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_1_attn2_to_out_0_bias_to_fp16, dilations = var_41447, groups = var_30355, pad = var_41449_pad_0, pad_type = var_41449_pad_type_0, strides = var_41445, weight = up_blocks_0_attentions_1_transformer_blocks_1_attn2_to_out_0_weight_to_fp16, x = input_547_cast_fp16)[name = tensor("op_41449_cast_fp16")]; + tensor inputs_275_cast_fp16 = add(x = var_41449_cast_fp16, y = inputs_273_cast_fp16)[name = tensor("inputs_275_cast_fp16")]; + tensor input_549_axes_0 = const()[name = tensor("input_549_axes_0"), val = tensor([1])]; + tensor input_549_gamma_0_to_fp16 = const()[name = tensor("input_549_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3488121344)))]; + tensor input_549_beta_0_to_fp16 = const()[name = tensor("input_549_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3488123968)))]; + tensor var_41459_to_fp16 = const()[name = tensor("op_41459_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_549_cast_fp16 = layer_norm(axes = input_549_axes_0, beta = input_549_beta_0_to_fp16, epsilon = var_41459_to_fp16, gamma = input_549_gamma_0_to_fp16, x = inputs_275_cast_fp16)[name = tensor("input_549_cast_fp16")]; + tensor var_41475 = const()[name = tensor("op_41475"), val = tensor([1, 1])]; + tensor var_41477 = const()[name = tensor("op_41477"), val = tensor([1, 1])]; + tensor var_41479_pad_type_0 = const()[name = tensor("op_41479_pad_type_0"), val = tensor("custom")]; + tensor var_41479_pad_0 = const()[name = tensor("op_41479_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_1_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_1_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3488126592)))]; + tensor up_blocks_0_attentions_1_transformer_blocks_1_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_1_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3514341056)))]; + tensor var_41479_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_1_ff_net_0_proj_bias_to_fp16, dilations = var_41477, groups = var_30355, pad = var_41479_pad_0, pad_type = var_41479_pad_type_0, strides = var_41475, weight = up_blocks_0_attentions_1_transformer_blocks_1_ff_net_0_proj_weight_to_fp16, x = input_549_cast_fp16)[name = tensor("op_41479_cast_fp16")]; + tensor var_41480_split_sizes_0 = const()[name = tensor("op_41480_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_41480_axis_0 = const()[name = tensor("op_41480_axis_0"), val = tensor(1)]; + tensor var_41480_cast_fp16_0, tensor var_41480_cast_fp16_1 = split(axis = var_41480_axis_0, split_sizes = var_41480_split_sizes_0, x = var_41479_cast_fp16)[name = tensor("op_41480_cast_fp16")]; + tensor var_41482_mode_0 = const()[name = tensor("op_41482_mode_0"), val = tensor("EXACT")]; + tensor var_41482_cast_fp16 = gelu(mode = var_41482_mode_0, x = var_41480_cast_fp16_1)[name = tensor("op_41482_cast_fp16")]; + tensor input_551_cast_fp16 = mul(x = var_41480_cast_fp16_0, y = var_41482_cast_fp16)[name = tensor("input_551_cast_fp16")]; + tensor var_41486 = const()[name = tensor("op_41486"), val = tensor([1, 1])]; + tensor var_41488 = const()[name = tensor("op_41488"), val = tensor([1, 1])]; + tensor var_41490_pad_type_0 = const()[name = tensor("op_41490_pad_type_0"), val = tensor("custom")]; + tensor var_41490_pad_0 = const()[name = tensor("op_41490_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_1_ff_net_2_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_1_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3514361600)))]; + tensor up_blocks_0_attentions_1_transformer_blocks_1_ff_net_2_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_1_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3527468864)))]; + tensor var_41490_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_1_ff_net_2_bias_to_fp16, dilations = var_41488, groups = var_30355, pad = var_41490_pad_0, pad_type = var_41490_pad_type_0, strides = var_41486, weight = up_blocks_0_attentions_1_transformer_blocks_1_ff_net_2_weight_to_fp16, x = input_551_cast_fp16)[name = tensor("op_41490_cast_fp16")]; + tensor inputs_277_cast_fp16 = add(x = var_41490_cast_fp16, y = inputs_275_cast_fp16)[name = tensor("inputs_277_cast_fp16")]; + tensor hidden_states_373_axes_0 = const()[name = tensor("hidden_states_373_axes_0"), val = tensor([1])]; + tensor hidden_states_373_gamma_0_to_fp16 = const()[name = tensor("hidden_states_373_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3527471488)))]; + tensor hidden_states_373_beta_0_to_fp16 = const()[name = tensor("hidden_states_373_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3527474112)))]; + tensor var_41506_to_fp16 = const()[name = tensor("op_41506_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_373_cast_fp16 = layer_norm(axes = hidden_states_373_axes_0, beta = hidden_states_373_beta_0_to_fp16, epsilon = var_41506_to_fp16, gamma = hidden_states_373_gamma_0_to_fp16, x = inputs_277_cast_fp16)[name = tensor("hidden_states_373_cast_fp16")]; + tensor var_41521 = const()[name = tensor("op_41521"), val = tensor([1, 1])]; + tensor var_41523 = const()[name = tensor("op_41523"), val = tensor([1, 1])]; + tensor q_185_pad_type_0 = const()[name = tensor("q_185_pad_type_0"), val = tensor("custom")]; + tensor q_185_pad_0 = const()[name = tensor("q_185_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_2_attn1_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_2_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3527476736)))]; + tensor q_185_cast_fp16 = conv(dilations = var_41523, groups = var_30355, pad = q_185_pad_0, pad_type = q_185_pad_type_0, strides = var_41521, weight = up_blocks_0_attentions_1_transformer_blocks_2_attn1_to_q_weight_to_fp16, x = hidden_states_373_cast_fp16)[name = tensor("q_185_cast_fp16")]; + tensor var_41527 = const()[name = tensor("op_41527"), val = tensor([1, 1])]; + tensor var_41529 = const()[name = tensor("op_41529"), val = tensor([1, 1])]; + tensor k_369_pad_type_0 = const()[name = tensor("k_369_pad_type_0"), val = tensor("custom")]; + tensor k_369_pad_0 = const()[name = tensor("k_369_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_2_attn1_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_2_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3530753600)))]; + tensor k_369_cast_fp16 = conv(dilations = var_41529, groups = var_30355, pad = k_369_pad_0, pad_type = k_369_pad_type_0, strides = var_41527, weight = up_blocks_0_attentions_1_transformer_blocks_2_attn1_to_k_weight_to_fp16, x = hidden_states_373_cast_fp16)[name = tensor("k_369_cast_fp16")]; + tensor var_41533 = const()[name = tensor("op_41533"), val = tensor([1, 1])]; + tensor var_41535 = const()[name = tensor("op_41535"), val = tensor([1, 1])]; + tensor v_185_pad_type_0 = const()[name = tensor("v_185_pad_type_0"), val = tensor("custom")]; + tensor v_185_pad_0 = const()[name = tensor("v_185_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_2_attn1_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_2_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3534030464)))]; + tensor v_185_cast_fp16 = conv(dilations = var_41535, groups = var_30355, pad = v_185_pad_0, pad_type = v_185_pad_type_0, strides = var_41533, weight = up_blocks_0_attentions_1_transformer_blocks_2_attn1_to_v_weight_to_fp16, x = hidden_states_373_cast_fp16)[name = tensor("v_185_cast_fp16")]; + tensor var_41539_begin_0 = const()[name = tensor("op_41539_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_41539_end_0 = const()[name = tensor("op_41539_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_41539_end_mask_0 = const()[name = tensor("op_41539_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41539_cast_fp16 = slice_by_index(begin = var_41539_begin_0, end = var_41539_end_0, end_mask = var_41539_end_mask_0, x = q_185_cast_fp16)[name = tensor("op_41539_cast_fp16")]; + tensor var_41543_begin_0 = const()[name = tensor("op_41543_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_41543_end_0 = const()[name = tensor("op_41543_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_41543_end_mask_0 = const()[name = tensor("op_41543_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41543_cast_fp16 = slice_by_index(begin = var_41543_begin_0, end = var_41543_end_0, end_mask = var_41543_end_mask_0, x = q_185_cast_fp16)[name = tensor("op_41543_cast_fp16")]; + tensor var_41547_begin_0 = const()[name = tensor("op_41547_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_41547_end_0 = const()[name = tensor("op_41547_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_41547_end_mask_0 = const()[name = tensor("op_41547_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41547_cast_fp16 = slice_by_index(begin = var_41547_begin_0, end = var_41547_end_0, end_mask = var_41547_end_mask_0, x = q_185_cast_fp16)[name = tensor("op_41547_cast_fp16")]; + tensor var_41551_begin_0 = const()[name = tensor("op_41551_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_41551_end_0 = const()[name = tensor("op_41551_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_41551_end_mask_0 = const()[name = tensor("op_41551_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41551_cast_fp16 = slice_by_index(begin = var_41551_begin_0, end = var_41551_end_0, end_mask = var_41551_end_mask_0, x = q_185_cast_fp16)[name = tensor("op_41551_cast_fp16")]; + tensor var_41555_begin_0 = const()[name = tensor("op_41555_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_41555_end_0 = const()[name = tensor("op_41555_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_41555_end_mask_0 = const()[name = tensor("op_41555_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41555_cast_fp16 = slice_by_index(begin = var_41555_begin_0, end = var_41555_end_0, end_mask = var_41555_end_mask_0, x = q_185_cast_fp16)[name = tensor("op_41555_cast_fp16")]; + tensor var_41559_begin_0 = const()[name = tensor("op_41559_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_41559_end_0 = const()[name = tensor("op_41559_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_41559_end_mask_0 = const()[name = tensor("op_41559_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41559_cast_fp16 = slice_by_index(begin = var_41559_begin_0, end = var_41559_end_0, end_mask = var_41559_end_mask_0, x = q_185_cast_fp16)[name = tensor("op_41559_cast_fp16")]; + tensor var_41563_begin_0 = const()[name = tensor("op_41563_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_41563_end_0 = const()[name = tensor("op_41563_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_41563_end_mask_0 = const()[name = tensor("op_41563_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41563_cast_fp16 = slice_by_index(begin = var_41563_begin_0, end = var_41563_end_0, end_mask = var_41563_end_mask_0, x = q_185_cast_fp16)[name = tensor("op_41563_cast_fp16")]; + tensor var_41567_begin_0 = const()[name = tensor("op_41567_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_41567_end_0 = const()[name = tensor("op_41567_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_41567_end_mask_0 = const()[name = tensor("op_41567_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41567_cast_fp16 = slice_by_index(begin = var_41567_begin_0, end = var_41567_end_0, end_mask = var_41567_end_mask_0, x = q_185_cast_fp16)[name = tensor("op_41567_cast_fp16")]; + tensor var_41571_begin_0 = const()[name = tensor("op_41571_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_41571_end_0 = const()[name = tensor("op_41571_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_41571_end_mask_0 = const()[name = tensor("op_41571_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41571_cast_fp16 = slice_by_index(begin = var_41571_begin_0, end = var_41571_end_0, end_mask = var_41571_end_mask_0, x = q_185_cast_fp16)[name = tensor("op_41571_cast_fp16")]; + tensor var_41575_begin_0 = const()[name = tensor("op_41575_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_41575_end_0 = const()[name = tensor("op_41575_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_41575_end_mask_0 = const()[name = tensor("op_41575_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41575_cast_fp16 = slice_by_index(begin = var_41575_begin_0, end = var_41575_end_0, end_mask = var_41575_end_mask_0, x = q_185_cast_fp16)[name = tensor("op_41575_cast_fp16")]; + tensor var_41579_begin_0 = const()[name = tensor("op_41579_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_41579_end_0 = const()[name = tensor("op_41579_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_41579_end_mask_0 = const()[name = tensor("op_41579_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41579_cast_fp16 = slice_by_index(begin = var_41579_begin_0, end = var_41579_end_0, end_mask = var_41579_end_mask_0, x = q_185_cast_fp16)[name = tensor("op_41579_cast_fp16")]; + tensor var_41583_begin_0 = const()[name = tensor("op_41583_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_41583_end_0 = const()[name = tensor("op_41583_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_41583_end_mask_0 = const()[name = tensor("op_41583_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41583_cast_fp16 = slice_by_index(begin = var_41583_begin_0, end = var_41583_end_0, end_mask = var_41583_end_mask_0, x = q_185_cast_fp16)[name = tensor("op_41583_cast_fp16")]; + tensor var_41587_begin_0 = const()[name = tensor("op_41587_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_41587_end_0 = const()[name = tensor("op_41587_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_41587_end_mask_0 = const()[name = tensor("op_41587_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41587_cast_fp16 = slice_by_index(begin = var_41587_begin_0, end = var_41587_end_0, end_mask = var_41587_end_mask_0, x = q_185_cast_fp16)[name = tensor("op_41587_cast_fp16")]; + tensor var_41591_begin_0 = const()[name = tensor("op_41591_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_41591_end_0 = const()[name = tensor("op_41591_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_41591_end_mask_0 = const()[name = tensor("op_41591_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41591_cast_fp16 = slice_by_index(begin = var_41591_begin_0, end = var_41591_end_0, end_mask = var_41591_end_mask_0, x = q_185_cast_fp16)[name = tensor("op_41591_cast_fp16")]; + tensor var_41595_begin_0 = const()[name = tensor("op_41595_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_41595_end_0 = const()[name = tensor("op_41595_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_41595_end_mask_0 = const()[name = tensor("op_41595_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41595_cast_fp16 = slice_by_index(begin = var_41595_begin_0, end = var_41595_end_0, end_mask = var_41595_end_mask_0, x = q_185_cast_fp16)[name = tensor("op_41595_cast_fp16")]; + tensor var_41599_begin_0 = const()[name = tensor("op_41599_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_41599_end_0 = const()[name = tensor("op_41599_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_41599_end_mask_0 = const()[name = tensor("op_41599_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41599_cast_fp16 = slice_by_index(begin = var_41599_begin_0, end = var_41599_end_0, end_mask = var_41599_end_mask_0, x = q_185_cast_fp16)[name = tensor("op_41599_cast_fp16")]; + tensor var_41603_begin_0 = const()[name = tensor("op_41603_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_41603_end_0 = const()[name = tensor("op_41603_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_41603_end_mask_0 = const()[name = tensor("op_41603_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41603_cast_fp16 = slice_by_index(begin = var_41603_begin_0, end = var_41603_end_0, end_mask = var_41603_end_mask_0, x = q_185_cast_fp16)[name = tensor("op_41603_cast_fp16")]; + tensor var_41607_begin_0 = const()[name = tensor("op_41607_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_41607_end_0 = const()[name = tensor("op_41607_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_41607_end_mask_0 = const()[name = tensor("op_41607_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41607_cast_fp16 = slice_by_index(begin = var_41607_begin_0, end = var_41607_end_0, end_mask = var_41607_end_mask_0, x = q_185_cast_fp16)[name = tensor("op_41607_cast_fp16")]; + tensor var_41611_begin_0 = const()[name = tensor("op_41611_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_41611_end_0 = const()[name = tensor("op_41611_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_41611_end_mask_0 = const()[name = tensor("op_41611_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41611_cast_fp16 = slice_by_index(begin = var_41611_begin_0, end = var_41611_end_0, end_mask = var_41611_end_mask_0, x = q_185_cast_fp16)[name = tensor("op_41611_cast_fp16")]; + tensor var_41615_begin_0 = const()[name = tensor("op_41615_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_41615_end_0 = const()[name = tensor("op_41615_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_41615_end_mask_0 = const()[name = tensor("op_41615_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41615_cast_fp16 = slice_by_index(begin = var_41615_begin_0, end = var_41615_end_0, end_mask = var_41615_end_mask_0, x = q_185_cast_fp16)[name = tensor("op_41615_cast_fp16")]; + tensor k_371_perm_0 = const()[name = tensor("k_371_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_41622_begin_0 = const()[name = tensor("op_41622_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_41622_end_0 = const()[name = tensor("op_41622_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_41622_end_mask_0 = const()[name = tensor("op_41622_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_371_cast_fp16 = transpose(perm = k_371_perm_0, x = k_369_cast_fp16)[name = tensor("transpose_47")]; + tensor var_41622_cast_fp16 = slice_by_index(begin = var_41622_begin_0, end = var_41622_end_0, end_mask = var_41622_end_mask_0, x = k_371_cast_fp16)[name = tensor("op_41622_cast_fp16")]; + tensor var_41626_begin_0 = const()[name = tensor("op_41626_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_41626_end_0 = const()[name = tensor("op_41626_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_41626_end_mask_0 = const()[name = tensor("op_41626_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41626_cast_fp16 = slice_by_index(begin = var_41626_begin_0, end = var_41626_end_0, end_mask = var_41626_end_mask_0, x = k_371_cast_fp16)[name = tensor("op_41626_cast_fp16")]; + tensor var_41630_begin_0 = const()[name = tensor("op_41630_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_41630_end_0 = const()[name = tensor("op_41630_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_41630_end_mask_0 = const()[name = tensor("op_41630_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41630_cast_fp16 = slice_by_index(begin = var_41630_begin_0, end = var_41630_end_0, end_mask = var_41630_end_mask_0, x = k_371_cast_fp16)[name = tensor("op_41630_cast_fp16")]; + tensor var_41634_begin_0 = const()[name = tensor("op_41634_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_41634_end_0 = const()[name = tensor("op_41634_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_41634_end_mask_0 = const()[name = tensor("op_41634_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41634_cast_fp16 = slice_by_index(begin = var_41634_begin_0, end = var_41634_end_0, end_mask = var_41634_end_mask_0, x = k_371_cast_fp16)[name = tensor("op_41634_cast_fp16")]; + tensor var_41638_begin_0 = const()[name = tensor("op_41638_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_41638_end_0 = const()[name = tensor("op_41638_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_41638_end_mask_0 = const()[name = tensor("op_41638_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41638_cast_fp16 = slice_by_index(begin = var_41638_begin_0, end = var_41638_end_0, end_mask = var_41638_end_mask_0, x = k_371_cast_fp16)[name = tensor("op_41638_cast_fp16")]; + tensor var_41642_begin_0 = const()[name = tensor("op_41642_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_41642_end_0 = const()[name = tensor("op_41642_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_41642_end_mask_0 = const()[name = tensor("op_41642_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41642_cast_fp16 = slice_by_index(begin = var_41642_begin_0, end = var_41642_end_0, end_mask = var_41642_end_mask_0, x = k_371_cast_fp16)[name = tensor("op_41642_cast_fp16")]; + tensor var_41646_begin_0 = const()[name = tensor("op_41646_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_41646_end_0 = const()[name = tensor("op_41646_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_41646_end_mask_0 = const()[name = tensor("op_41646_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41646_cast_fp16 = slice_by_index(begin = var_41646_begin_0, end = var_41646_end_0, end_mask = var_41646_end_mask_0, x = k_371_cast_fp16)[name = tensor("op_41646_cast_fp16")]; + tensor var_41650_begin_0 = const()[name = tensor("op_41650_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_41650_end_0 = const()[name = tensor("op_41650_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_41650_end_mask_0 = const()[name = tensor("op_41650_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41650_cast_fp16 = slice_by_index(begin = var_41650_begin_0, end = var_41650_end_0, end_mask = var_41650_end_mask_0, x = k_371_cast_fp16)[name = tensor("op_41650_cast_fp16")]; + tensor var_41654_begin_0 = const()[name = tensor("op_41654_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_41654_end_0 = const()[name = tensor("op_41654_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_41654_end_mask_0 = const()[name = tensor("op_41654_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41654_cast_fp16 = slice_by_index(begin = var_41654_begin_0, end = var_41654_end_0, end_mask = var_41654_end_mask_0, x = k_371_cast_fp16)[name = tensor("op_41654_cast_fp16")]; + tensor var_41658_begin_0 = const()[name = tensor("op_41658_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_41658_end_0 = const()[name = tensor("op_41658_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_41658_end_mask_0 = const()[name = tensor("op_41658_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41658_cast_fp16 = slice_by_index(begin = var_41658_begin_0, end = var_41658_end_0, end_mask = var_41658_end_mask_0, x = k_371_cast_fp16)[name = tensor("op_41658_cast_fp16")]; + tensor var_41662_begin_0 = const()[name = tensor("op_41662_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_41662_end_0 = const()[name = tensor("op_41662_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_41662_end_mask_0 = const()[name = tensor("op_41662_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41662_cast_fp16 = slice_by_index(begin = var_41662_begin_0, end = var_41662_end_0, end_mask = var_41662_end_mask_0, x = k_371_cast_fp16)[name = tensor("op_41662_cast_fp16")]; + tensor var_41666_begin_0 = const()[name = tensor("op_41666_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_41666_end_0 = const()[name = tensor("op_41666_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_41666_end_mask_0 = const()[name = tensor("op_41666_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41666_cast_fp16 = slice_by_index(begin = var_41666_begin_0, end = var_41666_end_0, end_mask = var_41666_end_mask_0, x = k_371_cast_fp16)[name = tensor("op_41666_cast_fp16")]; + tensor var_41670_begin_0 = const()[name = tensor("op_41670_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_41670_end_0 = const()[name = tensor("op_41670_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_41670_end_mask_0 = const()[name = tensor("op_41670_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41670_cast_fp16 = slice_by_index(begin = var_41670_begin_0, end = var_41670_end_0, end_mask = var_41670_end_mask_0, x = k_371_cast_fp16)[name = tensor("op_41670_cast_fp16")]; + tensor var_41674_begin_0 = const()[name = tensor("op_41674_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_41674_end_0 = const()[name = tensor("op_41674_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_41674_end_mask_0 = const()[name = tensor("op_41674_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41674_cast_fp16 = slice_by_index(begin = var_41674_begin_0, end = var_41674_end_0, end_mask = var_41674_end_mask_0, x = k_371_cast_fp16)[name = tensor("op_41674_cast_fp16")]; + tensor var_41678_begin_0 = const()[name = tensor("op_41678_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_41678_end_0 = const()[name = tensor("op_41678_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_41678_end_mask_0 = const()[name = tensor("op_41678_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41678_cast_fp16 = slice_by_index(begin = var_41678_begin_0, end = var_41678_end_0, end_mask = var_41678_end_mask_0, x = k_371_cast_fp16)[name = tensor("op_41678_cast_fp16")]; + tensor var_41682_begin_0 = const()[name = tensor("op_41682_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_41682_end_0 = const()[name = tensor("op_41682_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_41682_end_mask_0 = const()[name = tensor("op_41682_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41682_cast_fp16 = slice_by_index(begin = var_41682_begin_0, end = var_41682_end_0, end_mask = var_41682_end_mask_0, x = k_371_cast_fp16)[name = tensor("op_41682_cast_fp16")]; + tensor var_41686_begin_0 = const()[name = tensor("op_41686_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_41686_end_0 = const()[name = tensor("op_41686_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_41686_end_mask_0 = const()[name = tensor("op_41686_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41686_cast_fp16 = slice_by_index(begin = var_41686_begin_0, end = var_41686_end_0, end_mask = var_41686_end_mask_0, x = k_371_cast_fp16)[name = tensor("op_41686_cast_fp16")]; + tensor var_41690_begin_0 = const()[name = tensor("op_41690_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_41690_end_0 = const()[name = tensor("op_41690_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_41690_end_mask_0 = const()[name = tensor("op_41690_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41690_cast_fp16 = slice_by_index(begin = var_41690_begin_0, end = var_41690_end_0, end_mask = var_41690_end_mask_0, x = k_371_cast_fp16)[name = tensor("op_41690_cast_fp16")]; + tensor var_41694_begin_0 = const()[name = tensor("op_41694_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_41694_end_0 = const()[name = tensor("op_41694_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_41694_end_mask_0 = const()[name = tensor("op_41694_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41694_cast_fp16 = slice_by_index(begin = var_41694_begin_0, end = var_41694_end_0, end_mask = var_41694_end_mask_0, x = k_371_cast_fp16)[name = tensor("op_41694_cast_fp16")]; + tensor var_41698_begin_0 = const()[name = tensor("op_41698_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_41698_end_0 = const()[name = tensor("op_41698_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_41698_end_mask_0 = const()[name = tensor("op_41698_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41698_cast_fp16 = slice_by_index(begin = var_41698_begin_0, end = var_41698_end_0, end_mask = var_41698_end_mask_0, x = k_371_cast_fp16)[name = tensor("op_41698_cast_fp16")]; + tensor var_41700_begin_0 = const()[name = tensor("op_41700_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_41700_end_0 = const()[name = tensor("op_41700_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_41700_end_mask_0 = const()[name = tensor("op_41700_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41700_cast_fp16 = slice_by_index(begin = var_41700_begin_0, end = var_41700_end_0, end_mask = var_41700_end_mask_0, x = v_185_cast_fp16)[name = tensor("op_41700_cast_fp16")]; + tensor var_41704_begin_0 = const()[name = tensor("op_41704_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_41704_end_0 = const()[name = tensor("op_41704_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_41704_end_mask_0 = const()[name = tensor("op_41704_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41704_cast_fp16 = slice_by_index(begin = var_41704_begin_0, end = var_41704_end_0, end_mask = var_41704_end_mask_0, x = v_185_cast_fp16)[name = tensor("op_41704_cast_fp16")]; + tensor var_41708_begin_0 = const()[name = tensor("op_41708_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_41708_end_0 = const()[name = tensor("op_41708_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_41708_end_mask_0 = const()[name = tensor("op_41708_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41708_cast_fp16 = slice_by_index(begin = var_41708_begin_0, end = var_41708_end_0, end_mask = var_41708_end_mask_0, x = v_185_cast_fp16)[name = tensor("op_41708_cast_fp16")]; + tensor var_41712_begin_0 = const()[name = tensor("op_41712_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_41712_end_0 = const()[name = tensor("op_41712_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_41712_end_mask_0 = const()[name = tensor("op_41712_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41712_cast_fp16 = slice_by_index(begin = var_41712_begin_0, end = var_41712_end_0, end_mask = var_41712_end_mask_0, x = v_185_cast_fp16)[name = tensor("op_41712_cast_fp16")]; + tensor var_41716_begin_0 = const()[name = tensor("op_41716_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_41716_end_0 = const()[name = tensor("op_41716_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_41716_end_mask_0 = const()[name = tensor("op_41716_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41716_cast_fp16 = slice_by_index(begin = var_41716_begin_0, end = var_41716_end_0, end_mask = var_41716_end_mask_0, x = v_185_cast_fp16)[name = tensor("op_41716_cast_fp16")]; + tensor var_41720_begin_0 = const()[name = tensor("op_41720_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_41720_end_0 = const()[name = tensor("op_41720_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_41720_end_mask_0 = const()[name = tensor("op_41720_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41720_cast_fp16 = slice_by_index(begin = var_41720_begin_0, end = var_41720_end_0, end_mask = var_41720_end_mask_0, x = v_185_cast_fp16)[name = tensor("op_41720_cast_fp16")]; + tensor var_41724_begin_0 = const()[name = tensor("op_41724_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_41724_end_0 = const()[name = tensor("op_41724_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_41724_end_mask_0 = const()[name = tensor("op_41724_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41724_cast_fp16 = slice_by_index(begin = var_41724_begin_0, end = var_41724_end_0, end_mask = var_41724_end_mask_0, x = v_185_cast_fp16)[name = tensor("op_41724_cast_fp16")]; + tensor var_41728_begin_0 = const()[name = tensor("op_41728_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_41728_end_0 = const()[name = tensor("op_41728_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_41728_end_mask_0 = const()[name = tensor("op_41728_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41728_cast_fp16 = slice_by_index(begin = var_41728_begin_0, end = var_41728_end_0, end_mask = var_41728_end_mask_0, x = v_185_cast_fp16)[name = tensor("op_41728_cast_fp16")]; + tensor var_41732_begin_0 = const()[name = tensor("op_41732_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_41732_end_0 = const()[name = tensor("op_41732_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_41732_end_mask_0 = const()[name = tensor("op_41732_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41732_cast_fp16 = slice_by_index(begin = var_41732_begin_0, end = var_41732_end_0, end_mask = var_41732_end_mask_0, x = v_185_cast_fp16)[name = tensor("op_41732_cast_fp16")]; + tensor var_41736_begin_0 = const()[name = tensor("op_41736_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_41736_end_0 = const()[name = tensor("op_41736_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_41736_end_mask_0 = const()[name = tensor("op_41736_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41736_cast_fp16 = slice_by_index(begin = var_41736_begin_0, end = var_41736_end_0, end_mask = var_41736_end_mask_0, x = v_185_cast_fp16)[name = tensor("op_41736_cast_fp16")]; + tensor var_41740_begin_0 = const()[name = tensor("op_41740_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_41740_end_0 = const()[name = tensor("op_41740_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_41740_end_mask_0 = const()[name = tensor("op_41740_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41740_cast_fp16 = slice_by_index(begin = var_41740_begin_0, end = var_41740_end_0, end_mask = var_41740_end_mask_0, x = v_185_cast_fp16)[name = tensor("op_41740_cast_fp16")]; + tensor var_41744_begin_0 = const()[name = tensor("op_41744_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_41744_end_0 = const()[name = tensor("op_41744_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_41744_end_mask_0 = const()[name = tensor("op_41744_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41744_cast_fp16 = slice_by_index(begin = var_41744_begin_0, end = var_41744_end_0, end_mask = var_41744_end_mask_0, x = v_185_cast_fp16)[name = tensor("op_41744_cast_fp16")]; + tensor var_41748_begin_0 = const()[name = tensor("op_41748_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_41748_end_0 = const()[name = tensor("op_41748_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_41748_end_mask_0 = const()[name = tensor("op_41748_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41748_cast_fp16 = slice_by_index(begin = var_41748_begin_0, end = var_41748_end_0, end_mask = var_41748_end_mask_0, x = v_185_cast_fp16)[name = tensor("op_41748_cast_fp16")]; + tensor var_41752_begin_0 = const()[name = tensor("op_41752_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_41752_end_0 = const()[name = tensor("op_41752_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_41752_end_mask_0 = const()[name = tensor("op_41752_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41752_cast_fp16 = slice_by_index(begin = var_41752_begin_0, end = var_41752_end_0, end_mask = var_41752_end_mask_0, x = v_185_cast_fp16)[name = tensor("op_41752_cast_fp16")]; + tensor var_41756_begin_0 = const()[name = tensor("op_41756_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_41756_end_0 = const()[name = tensor("op_41756_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_41756_end_mask_0 = const()[name = tensor("op_41756_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41756_cast_fp16 = slice_by_index(begin = var_41756_begin_0, end = var_41756_end_0, end_mask = var_41756_end_mask_0, x = v_185_cast_fp16)[name = tensor("op_41756_cast_fp16")]; + tensor var_41760_begin_0 = const()[name = tensor("op_41760_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_41760_end_0 = const()[name = tensor("op_41760_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_41760_end_mask_0 = const()[name = tensor("op_41760_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41760_cast_fp16 = slice_by_index(begin = var_41760_begin_0, end = var_41760_end_0, end_mask = var_41760_end_mask_0, x = v_185_cast_fp16)[name = tensor("op_41760_cast_fp16")]; + tensor var_41764_begin_0 = const()[name = tensor("op_41764_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_41764_end_0 = const()[name = tensor("op_41764_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_41764_end_mask_0 = const()[name = tensor("op_41764_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41764_cast_fp16 = slice_by_index(begin = var_41764_begin_0, end = var_41764_end_0, end_mask = var_41764_end_mask_0, x = v_185_cast_fp16)[name = tensor("op_41764_cast_fp16")]; + tensor var_41768_begin_0 = const()[name = tensor("op_41768_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_41768_end_0 = const()[name = tensor("op_41768_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_41768_end_mask_0 = const()[name = tensor("op_41768_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41768_cast_fp16 = slice_by_index(begin = var_41768_begin_0, end = var_41768_end_0, end_mask = var_41768_end_mask_0, x = v_185_cast_fp16)[name = tensor("op_41768_cast_fp16")]; + tensor var_41772_begin_0 = const()[name = tensor("op_41772_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_41772_end_0 = const()[name = tensor("op_41772_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_41772_end_mask_0 = const()[name = tensor("op_41772_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41772_cast_fp16 = slice_by_index(begin = var_41772_begin_0, end = var_41772_end_0, end_mask = var_41772_end_mask_0, x = v_185_cast_fp16)[name = tensor("op_41772_cast_fp16")]; + tensor var_41776_begin_0 = const()[name = tensor("op_41776_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_41776_end_0 = const()[name = tensor("op_41776_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_41776_end_mask_0 = const()[name = tensor("op_41776_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41776_cast_fp16 = slice_by_index(begin = var_41776_begin_0, end = var_41776_end_0, end_mask = var_41776_end_mask_0, x = v_185_cast_fp16)[name = tensor("op_41776_cast_fp16")]; + tensor var_41780_equation_0 = const()[name = tensor("op_41780_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_41780_cast_fp16 = einsum(equation = var_41780_equation_0, values = (var_41622_cast_fp16, var_41539_cast_fp16))[name = tensor("op_41780_cast_fp16")]; + tensor var_41781_to_fp16 = const()[name = tensor("op_41781_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3521_cast_fp16 = mul(x = var_41780_cast_fp16, y = var_41781_to_fp16)[name = tensor("aw_3521_cast_fp16")]; + tensor var_41784_equation_0 = const()[name = tensor("op_41784_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_41784_cast_fp16 = einsum(equation = var_41784_equation_0, values = (var_41626_cast_fp16, var_41543_cast_fp16))[name = tensor("op_41784_cast_fp16")]; + tensor var_41785_to_fp16 = const()[name = tensor("op_41785_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3523_cast_fp16 = mul(x = var_41784_cast_fp16, y = var_41785_to_fp16)[name = tensor("aw_3523_cast_fp16")]; + tensor var_41788_equation_0 = const()[name = tensor("op_41788_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_41788_cast_fp16 = einsum(equation = var_41788_equation_0, values = (var_41630_cast_fp16, var_41547_cast_fp16))[name = tensor("op_41788_cast_fp16")]; + tensor var_41789_to_fp16 = const()[name = tensor("op_41789_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3525_cast_fp16 = mul(x = var_41788_cast_fp16, y = var_41789_to_fp16)[name = tensor("aw_3525_cast_fp16")]; + tensor var_41792_equation_0 = const()[name = tensor("op_41792_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_41792_cast_fp16 = einsum(equation = var_41792_equation_0, values = (var_41634_cast_fp16, var_41551_cast_fp16))[name = tensor("op_41792_cast_fp16")]; + tensor var_41793_to_fp16 = const()[name = tensor("op_41793_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3527_cast_fp16 = mul(x = var_41792_cast_fp16, y = var_41793_to_fp16)[name = tensor("aw_3527_cast_fp16")]; + tensor var_41796_equation_0 = const()[name = tensor("op_41796_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_41796_cast_fp16 = einsum(equation = var_41796_equation_0, values = (var_41638_cast_fp16, var_41555_cast_fp16))[name = tensor("op_41796_cast_fp16")]; + tensor var_41797_to_fp16 = const()[name = tensor("op_41797_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3529_cast_fp16 = mul(x = var_41796_cast_fp16, y = var_41797_to_fp16)[name = tensor("aw_3529_cast_fp16")]; + tensor var_41800_equation_0 = const()[name = tensor("op_41800_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_41800_cast_fp16 = einsum(equation = var_41800_equation_0, values = (var_41642_cast_fp16, var_41559_cast_fp16))[name = tensor("op_41800_cast_fp16")]; + tensor var_41801_to_fp16 = const()[name = tensor("op_41801_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3531_cast_fp16 = mul(x = var_41800_cast_fp16, y = var_41801_to_fp16)[name = tensor("aw_3531_cast_fp16")]; + tensor var_41804_equation_0 = const()[name = tensor("op_41804_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_41804_cast_fp16 = einsum(equation = var_41804_equation_0, values = (var_41646_cast_fp16, var_41563_cast_fp16))[name = tensor("op_41804_cast_fp16")]; + tensor var_41805_to_fp16 = const()[name = tensor("op_41805_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3533_cast_fp16 = mul(x = var_41804_cast_fp16, y = var_41805_to_fp16)[name = tensor("aw_3533_cast_fp16")]; + tensor var_41808_equation_0 = const()[name = tensor("op_41808_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_41808_cast_fp16 = einsum(equation = var_41808_equation_0, values = (var_41650_cast_fp16, var_41567_cast_fp16))[name = tensor("op_41808_cast_fp16")]; + tensor var_41809_to_fp16 = const()[name = tensor("op_41809_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3535_cast_fp16 = mul(x = var_41808_cast_fp16, y = var_41809_to_fp16)[name = tensor("aw_3535_cast_fp16")]; + tensor var_41812_equation_0 = const()[name = tensor("op_41812_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_41812_cast_fp16 = einsum(equation = var_41812_equation_0, values = (var_41654_cast_fp16, var_41571_cast_fp16))[name = tensor("op_41812_cast_fp16")]; + tensor var_41813_to_fp16 = const()[name = tensor("op_41813_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3537_cast_fp16 = mul(x = var_41812_cast_fp16, y = var_41813_to_fp16)[name = tensor("aw_3537_cast_fp16")]; + tensor var_41816_equation_0 = const()[name = tensor("op_41816_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_41816_cast_fp16 = einsum(equation = var_41816_equation_0, values = (var_41658_cast_fp16, var_41575_cast_fp16))[name = tensor("op_41816_cast_fp16")]; + tensor var_41817_to_fp16 = const()[name = tensor("op_41817_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3539_cast_fp16 = mul(x = var_41816_cast_fp16, y = var_41817_to_fp16)[name = tensor("aw_3539_cast_fp16")]; + tensor var_41820_equation_0 = const()[name = tensor("op_41820_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_41820_cast_fp16 = einsum(equation = var_41820_equation_0, values = (var_41662_cast_fp16, var_41579_cast_fp16))[name = tensor("op_41820_cast_fp16")]; + tensor var_41821_to_fp16 = const()[name = tensor("op_41821_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3541_cast_fp16 = mul(x = var_41820_cast_fp16, y = var_41821_to_fp16)[name = tensor("aw_3541_cast_fp16")]; + tensor var_41824_equation_0 = const()[name = tensor("op_41824_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_41824_cast_fp16 = einsum(equation = var_41824_equation_0, values = (var_41666_cast_fp16, var_41583_cast_fp16))[name = tensor("op_41824_cast_fp16")]; + tensor var_41825_to_fp16 = const()[name = tensor("op_41825_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3543_cast_fp16 = mul(x = var_41824_cast_fp16, y = var_41825_to_fp16)[name = tensor("aw_3543_cast_fp16")]; + tensor var_41828_equation_0 = const()[name = tensor("op_41828_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_41828_cast_fp16 = einsum(equation = var_41828_equation_0, values = (var_41670_cast_fp16, var_41587_cast_fp16))[name = tensor("op_41828_cast_fp16")]; + tensor var_41829_to_fp16 = const()[name = tensor("op_41829_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3545_cast_fp16 = mul(x = var_41828_cast_fp16, y = var_41829_to_fp16)[name = tensor("aw_3545_cast_fp16")]; + tensor var_41832_equation_0 = const()[name = tensor("op_41832_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_41832_cast_fp16 = einsum(equation = var_41832_equation_0, values = (var_41674_cast_fp16, var_41591_cast_fp16))[name = tensor("op_41832_cast_fp16")]; + tensor var_41833_to_fp16 = const()[name = tensor("op_41833_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3547_cast_fp16 = mul(x = var_41832_cast_fp16, y = var_41833_to_fp16)[name = tensor("aw_3547_cast_fp16")]; + tensor var_41836_equation_0 = const()[name = tensor("op_41836_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_41836_cast_fp16 = einsum(equation = var_41836_equation_0, values = (var_41678_cast_fp16, var_41595_cast_fp16))[name = tensor("op_41836_cast_fp16")]; + tensor var_41837_to_fp16 = const()[name = tensor("op_41837_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3549_cast_fp16 = mul(x = var_41836_cast_fp16, y = var_41837_to_fp16)[name = tensor("aw_3549_cast_fp16")]; + tensor var_41840_equation_0 = const()[name = tensor("op_41840_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_41840_cast_fp16 = einsum(equation = var_41840_equation_0, values = (var_41682_cast_fp16, var_41599_cast_fp16))[name = tensor("op_41840_cast_fp16")]; + tensor var_41841_to_fp16 = const()[name = tensor("op_41841_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3551_cast_fp16 = mul(x = var_41840_cast_fp16, y = var_41841_to_fp16)[name = tensor("aw_3551_cast_fp16")]; + tensor var_41844_equation_0 = const()[name = tensor("op_41844_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_41844_cast_fp16 = einsum(equation = var_41844_equation_0, values = (var_41686_cast_fp16, var_41603_cast_fp16))[name = tensor("op_41844_cast_fp16")]; + tensor var_41845_to_fp16 = const()[name = tensor("op_41845_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3553_cast_fp16 = mul(x = var_41844_cast_fp16, y = var_41845_to_fp16)[name = tensor("aw_3553_cast_fp16")]; + tensor var_41848_equation_0 = const()[name = tensor("op_41848_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_41848_cast_fp16 = einsum(equation = var_41848_equation_0, values = (var_41690_cast_fp16, var_41607_cast_fp16))[name = tensor("op_41848_cast_fp16")]; + tensor var_41849_to_fp16 = const()[name = tensor("op_41849_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3555_cast_fp16 = mul(x = var_41848_cast_fp16, y = var_41849_to_fp16)[name = tensor("aw_3555_cast_fp16")]; + tensor var_41852_equation_0 = const()[name = tensor("op_41852_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_41852_cast_fp16 = einsum(equation = var_41852_equation_0, values = (var_41694_cast_fp16, var_41611_cast_fp16))[name = tensor("op_41852_cast_fp16")]; + tensor var_41853_to_fp16 = const()[name = tensor("op_41853_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3557_cast_fp16 = mul(x = var_41852_cast_fp16, y = var_41853_to_fp16)[name = tensor("aw_3557_cast_fp16")]; + tensor var_41856_equation_0 = const()[name = tensor("op_41856_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_41856_cast_fp16 = einsum(equation = var_41856_equation_0, values = (var_41698_cast_fp16, var_41615_cast_fp16))[name = tensor("op_41856_cast_fp16")]; + tensor var_41857_to_fp16 = const()[name = tensor("op_41857_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3559_cast_fp16 = mul(x = var_41856_cast_fp16, y = var_41857_to_fp16)[name = tensor("aw_3559_cast_fp16")]; + tensor var_41859_cast_fp16 = softmax(axis = var_30355, x = aw_3521_cast_fp16)[name = tensor("op_41859_cast_fp16")]; + tensor var_41860_cast_fp16 = softmax(axis = var_30355, x = aw_3523_cast_fp16)[name = tensor("op_41860_cast_fp16")]; + tensor var_41861_cast_fp16 = softmax(axis = var_30355, x = aw_3525_cast_fp16)[name = tensor("op_41861_cast_fp16")]; + tensor var_41862_cast_fp16 = softmax(axis = var_30355, x = aw_3527_cast_fp16)[name = tensor("op_41862_cast_fp16")]; + tensor var_41863_cast_fp16 = softmax(axis = var_30355, x = aw_3529_cast_fp16)[name = tensor("op_41863_cast_fp16")]; + tensor var_41864_cast_fp16 = softmax(axis = var_30355, x = aw_3531_cast_fp16)[name = tensor("op_41864_cast_fp16")]; + tensor var_41865_cast_fp16 = softmax(axis = var_30355, x = aw_3533_cast_fp16)[name = tensor("op_41865_cast_fp16")]; + tensor var_41866_cast_fp16 = softmax(axis = var_30355, x = aw_3535_cast_fp16)[name = tensor("op_41866_cast_fp16")]; + tensor var_41867_cast_fp16 = softmax(axis = var_30355, x = aw_3537_cast_fp16)[name = tensor("op_41867_cast_fp16")]; + tensor var_41868_cast_fp16 = softmax(axis = var_30355, x = aw_3539_cast_fp16)[name = tensor("op_41868_cast_fp16")]; + tensor var_41869_cast_fp16 = softmax(axis = var_30355, x = aw_3541_cast_fp16)[name = tensor("op_41869_cast_fp16")]; + tensor var_41870_cast_fp16 = softmax(axis = var_30355, x = aw_3543_cast_fp16)[name = tensor("op_41870_cast_fp16")]; + tensor var_41871_cast_fp16 = softmax(axis = var_30355, x = aw_3545_cast_fp16)[name = tensor("op_41871_cast_fp16")]; + tensor var_41872_cast_fp16 = softmax(axis = var_30355, x = aw_3547_cast_fp16)[name = tensor("op_41872_cast_fp16")]; + tensor var_41873_cast_fp16 = softmax(axis = var_30355, x = aw_3549_cast_fp16)[name = tensor("op_41873_cast_fp16")]; + tensor var_41874_cast_fp16 = softmax(axis = var_30355, x = aw_3551_cast_fp16)[name = tensor("op_41874_cast_fp16")]; + tensor var_41875_cast_fp16 = softmax(axis = var_30355, x = aw_3553_cast_fp16)[name = tensor("op_41875_cast_fp16")]; + tensor var_41876_cast_fp16 = softmax(axis = var_30355, x = aw_3555_cast_fp16)[name = tensor("op_41876_cast_fp16")]; + tensor var_41877_cast_fp16 = softmax(axis = var_30355, x = aw_3557_cast_fp16)[name = tensor("op_41877_cast_fp16")]; + tensor var_41878_cast_fp16 = softmax(axis = var_30355, x = aw_3559_cast_fp16)[name = tensor("op_41878_cast_fp16")]; + tensor var_41880_equation_0 = const()[name = tensor("op_41880_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_41880_cast_fp16 = einsum(equation = var_41880_equation_0, values = (var_41700_cast_fp16, var_41859_cast_fp16))[name = tensor("op_41880_cast_fp16")]; + tensor var_41882_equation_0 = const()[name = tensor("op_41882_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_41882_cast_fp16 = einsum(equation = var_41882_equation_0, values = (var_41704_cast_fp16, var_41860_cast_fp16))[name = tensor("op_41882_cast_fp16")]; + tensor var_41884_equation_0 = const()[name = tensor("op_41884_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_41884_cast_fp16 = einsum(equation = var_41884_equation_0, values = (var_41708_cast_fp16, var_41861_cast_fp16))[name = tensor("op_41884_cast_fp16")]; + tensor var_41886_equation_0 = const()[name = tensor("op_41886_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_41886_cast_fp16 = einsum(equation = var_41886_equation_0, values = (var_41712_cast_fp16, var_41862_cast_fp16))[name = tensor("op_41886_cast_fp16")]; + tensor var_41888_equation_0 = const()[name = tensor("op_41888_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_41888_cast_fp16 = einsum(equation = var_41888_equation_0, values = (var_41716_cast_fp16, var_41863_cast_fp16))[name = tensor("op_41888_cast_fp16")]; + tensor var_41890_equation_0 = const()[name = tensor("op_41890_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_41890_cast_fp16 = einsum(equation = var_41890_equation_0, values = (var_41720_cast_fp16, var_41864_cast_fp16))[name = tensor("op_41890_cast_fp16")]; + tensor var_41892_equation_0 = const()[name = tensor("op_41892_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_41892_cast_fp16 = einsum(equation = var_41892_equation_0, values = (var_41724_cast_fp16, var_41865_cast_fp16))[name = tensor("op_41892_cast_fp16")]; + tensor var_41894_equation_0 = const()[name = tensor("op_41894_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_41894_cast_fp16 = einsum(equation = var_41894_equation_0, values = (var_41728_cast_fp16, var_41866_cast_fp16))[name = tensor("op_41894_cast_fp16")]; + tensor var_41896_equation_0 = const()[name = tensor("op_41896_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_41896_cast_fp16 = einsum(equation = var_41896_equation_0, values = (var_41732_cast_fp16, var_41867_cast_fp16))[name = tensor("op_41896_cast_fp16")]; + tensor var_41898_equation_0 = const()[name = tensor("op_41898_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_41898_cast_fp16 = einsum(equation = var_41898_equation_0, values = (var_41736_cast_fp16, var_41868_cast_fp16))[name = tensor("op_41898_cast_fp16")]; + tensor var_41900_equation_0 = const()[name = tensor("op_41900_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_41900_cast_fp16 = einsum(equation = var_41900_equation_0, values = (var_41740_cast_fp16, var_41869_cast_fp16))[name = tensor("op_41900_cast_fp16")]; + tensor var_41902_equation_0 = const()[name = tensor("op_41902_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_41902_cast_fp16 = einsum(equation = var_41902_equation_0, values = (var_41744_cast_fp16, var_41870_cast_fp16))[name = tensor("op_41902_cast_fp16")]; + tensor var_41904_equation_0 = const()[name = tensor("op_41904_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_41904_cast_fp16 = einsum(equation = var_41904_equation_0, values = (var_41748_cast_fp16, var_41871_cast_fp16))[name = tensor("op_41904_cast_fp16")]; + tensor var_41906_equation_0 = const()[name = tensor("op_41906_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_41906_cast_fp16 = einsum(equation = var_41906_equation_0, values = (var_41752_cast_fp16, var_41872_cast_fp16))[name = tensor("op_41906_cast_fp16")]; + tensor var_41908_equation_0 = const()[name = tensor("op_41908_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_41908_cast_fp16 = einsum(equation = var_41908_equation_0, values = (var_41756_cast_fp16, var_41873_cast_fp16))[name = tensor("op_41908_cast_fp16")]; + tensor var_41910_equation_0 = const()[name = tensor("op_41910_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_41910_cast_fp16 = einsum(equation = var_41910_equation_0, values = (var_41760_cast_fp16, var_41874_cast_fp16))[name = tensor("op_41910_cast_fp16")]; + tensor var_41912_equation_0 = const()[name = tensor("op_41912_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_41912_cast_fp16 = einsum(equation = var_41912_equation_0, values = (var_41764_cast_fp16, var_41875_cast_fp16))[name = tensor("op_41912_cast_fp16")]; + tensor var_41914_equation_0 = const()[name = tensor("op_41914_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_41914_cast_fp16 = einsum(equation = var_41914_equation_0, values = (var_41768_cast_fp16, var_41876_cast_fp16))[name = tensor("op_41914_cast_fp16")]; + tensor var_41916_equation_0 = const()[name = tensor("op_41916_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_41916_cast_fp16 = einsum(equation = var_41916_equation_0, values = (var_41772_cast_fp16, var_41877_cast_fp16))[name = tensor("op_41916_cast_fp16")]; + tensor var_41918_equation_0 = const()[name = tensor("op_41918_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_41918_cast_fp16 = einsum(equation = var_41918_equation_0, values = (var_41776_cast_fp16, var_41878_cast_fp16))[name = tensor("op_41918_cast_fp16")]; + tensor input_553_interleave_0 = const()[name = tensor("input_553_interleave_0"), val = tensor(false)]; + tensor input_553_cast_fp16 = concat(axis = var_30355, interleave = input_553_interleave_0, values = (var_41880_cast_fp16, var_41882_cast_fp16, var_41884_cast_fp16, var_41886_cast_fp16, var_41888_cast_fp16, var_41890_cast_fp16, var_41892_cast_fp16, var_41894_cast_fp16, var_41896_cast_fp16, var_41898_cast_fp16, var_41900_cast_fp16, var_41902_cast_fp16, var_41904_cast_fp16, var_41906_cast_fp16, var_41908_cast_fp16, var_41910_cast_fp16, var_41912_cast_fp16, var_41914_cast_fp16, var_41916_cast_fp16, var_41918_cast_fp16))[name = tensor("input_553_cast_fp16")]; + tensor var_41924 = const()[name = tensor("op_41924"), val = tensor([1, 1])]; + tensor var_41926 = const()[name = tensor("op_41926"), val = tensor([1, 1])]; + tensor var_41928_pad_type_0 = const()[name = tensor("op_41928_pad_type_0"), val = tensor("custom")]; + tensor var_41928_pad_0 = const()[name = tensor("op_41928_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_2_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_2_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3537307328)))]; + tensor up_blocks_0_attentions_1_transformer_blocks_2_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_2_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3540584192)))]; + tensor var_41928_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_2_attn1_to_out_0_bias_to_fp16, dilations = var_41926, groups = var_30355, pad = var_41928_pad_0, pad_type = var_41928_pad_type_0, strides = var_41924, weight = up_blocks_0_attentions_1_transformer_blocks_2_attn1_to_out_0_weight_to_fp16, x = input_553_cast_fp16)[name = tensor("op_41928_cast_fp16")]; + tensor inputs_279_cast_fp16 = add(x = var_41928_cast_fp16, y = inputs_277_cast_fp16)[name = tensor("inputs_279_cast_fp16")]; + tensor hidden_states_375_axes_0 = const()[name = tensor("hidden_states_375_axes_0"), val = tensor([1])]; + tensor hidden_states_375_gamma_0_to_fp16 = const()[name = tensor("hidden_states_375_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3540586816)))]; + tensor hidden_states_375_beta_0_to_fp16 = const()[name = tensor("hidden_states_375_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3540589440)))]; + tensor var_41938_to_fp16 = const()[name = tensor("op_41938_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_375_cast_fp16 = layer_norm(axes = hidden_states_375_axes_0, beta = hidden_states_375_beta_0_to_fp16, epsilon = var_41938_to_fp16, gamma = hidden_states_375_gamma_0_to_fp16, x = inputs_279_cast_fp16)[name = tensor("hidden_states_375_cast_fp16")]; + tensor var_41953 = const()[name = tensor("op_41953"), val = tensor([1, 1])]; + tensor var_41955 = const()[name = tensor("op_41955"), val = tensor([1, 1])]; + tensor q_187_pad_type_0 = const()[name = tensor("q_187_pad_type_0"), val = tensor("custom")]; + tensor q_187_pad_0 = const()[name = tensor("q_187_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_2_attn2_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_2_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3540592064)))]; + tensor q_187_cast_fp16 = conv(dilations = var_41955, groups = var_30355, pad = q_187_pad_0, pad_type = q_187_pad_type_0, strides = var_41953, weight = up_blocks_0_attentions_1_transformer_blocks_2_attn2_to_q_weight_to_fp16, x = hidden_states_375_cast_fp16)[name = tensor("q_187_cast_fp16")]; + tensor var_41959 = const()[name = tensor("op_41959"), val = tensor([1, 1])]; + tensor var_41961 = const()[name = tensor("op_41961"), val = tensor([1, 1])]; + tensor k_373_pad_type_0 = const()[name = tensor("k_373_pad_type_0"), val = tensor("custom")]; + tensor k_373_pad_0 = const()[name = tensor("k_373_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_2_attn2_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_2_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3543868928)))]; + tensor k_373_cast_fp16 = conv(dilations = var_41961, groups = var_30355, pad = k_373_pad_0, pad_type = k_373_pad_type_0, strides = var_41959, weight = up_blocks_0_attentions_1_transformer_blocks_2_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_373_cast_fp16")]; + tensor var_41965 = const()[name = tensor("op_41965"), val = tensor([1, 1])]; + tensor var_41967 = const()[name = tensor("op_41967"), val = tensor([1, 1])]; + tensor v_187_pad_type_0 = const()[name = tensor("v_187_pad_type_0"), val = tensor("custom")]; + tensor v_187_pad_0 = const()[name = tensor("v_187_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_2_attn2_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_2_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3549111872)))]; + tensor v_187_cast_fp16 = conv(dilations = var_41967, groups = var_30355, pad = v_187_pad_0, pad_type = v_187_pad_type_0, strides = var_41965, weight = up_blocks_0_attentions_1_transformer_blocks_2_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_187_cast_fp16")]; + tensor var_41971_begin_0 = const()[name = tensor("op_41971_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_41971_end_0 = const()[name = tensor("op_41971_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_41971_end_mask_0 = const()[name = tensor("op_41971_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41971_cast_fp16 = slice_by_index(begin = var_41971_begin_0, end = var_41971_end_0, end_mask = var_41971_end_mask_0, x = q_187_cast_fp16)[name = tensor("op_41971_cast_fp16")]; + tensor var_41975_begin_0 = const()[name = tensor("op_41975_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_41975_end_0 = const()[name = tensor("op_41975_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_41975_end_mask_0 = const()[name = tensor("op_41975_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41975_cast_fp16 = slice_by_index(begin = var_41975_begin_0, end = var_41975_end_0, end_mask = var_41975_end_mask_0, x = q_187_cast_fp16)[name = tensor("op_41975_cast_fp16")]; + tensor var_41979_begin_0 = const()[name = tensor("op_41979_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_41979_end_0 = const()[name = tensor("op_41979_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_41979_end_mask_0 = const()[name = tensor("op_41979_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41979_cast_fp16 = slice_by_index(begin = var_41979_begin_0, end = var_41979_end_0, end_mask = var_41979_end_mask_0, x = q_187_cast_fp16)[name = tensor("op_41979_cast_fp16")]; + tensor var_41983_begin_0 = const()[name = tensor("op_41983_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_41983_end_0 = const()[name = tensor("op_41983_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_41983_end_mask_0 = const()[name = tensor("op_41983_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41983_cast_fp16 = slice_by_index(begin = var_41983_begin_0, end = var_41983_end_0, end_mask = var_41983_end_mask_0, x = q_187_cast_fp16)[name = tensor("op_41983_cast_fp16")]; + tensor var_41987_begin_0 = const()[name = tensor("op_41987_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_41987_end_0 = const()[name = tensor("op_41987_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_41987_end_mask_0 = const()[name = tensor("op_41987_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41987_cast_fp16 = slice_by_index(begin = var_41987_begin_0, end = var_41987_end_0, end_mask = var_41987_end_mask_0, x = q_187_cast_fp16)[name = tensor("op_41987_cast_fp16")]; + tensor var_41991_begin_0 = const()[name = tensor("op_41991_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_41991_end_0 = const()[name = tensor("op_41991_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_41991_end_mask_0 = const()[name = tensor("op_41991_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41991_cast_fp16 = slice_by_index(begin = var_41991_begin_0, end = var_41991_end_0, end_mask = var_41991_end_mask_0, x = q_187_cast_fp16)[name = tensor("op_41991_cast_fp16")]; + tensor var_41995_begin_0 = const()[name = tensor("op_41995_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_41995_end_0 = const()[name = tensor("op_41995_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_41995_end_mask_0 = const()[name = tensor("op_41995_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41995_cast_fp16 = slice_by_index(begin = var_41995_begin_0, end = var_41995_end_0, end_mask = var_41995_end_mask_0, x = q_187_cast_fp16)[name = tensor("op_41995_cast_fp16")]; + tensor var_41999_begin_0 = const()[name = tensor("op_41999_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_41999_end_0 = const()[name = tensor("op_41999_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_41999_end_mask_0 = const()[name = tensor("op_41999_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41999_cast_fp16 = slice_by_index(begin = var_41999_begin_0, end = var_41999_end_0, end_mask = var_41999_end_mask_0, x = q_187_cast_fp16)[name = tensor("op_41999_cast_fp16")]; + tensor var_42003_begin_0 = const()[name = tensor("op_42003_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_42003_end_0 = const()[name = tensor("op_42003_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_42003_end_mask_0 = const()[name = tensor("op_42003_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42003_cast_fp16 = slice_by_index(begin = var_42003_begin_0, end = var_42003_end_0, end_mask = var_42003_end_mask_0, x = q_187_cast_fp16)[name = tensor("op_42003_cast_fp16")]; + tensor var_42007_begin_0 = const()[name = tensor("op_42007_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_42007_end_0 = const()[name = tensor("op_42007_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_42007_end_mask_0 = const()[name = tensor("op_42007_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42007_cast_fp16 = slice_by_index(begin = var_42007_begin_0, end = var_42007_end_0, end_mask = var_42007_end_mask_0, x = q_187_cast_fp16)[name = tensor("op_42007_cast_fp16")]; + tensor var_42011_begin_0 = const()[name = tensor("op_42011_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_42011_end_0 = const()[name = tensor("op_42011_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_42011_end_mask_0 = const()[name = tensor("op_42011_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42011_cast_fp16 = slice_by_index(begin = var_42011_begin_0, end = var_42011_end_0, end_mask = var_42011_end_mask_0, x = q_187_cast_fp16)[name = tensor("op_42011_cast_fp16")]; + tensor var_42015_begin_0 = const()[name = tensor("op_42015_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_42015_end_0 = const()[name = tensor("op_42015_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_42015_end_mask_0 = const()[name = tensor("op_42015_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42015_cast_fp16 = slice_by_index(begin = var_42015_begin_0, end = var_42015_end_0, end_mask = var_42015_end_mask_0, x = q_187_cast_fp16)[name = tensor("op_42015_cast_fp16")]; + tensor var_42019_begin_0 = const()[name = tensor("op_42019_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_42019_end_0 = const()[name = tensor("op_42019_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_42019_end_mask_0 = const()[name = tensor("op_42019_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42019_cast_fp16 = slice_by_index(begin = var_42019_begin_0, end = var_42019_end_0, end_mask = var_42019_end_mask_0, x = q_187_cast_fp16)[name = tensor("op_42019_cast_fp16")]; + tensor var_42023_begin_0 = const()[name = tensor("op_42023_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_42023_end_0 = const()[name = tensor("op_42023_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_42023_end_mask_0 = const()[name = tensor("op_42023_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42023_cast_fp16 = slice_by_index(begin = var_42023_begin_0, end = var_42023_end_0, end_mask = var_42023_end_mask_0, x = q_187_cast_fp16)[name = tensor("op_42023_cast_fp16")]; + tensor var_42027_begin_0 = const()[name = tensor("op_42027_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_42027_end_0 = const()[name = tensor("op_42027_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_42027_end_mask_0 = const()[name = tensor("op_42027_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42027_cast_fp16 = slice_by_index(begin = var_42027_begin_0, end = var_42027_end_0, end_mask = var_42027_end_mask_0, x = q_187_cast_fp16)[name = tensor("op_42027_cast_fp16")]; + tensor var_42031_begin_0 = const()[name = tensor("op_42031_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_42031_end_0 = const()[name = tensor("op_42031_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_42031_end_mask_0 = const()[name = tensor("op_42031_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42031_cast_fp16 = slice_by_index(begin = var_42031_begin_0, end = var_42031_end_0, end_mask = var_42031_end_mask_0, x = q_187_cast_fp16)[name = tensor("op_42031_cast_fp16")]; + tensor var_42035_begin_0 = const()[name = tensor("op_42035_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_42035_end_0 = const()[name = tensor("op_42035_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_42035_end_mask_0 = const()[name = tensor("op_42035_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42035_cast_fp16 = slice_by_index(begin = var_42035_begin_0, end = var_42035_end_0, end_mask = var_42035_end_mask_0, x = q_187_cast_fp16)[name = tensor("op_42035_cast_fp16")]; + tensor var_42039_begin_0 = const()[name = tensor("op_42039_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_42039_end_0 = const()[name = tensor("op_42039_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_42039_end_mask_0 = const()[name = tensor("op_42039_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42039_cast_fp16 = slice_by_index(begin = var_42039_begin_0, end = var_42039_end_0, end_mask = var_42039_end_mask_0, x = q_187_cast_fp16)[name = tensor("op_42039_cast_fp16")]; + tensor var_42043_begin_0 = const()[name = tensor("op_42043_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_42043_end_0 = const()[name = tensor("op_42043_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_42043_end_mask_0 = const()[name = tensor("op_42043_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42043_cast_fp16 = slice_by_index(begin = var_42043_begin_0, end = var_42043_end_0, end_mask = var_42043_end_mask_0, x = q_187_cast_fp16)[name = tensor("op_42043_cast_fp16")]; + tensor var_42047_begin_0 = const()[name = tensor("op_42047_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_42047_end_0 = const()[name = tensor("op_42047_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_42047_end_mask_0 = const()[name = tensor("op_42047_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42047_cast_fp16 = slice_by_index(begin = var_42047_begin_0, end = var_42047_end_0, end_mask = var_42047_end_mask_0, x = q_187_cast_fp16)[name = tensor("op_42047_cast_fp16")]; + tensor k_375_perm_0 = const()[name = tensor("k_375_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_42054_begin_0 = const()[name = tensor("op_42054_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_42054_end_0 = const()[name = tensor("op_42054_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_42054_end_mask_0 = const()[name = tensor("op_42054_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_375_cast_fp16 = transpose(perm = k_375_perm_0, x = k_373_cast_fp16)[name = tensor("transpose_46")]; + tensor var_42054_cast_fp16 = slice_by_index(begin = var_42054_begin_0, end = var_42054_end_0, end_mask = var_42054_end_mask_0, x = k_375_cast_fp16)[name = tensor("op_42054_cast_fp16")]; + tensor var_42058_begin_0 = const()[name = tensor("op_42058_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_42058_end_0 = const()[name = tensor("op_42058_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_42058_end_mask_0 = const()[name = tensor("op_42058_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42058_cast_fp16 = slice_by_index(begin = var_42058_begin_0, end = var_42058_end_0, end_mask = var_42058_end_mask_0, x = k_375_cast_fp16)[name = tensor("op_42058_cast_fp16")]; + tensor var_42062_begin_0 = const()[name = tensor("op_42062_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_42062_end_0 = const()[name = tensor("op_42062_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_42062_end_mask_0 = const()[name = tensor("op_42062_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42062_cast_fp16 = slice_by_index(begin = var_42062_begin_0, end = var_42062_end_0, end_mask = var_42062_end_mask_0, x = k_375_cast_fp16)[name = tensor("op_42062_cast_fp16")]; + tensor var_42066_begin_0 = const()[name = tensor("op_42066_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_42066_end_0 = const()[name = tensor("op_42066_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_42066_end_mask_0 = const()[name = tensor("op_42066_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42066_cast_fp16 = slice_by_index(begin = var_42066_begin_0, end = var_42066_end_0, end_mask = var_42066_end_mask_0, x = k_375_cast_fp16)[name = tensor("op_42066_cast_fp16")]; + tensor var_42070_begin_0 = const()[name = tensor("op_42070_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_42070_end_0 = const()[name = tensor("op_42070_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_42070_end_mask_0 = const()[name = tensor("op_42070_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42070_cast_fp16 = slice_by_index(begin = var_42070_begin_0, end = var_42070_end_0, end_mask = var_42070_end_mask_0, x = k_375_cast_fp16)[name = tensor("op_42070_cast_fp16")]; + tensor var_42074_begin_0 = const()[name = tensor("op_42074_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_42074_end_0 = const()[name = tensor("op_42074_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_42074_end_mask_0 = const()[name = tensor("op_42074_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42074_cast_fp16 = slice_by_index(begin = var_42074_begin_0, end = var_42074_end_0, end_mask = var_42074_end_mask_0, x = k_375_cast_fp16)[name = tensor("op_42074_cast_fp16")]; + tensor var_42078_begin_0 = const()[name = tensor("op_42078_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_42078_end_0 = const()[name = tensor("op_42078_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_42078_end_mask_0 = const()[name = tensor("op_42078_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42078_cast_fp16 = slice_by_index(begin = var_42078_begin_0, end = var_42078_end_0, end_mask = var_42078_end_mask_0, x = k_375_cast_fp16)[name = tensor("op_42078_cast_fp16")]; + tensor var_42082_begin_0 = const()[name = tensor("op_42082_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_42082_end_0 = const()[name = tensor("op_42082_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_42082_end_mask_0 = const()[name = tensor("op_42082_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42082_cast_fp16 = slice_by_index(begin = var_42082_begin_0, end = var_42082_end_0, end_mask = var_42082_end_mask_0, x = k_375_cast_fp16)[name = tensor("op_42082_cast_fp16")]; + tensor var_42086_begin_0 = const()[name = tensor("op_42086_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_42086_end_0 = const()[name = tensor("op_42086_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_42086_end_mask_0 = const()[name = tensor("op_42086_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42086_cast_fp16 = slice_by_index(begin = var_42086_begin_0, end = var_42086_end_0, end_mask = var_42086_end_mask_0, x = k_375_cast_fp16)[name = tensor("op_42086_cast_fp16")]; + tensor var_42090_begin_0 = const()[name = tensor("op_42090_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_42090_end_0 = const()[name = tensor("op_42090_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_42090_end_mask_0 = const()[name = tensor("op_42090_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42090_cast_fp16 = slice_by_index(begin = var_42090_begin_0, end = var_42090_end_0, end_mask = var_42090_end_mask_0, x = k_375_cast_fp16)[name = tensor("op_42090_cast_fp16")]; + tensor var_42094_begin_0 = const()[name = tensor("op_42094_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_42094_end_0 = const()[name = tensor("op_42094_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_42094_end_mask_0 = const()[name = tensor("op_42094_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42094_cast_fp16 = slice_by_index(begin = var_42094_begin_0, end = var_42094_end_0, end_mask = var_42094_end_mask_0, x = k_375_cast_fp16)[name = tensor("op_42094_cast_fp16")]; + tensor var_42098_begin_0 = const()[name = tensor("op_42098_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_42098_end_0 = const()[name = tensor("op_42098_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_42098_end_mask_0 = const()[name = tensor("op_42098_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42098_cast_fp16 = slice_by_index(begin = var_42098_begin_0, end = var_42098_end_0, end_mask = var_42098_end_mask_0, x = k_375_cast_fp16)[name = tensor("op_42098_cast_fp16")]; + tensor var_42102_begin_0 = const()[name = tensor("op_42102_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_42102_end_0 = const()[name = tensor("op_42102_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_42102_end_mask_0 = const()[name = tensor("op_42102_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42102_cast_fp16 = slice_by_index(begin = var_42102_begin_0, end = var_42102_end_0, end_mask = var_42102_end_mask_0, x = k_375_cast_fp16)[name = tensor("op_42102_cast_fp16")]; + tensor var_42106_begin_0 = const()[name = tensor("op_42106_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_42106_end_0 = const()[name = tensor("op_42106_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_42106_end_mask_0 = const()[name = tensor("op_42106_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42106_cast_fp16 = slice_by_index(begin = var_42106_begin_0, end = var_42106_end_0, end_mask = var_42106_end_mask_0, x = k_375_cast_fp16)[name = tensor("op_42106_cast_fp16")]; + tensor var_42110_begin_0 = const()[name = tensor("op_42110_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_42110_end_0 = const()[name = tensor("op_42110_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_42110_end_mask_0 = const()[name = tensor("op_42110_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42110_cast_fp16 = slice_by_index(begin = var_42110_begin_0, end = var_42110_end_0, end_mask = var_42110_end_mask_0, x = k_375_cast_fp16)[name = tensor("op_42110_cast_fp16")]; + tensor var_42114_begin_0 = const()[name = tensor("op_42114_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_42114_end_0 = const()[name = tensor("op_42114_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_42114_end_mask_0 = const()[name = tensor("op_42114_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42114_cast_fp16 = slice_by_index(begin = var_42114_begin_0, end = var_42114_end_0, end_mask = var_42114_end_mask_0, x = k_375_cast_fp16)[name = tensor("op_42114_cast_fp16")]; + tensor var_42118_begin_0 = const()[name = tensor("op_42118_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_42118_end_0 = const()[name = tensor("op_42118_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_42118_end_mask_0 = const()[name = tensor("op_42118_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42118_cast_fp16 = slice_by_index(begin = var_42118_begin_0, end = var_42118_end_0, end_mask = var_42118_end_mask_0, x = k_375_cast_fp16)[name = tensor("op_42118_cast_fp16")]; + tensor var_42122_begin_0 = const()[name = tensor("op_42122_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_42122_end_0 = const()[name = tensor("op_42122_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_42122_end_mask_0 = const()[name = tensor("op_42122_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42122_cast_fp16 = slice_by_index(begin = var_42122_begin_0, end = var_42122_end_0, end_mask = var_42122_end_mask_0, x = k_375_cast_fp16)[name = tensor("op_42122_cast_fp16")]; + tensor var_42126_begin_0 = const()[name = tensor("op_42126_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_42126_end_0 = const()[name = tensor("op_42126_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_42126_end_mask_0 = const()[name = tensor("op_42126_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42126_cast_fp16 = slice_by_index(begin = var_42126_begin_0, end = var_42126_end_0, end_mask = var_42126_end_mask_0, x = k_375_cast_fp16)[name = tensor("op_42126_cast_fp16")]; + tensor var_42130_begin_0 = const()[name = tensor("op_42130_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_42130_end_0 = const()[name = tensor("op_42130_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_42130_end_mask_0 = const()[name = tensor("op_42130_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42130_cast_fp16 = slice_by_index(begin = var_42130_begin_0, end = var_42130_end_0, end_mask = var_42130_end_mask_0, x = k_375_cast_fp16)[name = tensor("op_42130_cast_fp16")]; + tensor var_42132_begin_0 = const()[name = tensor("op_42132_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_42132_end_0 = const()[name = tensor("op_42132_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_42132_end_mask_0 = const()[name = tensor("op_42132_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42132_cast_fp16 = slice_by_index(begin = var_42132_begin_0, end = var_42132_end_0, end_mask = var_42132_end_mask_0, x = v_187_cast_fp16)[name = tensor("op_42132_cast_fp16")]; + tensor var_42136_begin_0 = const()[name = tensor("op_42136_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_42136_end_0 = const()[name = tensor("op_42136_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_42136_end_mask_0 = const()[name = tensor("op_42136_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42136_cast_fp16 = slice_by_index(begin = var_42136_begin_0, end = var_42136_end_0, end_mask = var_42136_end_mask_0, x = v_187_cast_fp16)[name = tensor("op_42136_cast_fp16")]; + tensor var_42140_begin_0 = const()[name = tensor("op_42140_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_42140_end_0 = const()[name = tensor("op_42140_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_42140_end_mask_0 = const()[name = tensor("op_42140_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42140_cast_fp16 = slice_by_index(begin = var_42140_begin_0, end = var_42140_end_0, end_mask = var_42140_end_mask_0, x = v_187_cast_fp16)[name = tensor("op_42140_cast_fp16")]; + tensor var_42144_begin_0 = const()[name = tensor("op_42144_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_42144_end_0 = const()[name = tensor("op_42144_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_42144_end_mask_0 = const()[name = tensor("op_42144_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42144_cast_fp16 = slice_by_index(begin = var_42144_begin_0, end = var_42144_end_0, end_mask = var_42144_end_mask_0, x = v_187_cast_fp16)[name = tensor("op_42144_cast_fp16")]; + tensor var_42148_begin_0 = const()[name = tensor("op_42148_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_42148_end_0 = const()[name = tensor("op_42148_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_42148_end_mask_0 = const()[name = tensor("op_42148_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42148_cast_fp16 = slice_by_index(begin = var_42148_begin_0, end = var_42148_end_0, end_mask = var_42148_end_mask_0, x = v_187_cast_fp16)[name = tensor("op_42148_cast_fp16")]; + tensor var_42152_begin_0 = const()[name = tensor("op_42152_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_42152_end_0 = const()[name = tensor("op_42152_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_42152_end_mask_0 = const()[name = tensor("op_42152_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42152_cast_fp16 = slice_by_index(begin = var_42152_begin_0, end = var_42152_end_0, end_mask = var_42152_end_mask_0, x = v_187_cast_fp16)[name = tensor("op_42152_cast_fp16")]; + tensor var_42156_begin_0 = const()[name = tensor("op_42156_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_42156_end_0 = const()[name = tensor("op_42156_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_42156_end_mask_0 = const()[name = tensor("op_42156_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42156_cast_fp16 = slice_by_index(begin = var_42156_begin_0, end = var_42156_end_0, end_mask = var_42156_end_mask_0, x = v_187_cast_fp16)[name = tensor("op_42156_cast_fp16")]; + tensor var_42160_begin_0 = const()[name = tensor("op_42160_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_42160_end_0 = const()[name = tensor("op_42160_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_42160_end_mask_0 = const()[name = tensor("op_42160_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42160_cast_fp16 = slice_by_index(begin = var_42160_begin_0, end = var_42160_end_0, end_mask = var_42160_end_mask_0, x = v_187_cast_fp16)[name = tensor("op_42160_cast_fp16")]; + tensor var_42164_begin_0 = const()[name = tensor("op_42164_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_42164_end_0 = const()[name = tensor("op_42164_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_42164_end_mask_0 = const()[name = tensor("op_42164_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42164_cast_fp16 = slice_by_index(begin = var_42164_begin_0, end = var_42164_end_0, end_mask = var_42164_end_mask_0, x = v_187_cast_fp16)[name = tensor("op_42164_cast_fp16")]; + tensor var_42168_begin_0 = const()[name = tensor("op_42168_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_42168_end_0 = const()[name = tensor("op_42168_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_42168_end_mask_0 = const()[name = tensor("op_42168_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42168_cast_fp16 = slice_by_index(begin = var_42168_begin_0, end = var_42168_end_0, end_mask = var_42168_end_mask_0, x = v_187_cast_fp16)[name = tensor("op_42168_cast_fp16")]; + tensor var_42172_begin_0 = const()[name = tensor("op_42172_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_42172_end_0 = const()[name = tensor("op_42172_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_42172_end_mask_0 = const()[name = tensor("op_42172_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42172_cast_fp16 = slice_by_index(begin = var_42172_begin_0, end = var_42172_end_0, end_mask = var_42172_end_mask_0, x = v_187_cast_fp16)[name = tensor("op_42172_cast_fp16")]; + tensor var_42176_begin_0 = const()[name = tensor("op_42176_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_42176_end_0 = const()[name = tensor("op_42176_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_42176_end_mask_0 = const()[name = tensor("op_42176_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42176_cast_fp16 = slice_by_index(begin = var_42176_begin_0, end = var_42176_end_0, end_mask = var_42176_end_mask_0, x = v_187_cast_fp16)[name = tensor("op_42176_cast_fp16")]; + tensor var_42180_begin_0 = const()[name = tensor("op_42180_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_42180_end_0 = const()[name = tensor("op_42180_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_42180_end_mask_0 = const()[name = tensor("op_42180_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42180_cast_fp16 = slice_by_index(begin = var_42180_begin_0, end = var_42180_end_0, end_mask = var_42180_end_mask_0, x = v_187_cast_fp16)[name = tensor("op_42180_cast_fp16")]; + tensor var_42184_begin_0 = const()[name = tensor("op_42184_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_42184_end_0 = const()[name = tensor("op_42184_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_42184_end_mask_0 = const()[name = tensor("op_42184_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42184_cast_fp16 = slice_by_index(begin = var_42184_begin_0, end = var_42184_end_0, end_mask = var_42184_end_mask_0, x = v_187_cast_fp16)[name = tensor("op_42184_cast_fp16")]; + tensor var_42188_begin_0 = const()[name = tensor("op_42188_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_42188_end_0 = const()[name = tensor("op_42188_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_42188_end_mask_0 = const()[name = tensor("op_42188_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42188_cast_fp16 = slice_by_index(begin = var_42188_begin_0, end = var_42188_end_0, end_mask = var_42188_end_mask_0, x = v_187_cast_fp16)[name = tensor("op_42188_cast_fp16")]; + tensor var_42192_begin_0 = const()[name = tensor("op_42192_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_42192_end_0 = const()[name = tensor("op_42192_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_42192_end_mask_0 = const()[name = tensor("op_42192_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42192_cast_fp16 = slice_by_index(begin = var_42192_begin_0, end = var_42192_end_0, end_mask = var_42192_end_mask_0, x = v_187_cast_fp16)[name = tensor("op_42192_cast_fp16")]; + tensor var_42196_begin_0 = const()[name = tensor("op_42196_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_42196_end_0 = const()[name = tensor("op_42196_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_42196_end_mask_0 = const()[name = tensor("op_42196_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42196_cast_fp16 = slice_by_index(begin = var_42196_begin_0, end = var_42196_end_0, end_mask = var_42196_end_mask_0, x = v_187_cast_fp16)[name = tensor("op_42196_cast_fp16")]; + tensor var_42200_begin_0 = const()[name = tensor("op_42200_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_42200_end_0 = const()[name = tensor("op_42200_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_42200_end_mask_0 = const()[name = tensor("op_42200_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42200_cast_fp16 = slice_by_index(begin = var_42200_begin_0, end = var_42200_end_0, end_mask = var_42200_end_mask_0, x = v_187_cast_fp16)[name = tensor("op_42200_cast_fp16")]; + tensor var_42204_begin_0 = const()[name = tensor("op_42204_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_42204_end_0 = const()[name = tensor("op_42204_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_42204_end_mask_0 = const()[name = tensor("op_42204_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42204_cast_fp16 = slice_by_index(begin = var_42204_begin_0, end = var_42204_end_0, end_mask = var_42204_end_mask_0, x = v_187_cast_fp16)[name = tensor("op_42204_cast_fp16")]; + tensor var_42208_begin_0 = const()[name = tensor("op_42208_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_42208_end_0 = const()[name = tensor("op_42208_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_42208_end_mask_0 = const()[name = tensor("op_42208_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42208_cast_fp16 = slice_by_index(begin = var_42208_begin_0, end = var_42208_end_0, end_mask = var_42208_end_mask_0, x = v_187_cast_fp16)[name = tensor("op_42208_cast_fp16")]; + tensor var_42212_equation_0 = const()[name = tensor("op_42212_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42212_cast_fp16 = einsum(equation = var_42212_equation_0, values = (var_42054_cast_fp16, var_41971_cast_fp16))[name = tensor("op_42212_cast_fp16")]; + tensor var_42213_to_fp16 = const()[name = tensor("op_42213_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3561_cast_fp16 = mul(x = var_42212_cast_fp16, y = var_42213_to_fp16)[name = tensor("aw_3561_cast_fp16")]; + tensor var_42216_equation_0 = const()[name = tensor("op_42216_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42216_cast_fp16 = einsum(equation = var_42216_equation_0, values = (var_42058_cast_fp16, var_41975_cast_fp16))[name = tensor("op_42216_cast_fp16")]; + tensor var_42217_to_fp16 = const()[name = tensor("op_42217_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3563_cast_fp16 = mul(x = var_42216_cast_fp16, y = var_42217_to_fp16)[name = tensor("aw_3563_cast_fp16")]; + tensor var_42220_equation_0 = const()[name = tensor("op_42220_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42220_cast_fp16 = einsum(equation = var_42220_equation_0, values = (var_42062_cast_fp16, var_41979_cast_fp16))[name = tensor("op_42220_cast_fp16")]; + tensor var_42221_to_fp16 = const()[name = tensor("op_42221_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3565_cast_fp16 = mul(x = var_42220_cast_fp16, y = var_42221_to_fp16)[name = tensor("aw_3565_cast_fp16")]; + tensor var_42224_equation_0 = const()[name = tensor("op_42224_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42224_cast_fp16 = einsum(equation = var_42224_equation_0, values = (var_42066_cast_fp16, var_41983_cast_fp16))[name = tensor("op_42224_cast_fp16")]; + tensor var_42225_to_fp16 = const()[name = tensor("op_42225_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3567_cast_fp16 = mul(x = var_42224_cast_fp16, y = var_42225_to_fp16)[name = tensor("aw_3567_cast_fp16")]; + tensor var_42228_equation_0 = const()[name = tensor("op_42228_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42228_cast_fp16 = einsum(equation = var_42228_equation_0, values = (var_42070_cast_fp16, var_41987_cast_fp16))[name = tensor("op_42228_cast_fp16")]; + tensor var_42229_to_fp16 = const()[name = tensor("op_42229_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3569_cast_fp16 = mul(x = var_42228_cast_fp16, y = var_42229_to_fp16)[name = tensor("aw_3569_cast_fp16")]; + tensor var_42232_equation_0 = const()[name = tensor("op_42232_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42232_cast_fp16 = einsum(equation = var_42232_equation_0, values = (var_42074_cast_fp16, var_41991_cast_fp16))[name = tensor("op_42232_cast_fp16")]; + tensor var_42233_to_fp16 = const()[name = tensor("op_42233_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3571_cast_fp16 = mul(x = var_42232_cast_fp16, y = var_42233_to_fp16)[name = tensor("aw_3571_cast_fp16")]; + tensor var_42236_equation_0 = const()[name = tensor("op_42236_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42236_cast_fp16 = einsum(equation = var_42236_equation_0, values = (var_42078_cast_fp16, var_41995_cast_fp16))[name = tensor("op_42236_cast_fp16")]; + tensor var_42237_to_fp16 = const()[name = tensor("op_42237_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3573_cast_fp16 = mul(x = var_42236_cast_fp16, y = var_42237_to_fp16)[name = tensor("aw_3573_cast_fp16")]; + tensor var_42240_equation_0 = const()[name = tensor("op_42240_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42240_cast_fp16 = einsum(equation = var_42240_equation_0, values = (var_42082_cast_fp16, var_41999_cast_fp16))[name = tensor("op_42240_cast_fp16")]; + tensor var_42241_to_fp16 = const()[name = tensor("op_42241_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3575_cast_fp16 = mul(x = var_42240_cast_fp16, y = var_42241_to_fp16)[name = tensor("aw_3575_cast_fp16")]; + tensor var_42244_equation_0 = const()[name = tensor("op_42244_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42244_cast_fp16 = einsum(equation = var_42244_equation_0, values = (var_42086_cast_fp16, var_42003_cast_fp16))[name = tensor("op_42244_cast_fp16")]; + tensor var_42245_to_fp16 = const()[name = tensor("op_42245_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3577_cast_fp16 = mul(x = var_42244_cast_fp16, y = var_42245_to_fp16)[name = tensor("aw_3577_cast_fp16")]; + tensor var_42248_equation_0 = const()[name = tensor("op_42248_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42248_cast_fp16 = einsum(equation = var_42248_equation_0, values = (var_42090_cast_fp16, var_42007_cast_fp16))[name = tensor("op_42248_cast_fp16")]; + tensor var_42249_to_fp16 = const()[name = tensor("op_42249_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3579_cast_fp16 = mul(x = var_42248_cast_fp16, y = var_42249_to_fp16)[name = tensor("aw_3579_cast_fp16")]; + tensor var_42252_equation_0 = const()[name = tensor("op_42252_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42252_cast_fp16 = einsum(equation = var_42252_equation_0, values = (var_42094_cast_fp16, var_42011_cast_fp16))[name = tensor("op_42252_cast_fp16")]; + tensor var_42253_to_fp16 = const()[name = tensor("op_42253_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3581_cast_fp16 = mul(x = var_42252_cast_fp16, y = var_42253_to_fp16)[name = tensor("aw_3581_cast_fp16")]; + tensor var_42256_equation_0 = const()[name = tensor("op_42256_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42256_cast_fp16 = einsum(equation = var_42256_equation_0, values = (var_42098_cast_fp16, var_42015_cast_fp16))[name = tensor("op_42256_cast_fp16")]; + tensor var_42257_to_fp16 = const()[name = tensor("op_42257_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3583_cast_fp16 = mul(x = var_42256_cast_fp16, y = var_42257_to_fp16)[name = tensor("aw_3583_cast_fp16")]; + tensor var_42260_equation_0 = const()[name = tensor("op_42260_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42260_cast_fp16 = einsum(equation = var_42260_equation_0, values = (var_42102_cast_fp16, var_42019_cast_fp16))[name = tensor("op_42260_cast_fp16")]; + tensor var_42261_to_fp16 = const()[name = tensor("op_42261_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3585_cast_fp16 = mul(x = var_42260_cast_fp16, y = var_42261_to_fp16)[name = tensor("aw_3585_cast_fp16")]; + tensor var_42264_equation_0 = const()[name = tensor("op_42264_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42264_cast_fp16 = einsum(equation = var_42264_equation_0, values = (var_42106_cast_fp16, var_42023_cast_fp16))[name = tensor("op_42264_cast_fp16")]; + tensor var_42265_to_fp16 = const()[name = tensor("op_42265_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3587_cast_fp16 = mul(x = var_42264_cast_fp16, y = var_42265_to_fp16)[name = tensor("aw_3587_cast_fp16")]; + tensor var_42268_equation_0 = const()[name = tensor("op_42268_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42268_cast_fp16 = einsum(equation = var_42268_equation_0, values = (var_42110_cast_fp16, var_42027_cast_fp16))[name = tensor("op_42268_cast_fp16")]; + tensor var_42269_to_fp16 = const()[name = tensor("op_42269_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3589_cast_fp16 = mul(x = var_42268_cast_fp16, y = var_42269_to_fp16)[name = tensor("aw_3589_cast_fp16")]; + tensor var_42272_equation_0 = const()[name = tensor("op_42272_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42272_cast_fp16 = einsum(equation = var_42272_equation_0, values = (var_42114_cast_fp16, var_42031_cast_fp16))[name = tensor("op_42272_cast_fp16")]; + tensor var_42273_to_fp16 = const()[name = tensor("op_42273_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3591_cast_fp16 = mul(x = var_42272_cast_fp16, y = var_42273_to_fp16)[name = tensor("aw_3591_cast_fp16")]; + tensor var_42276_equation_0 = const()[name = tensor("op_42276_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42276_cast_fp16 = einsum(equation = var_42276_equation_0, values = (var_42118_cast_fp16, var_42035_cast_fp16))[name = tensor("op_42276_cast_fp16")]; + tensor var_42277_to_fp16 = const()[name = tensor("op_42277_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3593_cast_fp16 = mul(x = var_42276_cast_fp16, y = var_42277_to_fp16)[name = tensor("aw_3593_cast_fp16")]; + tensor var_42280_equation_0 = const()[name = tensor("op_42280_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42280_cast_fp16 = einsum(equation = var_42280_equation_0, values = (var_42122_cast_fp16, var_42039_cast_fp16))[name = tensor("op_42280_cast_fp16")]; + tensor var_42281_to_fp16 = const()[name = tensor("op_42281_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3595_cast_fp16 = mul(x = var_42280_cast_fp16, y = var_42281_to_fp16)[name = tensor("aw_3595_cast_fp16")]; + tensor var_42284_equation_0 = const()[name = tensor("op_42284_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42284_cast_fp16 = einsum(equation = var_42284_equation_0, values = (var_42126_cast_fp16, var_42043_cast_fp16))[name = tensor("op_42284_cast_fp16")]; + tensor var_42285_to_fp16 = const()[name = tensor("op_42285_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3597_cast_fp16 = mul(x = var_42284_cast_fp16, y = var_42285_to_fp16)[name = tensor("aw_3597_cast_fp16")]; + tensor var_42288_equation_0 = const()[name = tensor("op_42288_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42288_cast_fp16 = einsum(equation = var_42288_equation_0, values = (var_42130_cast_fp16, var_42047_cast_fp16))[name = tensor("op_42288_cast_fp16")]; + tensor var_42289_to_fp16 = const()[name = tensor("op_42289_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3599_cast_fp16 = mul(x = var_42288_cast_fp16, y = var_42289_to_fp16)[name = tensor("aw_3599_cast_fp16")]; + tensor var_42291_cast_fp16 = softmax(axis = var_30355, x = aw_3561_cast_fp16)[name = tensor("op_42291_cast_fp16")]; + tensor var_42292_cast_fp16 = softmax(axis = var_30355, x = aw_3563_cast_fp16)[name = tensor("op_42292_cast_fp16")]; + tensor var_42293_cast_fp16 = softmax(axis = var_30355, x = aw_3565_cast_fp16)[name = tensor("op_42293_cast_fp16")]; + tensor var_42294_cast_fp16 = softmax(axis = var_30355, x = aw_3567_cast_fp16)[name = tensor("op_42294_cast_fp16")]; + tensor var_42295_cast_fp16 = softmax(axis = var_30355, x = aw_3569_cast_fp16)[name = tensor("op_42295_cast_fp16")]; + tensor var_42296_cast_fp16 = softmax(axis = var_30355, x = aw_3571_cast_fp16)[name = tensor("op_42296_cast_fp16")]; + tensor var_42297_cast_fp16 = softmax(axis = var_30355, x = aw_3573_cast_fp16)[name = tensor("op_42297_cast_fp16")]; + tensor var_42298_cast_fp16 = softmax(axis = var_30355, x = aw_3575_cast_fp16)[name = tensor("op_42298_cast_fp16")]; + tensor var_42299_cast_fp16 = softmax(axis = var_30355, x = aw_3577_cast_fp16)[name = tensor("op_42299_cast_fp16")]; + tensor var_42300_cast_fp16 = softmax(axis = var_30355, x = aw_3579_cast_fp16)[name = tensor("op_42300_cast_fp16")]; + tensor var_42301_cast_fp16 = softmax(axis = var_30355, x = aw_3581_cast_fp16)[name = tensor("op_42301_cast_fp16")]; + tensor var_42302_cast_fp16 = softmax(axis = var_30355, x = aw_3583_cast_fp16)[name = tensor("op_42302_cast_fp16")]; + tensor var_42303_cast_fp16 = softmax(axis = var_30355, x = aw_3585_cast_fp16)[name = tensor("op_42303_cast_fp16")]; + tensor var_42304_cast_fp16 = softmax(axis = var_30355, x = aw_3587_cast_fp16)[name = tensor("op_42304_cast_fp16")]; + tensor var_42305_cast_fp16 = softmax(axis = var_30355, x = aw_3589_cast_fp16)[name = tensor("op_42305_cast_fp16")]; + tensor var_42306_cast_fp16 = softmax(axis = var_30355, x = aw_3591_cast_fp16)[name = tensor("op_42306_cast_fp16")]; + tensor var_42307_cast_fp16 = softmax(axis = var_30355, x = aw_3593_cast_fp16)[name = tensor("op_42307_cast_fp16")]; + tensor var_42308_cast_fp16 = softmax(axis = var_30355, x = aw_3595_cast_fp16)[name = tensor("op_42308_cast_fp16")]; + tensor var_42309_cast_fp16 = softmax(axis = var_30355, x = aw_3597_cast_fp16)[name = tensor("op_42309_cast_fp16")]; + tensor var_42310_cast_fp16 = softmax(axis = var_30355, x = aw_3599_cast_fp16)[name = tensor("op_42310_cast_fp16")]; + tensor var_42312_equation_0 = const()[name = tensor("op_42312_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42312_cast_fp16 = einsum(equation = var_42312_equation_0, values = (var_42132_cast_fp16, var_42291_cast_fp16))[name = tensor("op_42312_cast_fp16")]; + tensor var_42314_equation_0 = const()[name = tensor("op_42314_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42314_cast_fp16 = einsum(equation = var_42314_equation_0, values = (var_42136_cast_fp16, var_42292_cast_fp16))[name = tensor("op_42314_cast_fp16")]; + tensor var_42316_equation_0 = const()[name = tensor("op_42316_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42316_cast_fp16 = einsum(equation = var_42316_equation_0, values = (var_42140_cast_fp16, var_42293_cast_fp16))[name = tensor("op_42316_cast_fp16")]; + tensor var_42318_equation_0 = const()[name = tensor("op_42318_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42318_cast_fp16 = einsum(equation = var_42318_equation_0, values = (var_42144_cast_fp16, var_42294_cast_fp16))[name = tensor("op_42318_cast_fp16")]; + tensor var_42320_equation_0 = const()[name = tensor("op_42320_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42320_cast_fp16 = einsum(equation = var_42320_equation_0, values = (var_42148_cast_fp16, var_42295_cast_fp16))[name = tensor("op_42320_cast_fp16")]; + tensor var_42322_equation_0 = const()[name = tensor("op_42322_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42322_cast_fp16 = einsum(equation = var_42322_equation_0, values = (var_42152_cast_fp16, var_42296_cast_fp16))[name = tensor("op_42322_cast_fp16")]; + tensor var_42324_equation_0 = const()[name = tensor("op_42324_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42324_cast_fp16 = einsum(equation = var_42324_equation_0, values = (var_42156_cast_fp16, var_42297_cast_fp16))[name = tensor("op_42324_cast_fp16")]; + tensor var_42326_equation_0 = const()[name = tensor("op_42326_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42326_cast_fp16 = einsum(equation = var_42326_equation_0, values = (var_42160_cast_fp16, var_42298_cast_fp16))[name = tensor("op_42326_cast_fp16")]; + tensor var_42328_equation_0 = const()[name = tensor("op_42328_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42328_cast_fp16 = einsum(equation = var_42328_equation_0, values = (var_42164_cast_fp16, var_42299_cast_fp16))[name = tensor("op_42328_cast_fp16")]; + tensor var_42330_equation_0 = const()[name = tensor("op_42330_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42330_cast_fp16 = einsum(equation = var_42330_equation_0, values = (var_42168_cast_fp16, var_42300_cast_fp16))[name = tensor("op_42330_cast_fp16")]; + tensor var_42332_equation_0 = const()[name = tensor("op_42332_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42332_cast_fp16 = einsum(equation = var_42332_equation_0, values = (var_42172_cast_fp16, var_42301_cast_fp16))[name = tensor("op_42332_cast_fp16")]; + tensor var_42334_equation_0 = const()[name = tensor("op_42334_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42334_cast_fp16 = einsum(equation = var_42334_equation_0, values = (var_42176_cast_fp16, var_42302_cast_fp16))[name = tensor("op_42334_cast_fp16")]; + tensor var_42336_equation_0 = const()[name = tensor("op_42336_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42336_cast_fp16 = einsum(equation = var_42336_equation_0, values = (var_42180_cast_fp16, var_42303_cast_fp16))[name = tensor("op_42336_cast_fp16")]; + tensor var_42338_equation_0 = const()[name = tensor("op_42338_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42338_cast_fp16 = einsum(equation = var_42338_equation_0, values = (var_42184_cast_fp16, var_42304_cast_fp16))[name = tensor("op_42338_cast_fp16")]; + tensor var_42340_equation_0 = const()[name = tensor("op_42340_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42340_cast_fp16 = einsum(equation = var_42340_equation_0, values = (var_42188_cast_fp16, var_42305_cast_fp16))[name = tensor("op_42340_cast_fp16")]; + tensor var_42342_equation_0 = const()[name = tensor("op_42342_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42342_cast_fp16 = einsum(equation = var_42342_equation_0, values = (var_42192_cast_fp16, var_42306_cast_fp16))[name = tensor("op_42342_cast_fp16")]; + tensor var_42344_equation_0 = const()[name = tensor("op_42344_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42344_cast_fp16 = einsum(equation = var_42344_equation_0, values = (var_42196_cast_fp16, var_42307_cast_fp16))[name = tensor("op_42344_cast_fp16")]; + tensor var_42346_equation_0 = const()[name = tensor("op_42346_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42346_cast_fp16 = einsum(equation = var_42346_equation_0, values = (var_42200_cast_fp16, var_42308_cast_fp16))[name = tensor("op_42346_cast_fp16")]; + tensor var_42348_equation_0 = const()[name = tensor("op_42348_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42348_cast_fp16 = einsum(equation = var_42348_equation_0, values = (var_42204_cast_fp16, var_42309_cast_fp16))[name = tensor("op_42348_cast_fp16")]; + tensor var_42350_equation_0 = const()[name = tensor("op_42350_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42350_cast_fp16 = einsum(equation = var_42350_equation_0, values = (var_42208_cast_fp16, var_42310_cast_fp16))[name = tensor("op_42350_cast_fp16")]; + tensor input_555_interleave_0 = const()[name = tensor("input_555_interleave_0"), val = tensor(false)]; + tensor input_555_cast_fp16 = concat(axis = var_30355, interleave = input_555_interleave_0, values = (var_42312_cast_fp16, var_42314_cast_fp16, var_42316_cast_fp16, var_42318_cast_fp16, var_42320_cast_fp16, var_42322_cast_fp16, var_42324_cast_fp16, var_42326_cast_fp16, var_42328_cast_fp16, var_42330_cast_fp16, var_42332_cast_fp16, var_42334_cast_fp16, var_42336_cast_fp16, var_42338_cast_fp16, var_42340_cast_fp16, var_42342_cast_fp16, var_42344_cast_fp16, var_42346_cast_fp16, var_42348_cast_fp16, var_42350_cast_fp16))[name = tensor("input_555_cast_fp16")]; + tensor var_42356 = const()[name = tensor("op_42356"), val = tensor([1, 1])]; + tensor var_42358 = const()[name = tensor("op_42358"), val = tensor([1, 1])]; + tensor var_42360_pad_type_0 = const()[name = tensor("op_42360_pad_type_0"), val = tensor("custom")]; + tensor var_42360_pad_0 = const()[name = tensor("op_42360_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_2_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_2_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3554354816)))]; + tensor up_blocks_0_attentions_1_transformer_blocks_2_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_2_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3557631680)))]; + tensor var_42360_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_2_attn2_to_out_0_bias_to_fp16, dilations = var_42358, groups = var_30355, pad = var_42360_pad_0, pad_type = var_42360_pad_type_0, strides = var_42356, weight = up_blocks_0_attentions_1_transformer_blocks_2_attn2_to_out_0_weight_to_fp16, x = input_555_cast_fp16)[name = tensor("op_42360_cast_fp16")]; + tensor inputs_281_cast_fp16 = add(x = var_42360_cast_fp16, y = inputs_279_cast_fp16)[name = tensor("inputs_281_cast_fp16")]; + tensor input_557_axes_0 = const()[name = tensor("input_557_axes_0"), val = tensor([1])]; + tensor input_557_gamma_0_to_fp16 = const()[name = tensor("input_557_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3557634304)))]; + tensor input_557_beta_0_to_fp16 = const()[name = tensor("input_557_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3557636928)))]; + tensor var_42370_to_fp16 = const()[name = tensor("op_42370_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_557_cast_fp16 = layer_norm(axes = input_557_axes_0, beta = input_557_beta_0_to_fp16, epsilon = var_42370_to_fp16, gamma = input_557_gamma_0_to_fp16, x = inputs_281_cast_fp16)[name = tensor("input_557_cast_fp16")]; + tensor var_42386 = const()[name = tensor("op_42386"), val = tensor([1, 1])]; + tensor var_42388 = const()[name = tensor("op_42388"), val = tensor([1, 1])]; + tensor var_42390_pad_type_0 = const()[name = tensor("op_42390_pad_type_0"), val = tensor("custom")]; + tensor var_42390_pad_0 = const()[name = tensor("op_42390_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_2_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_2_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3557639552)))]; + tensor up_blocks_0_attentions_1_transformer_blocks_2_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_2_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3583854016)))]; + tensor var_42390_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_2_ff_net_0_proj_bias_to_fp16, dilations = var_42388, groups = var_30355, pad = var_42390_pad_0, pad_type = var_42390_pad_type_0, strides = var_42386, weight = up_blocks_0_attentions_1_transformer_blocks_2_ff_net_0_proj_weight_to_fp16, x = input_557_cast_fp16)[name = tensor("op_42390_cast_fp16")]; + tensor var_42391_split_sizes_0 = const()[name = tensor("op_42391_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_42391_axis_0 = const()[name = tensor("op_42391_axis_0"), val = tensor(1)]; + tensor var_42391_cast_fp16_0, tensor var_42391_cast_fp16_1 = split(axis = var_42391_axis_0, split_sizes = var_42391_split_sizes_0, x = var_42390_cast_fp16)[name = tensor("op_42391_cast_fp16")]; + tensor var_42393_mode_0 = const()[name = tensor("op_42393_mode_0"), val = tensor("EXACT")]; + tensor var_42393_cast_fp16 = gelu(mode = var_42393_mode_0, x = var_42391_cast_fp16_1)[name = tensor("op_42393_cast_fp16")]; + tensor input_559_cast_fp16 = mul(x = var_42391_cast_fp16_0, y = var_42393_cast_fp16)[name = tensor("input_559_cast_fp16")]; + tensor var_42397 = const()[name = tensor("op_42397"), val = tensor([1, 1])]; + tensor var_42399 = const()[name = tensor("op_42399"), val = tensor([1, 1])]; + tensor var_42401_pad_type_0 = const()[name = tensor("op_42401_pad_type_0"), val = tensor("custom")]; + tensor var_42401_pad_0 = const()[name = tensor("op_42401_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_2_ff_net_2_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_2_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3583874560)))]; + tensor up_blocks_0_attentions_1_transformer_blocks_2_ff_net_2_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_2_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3596981824)))]; + tensor var_42401_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_2_ff_net_2_bias_to_fp16, dilations = var_42399, groups = var_30355, pad = var_42401_pad_0, pad_type = var_42401_pad_type_0, strides = var_42397, weight = up_blocks_0_attentions_1_transformer_blocks_2_ff_net_2_weight_to_fp16, x = input_559_cast_fp16)[name = tensor("op_42401_cast_fp16")]; + tensor inputs_283_cast_fp16 = add(x = var_42401_cast_fp16, y = inputs_281_cast_fp16)[name = tensor("inputs_283_cast_fp16")]; + tensor hidden_states_379_axes_0 = const()[name = tensor("hidden_states_379_axes_0"), val = tensor([1])]; + tensor hidden_states_379_gamma_0_to_fp16 = const()[name = tensor("hidden_states_379_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3596984448)))]; + tensor hidden_states_379_beta_0_to_fp16 = const()[name = tensor("hidden_states_379_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3596987072)))]; + tensor var_42417_to_fp16 = const()[name = tensor("op_42417_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_379_cast_fp16 = layer_norm(axes = hidden_states_379_axes_0, beta = hidden_states_379_beta_0_to_fp16, epsilon = var_42417_to_fp16, gamma = hidden_states_379_gamma_0_to_fp16, x = inputs_283_cast_fp16)[name = tensor("hidden_states_379_cast_fp16")]; + tensor var_42432 = const()[name = tensor("op_42432"), val = tensor([1, 1])]; + tensor var_42434 = const()[name = tensor("op_42434"), val = tensor([1, 1])]; + tensor q_189_pad_type_0 = const()[name = tensor("q_189_pad_type_0"), val = tensor("custom")]; + tensor q_189_pad_0 = const()[name = tensor("q_189_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_3_attn1_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_3_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3596989696)))]; + tensor q_189_cast_fp16 = conv(dilations = var_42434, groups = var_30355, pad = q_189_pad_0, pad_type = q_189_pad_type_0, strides = var_42432, weight = up_blocks_0_attentions_1_transformer_blocks_3_attn1_to_q_weight_to_fp16, x = hidden_states_379_cast_fp16)[name = tensor("q_189_cast_fp16")]; + tensor var_42438 = const()[name = tensor("op_42438"), val = tensor([1, 1])]; + tensor var_42440 = const()[name = tensor("op_42440"), val = tensor([1, 1])]; + tensor k_377_pad_type_0 = const()[name = tensor("k_377_pad_type_0"), val = tensor("custom")]; + tensor k_377_pad_0 = const()[name = tensor("k_377_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_3_attn1_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_3_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3600266560)))]; + tensor k_377_cast_fp16 = conv(dilations = var_42440, groups = var_30355, pad = k_377_pad_0, pad_type = k_377_pad_type_0, strides = var_42438, weight = up_blocks_0_attentions_1_transformer_blocks_3_attn1_to_k_weight_to_fp16, x = hidden_states_379_cast_fp16)[name = tensor("k_377_cast_fp16")]; + tensor var_42444 = const()[name = tensor("op_42444"), val = tensor([1, 1])]; + tensor var_42446 = const()[name = tensor("op_42446"), val = tensor([1, 1])]; + tensor v_189_pad_type_0 = const()[name = tensor("v_189_pad_type_0"), val = tensor("custom")]; + tensor v_189_pad_0 = const()[name = tensor("v_189_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_3_attn1_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_3_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3603543424)))]; + tensor v_189_cast_fp16 = conv(dilations = var_42446, groups = var_30355, pad = v_189_pad_0, pad_type = v_189_pad_type_0, strides = var_42444, weight = up_blocks_0_attentions_1_transformer_blocks_3_attn1_to_v_weight_to_fp16, x = hidden_states_379_cast_fp16)[name = tensor("v_189_cast_fp16")]; + tensor var_42450_begin_0 = const()[name = tensor("op_42450_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_42450_end_0 = const()[name = tensor("op_42450_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_42450_end_mask_0 = const()[name = tensor("op_42450_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42450_cast_fp16 = slice_by_index(begin = var_42450_begin_0, end = var_42450_end_0, end_mask = var_42450_end_mask_0, x = q_189_cast_fp16)[name = tensor("op_42450_cast_fp16")]; + tensor var_42454_begin_0 = const()[name = tensor("op_42454_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_42454_end_0 = const()[name = tensor("op_42454_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_42454_end_mask_0 = const()[name = tensor("op_42454_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42454_cast_fp16 = slice_by_index(begin = var_42454_begin_0, end = var_42454_end_0, end_mask = var_42454_end_mask_0, x = q_189_cast_fp16)[name = tensor("op_42454_cast_fp16")]; + tensor var_42458_begin_0 = const()[name = tensor("op_42458_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_42458_end_0 = const()[name = tensor("op_42458_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_42458_end_mask_0 = const()[name = tensor("op_42458_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42458_cast_fp16 = slice_by_index(begin = var_42458_begin_0, end = var_42458_end_0, end_mask = var_42458_end_mask_0, x = q_189_cast_fp16)[name = tensor("op_42458_cast_fp16")]; + tensor var_42462_begin_0 = const()[name = tensor("op_42462_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_42462_end_0 = const()[name = tensor("op_42462_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_42462_end_mask_0 = const()[name = tensor("op_42462_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42462_cast_fp16 = slice_by_index(begin = var_42462_begin_0, end = var_42462_end_0, end_mask = var_42462_end_mask_0, x = q_189_cast_fp16)[name = tensor("op_42462_cast_fp16")]; + tensor var_42466_begin_0 = const()[name = tensor("op_42466_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_42466_end_0 = const()[name = tensor("op_42466_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_42466_end_mask_0 = const()[name = tensor("op_42466_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42466_cast_fp16 = slice_by_index(begin = var_42466_begin_0, end = var_42466_end_0, end_mask = var_42466_end_mask_0, x = q_189_cast_fp16)[name = tensor("op_42466_cast_fp16")]; + tensor var_42470_begin_0 = const()[name = tensor("op_42470_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_42470_end_0 = const()[name = tensor("op_42470_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_42470_end_mask_0 = const()[name = tensor("op_42470_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42470_cast_fp16 = slice_by_index(begin = var_42470_begin_0, end = var_42470_end_0, end_mask = var_42470_end_mask_0, x = q_189_cast_fp16)[name = tensor("op_42470_cast_fp16")]; + tensor var_42474_begin_0 = const()[name = tensor("op_42474_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_42474_end_0 = const()[name = tensor("op_42474_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_42474_end_mask_0 = const()[name = tensor("op_42474_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42474_cast_fp16 = slice_by_index(begin = var_42474_begin_0, end = var_42474_end_0, end_mask = var_42474_end_mask_0, x = q_189_cast_fp16)[name = tensor("op_42474_cast_fp16")]; + tensor var_42478_begin_0 = const()[name = tensor("op_42478_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_42478_end_0 = const()[name = tensor("op_42478_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_42478_end_mask_0 = const()[name = tensor("op_42478_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42478_cast_fp16 = slice_by_index(begin = var_42478_begin_0, end = var_42478_end_0, end_mask = var_42478_end_mask_0, x = q_189_cast_fp16)[name = tensor("op_42478_cast_fp16")]; + tensor var_42482_begin_0 = const()[name = tensor("op_42482_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_42482_end_0 = const()[name = tensor("op_42482_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_42482_end_mask_0 = const()[name = tensor("op_42482_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42482_cast_fp16 = slice_by_index(begin = var_42482_begin_0, end = var_42482_end_0, end_mask = var_42482_end_mask_0, x = q_189_cast_fp16)[name = tensor("op_42482_cast_fp16")]; + tensor var_42486_begin_0 = const()[name = tensor("op_42486_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_42486_end_0 = const()[name = tensor("op_42486_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_42486_end_mask_0 = const()[name = tensor("op_42486_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42486_cast_fp16 = slice_by_index(begin = var_42486_begin_0, end = var_42486_end_0, end_mask = var_42486_end_mask_0, x = q_189_cast_fp16)[name = tensor("op_42486_cast_fp16")]; + tensor var_42490_begin_0 = const()[name = tensor("op_42490_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_42490_end_0 = const()[name = tensor("op_42490_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_42490_end_mask_0 = const()[name = tensor("op_42490_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42490_cast_fp16 = slice_by_index(begin = var_42490_begin_0, end = var_42490_end_0, end_mask = var_42490_end_mask_0, x = q_189_cast_fp16)[name = tensor("op_42490_cast_fp16")]; + tensor var_42494_begin_0 = const()[name = tensor("op_42494_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_42494_end_0 = const()[name = tensor("op_42494_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_42494_end_mask_0 = const()[name = tensor("op_42494_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42494_cast_fp16 = slice_by_index(begin = var_42494_begin_0, end = var_42494_end_0, end_mask = var_42494_end_mask_0, x = q_189_cast_fp16)[name = tensor("op_42494_cast_fp16")]; + tensor var_42498_begin_0 = const()[name = tensor("op_42498_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_42498_end_0 = const()[name = tensor("op_42498_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_42498_end_mask_0 = const()[name = tensor("op_42498_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42498_cast_fp16 = slice_by_index(begin = var_42498_begin_0, end = var_42498_end_0, end_mask = var_42498_end_mask_0, x = q_189_cast_fp16)[name = tensor("op_42498_cast_fp16")]; + tensor var_42502_begin_0 = const()[name = tensor("op_42502_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_42502_end_0 = const()[name = tensor("op_42502_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_42502_end_mask_0 = const()[name = tensor("op_42502_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42502_cast_fp16 = slice_by_index(begin = var_42502_begin_0, end = var_42502_end_0, end_mask = var_42502_end_mask_0, x = q_189_cast_fp16)[name = tensor("op_42502_cast_fp16")]; + tensor var_42506_begin_0 = const()[name = tensor("op_42506_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_42506_end_0 = const()[name = tensor("op_42506_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_42506_end_mask_0 = const()[name = tensor("op_42506_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42506_cast_fp16 = slice_by_index(begin = var_42506_begin_0, end = var_42506_end_0, end_mask = var_42506_end_mask_0, x = q_189_cast_fp16)[name = tensor("op_42506_cast_fp16")]; + tensor var_42510_begin_0 = const()[name = tensor("op_42510_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_42510_end_0 = const()[name = tensor("op_42510_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_42510_end_mask_0 = const()[name = tensor("op_42510_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42510_cast_fp16 = slice_by_index(begin = var_42510_begin_0, end = var_42510_end_0, end_mask = var_42510_end_mask_0, x = q_189_cast_fp16)[name = tensor("op_42510_cast_fp16")]; + tensor var_42514_begin_0 = const()[name = tensor("op_42514_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_42514_end_0 = const()[name = tensor("op_42514_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_42514_end_mask_0 = const()[name = tensor("op_42514_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42514_cast_fp16 = slice_by_index(begin = var_42514_begin_0, end = var_42514_end_0, end_mask = var_42514_end_mask_0, x = q_189_cast_fp16)[name = tensor("op_42514_cast_fp16")]; + tensor var_42518_begin_0 = const()[name = tensor("op_42518_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_42518_end_0 = const()[name = tensor("op_42518_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_42518_end_mask_0 = const()[name = tensor("op_42518_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42518_cast_fp16 = slice_by_index(begin = var_42518_begin_0, end = var_42518_end_0, end_mask = var_42518_end_mask_0, x = q_189_cast_fp16)[name = tensor("op_42518_cast_fp16")]; + tensor var_42522_begin_0 = const()[name = tensor("op_42522_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_42522_end_0 = const()[name = tensor("op_42522_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_42522_end_mask_0 = const()[name = tensor("op_42522_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42522_cast_fp16 = slice_by_index(begin = var_42522_begin_0, end = var_42522_end_0, end_mask = var_42522_end_mask_0, x = q_189_cast_fp16)[name = tensor("op_42522_cast_fp16")]; + tensor var_42526_begin_0 = const()[name = tensor("op_42526_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_42526_end_0 = const()[name = tensor("op_42526_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_42526_end_mask_0 = const()[name = tensor("op_42526_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42526_cast_fp16 = slice_by_index(begin = var_42526_begin_0, end = var_42526_end_0, end_mask = var_42526_end_mask_0, x = q_189_cast_fp16)[name = tensor("op_42526_cast_fp16")]; + tensor k_379_perm_0 = const()[name = tensor("k_379_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_42533_begin_0 = const()[name = tensor("op_42533_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_42533_end_0 = const()[name = tensor("op_42533_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_42533_end_mask_0 = const()[name = tensor("op_42533_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_379_cast_fp16 = transpose(perm = k_379_perm_0, x = k_377_cast_fp16)[name = tensor("transpose_45")]; + tensor var_42533_cast_fp16 = slice_by_index(begin = var_42533_begin_0, end = var_42533_end_0, end_mask = var_42533_end_mask_0, x = k_379_cast_fp16)[name = tensor("op_42533_cast_fp16")]; + tensor var_42537_begin_0 = const()[name = tensor("op_42537_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_42537_end_0 = const()[name = tensor("op_42537_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_42537_end_mask_0 = const()[name = tensor("op_42537_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42537_cast_fp16 = slice_by_index(begin = var_42537_begin_0, end = var_42537_end_0, end_mask = var_42537_end_mask_0, x = k_379_cast_fp16)[name = tensor("op_42537_cast_fp16")]; + tensor var_42541_begin_0 = const()[name = tensor("op_42541_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_42541_end_0 = const()[name = tensor("op_42541_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_42541_end_mask_0 = const()[name = tensor("op_42541_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42541_cast_fp16 = slice_by_index(begin = var_42541_begin_0, end = var_42541_end_0, end_mask = var_42541_end_mask_0, x = k_379_cast_fp16)[name = tensor("op_42541_cast_fp16")]; + tensor var_42545_begin_0 = const()[name = tensor("op_42545_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_42545_end_0 = const()[name = tensor("op_42545_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_42545_end_mask_0 = const()[name = tensor("op_42545_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42545_cast_fp16 = slice_by_index(begin = var_42545_begin_0, end = var_42545_end_0, end_mask = var_42545_end_mask_0, x = k_379_cast_fp16)[name = tensor("op_42545_cast_fp16")]; + tensor var_42549_begin_0 = const()[name = tensor("op_42549_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_42549_end_0 = const()[name = tensor("op_42549_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_42549_end_mask_0 = const()[name = tensor("op_42549_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42549_cast_fp16 = slice_by_index(begin = var_42549_begin_0, end = var_42549_end_0, end_mask = var_42549_end_mask_0, x = k_379_cast_fp16)[name = tensor("op_42549_cast_fp16")]; + tensor var_42553_begin_0 = const()[name = tensor("op_42553_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_42553_end_0 = const()[name = tensor("op_42553_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_42553_end_mask_0 = const()[name = tensor("op_42553_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42553_cast_fp16 = slice_by_index(begin = var_42553_begin_0, end = var_42553_end_0, end_mask = var_42553_end_mask_0, x = k_379_cast_fp16)[name = tensor("op_42553_cast_fp16")]; + tensor var_42557_begin_0 = const()[name = tensor("op_42557_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_42557_end_0 = const()[name = tensor("op_42557_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_42557_end_mask_0 = const()[name = tensor("op_42557_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42557_cast_fp16 = slice_by_index(begin = var_42557_begin_0, end = var_42557_end_0, end_mask = var_42557_end_mask_0, x = k_379_cast_fp16)[name = tensor("op_42557_cast_fp16")]; + tensor var_42561_begin_0 = const()[name = tensor("op_42561_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_42561_end_0 = const()[name = tensor("op_42561_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_42561_end_mask_0 = const()[name = tensor("op_42561_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42561_cast_fp16 = slice_by_index(begin = var_42561_begin_0, end = var_42561_end_0, end_mask = var_42561_end_mask_0, x = k_379_cast_fp16)[name = tensor("op_42561_cast_fp16")]; + tensor var_42565_begin_0 = const()[name = tensor("op_42565_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_42565_end_0 = const()[name = tensor("op_42565_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_42565_end_mask_0 = const()[name = tensor("op_42565_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42565_cast_fp16 = slice_by_index(begin = var_42565_begin_0, end = var_42565_end_0, end_mask = var_42565_end_mask_0, x = k_379_cast_fp16)[name = tensor("op_42565_cast_fp16")]; + tensor var_42569_begin_0 = const()[name = tensor("op_42569_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_42569_end_0 = const()[name = tensor("op_42569_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_42569_end_mask_0 = const()[name = tensor("op_42569_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42569_cast_fp16 = slice_by_index(begin = var_42569_begin_0, end = var_42569_end_0, end_mask = var_42569_end_mask_0, x = k_379_cast_fp16)[name = tensor("op_42569_cast_fp16")]; + tensor var_42573_begin_0 = const()[name = tensor("op_42573_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_42573_end_0 = const()[name = tensor("op_42573_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_42573_end_mask_0 = const()[name = tensor("op_42573_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42573_cast_fp16 = slice_by_index(begin = var_42573_begin_0, end = var_42573_end_0, end_mask = var_42573_end_mask_0, x = k_379_cast_fp16)[name = tensor("op_42573_cast_fp16")]; + tensor var_42577_begin_0 = const()[name = tensor("op_42577_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_42577_end_0 = const()[name = tensor("op_42577_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_42577_end_mask_0 = const()[name = tensor("op_42577_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42577_cast_fp16 = slice_by_index(begin = var_42577_begin_0, end = var_42577_end_0, end_mask = var_42577_end_mask_0, x = k_379_cast_fp16)[name = tensor("op_42577_cast_fp16")]; + tensor var_42581_begin_0 = const()[name = tensor("op_42581_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_42581_end_0 = const()[name = tensor("op_42581_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_42581_end_mask_0 = const()[name = tensor("op_42581_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42581_cast_fp16 = slice_by_index(begin = var_42581_begin_0, end = var_42581_end_0, end_mask = var_42581_end_mask_0, x = k_379_cast_fp16)[name = tensor("op_42581_cast_fp16")]; + tensor var_42585_begin_0 = const()[name = tensor("op_42585_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_42585_end_0 = const()[name = tensor("op_42585_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_42585_end_mask_0 = const()[name = tensor("op_42585_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42585_cast_fp16 = slice_by_index(begin = var_42585_begin_0, end = var_42585_end_0, end_mask = var_42585_end_mask_0, x = k_379_cast_fp16)[name = tensor("op_42585_cast_fp16")]; + tensor var_42589_begin_0 = const()[name = tensor("op_42589_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_42589_end_0 = const()[name = tensor("op_42589_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_42589_end_mask_0 = const()[name = tensor("op_42589_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42589_cast_fp16 = slice_by_index(begin = var_42589_begin_0, end = var_42589_end_0, end_mask = var_42589_end_mask_0, x = k_379_cast_fp16)[name = tensor("op_42589_cast_fp16")]; + tensor var_42593_begin_0 = const()[name = tensor("op_42593_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_42593_end_0 = const()[name = tensor("op_42593_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_42593_end_mask_0 = const()[name = tensor("op_42593_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42593_cast_fp16 = slice_by_index(begin = var_42593_begin_0, end = var_42593_end_0, end_mask = var_42593_end_mask_0, x = k_379_cast_fp16)[name = tensor("op_42593_cast_fp16")]; + tensor var_42597_begin_0 = const()[name = tensor("op_42597_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_42597_end_0 = const()[name = tensor("op_42597_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_42597_end_mask_0 = const()[name = tensor("op_42597_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42597_cast_fp16 = slice_by_index(begin = var_42597_begin_0, end = var_42597_end_0, end_mask = var_42597_end_mask_0, x = k_379_cast_fp16)[name = tensor("op_42597_cast_fp16")]; + tensor var_42601_begin_0 = const()[name = tensor("op_42601_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_42601_end_0 = const()[name = tensor("op_42601_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_42601_end_mask_0 = const()[name = tensor("op_42601_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42601_cast_fp16 = slice_by_index(begin = var_42601_begin_0, end = var_42601_end_0, end_mask = var_42601_end_mask_0, x = k_379_cast_fp16)[name = tensor("op_42601_cast_fp16")]; + tensor var_42605_begin_0 = const()[name = tensor("op_42605_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_42605_end_0 = const()[name = tensor("op_42605_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_42605_end_mask_0 = const()[name = tensor("op_42605_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42605_cast_fp16 = slice_by_index(begin = var_42605_begin_0, end = var_42605_end_0, end_mask = var_42605_end_mask_0, x = k_379_cast_fp16)[name = tensor("op_42605_cast_fp16")]; + tensor var_42609_begin_0 = const()[name = tensor("op_42609_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_42609_end_0 = const()[name = tensor("op_42609_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_42609_end_mask_0 = const()[name = tensor("op_42609_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42609_cast_fp16 = slice_by_index(begin = var_42609_begin_0, end = var_42609_end_0, end_mask = var_42609_end_mask_0, x = k_379_cast_fp16)[name = tensor("op_42609_cast_fp16")]; + tensor var_42611_begin_0 = const()[name = tensor("op_42611_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_42611_end_0 = const()[name = tensor("op_42611_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_42611_end_mask_0 = const()[name = tensor("op_42611_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42611_cast_fp16 = slice_by_index(begin = var_42611_begin_0, end = var_42611_end_0, end_mask = var_42611_end_mask_0, x = v_189_cast_fp16)[name = tensor("op_42611_cast_fp16")]; + tensor var_42615_begin_0 = const()[name = tensor("op_42615_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_42615_end_0 = const()[name = tensor("op_42615_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_42615_end_mask_0 = const()[name = tensor("op_42615_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42615_cast_fp16 = slice_by_index(begin = var_42615_begin_0, end = var_42615_end_0, end_mask = var_42615_end_mask_0, x = v_189_cast_fp16)[name = tensor("op_42615_cast_fp16")]; + tensor var_42619_begin_0 = const()[name = tensor("op_42619_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_42619_end_0 = const()[name = tensor("op_42619_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_42619_end_mask_0 = const()[name = tensor("op_42619_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42619_cast_fp16 = slice_by_index(begin = var_42619_begin_0, end = var_42619_end_0, end_mask = var_42619_end_mask_0, x = v_189_cast_fp16)[name = tensor("op_42619_cast_fp16")]; + tensor var_42623_begin_0 = const()[name = tensor("op_42623_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_42623_end_0 = const()[name = tensor("op_42623_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_42623_end_mask_0 = const()[name = tensor("op_42623_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42623_cast_fp16 = slice_by_index(begin = var_42623_begin_0, end = var_42623_end_0, end_mask = var_42623_end_mask_0, x = v_189_cast_fp16)[name = tensor("op_42623_cast_fp16")]; + tensor var_42627_begin_0 = const()[name = tensor("op_42627_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_42627_end_0 = const()[name = tensor("op_42627_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_42627_end_mask_0 = const()[name = tensor("op_42627_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42627_cast_fp16 = slice_by_index(begin = var_42627_begin_0, end = var_42627_end_0, end_mask = var_42627_end_mask_0, x = v_189_cast_fp16)[name = tensor("op_42627_cast_fp16")]; + tensor var_42631_begin_0 = const()[name = tensor("op_42631_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_42631_end_0 = const()[name = tensor("op_42631_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_42631_end_mask_0 = const()[name = tensor("op_42631_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42631_cast_fp16 = slice_by_index(begin = var_42631_begin_0, end = var_42631_end_0, end_mask = var_42631_end_mask_0, x = v_189_cast_fp16)[name = tensor("op_42631_cast_fp16")]; + tensor var_42635_begin_0 = const()[name = tensor("op_42635_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_42635_end_0 = const()[name = tensor("op_42635_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_42635_end_mask_0 = const()[name = tensor("op_42635_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42635_cast_fp16 = slice_by_index(begin = var_42635_begin_0, end = var_42635_end_0, end_mask = var_42635_end_mask_0, x = v_189_cast_fp16)[name = tensor("op_42635_cast_fp16")]; + tensor var_42639_begin_0 = const()[name = tensor("op_42639_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_42639_end_0 = const()[name = tensor("op_42639_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_42639_end_mask_0 = const()[name = tensor("op_42639_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42639_cast_fp16 = slice_by_index(begin = var_42639_begin_0, end = var_42639_end_0, end_mask = var_42639_end_mask_0, x = v_189_cast_fp16)[name = tensor("op_42639_cast_fp16")]; + tensor var_42643_begin_0 = const()[name = tensor("op_42643_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_42643_end_0 = const()[name = tensor("op_42643_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_42643_end_mask_0 = const()[name = tensor("op_42643_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42643_cast_fp16 = slice_by_index(begin = var_42643_begin_0, end = var_42643_end_0, end_mask = var_42643_end_mask_0, x = v_189_cast_fp16)[name = tensor("op_42643_cast_fp16")]; + tensor var_42647_begin_0 = const()[name = tensor("op_42647_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_42647_end_0 = const()[name = tensor("op_42647_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_42647_end_mask_0 = const()[name = tensor("op_42647_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42647_cast_fp16 = slice_by_index(begin = var_42647_begin_0, end = var_42647_end_0, end_mask = var_42647_end_mask_0, x = v_189_cast_fp16)[name = tensor("op_42647_cast_fp16")]; + tensor var_42651_begin_0 = const()[name = tensor("op_42651_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_42651_end_0 = const()[name = tensor("op_42651_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_42651_end_mask_0 = const()[name = tensor("op_42651_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42651_cast_fp16 = slice_by_index(begin = var_42651_begin_0, end = var_42651_end_0, end_mask = var_42651_end_mask_0, x = v_189_cast_fp16)[name = tensor("op_42651_cast_fp16")]; + tensor var_42655_begin_0 = const()[name = tensor("op_42655_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_42655_end_0 = const()[name = tensor("op_42655_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_42655_end_mask_0 = const()[name = tensor("op_42655_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42655_cast_fp16 = slice_by_index(begin = var_42655_begin_0, end = var_42655_end_0, end_mask = var_42655_end_mask_0, x = v_189_cast_fp16)[name = tensor("op_42655_cast_fp16")]; + tensor var_42659_begin_0 = const()[name = tensor("op_42659_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_42659_end_0 = const()[name = tensor("op_42659_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_42659_end_mask_0 = const()[name = tensor("op_42659_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42659_cast_fp16 = slice_by_index(begin = var_42659_begin_0, end = var_42659_end_0, end_mask = var_42659_end_mask_0, x = v_189_cast_fp16)[name = tensor("op_42659_cast_fp16")]; + tensor var_42663_begin_0 = const()[name = tensor("op_42663_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_42663_end_0 = const()[name = tensor("op_42663_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_42663_end_mask_0 = const()[name = tensor("op_42663_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42663_cast_fp16 = slice_by_index(begin = var_42663_begin_0, end = var_42663_end_0, end_mask = var_42663_end_mask_0, x = v_189_cast_fp16)[name = tensor("op_42663_cast_fp16")]; + tensor var_42667_begin_0 = const()[name = tensor("op_42667_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_42667_end_0 = const()[name = tensor("op_42667_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_42667_end_mask_0 = const()[name = tensor("op_42667_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42667_cast_fp16 = slice_by_index(begin = var_42667_begin_0, end = var_42667_end_0, end_mask = var_42667_end_mask_0, x = v_189_cast_fp16)[name = tensor("op_42667_cast_fp16")]; + tensor var_42671_begin_0 = const()[name = tensor("op_42671_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_42671_end_0 = const()[name = tensor("op_42671_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_42671_end_mask_0 = const()[name = tensor("op_42671_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42671_cast_fp16 = slice_by_index(begin = var_42671_begin_0, end = var_42671_end_0, end_mask = var_42671_end_mask_0, x = v_189_cast_fp16)[name = tensor("op_42671_cast_fp16")]; + tensor var_42675_begin_0 = const()[name = tensor("op_42675_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_42675_end_0 = const()[name = tensor("op_42675_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_42675_end_mask_0 = const()[name = tensor("op_42675_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42675_cast_fp16 = slice_by_index(begin = var_42675_begin_0, end = var_42675_end_0, end_mask = var_42675_end_mask_0, x = v_189_cast_fp16)[name = tensor("op_42675_cast_fp16")]; + tensor var_42679_begin_0 = const()[name = tensor("op_42679_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_42679_end_0 = const()[name = tensor("op_42679_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_42679_end_mask_0 = const()[name = tensor("op_42679_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42679_cast_fp16 = slice_by_index(begin = var_42679_begin_0, end = var_42679_end_0, end_mask = var_42679_end_mask_0, x = v_189_cast_fp16)[name = tensor("op_42679_cast_fp16")]; + tensor var_42683_begin_0 = const()[name = tensor("op_42683_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_42683_end_0 = const()[name = tensor("op_42683_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_42683_end_mask_0 = const()[name = tensor("op_42683_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42683_cast_fp16 = slice_by_index(begin = var_42683_begin_0, end = var_42683_end_0, end_mask = var_42683_end_mask_0, x = v_189_cast_fp16)[name = tensor("op_42683_cast_fp16")]; + tensor var_42687_begin_0 = const()[name = tensor("op_42687_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_42687_end_0 = const()[name = tensor("op_42687_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_42687_end_mask_0 = const()[name = tensor("op_42687_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42687_cast_fp16 = slice_by_index(begin = var_42687_begin_0, end = var_42687_end_0, end_mask = var_42687_end_mask_0, x = v_189_cast_fp16)[name = tensor("op_42687_cast_fp16")]; + tensor var_42691_equation_0 = const()[name = tensor("op_42691_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42691_cast_fp16 = einsum(equation = var_42691_equation_0, values = (var_42533_cast_fp16, var_42450_cast_fp16))[name = tensor("op_42691_cast_fp16")]; + tensor var_42692_to_fp16 = const()[name = tensor("op_42692_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3601_cast_fp16 = mul(x = var_42691_cast_fp16, y = var_42692_to_fp16)[name = tensor("aw_3601_cast_fp16")]; + tensor var_42695_equation_0 = const()[name = tensor("op_42695_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42695_cast_fp16 = einsum(equation = var_42695_equation_0, values = (var_42537_cast_fp16, var_42454_cast_fp16))[name = tensor("op_42695_cast_fp16")]; + tensor var_42696_to_fp16 = const()[name = tensor("op_42696_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3603_cast_fp16 = mul(x = var_42695_cast_fp16, y = var_42696_to_fp16)[name = tensor("aw_3603_cast_fp16")]; + tensor var_42699_equation_0 = const()[name = tensor("op_42699_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42699_cast_fp16 = einsum(equation = var_42699_equation_0, values = (var_42541_cast_fp16, var_42458_cast_fp16))[name = tensor("op_42699_cast_fp16")]; + tensor var_42700_to_fp16 = const()[name = tensor("op_42700_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3605_cast_fp16 = mul(x = var_42699_cast_fp16, y = var_42700_to_fp16)[name = tensor("aw_3605_cast_fp16")]; + tensor var_42703_equation_0 = const()[name = tensor("op_42703_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42703_cast_fp16 = einsum(equation = var_42703_equation_0, values = (var_42545_cast_fp16, var_42462_cast_fp16))[name = tensor("op_42703_cast_fp16")]; + tensor var_42704_to_fp16 = const()[name = tensor("op_42704_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3607_cast_fp16 = mul(x = var_42703_cast_fp16, y = var_42704_to_fp16)[name = tensor("aw_3607_cast_fp16")]; + tensor var_42707_equation_0 = const()[name = tensor("op_42707_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42707_cast_fp16 = einsum(equation = var_42707_equation_0, values = (var_42549_cast_fp16, var_42466_cast_fp16))[name = tensor("op_42707_cast_fp16")]; + tensor var_42708_to_fp16 = const()[name = tensor("op_42708_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3609_cast_fp16 = mul(x = var_42707_cast_fp16, y = var_42708_to_fp16)[name = tensor("aw_3609_cast_fp16")]; + tensor var_42711_equation_0 = const()[name = tensor("op_42711_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42711_cast_fp16 = einsum(equation = var_42711_equation_0, values = (var_42553_cast_fp16, var_42470_cast_fp16))[name = tensor("op_42711_cast_fp16")]; + tensor var_42712_to_fp16 = const()[name = tensor("op_42712_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3611_cast_fp16 = mul(x = var_42711_cast_fp16, y = var_42712_to_fp16)[name = tensor("aw_3611_cast_fp16")]; + tensor var_42715_equation_0 = const()[name = tensor("op_42715_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42715_cast_fp16 = einsum(equation = var_42715_equation_0, values = (var_42557_cast_fp16, var_42474_cast_fp16))[name = tensor("op_42715_cast_fp16")]; + tensor var_42716_to_fp16 = const()[name = tensor("op_42716_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3613_cast_fp16 = mul(x = var_42715_cast_fp16, y = var_42716_to_fp16)[name = tensor("aw_3613_cast_fp16")]; + tensor var_42719_equation_0 = const()[name = tensor("op_42719_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42719_cast_fp16 = einsum(equation = var_42719_equation_0, values = (var_42561_cast_fp16, var_42478_cast_fp16))[name = tensor("op_42719_cast_fp16")]; + tensor var_42720_to_fp16 = const()[name = tensor("op_42720_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3615_cast_fp16 = mul(x = var_42719_cast_fp16, y = var_42720_to_fp16)[name = tensor("aw_3615_cast_fp16")]; + tensor var_42723_equation_0 = const()[name = tensor("op_42723_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42723_cast_fp16 = einsum(equation = var_42723_equation_0, values = (var_42565_cast_fp16, var_42482_cast_fp16))[name = tensor("op_42723_cast_fp16")]; + tensor var_42724_to_fp16 = const()[name = tensor("op_42724_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3617_cast_fp16 = mul(x = var_42723_cast_fp16, y = var_42724_to_fp16)[name = tensor("aw_3617_cast_fp16")]; + tensor var_42727_equation_0 = const()[name = tensor("op_42727_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42727_cast_fp16 = einsum(equation = var_42727_equation_0, values = (var_42569_cast_fp16, var_42486_cast_fp16))[name = tensor("op_42727_cast_fp16")]; + tensor var_42728_to_fp16 = const()[name = tensor("op_42728_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3619_cast_fp16 = mul(x = var_42727_cast_fp16, y = var_42728_to_fp16)[name = tensor("aw_3619_cast_fp16")]; + tensor var_42731_equation_0 = const()[name = tensor("op_42731_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42731_cast_fp16 = einsum(equation = var_42731_equation_0, values = (var_42573_cast_fp16, var_42490_cast_fp16))[name = tensor("op_42731_cast_fp16")]; + tensor var_42732_to_fp16 = const()[name = tensor("op_42732_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3621_cast_fp16 = mul(x = var_42731_cast_fp16, y = var_42732_to_fp16)[name = tensor("aw_3621_cast_fp16")]; + tensor var_42735_equation_0 = const()[name = tensor("op_42735_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42735_cast_fp16 = einsum(equation = var_42735_equation_0, values = (var_42577_cast_fp16, var_42494_cast_fp16))[name = tensor("op_42735_cast_fp16")]; + tensor var_42736_to_fp16 = const()[name = tensor("op_42736_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3623_cast_fp16 = mul(x = var_42735_cast_fp16, y = var_42736_to_fp16)[name = tensor("aw_3623_cast_fp16")]; + tensor var_42739_equation_0 = const()[name = tensor("op_42739_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42739_cast_fp16 = einsum(equation = var_42739_equation_0, values = (var_42581_cast_fp16, var_42498_cast_fp16))[name = tensor("op_42739_cast_fp16")]; + tensor var_42740_to_fp16 = const()[name = tensor("op_42740_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3625_cast_fp16 = mul(x = var_42739_cast_fp16, y = var_42740_to_fp16)[name = tensor("aw_3625_cast_fp16")]; + tensor var_42743_equation_0 = const()[name = tensor("op_42743_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42743_cast_fp16 = einsum(equation = var_42743_equation_0, values = (var_42585_cast_fp16, var_42502_cast_fp16))[name = tensor("op_42743_cast_fp16")]; + tensor var_42744_to_fp16 = const()[name = tensor("op_42744_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3627_cast_fp16 = mul(x = var_42743_cast_fp16, y = var_42744_to_fp16)[name = tensor("aw_3627_cast_fp16")]; + tensor var_42747_equation_0 = const()[name = tensor("op_42747_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42747_cast_fp16 = einsum(equation = var_42747_equation_0, values = (var_42589_cast_fp16, var_42506_cast_fp16))[name = tensor("op_42747_cast_fp16")]; + tensor var_42748_to_fp16 = const()[name = tensor("op_42748_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3629_cast_fp16 = mul(x = var_42747_cast_fp16, y = var_42748_to_fp16)[name = tensor("aw_3629_cast_fp16")]; + tensor var_42751_equation_0 = const()[name = tensor("op_42751_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42751_cast_fp16 = einsum(equation = var_42751_equation_0, values = (var_42593_cast_fp16, var_42510_cast_fp16))[name = tensor("op_42751_cast_fp16")]; + tensor var_42752_to_fp16 = const()[name = tensor("op_42752_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3631_cast_fp16 = mul(x = var_42751_cast_fp16, y = var_42752_to_fp16)[name = tensor("aw_3631_cast_fp16")]; + tensor var_42755_equation_0 = const()[name = tensor("op_42755_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42755_cast_fp16 = einsum(equation = var_42755_equation_0, values = (var_42597_cast_fp16, var_42514_cast_fp16))[name = tensor("op_42755_cast_fp16")]; + tensor var_42756_to_fp16 = const()[name = tensor("op_42756_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3633_cast_fp16 = mul(x = var_42755_cast_fp16, y = var_42756_to_fp16)[name = tensor("aw_3633_cast_fp16")]; + tensor var_42759_equation_0 = const()[name = tensor("op_42759_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42759_cast_fp16 = einsum(equation = var_42759_equation_0, values = (var_42601_cast_fp16, var_42518_cast_fp16))[name = tensor("op_42759_cast_fp16")]; + tensor var_42760_to_fp16 = const()[name = tensor("op_42760_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3635_cast_fp16 = mul(x = var_42759_cast_fp16, y = var_42760_to_fp16)[name = tensor("aw_3635_cast_fp16")]; + tensor var_42763_equation_0 = const()[name = tensor("op_42763_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42763_cast_fp16 = einsum(equation = var_42763_equation_0, values = (var_42605_cast_fp16, var_42522_cast_fp16))[name = tensor("op_42763_cast_fp16")]; + tensor var_42764_to_fp16 = const()[name = tensor("op_42764_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3637_cast_fp16 = mul(x = var_42763_cast_fp16, y = var_42764_to_fp16)[name = tensor("aw_3637_cast_fp16")]; + tensor var_42767_equation_0 = const()[name = tensor("op_42767_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42767_cast_fp16 = einsum(equation = var_42767_equation_0, values = (var_42609_cast_fp16, var_42526_cast_fp16))[name = tensor("op_42767_cast_fp16")]; + tensor var_42768_to_fp16 = const()[name = tensor("op_42768_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3639_cast_fp16 = mul(x = var_42767_cast_fp16, y = var_42768_to_fp16)[name = tensor("aw_3639_cast_fp16")]; + tensor var_42770_cast_fp16 = softmax(axis = var_30355, x = aw_3601_cast_fp16)[name = tensor("op_42770_cast_fp16")]; + tensor var_42771_cast_fp16 = softmax(axis = var_30355, x = aw_3603_cast_fp16)[name = tensor("op_42771_cast_fp16")]; + tensor var_42772_cast_fp16 = softmax(axis = var_30355, x = aw_3605_cast_fp16)[name = tensor("op_42772_cast_fp16")]; + tensor var_42773_cast_fp16 = softmax(axis = var_30355, x = aw_3607_cast_fp16)[name = tensor("op_42773_cast_fp16")]; + tensor var_42774_cast_fp16 = softmax(axis = var_30355, x = aw_3609_cast_fp16)[name = tensor("op_42774_cast_fp16")]; + tensor var_42775_cast_fp16 = softmax(axis = var_30355, x = aw_3611_cast_fp16)[name = tensor("op_42775_cast_fp16")]; + tensor var_42776_cast_fp16 = softmax(axis = var_30355, x = aw_3613_cast_fp16)[name = tensor("op_42776_cast_fp16")]; + tensor var_42777_cast_fp16 = softmax(axis = var_30355, x = aw_3615_cast_fp16)[name = tensor("op_42777_cast_fp16")]; + tensor var_42778_cast_fp16 = softmax(axis = var_30355, x = aw_3617_cast_fp16)[name = tensor("op_42778_cast_fp16")]; + tensor var_42779_cast_fp16 = softmax(axis = var_30355, x = aw_3619_cast_fp16)[name = tensor("op_42779_cast_fp16")]; + tensor var_42780_cast_fp16 = softmax(axis = var_30355, x = aw_3621_cast_fp16)[name = tensor("op_42780_cast_fp16")]; + tensor var_42781_cast_fp16 = softmax(axis = var_30355, x = aw_3623_cast_fp16)[name = tensor("op_42781_cast_fp16")]; + tensor var_42782_cast_fp16 = softmax(axis = var_30355, x = aw_3625_cast_fp16)[name = tensor("op_42782_cast_fp16")]; + tensor var_42783_cast_fp16 = softmax(axis = var_30355, x = aw_3627_cast_fp16)[name = tensor("op_42783_cast_fp16")]; + tensor var_42784_cast_fp16 = softmax(axis = var_30355, x = aw_3629_cast_fp16)[name = tensor("op_42784_cast_fp16")]; + tensor var_42785_cast_fp16 = softmax(axis = var_30355, x = aw_3631_cast_fp16)[name = tensor("op_42785_cast_fp16")]; + tensor var_42786_cast_fp16 = softmax(axis = var_30355, x = aw_3633_cast_fp16)[name = tensor("op_42786_cast_fp16")]; + tensor var_42787_cast_fp16 = softmax(axis = var_30355, x = aw_3635_cast_fp16)[name = tensor("op_42787_cast_fp16")]; + tensor var_42788_cast_fp16 = softmax(axis = var_30355, x = aw_3637_cast_fp16)[name = tensor("op_42788_cast_fp16")]; + tensor var_42789_cast_fp16 = softmax(axis = var_30355, x = aw_3639_cast_fp16)[name = tensor("op_42789_cast_fp16")]; + tensor var_42791_equation_0 = const()[name = tensor("op_42791_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42791_cast_fp16 = einsum(equation = var_42791_equation_0, values = (var_42611_cast_fp16, var_42770_cast_fp16))[name = tensor("op_42791_cast_fp16")]; + tensor var_42793_equation_0 = const()[name = tensor("op_42793_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42793_cast_fp16 = einsum(equation = var_42793_equation_0, values = (var_42615_cast_fp16, var_42771_cast_fp16))[name = tensor("op_42793_cast_fp16")]; + tensor var_42795_equation_0 = const()[name = tensor("op_42795_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42795_cast_fp16 = einsum(equation = var_42795_equation_0, values = (var_42619_cast_fp16, var_42772_cast_fp16))[name = tensor("op_42795_cast_fp16")]; + tensor var_42797_equation_0 = const()[name = tensor("op_42797_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42797_cast_fp16 = einsum(equation = var_42797_equation_0, values = (var_42623_cast_fp16, var_42773_cast_fp16))[name = tensor("op_42797_cast_fp16")]; + tensor var_42799_equation_0 = const()[name = tensor("op_42799_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42799_cast_fp16 = einsum(equation = var_42799_equation_0, values = (var_42627_cast_fp16, var_42774_cast_fp16))[name = tensor("op_42799_cast_fp16")]; + tensor var_42801_equation_0 = const()[name = tensor("op_42801_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42801_cast_fp16 = einsum(equation = var_42801_equation_0, values = (var_42631_cast_fp16, var_42775_cast_fp16))[name = tensor("op_42801_cast_fp16")]; + tensor var_42803_equation_0 = const()[name = tensor("op_42803_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42803_cast_fp16 = einsum(equation = var_42803_equation_0, values = (var_42635_cast_fp16, var_42776_cast_fp16))[name = tensor("op_42803_cast_fp16")]; + tensor var_42805_equation_0 = const()[name = tensor("op_42805_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42805_cast_fp16 = einsum(equation = var_42805_equation_0, values = (var_42639_cast_fp16, var_42777_cast_fp16))[name = tensor("op_42805_cast_fp16")]; + tensor var_42807_equation_0 = const()[name = tensor("op_42807_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42807_cast_fp16 = einsum(equation = var_42807_equation_0, values = (var_42643_cast_fp16, var_42778_cast_fp16))[name = tensor("op_42807_cast_fp16")]; + tensor var_42809_equation_0 = const()[name = tensor("op_42809_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42809_cast_fp16 = einsum(equation = var_42809_equation_0, values = (var_42647_cast_fp16, var_42779_cast_fp16))[name = tensor("op_42809_cast_fp16")]; + tensor var_42811_equation_0 = const()[name = tensor("op_42811_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42811_cast_fp16 = einsum(equation = var_42811_equation_0, values = (var_42651_cast_fp16, var_42780_cast_fp16))[name = tensor("op_42811_cast_fp16")]; + tensor var_42813_equation_0 = const()[name = tensor("op_42813_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42813_cast_fp16 = einsum(equation = var_42813_equation_0, values = (var_42655_cast_fp16, var_42781_cast_fp16))[name = tensor("op_42813_cast_fp16")]; + tensor var_42815_equation_0 = const()[name = tensor("op_42815_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42815_cast_fp16 = einsum(equation = var_42815_equation_0, values = (var_42659_cast_fp16, var_42782_cast_fp16))[name = tensor("op_42815_cast_fp16")]; + tensor var_42817_equation_0 = const()[name = tensor("op_42817_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42817_cast_fp16 = einsum(equation = var_42817_equation_0, values = (var_42663_cast_fp16, var_42783_cast_fp16))[name = tensor("op_42817_cast_fp16")]; + tensor var_42819_equation_0 = const()[name = tensor("op_42819_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42819_cast_fp16 = einsum(equation = var_42819_equation_0, values = (var_42667_cast_fp16, var_42784_cast_fp16))[name = tensor("op_42819_cast_fp16")]; + tensor var_42821_equation_0 = const()[name = tensor("op_42821_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42821_cast_fp16 = einsum(equation = var_42821_equation_0, values = (var_42671_cast_fp16, var_42785_cast_fp16))[name = tensor("op_42821_cast_fp16")]; + tensor var_42823_equation_0 = const()[name = tensor("op_42823_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42823_cast_fp16 = einsum(equation = var_42823_equation_0, values = (var_42675_cast_fp16, var_42786_cast_fp16))[name = tensor("op_42823_cast_fp16")]; + tensor var_42825_equation_0 = const()[name = tensor("op_42825_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42825_cast_fp16 = einsum(equation = var_42825_equation_0, values = (var_42679_cast_fp16, var_42787_cast_fp16))[name = tensor("op_42825_cast_fp16")]; + tensor var_42827_equation_0 = const()[name = tensor("op_42827_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42827_cast_fp16 = einsum(equation = var_42827_equation_0, values = (var_42683_cast_fp16, var_42788_cast_fp16))[name = tensor("op_42827_cast_fp16")]; + tensor var_42829_equation_0 = const()[name = tensor("op_42829_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42829_cast_fp16 = einsum(equation = var_42829_equation_0, values = (var_42687_cast_fp16, var_42789_cast_fp16))[name = tensor("op_42829_cast_fp16")]; + tensor input_561_interleave_0 = const()[name = tensor("input_561_interleave_0"), val = tensor(false)]; + tensor input_561_cast_fp16 = concat(axis = var_30355, interleave = input_561_interleave_0, values = (var_42791_cast_fp16, var_42793_cast_fp16, var_42795_cast_fp16, var_42797_cast_fp16, var_42799_cast_fp16, var_42801_cast_fp16, var_42803_cast_fp16, var_42805_cast_fp16, var_42807_cast_fp16, var_42809_cast_fp16, var_42811_cast_fp16, var_42813_cast_fp16, var_42815_cast_fp16, var_42817_cast_fp16, var_42819_cast_fp16, var_42821_cast_fp16, var_42823_cast_fp16, var_42825_cast_fp16, var_42827_cast_fp16, var_42829_cast_fp16))[name = tensor("input_561_cast_fp16")]; + tensor var_42835 = const()[name = tensor("op_42835"), val = tensor([1, 1])]; + tensor var_42837 = const()[name = tensor("op_42837"), val = tensor([1, 1])]; + tensor var_42839_pad_type_0 = const()[name = tensor("op_42839_pad_type_0"), val = tensor("custom")]; + tensor var_42839_pad_0 = const()[name = tensor("op_42839_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_3_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_3_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3606820288)))]; + tensor up_blocks_0_attentions_1_transformer_blocks_3_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_3_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3610097152)))]; + tensor var_42839_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_3_attn1_to_out_0_bias_to_fp16, dilations = var_42837, groups = var_30355, pad = var_42839_pad_0, pad_type = var_42839_pad_type_0, strides = var_42835, weight = up_blocks_0_attentions_1_transformer_blocks_3_attn1_to_out_0_weight_to_fp16, x = input_561_cast_fp16)[name = tensor("op_42839_cast_fp16")]; + tensor inputs_285_cast_fp16 = add(x = var_42839_cast_fp16, y = inputs_283_cast_fp16)[name = tensor("inputs_285_cast_fp16")]; + tensor hidden_states_381_axes_0 = const()[name = tensor("hidden_states_381_axes_0"), val = tensor([1])]; + tensor hidden_states_381_gamma_0_to_fp16 = const()[name = tensor("hidden_states_381_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3610099776)))]; + tensor hidden_states_381_beta_0_to_fp16 = const()[name = tensor("hidden_states_381_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3610102400)))]; + tensor var_42849_to_fp16 = const()[name = tensor("op_42849_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_381_cast_fp16 = layer_norm(axes = hidden_states_381_axes_0, beta = hidden_states_381_beta_0_to_fp16, epsilon = var_42849_to_fp16, gamma = hidden_states_381_gamma_0_to_fp16, x = inputs_285_cast_fp16)[name = tensor("hidden_states_381_cast_fp16")]; + tensor var_42864 = const()[name = tensor("op_42864"), val = tensor([1, 1])]; + tensor var_42866 = const()[name = tensor("op_42866"), val = tensor([1, 1])]; + tensor q_191_pad_type_0 = const()[name = tensor("q_191_pad_type_0"), val = tensor("custom")]; + tensor q_191_pad_0 = const()[name = tensor("q_191_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_3_attn2_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_3_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3610105024)))]; + tensor q_191_cast_fp16 = conv(dilations = var_42866, groups = var_30355, pad = q_191_pad_0, pad_type = q_191_pad_type_0, strides = var_42864, weight = up_blocks_0_attentions_1_transformer_blocks_3_attn2_to_q_weight_to_fp16, x = hidden_states_381_cast_fp16)[name = tensor("q_191_cast_fp16")]; + tensor var_42870 = const()[name = tensor("op_42870"), val = tensor([1, 1])]; + tensor var_42872 = const()[name = tensor("op_42872"), val = tensor([1, 1])]; + tensor k_381_pad_type_0 = const()[name = tensor("k_381_pad_type_0"), val = tensor("custom")]; + tensor k_381_pad_0 = const()[name = tensor("k_381_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_3_attn2_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_3_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3613381888)))]; + tensor k_381_cast_fp16 = conv(dilations = var_42872, groups = var_30355, pad = k_381_pad_0, pad_type = k_381_pad_type_0, strides = var_42870, weight = up_blocks_0_attentions_1_transformer_blocks_3_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_381_cast_fp16")]; + tensor var_42876 = const()[name = tensor("op_42876"), val = tensor([1, 1])]; + tensor var_42878 = const()[name = tensor("op_42878"), val = tensor([1, 1])]; + tensor v_191_pad_type_0 = const()[name = tensor("v_191_pad_type_0"), val = tensor("custom")]; + tensor v_191_pad_0 = const()[name = tensor("v_191_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_3_attn2_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_3_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3618624832)))]; + tensor v_191_cast_fp16 = conv(dilations = var_42878, groups = var_30355, pad = v_191_pad_0, pad_type = v_191_pad_type_0, strides = var_42876, weight = up_blocks_0_attentions_1_transformer_blocks_3_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_191_cast_fp16")]; + tensor var_42882_begin_0 = const()[name = tensor("op_42882_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_42882_end_0 = const()[name = tensor("op_42882_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_42882_end_mask_0 = const()[name = tensor("op_42882_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42882_cast_fp16 = slice_by_index(begin = var_42882_begin_0, end = var_42882_end_0, end_mask = var_42882_end_mask_0, x = q_191_cast_fp16)[name = tensor("op_42882_cast_fp16")]; + tensor var_42886_begin_0 = const()[name = tensor("op_42886_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_42886_end_0 = const()[name = tensor("op_42886_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_42886_end_mask_0 = const()[name = tensor("op_42886_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42886_cast_fp16 = slice_by_index(begin = var_42886_begin_0, end = var_42886_end_0, end_mask = var_42886_end_mask_0, x = q_191_cast_fp16)[name = tensor("op_42886_cast_fp16")]; + tensor var_42890_begin_0 = const()[name = tensor("op_42890_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_42890_end_0 = const()[name = tensor("op_42890_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_42890_end_mask_0 = const()[name = tensor("op_42890_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42890_cast_fp16 = slice_by_index(begin = var_42890_begin_0, end = var_42890_end_0, end_mask = var_42890_end_mask_0, x = q_191_cast_fp16)[name = tensor("op_42890_cast_fp16")]; + tensor var_42894_begin_0 = const()[name = tensor("op_42894_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_42894_end_0 = const()[name = tensor("op_42894_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_42894_end_mask_0 = const()[name = tensor("op_42894_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42894_cast_fp16 = slice_by_index(begin = var_42894_begin_0, end = var_42894_end_0, end_mask = var_42894_end_mask_0, x = q_191_cast_fp16)[name = tensor("op_42894_cast_fp16")]; + tensor var_42898_begin_0 = const()[name = tensor("op_42898_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_42898_end_0 = const()[name = tensor("op_42898_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_42898_end_mask_0 = const()[name = tensor("op_42898_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42898_cast_fp16 = slice_by_index(begin = var_42898_begin_0, end = var_42898_end_0, end_mask = var_42898_end_mask_0, x = q_191_cast_fp16)[name = tensor("op_42898_cast_fp16")]; + tensor var_42902_begin_0 = const()[name = tensor("op_42902_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_42902_end_0 = const()[name = tensor("op_42902_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_42902_end_mask_0 = const()[name = tensor("op_42902_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42902_cast_fp16 = slice_by_index(begin = var_42902_begin_0, end = var_42902_end_0, end_mask = var_42902_end_mask_0, x = q_191_cast_fp16)[name = tensor("op_42902_cast_fp16")]; + tensor var_42906_begin_0 = const()[name = tensor("op_42906_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_42906_end_0 = const()[name = tensor("op_42906_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_42906_end_mask_0 = const()[name = tensor("op_42906_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42906_cast_fp16 = slice_by_index(begin = var_42906_begin_0, end = var_42906_end_0, end_mask = var_42906_end_mask_0, x = q_191_cast_fp16)[name = tensor("op_42906_cast_fp16")]; + tensor var_42910_begin_0 = const()[name = tensor("op_42910_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_42910_end_0 = const()[name = tensor("op_42910_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_42910_end_mask_0 = const()[name = tensor("op_42910_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42910_cast_fp16 = slice_by_index(begin = var_42910_begin_0, end = var_42910_end_0, end_mask = var_42910_end_mask_0, x = q_191_cast_fp16)[name = tensor("op_42910_cast_fp16")]; + tensor var_42914_begin_0 = const()[name = tensor("op_42914_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_42914_end_0 = const()[name = tensor("op_42914_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_42914_end_mask_0 = const()[name = tensor("op_42914_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42914_cast_fp16 = slice_by_index(begin = var_42914_begin_0, end = var_42914_end_0, end_mask = var_42914_end_mask_0, x = q_191_cast_fp16)[name = tensor("op_42914_cast_fp16")]; + tensor var_42918_begin_0 = const()[name = tensor("op_42918_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_42918_end_0 = const()[name = tensor("op_42918_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_42918_end_mask_0 = const()[name = tensor("op_42918_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42918_cast_fp16 = slice_by_index(begin = var_42918_begin_0, end = var_42918_end_0, end_mask = var_42918_end_mask_0, x = q_191_cast_fp16)[name = tensor("op_42918_cast_fp16")]; + tensor var_42922_begin_0 = const()[name = tensor("op_42922_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_42922_end_0 = const()[name = tensor("op_42922_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_42922_end_mask_0 = const()[name = tensor("op_42922_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42922_cast_fp16 = slice_by_index(begin = var_42922_begin_0, end = var_42922_end_0, end_mask = var_42922_end_mask_0, x = q_191_cast_fp16)[name = tensor("op_42922_cast_fp16")]; + tensor var_42926_begin_0 = const()[name = tensor("op_42926_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_42926_end_0 = const()[name = tensor("op_42926_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_42926_end_mask_0 = const()[name = tensor("op_42926_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42926_cast_fp16 = slice_by_index(begin = var_42926_begin_0, end = var_42926_end_0, end_mask = var_42926_end_mask_0, x = q_191_cast_fp16)[name = tensor("op_42926_cast_fp16")]; + tensor var_42930_begin_0 = const()[name = tensor("op_42930_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_42930_end_0 = const()[name = tensor("op_42930_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_42930_end_mask_0 = const()[name = tensor("op_42930_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42930_cast_fp16 = slice_by_index(begin = var_42930_begin_0, end = var_42930_end_0, end_mask = var_42930_end_mask_0, x = q_191_cast_fp16)[name = tensor("op_42930_cast_fp16")]; + tensor var_42934_begin_0 = const()[name = tensor("op_42934_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_42934_end_0 = const()[name = tensor("op_42934_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_42934_end_mask_0 = const()[name = tensor("op_42934_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42934_cast_fp16 = slice_by_index(begin = var_42934_begin_0, end = var_42934_end_0, end_mask = var_42934_end_mask_0, x = q_191_cast_fp16)[name = tensor("op_42934_cast_fp16")]; + tensor var_42938_begin_0 = const()[name = tensor("op_42938_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_42938_end_0 = const()[name = tensor("op_42938_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_42938_end_mask_0 = const()[name = tensor("op_42938_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42938_cast_fp16 = slice_by_index(begin = var_42938_begin_0, end = var_42938_end_0, end_mask = var_42938_end_mask_0, x = q_191_cast_fp16)[name = tensor("op_42938_cast_fp16")]; + tensor var_42942_begin_0 = const()[name = tensor("op_42942_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_42942_end_0 = const()[name = tensor("op_42942_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_42942_end_mask_0 = const()[name = tensor("op_42942_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42942_cast_fp16 = slice_by_index(begin = var_42942_begin_0, end = var_42942_end_0, end_mask = var_42942_end_mask_0, x = q_191_cast_fp16)[name = tensor("op_42942_cast_fp16")]; + tensor var_42946_begin_0 = const()[name = tensor("op_42946_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_42946_end_0 = const()[name = tensor("op_42946_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_42946_end_mask_0 = const()[name = tensor("op_42946_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42946_cast_fp16 = slice_by_index(begin = var_42946_begin_0, end = var_42946_end_0, end_mask = var_42946_end_mask_0, x = q_191_cast_fp16)[name = tensor("op_42946_cast_fp16")]; + tensor var_42950_begin_0 = const()[name = tensor("op_42950_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_42950_end_0 = const()[name = tensor("op_42950_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_42950_end_mask_0 = const()[name = tensor("op_42950_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42950_cast_fp16 = slice_by_index(begin = var_42950_begin_0, end = var_42950_end_0, end_mask = var_42950_end_mask_0, x = q_191_cast_fp16)[name = tensor("op_42950_cast_fp16")]; + tensor var_42954_begin_0 = const()[name = tensor("op_42954_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_42954_end_0 = const()[name = tensor("op_42954_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_42954_end_mask_0 = const()[name = tensor("op_42954_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42954_cast_fp16 = slice_by_index(begin = var_42954_begin_0, end = var_42954_end_0, end_mask = var_42954_end_mask_0, x = q_191_cast_fp16)[name = tensor("op_42954_cast_fp16")]; + tensor var_42958_begin_0 = const()[name = tensor("op_42958_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_42958_end_0 = const()[name = tensor("op_42958_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_42958_end_mask_0 = const()[name = tensor("op_42958_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42958_cast_fp16 = slice_by_index(begin = var_42958_begin_0, end = var_42958_end_0, end_mask = var_42958_end_mask_0, x = q_191_cast_fp16)[name = tensor("op_42958_cast_fp16")]; + tensor k_383_perm_0 = const()[name = tensor("k_383_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_42965_begin_0 = const()[name = tensor("op_42965_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_42965_end_0 = const()[name = tensor("op_42965_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_42965_end_mask_0 = const()[name = tensor("op_42965_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_383_cast_fp16 = transpose(perm = k_383_perm_0, x = k_381_cast_fp16)[name = tensor("transpose_44")]; + tensor var_42965_cast_fp16 = slice_by_index(begin = var_42965_begin_0, end = var_42965_end_0, end_mask = var_42965_end_mask_0, x = k_383_cast_fp16)[name = tensor("op_42965_cast_fp16")]; + tensor var_42969_begin_0 = const()[name = tensor("op_42969_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_42969_end_0 = const()[name = tensor("op_42969_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_42969_end_mask_0 = const()[name = tensor("op_42969_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42969_cast_fp16 = slice_by_index(begin = var_42969_begin_0, end = var_42969_end_0, end_mask = var_42969_end_mask_0, x = k_383_cast_fp16)[name = tensor("op_42969_cast_fp16")]; + tensor var_42973_begin_0 = const()[name = tensor("op_42973_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_42973_end_0 = const()[name = tensor("op_42973_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_42973_end_mask_0 = const()[name = tensor("op_42973_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42973_cast_fp16 = slice_by_index(begin = var_42973_begin_0, end = var_42973_end_0, end_mask = var_42973_end_mask_0, x = k_383_cast_fp16)[name = tensor("op_42973_cast_fp16")]; + tensor var_42977_begin_0 = const()[name = tensor("op_42977_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_42977_end_0 = const()[name = tensor("op_42977_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_42977_end_mask_0 = const()[name = tensor("op_42977_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42977_cast_fp16 = slice_by_index(begin = var_42977_begin_0, end = var_42977_end_0, end_mask = var_42977_end_mask_0, x = k_383_cast_fp16)[name = tensor("op_42977_cast_fp16")]; + tensor var_42981_begin_0 = const()[name = tensor("op_42981_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_42981_end_0 = const()[name = tensor("op_42981_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_42981_end_mask_0 = const()[name = tensor("op_42981_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42981_cast_fp16 = slice_by_index(begin = var_42981_begin_0, end = var_42981_end_0, end_mask = var_42981_end_mask_0, x = k_383_cast_fp16)[name = tensor("op_42981_cast_fp16")]; + tensor var_42985_begin_0 = const()[name = tensor("op_42985_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_42985_end_0 = const()[name = tensor("op_42985_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_42985_end_mask_0 = const()[name = tensor("op_42985_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42985_cast_fp16 = slice_by_index(begin = var_42985_begin_0, end = var_42985_end_0, end_mask = var_42985_end_mask_0, x = k_383_cast_fp16)[name = tensor("op_42985_cast_fp16")]; + tensor var_42989_begin_0 = const()[name = tensor("op_42989_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_42989_end_0 = const()[name = tensor("op_42989_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_42989_end_mask_0 = const()[name = tensor("op_42989_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42989_cast_fp16 = slice_by_index(begin = var_42989_begin_0, end = var_42989_end_0, end_mask = var_42989_end_mask_0, x = k_383_cast_fp16)[name = tensor("op_42989_cast_fp16")]; + tensor var_42993_begin_0 = const()[name = tensor("op_42993_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_42993_end_0 = const()[name = tensor("op_42993_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_42993_end_mask_0 = const()[name = tensor("op_42993_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42993_cast_fp16 = slice_by_index(begin = var_42993_begin_0, end = var_42993_end_0, end_mask = var_42993_end_mask_0, x = k_383_cast_fp16)[name = tensor("op_42993_cast_fp16")]; + tensor var_42997_begin_0 = const()[name = tensor("op_42997_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_42997_end_0 = const()[name = tensor("op_42997_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_42997_end_mask_0 = const()[name = tensor("op_42997_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42997_cast_fp16 = slice_by_index(begin = var_42997_begin_0, end = var_42997_end_0, end_mask = var_42997_end_mask_0, x = k_383_cast_fp16)[name = tensor("op_42997_cast_fp16")]; + tensor var_43001_begin_0 = const()[name = tensor("op_43001_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_43001_end_0 = const()[name = tensor("op_43001_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_43001_end_mask_0 = const()[name = tensor("op_43001_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43001_cast_fp16 = slice_by_index(begin = var_43001_begin_0, end = var_43001_end_0, end_mask = var_43001_end_mask_0, x = k_383_cast_fp16)[name = tensor("op_43001_cast_fp16")]; + tensor var_43005_begin_0 = const()[name = tensor("op_43005_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_43005_end_0 = const()[name = tensor("op_43005_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_43005_end_mask_0 = const()[name = tensor("op_43005_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43005_cast_fp16 = slice_by_index(begin = var_43005_begin_0, end = var_43005_end_0, end_mask = var_43005_end_mask_0, x = k_383_cast_fp16)[name = tensor("op_43005_cast_fp16")]; + tensor var_43009_begin_0 = const()[name = tensor("op_43009_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_43009_end_0 = const()[name = tensor("op_43009_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_43009_end_mask_0 = const()[name = tensor("op_43009_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43009_cast_fp16 = slice_by_index(begin = var_43009_begin_0, end = var_43009_end_0, end_mask = var_43009_end_mask_0, x = k_383_cast_fp16)[name = tensor("op_43009_cast_fp16")]; + tensor var_43013_begin_0 = const()[name = tensor("op_43013_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_43013_end_0 = const()[name = tensor("op_43013_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_43013_end_mask_0 = const()[name = tensor("op_43013_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43013_cast_fp16 = slice_by_index(begin = var_43013_begin_0, end = var_43013_end_0, end_mask = var_43013_end_mask_0, x = k_383_cast_fp16)[name = tensor("op_43013_cast_fp16")]; + tensor var_43017_begin_0 = const()[name = tensor("op_43017_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_43017_end_0 = const()[name = tensor("op_43017_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_43017_end_mask_0 = const()[name = tensor("op_43017_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43017_cast_fp16 = slice_by_index(begin = var_43017_begin_0, end = var_43017_end_0, end_mask = var_43017_end_mask_0, x = k_383_cast_fp16)[name = tensor("op_43017_cast_fp16")]; + tensor var_43021_begin_0 = const()[name = tensor("op_43021_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_43021_end_0 = const()[name = tensor("op_43021_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_43021_end_mask_0 = const()[name = tensor("op_43021_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43021_cast_fp16 = slice_by_index(begin = var_43021_begin_0, end = var_43021_end_0, end_mask = var_43021_end_mask_0, x = k_383_cast_fp16)[name = tensor("op_43021_cast_fp16")]; + tensor var_43025_begin_0 = const()[name = tensor("op_43025_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_43025_end_0 = const()[name = tensor("op_43025_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_43025_end_mask_0 = const()[name = tensor("op_43025_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43025_cast_fp16 = slice_by_index(begin = var_43025_begin_0, end = var_43025_end_0, end_mask = var_43025_end_mask_0, x = k_383_cast_fp16)[name = tensor("op_43025_cast_fp16")]; + tensor var_43029_begin_0 = const()[name = tensor("op_43029_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_43029_end_0 = const()[name = tensor("op_43029_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_43029_end_mask_0 = const()[name = tensor("op_43029_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43029_cast_fp16 = slice_by_index(begin = var_43029_begin_0, end = var_43029_end_0, end_mask = var_43029_end_mask_0, x = k_383_cast_fp16)[name = tensor("op_43029_cast_fp16")]; + tensor var_43033_begin_0 = const()[name = tensor("op_43033_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_43033_end_0 = const()[name = tensor("op_43033_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_43033_end_mask_0 = const()[name = tensor("op_43033_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43033_cast_fp16 = slice_by_index(begin = var_43033_begin_0, end = var_43033_end_0, end_mask = var_43033_end_mask_0, x = k_383_cast_fp16)[name = tensor("op_43033_cast_fp16")]; + tensor var_43037_begin_0 = const()[name = tensor("op_43037_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_43037_end_0 = const()[name = tensor("op_43037_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_43037_end_mask_0 = const()[name = tensor("op_43037_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43037_cast_fp16 = slice_by_index(begin = var_43037_begin_0, end = var_43037_end_0, end_mask = var_43037_end_mask_0, x = k_383_cast_fp16)[name = tensor("op_43037_cast_fp16")]; + tensor var_43041_begin_0 = const()[name = tensor("op_43041_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_43041_end_0 = const()[name = tensor("op_43041_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_43041_end_mask_0 = const()[name = tensor("op_43041_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43041_cast_fp16 = slice_by_index(begin = var_43041_begin_0, end = var_43041_end_0, end_mask = var_43041_end_mask_0, x = k_383_cast_fp16)[name = tensor("op_43041_cast_fp16")]; + tensor var_43043_begin_0 = const()[name = tensor("op_43043_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43043_end_0 = const()[name = tensor("op_43043_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_43043_end_mask_0 = const()[name = tensor("op_43043_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43043_cast_fp16 = slice_by_index(begin = var_43043_begin_0, end = var_43043_end_0, end_mask = var_43043_end_mask_0, x = v_191_cast_fp16)[name = tensor("op_43043_cast_fp16")]; + tensor var_43047_begin_0 = const()[name = tensor("op_43047_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_43047_end_0 = const()[name = tensor("op_43047_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_43047_end_mask_0 = const()[name = tensor("op_43047_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43047_cast_fp16 = slice_by_index(begin = var_43047_begin_0, end = var_43047_end_0, end_mask = var_43047_end_mask_0, x = v_191_cast_fp16)[name = tensor("op_43047_cast_fp16")]; + tensor var_43051_begin_0 = const()[name = tensor("op_43051_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_43051_end_0 = const()[name = tensor("op_43051_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_43051_end_mask_0 = const()[name = tensor("op_43051_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43051_cast_fp16 = slice_by_index(begin = var_43051_begin_0, end = var_43051_end_0, end_mask = var_43051_end_mask_0, x = v_191_cast_fp16)[name = tensor("op_43051_cast_fp16")]; + tensor var_43055_begin_0 = const()[name = tensor("op_43055_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_43055_end_0 = const()[name = tensor("op_43055_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_43055_end_mask_0 = const()[name = tensor("op_43055_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43055_cast_fp16 = slice_by_index(begin = var_43055_begin_0, end = var_43055_end_0, end_mask = var_43055_end_mask_0, x = v_191_cast_fp16)[name = tensor("op_43055_cast_fp16")]; + tensor var_43059_begin_0 = const()[name = tensor("op_43059_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_43059_end_0 = const()[name = tensor("op_43059_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_43059_end_mask_0 = const()[name = tensor("op_43059_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43059_cast_fp16 = slice_by_index(begin = var_43059_begin_0, end = var_43059_end_0, end_mask = var_43059_end_mask_0, x = v_191_cast_fp16)[name = tensor("op_43059_cast_fp16")]; + tensor var_43063_begin_0 = const()[name = tensor("op_43063_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_43063_end_0 = const()[name = tensor("op_43063_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_43063_end_mask_0 = const()[name = tensor("op_43063_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43063_cast_fp16 = slice_by_index(begin = var_43063_begin_0, end = var_43063_end_0, end_mask = var_43063_end_mask_0, x = v_191_cast_fp16)[name = tensor("op_43063_cast_fp16")]; + tensor var_43067_begin_0 = const()[name = tensor("op_43067_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_43067_end_0 = const()[name = tensor("op_43067_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_43067_end_mask_0 = const()[name = tensor("op_43067_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43067_cast_fp16 = slice_by_index(begin = var_43067_begin_0, end = var_43067_end_0, end_mask = var_43067_end_mask_0, x = v_191_cast_fp16)[name = tensor("op_43067_cast_fp16")]; + tensor var_43071_begin_0 = const()[name = tensor("op_43071_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_43071_end_0 = const()[name = tensor("op_43071_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_43071_end_mask_0 = const()[name = tensor("op_43071_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43071_cast_fp16 = slice_by_index(begin = var_43071_begin_0, end = var_43071_end_0, end_mask = var_43071_end_mask_0, x = v_191_cast_fp16)[name = tensor("op_43071_cast_fp16")]; + tensor var_43075_begin_0 = const()[name = tensor("op_43075_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_43075_end_0 = const()[name = tensor("op_43075_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_43075_end_mask_0 = const()[name = tensor("op_43075_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43075_cast_fp16 = slice_by_index(begin = var_43075_begin_0, end = var_43075_end_0, end_mask = var_43075_end_mask_0, x = v_191_cast_fp16)[name = tensor("op_43075_cast_fp16")]; + tensor var_43079_begin_0 = const()[name = tensor("op_43079_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_43079_end_0 = const()[name = tensor("op_43079_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_43079_end_mask_0 = const()[name = tensor("op_43079_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43079_cast_fp16 = slice_by_index(begin = var_43079_begin_0, end = var_43079_end_0, end_mask = var_43079_end_mask_0, x = v_191_cast_fp16)[name = tensor("op_43079_cast_fp16")]; + tensor var_43083_begin_0 = const()[name = tensor("op_43083_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_43083_end_0 = const()[name = tensor("op_43083_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_43083_end_mask_0 = const()[name = tensor("op_43083_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43083_cast_fp16 = slice_by_index(begin = var_43083_begin_0, end = var_43083_end_0, end_mask = var_43083_end_mask_0, x = v_191_cast_fp16)[name = tensor("op_43083_cast_fp16")]; + tensor var_43087_begin_0 = const()[name = tensor("op_43087_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_43087_end_0 = const()[name = tensor("op_43087_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_43087_end_mask_0 = const()[name = tensor("op_43087_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43087_cast_fp16 = slice_by_index(begin = var_43087_begin_0, end = var_43087_end_0, end_mask = var_43087_end_mask_0, x = v_191_cast_fp16)[name = tensor("op_43087_cast_fp16")]; + tensor var_43091_begin_0 = const()[name = tensor("op_43091_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_43091_end_0 = const()[name = tensor("op_43091_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_43091_end_mask_0 = const()[name = tensor("op_43091_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43091_cast_fp16 = slice_by_index(begin = var_43091_begin_0, end = var_43091_end_0, end_mask = var_43091_end_mask_0, x = v_191_cast_fp16)[name = tensor("op_43091_cast_fp16")]; + tensor var_43095_begin_0 = const()[name = tensor("op_43095_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_43095_end_0 = const()[name = tensor("op_43095_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_43095_end_mask_0 = const()[name = tensor("op_43095_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43095_cast_fp16 = slice_by_index(begin = var_43095_begin_0, end = var_43095_end_0, end_mask = var_43095_end_mask_0, x = v_191_cast_fp16)[name = tensor("op_43095_cast_fp16")]; + tensor var_43099_begin_0 = const()[name = tensor("op_43099_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_43099_end_0 = const()[name = tensor("op_43099_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_43099_end_mask_0 = const()[name = tensor("op_43099_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43099_cast_fp16 = slice_by_index(begin = var_43099_begin_0, end = var_43099_end_0, end_mask = var_43099_end_mask_0, x = v_191_cast_fp16)[name = tensor("op_43099_cast_fp16")]; + tensor var_43103_begin_0 = const()[name = tensor("op_43103_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_43103_end_0 = const()[name = tensor("op_43103_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_43103_end_mask_0 = const()[name = tensor("op_43103_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43103_cast_fp16 = slice_by_index(begin = var_43103_begin_0, end = var_43103_end_0, end_mask = var_43103_end_mask_0, x = v_191_cast_fp16)[name = tensor("op_43103_cast_fp16")]; + tensor var_43107_begin_0 = const()[name = tensor("op_43107_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_43107_end_0 = const()[name = tensor("op_43107_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_43107_end_mask_0 = const()[name = tensor("op_43107_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43107_cast_fp16 = slice_by_index(begin = var_43107_begin_0, end = var_43107_end_0, end_mask = var_43107_end_mask_0, x = v_191_cast_fp16)[name = tensor("op_43107_cast_fp16")]; + tensor var_43111_begin_0 = const()[name = tensor("op_43111_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_43111_end_0 = const()[name = tensor("op_43111_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_43111_end_mask_0 = const()[name = tensor("op_43111_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43111_cast_fp16 = slice_by_index(begin = var_43111_begin_0, end = var_43111_end_0, end_mask = var_43111_end_mask_0, x = v_191_cast_fp16)[name = tensor("op_43111_cast_fp16")]; + tensor var_43115_begin_0 = const()[name = tensor("op_43115_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_43115_end_0 = const()[name = tensor("op_43115_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_43115_end_mask_0 = const()[name = tensor("op_43115_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43115_cast_fp16 = slice_by_index(begin = var_43115_begin_0, end = var_43115_end_0, end_mask = var_43115_end_mask_0, x = v_191_cast_fp16)[name = tensor("op_43115_cast_fp16")]; + tensor var_43119_begin_0 = const()[name = tensor("op_43119_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_43119_end_0 = const()[name = tensor("op_43119_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_43119_end_mask_0 = const()[name = tensor("op_43119_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43119_cast_fp16 = slice_by_index(begin = var_43119_begin_0, end = var_43119_end_0, end_mask = var_43119_end_mask_0, x = v_191_cast_fp16)[name = tensor("op_43119_cast_fp16")]; + tensor var_43123_equation_0 = const()[name = tensor("op_43123_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43123_cast_fp16 = einsum(equation = var_43123_equation_0, values = (var_42965_cast_fp16, var_42882_cast_fp16))[name = tensor("op_43123_cast_fp16")]; + tensor var_43124_to_fp16 = const()[name = tensor("op_43124_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3641_cast_fp16 = mul(x = var_43123_cast_fp16, y = var_43124_to_fp16)[name = tensor("aw_3641_cast_fp16")]; + tensor var_43127_equation_0 = const()[name = tensor("op_43127_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43127_cast_fp16 = einsum(equation = var_43127_equation_0, values = (var_42969_cast_fp16, var_42886_cast_fp16))[name = tensor("op_43127_cast_fp16")]; + tensor var_43128_to_fp16 = const()[name = tensor("op_43128_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3643_cast_fp16 = mul(x = var_43127_cast_fp16, y = var_43128_to_fp16)[name = tensor("aw_3643_cast_fp16")]; + tensor var_43131_equation_0 = const()[name = tensor("op_43131_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43131_cast_fp16 = einsum(equation = var_43131_equation_0, values = (var_42973_cast_fp16, var_42890_cast_fp16))[name = tensor("op_43131_cast_fp16")]; + tensor var_43132_to_fp16 = const()[name = tensor("op_43132_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3645_cast_fp16 = mul(x = var_43131_cast_fp16, y = var_43132_to_fp16)[name = tensor("aw_3645_cast_fp16")]; + tensor var_43135_equation_0 = const()[name = tensor("op_43135_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43135_cast_fp16 = einsum(equation = var_43135_equation_0, values = (var_42977_cast_fp16, var_42894_cast_fp16))[name = tensor("op_43135_cast_fp16")]; + tensor var_43136_to_fp16 = const()[name = tensor("op_43136_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3647_cast_fp16 = mul(x = var_43135_cast_fp16, y = var_43136_to_fp16)[name = tensor("aw_3647_cast_fp16")]; + tensor var_43139_equation_0 = const()[name = tensor("op_43139_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43139_cast_fp16 = einsum(equation = var_43139_equation_0, values = (var_42981_cast_fp16, var_42898_cast_fp16))[name = tensor("op_43139_cast_fp16")]; + tensor var_43140_to_fp16 = const()[name = tensor("op_43140_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3649_cast_fp16 = mul(x = var_43139_cast_fp16, y = var_43140_to_fp16)[name = tensor("aw_3649_cast_fp16")]; + tensor var_43143_equation_0 = const()[name = tensor("op_43143_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43143_cast_fp16 = einsum(equation = var_43143_equation_0, values = (var_42985_cast_fp16, var_42902_cast_fp16))[name = tensor("op_43143_cast_fp16")]; + tensor var_43144_to_fp16 = const()[name = tensor("op_43144_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3651_cast_fp16 = mul(x = var_43143_cast_fp16, y = var_43144_to_fp16)[name = tensor("aw_3651_cast_fp16")]; + tensor var_43147_equation_0 = const()[name = tensor("op_43147_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43147_cast_fp16 = einsum(equation = var_43147_equation_0, values = (var_42989_cast_fp16, var_42906_cast_fp16))[name = tensor("op_43147_cast_fp16")]; + tensor var_43148_to_fp16 = const()[name = tensor("op_43148_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3653_cast_fp16 = mul(x = var_43147_cast_fp16, y = var_43148_to_fp16)[name = tensor("aw_3653_cast_fp16")]; + tensor var_43151_equation_0 = const()[name = tensor("op_43151_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43151_cast_fp16 = einsum(equation = var_43151_equation_0, values = (var_42993_cast_fp16, var_42910_cast_fp16))[name = tensor("op_43151_cast_fp16")]; + tensor var_43152_to_fp16 = const()[name = tensor("op_43152_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3655_cast_fp16 = mul(x = var_43151_cast_fp16, y = var_43152_to_fp16)[name = tensor("aw_3655_cast_fp16")]; + tensor var_43155_equation_0 = const()[name = tensor("op_43155_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43155_cast_fp16 = einsum(equation = var_43155_equation_0, values = (var_42997_cast_fp16, var_42914_cast_fp16))[name = tensor("op_43155_cast_fp16")]; + tensor var_43156_to_fp16 = const()[name = tensor("op_43156_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3657_cast_fp16 = mul(x = var_43155_cast_fp16, y = var_43156_to_fp16)[name = tensor("aw_3657_cast_fp16")]; + tensor var_43159_equation_0 = const()[name = tensor("op_43159_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43159_cast_fp16 = einsum(equation = var_43159_equation_0, values = (var_43001_cast_fp16, var_42918_cast_fp16))[name = tensor("op_43159_cast_fp16")]; + tensor var_43160_to_fp16 = const()[name = tensor("op_43160_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3659_cast_fp16 = mul(x = var_43159_cast_fp16, y = var_43160_to_fp16)[name = tensor("aw_3659_cast_fp16")]; + tensor var_43163_equation_0 = const()[name = tensor("op_43163_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43163_cast_fp16 = einsum(equation = var_43163_equation_0, values = (var_43005_cast_fp16, var_42922_cast_fp16))[name = tensor("op_43163_cast_fp16")]; + tensor var_43164_to_fp16 = const()[name = tensor("op_43164_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3661_cast_fp16 = mul(x = var_43163_cast_fp16, y = var_43164_to_fp16)[name = tensor("aw_3661_cast_fp16")]; + tensor var_43167_equation_0 = const()[name = tensor("op_43167_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43167_cast_fp16 = einsum(equation = var_43167_equation_0, values = (var_43009_cast_fp16, var_42926_cast_fp16))[name = tensor("op_43167_cast_fp16")]; + tensor var_43168_to_fp16 = const()[name = tensor("op_43168_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3663_cast_fp16 = mul(x = var_43167_cast_fp16, y = var_43168_to_fp16)[name = tensor("aw_3663_cast_fp16")]; + tensor var_43171_equation_0 = const()[name = tensor("op_43171_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43171_cast_fp16 = einsum(equation = var_43171_equation_0, values = (var_43013_cast_fp16, var_42930_cast_fp16))[name = tensor("op_43171_cast_fp16")]; + tensor var_43172_to_fp16 = const()[name = tensor("op_43172_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3665_cast_fp16 = mul(x = var_43171_cast_fp16, y = var_43172_to_fp16)[name = tensor("aw_3665_cast_fp16")]; + tensor var_43175_equation_0 = const()[name = tensor("op_43175_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43175_cast_fp16 = einsum(equation = var_43175_equation_0, values = (var_43017_cast_fp16, var_42934_cast_fp16))[name = tensor("op_43175_cast_fp16")]; + tensor var_43176_to_fp16 = const()[name = tensor("op_43176_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3667_cast_fp16 = mul(x = var_43175_cast_fp16, y = var_43176_to_fp16)[name = tensor("aw_3667_cast_fp16")]; + tensor var_43179_equation_0 = const()[name = tensor("op_43179_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43179_cast_fp16 = einsum(equation = var_43179_equation_0, values = (var_43021_cast_fp16, var_42938_cast_fp16))[name = tensor("op_43179_cast_fp16")]; + tensor var_43180_to_fp16 = const()[name = tensor("op_43180_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3669_cast_fp16 = mul(x = var_43179_cast_fp16, y = var_43180_to_fp16)[name = tensor("aw_3669_cast_fp16")]; + tensor var_43183_equation_0 = const()[name = tensor("op_43183_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43183_cast_fp16 = einsum(equation = var_43183_equation_0, values = (var_43025_cast_fp16, var_42942_cast_fp16))[name = tensor("op_43183_cast_fp16")]; + tensor var_43184_to_fp16 = const()[name = tensor("op_43184_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3671_cast_fp16 = mul(x = var_43183_cast_fp16, y = var_43184_to_fp16)[name = tensor("aw_3671_cast_fp16")]; + tensor var_43187_equation_0 = const()[name = tensor("op_43187_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43187_cast_fp16 = einsum(equation = var_43187_equation_0, values = (var_43029_cast_fp16, var_42946_cast_fp16))[name = tensor("op_43187_cast_fp16")]; + tensor var_43188_to_fp16 = const()[name = tensor("op_43188_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3673_cast_fp16 = mul(x = var_43187_cast_fp16, y = var_43188_to_fp16)[name = tensor("aw_3673_cast_fp16")]; + tensor var_43191_equation_0 = const()[name = tensor("op_43191_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43191_cast_fp16 = einsum(equation = var_43191_equation_0, values = (var_43033_cast_fp16, var_42950_cast_fp16))[name = tensor("op_43191_cast_fp16")]; + tensor var_43192_to_fp16 = const()[name = tensor("op_43192_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3675_cast_fp16 = mul(x = var_43191_cast_fp16, y = var_43192_to_fp16)[name = tensor("aw_3675_cast_fp16")]; + tensor var_43195_equation_0 = const()[name = tensor("op_43195_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43195_cast_fp16 = einsum(equation = var_43195_equation_0, values = (var_43037_cast_fp16, var_42954_cast_fp16))[name = tensor("op_43195_cast_fp16")]; + tensor var_43196_to_fp16 = const()[name = tensor("op_43196_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3677_cast_fp16 = mul(x = var_43195_cast_fp16, y = var_43196_to_fp16)[name = tensor("aw_3677_cast_fp16")]; + tensor var_43199_equation_0 = const()[name = tensor("op_43199_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43199_cast_fp16 = einsum(equation = var_43199_equation_0, values = (var_43041_cast_fp16, var_42958_cast_fp16))[name = tensor("op_43199_cast_fp16")]; + tensor var_43200_to_fp16 = const()[name = tensor("op_43200_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3679_cast_fp16 = mul(x = var_43199_cast_fp16, y = var_43200_to_fp16)[name = tensor("aw_3679_cast_fp16")]; + tensor var_43202_cast_fp16 = softmax(axis = var_30355, x = aw_3641_cast_fp16)[name = tensor("op_43202_cast_fp16")]; + tensor var_43203_cast_fp16 = softmax(axis = var_30355, x = aw_3643_cast_fp16)[name = tensor("op_43203_cast_fp16")]; + tensor var_43204_cast_fp16 = softmax(axis = var_30355, x = aw_3645_cast_fp16)[name = tensor("op_43204_cast_fp16")]; + tensor var_43205_cast_fp16 = softmax(axis = var_30355, x = aw_3647_cast_fp16)[name = tensor("op_43205_cast_fp16")]; + tensor var_43206_cast_fp16 = softmax(axis = var_30355, x = aw_3649_cast_fp16)[name = tensor("op_43206_cast_fp16")]; + tensor var_43207_cast_fp16 = softmax(axis = var_30355, x = aw_3651_cast_fp16)[name = tensor("op_43207_cast_fp16")]; + tensor var_43208_cast_fp16 = softmax(axis = var_30355, x = aw_3653_cast_fp16)[name = tensor("op_43208_cast_fp16")]; + tensor var_43209_cast_fp16 = softmax(axis = var_30355, x = aw_3655_cast_fp16)[name = tensor("op_43209_cast_fp16")]; + tensor var_43210_cast_fp16 = softmax(axis = var_30355, x = aw_3657_cast_fp16)[name = tensor("op_43210_cast_fp16")]; + tensor var_43211_cast_fp16 = softmax(axis = var_30355, x = aw_3659_cast_fp16)[name = tensor("op_43211_cast_fp16")]; + tensor var_43212_cast_fp16 = softmax(axis = var_30355, x = aw_3661_cast_fp16)[name = tensor("op_43212_cast_fp16")]; + tensor var_43213_cast_fp16 = softmax(axis = var_30355, x = aw_3663_cast_fp16)[name = tensor("op_43213_cast_fp16")]; + tensor var_43214_cast_fp16 = softmax(axis = var_30355, x = aw_3665_cast_fp16)[name = tensor("op_43214_cast_fp16")]; + tensor var_43215_cast_fp16 = softmax(axis = var_30355, x = aw_3667_cast_fp16)[name = tensor("op_43215_cast_fp16")]; + tensor var_43216_cast_fp16 = softmax(axis = var_30355, x = aw_3669_cast_fp16)[name = tensor("op_43216_cast_fp16")]; + tensor var_43217_cast_fp16 = softmax(axis = var_30355, x = aw_3671_cast_fp16)[name = tensor("op_43217_cast_fp16")]; + tensor var_43218_cast_fp16 = softmax(axis = var_30355, x = aw_3673_cast_fp16)[name = tensor("op_43218_cast_fp16")]; + tensor var_43219_cast_fp16 = softmax(axis = var_30355, x = aw_3675_cast_fp16)[name = tensor("op_43219_cast_fp16")]; + tensor var_43220_cast_fp16 = softmax(axis = var_30355, x = aw_3677_cast_fp16)[name = tensor("op_43220_cast_fp16")]; + tensor var_43221_cast_fp16 = softmax(axis = var_30355, x = aw_3679_cast_fp16)[name = tensor("op_43221_cast_fp16")]; + tensor var_43223_equation_0 = const()[name = tensor("op_43223_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_43223_cast_fp16 = einsum(equation = var_43223_equation_0, values = (var_43043_cast_fp16, var_43202_cast_fp16))[name = tensor("op_43223_cast_fp16")]; + tensor var_43225_equation_0 = const()[name = tensor("op_43225_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_43225_cast_fp16 = einsum(equation = var_43225_equation_0, values = (var_43047_cast_fp16, var_43203_cast_fp16))[name = tensor("op_43225_cast_fp16")]; + tensor var_43227_equation_0 = const()[name = tensor("op_43227_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_43227_cast_fp16 = einsum(equation = var_43227_equation_0, values = (var_43051_cast_fp16, var_43204_cast_fp16))[name = tensor("op_43227_cast_fp16")]; + tensor var_43229_equation_0 = const()[name = tensor("op_43229_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_43229_cast_fp16 = einsum(equation = var_43229_equation_0, values = (var_43055_cast_fp16, var_43205_cast_fp16))[name = tensor("op_43229_cast_fp16")]; + tensor var_43231_equation_0 = const()[name = tensor("op_43231_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_43231_cast_fp16 = einsum(equation = var_43231_equation_0, values = (var_43059_cast_fp16, var_43206_cast_fp16))[name = tensor("op_43231_cast_fp16")]; + tensor var_43233_equation_0 = const()[name = tensor("op_43233_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_43233_cast_fp16 = einsum(equation = var_43233_equation_0, values = (var_43063_cast_fp16, var_43207_cast_fp16))[name = tensor("op_43233_cast_fp16")]; + tensor var_43235_equation_0 = const()[name = tensor("op_43235_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_43235_cast_fp16 = einsum(equation = var_43235_equation_0, values = (var_43067_cast_fp16, var_43208_cast_fp16))[name = tensor("op_43235_cast_fp16")]; + tensor var_43237_equation_0 = const()[name = tensor("op_43237_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_43237_cast_fp16 = einsum(equation = var_43237_equation_0, values = (var_43071_cast_fp16, var_43209_cast_fp16))[name = tensor("op_43237_cast_fp16")]; + tensor var_43239_equation_0 = const()[name = tensor("op_43239_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_43239_cast_fp16 = einsum(equation = var_43239_equation_0, values = (var_43075_cast_fp16, var_43210_cast_fp16))[name = tensor("op_43239_cast_fp16")]; + tensor var_43241_equation_0 = const()[name = tensor("op_43241_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_43241_cast_fp16 = einsum(equation = var_43241_equation_0, values = (var_43079_cast_fp16, var_43211_cast_fp16))[name = tensor("op_43241_cast_fp16")]; + tensor var_43243_equation_0 = const()[name = tensor("op_43243_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_43243_cast_fp16 = einsum(equation = var_43243_equation_0, values = (var_43083_cast_fp16, var_43212_cast_fp16))[name = tensor("op_43243_cast_fp16")]; + tensor var_43245_equation_0 = const()[name = tensor("op_43245_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_43245_cast_fp16 = einsum(equation = var_43245_equation_0, values = (var_43087_cast_fp16, var_43213_cast_fp16))[name = tensor("op_43245_cast_fp16")]; + tensor var_43247_equation_0 = const()[name = tensor("op_43247_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_43247_cast_fp16 = einsum(equation = var_43247_equation_0, values = (var_43091_cast_fp16, var_43214_cast_fp16))[name = tensor("op_43247_cast_fp16")]; + tensor var_43249_equation_0 = const()[name = tensor("op_43249_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_43249_cast_fp16 = einsum(equation = var_43249_equation_0, values = (var_43095_cast_fp16, var_43215_cast_fp16))[name = tensor("op_43249_cast_fp16")]; + tensor var_43251_equation_0 = const()[name = tensor("op_43251_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_43251_cast_fp16 = einsum(equation = var_43251_equation_0, values = (var_43099_cast_fp16, var_43216_cast_fp16))[name = tensor("op_43251_cast_fp16")]; + tensor var_43253_equation_0 = const()[name = tensor("op_43253_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_43253_cast_fp16 = einsum(equation = var_43253_equation_0, values = (var_43103_cast_fp16, var_43217_cast_fp16))[name = tensor("op_43253_cast_fp16")]; + tensor var_43255_equation_0 = const()[name = tensor("op_43255_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_43255_cast_fp16 = einsum(equation = var_43255_equation_0, values = (var_43107_cast_fp16, var_43218_cast_fp16))[name = tensor("op_43255_cast_fp16")]; + tensor var_43257_equation_0 = const()[name = tensor("op_43257_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_43257_cast_fp16 = einsum(equation = var_43257_equation_0, values = (var_43111_cast_fp16, var_43219_cast_fp16))[name = tensor("op_43257_cast_fp16")]; + tensor var_43259_equation_0 = const()[name = tensor("op_43259_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_43259_cast_fp16 = einsum(equation = var_43259_equation_0, values = (var_43115_cast_fp16, var_43220_cast_fp16))[name = tensor("op_43259_cast_fp16")]; + tensor var_43261_equation_0 = const()[name = tensor("op_43261_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_43261_cast_fp16 = einsum(equation = var_43261_equation_0, values = (var_43119_cast_fp16, var_43221_cast_fp16))[name = tensor("op_43261_cast_fp16")]; + tensor input_563_interleave_0 = const()[name = tensor("input_563_interleave_0"), val = tensor(false)]; + tensor input_563_cast_fp16 = concat(axis = var_30355, interleave = input_563_interleave_0, values = (var_43223_cast_fp16, var_43225_cast_fp16, var_43227_cast_fp16, var_43229_cast_fp16, var_43231_cast_fp16, var_43233_cast_fp16, var_43235_cast_fp16, var_43237_cast_fp16, var_43239_cast_fp16, var_43241_cast_fp16, var_43243_cast_fp16, var_43245_cast_fp16, var_43247_cast_fp16, var_43249_cast_fp16, var_43251_cast_fp16, var_43253_cast_fp16, var_43255_cast_fp16, var_43257_cast_fp16, var_43259_cast_fp16, var_43261_cast_fp16))[name = tensor("input_563_cast_fp16")]; + tensor var_43267 = const()[name = tensor("op_43267"), val = tensor([1, 1])]; + tensor var_43269 = const()[name = tensor("op_43269"), val = tensor([1, 1])]; + tensor var_43271_pad_type_0 = const()[name = tensor("op_43271_pad_type_0"), val = tensor("custom")]; + tensor var_43271_pad_0 = const()[name = tensor("op_43271_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_3_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_3_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3623867776)))]; + tensor up_blocks_0_attentions_1_transformer_blocks_3_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_3_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3627144640)))]; + tensor var_43271_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_3_attn2_to_out_0_bias_to_fp16, dilations = var_43269, groups = var_30355, pad = var_43271_pad_0, pad_type = var_43271_pad_type_0, strides = var_43267, weight = up_blocks_0_attentions_1_transformer_blocks_3_attn2_to_out_0_weight_to_fp16, x = input_563_cast_fp16)[name = tensor("op_43271_cast_fp16")]; + tensor inputs_287_cast_fp16 = add(x = var_43271_cast_fp16, y = inputs_285_cast_fp16)[name = tensor("inputs_287_cast_fp16")]; + tensor input_565_axes_0 = const()[name = tensor("input_565_axes_0"), val = tensor([1])]; + tensor input_565_gamma_0_to_fp16 = const()[name = tensor("input_565_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3627147264)))]; + tensor input_565_beta_0_to_fp16 = const()[name = tensor("input_565_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3627149888)))]; + tensor var_43281_to_fp16 = const()[name = tensor("op_43281_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_565_cast_fp16 = layer_norm(axes = input_565_axes_0, beta = input_565_beta_0_to_fp16, epsilon = var_43281_to_fp16, gamma = input_565_gamma_0_to_fp16, x = inputs_287_cast_fp16)[name = tensor("input_565_cast_fp16")]; + tensor var_43297 = const()[name = tensor("op_43297"), val = tensor([1, 1])]; + tensor var_43299 = const()[name = tensor("op_43299"), val = tensor([1, 1])]; + tensor var_43301_pad_type_0 = const()[name = tensor("op_43301_pad_type_0"), val = tensor("custom")]; + tensor var_43301_pad_0 = const()[name = tensor("op_43301_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_3_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_3_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3627152512)))]; + tensor up_blocks_0_attentions_1_transformer_blocks_3_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_3_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3653366976)))]; + tensor var_43301_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_3_ff_net_0_proj_bias_to_fp16, dilations = var_43299, groups = var_30355, pad = var_43301_pad_0, pad_type = var_43301_pad_type_0, strides = var_43297, weight = up_blocks_0_attentions_1_transformer_blocks_3_ff_net_0_proj_weight_to_fp16, x = input_565_cast_fp16)[name = tensor("op_43301_cast_fp16")]; + tensor var_43302_split_sizes_0 = const()[name = tensor("op_43302_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_43302_axis_0 = const()[name = tensor("op_43302_axis_0"), val = tensor(1)]; + tensor var_43302_cast_fp16_0, tensor var_43302_cast_fp16_1 = split(axis = var_43302_axis_0, split_sizes = var_43302_split_sizes_0, x = var_43301_cast_fp16)[name = tensor("op_43302_cast_fp16")]; + tensor var_43304_mode_0 = const()[name = tensor("op_43304_mode_0"), val = tensor("EXACT")]; + tensor var_43304_cast_fp16 = gelu(mode = var_43304_mode_0, x = var_43302_cast_fp16_1)[name = tensor("op_43304_cast_fp16")]; + tensor input_567_cast_fp16 = mul(x = var_43302_cast_fp16_0, y = var_43304_cast_fp16)[name = tensor("input_567_cast_fp16")]; + tensor var_43308 = const()[name = tensor("op_43308"), val = tensor([1, 1])]; + tensor var_43310 = const()[name = tensor("op_43310"), val = tensor([1, 1])]; + tensor var_43312_pad_type_0 = const()[name = tensor("op_43312_pad_type_0"), val = tensor("custom")]; + tensor var_43312_pad_0 = const()[name = tensor("op_43312_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_3_ff_net_2_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_3_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3653387520)))]; + tensor up_blocks_0_attentions_1_transformer_blocks_3_ff_net_2_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_3_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3666494784)))]; + tensor var_43312_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_3_ff_net_2_bias_to_fp16, dilations = var_43310, groups = var_30355, pad = var_43312_pad_0, pad_type = var_43312_pad_type_0, strides = var_43308, weight = up_blocks_0_attentions_1_transformer_blocks_3_ff_net_2_weight_to_fp16, x = input_567_cast_fp16)[name = tensor("op_43312_cast_fp16")]; + tensor inputs_289_cast_fp16 = add(x = var_43312_cast_fp16, y = inputs_287_cast_fp16)[name = tensor("inputs_289_cast_fp16")]; + tensor hidden_states_385_axes_0 = const()[name = tensor("hidden_states_385_axes_0"), val = tensor([1])]; + tensor hidden_states_385_gamma_0_to_fp16 = const()[name = tensor("hidden_states_385_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3666497408)))]; + tensor hidden_states_385_beta_0_to_fp16 = const()[name = tensor("hidden_states_385_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3666500032)))]; + tensor var_43328_to_fp16 = const()[name = tensor("op_43328_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_385_cast_fp16 = layer_norm(axes = hidden_states_385_axes_0, beta = hidden_states_385_beta_0_to_fp16, epsilon = var_43328_to_fp16, gamma = hidden_states_385_gamma_0_to_fp16, x = inputs_289_cast_fp16)[name = tensor("hidden_states_385_cast_fp16")]; + tensor var_43343 = const()[name = tensor("op_43343"), val = tensor([1, 1])]; + tensor var_43345 = const()[name = tensor("op_43345"), val = tensor([1, 1])]; + tensor q_193_pad_type_0 = const()[name = tensor("q_193_pad_type_0"), val = tensor("custom")]; + tensor q_193_pad_0 = const()[name = tensor("q_193_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_4_attn1_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_4_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3666502656)))]; + tensor q_193_cast_fp16 = conv(dilations = var_43345, groups = var_30355, pad = q_193_pad_0, pad_type = q_193_pad_type_0, strides = var_43343, weight = up_blocks_0_attentions_1_transformer_blocks_4_attn1_to_q_weight_to_fp16, x = hidden_states_385_cast_fp16)[name = tensor("q_193_cast_fp16")]; + tensor var_43349 = const()[name = tensor("op_43349"), val = tensor([1, 1])]; + tensor var_43351 = const()[name = tensor("op_43351"), val = tensor([1, 1])]; + tensor k_385_pad_type_0 = const()[name = tensor("k_385_pad_type_0"), val = tensor("custom")]; + tensor k_385_pad_0 = const()[name = tensor("k_385_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_4_attn1_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_4_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3669779520)))]; + tensor k_385_cast_fp16 = conv(dilations = var_43351, groups = var_30355, pad = k_385_pad_0, pad_type = k_385_pad_type_0, strides = var_43349, weight = up_blocks_0_attentions_1_transformer_blocks_4_attn1_to_k_weight_to_fp16, x = hidden_states_385_cast_fp16)[name = tensor("k_385_cast_fp16")]; + tensor var_43355 = const()[name = tensor("op_43355"), val = tensor([1, 1])]; + tensor var_43357 = const()[name = tensor("op_43357"), val = tensor([1, 1])]; + tensor v_193_pad_type_0 = const()[name = tensor("v_193_pad_type_0"), val = tensor("custom")]; + tensor v_193_pad_0 = const()[name = tensor("v_193_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_4_attn1_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_4_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3673056384)))]; + tensor v_193_cast_fp16 = conv(dilations = var_43357, groups = var_30355, pad = v_193_pad_0, pad_type = v_193_pad_type_0, strides = var_43355, weight = up_blocks_0_attentions_1_transformer_blocks_4_attn1_to_v_weight_to_fp16, x = hidden_states_385_cast_fp16)[name = tensor("v_193_cast_fp16")]; + tensor var_43361_begin_0 = const()[name = tensor("op_43361_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43361_end_0 = const()[name = tensor("op_43361_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_43361_end_mask_0 = const()[name = tensor("op_43361_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43361_cast_fp16 = slice_by_index(begin = var_43361_begin_0, end = var_43361_end_0, end_mask = var_43361_end_mask_0, x = q_193_cast_fp16)[name = tensor("op_43361_cast_fp16")]; + tensor var_43365_begin_0 = const()[name = tensor("op_43365_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_43365_end_0 = const()[name = tensor("op_43365_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_43365_end_mask_0 = const()[name = tensor("op_43365_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43365_cast_fp16 = slice_by_index(begin = var_43365_begin_0, end = var_43365_end_0, end_mask = var_43365_end_mask_0, x = q_193_cast_fp16)[name = tensor("op_43365_cast_fp16")]; + tensor var_43369_begin_0 = const()[name = tensor("op_43369_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_43369_end_0 = const()[name = tensor("op_43369_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_43369_end_mask_0 = const()[name = tensor("op_43369_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43369_cast_fp16 = slice_by_index(begin = var_43369_begin_0, end = var_43369_end_0, end_mask = var_43369_end_mask_0, x = q_193_cast_fp16)[name = tensor("op_43369_cast_fp16")]; + tensor var_43373_begin_0 = const()[name = tensor("op_43373_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_43373_end_0 = const()[name = tensor("op_43373_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_43373_end_mask_0 = const()[name = tensor("op_43373_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43373_cast_fp16 = slice_by_index(begin = var_43373_begin_0, end = var_43373_end_0, end_mask = var_43373_end_mask_0, x = q_193_cast_fp16)[name = tensor("op_43373_cast_fp16")]; + tensor var_43377_begin_0 = const()[name = tensor("op_43377_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_43377_end_0 = const()[name = tensor("op_43377_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_43377_end_mask_0 = const()[name = tensor("op_43377_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43377_cast_fp16 = slice_by_index(begin = var_43377_begin_0, end = var_43377_end_0, end_mask = var_43377_end_mask_0, x = q_193_cast_fp16)[name = tensor("op_43377_cast_fp16")]; + tensor var_43381_begin_0 = const()[name = tensor("op_43381_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_43381_end_0 = const()[name = tensor("op_43381_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_43381_end_mask_0 = const()[name = tensor("op_43381_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43381_cast_fp16 = slice_by_index(begin = var_43381_begin_0, end = var_43381_end_0, end_mask = var_43381_end_mask_0, x = q_193_cast_fp16)[name = tensor("op_43381_cast_fp16")]; + tensor var_43385_begin_0 = const()[name = tensor("op_43385_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_43385_end_0 = const()[name = tensor("op_43385_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_43385_end_mask_0 = const()[name = tensor("op_43385_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43385_cast_fp16 = slice_by_index(begin = var_43385_begin_0, end = var_43385_end_0, end_mask = var_43385_end_mask_0, x = q_193_cast_fp16)[name = tensor("op_43385_cast_fp16")]; + tensor var_43389_begin_0 = const()[name = tensor("op_43389_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_43389_end_0 = const()[name = tensor("op_43389_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_43389_end_mask_0 = const()[name = tensor("op_43389_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43389_cast_fp16 = slice_by_index(begin = var_43389_begin_0, end = var_43389_end_0, end_mask = var_43389_end_mask_0, x = q_193_cast_fp16)[name = tensor("op_43389_cast_fp16")]; + tensor var_43393_begin_0 = const()[name = tensor("op_43393_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_43393_end_0 = const()[name = tensor("op_43393_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_43393_end_mask_0 = const()[name = tensor("op_43393_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43393_cast_fp16 = slice_by_index(begin = var_43393_begin_0, end = var_43393_end_0, end_mask = var_43393_end_mask_0, x = q_193_cast_fp16)[name = tensor("op_43393_cast_fp16")]; + tensor var_43397_begin_0 = const()[name = tensor("op_43397_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_43397_end_0 = const()[name = tensor("op_43397_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_43397_end_mask_0 = const()[name = tensor("op_43397_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43397_cast_fp16 = slice_by_index(begin = var_43397_begin_0, end = var_43397_end_0, end_mask = var_43397_end_mask_0, x = q_193_cast_fp16)[name = tensor("op_43397_cast_fp16")]; + tensor var_43401_begin_0 = const()[name = tensor("op_43401_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_43401_end_0 = const()[name = tensor("op_43401_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_43401_end_mask_0 = const()[name = tensor("op_43401_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43401_cast_fp16 = slice_by_index(begin = var_43401_begin_0, end = var_43401_end_0, end_mask = var_43401_end_mask_0, x = q_193_cast_fp16)[name = tensor("op_43401_cast_fp16")]; + tensor var_43405_begin_0 = const()[name = tensor("op_43405_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_43405_end_0 = const()[name = tensor("op_43405_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_43405_end_mask_0 = const()[name = tensor("op_43405_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43405_cast_fp16 = slice_by_index(begin = var_43405_begin_0, end = var_43405_end_0, end_mask = var_43405_end_mask_0, x = q_193_cast_fp16)[name = tensor("op_43405_cast_fp16")]; + tensor var_43409_begin_0 = const()[name = tensor("op_43409_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_43409_end_0 = const()[name = tensor("op_43409_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_43409_end_mask_0 = const()[name = tensor("op_43409_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43409_cast_fp16 = slice_by_index(begin = var_43409_begin_0, end = var_43409_end_0, end_mask = var_43409_end_mask_0, x = q_193_cast_fp16)[name = tensor("op_43409_cast_fp16")]; + tensor var_43413_begin_0 = const()[name = tensor("op_43413_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_43413_end_0 = const()[name = tensor("op_43413_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_43413_end_mask_0 = const()[name = tensor("op_43413_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43413_cast_fp16 = slice_by_index(begin = var_43413_begin_0, end = var_43413_end_0, end_mask = var_43413_end_mask_0, x = q_193_cast_fp16)[name = tensor("op_43413_cast_fp16")]; + tensor var_43417_begin_0 = const()[name = tensor("op_43417_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_43417_end_0 = const()[name = tensor("op_43417_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_43417_end_mask_0 = const()[name = tensor("op_43417_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43417_cast_fp16 = slice_by_index(begin = var_43417_begin_0, end = var_43417_end_0, end_mask = var_43417_end_mask_0, x = q_193_cast_fp16)[name = tensor("op_43417_cast_fp16")]; + tensor var_43421_begin_0 = const()[name = tensor("op_43421_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_43421_end_0 = const()[name = tensor("op_43421_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_43421_end_mask_0 = const()[name = tensor("op_43421_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43421_cast_fp16 = slice_by_index(begin = var_43421_begin_0, end = var_43421_end_0, end_mask = var_43421_end_mask_0, x = q_193_cast_fp16)[name = tensor("op_43421_cast_fp16")]; + tensor var_43425_begin_0 = const()[name = tensor("op_43425_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_43425_end_0 = const()[name = tensor("op_43425_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_43425_end_mask_0 = const()[name = tensor("op_43425_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43425_cast_fp16 = slice_by_index(begin = var_43425_begin_0, end = var_43425_end_0, end_mask = var_43425_end_mask_0, x = q_193_cast_fp16)[name = tensor("op_43425_cast_fp16")]; + tensor var_43429_begin_0 = const()[name = tensor("op_43429_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_43429_end_0 = const()[name = tensor("op_43429_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_43429_end_mask_0 = const()[name = tensor("op_43429_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43429_cast_fp16 = slice_by_index(begin = var_43429_begin_0, end = var_43429_end_0, end_mask = var_43429_end_mask_0, x = q_193_cast_fp16)[name = tensor("op_43429_cast_fp16")]; + tensor var_43433_begin_0 = const()[name = tensor("op_43433_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_43433_end_0 = const()[name = tensor("op_43433_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_43433_end_mask_0 = const()[name = tensor("op_43433_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43433_cast_fp16 = slice_by_index(begin = var_43433_begin_0, end = var_43433_end_0, end_mask = var_43433_end_mask_0, x = q_193_cast_fp16)[name = tensor("op_43433_cast_fp16")]; + tensor var_43437_begin_0 = const()[name = tensor("op_43437_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_43437_end_0 = const()[name = tensor("op_43437_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_43437_end_mask_0 = const()[name = tensor("op_43437_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43437_cast_fp16 = slice_by_index(begin = var_43437_begin_0, end = var_43437_end_0, end_mask = var_43437_end_mask_0, x = q_193_cast_fp16)[name = tensor("op_43437_cast_fp16")]; + tensor k_387_perm_0 = const()[name = tensor("k_387_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_43444_begin_0 = const()[name = tensor("op_43444_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43444_end_0 = const()[name = tensor("op_43444_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_43444_end_mask_0 = const()[name = tensor("op_43444_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_387_cast_fp16 = transpose(perm = k_387_perm_0, x = k_385_cast_fp16)[name = tensor("transpose_43")]; + tensor var_43444_cast_fp16 = slice_by_index(begin = var_43444_begin_0, end = var_43444_end_0, end_mask = var_43444_end_mask_0, x = k_387_cast_fp16)[name = tensor("op_43444_cast_fp16")]; + tensor var_43448_begin_0 = const()[name = tensor("op_43448_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_43448_end_0 = const()[name = tensor("op_43448_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_43448_end_mask_0 = const()[name = tensor("op_43448_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43448_cast_fp16 = slice_by_index(begin = var_43448_begin_0, end = var_43448_end_0, end_mask = var_43448_end_mask_0, x = k_387_cast_fp16)[name = tensor("op_43448_cast_fp16")]; + tensor var_43452_begin_0 = const()[name = tensor("op_43452_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_43452_end_0 = const()[name = tensor("op_43452_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_43452_end_mask_0 = const()[name = tensor("op_43452_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43452_cast_fp16 = slice_by_index(begin = var_43452_begin_0, end = var_43452_end_0, end_mask = var_43452_end_mask_0, x = k_387_cast_fp16)[name = tensor("op_43452_cast_fp16")]; + tensor var_43456_begin_0 = const()[name = tensor("op_43456_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_43456_end_0 = const()[name = tensor("op_43456_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_43456_end_mask_0 = const()[name = tensor("op_43456_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43456_cast_fp16 = slice_by_index(begin = var_43456_begin_0, end = var_43456_end_0, end_mask = var_43456_end_mask_0, x = k_387_cast_fp16)[name = tensor("op_43456_cast_fp16")]; + tensor var_43460_begin_0 = const()[name = tensor("op_43460_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_43460_end_0 = const()[name = tensor("op_43460_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_43460_end_mask_0 = const()[name = tensor("op_43460_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43460_cast_fp16 = slice_by_index(begin = var_43460_begin_0, end = var_43460_end_0, end_mask = var_43460_end_mask_0, x = k_387_cast_fp16)[name = tensor("op_43460_cast_fp16")]; + tensor var_43464_begin_0 = const()[name = tensor("op_43464_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_43464_end_0 = const()[name = tensor("op_43464_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_43464_end_mask_0 = const()[name = tensor("op_43464_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43464_cast_fp16 = slice_by_index(begin = var_43464_begin_0, end = var_43464_end_0, end_mask = var_43464_end_mask_0, x = k_387_cast_fp16)[name = tensor("op_43464_cast_fp16")]; + tensor var_43468_begin_0 = const()[name = tensor("op_43468_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_43468_end_0 = const()[name = tensor("op_43468_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_43468_end_mask_0 = const()[name = tensor("op_43468_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43468_cast_fp16 = slice_by_index(begin = var_43468_begin_0, end = var_43468_end_0, end_mask = var_43468_end_mask_0, x = k_387_cast_fp16)[name = tensor("op_43468_cast_fp16")]; + tensor var_43472_begin_0 = const()[name = tensor("op_43472_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_43472_end_0 = const()[name = tensor("op_43472_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_43472_end_mask_0 = const()[name = tensor("op_43472_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43472_cast_fp16 = slice_by_index(begin = var_43472_begin_0, end = var_43472_end_0, end_mask = var_43472_end_mask_0, x = k_387_cast_fp16)[name = tensor("op_43472_cast_fp16")]; + tensor var_43476_begin_0 = const()[name = tensor("op_43476_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_43476_end_0 = const()[name = tensor("op_43476_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_43476_end_mask_0 = const()[name = tensor("op_43476_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43476_cast_fp16 = slice_by_index(begin = var_43476_begin_0, end = var_43476_end_0, end_mask = var_43476_end_mask_0, x = k_387_cast_fp16)[name = tensor("op_43476_cast_fp16")]; + tensor var_43480_begin_0 = const()[name = tensor("op_43480_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_43480_end_0 = const()[name = tensor("op_43480_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_43480_end_mask_0 = const()[name = tensor("op_43480_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43480_cast_fp16 = slice_by_index(begin = var_43480_begin_0, end = var_43480_end_0, end_mask = var_43480_end_mask_0, x = k_387_cast_fp16)[name = tensor("op_43480_cast_fp16")]; + tensor var_43484_begin_0 = const()[name = tensor("op_43484_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_43484_end_0 = const()[name = tensor("op_43484_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_43484_end_mask_0 = const()[name = tensor("op_43484_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43484_cast_fp16 = slice_by_index(begin = var_43484_begin_0, end = var_43484_end_0, end_mask = var_43484_end_mask_0, x = k_387_cast_fp16)[name = tensor("op_43484_cast_fp16")]; + tensor var_43488_begin_0 = const()[name = tensor("op_43488_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_43488_end_0 = const()[name = tensor("op_43488_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_43488_end_mask_0 = const()[name = tensor("op_43488_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43488_cast_fp16 = slice_by_index(begin = var_43488_begin_0, end = var_43488_end_0, end_mask = var_43488_end_mask_0, x = k_387_cast_fp16)[name = tensor("op_43488_cast_fp16")]; + tensor var_43492_begin_0 = const()[name = tensor("op_43492_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_43492_end_0 = const()[name = tensor("op_43492_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_43492_end_mask_0 = const()[name = tensor("op_43492_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43492_cast_fp16 = slice_by_index(begin = var_43492_begin_0, end = var_43492_end_0, end_mask = var_43492_end_mask_0, x = k_387_cast_fp16)[name = tensor("op_43492_cast_fp16")]; + tensor var_43496_begin_0 = const()[name = tensor("op_43496_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_43496_end_0 = const()[name = tensor("op_43496_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_43496_end_mask_0 = const()[name = tensor("op_43496_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43496_cast_fp16 = slice_by_index(begin = var_43496_begin_0, end = var_43496_end_0, end_mask = var_43496_end_mask_0, x = k_387_cast_fp16)[name = tensor("op_43496_cast_fp16")]; + tensor var_43500_begin_0 = const()[name = tensor("op_43500_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_43500_end_0 = const()[name = tensor("op_43500_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_43500_end_mask_0 = const()[name = tensor("op_43500_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43500_cast_fp16 = slice_by_index(begin = var_43500_begin_0, end = var_43500_end_0, end_mask = var_43500_end_mask_0, x = k_387_cast_fp16)[name = tensor("op_43500_cast_fp16")]; + tensor var_43504_begin_0 = const()[name = tensor("op_43504_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_43504_end_0 = const()[name = tensor("op_43504_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_43504_end_mask_0 = const()[name = tensor("op_43504_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43504_cast_fp16 = slice_by_index(begin = var_43504_begin_0, end = var_43504_end_0, end_mask = var_43504_end_mask_0, x = k_387_cast_fp16)[name = tensor("op_43504_cast_fp16")]; + tensor var_43508_begin_0 = const()[name = tensor("op_43508_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_43508_end_0 = const()[name = tensor("op_43508_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_43508_end_mask_0 = const()[name = tensor("op_43508_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43508_cast_fp16 = slice_by_index(begin = var_43508_begin_0, end = var_43508_end_0, end_mask = var_43508_end_mask_0, x = k_387_cast_fp16)[name = tensor("op_43508_cast_fp16")]; + tensor var_43512_begin_0 = const()[name = tensor("op_43512_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_43512_end_0 = const()[name = tensor("op_43512_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_43512_end_mask_0 = const()[name = tensor("op_43512_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43512_cast_fp16 = slice_by_index(begin = var_43512_begin_0, end = var_43512_end_0, end_mask = var_43512_end_mask_0, x = k_387_cast_fp16)[name = tensor("op_43512_cast_fp16")]; + tensor var_43516_begin_0 = const()[name = tensor("op_43516_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_43516_end_0 = const()[name = tensor("op_43516_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_43516_end_mask_0 = const()[name = tensor("op_43516_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43516_cast_fp16 = slice_by_index(begin = var_43516_begin_0, end = var_43516_end_0, end_mask = var_43516_end_mask_0, x = k_387_cast_fp16)[name = tensor("op_43516_cast_fp16")]; + tensor var_43520_begin_0 = const()[name = tensor("op_43520_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_43520_end_0 = const()[name = tensor("op_43520_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_43520_end_mask_0 = const()[name = tensor("op_43520_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43520_cast_fp16 = slice_by_index(begin = var_43520_begin_0, end = var_43520_end_0, end_mask = var_43520_end_mask_0, x = k_387_cast_fp16)[name = tensor("op_43520_cast_fp16")]; + tensor var_43522_begin_0 = const()[name = tensor("op_43522_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43522_end_0 = const()[name = tensor("op_43522_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_43522_end_mask_0 = const()[name = tensor("op_43522_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43522_cast_fp16 = slice_by_index(begin = var_43522_begin_0, end = var_43522_end_0, end_mask = var_43522_end_mask_0, x = v_193_cast_fp16)[name = tensor("op_43522_cast_fp16")]; + tensor var_43526_begin_0 = const()[name = tensor("op_43526_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_43526_end_0 = const()[name = tensor("op_43526_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_43526_end_mask_0 = const()[name = tensor("op_43526_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43526_cast_fp16 = slice_by_index(begin = var_43526_begin_0, end = var_43526_end_0, end_mask = var_43526_end_mask_0, x = v_193_cast_fp16)[name = tensor("op_43526_cast_fp16")]; + tensor var_43530_begin_0 = const()[name = tensor("op_43530_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_43530_end_0 = const()[name = tensor("op_43530_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_43530_end_mask_0 = const()[name = tensor("op_43530_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43530_cast_fp16 = slice_by_index(begin = var_43530_begin_0, end = var_43530_end_0, end_mask = var_43530_end_mask_0, x = v_193_cast_fp16)[name = tensor("op_43530_cast_fp16")]; + tensor var_43534_begin_0 = const()[name = tensor("op_43534_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_43534_end_0 = const()[name = tensor("op_43534_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_43534_end_mask_0 = const()[name = tensor("op_43534_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43534_cast_fp16 = slice_by_index(begin = var_43534_begin_0, end = var_43534_end_0, end_mask = var_43534_end_mask_0, x = v_193_cast_fp16)[name = tensor("op_43534_cast_fp16")]; + tensor var_43538_begin_0 = const()[name = tensor("op_43538_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_43538_end_0 = const()[name = tensor("op_43538_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_43538_end_mask_0 = const()[name = tensor("op_43538_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43538_cast_fp16 = slice_by_index(begin = var_43538_begin_0, end = var_43538_end_0, end_mask = var_43538_end_mask_0, x = v_193_cast_fp16)[name = tensor("op_43538_cast_fp16")]; + tensor var_43542_begin_0 = const()[name = tensor("op_43542_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_43542_end_0 = const()[name = tensor("op_43542_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_43542_end_mask_0 = const()[name = tensor("op_43542_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43542_cast_fp16 = slice_by_index(begin = var_43542_begin_0, end = var_43542_end_0, end_mask = var_43542_end_mask_0, x = v_193_cast_fp16)[name = tensor("op_43542_cast_fp16")]; + tensor var_43546_begin_0 = const()[name = tensor("op_43546_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_43546_end_0 = const()[name = tensor("op_43546_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_43546_end_mask_0 = const()[name = tensor("op_43546_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43546_cast_fp16 = slice_by_index(begin = var_43546_begin_0, end = var_43546_end_0, end_mask = var_43546_end_mask_0, x = v_193_cast_fp16)[name = tensor("op_43546_cast_fp16")]; + tensor var_43550_begin_0 = const()[name = tensor("op_43550_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_43550_end_0 = const()[name = tensor("op_43550_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_43550_end_mask_0 = const()[name = tensor("op_43550_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43550_cast_fp16 = slice_by_index(begin = var_43550_begin_0, end = var_43550_end_0, end_mask = var_43550_end_mask_0, x = v_193_cast_fp16)[name = tensor("op_43550_cast_fp16")]; + tensor var_43554_begin_0 = const()[name = tensor("op_43554_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_43554_end_0 = const()[name = tensor("op_43554_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_43554_end_mask_0 = const()[name = tensor("op_43554_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43554_cast_fp16 = slice_by_index(begin = var_43554_begin_0, end = var_43554_end_0, end_mask = var_43554_end_mask_0, x = v_193_cast_fp16)[name = tensor("op_43554_cast_fp16")]; + tensor var_43558_begin_0 = const()[name = tensor("op_43558_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_43558_end_0 = const()[name = tensor("op_43558_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_43558_end_mask_0 = const()[name = tensor("op_43558_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43558_cast_fp16 = slice_by_index(begin = var_43558_begin_0, end = var_43558_end_0, end_mask = var_43558_end_mask_0, x = v_193_cast_fp16)[name = tensor("op_43558_cast_fp16")]; + tensor var_43562_begin_0 = const()[name = tensor("op_43562_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_43562_end_0 = const()[name = tensor("op_43562_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_43562_end_mask_0 = const()[name = tensor("op_43562_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43562_cast_fp16 = slice_by_index(begin = var_43562_begin_0, end = var_43562_end_0, end_mask = var_43562_end_mask_0, x = v_193_cast_fp16)[name = tensor("op_43562_cast_fp16")]; + tensor var_43566_begin_0 = const()[name = tensor("op_43566_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_43566_end_0 = const()[name = tensor("op_43566_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_43566_end_mask_0 = const()[name = tensor("op_43566_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43566_cast_fp16 = slice_by_index(begin = var_43566_begin_0, end = var_43566_end_0, end_mask = var_43566_end_mask_0, x = v_193_cast_fp16)[name = tensor("op_43566_cast_fp16")]; + tensor var_43570_begin_0 = const()[name = tensor("op_43570_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_43570_end_0 = const()[name = tensor("op_43570_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_43570_end_mask_0 = const()[name = tensor("op_43570_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43570_cast_fp16 = slice_by_index(begin = var_43570_begin_0, end = var_43570_end_0, end_mask = var_43570_end_mask_0, x = v_193_cast_fp16)[name = tensor("op_43570_cast_fp16")]; + tensor var_43574_begin_0 = const()[name = tensor("op_43574_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_43574_end_0 = const()[name = tensor("op_43574_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_43574_end_mask_0 = const()[name = tensor("op_43574_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43574_cast_fp16 = slice_by_index(begin = var_43574_begin_0, end = var_43574_end_0, end_mask = var_43574_end_mask_0, x = v_193_cast_fp16)[name = tensor("op_43574_cast_fp16")]; + tensor var_43578_begin_0 = const()[name = tensor("op_43578_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_43578_end_0 = const()[name = tensor("op_43578_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_43578_end_mask_0 = const()[name = tensor("op_43578_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43578_cast_fp16 = slice_by_index(begin = var_43578_begin_0, end = var_43578_end_0, end_mask = var_43578_end_mask_0, x = v_193_cast_fp16)[name = tensor("op_43578_cast_fp16")]; + tensor var_43582_begin_0 = const()[name = tensor("op_43582_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_43582_end_0 = const()[name = tensor("op_43582_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_43582_end_mask_0 = const()[name = tensor("op_43582_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43582_cast_fp16 = slice_by_index(begin = var_43582_begin_0, end = var_43582_end_0, end_mask = var_43582_end_mask_0, x = v_193_cast_fp16)[name = tensor("op_43582_cast_fp16")]; + tensor var_43586_begin_0 = const()[name = tensor("op_43586_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_43586_end_0 = const()[name = tensor("op_43586_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_43586_end_mask_0 = const()[name = tensor("op_43586_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43586_cast_fp16 = slice_by_index(begin = var_43586_begin_0, end = var_43586_end_0, end_mask = var_43586_end_mask_0, x = v_193_cast_fp16)[name = tensor("op_43586_cast_fp16")]; + tensor var_43590_begin_0 = const()[name = tensor("op_43590_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_43590_end_0 = const()[name = tensor("op_43590_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_43590_end_mask_0 = const()[name = tensor("op_43590_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43590_cast_fp16 = slice_by_index(begin = var_43590_begin_0, end = var_43590_end_0, end_mask = var_43590_end_mask_0, x = v_193_cast_fp16)[name = tensor("op_43590_cast_fp16")]; + tensor var_43594_begin_0 = const()[name = tensor("op_43594_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_43594_end_0 = const()[name = tensor("op_43594_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_43594_end_mask_0 = const()[name = tensor("op_43594_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43594_cast_fp16 = slice_by_index(begin = var_43594_begin_0, end = var_43594_end_0, end_mask = var_43594_end_mask_0, x = v_193_cast_fp16)[name = tensor("op_43594_cast_fp16")]; + tensor var_43598_begin_0 = const()[name = tensor("op_43598_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_43598_end_0 = const()[name = tensor("op_43598_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_43598_end_mask_0 = const()[name = tensor("op_43598_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43598_cast_fp16 = slice_by_index(begin = var_43598_begin_0, end = var_43598_end_0, end_mask = var_43598_end_mask_0, x = v_193_cast_fp16)[name = tensor("op_43598_cast_fp16")]; + tensor var_43602_equation_0 = const()[name = tensor("op_43602_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43602_cast_fp16 = einsum(equation = var_43602_equation_0, values = (var_43444_cast_fp16, var_43361_cast_fp16))[name = tensor("op_43602_cast_fp16")]; + tensor var_43603_to_fp16 = const()[name = tensor("op_43603_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3681_cast_fp16 = mul(x = var_43602_cast_fp16, y = var_43603_to_fp16)[name = tensor("aw_3681_cast_fp16")]; + tensor var_43606_equation_0 = const()[name = tensor("op_43606_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43606_cast_fp16 = einsum(equation = var_43606_equation_0, values = (var_43448_cast_fp16, var_43365_cast_fp16))[name = tensor("op_43606_cast_fp16")]; + tensor var_43607_to_fp16 = const()[name = tensor("op_43607_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3683_cast_fp16 = mul(x = var_43606_cast_fp16, y = var_43607_to_fp16)[name = tensor("aw_3683_cast_fp16")]; + tensor var_43610_equation_0 = const()[name = tensor("op_43610_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43610_cast_fp16 = einsum(equation = var_43610_equation_0, values = (var_43452_cast_fp16, var_43369_cast_fp16))[name = tensor("op_43610_cast_fp16")]; + tensor var_43611_to_fp16 = const()[name = tensor("op_43611_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3685_cast_fp16 = mul(x = var_43610_cast_fp16, y = var_43611_to_fp16)[name = tensor("aw_3685_cast_fp16")]; + tensor var_43614_equation_0 = const()[name = tensor("op_43614_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43614_cast_fp16 = einsum(equation = var_43614_equation_0, values = (var_43456_cast_fp16, var_43373_cast_fp16))[name = tensor("op_43614_cast_fp16")]; + tensor var_43615_to_fp16 = const()[name = tensor("op_43615_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3687_cast_fp16 = mul(x = var_43614_cast_fp16, y = var_43615_to_fp16)[name = tensor("aw_3687_cast_fp16")]; + tensor var_43618_equation_0 = const()[name = tensor("op_43618_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43618_cast_fp16 = einsum(equation = var_43618_equation_0, values = (var_43460_cast_fp16, var_43377_cast_fp16))[name = tensor("op_43618_cast_fp16")]; + tensor var_43619_to_fp16 = const()[name = tensor("op_43619_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3689_cast_fp16 = mul(x = var_43618_cast_fp16, y = var_43619_to_fp16)[name = tensor("aw_3689_cast_fp16")]; + tensor var_43622_equation_0 = const()[name = tensor("op_43622_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43622_cast_fp16 = einsum(equation = var_43622_equation_0, values = (var_43464_cast_fp16, var_43381_cast_fp16))[name = tensor("op_43622_cast_fp16")]; + tensor var_43623_to_fp16 = const()[name = tensor("op_43623_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3691_cast_fp16 = mul(x = var_43622_cast_fp16, y = var_43623_to_fp16)[name = tensor("aw_3691_cast_fp16")]; + tensor var_43626_equation_0 = const()[name = tensor("op_43626_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43626_cast_fp16 = einsum(equation = var_43626_equation_0, values = (var_43468_cast_fp16, var_43385_cast_fp16))[name = tensor("op_43626_cast_fp16")]; + tensor var_43627_to_fp16 = const()[name = tensor("op_43627_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3693_cast_fp16 = mul(x = var_43626_cast_fp16, y = var_43627_to_fp16)[name = tensor("aw_3693_cast_fp16")]; + tensor var_43630_equation_0 = const()[name = tensor("op_43630_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43630_cast_fp16 = einsum(equation = var_43630_equation_0, values = (var_43472_cast_fp16, var_43389_cast_fp16))[name = tensor("op_43630_cast_fp16")]; + tensor var_43631_to_fp16 = const()[name = tensor("op_43631_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3695_cast_fp16 = mul(x = var_43630_cast_fp16, y = var_43631_to_fp16)[name = tensor("aw_3695_cast_fp16")]; + tensor var_43634_equation_0 = const()[name = tensor("op_43634_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43634_cast_fp16 = einsum(equation = var_43634_equation_0, values = (var_43476_cast_fp16, var_43393_cast_fp16))[name = tensor("op_43634_cast_fp16")]; + tensor var_43635_to_fp16 = const()[name = tensor("op_43635_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3697_cast_fp16 = mul(x = var_43634_cast_fp16, y = var_43635_to_fp16)[name = tensor("aw_3697_cast_fp16")]; + tensor var_43638_equation_0 = const()[name = tensor("op_43638_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43638_cast_fp16 = einsum(equation = var_43638_equation_0, values = (var_43480_cast_fp16, var_43397_cast_fp16))[name = tensor("op_43638_cast_fp16")]; + tensor var_43639_to_fp16 = const()[name = tensor("op_43639_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3699_cast_fp16 = mul(x = var_43638_cast_fp16, y = var_43639_to_fp16)[name = tensor("aw_3699_cast_fp16")]; + tensor var_43642_equation_0 = const()[name = tensor("op_43642_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43642_cast_fp16 = einsum(equation = var_43642_equation_0, values = (var_43484_cast_fp16, var_43401_cast_fp16))[name = tensor("op_43642_cast_fp16")]; + tensor var_43643_to_fp16 = const()[name = tensor("op_43643_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3701_cast_fp16 = mul(x = var_43642_cast_fp16, y = var_43643_to_fp16)[name = tensor("aw_3701_cast_fp16")]; + tensor var_43646_equation_0 = const()[name = tensor("op_43646_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43646_cast_fp16 = einsum(equation = var_43646_equation_0, values = (var_43488_cast_fp16, var_43405_cast_fp16))[name = tensor("op_43646_cast_fp16")]; + tensor var_43647_to_fp16 = const()[name = tensor("op_43647_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3703_cast_fp16 = mul(x = var_43646_cast_fp16, y = var_43647_to_fp16)[name = tensor("aw_3703_cast_fp16")]; + tensor var_43650_equation_0 = const()[name = tensor("op_43650_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43650_cast_fp16 = einsum(equation = var_43650_equation_0, values = (var_43492_cast_fp16, var_43409_cast_fp16))[name = tensor("op_43650_cast_fp16")]; + tensor var_43651_to_fp16 = const()[name = tensor("op_43651_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3705_cast_fp16 = mul(x = var_43650_cast_fp16, y = var_43651_to_fp16)[name = tensor("aw_3705_cast_fp16")]; + tensor var_43654_equation_0 = const()[name = tensor("op_43654_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43654_cast_fp16 = einsum(equation = var_43654_equation_0, values = (var_43496_cast_fp16, var_43413_cast_fp16))[name = tensor("op_43654_cast_fp16")]; + tensor var_43655_to_fp16 = const()[name = tensor("op_43655_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3707_cast_fp16 = mul(x = var_43654_cast_fp16, y = var_43655_to_fp16)[name = tensor("aw_3707_cast_fp16")]; + tensor var_43658_equation_0 = const()[name = tensor("op_43658_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43658_cast_fp16 = einsum(equation = var_43658_equation_0, values = (var_43500_cast_fp16, var_43417_cast_fp16))[name = tensor("op_43658_cast_fp16")]; + tensor var_43659_to_fp16 = const()[name = tensor("op_43659_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3709_cast_fp16 = mul(x = var_43658_cast_fp16, y = var_43659_to_fp16)[name = tensor("aw_3709_cast_fp16")]; + tensor var_43662_equation_0 = const()[name = tensor("op_43662_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43662_cast_fp16 = einsum(equation = var_43662_equation_0, values = (var_43504_cast_fp16, var_43421_cast_fp16))[name = tensor("op_43662_cast_fp16")]; + tensor var_43663_to_fp16 = const()[name = tensor("op_43663_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3711_cast_fp16 = mul(x = var_43662_cast_fp16, y = var_43663_to_fp16)[name = tensor("aw_3711_cast_fp16")]; + tensor var_43666_equation_0 = const()[name = tensor("op_43666_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43666_cast_fp16 = einsum(equation = var_43666_equation_0, values = (var_43508_cast_fp16, var_43425_cast_fp16))[name = tensor("op_43666_cast_fp16")]; + tensor var_43667_to_fp16 = const()[name = tensor("op_43667_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3713_cast_fp16 = mul(x = var_43666_cast_fp16, y = var_43667_to_fp16)[name = tensor("aw_3713_cast_fp16")]; + tensor var_43670_equation_0 = const()[name = tensor("op_43670_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43670_cast_fp16 = einsum(equation = var_43670_equation_0, values = (var_43512_cast_fp16, var_43429_cast_fp16))[name = tensor("op_43670_cast_fp16")]; + tensor var_43671_to_fp16 = const()[name = tensor("op_43671_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3715_cast_fp16 = mul(x = var_43670_cast_fp16, y = var_43671_to_fp16)[name = tensor("aw_3715_cast_fp16")]; + tensor var_43674_equation_0 = const()[name = tensor("op_43674_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43674_cast_fp16 = einsum(equation = var_43674_equation_0, values = (var_43516_cast_fp16, var_43433_cast_fp16))[name = tensor("op_43674_cast_fp16")]; + tensor var_43675_to_fp16 = const()[name = tensor("op_43675_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3717_cast_fp16 = mul(x = var_43674_cast_fp16, y = var_43675_to_fp16)[name = tensor("aw_3717_cast_fp16")]; + tensor var_43678_equation_0 = const()[name = tensor("op_43678_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43678_cast_fp16 = einsum(equation = var_43678_equation_0, values = (var_43520_cast_fp16, var_43437_cast_fp16))[name = tensor("op_43678_cast_fp16")]; + tensor var_43679_to_fp16 = const()[name = tensor("op_43679_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3719_cast_fp16 = mul(x = var_43678_cast_fp16, y = var_43679_to_fp16)[name = tensor("aw_3719_cast_fp16")]; + tensor var_43681_cast_fp16 = softmax(axis = var_30355, x = aw_3681_cast_fp16)[name = tensor("op_43681_cast_fp16")]; + tensor var_43682_cast_fp16 = softmax(axis = var_30355, x = aw_3683_cast_fp16)[name = tensor("op_43682_cast_fp16")]; + tensor var_43683_cast_fp16 = softmax(axis = var_30355, x = aw_3685_cast_fp16)[name = tensor("op_43683_cast_fp16")]; + tensor var_43684_cast_fp16 = softmax(axis = var_30355, x = aw_3687_cast_fp16)[name = tensor("op_43684_cast_fp16")]; + tensor var_43685_cast_fp16 = softmax(axis = var_30355, x = aw_3689_cast_fp16)[name = tensor("op_43685_cast_fp16")]; + tensor var_43686_cast_fp16 = softmax(axis = var_30355, x = aw_3691_cast_fp16)[name = tensor("op_43686_cast_fp16")]; + tensor var_43687_cast_fp16 = softmax(axis = var_30355, x = aw_3693_cast_fp16)[name = tensor("op_43687_cast_fp16")]; + tensor var_43688_cast_fp16 = softmax(axis = var_30355, x = aw_3695_cast_fp16)[name = tensor("op_43688_cast_fp16")]; + tensor var_43689_cast_fp16 = softmax(axis = var_30355, x = aw_3697_cast_fp16)[name = tensor("op_43689_cast_fp16")]; + tensor var_43690_cast_fp16 = softmax(axis = var_30355, x = aw_3699_cast_fp16)[name = tensor("op_43690_cast_fp16")]; + tensor var_43691_cast_fp16 = softmax(axis = var_30355, x = aw_3701_cast_fp16)[name = tensor("op_43691_cast_fp16")]; + tensor var_43692_cast_fp16 = softmax(axis = var_30355, x = aw_3703_cast_fp16)[name = tensor("op_43692_cast_fp16")]; + tensor var_43693_cast_fp16 = softmax(axis = var_30355, x = aw_3705_cast_fp16)[name = tensor("op_43693_cast_fp16")]; + tensor var_43694_cast_fp16 = softmax(axis = var_30355, x = aw_3707_cast_fp16)[name = tensor("op_43694_cast_fp16")]; + tensor var_43695_cast_fp16 = softmax(axis = var_30355, x = aw_3709_cast_fp16)[name = tensor("op_43695_cast_fp16")]; + tensor var_43696_cast_fp16 = softmax(axis = var_30355, x = aw_3711_cast_fp16)[name = tensor("op_43696_cast_fp16")]; + tensor var_43697_cast_fp16 = softmax(axis = var_30355, x = aw_3713_cast_fp16)[name = tensor("op_43697_cast_fp16")]; + tensor var_43698_cast_fp16 = softmax(axis = var_30355, x = aw_3715_cast_fp16)[name = tensor("op_43698_cast_fp16")]; + tensor var_43699_cast_fp16 = softmax(axis = var_30355, x = aw_3717_cast_fp16)[name = tensor("op_43699_cast_fp16")]; + tensor var_43700_cast_fp16 = softmax(axis = var_30355, x = aw_3719_cast_fp16)[name = tensor("op_43700_cast_fp16")]; + tensor var_43702_equation_0 = const()[name = tensor("op_43702_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_43702_cast_fp16 = einsum(equation = var_43702_equation_0, values = (var_43522_cast_fp16, var_43681_cast_fp16))[name = tensor("op_43702_cast_fp16")]; + tensor var_43704_equation_0 = const()[name = tensor("op_43704_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_43704_cast_fp16 = einsum(equation = var_43704_equation_0, values = (var_43526_cast_fp16, var_43682_cast_fp16))[name = tensor("op_43704_cast_fp16")]; + tensor var_43706_equation_0 = const()[name = tensor("op_43706_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_43706_cast_fp16 = einsum(equation = var_43706_equation_0, values = (var_43530_cast_fp16, var_43683_cast_fp16))[name = tensor("op_43706_cast_fp16")]; + tensor var_43708_equation_0 = const()[name = tensor("op_43708_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_43708_cast_fp16 = einsum(equation = var_43708_equation_0, values = (var_43534_cast_fp16, var_43684_cast_fp16))[name = tensor("op_43708_cast_fp16")]; + tensor var_43710_equation_0 = const()[name = tensor("op_43710_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_43710_cast_fp16 = einsum(equation = var_43710_equation_0, values = (var_43538_cast_fp16, var_43685_cast_fp16))[name = tensor("op_43710_cast_fp16")]; + tensor var_43712_equation_0 = const()[name = tensor("op_43712_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_43712_cast_fp16 = einsum(equation = var_43712_equation_0, values = (var_43542_cast_fp16, var_43686_cast_fp16))[name = tensor("op_43712_cast_fp16")]; + tensor var_43714_equation_0 = const()[name = tensor("op_43714_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_43714_cast_fp16 = einsum(equation = var_43714_equation_0, values = (var_43546_cast_fp16, var_43687_cast_fp16))[name = tensor("op_43714_cast_fp16")]; + tensor var_43716_equation_0 = const()[name = tensor("op_43716_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_43716_cast_fp16 = einsum(equation = var_43716_equation_0, values = (var_43550_cast_fp16, var_43688_cast_fp16))[name = tensor("op_43716_cast_fp16")]; + tensor var_43718_equation_0 = const()[name = tensor("op_43718_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_43718_cast_fp16 = einsum(equation = var_43718_equation_0, values = (var_43554_cast_fp16, var_43689_cast_fp16))[name = tensor("op_43718_cast_fp16")]; + tensor var_43720_equation_0 = const()[name = tensor("op_43720_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_43720_cast_fp16 = einsum(equation = var_43720_equation_0, values = (var_43558_cast_fp16, var_43690_cast_fp16))[name = tensor("op_43720_cast_fp16")]; + tensor var_43722_equation_0 = const()[name = tensor("op_43722_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_43722_cast_fp16 = einsum(equation = var_43722_equation_0, values = (var_43562_cast_fp16, var_43691_cast_fp16))[name = tensor("op_43722_cast_fp16")]; + tensor var_43724_equation_0 = const()[name = tensor("op_43724_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_43724_cast_fp16 = einsum(equation = var_43724_equation_0, values = (var_43566_cast_fp16, var_43692_cast_fp16))[name = tensor("op_43724_cast_fp16")]; + tensor var_43726_equation_0 = const()[name = tensor("op_43726_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_43726_cast_fp16 = einsum(equation = var_43726_equation_0, values = (var_43570_cast_fp16, var_43693_cast_fp16))[name = tensor("op_43726_cast_fp16")]; + tensor var_43728_equation_0 = const()[name = tensor("op_43728_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_43728_cast_fp16 = einsum(equation = var_43728_equation_0, values = (var_43574_cast_fp16, var_43694_cast_fp16))[name = tensor("op_43728_cast_fp16")]; + tensor var_43730_equation_0 = const()[name = tensor("op_43730_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_43730_cast_fp16 = einsum(equation = var_43730_equation_0, values = (var_43578_cast_fp16, var_43695_cast_fp16))[name = tensor("op_43730_cast_fp16")]; + tensor var_43732_equation_0 = const()[name = tensor("op_43732_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_43732_cast_fp16 = einsum(equation = var_43732_equation_0, values = (var_43582_cast_fp16, var_43696_cast_fp16))[name = tensor("op_43732_cast_fp16")]; + tensor var_43734_equation_0 = const()[name = tensor("op_43734_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_43734_cast_fp16 = einsum(equation = var_43734_equation_0, values = (var_43586_cast_fp16, var_43697_cast_fp16))[name = tensor("op_43734_cast_fp16")]; + tensor var_43736_equation_0 = const()[name = tensor("op_43736_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_43736_cast_fp16 = einsum(equation = var_43736_equation_0, values = (var_43590_cast_fp16, var_43698_cast_fp16))[name = tensor("op_43736_cast_fp16")]; + tensor var_43738_equation_0 = const()[name = tensor("op_43738_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_43738_cast_fp16 = einsum(equation = var_43738_equation_0, values = (var_43594_cast_fp16, var_43699_cast_fp16))[name = tensor("op_43738_cast_fp16")]; + tensor var_43740_equation_0 = const()[name = tensor("op_43740_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_43740_cast_fp16 = einsum(equation = var_43740_equation_0, values = (var_43598_cast_fp16, var_43700_cast_fp16))[name = tensor("op_43740_cast_fp16")]; + tensor input_569_interleave_0 = const()[name = tensor("input_569_interleave_0"), val = tensor(false)]; + tensor input_569_cast_fp16 = concat(axis = var_30355, interleave = input_569_interleave_0, values = (var_43702_cast_fp16, var_43704_cast_fp16, var_43706_cast_fp16, var_43708_cast_fp16, var_43710_cast_fp16, var_43712_cast_fp16, var_43714_cast_fp16, var_43716_cast_fp16, var_43718_cast_fp16, var_43720_cast_fp16, var_43722_cast_fp16, var_43724_cast_fp16, var_43726_cast_fp16, var_43728_cast_fp16, var_43730_cast_fp16, var_43732_cast_fp16, var_43734_cast_fp16, var_43736_cast_fp16, var_43738_cast_fp16, var_43740_cast_fp16))[name = tensor("input_569_cast_fp16")]; + tensor var_43746 = const()[name = tensor("op_43746"), val = tensor([1, 1])]; + tensor var_43748 = const()[name = tensor("op_43748"), val = tensor([1, 1])]; + tensor var_43750_pad_type_0 = const()[name = tensor("op_43750_pad_type_0"), val = tensor("custom")]; + tensor var_43750_pad_0 = const()[name = tensor("op_43750_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_4_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_4_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3676333248)))]; + tensor up_blocks_0_attentions_1_transformer_blocks_4_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_4_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3679610112)))]; + tensor var_43750_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_4_attn1_to_out_0_bias_to_fp16, dilations = var_43748, groups = var_30355, pad = var_43750_pad_0, pad_type = var_43750_pad_type_0, strides = var_43746, weight = up_blocks_0_attentions_1_transformer_blocks_4_attn1_to_out_0_weight_to_fp16, x = input_569_cast_fp16)[name = tensor("op_43750_cast_fp16")]; + tensor inputs_291_cast_fp16 = add(x = var_43750_cast_fp16, y = inputs_289_cast_fp16)[name = tensor("inputs_291_cast_fp16")]; + tensor hidden_states_387_axes_0 = const()[name = tensor("hidden_states_387_axes_0"), val = tensor([1])]; + tensor hidden_states_387_gamma_0_to_fp16 = const()[name = tensor("hidden_states_387_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3679612736)))]; + tensor hidden_states_387_beta_0_to_fp16 = const()[name = tensor("hidden_states_387_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3679615360)))]; + tensor var_43760_to_fp16 = const()[name = tensor("op_43760_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_387_cast_fp16 = layer_norm(axes = hidden_states_387_axes_0, beta = hidden_states_387_beta_0_to_fp16, epsilon = var_43760_to_fp16, gamma = hidden_states_387_gamma_0_to_fp16, x = inputs_291_cast_fp16)[name = tensor("hidden_states_387_cast_fp16")]; + tensor var_43775 = const()[name = tensor("op_43775"), val = tensor([1, 1])]; + tensor var_43777 = const()[name = tensor("op_43777"), val = tensor([1, 1])]; + tensor q_195_pad_type_0 = const()[name = tensor("q_195_pad_type_0"), val = tensor("custom")]; + tensor q_195_pad_0 = const()[name = tensor("q_195_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_4_attn2_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_4_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3679617984)))]; + tensor q_195_cast_fp16 = conv(dilations = var_43777, groups = var_30355, pad = q_195_pad_0, pad_type = q_195_pad_type_0, strides = var_43775, weight = up_blocks_0_attentions_1_transformer_blocks_4_attn2_to_q_weight_to_fp16, x = hidden_states_387_cast_fp16)[name = tensor("q_195_cast_fp16")]; + tensor var_43781 = const()[name = tensor("op_43781"), val = tensor([1, 1])]; + tensor var_43783 = const()[name = tensor("op_43783"), val = tensor([1, 1])]; + tensor k_389_pad_type_0 = const()[name = tensor("k_389_pad_type_0"), val = tensor("custom")]; + tensor k_389_pad_0 = const()[name = tensor("k_389_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_4_attn2_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_4_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3682894848)))]; + tensor k_389_cast_fp16 = conv(dilations = var_43783, groups = var_30355, pad = k_389_pad_0, pad_type = k_389_pad_type_0, strides = var_43781, weight = up_blocks_0_attentions_1_transformer_blocks_4_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_389_cast_fp16")]; + tensor var_43787 = const()[name = tensor("op_43787"), val = tensor([1, 1])]; + tensor var_43789 = const()[name = tensor("op_43789"), val = tensor([1, 1])]; + tensor v_195_pad_type_0 = const()[name = tensor("v_195_pad_type_0"), val = tensor("custom")]; + tensor v_195_pad_0 = const()[name = tensor("v_195_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_4_attn2_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_4_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3688137792)))]; + tensor v_195_cast_fp16 = conv(dilations = var_43789, groups = var_30355, pad = v_195_pad_0, pad_type = v_195_pad_type_0, strides = var_43787, weight = up_blocks_0_attentions_1_transformer_blocks_4_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_195_cast_fp16")]; + tensor var_43793_begin_0 = const()[name = tensor("op_43793_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43793_end_0 = const()[name = tensor("op_43793_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_43793_end_mask_0 = const()[name = tensor("op_43793_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43793_cast_fp16 = slice_by_index(begin = var_43793_begin_0, end = var_43793_end_0, end_mask = var_43793_end_mask_0, x = q_195_cast_fp16)[name = tensor("op_43793_cast_fp16")]; + tensor var_43797_begin_0 = const()[name = tensor("op_43797_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_43797_end_0 = const()[name = tensor("op_43797_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_43797_end_mask_0 = const()[name = tensor("op_43797_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43797_cast_fp16 = slice_by_index(begin = var_43797_begin_0, end = var_43797_end_0, end_mask = var_43797_end_mask_0, x = q_195_cast_fp16)[name = tensor("op_43797_cast_fp16")]; + tensor var_43801_begin_0 = const()[name = tensor("op_43801_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_43801_end_0 = const()[name = tensor("op_43801_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_43801_end_mask_0 = const()[name = tensor("op_43801_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43801_cast_fp16 = slice_by_index(begin = var_43801_begin_0, end = var_43801_end_0, end_mask = var_43801_end_mask_0, x = q_195_cast_fp16)[name = tensor("op_43801_cast_fp16")]; + tensor var_43805_begin_0 = const()[name = tensor("op_43805_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_43805_end_0 = const()[name = tensor("op_43805_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_43805_end_mask_0 = const()[name = tensor("op_43805_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43805_cast_fp16 = slice_by_index(begin = var_43805_begin_0, end = var_43805_end_0, end_mask = var_43805_end_mask_0, x = q_195_cast_fp16)[name = tensor("op_43805_cast_fp16")]; + tensor var_43809_begin_0 = const()[name = tensor("op_43809_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_43809_end_0 = const()[name = tensor("op_43809_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_43809_end_mask_0 = const()[name = tensor("op_43809_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43809_cast_fp16 = slice_by_index(begin = var_43809_begin_0, end = var_43809_end_0, end_mask = var_43809_end_mask_0, x = q_195_cast_fp16)[name = tensor("op_43809_cast_fp16")]; + tensor var_43813_begin_0 = const()[name = tensor("op_43813_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_43813_end_0 = const()[name = tensor("op_43813_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_43813_end_mask_0 = const()[name = tensor("op_43813_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43813_cast_fp16 = slice_by_index(begin = var_43813_begin_0, end = var_43813_end_0, end_mask = var_43813_end_mask_0, x = q_195_cast_fp16)[name = tensor("op_43813_cast_fp16")]; + tensor var_43817_begin_0 = const()[name = tensor("op_43817_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_43817_end_0 = const()[name = tensor("op_43817_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_43817_end_mask_0 = const()[name = tensor("op_43817_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43817_cast_fp16 = slice_by_index(begin = var_43817_begin_0, end = var_43817_end_0, end_mask = var_43817_end_mask_0, x = q_195_cast_fp16)[name = tensor("op_43817_cast_fp16")]; + tensor var_43821_begin_0 = const()[name = tensor("op_43821_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_43821_end_0 = const()[name = tensor("op_43821_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_43821_end_mask_0 = const()[name = tensor("op_43821_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43821_cast_fp16 = slice_by_index(begin = var_43821_begin_0, end = var_43821_end_0, end_mask = var_43821_end_mask_0, x = q_195_cast_fp16)[name = tensor("op_43821_cast_fp16")]; + tensor var_43825_begin_0 = const()[name = tensor("op_43825_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_43825_end_0 = const()[name = tensor("op_43825_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_43825_end_mask_0 = const()[name = tensor("op_43825_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43825_cast_fp16 = slice_by_index(begin = var_43825_begin_0, end = var_43825_end_0, end_mask = var_43825_end_mask_0, x = q_195_cast_fp16)[name = tensor("op_43825_cast_fp16")]; + tensor var_43829_begin_0 = const()[name = tensor("op_43829_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_43829_end_0 = const()[name = tensor("op_43829_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_43829_end_mask_0 = const()[name = tensor("op_43829_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43829_cast_fp16 = slice_by_index(begin = var_43829_begin_0, end = var_43829_end_0, end_mask = var_43829_end_mask_0, x = q_195_cast_fp16)[name = tensor("op_43829_cast_fp16")]; + tensor var_43833_begin_0 = const()[name = tensor("op_43833_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_43833_end_0 = const()[name = tensor("op_43833_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_43833_end_mask_0 = const()[name = tensor("op_43833_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43833_cast_fp16 = slice_by_index(begin = var_43833_begin_0, end = var_43833_end_0, end_mask = var_43833_end_mask_0, x = q_195_cast_fp16)[name = tensor("op_43833_cast_fp16")]; + tensor var_43837_begin_0 = const()[name = tensor("op_43837_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_43837_end_0 = const()[name = tensor("op_43837_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_43837_end_mask_0 = const()[name = tensor("op_43837_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43837_cast_fp16 = slice_by_index(begin = var_43837_begin_0, end = var_43837_end_0, end_mask = var_43837_end_mask_0, x = q_195_cast_fp16)[name = tensor("op_43837_cast_fp16")]; + tensor var_43841_begin_0 = const()[name = tensor("op_43841_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_43841_end_0 = const()[name = tensor("op_43841_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_43841_end_mask_0 = const()[name = tensor("op_43841_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43841_cast_fp16 = slice_by_index(begin = var_43841_begin_0, end = var_43841_end_0, end_mask = var_43841_end_mask_0, x = q_195_cast_fp16)[name = tensor("op_43841_cast_fp16")]; + tensor var_43845_begin_0 = const()[name = tensor("op_43845_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_43845_end_0 = const()[name = tensor("op_43845_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_43845_end_mask_0 = const()[name = tensor("op_43845_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43845_cast_fp16 = slice_by_index(begin = var_43845_begin_0, end = var_43845_end_0, end_mask = var_43845_end_mask_0, x = q_195_cast_fp16)[name = tensor("op_43845_cast_fp16")]; + tensor var_43849_begin_0 = const()[name = tensor("op_43849_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_43849_end_0 = const()[name = tensor("op_43849_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_43849_end_mask_0 = const()[name = tensor("op_43849_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43849_cast_fp16 = slice_by_index(begin = var_43849_begin_0, end = var_43849_end_0, end_mask = var_43849_end_mask_0, x = q_195_cast_fp16)[name = tensor("op_43849_cast_fp16")]; + tensor var_43853_begin_0 = const()[name = tensor("op_43853_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_43853_end_0 = const()[name = tensor("op_43853_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_43853_end_mask_0 = const()[name = tensor("op_43853_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43853_cast_fp16 = slice_by_index(begin = var_43853_begin_0, end = var_43853_end_0, end_mask = var_43853_end_mask_0, x = q_195_cast_fp16)[name = tensor("op_43853_cast_fp16")]; + tensor var_43857_begin_0 = const()[name = tensor("op_43857_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_43857_end_0 = const()[name = tensor("op_43857_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_43857_end_mask_0 = const()[name = tensor("op_43857_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43857_cast_fp16 = slice_by_index(begin = var_43857_begin_0, end = var_43857_end_0, end_mask = var_43857_end_mask_0, x = q_195_cast_fp16)[name = tensor("op_43857_cast_fp16")]; + tensor var_43861_begin_0 = const()[name = tensor("op_43861_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_43861_end_0 = const()[name = tensor("op_43861_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_43861_end_mask_0 = const()[name = tensor("op_43861_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43861_cast_fp16 = slice_by_index(begin = var_43861_begin_0, end = var_43861_end_0, end_mask = var_43861_end_mask_0, x = q_195_cast_fp16)[name = tensor("op_43861_cast_fp16")]; + tensor var_43865_begin_0 = const()[name = tensor("op_43865_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_43865_end_0 = const()[name = tensor("op_43865_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_43865_end_mask_0 = const()[name = tensor("op_43865_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43865_cast_fp16 = slice_by_index(begin = var_43865_begin_0, end = var_43865_end_0, end_mask = var_43865_end_mask_0, x = q_195_cast_fp16)[name = tensor("op_43865_cast_fp16")]; + tensor var_43869_begin_0 = const()[name = tensor("op_43869_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_43869_end_0 = const()[name = tensor("op_43869_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_43869_end_mask_0 = const()[name = tensor("op_43869_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43869_cast_fp16 = slice_by_index(begin = var_43869_begin_0, end = var_43869_end_0, end_mask = var_43869_end_mask_0, x = q_195_cast_fp16)[name = tensor("op_43869_cast_fp16")]; + tensor k_391_perm_0 = const()[name = tensor("k_391_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_43876_begin_0 = const()[name = tensor("op_43876_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43876_end_0 = const()[name = tensor("op_43876_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_43876_end_mask_0 = const()[name = tensor("op_43876_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_391_cast_fp16 = transpose(perm = k_391_perm_0, x = k_389_cast_fp16)[name = tensor("transpose_42")]; + tensor var_43876_cast_fp16 = slice_by_index(begin = var_43876_begin_0, end = var_43876_end_0, end_mask = var_43876_end_mask_0, x = k_391_cast_fp16)[name = tensor("op_43876_cast_fp16")]; + tensor var_43880_begin_0 = const()[name = tensor("op_43880_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_43880_end_0 = const()[name = tensor("op_43880_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_43880_end_mask_0 = const()[name = tensor("op_43880_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43880_cast_fp16 = slice_by_index(begin = var_43880_begin_0, end = var_43880_end_0, end_mask = var_43880_end_mask_0, x = k_391_cast_fp16)[name = tensor("op_43880_cast_fp16")]; + tensor var_43884_begin_0 = const()[name = tensor("op_43884_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_43884_end_0 = const()[name = tensor("op_43884_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_43884_end_mask_0 = const()[name = tensor("op_43884_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43884_cast_fp16 = slice_by_index(begin = var_43884_begin_0, end = var_43884_end_0, end_mask = var_43884_end_mask_0, x = k_391_cast_fp16)[name = tensor("op_43884_cast_fp16")]; + tensor var_43888_begin_0 = const()[name = tensor("op_43888_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_43888_end_0 = const()[name = tensor("op_43888_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_43888_end_mask_0 = const()[name = tensor("op_43888_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43888_cast_fp16 = slice_by_index(begin = var_43888_begin_0, end = var_43888_end_0, end_mask = var_43888_end_mask_0, x = k_391_cast_fp16)[name = tensor("op_43888_cast_fp16")]; + tensor var_43892_begin_0 = const()[name = tensor("op_43892_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_43892_end_0 = const()[name = tensor("op_43892_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_43892_end_mask_0 = const()[name = tensor("op_43892_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43892_cast_fp16 = slice_by_index(begin = var_43892_begin_0, end = var_43892_end_0, end_mask = var_43892_end_mask_0, x = k_391_cast_fp16)[name = tensor("op_43892_cast_fp16")]; + tensor var_43896_begin_0 = const()[name = tensor("op_43896_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_43896_end_0 = const()[name = tensor("op_43896_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_43896_end_mask_0 = const()[name = tensor("op_43896_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43896_cast_fp16 = slice_by_index(begin = var_43896_begin_0, end = var_43896_end_0, end_mask = var_43896_end_mask_0, x = k_391_cast_fp16)[name = tensor("op_43896_cast_fp16")]; + tensor var_43900_begin_0 = const()[name = tensor("op_43900_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_43900_end_0 = const()[name = tensor("op_43900_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_43900_end_mask_0 = const()[name = tensor("op_43900_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43900_cast_fp16 = slice_by_index(begin = var_43900_begin_0, end = var_43900_end_0, end_mask = var_43900_end_mask_0, x = k_391_cast_fp16)[name = tensor("op_43900_cast_fp16")]; + tensor var_43904_begin_0 = const()[name = tensor("op_43904_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_43904_end_0 = const()[name = tensor("op_43904_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_43904_end_mask_0 = const()[name = tensor("op_43904_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43904_cast_fp16 = slice_by_index(begin = var_43904_begin_0, end = var_43904_end_0, end_mask = var_43904_end_mask_0, x = k_391_cast_fp16)[name = tensor("op_43904_cast_fp16")]; + tensor var_43908_begin_0 = const()[name = tensor("op_43908_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_43908_end_0 = const()[name = tensor("op_43908_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_43908_end_mask_0 = const()[name = tensor("op_43908_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43908_cast_fp16 = slice_by_index(begin = var_43908_begin_0, end = var_43908_end_0, end_mask = var_43908_end_mask_0, x = k_391_cast_fp16)[name = tensor("op_43908_cast_fp16")]; + tensor var_43912_begin_0 = const()[name = tensor("op_43912_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_43912_end_0 = const()[name = tensor("op_43912_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_43912_end_mask_0 = const()[name = tensor("op_43912_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43912_cast_fp16 = slice_by_index(begin = var_43912_begin_0, end = var_43912_end_0, end_mask = var_43912_end_mask_0, x = k_391_cast_fp16)[name = tensor("op_43912_cast_fp16")]; + tensor var_43916_begin_0 = const()[name = tensor("op_43916_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_43916_end_0 = const()[name = tensor("op_43916_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_43916_end_mask_0 = const()[name = tensor("op_43916_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43916_cast_fp16 = slice_by_index(begin = var_43916_begin_0, end = var_43916_end_0, end_mask = var_43916_end_mask_0, x = k_391_cast_fp16)[name = tensor("op_43916_cast_fp16")]; + tensor var_43920_begin_0 = const()[name = tensor("op_43920_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_43920_end_0 = const()[name = tensor("op_43920_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_43920_end_mask_0 = const()[name = tensor("op_43920_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43920_cast_fp16 = slice_by_index(begin = var_43920_begin_0, end = var_43920_end_0, end_mask = var_43920_end_mask_0, x = k_391_cast_fp16)[name = tensor("op_43920_cast_fp16")]; + tensor var_43924_begin_0 = const()[name = tensor("op_43924_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_43924_end_0 = const()[name = tensor("op_43924_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_43924_end_mask_0 = const()[name = tensor("op_43924_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43924_cast_fp16 = slice_by_index(begin = var_43924_begin_0, end = var_43924_end_0, end_mask = var_43924_end_mask_0, x = k_391_cast_fp16)[name = tensor("op_43924_cast_fp16")]; + tensor var_43928_begin_0 = const()[name = tensor("op_43928_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_43928_end_0 = const()[name = tensor("op_43928_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_43928_end_mask_0 = const()[name = tensor("op_43928_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43928_cast_fp16 = slice_by_index(begin = var_43928_begin_0, end = var_43928_end_0, end_mask = var_43928_end_mask_0, x = k_391_cast_fp16)[name = tensor("op_43928_cast_fp16")]; + tensor var_43932_begin_0 = const()[name = tensor("op_43932_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_43932_end_0 = const()[name = tensor("op_43932_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_43932_end_mask_0 = const()[name = tensor("op_43932_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43932_cast_fp16 = slice_by_index(begin = var_43932_begin_0, end = var_43932_end_0, end_mask = var_43932_end_mask_0, x = k_391_cast_fp16)[name = tensor("op_43932_cast_fp16")]; + tensor var_43936_begin_0 = const()[name = tensor("op_43936_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_43936_end_0 = const()[name = tensor("op_43936_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_43936_end_mask_0 = const()[name = tensor("op_43936_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43936_cast_fp16 = slice_by_index(begin = var_43936_begin_0, end = var_43936_end_0, end_mask = var_43936_end_mask_0, x = k_391_cast_fp16)[name = tensor("op_43936_cast_fp16")]; + tensor var_43940_begin_0 = const()[name = tensor("op_43940_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_43940_end_0 = const()[name = tensor("op_43940_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_43940_end_mask_0 = const()[name = tensor("op_43940_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43940_cast_fp16 = slice_by_index(begin = var_43940_begin_0, end = var_43940_end_0, end_mask = var_43940_end_mask_0, x = k_391_cast_fp16)[name = tensor("op_43940_cast_fp16")]; + tensor var_43944_begin_0 = const()[name = tensor("op_43944_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_43944_end_0 = const()[name = tensor("op_43944_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_43944_end_mask_0 = const()[name = tensor("op_43944_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43944_cast_fp16 = slice_by_index(begin = var_43944_begin_0, end = var_43944_end_0, end_mask = var_43944_end_mask_0, x = k_391_cast_fp16)[name = tensor("op_43944_cast_fp16")]; + tensor var_43948_begin_0 = const()[name = tensor("op_43948_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_43948_end_0 = const()[name = tensor("op_43948_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_43948_end_mask_0 = const()[name = tensor("op_43948_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43948_cast_fp16 = slice_by_index(begin = var_43948_begin_0, end = var_43948_end_0, end_mask = var_43948_end_mask_0, x = k_391_cast_fp16)[name = tensor("op_43948_cast_fp16")]; + tensor var_43952_begin_0 = const()[name = tensor("op_43952_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_43952_end_0 = const()[name = tensor("op_43952_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_43952_end_mask_0 = const()[name = tensor("op_43952_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43952_cast_fp16 = slice_by_index(begin = var_43952_begin_0, end = var_43952_end_0, end_mask = var_43952_end_mask_0, x = k_391_cast_fp16)[name = tensor("op_43952_cast_fp16")]; + tensor var_43954_begin_0 = const()[name = tensor("op_43954_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43954_end_0 = const()[name = tensor("op_43954_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_43954_end_mask_0 = const()[name = tensor("op_43954_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43954_cast_fp16 = slice_by_index(begin = var_43954_begin_0, end = var_43954_end_0, end_mask = var_43954_end_mask_0, x = v_195_cast_fp16)[name = tensor("op_43954_cast_fp16")]; + tensor var_43958_begin_0 = const()[name = tensor("op_43958_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_43958_end_0 = const()[name = tensor("op_43958_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_43958_end_mask_0 = const()[name = tensor("op_43958_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43958_cast_fp16 = slice_by_index(begin = var_43958_begin_0, end = var_43958_end_0, end_mask = var_43958_end_mask_0, x = v_195_cast_fp16)[name = tensor("op_43958_cast_fp16")]; + tensor var_43962_begin_0 = const()[name = tensor("op_43962_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_43962_end_0 = const()[name = tensor("op_43962_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_43962_end_mask_0 = const()[name = tensor("op_43962_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43962_cast_fp16 = slice_by_index(begin = var_43962_begin_0, end = var_43962_end_0, end_mask = var_43962_end_mask_0, x = v_195_cast_fp16)[name = tensor("op_43962_cast_fp16")]; + tensor var_43966_begin_0 = const()[name = tensor("op_43966_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_43966_end_0 = const()[name = tensor("op_43966_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_43966_end_mask_0 = const()[name = tensor("op_43966_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43966_cast_fp16 = slice_by_index(begin = var_43966_begin_0, end = var_43966_end_0, end_mask = var_43966_end_mask_0, x = v_195_cast_fp16)[name = tensor("op_43966_cast_fp16")]; + tensor var_43970_begin_0 = const()[name = tensor("op_43970_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_43970_end_0 = const()[name = tensor("op_43970_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_43970_end_mask_0 = const()[name = tensor("op_43970_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43970_cast_fp16 = slice_by_index(begin = var_43970_begin_0, end = var_43970_end_0, end_mask = var_43970_end_mask_0, x = v_195_cast_fp16)[name = tensor("op_43970_cast_fp16")]; + tensor var_43974_begin_0 = const()[name = tensor("op_43974_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_43974_end_0 = const()[name = tensor("op_43974_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_43974_end_mask_0 = const()[name = tensor("op_43974_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43974_cast_fp16 = slice_by_index(begin = var_43974_begin_0, end = var_43974_end_0, end_mask = var_43974_end_mask_0, x = v_195_cast_fp16)[name = tensor("op_43974_cast_fp16")]; + tensor var_43978_begin_0 = const()[name = tensor("op_43978_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_43978_end_0 = const()[name = tensor("op_43978_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_43978_end_mask_0 = const()[name = tensor("op_43978_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43978_cast_fp16 = slice_by_index(begin = var_43978_begin_0, end = var_43978_end_0, end_mask = var_43978_end_mask_0, x = v_195_cast_fp16)[name = tensor("op_43978_cast_fp16")]; + tensor var_43982_begin_0 = const()[name = tensor("op_43982_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_43982_end_0 = const()[name = tensor("op_43982_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_43982_end_mask_0 = const()[name = tensor("op_43982_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43982_cast_fp16 = slice_by_index(begin = var_43982_begin_0, end = var_43982_end_0, end_mask = var_43982_end_mask_0, x = v_195_cast_fp16)[name = tensor("op_43982_cast_fp16")]; + tensor var_43986_begin_0 = const()[name = tensor("op_43986_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_43986_end_0 = const()[name = tensor("op_43986_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_43986_end_mask_0 = const()[name = tensor("op_43986_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43986_cast_fp16 = slice_by_index(begin = var_43986_begin_0, end = var_43986_end_0, end_mask = var_43986_end_mask_0, x = v_195_cast_fp16)[name = tensor("op_43986_cast_fp16")]; + tensor var_43990_begin_0 = const()[name = tensor("op_43990_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_43990_end_0 = const()[name = tensor("op_43990_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_43990_end_mask_0 = const()[name = tensor("op_43990_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43990_cast_fp16 = slice_by_index(begin = var_43990_begin_0, end = var_43990_end_0, end_mask = var_43990_end_mask_0, x = v_195_cast_fp16)[name = tensor("op_43990_cast_fp16")]; + tensor var_43994_begin_0 = const()[name = tensor("op_43994_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_43994_end_0 = const()[name = tensor("op_43994_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_43994_end_mask_0 = const()[name = tensor("op_43994_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43994_cast_fp16 = slice_by_index(begin = var_43994_begin_0, end = var_43994_end_0, end_mask = var_43994_end_mask_0, x = v_195_cast_fp16)[name = tensor("op_43994_cast_fp16")]; + tensor var_43998_begin_0 = const()[name = tensor("op_43998_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_43998_end_0 = const()[name = tensor("op_43998_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_43998_end_mask_0 = const()[name = tensor("op_43998_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43998_cast_fp16 = slice_by_index(begin = var_43998_begin_0, end = var_43998_end_0, end_mask = var_43998_end_mask_0, x = v_195_cast_fp16)[name = tensor("op_43998_cast_fp16")]; + tensor var_44002_begin_0 = const()[name = tensor("op_44002_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_44002_end_0 = const()[name = tensor("op_44002_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_44002_end_mask_0 = const()[name = tensor("op_44002_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44002_cast_fp16 = slice_by_index(begin = var_44002_begin_0, end = var_44002_end_0, end_mask = var_44002_end_mask_0, x = v_195_cast_fp16)[name = tensor("op_44002_cast_fp16")]; + tensor var_44006_begin_0 = const()[name = tensor("op_44006_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_44006_end_0 = const()[name = tensor("op_44006_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_44006_end_mask_0 = const()[name = tensor("op_44006_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44006_cast_fp16 = slice_by_index(begin = var_44006_begin_0, end = var_44006_end_0, end_mask = var_44006_end_mask_0, x = v_195_cast_fp16)[name = tensor("op_44006_cast_fp16")]; + tensor var_44010_begin_0 = const()[name = tensor("op_44010_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_44010_end_0 = const()[name = tensor("op_44010_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_44010_end_mask_0 = const()[name = tensor("op_44010_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44010_cast_fp16 = slice_by_index(begin = var_44010_begin_0, end = var_44010_end_0, end_mask = var_44010_end_mask_0, x = v_195_cast_fp16)[name = tensor("op_44010_cast_fp16")]; + tensor var_44014_begin_0 = const()[name = tensor("op_44014_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_44014_end_0 = const()[name = tensor("op_44014_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_44014_end_mask_0 = const()[name = tensor("op_44014_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44014_cast_fp16 = slice_by_index(begin = var_44014_begin_0, end = var_44014_end_0, end_mask = var_44014_end_mask_0, x = v_195_cast_fp16)[name = tensor("op_44014_cast_fp16")]; + tensor var_44018_begin_0 = const()[name = tensor("op_44018_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_44018_end_0 = const()[name = tensor("op_44018_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_44018_end_mask_0 = const()[name = tensor("op_44018_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44018_cast_fp16 = slice_by_index(begin = var_44018_begin_0, end = var_44018_end_0, end_mask = var_44018_end_mask_0, x = v_195_cast_fp16)[name = tensor("op_44018_cast_fp16")]; + tensor var_44022_begin_0 = const()[name = tensor("op_44022_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_44022_end_0 = const()[name = tensor("op_44022_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_44022_end_mask_0 = const()[name = tensor("op_44022_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44022_cast_fp16 = slice_by_index(begin = var_44022_begin_0, end = var_44022_end_0, end_mask = var_44022_end_mask_0, x = v_195_cast_fp16)[name = tensor("op_44022_cast_fp16")]; + tensor var_44026_begin_0 = const()[name = tensor("op_44026_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_44026_end_0 = const()[name = tensor("op_44026_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_44026_end_mask_0 = const()[name = tensor("op_44026_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44026_cast_fp16 = slice_by_index(begin = var_44026_begin_0, end = var_44026_end_0, end_mask = var_44026_end_mask_0, x = v_195_cast_fp16)[name = tensor("op_44026_cast_fp16")]; + tensor var_44030_begin_0 = const()[name = tensor("op_44030_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_44030_end_0 = const()[name = tensor("op_44030_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_44030_end_mask_0 = const()[name = tensor("op_44030_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44030_cast_fp16 = slice_by_index(begin = var_44030_begin_0, end = var_44030_end_0, end_mask = var_44030_end_mask_0, x = v_195_cast_fp16)[name = tensor("op_44030_cast_fp16")]; + tensor var_44034_equation_0 = const()[name = tensor("op_44034_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_44034_cast_fp16 = einsum(equation = var_44034_equation_0, values = (var_43876_cast_fp16, var_43793_cast_fp16))[name = tensor("op_44034_cast_fp16")]; + tensor var_44035_to_fp16 = const()[name = tensor("op_44035_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3721_cast_fp16 = mul(x = var_44034_cast_fp16, y = var_44035_to_fp16)[name = tensor("aw_3721_cast_fp16")]; + tensor var_44038_equation_0 = const()[name = tensor("op_44038_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_44038_cast_fp16 = einsum(equation = var_44038_equation_0, values = (var_43880_cast_fp16, var_43797_cast_fp16))[name = tensor("op_44038_cast_fp16")]; + tensor var_44039_to_fp16 = const()[name = tensor("op_44039_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3723_cast_fp16 = mul(x = var_44038_cast_fp16, y = var_44039_to_fp16)[name = tensor("aw_3723_cast_fp16")]; + tensor var_44042_equation_0 = const()[name = tensor("op_44042_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_44042_cast_fp16 = einsum(equation = var_44042_equation_0, values = (var_43884_cast_fp16, var_43801_cast_fp16))[name = tensor("op_44042_cast_fp16")]; + tensor var_44043_to_fp16 = const()[name = tensor("op_44043_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3725_cast_fp16 = mul(x = var_44042_cast_fp16, y = var_44043_to_fp16)[name = tensor("aw_3725_cast_fp16")]; + tensor var_44046_equation_0 = const()[name = tensor("op_44046_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_44046_cast_fp16 = einsum(equation = var_44046_equation_0, values = (var_43888_cast_fp16, var_43805_cast_fp16))[name = tensor("op_44046_cast_fp16")]; + tensor var_44047_to_fp16 = const()[name = tensor("op_44047_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3727_cast_fp16 = mul(x = var_44046_cast_fp16, y = var_44047_to_fp16)[name = tensor("aw_3727_cast_fp16")]; + tensor var_44050_equation_0 = const()[name = tensor("op_44050_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_44050_cast_fp16 = einsum(equation = var_44050_equation_0, values = (var_43892_cast_fp16, var_43809_cast_fp16))[name = tensor("op_44050_cast_fp16")]; + tensor var_44051_to_fp16 = const()[name = tensor("op_44051_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3729_cast_fp16 = mul(x = var_44050_cast_fp16, y = var_44051_to_fp16)[name = tensor("aw_3729_cast_fp16")]; + tensor var_44054_equation_0 = const()[name = tensor("op_44054_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_44054_cast_fp16 = einsum(equation = var_44054_equation_0, values = (var_43896_cast_fp16, var_43813_cast_fp16))[name = tensor("op_44054_cast_fp16")]; + tensor var_44055_to_fp16 = const()[name = tensor("op_44055_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3731_cast_fp16 = mul(x = var_44054_cast_fp16, y = var_44055_to_fp16)[name = tensor("aw_3731_cast_fp16")]; + tensor var_44058_equation_0 = const()[name = tensor("op_44058_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_44058_cast_fp16 = einsum(equation = var_44058_equation_0, values = (var_43900_cast_fp16, var_43817_cast_fp16))[name = tensor("op_44058_cast_fp16")]; + tensor var_44059_to_fp16 = const()[name = tensor("op_44059_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3733_cast_fp16 = mul(x = var_44058_cast_fp16, y = var_44059_to_fp16)[name = tensor("aw_3733_cast_fp16")]; + tensor var_44062_equation_0 = const()[name = tensor("op_44062_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_44062_cast_fp16 = einsum(equation = var_44062_equation_0, values = (var_43904_cast_fp16, var_43821_cast_fp16))[name = tensor("op_44062_cast_fp16")]; + tensor var_44063_to_fp16 = const()[name = tensor("op_44063_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3735_cast_fp16 = mul(x = var_44062_cast_fp16, y = var_44063_to_fp16)[name = tensor("aw_3735_cast_fp16")]; + tensor var_44066_equation_0 = const()[name = tensor("op_44066_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_44066_cast_fp16 = einsum(equation = var_44066_equation_0, values = (var_43908_cast_fp16, var_43825_cast_fp16))[name = tensor("op_44066_cast_fp16")]; + tensor var_44067_to_fp16 = const()[name = tensor("op_44067_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3737_cast_fp16 = mul(x = var_44066_cast_fp16, y = var_44067_to_fp16)[name = tensor("aw_3737_cast_fp16")]; + tensor var_44070_equation_0 = const()[name = tensor("op_44070_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_44070_cast_fp16 = einsum(equation = var_44070_equation_0, values = (var_43912_cast_fp16, var_43829_cast_fp16))[name = tensor("op_44070_cast_fp16")]; + tensor var_44071_to_fp16 = const()[name = tensor("op_44071_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3739_cast_fp16 = mul(x = var_44070_cast_fp16, y = var_44071_to_fp16)[name = tensor("aw_3739_cast_fp16")]; + tensor var_44074_equation_0 = const()[name = tensor("op_44074_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_44074_cast_fp16 = einsum(equation = var_44074_equation_0, values = (var_43916_cast_fp16, var_43833_cast_fp16))[name = tensor("op_44074_cast_fp16")]; + tensor var_44075_to_fp16 = const()[name = tensor("op_44075_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3741_cast_fp16 = mul(x = var_44074_cast_fp16, y = var_44075_to_fp16)[name = tensor("aw_3741_cast_fp16")]; + tensor var_44078_equation_0 = const()[name = tensor("op_44078_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_44078_cast_fp16 = einsum(equation = var_44078_equation_0, values = (var_43920_cast_fp16, var_43837_cast_fp16))[name = tensor("op_44078_cast_fp16")]; + tensor var_44079_to_fp16 = const()[name = tensor("op_44079_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3743_cast_fp16 = mul(x = var_44078_cast_fp16, y = var_44079_to_fp16)[name = tensor("aw_3743_cast_fp16")]; + tensor var_44082_equation_0 = const()[name = tensor("op_44082_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_44082_cast_fp16 = einsum(equation = var_44082_equation_0, values = (var_43924_cast_fp16, var_43841_cast_fp16))[name = tensor("op_44082_cast_fp16")]; + tensor var_44083_to_fp16 = const()[name = tensor("op_44083_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3745_cast_fp16 = mul(x = var_44082_cast_fp16, y = var_44083_to_fp16)[name = tensor("aw_3745_cast_fp16")]; + tensor var_44086_equation_0 = const()[name = tensor("op_44086_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_44086_cast_fp16 = einsum(equation = var_44086_equation_0, values = (var_43928_cast_fp16, var_43845_cast_fp16))[name = tensor("op_44086_cast_fp16")]; + tensor var_44087_to_fp16 = const()[name = tensor("op_44087_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3747_cast_fp16 = mul(x = var_44086_cast_fp16, y = var_44087_to_fp16)[name = tensor("aw_3747_cast_fp16")]; + tensor var_44090_equation_0 = const()[name = tensor("op_44090_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_44090_cast_fp16 = einsum(equation = var_44090_equation_0, values = (var_43932_cast_fp16, var_43849_cast_fp16))[name = tensor("op_44090_cast_fp16")]; + tensor var_44091_to_fp16 = const()[name = tensor("op_44091_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3749_cast_fp16 = mul(x = var_44090_cast_fp16, y = var_44091_to_fp16)[name = tensor("aw_3749_cast_fp16")]; + tensor var_44094_equation_0 = const()[name = tensor("op_44094_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_44094_cast_fp16 = einsum(equation = var_44094_equation_0, values = (var_43936_cast_fp16, var_43853_cast_fp16))[name = tensor("op_44094_cast_fp16")]; + tensor var_44095_to_fp16 = const()[name = tensor("op_44095_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3751_cast_fp16 = mul(x = var_44094_cast_fp16, y = var_44095_to_fp16)[name = tensor("aw_3751_cast_fp16")]; + tensor var_44098_equation_0 = const()[name = tensor("op_44098_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_44098_cast_fp16 = einsum(equation = var_44098_equation_0, values = (var_43940_cast_fp16, var_43857_cast_fp16))[name = tensor("op_44098_cast_fp16")]; + tensor var_44099_to_fp16 = const()[name = tensor("op_44099_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3753_cast_fp16 = mul(x = var_44098_cast_fp16, y = var_44099_to_fp16)[name = tensor("aw_3753_cast_fp16")]; + tensor var_44102_equation_0 = const()[name = tensor("op_44102_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_44102_cast_fp16 = einsum(equation = var_44102_equation_0, values = (var_43944_cast_fp16, var_43861_cast_fp16))[name = tensor("op_44102_cast_fp16")]; + tensor var_44103_to_fp16 = const()[name = tensor("op_44103_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3755_cast_fp16 = mul(x = var_44102_cast_fp16, y = var_44103_to_fp16)[name = tensor("aw_3755_cast_fp16")]; + tensor var_44106_equation_0 = const()[name = tensor("op_44106_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_44106_cast_fp16 = einsum(equation = var_44106_equation_0, values = (var_43948_cast_fp16, var_43865_cast_fp16))[name = tensor("op_44106_cast_fp16")]; + tensor var_44107_to_fp16 = const()[name = tensor("op_44107_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3757_cast_fp16 = mul(x = var_44106_cast_fp16, y = var_44107_to_fp16)[name = tensor("aw_3757_cast_fp16")]; + tensor var_44110_equation_0 = const()[name = tensor("op_44110_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_44110_cast_fp16 = einsum(equation = var_44110_equation_0, values = (var_43952_cast_fp16, var_43869_cast_fp16))[name = tensor("op_44110_cast_fp16")]; + tensor var_44111_to_fp16 = const()[name = tensor("op_44111_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3759_cast_fp16 = mul(x = var_44110_cast_fp16, y = var_44111_to_fp16)[name = tensor("aw_3759_cast_fp16")]; + tensor var_44113_cast_fp16 = softmax(axis = var_30355, x = aw_3721_cast_fp16)[name = tensor("op_44113_cast_fp16")]; + tensor var_44114_cast_fp16 = softmax(axis = var_30355, x = aw_3723_cast_fp16)[name = tensor("op_44114_cast_fp16")]; + tensor var_44115_cast_fp16 = softmax(axis = var_30355, x = aw_3725_cast_fp16)[name = tensor("op_44115_cast_fp16")]; + tensor var_44116_cast_fp16 = softmax(axis = var_30355, x = aw_3727_cast_fp16)[name = tensor("op_44116_cast_fp16")]; + tensor var_44117_cast_fp16 = softmax(axis = var_30355, x = aw_3729_cast_fp16)[name = tensor("op_44117_cast_fp16")]; + tensor var_44118_cast_fp16 = softmax(axis = var_30355, x = aw_3731_cast_fp16)[name = tensor("op_44118_cast_fp16")]; + tensor var_44119_cast_fp16 = softmax(axis = var_30355, x = aw_3733_cast_fp16)[name = tensor("op_44119_cast_fp16")]; + tensor var_44120_cast_fp16 = softmax(axis = var_30355, x = aw_3735_cast_fp16)[name = tensor("op_44120_cast_fp16")]; + tensor var_44121_cast_fp16 = softmax(axis = var_30355, x = aw_3737_cast_fp16)[name = tensor("op_44121_cast_fp16")]; + tensor var_44122_cast_fp16 = softmax(axis = var_30355, x = aw_3739_cast_fp16)[name = tensor("op_44122_cast_fp16")]; + tensor var_44123_cast_fp16 = softmax(axis = var_30355, x = aw_3741_cast_fp16)[name = tensor("op_44123_cast_fp16")]; + tensor var_44124_cast_fp16 = softmax(axis = var_30355, x = aw_3743_cast_fp16)[name = tensor("op_44124_cast_fp16")]; + tensor var_44125_cast_fp16 = softmax(axis = var_30355, x = aw_3745_cast_fp16)[name = tensor("op_44125_cast_fp16")]; + tensor var_44126_cast_fp16 = softmax(axis = var_30355, x = aw_3747_cast_fp16)[name = tensor("op_44126_cast_fp16")]; + tensor var_44127_cast_fp16 = softmax(axis = var_30355, x = aw_3749_cast_fp16)[name = tensor("op_44127_cast_fp16")]; + tensor var_44128_cast_fp16 = softmax(axis = var_30355, x = aw_3751_cast_fp16)[name = tensor("op_44128_cast_fp16")]; + tensor var_44129_cast_fp16 = softmax(axis = var_30355, x = aw_3753_cast_fp16)[name = tensor("op_44129_cast_fp16")]; + tensor var_44130_cast_fp16 = softmax(axis = var_30355, x = aw_3755_cast_fp16)[name = tensor("op_44130_cast_fp16")]; + tensor var_44131_cast_fp16 = softmax(axis = var_30355, x = aw_3757_cast_fp16)[name = tensor("op_44131_cast_fp16")]; + tensor var_44132_cast_fp16 = softmax(axis = var_30355, x = aw_3759_cast_fp16)[name = tensor("op_44132_cast_fp16")]; + tensor var_44134_equation_0 = const()[name = tensor("op_44134_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44134_cast_fp16 = einsum(equation = var_44134_equation_0, values = (var_43954_cast_fp16, var_44113_cast_fp16))[name = tensor("op_44134_cast_fp16")]; + tensor var_44136_equation_0 = const()[name = tensor("op_44136_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44136_cast_fp16 = einsum(equation = var_44136_equation_0, values = (var_43958_cast_fp16, var_44114_cast_fp16))[name = tensor("op_44136_cast_fp16")]; + tensor var_44138_equation_0 = const()[name = tensor("op_44138_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44138_cast_fp16 = einsum(equation = var_44138_equation_0, values = (var_43962_cast_fp16, var_44115_cast_fp16))[name = tensor("op_44138_cast_fp16")]; + tensor var_44140_equation_0 = const()[name = tensor("op_44140_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44140_cast_fp16 = einsum(equation = var_44140_equation_0, values = (var_43966_cast_fp16, var_44116_cast_fp16))[name = tensor("op_44140_cast_fp16")]; + tensor var_44142_equation_0 = const()[name = tensor("op_44142_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44142_cast_fp16 = einsum(equation = var_44142_equation_0, values = (var_43970_cast_fp16, var_44117_cast_fp16))[name = tensor("op_44142_cast_fp16")]; + tensor var_44144_equation_0 = const()[name = tensor("op_44144_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44144_cast_fp16 = einsum(equation = var_44144_equation_0, values = (var_43974_cast_fp16, var_44118_cast_fp16))[name = tensor("op_44144_cast_fp16")]; + tensor var_44146_equation_0 = const()[name = tensor("op_44146_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44146_cast_fp16 = einsum(equation = var_44146_equation_0, values = (var_43978_cast_fp16, var_44119_cast_fp16))[name = tensor("op_44146_cast_fp16")]; + tensor var_44148_equation_0 = const()[name = tensor("op_44148_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44148_cast_fp16 = einsum(equation = var_44148_equation_0, values = (var_43982_cast_fp16, var_44120_cast_fp16))[name = tensor("op_44148_cast_fp16")]; + tensor var_44150_equation_0 = const()[name = tensor("op_44150_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44150_cast_fp16 = einsum(equation = var_44150_equation_0, values = (var_43986_cast_fp16, var_44121_cast_fp16))[name = tensor("op_44150_cast_fp16")]; + tensor var_44152_equation_0 = const()[name = tensor("op_44152_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44152_cast_fp16 = einsum(equation = var_44152_equation_0, values = (var_43990_cast_fp16, var_44122_cast_fp16))[name = tensor("op_44152_cast_fp16")]; + tensor var_44154_equation_0 = const()[name = tensor("op_44154_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44154_cast_fp16 = einsum(equation = var_44154_equation_0, values = (var_43994_cast_fp16, var_44123_cast_fp16))[name = tensor("op_44154_cast_fp16")]; + tensor var_44156_equation_0 = const()[name = tensor("op_44156_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44156_cast_fp16 = einsum(equation = var_44156_equation_0, values = (var_43998_cast_fp16, var_44124_cast_fp16))[name = tensor("op_44156_cast_fp16")]; + tensor var_44158_equation_0 = const()[name = tensor("op_44158_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44158_cast_fp16 = einsum(equation = var_44158_equation_0, values = (var_44002_cast_fp16, var_44125_cast_fp16))[name = tensor("op_44158_cast_fp16")]; + tensor var_44160_equation_0 = const()[name = tensor("op_44160_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44160_cast_fp16 = einsum(equation = var_44160_equation_0, values = (var_44006_cast_fp16, var_44126_cast_fp16))[name = tensor("op_44160_cast_fp16")]; + tensor var_44162_equation_0 = const()[name = tensor("op_44162_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44162_cast_fp16 = einsum(equation = var_44162_equation_0, values = (var_44010_cast_fp16, var_44127_cast_fp16))[name = tensor("op_44162_cast_fp16")]; + tensor var_44164_equation_0 = const()[name = tensor("op_44164_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44164_cast_fp16 = einsum(equation = var_44164_equation_0, values = (var_44014_cast_fp16, var_44128_cast_fp16))[name = tensor("op_44164_cast_fp16")]; + tensor var_44166_equation_0 = const()[name = tensor("op_44166_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44166_cast_fp16 = einsum(equation = var_44166_equation_0, values = (var_44018_cast_fp16, var_44129_cast_fp16))[name = tensor("op_44166_cast_fp16")]; + tensor var_44168_equation_0 = const()[name = tensor("op_44168_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44168_cast_fp16 = einsum(equation = var_44168_equation_0, values = (var_44022_cast_fp16, var_44130_cast_fp16))[name = tensor("op_44168_cast_fp16")]; + tensor var_44170_equation_0 = const()[name = tensor("op_44170_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44170_cast_fp16 = einsum(equation = var_44170_equation_0, values = (var_44026_cast_fp16, var_44131_cast_fp16))[name = tensor("op_44170_cast_fp16")]; + tensor var_44172_equation_0 = const()[name = tensor("op_44172_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44172_cast_fp16 = einsum(equation = var_44172_equation_0, values = (var_44030_cast_fp16, var_44132_cast_fp16))[name = tensor("op_44172_cast_fp16")]; + tensor input_571_interleave_0 = const()[name = tensor("input_571_interleave_0"), val = tensor(false)]; + tensor input_571_cast_fp16 = concat(axis = var_30355, interleave = input_571_interleave_0, values = (var_44134_cast_fp16, var_44136_cast_fp16, var_44138_cast_fp16, var_44140_cast_fp16, var_44142_cast_fp16, var_44144_cast_fp16, var_44146_cast_fp16, var_44148_cast_fp16, var_44150_cast_fp16, var_44152_cast_fp16, var_44154_cast_fp16, var_44156_cast_fp16, var_44158_cast_fp16, var_44160_cast_fp16, var_44162_cast_fp16, var_44164_cast_fp16, var_44166_cast_fp16, var_44168_cast_fp16, var_44170_cast_fp16, var_44172_cast_fp16))[name = tensor("input_571_cast_fp16")]; + tensor var_44178 = const()[name = tensor("op_44178"), val = tensor([1, 1])]; + tensor var_44180 = const()[name = tensor("op_44180"), val = tensor([1, 1])]; + tensor var_44182_pad_type_0 = const()[name = tensor("op_44182_pad_type_0"), val = tensor("custom")]; + tensor var_44182_pad_0 = const()[name = tensor("op_44182_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_4_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_4_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3693380736)))]; + tensor up_blocks_0_attentions_1_transformer_blocks_4_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_4_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3696657600)))]; + tensor var_44182_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_4_attn2_to_out_0_bias_to_fp16, dilations = var_44180, groups = var_30355, pad = var_44182_pad_0, pad_type = var_44182_pad_type_0, strides = var_44178, weight = up_blocks_0_attentions_1_transformer_blocks_4_attn2_to_out_0_weight_to_fp16, x = input_571_cast_fp16)[name = tensor("op_44182_cast_fp16")]; + tensor inputs_293_cast_fp16 = add(x = var_44182_cast_fp16, y = inputs_291_cast_fp16)[name = tensor("inputs_293_cast_fp16")]; + tensor input_573_axes_0 = const()[name = tensor("input_573_axes_0"), val = tensor([1])]; + tensor input_573_gamma_0_to_fp16 = const()[name = tensor("input_573_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3696660224)))]; + tensor input_573_beta_0_to_fp16 = const()[name = tensor("input_573_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3696662848)))]; + tensor var_44192_to_fp16 = const()[name = tensor("op_44192_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_573_cast_fp16 = layer_norm(axes = input_573_axes_0, beta = input_573_beta_0_to_fp16, epsilon = var_44192_to_fp16, gamma = input_573_gamma_0_to_fp16, x = inputs_293_cast_fp16)[name = tensor("input_573_cast_fp16")]; + tensor var_44208 = const()[name = tensor("op_44208"), val = tensor([1, 1])]; + tensor var_44210 = const()[name = tensor("op_44210"), val = tensor([1, 1])]; + tensor var_44212_pad_type_0 = const()[name = tensor("op_44212_pad_type_0"), val = tensor("custom")]; + tensor var_44212_pad_0 = const()[name = tensor("op_44212_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_4_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_4_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3696665472)))]; + tensor up_blocks_0_attentions_1_transformer_blocks_4_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_4_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3722879936)))]; + tensor var_44212_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_4_ff_net_0_proj_bias_to_fp16, dilations = var_44210, groups = var_30355, pad = var_44212_pad_0, pad_type = var_44212_pad_type_0, strides = var_44208, weight = up_blocks_0_attentions_1_transformer_blocks_4_ff_net_0_proj_weight_to_fp16, x = input_573_cast_fp16)[name = tensor("op_44212_cast_fp16")]; + tensor var_44213_split_sizes_0 = const()[name = tensor("op_44213_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_44213_axis_0 = const()[name = tensor("op_44213_axis_0"), val = tensor(1)]; + tensor var_44213_cast_fp16_0, tensor var_44213_cast_fp16_1 = split(axis = var_44213_axis_0, split_sizes = var_44213_split_sizes_0, x = var_44212_cast_fp16)[name = tensor("op_44213_cast_fp16")]; + tensor var_44215_mode_0 = const()[name = tensor("op_44215_mode_0"), val = tensor("EXACT")]; + tensor var_44215_cast_fp16 = gelu(mode = var_44215_mode_0, x = var_44213_cast_fp16_1)[name = tensor("op_44215_cast_fp16")]; + tensor input_575_cast_fp16 = mul(x = var_44213_cast_fp16_0, y = var_44215_cast_fp16)[name = tensor("input_575_cast_fp16")]; + tensor var_44219 = const()[name = tensor("op_44219"), val = tensor([1, 1])]; + tensor var_44221 = const()[name = tensor("op_44221"), val = tensor([1, 1])]; + tensor var_44223_pad_type_0 = const()[name = tensor("op_44223_pad_type_0"), val = tensor("custom")]; + tensor var_44223_pad_0 = const()[name = tensor("op_44223_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_4_ff_net_2_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_4_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3722900480)))]; + tensor up_blocks_0_attentions_1_transformer_blocks_4_ff_net_2_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_4_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3736007744)))]; + tensor var_44223_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_4_ff_net_2_bias_to_fp16, dilations = var_44221, groups = var_30355, pad = var_44223_pad_0, pad_type = var_44223_pad_type_0, strides = var_44219, weight = up_blocks_0_attentions_1_transformer_blocks_4_ff_net_2_weight_to_fp16, x = input_575_cast_fp16)[name = tensor("op_44223_cast_fp16")]; + tensor inputs_295_cast_fp16 = add(x = var_44223_cast_fp16, y = inputs_293_cast_fp16)[name = tensor("inputs_295_cast_fp16")]; + tensor hidden_states_391_axes_0 = const()[name = tensor("hidden_states_391_axes_0"), val = tensor([1])]; + tensor hidden_states_391_gamma_0_to_fp16 = const()[name = tensor("hidden_states_391_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3736010368)))]; + tensor hidden_states_391_beta_0_to_fp16 = const()[name = tensor("hidden_states_391_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3736012992)))]; + tensor var_44239_to_fp16 = const()[name = tensor("op_44239_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_391_cast_fp16 = layer_norm(axes = hidden_states_391_axes_0, beta = hidden_states_391_beta_0_to_fp16, epsilon = var_44239_to_fp16, gamma = hidden_states_391_gamma_0_to_fp16, x = inputs_295_cast_fp16)[name = tensor("hidden_states_391_cast_fp16")]; + tensor var_44254 = const()[name = tensor("op_44254"), val = tensor([1, 1])]; + tensor var_44256 = const()[name = tensor("op_44256"), val = tensor([1, 1])]; + tensor q_197_pad_type_0 = const()[name = tensor("q_197_pad_type_0"), val = tensor("custom")]; + tensor q_197_pad_0 = const()[name = tensor("q_197_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_5_attn1_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_5_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3736015616)))]; + tensor q_197_cast_fp16 = conv(dilations = var_44256, groups = var_30355, pad = q_197_pad_0, pad_type = q_197_pad_type_0, strides = var_44254, weight = up_blocks_0_attentions_1_transformer_blocks_5_attn1_to_q_weight_to_fp16, x = hidden_states_391_cast_fp16)[name = tensor("q_197_cast_fp16")]; + tensor var_44260 = const()[name = tensor("op_44260"), val = tensor([1, 1])]; + tensor var_44262 = const()[name = tensor("op_44262"), val = tensor([1, 1])]; + tensor k_393_pad_type_0 = const()[name = tensor("k_393_pad_type_0"), val = tensor("custom")]; + tensor k_393_pad_0 = const()[name = tensor("k_393_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_5_attn1_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_5_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3739292480)))]; + tensor k_393_cast_fp16 = conv(dilations = var_44262, groups = var_30355, pad = k_393_pad_0, pad_type = k_393_pad_type_0, strides = var_44260, weight = up_blocks_0_attentions_1_transformer_blocks_5_attn1_to_k_weight_to_fp16, x = hidden_states_391_cast_fp16)[name = tensor("k_393_cast_fp16")]; + tensor var_44266 = const()[name = tensor("op_44266"), val = tensor([1, 1])]; + tensor var_44268 = const()[name = tensor("op_44268"), val = tensor([1, 1])]; + tensor v_197_pad_type_0 = const()[name = tensor("v_197_pad_type_0"), val = tensor("custom")]; + tensor v_197_pad_0 = const()[name = tensor("v_197_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_5_attn1_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_5_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3742569344)))]; + tensor v_197_cast_fp16 = conv(dilations = var_44268, groups = var_30355, pad = v_197_pad_0, pad_type = v_197_pad_type_0, strides = var_44266, weight = up_blocks_0_attentions_1_transformer_blocks_5_attn1_to_v_weight_to_fp16, x = hidden_states_391_cast_fp16)[name = tensor("v_197_cast_fp16")]; + tensor var_44272_begin_0 = const()[name = tensor("op_44272_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_44272_end_0 = const()[name = tensor("op_44272_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_44272_end_mask_0 = const()[name = tensor("op_44272_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44272_cast_fp16 = slice_by_index(begin = var_44272_begin_0, end = var_44272_end_0, end_mask = var_44272_end_mask_0, x = q_197_cast_fp16)[name = tensor("op_44272_cast_fp16")]; + tensor var_44276_begin_0 = const()[name = tensor("op_44276_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_44276_end_0 = const()[name = tensor("op_44276_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_44276_end_mask_0 = const()[name = tensor("op_44276_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44276_cast_fp16 = slice_by_index(begin = var_44276_begin_0, end = var_44276_end_0, end_mask = var_44276_end_mask_0, x = q_197_cast_fp16)[name = tensor("op_44276_cast_fp16")]; + tensor var_44280_begin_0 = const()[name = tensor("op_44280_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_44280_end_0 = const()[name = tensor("op_44280_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_44280_end_mask_0 = const()[name = tensor("op_44280_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44280_cast_fp16 = slice_by_index(begin = var_44280_begin_0, end = var_44280_end_0, end_mask = var_44280_end_mask_0, x = q_197_cast_fp16)[name = tensor("op_44280_cast_fp16")]; + tensor var_44284_begin_0 = const()[name = tensor("op_44284_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_44284_end_0 = const()[name = tensor("op_44284_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_44284_end_mask_0 = const()[name = tensor("op_44284_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44284_cast_fp16 = slice_by_index(begin = var_44284_begin_0, end = var_44284_end_0, end_mask = var_44284_end_mask_0, x = q_197_cast_fp16)[name = tensor("op_44284_cast_fp16")]; + tensor var_44288_begin_0 = const()[name = tensor("op_44288_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_44288_end_0 = const()[name = tensor("op_44288_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_44288_end_mask_0 = const()[name = tensor("op_44288_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44288_cast_fp16 = slice_by_index(begin = var_44288_begin_0, end = var_44288_end_0, end_mask = var_44288_end_mask_0, x = q_197_cast_fp16)[name = tensor("op_44288_cast_fp16")]; + tensor var_44292_begin_0 = const()[name = tensor("op_44292_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_44292_end_0 = const()[name = tensor("op_44292_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_44292_end_mask_0 = const()[name = tensor("op_44292_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44292_cast_fp16 = slice_by_index(begin = var_44292_begin_0, end = var_44292_end_0, end_mask = var_44292_end_mask_0, x = q_197_cast_fp16)[name = tensor("op_44292_cast_fp16")]; + tensor var_44296_begin_0 = const()[name = tensor("op_44296_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_44296_end_0 = const()[name = tensor("op_44296_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_44296_end_mask_0 = const()[name = tensor("op_44296_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44296_cast_fp16 = slice_by_index(begin = var_44296_begin_0, end = var_44296_end_0, end_mask = var_44296_end_mask_0, x = q_197_cast_fp16)[name = tensor("op_44296_cast_fp16")]; + tensor var_44300_begin_0 = const()[name = tensor("op_44300_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_44300_end_0 = const()[name = tensor("op_44300_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_44300_end_mask_0 = const()[name = tensor("op_44300_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44300_cast_fp16 = slice_by_index(begin = var_44300_begin_0, end = var_44300_end_0, end_mask = var_44300_end_mask_0, x = q_197_cast_fp16)[name = tensor("op_44300_cast_fp16")]; + tensor var_44304_begin_0 = const()[name = tensor("op_44304_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_44304_end_0 = const()[name = tensor("op_44304_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_44304_end_mask_0 = const()[name = tensor("op_44304_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44304_cast_fp16 = slice_by_index(begin = var_44304_begin_0, end = var_44304_end_0, end_mask = var_44304_end_mask_0, x = q_197_cast_fp16)[name = tensor("op_44304_cast_fp16")]; + tensor var_44308_begin_0 = const()[name = tensor("op_44308_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_44308_end_0 = const()[name = tensor("op_44308_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_44308_end_mask_0 = const()[name = tensor("op_44308_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44308_cast_fp16 = slice_by_index(begin = var_44308_begin_0, end = var_44308_end_0, end_mask = var_44308_end_mask_0, x = q_197_cast_fp16)[name = tensor("op_44308_cast_fp16")]; + tensor var_44312_begin_0 = const()[name = tensor("op_44312_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_44312_end_0 = const()[name = tensor("op_44312_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_44312_end_mask_0 = const()[name = tensor("op_44312_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44312_cast_fp16 = slice_by_index(begin = var_44312_begin_0, end = var_44312_end_0, end_mask = var_44312_end_mask_0, x = q_197_cast_fp16)[name = tensor("op_44312_cast_fp16")]; + tensor var_44316_begin_0 = const()[name = tensor("op_44316_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_44316_end_0 = const()[name = tensor("op_44316_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_44316_end_mask_0 = const()[name = tensor("op_44316_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44316_cast_fp16 = slice_by_index(begin = var_44316_begin_0, end = var_44316_end_0, end_mask = var_44316_end_mask_0, x = q_197_cast_fp16)[name = tensor("op_44316_cast_fp16")]; + tensor var_44320_begin_0 = const()[name = tensor("op_44320_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_44320_end_0 = const()[name = tensor("op_44320_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_44320_end_mask_0 = const()[name = tensor("op_44320_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44320_cast_fp16 = slice_by_index(begin = var_44320_begin_0, end = var_44320_end_0, end_mask = var_44320_end_mask_0, x = q_197_cast_fp16)[name = tensor("op_44320_cast_fp16")]; + tensor var_44324_begin_0 = const()[name = tensor("op_44324_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_44324_end_0 = const()[name = tensor("op_44324_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_44324_end_mask_0 = const()[name = tensor("op_44324_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44324_cast_fp16 = slice_by_index(begin = var_44324_begin_0, end = var_44324_end_0, end_mask = var_44324_end_mask_0, x = q_197_cast_fp16)[name = tensor("op_44324_cast_fp16")]; + tensor var_44328_begin_0 = const()[name = tensor("op_44328_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_44328_end_0 = const()[name = tensor("op_44328_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_44328_end_mask_0 = const()[name = tensor("op_44328_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44328_cast_fp16 = slice_by_index(begin = var_44328_begin_0, end = var_44328_end_0, end_mask = var_44328_end_mask_0, x = q_197_cast_fp16)[name = tensor("op_44328_cast_fp16")]; + tensor var_44332_begin_0 = const()[name = tensor("op_44332_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_44332_end_0 = const()[name = tensor("op_44332_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_44332_end_mask_0 = const()[name = tensor("op_44332_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44332_cast_fp16 = slice_by_index(begin = var_44332_begin_0, end = var_44332_end_0, end_mask = var_44332_end_mask_0, x = q_197_cast_fp16)[name = tensor("op_44332_cast_fp16")]; + tensor var_44336_begin_0 = const()[name = tensor("op_44336_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_44336_end_0 = const()[name = tensor("op_44336_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_44336_end_mask_0 = const()[name = tensor("op_44336_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44336_cast_fp16 = slice_by_index(begin = var_44336_begin_0, end = var_44336_end_0, end_mask = var_44336_end_mask_0, x = q_197_cast_fp16)[name = tensor("op_44336_cast_fp16")]; + tensor var_44340_begin_0 = const()[name = tensor("op_44340_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_44340_end_0 = const()[name = tensor("op_44340_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_44340_end_mask_0 = const()[name = tensor("op_44340_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44340_cast_fp16 = slice_by_index(begin = var_44340_begin_0, end = var_44340_end_0, end_mask = var_44340_end_mask_0, x = q_197_cast_fp16)[name = tensor("op_44340_cast_fp16")]; + tensor var_44344_begin_0 = const()[name = tensor("op_44344_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_44344_end_0 = const()[name = tensor("op_44344_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_44344_end_mask_0 = const()[name = tensor("op_44344_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44344_cast_fp16 = slice_by_index(begin = var_44344_begin_0, end = var_44344_end_0, end_mask = var_44344_end_mask_0, x = q_197_cast_fp16)[name = tensor("op_44344_cast_fp16")]; + tensor var_44348_begin_0 = const()[name = tensor("op_44348_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_44348_end_0 = const()[name = tensor("op_44348_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_44348_end_mask_0 = const()[name = tensor("op_44348_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44348_cast_fp16 = slice_by_index(begin = var_44348_begin_0, end = var_44348_end_0, end_mask = var_44348_end_mask_0, x = q_197_cast_fp16)[name = tensor("op_44348_cast_fp16")]; + tensor k_395_perm_0 = const()[name = tensor("k_395_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_44355_begin_0 = const()[name = tensor("op_44355_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_44355_end_0 = const()[name = tensor("op_44355_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_44355_end_mask_0 = const()[name = tensor("op_44355_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_395_cast_fp16 = transpose(perm = k_395_perm_0, x = k_393_cast_fp16)[name = tensor("transpose_41")]; + tensor var_44355_cast_fp16 = slice_by_index(begin = var_44355_begin_0, end = var_44355_end_0, end_mask = var_44355_end_mask_0, x = k_395_cast_fp16)[name = tensor("op_44355_cast_fp16")]; + tensor var_44359_begin_0 = const()[name = tensor("op_44359_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_44359_end_0 = const()[name = tensor("op_44359_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_44359_end_mask_0 = const()[name = tensor("op_44359_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44359_cast_fp16 = slice_by_index(begin = var_44359_begin_0, end = var_44359_end_0, end_mask = var_44359_end_mask_0, x = k_395_cast_fp16)[name = tensor("op_44359_cast_fp16")]; + tensor var_44363_begin_0 = const()[name = tensor("op_44363_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_44363_end_0 = const()[name = tensor("op_44363_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_44363_end_mask_0 = const()[name = tensor("op_44363_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44363_cast_fp16 = slice_by_index(begin = var_44363_begin_0, end = var_44363_end_0, end_mask = var_44363_end_mask_0, x = k_395_cast_fp16)[name = tensor("op_44363_cast_fp16")]; + tensor var_44367_begin_0 = const()[name = tensor("op_44367_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_44367_end_0 = const()[name = tensor("op_44367_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_44367_end_mask_0 = const()[name = tensor("op_44367_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44367_cast_fp16 = slice_by_index(begin = var_44367_begin_0, end = var_44367_end_0, end_mask = var_44367_end_mask_0, x = k_395_cast_fp16)[name = tensor("op_44367_cast_fp16")]; + tensor var_44371_begin_0 = const()[name = tensor("op_44371_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_44371_end_0 = const()[name = tensor("op_44371_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_44371_end_mask_0 = const()[name = tensor("op_44371_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44371_cast_fp16 = slice_by_index(begin = var_44371_begin_0, end = var_44371_end_0, end_mask = var_44371_end_mask_0, x = k_395_cast_fp16)[name = tensor("op_44371_cast_fp16")]; + tensor var_44375_begin_0 = const()[name = tensor("op_44375_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_44375_end_0 = const()[name = tensor("op_44375_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_44375_end_mask_0 = const()[name = tensor("op_44375_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44375_cast_fp16 = slice_by_index(begin = var_44375_begin_0, end = var_44375_end_0, end_mask = var_44375_end_mask_0, x = k_395_cast_fp16)[name = tensor("op_44375_cast_fp16")]; + tensor var_44379_begin_0 = const()[name = tensor("op_44379_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_44379_end_0 = const()[name = tensor("op_44379_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_44379_end_mask_0 = const()[name = tensor("op_44379_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44379_cast_fp16 = slice_by_index(begin = var_44379_begin_0, end = var_44379_end_0, end_mask = var_44379_end_mask_0, x = k_395_cast_fp16)[name = tensor("op_44379_cast_fp16")]; + tensor var_44383_begin_0 = const()[name = tensor("op_44383_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_44383_end_0 = const()[name = tensor("op_44383_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_44383_end_mask_0 = const()[name = tensor("op_44383_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44383_cast_fp16 = slice_by_index(begin = var_44383_begin_0, end = var_44383_end_0, end_mask = var_44383_end_mask_0, x = k_395_cast_fp16)[name = tensor("op_44383_cast_fp16")]; + tensor var_44387_begin_0 = const()[name = tensor("op_44387_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_44387_end_0 = const()[name = tensor("op_44387_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_44387_end_mask_0 = const()[name = tensor("op_44387_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44387_cast_fp16 = slice_by_index(begin = var_44387_begin_0, end = var_44387_end_0, end_mask = var_44387_end_mask_0, x = k_395_cast_fp16)[name = tensor("op_44387_cast_fp16")]; + tensor var_44391_begin_0 = const()[name = tensor("op_44391_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_44391_end_0 = const()[name = tensor("op_44391_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_44391_end_mask_0 = const()[name = tensor("op_44391_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44391_cast_fp16 = slice_by_index(begin = var_44391_begin_0, end = var_44391_end_0, end_mask = var_44391_end_mask_0, x = k_395_cast_fp16)[name = tensor("op_44391_cast_fp16")]; + tensor var_44395_begin_0 = const()[name = tensor("op_44395_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_44395_end_0 = const()[name = tensor("op_44395_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_44395_end_mask_0 = const()[name = tensor("op_44395_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44395_cast_fp16 = slice_by_index(begin = var_44395_begin_0, end = var_44395_end_0, end_mask = var_44395_end_mask_0, x = k_395_cast_fp16)[name = tensor("op_44395_cast_fp16")]; + tensor var_44399_begin_0 = const()[name = tensor("op_44399_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_44399_end_0 = const()[name = tensor("op_44399_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_44399_end_mask_0 = const()[name = tensor("op_44399_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44399_cast_fp16 = slice_by_index(begin = var_44399_begin_0, end = var_44399_end_0, end_mask = var_44399_end_mask_0, x = k_395_cast_fp16)[name = tensor("op_44399_cast_fp16")]; + tensor var_44403_begin_0 = const()[name = tensor("op_44403_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_44403_end_0 = const()[name = tensor("op_44403_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_44403_end_mask_0 = const()[name = tensor("op_44403_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44403_cast_fp16 = slice_by_index(begin = var_44403_begin_0, end = var_44403_end_0, end_mask = var_44403_end_mask_0, x = k_395_cast_fp16)[name = tensor("op_44403_cast_fp16")]; + tensor var_44407_begin_0 = const()[name = tensor("op_44407_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_44407_end_0 = const()[name = tensor("op_44407_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_44407_end_mask_0 = const()[name = tensor("op_44407_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44407_cast_fp16 = slice_by_index(begin = var_44407_begin_0, end = var_44407_end_0, end_mask = var_44407_end_mask_0, x = k_395_cast_fp16)[name = tensor("op_44407_cast_fp16")]; + tensor var_44411_begin_0 = const()[name = tensor("op_44411_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_44411_end_0 = const()[name = tensor("op_44411_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_44411_end_mask_0 = const()[name = tensor("op_44411_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44411_cast_fp16 = slice_by_index(begin = var_44411_begin_0, end = var_44411_end_0, end_mask = var_44411_end_mask_0, x = k_395_cast_fp16)[name = tensor("op_44411_cast_fp16")]; + tensor var_44415_begin_0 = const()[name = tensor("op_44415_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_44415_end_0 = const()[name = tensor("op_44415_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_44415_end_mask_0 = const()[name = tensor("op_44415_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44415_cast_fp16 = slice_by_index(begin = var_44415_begin_0, end = var_44415_end_0, end_mask = var_44415_end_mask_0, x = k_395_cast_fp16)[name = tensor("op_44415_cast_fp16")]; + tensor var_44419_begin_0 = const()[name = tensor("op_44419_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_44419_end_0 = const()[name = tensor("op_44419_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_44419_end_mask_0 = const()[name = tensor("op_44419_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44419_cast_fp16 = slice_by_index(begin = var_44419_begin_0, end = var_44419_end_0, end_mask = var_44419_end_mask_0, x = k_395_cast_fp16)[name = tensor("op_44419_cast_fp16")]; + tensor var_44423_begin_0 = const()[name = tensor("op_44423_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_44423_end_0 = const()[name = tensor("op_44423_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_44423_end_mask_0 = const()[name = tensor("op_44423_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44423_cast_fp16 = slice_by_index(begin = var_44423_begin_0, end = var_44423_end_0, end_mask = var_44423_end_mask_0, x = k_395_cast_fp16)[name = tensor("op_44423_cast_fp16")]; + tensor var_44427_begin_0 = const()[name = tensor("op_44427_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_44427_end_0 = const()[name = tensor("op_44427_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_44427_end_mask_0 = const()[name = tensor("op_44427_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44427_cast_fp16 = slice_by_index(begin = var_44427_begin_0, end = var_44427_end_0, end_mask = var_44427_end_mask_0, x = k_395_cast_fp16)[name = tensor("op_44427_cast_fp16")]; + tensor var_44431_begin_0 = const()[name = tensor("op_44431_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_44431_end_0 = const()[name = tensor("op_44431_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_44431_end_mask_0 = const()[name = tensor("op_44431_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44431_cast_fp16 = slice_by_index(begin = var_44431_begin_0, end = var_44431_end_0, end_mask = var_44431_end_mask_0, x = k_395_cast_fp16)[name = tensor("op_44431_cast_fp16")]; + tensor var_44433_begin_0 = const()[name = tensor("op_44433_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_44433_end_0 = const()[name = tensor("op_44433_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_44433_end_mask_0 = const()[name = tensor("op_44433_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44433_cast_fp16 = slice_by_index(begin = var_44433_begin_0, end = var_44433_end_0, end_mask = var_44433_end_mask_0, x = v_197_cast_fp16)[name = tensor("op_44433_cast_fp16")]; + tensor var_44437_begin_0 = const()[name = tensor("op_44437_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_44437_end_0 = const()[name = tensor("op_44437_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_44437_end_mask_0 = const()[name = tensor("op_44437_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44437_cast_fp16 = slice_by_index(begin = var_44437_begin_0, end = var_44437_end_0, end_mask = var_44437_end_mask_0, x = v_197_cast_fp16)[name = tensor("op_44437_cast_fp16")]; + tensor var_44441_begin_0 = const()[name = tensor("op_44441_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_44441_end_0 = const()[name = tensor("op_44441_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_44441_end_mask_0 = const()[name = tensor("op_44441_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44441_cast_fp16 = slice_by_index(begin = var_44441_begin_0, end = var_44441_end_0, end_mask = var_44441_end_mask_0, x = v_197_cast_fp16)[name = tensor("op_44441_cast_fp16")]; + tensor var_44445_begin_0 = const()[name = tensor("op_44445_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_44445_end_0 = const()[name = tensor("op_44445_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_44445_end_mask_0 = const()[name = tensor("op_44445_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44445_cast_fp16 = slice_by_index(begin = var_44445_begin_0, end = var_44445_end_0, end_mask = var_44445_end_mask_0, x = v_197_cast_fp16)[name = tensor("op_44445_cast_fp16")]; + tensor var_44449_begin_0 = const()[name = tensor("op_44449_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_44449_end_0 = const()[name = tensor("op_44449_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_44449_end_mask_0 = const()[name = tensor("op_44449_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44449_cast_fp16 = slice_by_index(begin = var_44449_begin_0, end = var_44449_end_0, end_mask = var_44449_end_mask_0, x = v_197_cast_fp16)[name = tensor("op_44449_cast_fp16")]; + tensor var_44453_begin_0 = const()[name = tensor("op_44453_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_44453_end_0 = const()[name = tensor("op_44453_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_44453_end_mask_0 = const()[name = tensor("op_44453_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44453_cast_fp16 = slice_by_index(begin = var_44453_begin_0, end = var_44453_end_0, end_mask = var_44453_end_mask_0, x = v_197_cast_fp16)[name = tensor("op_44453_cast_fp16")]; + tensor var_44457_begin_0 = const()[name = tensor("op_44457_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_44457_end_0 = const()[name = tensor("op_44457_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_44457_end_mask_0 = const()[name = tensor("op_44457_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44457_cast_fp16 = slice_by_index(begin = var_44457_begin_0, end = var_44457_end_0, end_mask = var_44457_end_mask_0, x = v_197_cast_fp16)[name = tensor("op_44457_cast_fp16")]; + tensor var_44461_begin_0 = const()[name = tensor("op_44461_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_44461_end_0 = const()[name = tensor("op_44461_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_44461_end_mask_0 = const()[name = tensor("op_44461_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44461_cast_fp16 = slice_by_index(begin = var_44461_begin_0, end = var_44461_end_0, end_mask = var_44461_end_mask_0, x = v_197_cast_fp16)[name = tensor("op_44461_cast_fp16")]; + tensor var_44465_begin_0 = const()[name = tensor("op_44465_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_44465_end_0 = const()[name = tensor("op_44465_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_44465_end_mask_0 = const()[name = tensor("op_44465_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44465_cast_fp16 = slice_by_index(begin = var_44465_begin_0, end = var_44465_end_0, end_mask = var_44465_end_mask_0, x = v_197_cast_fp16)[name = tensor("op_44465_cast_fp16")]; + tensor var_44469_begin_0 = const()[name = tensor("op_44469_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_44469_end_0 = const()[name = tensor("op_44469_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_44469_end_mask_0 = const()[name = tensor("op_44469_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44469_cast_fp16 = slice_by_index(begin = var_44469_begin_0, end = var_44469_end_0, end_mask = var_44469_end_mask_0, x = v_197_cast_fp16)[name = tensor("op_44469_cast_fp16")]; + tensor var_44473_begin_0 = const()[name = tensor("op_44473_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_44473_end_0 = const()[name = tensor("op_44473_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_44473_end_mask_0 = const()[name = tensor("op_44473_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44473_cast_fp16 = slice_by_index(begin = var_44473_begin_0, end = var_44473_end_0, end_mask = var_44473_end_mask_0, x = v_197_cast_fp16)[name = tensor("op_44473_cast_fp16")]; + tensor var_44477_begin_0 = const()[name = tensor("op_44477_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_44477_end_0 = const()[name = tensor("op_44477_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_44477_end_mask_0 = const()[name = tensor("op_44477_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44477_cast_fp16 = slice_by_index(begin = var_44477_begin_0, end = var_44477_end_0, end_mask = var_44477_end_mask_0, x = v_197_cast_fp16)[name = tensor("op_44477_cast_fp16")]; + tensor var_44481_begin_0 = const()[name = tensor("op_44481_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_44481_end_0 = const()[name = tensor("op_44481_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_44481_end_mask_0 = const()[name = tensor("op_44481_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44481_cast_fp16 = slice_by_index(begin = var_44481_begin_0, end = var_44481_end_0, end_mask = var_44481_end_mask_0, x = v_197_cast_fp16)[name = tensor("op_44481_cast_fp16")]; + tensor var_44485_begin_0 = const()[name = tensor("op_44485_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_44485_end_0 = const()[name = tensor("op_44485_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_44485_end_mask_0 = const()[name = tensor("op_44485_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44485_cast_fp16 = slice_by_index(begin = var_44485_begin_0, end = var_44485_end_0, end_mask = var_44485_end_mask_0, x = v_197_cast_fp16)[name = tensor("op_44485_cast_fp16")]; + tensor var_44489_begin_0 = const()[name = tensor("op_44489_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_44489_end_0 = const()[name = tensor("op_44489_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_44489_end_mask_0 = const()[name = tensor("op_44489_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44489_cast_fp16 = slice_by_index(begin = var_44489_begin_0, end = var_44489_end_0, end_mask = var_44489_end_mask_0, x = v_197_cast_fp16)[name = tensor("op_44489_cast_fp16")]; + tensor var_44493_begin_0 = const()[name = tensor("op_44493_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_44493_end_0 = const()[name = tensor("op_44493_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_44493_end_mask_0 = const()[name = tensor("op_44493_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44493_cast_fp16 = slice_by_index(begin = var_44493_begin_0, end = var_44493_end_0, end_mask = var_44493_end_mask_0, x = v_197_cast_fp16)[name = tensor("op_44493_cast_fp16")]; + tensor var_44497_begin_0 = const()[name = tensor("op_44497_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_44497_end_0 = const()[name = tensor("op_44497_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_44497_end_mask_0 = const()[name = tensor("op_44497_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44497_cast_fp16 = slice_by_index(begin = var_44497_begin_0, end = var_44497_end_0, end_mask = var_44497_end_mask_0, x = v_197_cast_fp16)[name = tensor("op_44497_cast_fp16")]; + tensor var_44501_begin_0 = const()[name = tensor("op_44501_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_44501_end_0 = const()[name = tensor("op_44501_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_44501_end_mask_0 = const()[name = tensor("op_44501_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44501_cast_fp16 = slice_by_index(begin = var_44501_begin_0, end = var_44501_end_0, end_mask = var_44501_end_mask_0, x = v_197_cast_fp16)[name = tensor("op_44501_cast_fp16")]; + tensor var_44505_begin_0 = const()[name = tensor("op_44505_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_44505_end_0 = const()[name = tensor("op_44505_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_44505_end_mask_0 = const()[name = tensor("op_44505_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44505_cast_fp16 = slice_by_index(begin = var_44505_begin_0, end = var_44505_end_0, end_mask = var_44505_end_mask_0, x = v_197_cast_fp16)[name = tensor("op_44505_cast_fp16")]; + tensor var_44509_begin_0 = const()[name = tensor("op_44509_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_44509_end_0 = const()[name = tensor("op_44509_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_44509_end_mask_0 = const()[name = tensor("op_44509_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44509_cast_fp16 = slice_by_index(begin = var_44509_begin_0, end = var_44509_end_0, end_mask = var_44509_end_mask_0, x = v_197_cast_fp16)[name = tensor("op_44509_cast_fp16")]; + tensor var_44513_equation_0 = const()[name = tensor("op_44513_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_44513_cast_fp16 = einsum(equation = var_44513_equation_0, values = (var_44355_cast_fp16, var_44272_cast_fp16))[name = tensor("op_44513_cast_fp16")]; + tensor var_44514_to_fp16 = const()[name = tensor("op_44514_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3761_cast_fp16 = mul(x = var_44513_cast_fp16, y = var_44514_to_fp16)[name = tensor("aw_3761_cast_fp16")]; + tensor var_44517_equation_0 = const()[name = tensor("op_44517_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_44517_cast_fp16 = einsum(equation = var_44517_equation_0, values = (var_44359_cast_fp16, var_44276_cast_fp16))[name = tensor("op_44517_cast_fp16")]; + tensor var_44518_to_fp16 = const()[name = tensor("op_44518_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3763_cast_fp16 = mul(x = var_44517_cast_fp16, y = var_44518_to_fp16)[name = tensor("aw_3763_cast_fp16")]; + tensor var_44521_equation_0 = const()[name = tensor("op_44521_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_44521_cast_fp16 = einsum(equation = var_44521_equation_0, values = (var_44363_cast_fp16, var_44280_cast_fp16))[name = tensor("op_44521_cast_fp16")]; + tensor var_44522_to_fp16 = const()[name = tensor("op_44522_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3765_cast_fp16 = mul(x = var_44521_cast_fp16, y = var_44522_to_fp16)[name = tensor("aw_3765_cast_fp16")]; + tensor var_44525_equation_0 = const()[name = tensor("op_44525_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_44525_cast_fp16 = einsum(equation = var_44525_equation_0, values = (var_44367_cast_fp16, var_44284_cast_fp16))[name = tensor("op_44525_cast_fp16")]; + tensor var_44526_to_fp16 = const()[name = tensor("op_44526_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3767_cast_fp16 = mul(x = var_44525_cast_fp16, y = var_44526_to_fp16)[name = tensor("aw_3767_cast_fp16")]; + tensor var_44529_equation_0 = const()[name = tensor("op_44529_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_44529_cast_fp16 = einsum(equation = var_44529_equation_0, values = (var_44371_cast_fp16, var_44288_cast_fp16))[name = tensor("op_44529_cast_fp16")]; + tensor var_44530_to_fp16 = const()[name = tensor("op_44530_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3769_cast_fp16 = mul(x = var_44529_cast_fp16, y = var_44530_to_fp16)[name = tensor("aw_3769_cast_fp16")]; + tensor var_44533_equation_0 = const()[name = tensor("op_44533_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_44533_cast_fp16 = einsum(equation = var_44533_equation_0, values = (var_44375_cast_fp16, var_44292_cast_fp16))[name = tensor("op_44533_cast_fp16")]; + tensor var_44534_to_fp16 = const()[name = tensor("op_44534_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3771_cast_fp16 = mul(x = var_44533_cast_fp16, y = var_44534_to_fp16)[name = tensor("aw_3771_cast_fp16")]; + tensor var_44537_equation_0 = const()[name = tensor("op_44537_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_44537_cast_fp16 = einsum(equation = var_44537_equation_0, values = (var_44379_cast_fp16, var_44296_cast_fp16))[name = tensor("op_44537_cast_fp16")]; + tensor var_44538_to_fp16 = const()[name = tensor("op_44538_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3773_cast_fp16 = mul(x = var_44537_cast_fp16, y = var_44538_to_fp16)[name = tensor("aw_3773_cast_fp16")]; + tensor var_44541_equation_0 = const()[name = tensor("op_44541_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_44541_cast_fp16 = einsum(equation = var_44541_equation_0, values = (var_44383_cast_fp16, var_44300_cast_fp16))[name = tensor("op_44541_cast_fp16")]; + tensor var_44542_to_fp16 = const()[name = tensor("op_44542_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3775_cast_fp16 = mul(x = var_44541_cast_fp16, y = var_44542_to_fp16)[name = tensor("aw_3775_cast_fp16")]; + tensor var_44545_equation_0 = const()[name = tensor("op_44545_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_44545_cast_fp16 = einsum(equation = var_44545_equation_0, values = (var_44387_cast_fp16, var_44304_cast_fp16))[name = tensor("op_44545_cast_fp16")]; + tensor var_44546_to_fp16 = const()[name = tensor("op_44546_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3777_cast_fp16 = mul(x = var_44545_cast_fp16, y = var_44546_to_fp16)[name = tensor("aw_3777_cast_fp16")]; + tensor var_44549_equation_0 = const()[name = tensor("op_44549_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_44549_cast_fp16 = einsum(equation = var_44549_equation_0, values = (var_44391_cast_fp16, var_44308_cast_fp16))[name = tensor("op_44549_cast_fp16")]; + tensor var_44550_to_fp16 = const()[name = tensor("op_44550_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3779_cast_fp16 = mul(x = var_44549_cast_fp16, y = var_44550_to_fp16)[name = tensor("aw_3779_cast_fp16")]; + tensor var_44553_equation_0 = const()[name = tensor("op_44553_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_44553_cast_fp16 = einsum(equation = var_44553_equation_0, values = (var_44395_cast_fp16, var_44312_cast_fp16))[name = tensor("op_44553_cast_fp16")]; + tensor var_44554_to_fp16 = const()[name = tensor("op_44554_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3781_cast_fp16 = mul(x = var_44553_cast_fp16, y = var_44554_to_fp16)[name = tensor("aw_3781_cast_fp16")]; + tensor var_44557_equation_0 = const()[name = tensor("op_44557_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_44557_cast_fp16 = einsum(equation = var_44557_equation_0, values = (var_44399_cast_fp16, var_44316_cast_fp16))[name = tensor("op_44557_cast_fp16")]; + tensor var_44558_to_fp16 = const()[name = tensor("op_44558_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3783_cast_fp16 = mul(x = var_44557_cast_fp16, y = var_44558_to_fp16)[name = tensor("aw_3783_cast_fp16")]; + tensor var_44561_equation_0 = const()[name = tensor("op_44561_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_44561_cast_fp16 = einsum(equation = var_44561_equation_0, values = (var_44403_cast_fp16, var_44320_cast_fp16))[name = tensor("op_44561_cast_fp16")]; + tensor var_44562_to_fp16 = const()[name = tensor("op_44562_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3785_cast_fp16 = mul(x = var_44561_cast_fp16, y = var_44562_to_fp16)[name = tensor("aw_3785_cast_fp16")]; + tensor var_44565_equation_0 = const()[name = tensor("op_44565_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_44565_cast_fp16 = einsum(equation = var_44565_equation_0, values = (var_44407_cast_fp16, var_44324_cast_fp16))[name = tensor("op_44565_cast_fp16")]; + tensor var_44566_to_fp16 = const()[name = tensor("op_44566_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3787_cast_fp16 = mul(x = var_44565_cast_fp16, y = var_44566_to_fp16)[name = tensor("aw_3787_cast_fp16")]; + tensor var_44569_equation_0 = const()[name = tensor("op_44569_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_44569_cast_fp16 = einsum(equation = var_44569_equation_0, values = (var_44411_cast_fp16, var_44328_cast_fp16))[name = tensor("op_44569_cast_fp16")]; + tensor var_44570_to_fp16 = const()[name = tensor("op_44570_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3789_cast_fp16 = mul(x = var_44569_cast_fp16, y = var_44570_to_fp16)[name = tensor("aw_3789_cast_fp16")]; + tensor var_44573_equation_0 = const()[name = tensor("op_44573_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_44573_cast_fp16 = einsum(equation = var_44573_equation_0, values = (var_44415_cast_fp16, var_44332_cast_fp16))[name = tensor("op_44573_cast_fp16")]; + tensor var_44574_to_fp16 = const()[name = tensor("op_44574_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3791_cast_fp16 = mul(x = var_44573_cast_fp16, y = var_44574_to_fp16)[name = tensor("aw_3791_cast_fp16")]; + tensor var_44577_equation_0 = const()[name = tensor("op_44577_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_44577_cast_fp16 = einsum(equation = var_44577_equation_0, values = (var_44419_cast_fp16, var_44336_cast_fp16))[name = tensor("op_44577_cast_fp16")]; + tensor var_44578_to_fp16 = const()[name = tensor("op_44578_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3793_cast_fp16 = mul(x = var_44577_cast_fp16, y = var_44578_to_fp16)[name = tensor("aw_3793_cast_fp16")]; + tensor var_44581_equation_0 = const()[name = tensor("op_44581_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_44581_cast_fp16 = einsum(equation = var_44581_equation_0, values = (var_44423_cast_fp16, var_44340_cast_fp16))[name = tensor("op_44581_cast_fp16")]; + tensor var_44582_to_fp16 = const()[name = tensor("op_44582_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3795_cast_fp16 = mul(x = var_44581_cast_fp16, y = var_44582_to_fp16)[name = tensor("aw_3795_cast_fp16")]; + tensor var_44585_equation_0 = const()[name = tensor("op_44585_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_44585_cast_fp16 = einsum(equation = var_44585_equation_0, values = (var_44427_cast_fp16, var_44344_cast_fp16))[name = tensor("op_44585_cast_fp16")]; + tensor var_44586_to_fp16 = const()[name = tensor("op_44586_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3797_cast_fp16 = mul(x = var_44585_cast_fp16, y = var_44586_to_fp16)[name = tensor("aw_3797_cast_fp16")]; + tensor var_44589_equation_0 = const()[name = tensor("op_44589_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_44589_cast_fp16 = einsum(equation = var_44589_equation_0, values = (var_44431_cast_fp16, var_44348_cast_fp16))[name = tensor("op_44589_cast_fp16")]; + tensor var_44590_to_fp16 = const()[name = tensor("op_44590_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3799_cast_fp16 = mul(x = var_44589_cast_fp16, y = var_44590_to_fp16)[name = tensor("aw_3799_cast_fp16")]; + tensor var_44592_cast_fp16 = softmax(axis = var_30355, x = aw_3761_cast_fp16)[name = tensor("op_44592_cast_fp16")]; + tensor var_44593_cast_fp16 = softmax(axis = var_30355, x = aw_3763_cast_fp16)[name = tensor("op_44593_cast_fp16")]; + tensor var_44594_cast_fp16 = softmax(axis = var_30355, x = aw_3765_cast_fp16)[name = tensor("op_44594_cast_fp16")]; + tensor var_44595_cast_fp16 = softmax(axis = var_30355, x = aw_3767_cast_fp16)[name = tensor("op_44595_cast_fp16")]; + tensor var_44596_cast_fp16 = softmax(axis = var_30355, x = aw_3769_cast_fp16)[name = tensor("op_44596_cast_fp16")]; + tensor var_44597_cast_fp16 = softmax(axis = var_30355, x = aw_3771_cast_fp16)[name = tensor("op_44597_cast_fp16")]; + tensor var_44598_cast_fp16 = softmax(axis = var_30355, x = aw_3773_cast_fp16)[name = tensor("op_44598_cast_fp16")]; + tensor var_44599_cast_fp16 = softmax(axis = var_30355, x = aw_3775_cast_fp16)[name = tensor("op_44599_cast_fp16")]; + tensor var_44600_cast_fp16 = softmax(axis = var_30355, x = aw_3777_cast_fp16)[name = tensor("op_44600_cast_fp16")]; + tensor var_44601_cast_fp16 = softmax(axis = var_30355, x = aw_3779_cast_fp16)[name = tensor("op_44601_cast_fp16")]; + tensor var_44602_cast_fp16 = softmax(axis = var_30355, x = aw_3781_cast_fp16)[name = tensor("op_44602_cast_fp16")]; + tensor var_44603_cast_fp16 = softmax(axis = var_30355, x = aw_3783_cast_fp16)[name = tensor("op_44603_cast_fp16")]; + tensor var_44604_cast_fp16 = softmax(axis = var_30355, x = aw_3785_cast_fp16)[name = tensor("op_44604_cast_fp16")]; + tensor var_44605_cast_fp16 = softmax(axis = var_30355, x = aw_3787_cast_fp16)[name = tensor("op_44605_cast_fp16")]; + tensor var_44606_cast_fp16 = softmax(axis = var_30355, x = aw_3789_cast_fp16)[name = tensor("op_44606_cast_fp16")]; + tensor var_44607_cast_fp16 = softmax(axis = var_30355, x = aw_3791_cast_fp16)[name = tensor("op_44607_cast_fp16")]; + tensor var_44608_cast_fp16 = softmax(axis = var_30355, x = aw_3793_cast_fp16)[name = tensor("op_44608_cast_fp16")]; + tensor var_44609_cast_fp16 = softmax(axis = var_30355, x = aw_3795_cast_fp16)[name = tensor("op_44609_cast_fp16")]; + tensor var_44610_cast_fp16 = softmax(axis = var_30355, x = aw_3797_cast_fp16)[name = tensor("op_44610_cast_fp16")]; + tensor var_44611_cast_fp16 = softmax(axis = var_30355, x = aw_3799_cast_fp16)[name = tensor("op_44611_cast_fp16")]; + tensor var_44613_equation_0 = const()[name = tensor("op_44613_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44613_cast_fp16 = einsum(equation = var_44613_equation_0, values = (var_44433_cast_fp16, var_44592_cast_fp16))[name = tensor("op_44613_cast_fp16")]; + tensor var_44615_equation_0 = const()[name = tensor("op_44615_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44615_cast_fp16 = einsum(equation = var_44615_equation_0, values = (var_44437_cast_fp16, var_44593_cast_fp16))[name = tensor("op_44615_cast_fp16")]; + tensor var_44617_equation_0 = const()[name = tensor("op_44617_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44617_cast_fp16 = einsum(equation = var_44617_equation_0, values = (var_44441_cast_fp16, var_44594_cast_fp16))[name = tensor("op_44617_cast_fp16")]; + tensor var_44619_equation_0 = const()[name = tensor("op_44619_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44619_cast_fp16 = einsum(equation = var_44619_equation_0, values = (var_44445_cast_fp16, var_44595_cast_fp16))[name = tensor("op_44619_cast_fp16")]; + tensor var_44621_equation_0 = const()[name = tensor("op_44621_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44621_cast_fp16 = einsum(equation = var_44621_equation_0, values = (var_44449_cast_fp16, var_44596_cast_fp16))[name = tensor("op_44621_cast_fp16")]; + tensor var_44623_equation_0 = const()[name = tensor("op_44623_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44623_cast_fp16 = einsum(equation = var_44623_equation_0, values = (var_44453_cast_fp16, var_44597_cast_fp16))[name = tensor("op_44623_cast_fp16")]; + tensor var_44625_equation_0 = const()[name = tensor("op_44625_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44625_cast_fp16 = einsum(equation = var_44625_equation_0, values = (var_44457_cast_fp16, var_44598_cast_fp16))[name = tensor("op_44625_cast_fp16")]; + tensor var_44627_equation_0 = const()[name = tensor("op_44627_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44627_cast_fp16 = einsum(equation = var_44627_equation_0, values = (var_44461_cast_fp16, var_44599_cast_fp16))[name = tensor("op_44627_cast_fp16")]; + tensor var_44629_equation_0 = const()[name = tensor("op_44629_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44629_cast_fp16 = einsum(equation = var_44629_equation_0, values = (var_44465_cast_fp16, var_44600_cast_fp16))[name = tensor("op_44629_cast_fp16")]; + tensor var_44631_equation_0 = const()[name = tensor("op_44631_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44631_cast_fp16 = einsum(equation = var_44631_equation_0, values = (var_44469_cast_fp16, var_44601_cast_fp16))[name = tensor("op_44631_cast_fp16")]; + tensor var_44633_equation_0 = const()[name = tensor("op_44633_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44633_cast_fp16 = einsum(equation = var_44633_equation_0, values = (var_44473_cast_fp16, var_44602_cast_fp16))[name = tensor("op_44633_cast_fp16")]; + tensor var_44635_equation_0 = const()[name = tensor("op_44635_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44635_cast_fp16 = einsum(equation = var_44635_equation_0, values = (var_44477_cast_fp16, var_44603_cast_fp16))[name = tensor("op_44635_cast_fp16")]; + tensor var_44637_equation_0 = const()[name = tensor("op_44637_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44637_cast_fp16 = einsum(equation = var_44637_equation_0, values = (var_44481_cast_fp16, var_44604_cast_fp16))[name = tensor("op_44637_cast_fp16")]; + tensor var_44639_equation_0 = const()[name = tensor("op_44639_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44639_cast_fp16 = einsum(equation = var_44639_equation_0, values = (var_44485_cast_fp16, var_44605_cast_fp16))[name = tensor("op_44639_cast_fp16")]; + tensor var_44641_equation_0 = const()[name = tensor("op_44641_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44641_cast_fp16 = einsum(equation = var_44641_equation_0, values = (var_44489_cast_fp16, var_44606_cast_fp16))[name = tensor("op_44641_cast_fp16")]; + tensor var_44643_equation_0 = const()[name = tensor("op_44643_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44643_cast_fp16 = einsum(equation = var_44643_equation_0, values = (var_44493_cast_fp16, var_44607_cast_fp16))[name = tensor("op_44643_cast_fp16")]; + tensor var_44645_equation_0 = const()[name = tensor("op_44645_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44645_cast_fp16 = einsum(equation = var_44645_equation_0, values = (var_44497_cast_fp16, var_44608_cast_fp16))[name = tensor("op_44645_cast_fp16")]; + tensor var_44647_equation_0 = const()[name = tensor("op_44647_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44647_cast_fp16 = einsum(equation = var_44647_equation_0, values = (var_44501_cast_fp16, var_44609_cast_fp16))[name = tensor("op_44647_cast_fp16")]; + tensor var_44649_equation_0 = const()[name = tensor("op_44649_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44649_cast_fp16 = einsum(equation = var_44649_equation_0, values = (var_44505_cast_fp16, var_44610_cast_fp16))[name = tensor("op_44649_cast_fp16")]; + tensor var_44651_equation_0 = const()[name = tensor("op_44651_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44651_cast_fp16 = einsum(equation = var_44651_equation_0, values = (var_44509_cast_fp16, var_44611_cast_fp16))[name = tensor("op_44651_cast_fp16")]; + tensor input_577_interleave_0 = const()[name = tensor("input_577_interleave_0"), val = tensor(false)]; + tensor input_577_cast_fp16 = concat(axis = var_30355, interleave = input_577_interleave_0, values = (var_44613_cast_fp16, var_44615_cast_fp16, var_44617_cast_fp16, var_44619_cast_fp16, var_44621_cast_fp16, var_44623_cast_fp16, var_44625_cast_fp16, var_44627_cast_fp16, var_44629_cast_fp16, var_44631_cast_fp16, var_44633_cast_fp16, var_44635_cast_fp16, var_44637_cast_fp16, var_44639_cast_fp16, var_44641_cast_fp16, var_44643_cast_fp16, var_44645_cast_fp16, var_44647_cast_fp16, var_44649_cast_fp16, var_44651_cast_fp16))[name = tensor("input_577_cast_fp16")]; + tensor var_44657 = const()[name = tensor("op_44657"), val = tensor([1, 1])]; + tensor var_44659 = const()[name = tensor("op_44659"), val = tensor([1, 1])]; + tensor var_44661_pad_type_0 = const()[name = tensor("op_44661_pad_type_0"), val = tensor("custom")]; + tensor var_44661_pad_0 = const()[name = tensor("op_44661_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_5_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_5_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3745846208)))]; + tensor up_blocks_0_attentions_1_transformer_blocks_5_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_5_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3749123072)))]; + tensor var_44661_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_5_attn1_to_out_0_bias_to_fp16, dilations = var_44659, groups = var_30355, pad = var_44661_pad_0, pad_type = var_44661_pad_type_0, strides = var_44657, weight = up_blocks_0_attentions_1_transformer_blocks_5_attn1_to_out_0_weight_to_fp16, x = input_577_cast_fp16)[name = tensor("op_44661_cast_fp16")]; + tensor inputs_297_cast_fp16 = add(x = var_44661_cast_fp16, y = inputs_295_cast_fp16)[name = tensor("inputs_297_cast_fp16")]; + tensor hidden_states_393_axes_0 = const()[name = tensor("hidden_states_393_axes_0"), val = tensor([1])]; + tensor hidden_states_393_gamma_0_to_fp16 = const()[name = tensor("hidden_states_393_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3749125696)))]; + tensor hidden_states_393_beta_0_to_fp16 = const()[name = tensor("hidden_states_393_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3749128320)))]; + tensor var_44671_to_fp16 = const()[name = tensor("op_44671_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_393_cast_fp16 = layer_norm(axes = hidden_states_393_axes_0, beta = hidden_states_393_beta_0_to_fp16, epsilon = var_44671_to_fp16, gamma = hidden_states_393_gamma_0_to_fp16, x = inputs_297_cast_fp16)[name = tensor("hidden_states_393_cast_fp16")]; + tensor var_44686 = const()[name = tensor("op_44686"), val = tensor([1, 1])]; + tensor var_44688 = const()[name = tensor("op_44688"), val = tensor([1, 1])]; + tensor q_199_pad_type_0 = const()[name = tensor("q_199_pad_type_0"), val = tensor("custom")]; + tensor q_199_pad_0 = const()[name = tensor("q_199_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_5_attn2_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_5_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3749130944)))]; + tensor q_199_cast_fp16 = conv(dilations = var_44688, groups = var_30355, pad = q_199_pad_0, pad_type = q_199_pad_type_0, strides = var_44686, weight = up_blocks_0_attentions_1_transformer_blocks_5_attn2_to_q_weight_to_fp16, x = hidden_states_393_cast_fp16)[name = tensor("q_199_cast_fp16")]; + tensor var_44692 = const()[name = tensor("op_44692"), val = tensor([1, 1])]; + tensor var_44694 = const()[name = tensor("op_44694"), val = tensor([1, 1])]; + tensor k_397_pad_type_0 = const()[name = tensor("k_397_pad_type_0"), val = tensor("custom")]; + tensor k_397_pad_0 = const()[name = tensor("k_397_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_5_attn2_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_5_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3752407808)))]; + tensor k_397_cast_fp16 = conv(dilations = var_44694, groups = var_30355, pad = k_397_pad_0, pad_type = k_397_pad_type_0, strides = var_44692, weight = up_blocks_0_attentions_1_transformer_blocks_5_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_397_cast_fp16")]; + tensor var_44698 = const()[name = tensor("op_44698"), val = tensor([1, 1])]; + tensor var_44700 = const()[name = tensor("op_44700"), val = tensor([1, 1])]; + tensor v_199_pad_type_0 = const()[name = tensor("v_199_pad_type_0"), val = tensor("custom")]; + tensor v_199_pad_0 = const()[name = tensor("v_199_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_5_attn2_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_5_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3757650752)))]; + tensor v_199_cast_fp16 = conv(dilations = var_44700, groups = var_30355, pad = v_199_pad_0, pad_type = v_199_pad_type_0, strides = var_44698, weight = up_blocks_0_attentions_1_transformer_blocks_5_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_199_cast_fp16")]; + tensor var_44704_begin_0 = const()[name = tensor("op_44704_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_44704_end_0 = const()[name = tensor("op_44704_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_44704_end_mask_0 = const()[name = tensor("op_44704_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44704_cast_fp16 = slice_by_index(begin = var_44704_begin_0, end = var_44704_end_0, end_mask = var_44704_end_mask_0, x = q_199_cast_fp16)[name = tensor("op_44704_cast_fp16")]; + tensor var_44708_begin_0 = const()[name = tensor("op_44708_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_44708_end_0 = const()[name = tensor("op_44708_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_44708_end_mask_0 = const()[name = tensor("op_44708_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44708_cast_fp16 = slice_by_index(begin = var_44708_begin_0, end = var_44708_end_0, end_mask = var_44708_end_mask_0, x = q_199_cast_fp16)[name = tensor("op_44708_cast_fp16")]; + tensor var_44712_begin_0 = const()[name = tensor("op_44712_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_44712_end_0 = const()[name = tensor("op_44712_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_44712_end_mask_0 = const()[name = tensor("op_44712_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44712_cast_fp16 = slice_by_index(begin = var_44712_begin_0, end = var_44712_end_0, end_mask = var_44712_end_mask_0, x = q_199_cast_fp16)[name = tensor("op_44712_cast_fp16")]; + tensor var_44716_begin_0 = const()[name = tensor("op_44716_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_44716_end_0 = const()[name = tensor("op_44716_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_44716_end_mask_0 = const()[name = tensor("op_44716_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44716_cast_fp16 = slice_by_index(begin = var_44716_begin_0, end = var_44716_end_0, end_mask = var_44716_end_mask_0, x = q_199_cast_fp16)[name = tensor("op_44716_cast_fp16")]; + tensor var_44720_begin_0 = const()[name = tensor("op_44720_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_44720_end_0 = const()[name = tensor("op_44720_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_44720_end_mask_0 = const()[name = tensor("op_44720_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44720_cast_fp16 = slice_by_index(begin = var_44720_begin_0, end = var_44720_end_0, end_mask = var_44720_end_mask_0, x = q_199_cast_fp16)[name = tensor("op_44720_cast_fp16")]; + tensor var_44724_begin_0 = const()[name = tensor("op_44724_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_44724_end_0 = const()[name = tensor("op_44724_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_44724_end_mask_0 = const()[name = tensor("op_44724_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44724_cast_fp16 = slice_by_index(begin = var_44724_begin_0, end = var_44724_end_0, end_mask = var_44724_end_mask_0, x = q_199_cast_fp16)[name = tensor("op_44724_cast_fp16")]; + tensor var_44728_begin_0 = const()[name = tensor("op_44728_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_44728_end_0 = const()[name = tensor("op_44728_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_44728_end_mask_0 = const()[name = tensor("op_44728_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44728_cast_fp16 = slice_by_index(begin = var_44728_begin_0, end = var_44728_end_0, end_mask = var_44728_end_mask_0, x = q_199_cast_fp16)[name = tensor("op_44728_cast_fp16")]; + tensor var_44732_begin_0 = const()[name = tensor("op_44732_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_44732_end_0 = const()[name = tensor("op_44732_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_44732_end_mask_0 = const()[name = tensor("op_44732_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44732_cast_fp16 = slice_by_index(begin = var_44732_begin_0, end = var_44732_end_0, end_mask = var_44732_end_mask_0, x = q_199_cast_fp16)[name = tensor("op_44732_cast_fp16")]; + tensor var_44736_begin_0 = const()[name = tensor("op_44736_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_44736_end_0 = const()[name = tensor("op_44736_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_44736_end_mask_0 = const()[name = tensor("op_44736_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44736_cast_fp16 = slice_by_index(begin = var_44736_begin_0, end = var_44736_end_0, end_mask = var_44736_end_mask_0, x = q_199_cast_fp16)[name = tensor("op_44736_cast_fp16")]; + tensor var_44740_begin_0 = const()[name = tensor("op_44740_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_44740_end_0 = const()[name = tensor("op_44740_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_44740_end_mask_0 = const()[name = tensor("op_44740_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44740_cast_fp16 = slice_by_index(begin = var_44740_begin_0, end = var_44740_end_0, end_mask = var_44740_end_mask_0, x = q_199_cast_fp16)[name = tensor("op_44740_cast_fp16")]; + tensor var_44744_begin_0 = const()[name = tensor("op_44744_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_44744_end_0 = const()[name = tensor("op_44744_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_44744_end_mask_0 = const()[name = tensor("op_44744_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44744_cast_fp16 = slice_by_index(begin = var_44744_begin_0, end = var_44744_end_0, end_mask = var_44744_end_mask_0, x = q_199_cast_fp16)[name = tensor("op_44744_cast_fp16")]; + tensor var_44748_begin_0 = const()[name = tensor("op_44748_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_44748_end_0 = const()[name = tensor("op_44748_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_44748_end_mask_0 = const()[name = tensor("op_44748_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44748_cast_fp16 = slice_by_index(begin = var_44748_begin_0, end = var_44748_end_0, end_mask = var_44748_end_mask_0, x = q_199_cast_fp16)[name = tensor("op_44748_cast_fp16")]; + tensor var_44752_begin_0 = const()[name = tensor("op_44752_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_44752_end_0 = const()[name = tensor("op_44752_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_44752_end_mask_0 = const()[name = tensor("op_44752_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44752_cast_fp16 = slice_by_index(begin = var_44752_begin_0, end = var_44752_end_0, end_mask = var_44752_end_mask_0, x = q_199_cast_fp16)[name = tensor("op_44752_cast_fp16")]; + tensor var_44756_begin_0 = const()[name = tensor("op_44756_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_44756_end_0 = const()[name = tensor("op_44756_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_44756_end_mask_0 = const()[name = tensor("op_44756_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44756_cast_fp16 = slice_by_index(begin = var_44756_begin_0, end = var_44756_end_0, end_mask = var_44756_end_mask_0, x = q_199_cast_fp16)[name = tensor("op_44756_cast_fp16")]; + tensor var_44760_begin_0 = const()[name = tensor("op_44760_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_44760_end_0 = const()[name = tensor("op_44760_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_44760_end_mask_0 = const()[name = tensor("op_44760_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44760_cast_fp16 = slice_by_index(begin = var_44760_begin_0, end = var_44760_end_0, end_mask = var_44760_end_mask_0, x = q_199_cast_fp16)[name = tensor("op_44760_cast_fp16")]; + tensor var_44764_begin_0 = const()[name = tensor("op_44764_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_44764_end_0 = const()[name = tensor("op_44764_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_44764_end_mask_0 = const()[name = tensor("op_44764_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44764_cast_fp16 = slice_by_index(begin = var_44764_begin_0, end = var_44764_end_0, end_mask = var_44764_end_mask_0, x = q_199_cast_fp16)[name = tensor("op_44764_cast_fp16")]; + tensor var_44768_begin_0 = const()[name = tensor("op_44768_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_44768_end_0 = const()[name = tensor("op_44768_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_44768_end_mask_0 = const()[name = tensor("op_44768_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44768_cast_fp16 = slice_by_index(begin = var_44768_begin_0, end = var_44768_end_0, end_mask = var_44768_end_mask_0, x = q_199_cast_fp16)[name = tensor("op_44768_cast_fp16")]; + tensor var_44772_begin_0 = const()[name = tensor("op_44772_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_44772_end_0 = const()[name = tensor("op_44772_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_44772_end_mask_0 = const()[name = tensor("op_44772_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44772_cast_fp16 = slice_by_index(begin = var_44772_begin_0, end = var_44772_end_0, end_mask = var_44772_end_mask_0, x = q_199_cast_fp16)[name = tensor("op_44772_cast_fp16")]; + tensor var_44776_begin_0 = const()[name = tensor("op_44776_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_44776_end_0 = const()[name = tensor("op_44776_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_44776_end_mask_0 = const()[name = tensor("op_44776_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44776_cast_fp16 = slice_by_index(begin = var_44776_begin_0, end = var_44776_end_0, end_mask = var_44776_end_mask_0, x = q_199_cast_fp16)[name = tensor("op_44776_cast_fp16")]; + tensor var_44780_begin_0 = const()[name = tensor("op_44780_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_44780_end_0 = const()[name = tensor("op_44780_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_44780_end_mask_0 = const()[name = tensor("op_44780_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44780_cast_fp16 = slice_by_index(begin = var_44780_begin_0, end = var_44780_end_0, end_mask = var_44780_end_mask_0, x = q_199_cast_fp16)[name = tensor("op_44780_cast_fp16")]; + tensor k_399_perm_0 = const()[name = tensor("k_399_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_44787_begin_0 = const()[name = tensor("op_44787_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_44787_end_0 = const()[name = tensor("op_44787_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_44787_end_mask_0 = const()[name = tensor("op_44787_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_399_cast_fp16 = transpose(perm = k_399_perm_0, x = k_397_cast_fp16)[name = tensor("transpose_40")]; + tensor var_44787_cast_fp16 = slice_by_index(begin = var_44787_begin_0, end = var_44787_end_0, end_mask = var_44787_end_mask_0, x = k_399_cast_fp16)[name = tensor("op_44787_cast_fp16")]; + tensor var_44791_begin_0 = const()[name = tensor("op_44791_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_44791_end_0 = const()[name = tensor("op_44791_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_44791_end_mask_0 = const()[name = tensor("op_44791_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44791_cast_fp16 = slice_by_index(begin = var_44791_begin_0, end = var_44791_end_0, end_mask = var_44791_end_mask_0, x = k_399_cast_fp16)[name = tensor("op_44791_cast_fp16")]; + tensor var_44795_begin_0 = const()[name = tensor("op_44795_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_44795_end_0 = const()[name = tensor("op_44795_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_44795_end_mask_0 = const()[name = tensor("op_44795_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44795_cast_fp16 = slice_by_index(begin = var_44795_begin_0, end = var_44795_end_0, end_mask = var_44795_end_mask_0, x = k_399_cast_fp16)[name = tensor("op_44795_cast_fp16")]; + tensor var_44799_begin_0 = const()[name = tensor("op_44799_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_44799_end_0 = const()[name = tensor("op_44799_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_44799_end_mask_0 = const()[name = tensor("op_44799_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44799_cast_fp16 = slice_by_index(begin = var_44799_begin_0, end = var_44799_end_0, end_mask = var_44799_end_mask_0, x = k_399_cast_fp16)[name = tensor("op_44799_cast_fp16")]; + tensor var_44803_begin_0 = const()[name = tensor("op_44803_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_44803_end_0 = const()[name = tensor("op_44803_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_44803_end_mask_0 = const()[name = tensor("op_44803_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44803_cast_fp16 = slice_by_index(begin = var_44803_begin_0, end = var_44803_end_0, end_mask = var_44803_end_mask_0, x = k_399_cast_fp16)[name = tensor("op_44803_cast_fp16")]; + tensor var_44807_begin_0 = const()[name = tensor("op_44807_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_44807_end_0 = const()[name = tensor("op_44807_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_44807_end_mask_0 = const()[name = tensor("op_44807_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44807_cast_fp16 = slice_by_index(begin = var_44807_begin_0, end = var_44807_end_0, end_mask = var_44807_end_mask_0, x = k_399_cast_fp16)[name = tensor("op_44807_cast_fp16")]; + tensor var_44811_begin_0 = const()[name = tensor("op_44811_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_44811_end_0 = const()[name = tensor("op_44811_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_44811_end_mask_0 = const()[name = tensor("op_44811_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44811_cast_fp16 = slice_by_index(begin = var_44811_begin_0, end = var_44811_end_0, end_mask = var_44811_end_mask_0, x = k_399_cast_fp16)[name = tensor("op_44811_cast_fp16")]; + tensor var_44815_begin_0 = const()[name = tensor("op_44815_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_44815_end_0 = const()[name = tensor("op_44815_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_44815_end_mask_0 = const()[name = tensor("op_44815_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44815_cast_fp16 = slice_by_index(begin = var_44815_begin_0, end = var_44815_end_0, end_mask = var_44815_end_mask_0, x = k_399_cast_fp16)[name = tensor("op_44815_cast_fp16")]; + tensor var_44819_begin_0 = const()[name = tensor("op_44819_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_44819_end_0 = const()[name = tensor("op_44819_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_44819_end_mask_0 = const()[name = tensor("op_44819_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44819_cast_fp16 = slice_by_index(begin = var_44819_begin_0, end = var_44819_end_0, end_mask = var_44819_end_mask_0, x = k_399_cast_fp16)[name = tensor("op_44819_cast_fp16")]; + tensor var_44823_begin_0 = const()[name = tensor("op_44823_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_44823_end_0 = const()[name = tensor("op_44823_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_44823_end_mask_0 = const()[name = tensor("op_44823_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44823_cast_fp16 = slice_by_index(begin = var_44823_begin_0, end = var_44823_end_0, end_mask = var_44823_end_mask_0, x = k_399_cast_fp16)[name = tensor("op_44823_cast_fp16")]; + tensor var_44827_begin_0 = const()[name = tensor("op_44827_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_44827_end_0 = const()[name = tensor("op_44827_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_44827_end_mask_0 = const()[name = tensor("op_44827_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44827_cast_fp16 = slice_by_index(begin = var_44827_begin_0, end = var_44827_end_0, end_mask = var_44827_end_mask_0, x = k_399_cast_fp16)[name = tensor("op_44827_cast_fp16")]; + tensor var_44831_begin_0 = const()[name = tensor("op_44831_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_44831_end_0 = const()[name = tensor("op_44831_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_44831_end_mask_0 = const()[name = tensor("op_44831_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44831_cast_fp16 = slice_by_index(begin = var_44831_begin_0, end = var_44831_end_0, end_mask = var_44831_end_mask_0, x = k_399_cast_fp16)[name = tensor("op_44831_cast_fp16")]; + tensor var_44835_begin_0 = const()[name = tensor("op_44835_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_44835_end_0 = const()[name = tensor("op_44835_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_44835_end_mask_0 = const()[name = tensor("op_44835_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44835_cast_fp16 = slice_by_index(begin = var_44835_begin_0, end = var_44835_end_0, end_mask = var_44835_end_mask_0, x = k_399_cast_fp16)[name = tensor("op_44835_cast_fp16")]; + tensor var_44839_begin_0 = const()[name = tensor("op_44839_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_44839_end_0 = const()[name = tensor("op_44839_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_44839_end_mask_0 = const()[name = tensor("op_44839_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44839_cast_fp16 = slice_by_index(begin = var_44839_begin_0, end = var_44839_end_0, end_mask = var_44839_end_mask_0, x = k_399_cast_fp16)[name = tensor("op_44839_cast_fp16")]; + tensor var_44843_begin_0 = const()[name = tensor("op_44843_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_44843_end_0 = const()[name = tensor("op_44843_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_44843_end_mask_0 = const()[name = tensor("op_44843_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44843_cast_fp16 = slice_by_index(begin = var_44843_begin_0, end = var_44843_end_0, end_mask = var_44843_end_mask_0, x = k_399_cast_fp16)[name = tensor("op_44843_cast_fp16")]; + tensor var_44847_begin_0 = const()[name = tensor("op_44847_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_44847_end_0 = const()[name = tensor("op_44847_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_44847_end_mask_0 = const()[name = tensor("op_44847_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44847_cast_fp16 = slice_by_index(begin = var_44847_begin_0, end = var_44847_end_0, end_mask = var_44847_end_mask_0, x = k_399_cast_fp16)[name = tensor("op_44847_cast_fp16")]; + tensor var_44851_begin_0 = const()[name = tensor("op_44851_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_44851_end_0 = const()[name = tensor("op_44851_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_44851_end_mask_0 = const()[name = tensor("op_44851_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44851_cast_fp16 = slice_by_index(begin = var_44851_begin_0, end = var_44851_end_0, end_mask = var_44851_end_mask_0, x = k_399_cast_fp16)[name = tensor("op_44851_cast_fp16")]; + tensor var_44855_begin_0 = const()[name = tensor("op_44855_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_44855_end_0 = const()[name = tensor("op_44855_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_44855_end_mask_0 = const()[name = tensor("op_44855_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44855_cast_fp16 = slice_by_index(begin = var_44855_begin_0, end = var_44855_end_0, end_mask = var_44855_end_mask_0, x = k_399_cast_fp16)[name = tensor("op_44855_cast_fp16")]; + tensor var_44859_begin_0 = const()[name = tensor("op_44859_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_44859_end_0 = const()[name = tensor("op_44859_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_44859_end_mask_0 = const()[name = tensor("op_44859_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44859_cast_fp16 = slice_by_index(begin = var_44859_begin_0, end = var_44859_end_0, end_mask = var_44859_end_mask_0, x = k_399_cast_fp16)[name = tensor("op_44859_cast_fp16")]; + tensor var_44863_begin_0 = const()[name = tensor("op_44863_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_44863_end_0 = const()[name = tensor("op_44863_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_44863_end_mask_0 = const()[name = tensor("op_44863_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44863_cast_fp16 = slice_by_index(begin = var_44863_begin_0, end = var_44863_end_0, end_mask = var_44863_end_mask_0, x = k_399_cast_fp16)[name = tensor("op_44863_cast_fp16")]; + tensor var_44865_begin_0 = const()[name = tensor("op_44865_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_44865_end_0 = const()[name = tensor("op_44865_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_44865_end_mask_0 = const()[name = tensor("op_44865_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44865_cast_fp16 = slice_by_index(begin = var_44865_begin_0, end = var_44865_end_0, end_mask = var_44865_end_mask_0, x = v_199_cast_fp16)[name = tensor("op_44865_cast_fp16")]; + tensor var_44869_begin_0 = const()[name = tensor("op_44869_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_44869_end_0 = const()[name = tensor("op_44869_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_44869_end_mask_0 = const()[name = tensor("op_44869_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44869_cast_fp16 = slice_by_index(begin = var_44869_begin_0, end = var_44869_end_0, end_mask = var_44869_end_mask_0, x = v_199_cast_fp16)[name = tensor("op_44869_cast_fp16")]; + tensor var_44873_begin_0 = const()[name = tensor("op_44873_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_44873_end_0 = const()[name = tensor("op_44873_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_44873_end_mask_0 = const()[name = tensor("op_44873_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44873_cast_fp16 = slice_by_index(begin = var_44873_begin_0, end = var_44873_end_0, end_mask = var_44873_end_mask_0, x = v_199_cast_fp16)[name = tensor("op_44873_cast_fp16")]; + tensor var_44877_begin_0 = const()[name = tensor("op_44877_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_44877_end_0 = const()[name = tensor("op_44877_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_44877_end_mask_0 = const()[name = tensor("op_44877_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44877_cast_fp16 = slice_by_index(begin = var_44877_begin_0, end = var_44877_end_0, end_mask = var_44877_end_mask_0, x = v_199_cast_fp16)[name = tensor("op_44877_cast_fp16")]; + tensor var_44881_begin_0 = const()[name = tensor("op_44881_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_44881_end_0 = const()[name = tensor("op_44881_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_44881_end_mask_0 = const()[name = tensor("op_44881_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44881_cast_fp16 = slice_by_index(begin = var_44881_begin_0, end = var_44881_end_0, end_mask = var_44881_end_mask_0, x = v_199_cast_fp16)[name = tensor("op_44881_cast_fp16")]; + tensor var_44885_begin_0 = const()[name = tensor("op_44885_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_44885_end_0 = const()[name = tensor("op_44885_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_44885_end_mask_0 = const()[name = tensor("op_44885_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44885_cast_fp16 = slice_by_index(begin = var_44885_begin_0, end = var_44885_end_0, end_mask = var_44885_end_mask_0, x = v_199_cast_fp16)[name = tensor("op_44885_cast_fp16")]; + tensor var_44889_begin_0 = const()[name = tensor("op_44889_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_44889_end_0 = const()[name = tensor("op_44889_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_44889_end_mask_0 = const()[name = tensor("op_44889_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44889_cast_fp16 = slice_by_index(begin = var_44889_begin_0, end = var_44889_end_0, end_mask = var_44889_end_mask_0, x = v_199_cast_fp16)[name = tensor("op_44889_cast_fp16")]; + tensor var_44893_begin_0 = const()[name = tensor("op_44893_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_44893_end_0 = const()[name = tensor("op_44893_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_44893_end_mask_0 = const()[name = tensor("op_44893_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44893_cast_fp16 = slice_by_index(begin = var_44893_begin_0, end = var_44893_end_0, end_mask = var_44893_end_mask_0, x = v_199_cast_fp16)[name = tensor("op_44893_cast_fp16")]; + tensor var_44897_begin_0 = const()[name = tensor("op_44897_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_44897_end_0 = const()[name = tensor("op_44897_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_44897_end_mask_0 = const()[name = tensor("op_44897_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44897_cast_fp16 = slice_by_index(begin = var_44897_begin_0, end = var_44897_end_0, end_mask = var_44897_end_mask_0, x = v_199_cast_fp16)[name = tensor("op_44897_cast_fp16")]; + tensor var_44901_begin_0 = const()[name = tensor("op_44901_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_44901_end_0 = const()[name = tensor("op_44901_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_44901_end_mask_0 = const()[name = tensor("op_44901_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44901_cast_fp16 = slice_by_index(begin = var_44901_begin_0, end = var_44901_end_0, end_mask = var_44901_end_mask_0, x = v_199_cast_fp16)[name = tensor("op_44901_cast_fp16")]; + tensor var_44905_begin_0 = const()[name = tensor("op_44905_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_44905_end_0 = const()[name = tensor("op_44905_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_44905_end_mask_0 = const()[name = tensor("op_44905_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44905_cast_fp16 = slice_by_index(begin = var_44905_begin_0, end = var_44905_end_0, end_mask = var_44905_end_mask_0, x = v_199_cast_fp16)[name = tensor("op_44905_cast_fp16")]; + tensor var_44909_begin_0 = const()[name = tensor("op_44909_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_44909_end_0 = const()[name = tensor("op_44909_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_44909_end_mask_0 = const()[name = tensor("op_44909_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44909_cast_fp16 = slice_by_index(begin = var_44909_begin_0, end = var_44909_end_0, end_mask = var_44909_end_mask_0, x = v_199_cast_fp16)[name = tensor("op_44909_cast_fp16")]; + tensor var_44913_begin_0 = const()[name = tensor("op_44913_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_44913_end_0 = const()[name = tensor("op_44913_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_44913_end_mask_0 = const()[name = tensor("op_44913_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44913_cast_fp16 = slice_by_index(begin = var_44913_begin_0, end = var_44913_end_0, end_mask = var_44913_end_mask_0, x = v_199_cast_fp16)[name = tensor("op_44913_cast_fp16")]; + tensor var_44917_begin_0 = const()[name = tensor("op_44917_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_44917_end_0 = const()[name = tensor("op_44917_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_44917_end_mask_0 = const()[name = tensor("op_44917_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44917_cast_fp16 = slice_by_index(begin = var_44917_begin_0, end = var_44917_end_0, end_mask = var_44917_end_mask_0, x = v_199_cast_fp16)[name = tensor("op_44917_cast_fp16")]; + tensor var_44921_begin_0 = const()[name = tensor("op_44921_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_44921_end_0 = const()[name = tensor("op_44921_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_44921_end_mask_0 = const()[name = tensor("op_44921_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44921_cast_fp16 = slice_by_index(begin = var_44921_begin_0, end = var_44921_end_0, end_mask = var_44921_end_mask_0, x = v_199_cast_fp16)[name = tensor("op_44921_cast_fp16")]; + tensor var_44925_begin_0 = const()[name = tensor("op_44925_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_44925_end_0 = const()[name = tensor("op_44925_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_44925_end_mask_0 = const()[name = tensor("op_44925_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44925_cast_fp16 = slice_by_index(begin = var_44925_begin_0, end = var_44925_end_0, end_mask = var_44925_end_mask_0, x = v_199_cast_fp16)[name = tensor("op_44925_cast_fp16")]; + tensor var_44929_begin_0 = const()[name = tensor("op_44929_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_44929_end_0 = const()[name = tensor("op_44929_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_44929_end_mask_0 = const()[name = tensor("op_44929_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44929_cast_fp16 = slice_by_index(begin = var_44929_begin_0, end = var_44929_end_0, end_mask = var_44929_end_mask_0, x = v_199_cast_fp16)[name = tensor("op_44929_cast_fp16")]; + tensor var_44933_begin_0 = const()[name = tensor("op_44933_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_44933_end_0 = const()[name = tensor("op_44933_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_44933_end_mask_0 = const()[name = tensor("op_44933_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44933_cast_fp16 = slice_by_index(begin = var_44933_begin_0, end = var_44933_end_0, end_mask = var_44933_end_mask_0, x = v_199_cast_fp16)[name = tensor("op_44933_cast_fp16")]; + tensor var_44937_begin_0 = const()[name = tensor("op_44937_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_44937_end_0 = const()[name = tensor("op_44937_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_44937_end_mask_0 = const()[name = tensor("op_44937_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44937_cast_fp16 = slice_by_index(begin = var_44937_begin_0, end = var_44937_end_0, end_mask = var_44937_end_mask_0, x = v_199_cast_fp16)[name = tensor("op_44937_cast_fp16")]; + tensor var_44941_begin_0 = const()[name = tensor("op_44941_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_44941_end_0 = const()[name = tensor("op_44941_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_44941_end_mask_0 = const()[name = tensor("op_44941_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44941_cast_fp16 = slice_by_index(begin = var_44941_begin_0, end = var_44941_end_0, end_mask = var_44941_end_mask_0, x = v_199_cast_fp16)[name = tensor("op_44941_cast_fp16")]; + tensor var_44945_equation_0 = const()[name = tensor("op_44945_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_44945_cast_fp16 = einsum(equation = var_44945_equation_0, values = (var_44787_cast_fp16, var_44704_cast_fp16))[name = tensor("op_44945_cast_fp16")]; + tensor var_44946_to_fp16 = const()[name = tensor("op_44946_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3801_cast_fp16 = mul(x = var_44945_cast_fp16, y = var_44946_to_fp16)[name = tensor("aw_3801_cast_fp16")]; + tensor var_44949_equation_0 = const()[name = tensor("op_44949_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_44949_cast_fp16 = einsum(equation = var_44949_equation_0, values = (var_44791_cast_fp16, var_44708_cast_fp16))[name = tensor("op_44949_cast_fp16")]; + tensor var_44950_to_fp16 = const()[name = tensor("op_44950_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3803_cast_fp16 = mul(x = var_44949_cast_fp16, y = var_44950_to_fp16)[name = tensor("aw_3803_cast_fp16")]; + tensor var_44953_equation_0 = const()[name = tensor("op_44953_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_44953_cast_fp16 = einsum(equation = var_44953_equation_0, values = (var_44795_cast_fp16, var_44712_cast_fp16))[name = tensor("op_44953_cast_fp16")]; + tensor var_44954_to_fp16 = const()[name = tensor("op_44954_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3805_cast_fp16 = mul(x = var_44953_cast_fp16, y = var_44954_to_fp16)[name = tensor("aw_3805_cast_fp16")]; + tensor var_44957_equation_0 = const()[name = tensor("op_44957_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_44957_cast_fp16 = einsum(equation = var_44957_equation_0, values = (var_44799_cast_fp16, var_44716_cast_fp16))[name = tensor("op_44957_cast_fp16")]; + tensor var_44958_to_fp16 = const()[name = tensor("op_44958_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3807_cast_fp16 = mul(x = var_44957_cast_fp16, y = var_44958_to_fp16)[name = tensor("aw_3807_cast_fp16")]; + tensor var_44961_equation_0 = const()[name = tensor("op_44961_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_44961_cast_fp16 = einsum(equation = var_44961_equation_0, values = (var_44803_cast_fp16, var_44720_cast_fp16))[name = tensor("op_44961_cast_fp16")]; + tensor var_44962_to_fp16 = const()[name = tensor("op_44962_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3809_cast_fp16 = mul(x = var_44961_cast_fp16, y = var_44962_to_fp16)[name = tensor("aw_3809_cast_fp16")]; + tensor var_44965_equation_0 = const()[name = tensor("op_44965_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_44965_cast_fp16 = einsum(equation = var_44965_equation_0, values = (var_44807_cast_fp16, var_44724_cast_fp16))[name = tensor("op_44965_cast_fp16")]; + tensor var_44966_to_fp16 = const()[name = tensor("op_44966_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3811_cast_fp16 = mul(x = var_44965_cast_fp16, y = var_44966_to_fp16)[name = tensor("aw_3811_cast_fp16")]; + tensor var_44969_equation_0 = const()[name = tensor("op_44969_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_44969_cast_fp16 = einsum(equation = var_44969_equation_0, values = (var_44811_cast_fp16, var_44728_cast_fp16))[name = tensor("op_44969_cast_fp16")]; + tensor var_44970_to_fp16 = const()[name = tensor("op_44970_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3813_cast_fp16 = mul(x = var_44969_cast_fp16, y = var_44970_to_fp16)[name = tensor("aw_3813_cast_fp16")]; + tensor var_44973_equation_0 = const()[name = tensor("op_44973_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_44973_cast_fp16 = einsum(equation = var_44973_equation_0, values = (var_44815_cast_fp16, var_44732_cast_fp16))[name = tensor("op_44973_cast_fp16")]; + tensor var_44974_to_fp16 = const()[name = tensor("op_44974_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3815_cast_fp16 = mul(x = var_44973_cast_fp16, y = var_44974_to_fp16)[name = tensor("aw_3815_cast_fp16")]; + tensor var_44977_equation_0 = const()[name = tensor("op_44977_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_44977_cast_fp16 = einsum(equation = var_44977_equation_0, values = (var_44819_cast_fp16, var_44736_cast_fp16))[name = tensor("op_44977_cast_fp16")]; + tensor var_44978_to_fp16 = const()[name = tensor("op_44978_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3817_cast_fp16 = mul(x = var_44977_cast_fp16, y = var_44978_to_fp16)[name = tensor("aw_3817_cast_fp16")]; + tensor var_44981_equation_0 = const()[name = tensor("op_44981_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_44981_cast_fp16 = einsum(equation = var_44981_equation_0, values = (var_44823_cast_fp16, var_44740_cast_fp16))[name = tensor("op_44981_cast_fp16")]; + tensor var_44982_to_fp16 = const()[name = tensor("op_44982_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3819_cast_fp16 = mul(x = var_44981_cast_fp16, y = var_44982_to_fp16)[name = tensor("aw_3819_cast_fp16")]; + tensor var_44985_equation_0 = const()[name = tensor("op_44985_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_44985_cast_fp16 = einsum(equation = var_44985_equation_0, values = (var_44827_cast_fp16, var_44744_cast_fp16))[name = tensor("op_44985_cast_fp16")]; + tensor var_44986_to_fp16 = const()[name = tensor("op_44986_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3821_cast_fp16 = mul(x = var_44985_cast_fp16, y = var_44986_to_fp16)[name = tensor("aw_3821_cast_fp16")]; + tensor var_44989_equation_0 = const()[name = tensor("op_44989_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_44989_cast_fp16 = einsum(equation = var_44989_equation_0, values = (var_44831_cast_fp16, var_44748_cast_fp16))[name = tensor("op_44989_cast_fp16")]; + tensor var_44990_to_fp16 = const()[name = tensor("op_44990_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3823_cast_fp16 = mul(x = var_44989_cast_fp16, y = var_44990_to_fp16)[name = tensor("aw_3823_cast_fp16")]; + tensor var_44993_equation_0 = const()[name = tensor("op_44993_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_44993_cast_fp16 = einsum(equation = var_44993_equation_0, values = (var_44835_cast_fp16, var_44752_cast_fp16))[name = tensor("op_44993_cast_fp16")]; + tensor var_44994_to_fp16 = const()[name = tensor("op_44994_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3825_cast_fp16 = mul(x = var_44993_cast_fp16, y = var_44994_to_fp16)[name = tensor("aw_3825_cast_fp16")]; + tensor var_44997_equation_0 = const()[name = tensor("op_44997_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_44997_cast_fp16 = einsum(equation = var_44997_equation_0, values = (var_44839_cast_fp16, var_44756_cast_fp16))[name = tensor("op_44997_cast_fp16")]; + tensor var_44998_to_fp16 = const()[name = tensor("op_44998_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3827_cast_fp16 = mul(x = var_44997_cast_fp16, y = var_44998_to_fp16)[name = tensor("aw_3827_cast_fp16")]; + tensor var_45001_equation_0 = const()[name = tensor("op_45001_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45001_cast_fp16 = einsum(equation = var_45001_equation_0, values = (var_44843_cast_fp16, var_44760_cast_fp16))[name = tensor("op_45001_cast_fp16")]; + tensor var_45002_to_fp16 = const()[name = tensor("op_45002_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3829_cast_fp16 = mul(x = var_45001_cast_fp16, y = var_45002_to_fp16)[name = tensor("aw_3829_cast_fp16")]; + tensor var_45005_equation_0 = const()[name = tensor("op_45005_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45005_cast_fp16 = einsum(equation = var_45005_equation_0, values = (var_44847_cast_fp16, var_44764_cast_fp16))[name = tensor("op_45005_cast_fp16")]; + tensor var_45006_to_fp16 = const()[name = tensor("op_45006_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3831_cast_fp16 = mul(x = var_45005_cast_fp16, y = var_45006_to_fp16)[name = tensor("aw_3831_cast_fp16")]; + tensor var_45009_equation_0 = const()[name = tensor("op_45009_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45009_cast_fp16 = einsum(equation = var_45009_equation_0, values = (var_44851_cast_fp16, var_44768_cast_fp16))[name = tensor("op_45009_cast_fp16")]; + tensor var_45010_to_fp16 = const()[name = tensor("op_45010_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3833_cast_fp16 = mul(x = var_45009_cast_fp16, y = var_45010_to_fp16)[name = tensor("aw_3833_cast_fp16")]; + tensor var_45013_equation_0 = const()[name = tensor("op_45013_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45013_cast_fp16 = einsum(equation = var_45013_equation_0, values = (var_44855_cast_fp16, var_44772_cast_fp16))[name = tensor("op_45013_cast_fp16")]; + tensor var_45014_to_fp16 = const()[name = tensor("op_45014_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3835_cast_fp16 = mul(x = var_45013_cast_fp16, y = var_45014_to_fp16)[name = tensor("aw_3835_cast_fp16")]; + tensor var_45017_equation_0 = const()[name = tensor("op_45017_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45017_cast_fp16 = einsum(equation = var_45017_equation_0, values = (var_44859_cast_fp16, var_44776_cast_fp16))[name = tensor("op_45017_cast_fp16")]; + tensor var_45018_to_fp16 = const()[name = tensor("op_45018_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3837_cast_fp16 = mul(x = var_45017_cast_fp16, y = var_45018_to_fp16)[name = tensor("aw_3837_cast_fp16")]; + tensor var_45021_equation_0 = const()[name = tensor("op_45021_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45021_cast_fp16 = einsum(equation = var_45021_equation_0, values = (var_44863_cast_fp16, var_44780_cast_fp16))[name = tensor("op_45021_cast_fp16")]; + tensor var_45022_to_fp16 = const()[name = tensor("op_45022_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3839_cast_fp16 = mul(x = var_45021_cast_fp16, y = var_45022_to_fp16)[name = tensor("aw_3839_cast_fp16")]; + tensor var_45024_cast_fp16 = softmax(axis = var_30355, x = aw_3801_cast_fp16)[name = tensor("op_45024_cast_fp16")]; + tensor var_45025_cast_fp16 = softmax(axis = var_30355, x = aw_3803_cast_fp16)[name = tensor("op_45025_cast_fp16")]; + tensor var_45026_cast_fp16 = softmax(axis = var_30355, x = aw_3805_cast_fp16)[name = tensor("op_45026_cast_fp16")]; + tensor var_45027_cast_fp16 = softmax(axis = var_30355, x = aw_3807_cast_fp16)[name = tensor("op_45027_cast_fp16")]; + tensor var_45028_cast_fp16 = softmax(axis = var_30355, x = aw_3809_cast_fp16)[name = tensor("op_45028_cast_fp16")]; + tensor var_45029_cast_fp16 = softmax(axis = var_30355, x = aw_3811_cast_fp16)[name = tensor("op_45029_cast_fp16")]; + tensor var_45030_cast_fp16 = softmax(axis = var_30355, x = aw_3813_cast_fp16)[name = tensor("op_45030_cast_fp16")]; + tensor var_45031_cast_fp16 = softmax(axis = var_30355, x = aw_3815_cast_fp16)[name = tensor("op_45031_cast_fp16")]; + tensor var_45032_cast_fp16 = softmax(axis = var_30355, x = aw_3817_cast_fp16)[name = tensor("op_45032_cast_fp16")]; + tensor var_45033_cast_fp16 = softmax(axis = var_30355, x = aw_3819_cast_fp16)[name = tensor("op_45033_cast_fp16")]; + tensor var_45034_cast_fp16 = softmax(axis = var_30355, x = aw_3821_cast_fp16)[name = tensor("op_45034_cast_fp16")]; + tensor var_45035_cast_fp16 = softmax(axis = var_30355, x = aw_3823_cast_fp16)[name = tensor("op_45035_cast_fp16")]; + tensor var_45036_cast_fp16 = softmax(axis = var_30355, x = aw_3825_cast_fp16)[name = tensor("op_45036_cast_fp16")]; + tensor var_45037_cast_fp16 = softmax(axis = var_30355, x = aw_3827_cast_fp16)[name = tensor("op_45037_cast_fp16")]; + tensor var_45038_cast_fp16 = softmax(axis = var_30355, x = aw_3829_cast_fp16)[name = tensor("op_45038_cast_fp16")]; + tensor var_45039_cast_fp16 = softmax(axis = var_30355, x = aw_3831_cast_fp16)[name = tensor("op_45039_cast_fp16")]; + tensor var_45040_cast_fp16 = softmax(axis = var_30355, x = aw_3833_cast_fp16)[name = tensor("op_45040_cast_fp16")]; + tensor var_45041_cast_fp16 = softmax(axis = var_30355, x = aw_3835_cast_fp16)[name = tensor("op_45041_cast_fp16")]; + tensor var_45042_cast_fp16 = softmax(axis = var_30355, x = aw_3837_cast_fp16)[name = tensor("op_45042_cast_fp16")]; + tensor var_45043_cast_fp16 = softmax(axis = var_30355, x = aw_3839_cast_fp16)[name = tensor("op_45043_cast_fp16")]; + tensor var_45045_equation_0 = const()[name = tensor("op_45045_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45045_cast_fp16 = einsum(equation = var_45045_equation_0, values = (var_44865_cast_fp16, var_45024_cast_fp16))[name = tensor("op_45045_cast_fp16")]; + tensor var_45047_equation_0 = const()[name = tensor("op_45047_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45047_cast_fp16 = einsum(equation = var_45047_equation_0, values = (var_44869_cast_fp16, var_45025_cast_fp16))[name = tensor("op_45047_cast_fp16")]; + tensor var_45049_equation_0 = const()[name = tensor("op_45049_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45049_cast_fp16 = einsum(equation = var_45049_equation_0, values = (var_44873_cast_fp16, var_45026_cast_fp16))[name = tensor("op_45049_cast_fp16")]; + tensor var_45051_equation_0 = const()[name = tensor("op_45051_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45051_cast_fp16 = einsum(equation = var_45051_equation_0, values = (var_44877_cast_fp16, var_45027_cast_fp16))[name = tensor("op_45051_cast_fp16")]; + tensor var_45053_equation_0 = const()[name = tensor("op_45053_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45053_cast_fp16 = einsum(equation = var_45053_equation_0, values = (var_44881_cast_fp16, var_45028_cast_fp16))[name = tensor("op_45053_cast_fp16")]; + tensor var_45055_equation_0 = const()[name = tensor("op_45055_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45055_cast_fp16 = einsum(equation = var_45055_equation_0, values = (var_44885_cast_fp16, var_45029_cast_fp16))[name = tensor("op_45055_cast_fp16")]; + tensor var_45057_equation_0 = const()[name = tensor("op_45057_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45057_cast_fp16 = einsum(equation = var_45057_equation_0, values = (var_44889_cast_fp16, var_45030_cast_fp16))[name = tensor("op_45057_cast_fp16")]; + tensor var_45059_equation_0 = const()[name = tensor("op_45059_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45059_cast_fp16 = einsum(equation = var_45059_equation_0, values = (var_44893_cast_fp16, var_45031_cast_fp16))[name = tensor("op_45059_cast_fp16")]; + tensor var_45061_equation_0 = const()[name = tensor("op_45061_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45061_cast_fp16 = einsum(equation = var_45061_equation_0, values = (var_44897_cast_fp16, var_45032_cast_fp16))[name = tensor("op_45061_cast_fp16")]; + tensor var_45063_equation_0 = const()[name = tensor("op_45063_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45063_cast_fp16 = einsum(equation = var_45063_equation_0, values = (var_44901_cast_fp16, var_45033_cast_fp16))[name = tensor("op_45063_cast_fp16")]; + tensor var_45065_equation_0 = const()[name = tensor("op_45065_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45065_cast_fp16 = einsum(equation = var_45065_equation_0, values = (var_44905_cast_fp16, var_45034_cast_fp16))[name = tensor("op_45065_cast_fp16")]; + tensor var_45067_equation_0 = const()[name = tensor("op_45067_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45067_cast_fp16 = einsum(equation = var_45067_equation_0, values = (var_44909_cast_fp16, var_45035_cast_fp16))[name = tensor("op_45067_cast_fp16")]; + tensor var_45069_equation_0 = const()[name = tensor("op_45069_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45069_cast_fp16 = einsum(equation = var_45069_equation_0, values = (var_44913_cast_fp16, var_45036_cast_fp16))[name = tensor("op_45069_cast_fp16")]; + tensor var_45071_equation_0 = const()[name = tensor("op_45071_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45071_cast_fp16 = einsum(equation = var_45071_equation_0, values = (var_44917_cast_fp16, var_45037_cast_fp16))[name = tensor("op_45071_cast_fp16")]; + tensor var_45073_equation_0 = const()[name = tensor("op_45073_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45073_cast_fp16 = einsum(equation = var_45073_equation_0, values = (var_44921_cast_fp16, var_45038_cast_fp16))[name = tensor("op_45073_cast_fp16")]; + tensor var_45075_equation_0 = const()[name = tensor("op_45075_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45075_cast_fp16 = einsum(equation = var_45075_equation_0, values = (var_44925_cast_fp16, var_45039_cast_fp16))[name = tensor("op_45075_cast_fp16")]; + tensor var_45077_equation_0 = const()[name = tensor("op_45077_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45077_cast_fp16 = einsum(equation = var_45077_equation_0, values = (var_44929_cast_fp16, var_45040_cast_fp16))[name = tensor("op_45077_cast_fp16")]; + tensor var_45079_equation_0 = const()[name = tensor("op_45079_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45079_cast_fp16 = einsum(equation = var_45079_equation_0, values = (var_44933_cast_fp16, var_45041_cast_fp16))[name = tensor("op_45079_cast_fp16")]; + tensor var_45081_equation_0 = const()[name = tensor("op_45081_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45081_cast_fp16 = einsum(equation = var_45081_equation_0, values = (var_44937_cast_fp16, var_45042_cast_fp16))[name = tensor("op_45081_cast_fp16")]; + tensor var_45083_equation_0 = const()[name = tensor("op_45083_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45083_cast_fp16 = einsum(equation = var_45083_equation_0, values = (var_44941_cast_fp16, var_45043_cast_fp16))[name = tensor("op_45083_cast_fp16")]; + tensor input_579_interleave_0 = const()[name = tensor("input_579_interleave_0"), val = tensor(false)]; + tensor input_579_cast_fp16 = concat(axis = var_30355, interleave = input_579_interleave_0, values = (var_45045_cast_fp16, var_45047_cast_fp16, var_45049_cast_fp16, var_45051_cast_fp16, var_45053_cast_fp16, var_45055_cast_fp16, var_45057_cast_fp16, var_45059_cast_fp16, var_45061_cast_fp16, var_45063_cast_fp16, var_45065_cast_fp16, var_45067_cast_fp16, var_45069_cast_fp16, var_45071_cast_fp16, var_45073_cast_fp16, var_45075_cast_fp16, var_45077_cast_fp16, var_45079_cast_fp16, var_45081_cast_fp16, var_45083_cast_fp16))[name = tensor("input_579_cast_fp16")]; + tensor var_45089 = const()[name = tensor("op_45089"), val = tensor([1, 1])]; + tensor var_45091 = const()[name = tensor("op_45091"), val = tensor([1, 1])]; + tensor var_45093_pad_type_0 = const()[name = tensor("op_45093_pad_type_0"), val = tensor("custom")]; + tensor var_45093_pad_0 = const()[name = tensor("op_45093_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_5_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_5_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3762893696)))]; + tensor up_blocks_0_attentions_1_transformer_blocks_5_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_5_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3766170560)))]; + tensor var_45093_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_5_attn2_to_out_0_bias_to_fp16, dilations = var_45091, groups = var_30355, pad = var_45093_pad_0, pad_type = var_45093_pad_type_0, strides = var_45089, weight = up_blocks_0_attentions_1_transformer_blocks_5_attn2_to_out_0_weight_to_fp16, x = input_579_cast_fp16)[name = tensor("op_45093_cast_fp16")]; + tensor inputs_299_cast_fp16 = add(x = var_45093_cast_fp16, y = inputs_297_cast_fp16)[name = tensor("inputs_299_cast_fp16")]; + tensor input_581_axes_0 = const()[name = tensor("input_581_axes_0"), val = tensor([1])]; + tensor input_581_gamma_0_to_fp16 = const()[name = tensor("input_581_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3766173184)))]; + tensor input_581_beta_0_to_fp16 = const()[name = tensor("input_581_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3766175808)))]; + tensor var_45103_to_fp16 = const()[name = tensor("op_45103_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_581_cast_fp16 = layer_norm(axes = input_581_axes_0, beta = input_581_beta_0_to_fp16, epsilon = var_45103_to_fp16, gamma = input_581_gamma_0_to_fp16, x = inputs_299_cast_fp16)[name = tensor("input_581_cast_fp16")]; + tensor var_45119 = const()[name = tensor("op_45119"), val = tensor([1, 1])]; + tensor var_45121 = const()[name = tensor("op_45121"), val = tensor([1, 1])]; + tensor var_45123_pad_type_0 = const()[name = tensor("op_45123_pad_type_0"), val = tensor("custom")]; + tensor var_45123_pad_0 = const()[name = tensor("op_45123_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_5_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_5_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3766178432)))]; + tensor up_blocks_0_attentions_1_transformer_blocks_5_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_5_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3792392896)))]; + tensor var_45123_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_5_ff_net_0_proj_bias_to_fp16, dilations = var_45121, groups = var_30355, pad = var_45123_pad_0, pad_type = var_45123_pad_type_0, strides = var_45119, weight = up_blocks_0_attentions_1_transformer_blocks_5_ff_net_0_proj_weight_to_fp16, x = input_581_cast_fp16)[name = tensor("op_45123_cast_fp16")]; + tensor var_45124_split_sizes_0 = const()[name = tensor("op_45124_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_45124_axis_0 = const()[name = tensor("op_45124_axis_0"), val = tensor(1)]; + tensor var_45124_cast_fp16_0, tensor var_45124_cast_fp16_1 = split(axis = var_45124_axis_0, split_sizes = var_45124_split_sizes_0, x = var_45123_cast_fp16)[name = tensor("op_45124_cast_fp16")]; + tensor var_45126_mode_0 = const()[name = tensor("op_45126_mode_0"), val = tensor("EXACT")]; + tensor var_45126_cast_fp16 = gelu(mode = var_45126_mode_0, x = var_45124_cast_fp16_1)[name = tensor("op_45126_cast_fp16")]; + tensor input_583_cast_fp16 = mul(x = var_45124_cast_fp16_0, y = var_45126_cast_fp16)[name = tensor("input_583_cast_fp16")]; + tensor var_45130 = const()[name = tensor("op_45130"), val = tensor([1, 1])]; + tensor var_45132 = const()[name = tensor("op_45132"), val = tensor([1, 1])]; + tensor var_45134_pad_type_0 = const()[name = tensor("op_45134_pad_type_0"), val = tensor("custom")]; + tensor var_45134_pad_0 = const()[name = tensor("op_45134_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_5_ff_net_2_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_5_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3792413440)))]; + tensor up_blocks_0_attentions_1_transformer_blocks_5_ff_net_2_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_5_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3805520704)))]; + tensor var_45134_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_5_ff_net_2_bias_to_fp16, dilations = var_45132, groups = var_30355, pad = var_45134_pad_0, pad_type = var_45134_pad_type_0, strides = var_45130, weight = up_blocks_0_attentions_1_transformer_blocks_5_ff_net_2_weight_to_fp16, x = input_583_cast_fp16)[name = tensor("op_45134_cast_fp16")]; + tensor inputs_301_cast_fp16 = add(x = var_45134_cast_fp16, y = inputs_299_cast_fp16)[name = tensor("inputs_301_cast_fp16")]; + tensor hidden_states_397_axes_0 = const()[name = tensor("hidden_states_397_axes_0"), val = tensor([1])]; + tensor hidden_states_397_gamma_0_to_fp16 = const()[name = tensor("hidden_states_397_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3805523328)))]; + tensor hidden_states_397_beta_0_to_fp16 = const()[name = tensor("hidden_states_397_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3805525952)))]; + tensor var_45150_to_fp16 = const()[name = tensor("op_45150_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_397_cast_fp16 = layer_norm(axes = hidden_states_397_axes_0, beta = hidden_states_397_beta_0_to_fp16, epsilon = var_45150_to_fp16, gamma = hidden_states_397_gamma_0_to_fp16, x = inputs_301_cast_fp16)[name = tensor("hidden_states_397_cast_fp16")]; + tensor var_45165 = const()[name = tensor("op_45165"), val = tensor([1, 1])]; + tensor var_45167 = const()[name = tensor("op_45167"), val = tensor([1, 1])]; + tensor q_201_pad_type_0 = const()[name = tensor("q_201_pad_type_0"), val = tensor("custom")]; + tensor q_201_pad_0 = const()[name = tensor("q_201_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_6_attn1_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_6_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3805528576)))]; + tensor q_201_cast_fp16 = conv(dilations = var_45167, groups = var_30355, pad = q_201_pad_0, pad_type = q_201_pad_type_0, strides = var_45165, weight = up_blocks_0_attentions_1_transformer_blocks_6_attn1_to_q_weight_to_fp16, x = hidden_states_397_cast_fp16)[name = tensor("q_201_cast_fp16")]; + tensor var_45171 = const()[name = tensor("op_45171"), val = tensor([1, 1])]; + tensor var_45173 = const()[name = tensor("op_45173"), val = tensor([1, 1])]; + tensor k_401_pad_type_0 = const()[name = tensor("k_401_pad_type_0"), val = tensor("custom")]; + tensor k_401_pad_0 = const()[name = tensor("k_401_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_6_attn1_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_6_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3808805440)))]; + tensor k_401_cast_fp16 = conv(dilations = var_45173, groups = var_30355, pad = k_401_pad_0, pad_type = k_401_pad_type_0, strides = var_45171, weight = up_blocks_0_attentions_1_transformer_blocks_6_attn1_to_k_weight_to_fp16, x = hidden_states_397_cast_fp16)[name = tensor("k_401_cast_fp16")]; + tensor var_45177 = const()[name = tensor("op_45177"), val = tensor([1, 1])]; + tensor var_45179 = const()[name = tensor("op_45179"), val = tensor([1, 1])]; + tensor v_201_pad_type_0 = const()[name = tensor("v_201_pad_type_0"), val = tensor("custom")]; + tensor v_201_pad_0 = const()[name = tensor("v_201_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_6_attn1_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_6_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3812082304)))]; + tensor v_201_cast_fp16 = conv(dilations = var_45179, groups = var_30355, pad = v_201_pad_0, pad_type = v_201_pad_type_0, strides = var_45177, weight = up_blocks_0_attentions_1_transformer_blocks_6_attn1_to_v_weight_to_fp16, x = hidden_states_397_cast_fp16)[name = tensor("v_201_cast_fp16")]; + tensor var_45183_begin_0 = const()[name = tensor("op_45183_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_45183_end_0 = const()[name = tensor("op_45183_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_45183_end_mask_0 = const()[name = tensor("op_45183_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45183_cast_fp16 = slice_by_index(begin = var_45183_begin_0, end = var_45183_end_0, end_mask = var_45183_end_mask_0, x = q_201_cast_fp16)[name = tensor("op_45183_cast_fp16")]; + tensor var_45187_begin_0 = const()[name = tensor("op_45187_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_45187_end_0 = const()[name = tensor("op_45187_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_45187_end_mask_0 = const()[name = tensor("op_45187_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45187_cast_fp16 = slice_by_index(begin = var_45187_begin_0, end = var_45187_end_0, end_mask = var_45187_end_mask_0, x = q_201_cast_fp16)[name = tensor("op_45187_cast_fp16")]; + tensor var_45191_begin_0 = const()[name = tensor("op_45191_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_45191_end_0 = const()[name = tensor("op_45191_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_45191_end_mask_0 = const()[name = tensor("op_45191_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45191_cast_fp16 = slice_by_index(begin = var_45191_begin_0, end = var_45191_end_0, end_mask = var_45191_end_mask_0, x = q_201_cast_fp16)[name = tensor("op_45191_cast_fp16")]; + tensor var_45195_begin_0 = const()[name = tensor("op_45195_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_45195_end_0 = const()[name = tensor("op_45195_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_45195_end_mask_0 = const()[name = tensor("op_45195_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45195_cast_fp16 = slice_by_index(begin = var_45195_begin_0, end = var_45195_end_0, end_mask = var_45195_end_mask_0, x = q_201_cast_fp16)[name = tensor("op_45195_cast_fp16")]; + tensor var_45199_begin_0 = const()[name = tensor("op_45199_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_45199_end_0 = const()[name = tensor("op_45199_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_45199_end_mask_0 = const()[name = tensor("op_45199_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45199_cast_fp16 = slice_by_index(begin = var_45199_begin_0, end = var_45199_end_0, end_mask = var_45199_end_mask_0, x = q_201_cast_fp16)[name = tensor("op_45199_cast_fp16")]; + tensor var_45203_begin_0 = const()[name = tensor("op_45203_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_45203_end_0 = const()[name = tensor("op_45203_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_45203_end_mask_0 = const()[name = tensor("op_45203_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45203_cast_fp16 = slice_by_index(begin = var_45203_begin_0, end = var_45203_end_0, end_mask = var_45203_end_mask_0, x = q_201_cast_fp16)[name = tensor("op_45203_cast_fp16")]; + tensor var_45207_begin_0 = const()[name = tensor("op_45207_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_45207_end_0 = const()[name = tensor("op_45207_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_45207_end_mask_0 = const()[name = tensor("op_45207_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45207_cast_fp16 = slice_by_index(begin = var_45207_begin_0, end = var_45207_end_0, end_mask = var_45207_end_mask_0, x = q_201_cast_fp16)[name = tensor("op_45207_cast_fp16")]; + tensor var_45211_begin_0 = const()[name = tensor("op_45211_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_45211_end_0 = const()[name = tensor("op_45211_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_45211_end_mask_0 = const()[name = tensor("op_45211_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45211_cast_fp16 = slice_by_index(begin = var_45211_begin_0, end = var_45211_end_0, end_mask = var_45211_end_mask_0, x = q_201_cast_fp16)[name = tensor("op_45211_cast_fp16")]; + tensor var_45215_begin_0 = const()[name = tensor("op_45215_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_45215_end_0 = const()[name = tensor("op_45215_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_45215_end_mask_0 = const()[name = tensor("op_45215_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45215_cast_fp16 = slice_by_index(begin = var_45215_begin_0, end = var_45215_end_0, end_mask = var_45215_end_mask_0, x = q_201_cast_fp16)[name = tensor("op_45215_cast_fp16")]; + tensor var_45219_begin_0 = const()[name = tensor("op_45219_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_45219_end_0 = const()[name = tensor("op_45219_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_45219_end_mask_0 = const()[name = tensor("op_45219_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45219_cast_fp16 = slice_by_index(begin = var_45219_begin_0, end = var_45219_end_0, end_mask = var_45219_end_mask_0, x = q_201_cast_fp16)[name = tensor("op_45219_cast_fp16")]; + tensor var_45223_begin_0 = const()[name = tensor("op_45223_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_45223_end_0 = const()[name = tensor("op_45223_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_45223_end_mask_0 = const()[name = tensor("op_45223_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45223_cast_fp16 = slice_by_index(begin = var_45223_begin_0, end = var_45223_end_0, end_mask = var_45223_end_mask_0, x = q_201_cast_fp16)[name = tensor("op_45223_cast_fp16")]; + tensor var_45227_begin_0 = const()[name = tensor("op_45227_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_45227_end_0 = const()[name = tensor("op_45227_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_45227_end_mask_0 = const()[name = tensor("op_45227_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45227_cast_fp16 = slice_by_index(begin = var_45227_begin_0, end = var_45227_end_0, end_mask = var_45227_end_mask_0, x = q_201_cast_fp16)[name = tensor("op_45227_cast_fp16")]; + tensor var_45231_begin_0 = const()[name = tensor("op_45231_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_45231_end_0 = const()[name = tensor("op_45231_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_45231_end_mask_0 = const()[name = tensor("op_45231_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45231_cast_fp16 = slice_by_index(begin = var_45231_begin_0, end = var_45231_end_0, end_mask = var_45231_end_mask_0, x = q_201_cast_fp16)[name = tensor("op_45231_cast_fp16")]; + tensor var_45235_begin_0 = const()[name = tensor("op_45235_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_45235_end_0 = const()[name = tensor("op_45235_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_45235_end_mask_0 = const()[name = tensor("op_45235_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45235_cast_fp16 = slice_by_index(begin = var_45235_begin_0, end = var_45235_end_0, end_mask = var_45235_end_mask_0, x = q_201_cast_fp16)[name = tensor("op_45235_cast_fp16")]; + tensor var_45239_begin_0 = const()[name = tensor("op_45239_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_45239_end_0 = const()[name = tensor("op_45239_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_45239_end_mask_0 = const()[name = tensor("op_45239_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45239_cast_fp16 = slice_by_index(begin = var_45239_begin_0, end = var_45239_end_0, end_mask = var_45239_end_mask_0, x = q_201_cast_fp16)[name = tensor("op_45239_cast_fp16")]; + tensor var_45243_begin_0 = const()[name = tensor("op_45243_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_45243_end_0 = const()[name = tensor("op_45243_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_45243_end_mask_0 = const()[name = tensor("op_45243_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45243_cast_fp16 = slice_by_index(begin = var_45243_begin_0, end = var_45243_end_0, end_mask = var_45243_end_mask_0, x = q_201_cast_fp16)[name = tensor("op_45243_cast_fp16")]; + tensor var_45247_begin_0 = const()[name = tensor("op_45247_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_45247_end_0 = const()[name = tensor("op_45247_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_45247_end_mask_0 = const()[name = tensor("op_45247_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45247_cast_fp16 = slice_by_index(begin = var_45247_begin_0, end = var_45247_end_0, end_mask = var_45247_end_mask_0, x = q_201_cast_fp16)[name = tensor("op_45247_cast_fp16")]; + tensor var_45251_begin_0 = const()[name = tensor("op_45251_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_45251_end_0 = const()[name = tensor("op_45251_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_45251_end_mask_0 = const()[name = tensor("op_45251_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45251_cast_fp16 = slice_by_index(begin = var_45251_begin_0, end = var_45251_end_0, end_mask = var_45251_end_mask_0, x = q_201_cast_fp16)[name = tensor("op_45251_cast_fp16")]; + tensor var_45255_begin_0 = const()[name = tensor("op_45255_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_45255_end_0 = const()[name = tensor("op_45255_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_45255_end_mask_0 = const()[name = tensor("op_45255_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45255_cast_fp16 = slice_by_index(begin = var_45255_begin_0, end = var_45255_end_0, end_mask = var_45255_end_mask_0, x = q_201_cast_fp16)[name = tensor("op_45255_cast_fp16")]; + tensor var_45259_begin_0 = const()[name = tensor("op_45259_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_45259_end_0 = const()[name = tensor("op_45259_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_45259_end_mask_0 = const()[name = tensor("op_45259_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45259_cast_fp16 = slice_by_index(begin = var_45259_begin_0, end = var_45259_end_0, end_mask = var_45259_end_mask_0, x = q_201_cast_fp16)[name = tensor("op_45259_cast_fp16")]; + tensor k_403_perm_0 = const()[name = tensor("k_403_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_45266_begin_0 = const()[name = tensor("op_45266_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_45266_end_0 = const()[name = tensor("op_45266_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_45266_end_mask_0 = const()[name = tensor("op_45266_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_403_cast_fp16 = transpose(perm = k_403_perm_0, x = k_401_cast_fp16)[name = tensor("transpose_39")]; + tensor var_45266_cast_fp16 = slice_by_index(begin = var_45266_begin_0, end = var_45266_end_0, end_mask = var_45266_end_mask_0, x = k_403_cast_fp16)[name = tensor("op_45266_cast_fp16")]; + tensor var_45270_begin_0 = const()[name = tensor("op_45270_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_45270_end_0 = const()[name = tensor("op_45270_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_45270_end_mask_0 = const()[name = tensor("op_45270_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45270_cast_fp16 = slice_by_index(begin = var_45270_begin_0, end = var_45270_end_0, end_mask = var_45270_end_mask_0, x = k_403_cast_fp16)[name = tensor("op_45270_cast_fp16")]; + tensor var_45274_begin_0 = const()[name = tensor("op_45274_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_45274_end_0 = const()[name = tensor("op_45274_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_45274_end_mask_0 = const()[name = tensor("op_45274_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45274_cast_fp16 = slice_by_index(begin = var_45274_begin_0, end = var_45274_end_0, end_mask = var_45274_end_mask_0, x = k_403_cast_fp16)[name = tensor("op_45274_cast_fp16")]; + tensor var_45278_begin_0 = const()[name = tensor("op_45278_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_45278_end_0 = const()[name = tensor("op_45278_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_45278_end_mask_0 = const()[name = tensor("op_45278_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45278_cast_fp16 = slice_by_index(begin = var_45278_begin_0, end = var_45278_end_0, end_mask = var_45278_end_mask_0, x = k_403_cast_fp16)[name = tensor("op_45278_cast_fp16")]; + tensor var_45282_begin_0 = const()[name = tensor("op_45282_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_45282_end_0 = const()[name = tensor("op_45282_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_45282_end_mask_0 = const()[name = tensor("op_45282_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45282_cast_fp16 = slice_by_index(begin = var_45282_begin_0, end = var_45282_end_0, end_mask = var_45282_end_mask_0, x = k_403_cast_fp16)[name = tensor("op_45282_cast_fp16")]; + tensor var_45286_begin_0 = const()[name = tensor("op_45286_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_45286_end_0 = const()[name = tensor("op_45286_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_45286_end_mask_0 = const()[name = tensor("op_45286_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45286_cast_fp16 = slice_by_index(begin = var_45286_begin_0, end = var_45286_end_0, end_mask = var_45286_end_mask_0, x = k_403_cast_fp16)[name = tensor("op_45286_cast_fp16")]; + tensor var_45290_begin_0 = const()[name = tensor("op_45290_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_45290_end_0 = const()[name = tensor("op_45290_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_45290_end_mask_0 = const()[name = tensor("op_45290_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45290_cast_fp16 = slice_by_index(begin = var_45290_begin_0, end = var_45290_end_0, end_mask = var_45290_end_mask_0, x = k_403_cast_fp16)[name = tensor("op_45290_cast_fp16")]; + tensor var_45294_begin_0 = const()[name = tensor("op_45294_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_45294_end_0 = const()[name = tensor("op_45294_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_45294_end_mask_0 = const()[name = tensor("op_45294_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45294_cast_fp16 = slice_by_index(begin = var_45294_begin_0, end = var_45294_end_0, end_mask = var_45294_end_mask_0, x = k_403_cast_fp16)[name = tensor("op_45294_cast_fp16")]; + tensor var_45298_begin_0 = const()[name = tensor("op_45298_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_45298_end_0 = const()[name = tensor("op_45298_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_45298_end_mask_0 = const()[name = tensor("op_45298_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45298_cast_fp16 = slice_by_index(begin = var_45298_begin_0, end = var_45298_end_0, end_mask = var_45298_end_mask_0, x = k_403_cast_fp16)[name = tensor("op_45298_cast_fp16")]; + tensor var_45302_begin_0 = const()[name = tensor("op_45302_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_45302_end_0 = const()[name = tensor("op_45302_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_45302_end_mask_0 = const()[name = tensor("op_45302_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45302_cast_fp16 = slice_by_index(begin = var_45302_begin_0, end = var_45302_end_0, end_mask = var_45302_end_mask_0, x = k_403_cast_fp16)[name = tensor("op_45302_cast_fp16")]; + tensor var_45306_begin_0 = const()[name = tensor("op_45306_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_45306_end_0 = const()[name = tensor("op_45306_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_45306_end_mask_0 = const()[name = tensor("op_45306_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45306_cast_fp16 = slice_by_index(begin = var_45306_begin_0, end = var_45306_end_0, end_mask = var_45306_end_mask_0, x = k_403_cast_fp16)[name = tensor("op_45306_cast_fp16")]; + tensor var_45310_begin_0 = const()[name = tensor("op_45310_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_45310_end_0 = const()[name = tensor("op_45310_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_45310_end_mask_0 = const()[name = tensor("op_45310_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45310_cast_fp16 = slice_by_index(begin = var_45310_begin_0, end = var_45310_end_0, end_mask = var_45310_end_mask_0, x = k_403_cast_fp16)[name = tensor("op_45310_cast_fp16")]; + tensor var_45314_begin_0 = const()[name = tensor("op_45314_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_45314_end_0 = const()[name = tensor("op_45314_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_45314_end_mask_0 = const()[name = tensor("op_45314_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45314_cast_fp16 = slice_by_index(begin = var_45314_begin_0, end = var_45314_end_0, end_mask = var_45314_end_mask_0, x = k_403_cast_fp16)[name = tensor("op_45314_cast_fp16")]; + tensor var_45318_begin_0 = const()[name = tensor("op_45318_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_45318_end_0 = const()[name = tensor("op_45318_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_45318_end_mask_0 = const()[name = tensor("op_45318_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45318_cast_fp16 = slice_by_index(begin = var_45318_begin_0, end = var_45318_end_0, end_mask = var_45318_end_mask_0, x = k_403_cast_fp16)[name = tensor("op_45318_cast_fp16")]; + tensor var_45322_begin_0 = const()[name = tensor("op_45322_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_45322_end_0 = const()[name = tensor("op_45322_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_45322_end_mask_0 = const()[name = tensor("op_45322_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45322_cast_fp16 = slice_by_index(begin = var_45322_begin_0, end = var_45322_end_0, end_mask = var_45322_end_mask_0, x = k_403_cast_fp16)[name = tensor("op_45322_cast_fp16")]; + tensor var_45326_begin_0 = const()[name = tensor("op_45326_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_45326_end_0 = const()[name = tensor("op_45326_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_45326_end_mask_0 = const()[name = tensor("op_45326_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45326_cast_fp16 = slice_by_index(begin = var_45326_begin_0, end = var_45326_end_0, end_mask = var_45326_end_mask_0, x = k_403_cast_fp16)[name = tensor("op_45326_cast_fp16")]; + tensor var_45330_begin_0 = const()[name = tensor("op_45330_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_45330_end_0 = const()[name = tensor("op_45330_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_45330_end_mask_0 = const()[name = tensor("op_45330_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45330_cast_fp16 = slice_by_index(begin = var_45330_begin_0, end = var_45330_end_0, end_mask = var_45330_end_mask_0, x = k_403_cast_fp16)[name = tensor("op_45330_cast_fp16")]; + tensor var_45334_begin_0 = const()[name = tensor("op_45334_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_45334_end_0 = const()[name = tensor("op_45334_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_45334_end_mask_0 = const()[name = tensor("op_45334_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45334_cast_fp16 = slice_by_index(begin = var_45334_begin_0, end = var_45334_end_0, end_mask = var_45334_end_mask_0, x = k_403_cast_fp16)[name = tensor("op_45334_cast_fp16")]; + tensor var_45338_begin_0 = const()[name = tensor("op_45338_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_45338_end_0 = const()[name = tensor("op_45338_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_45338_end_mask_0 = const()[name = tensor("op_45338_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45338_cast_fp16 = slice_by_index(begin = var_45338_begin_0, end = var_45338_end_0, end_mask = var_45338_end_mask_0, x = k_403_cast_fp16)[name = tensor("op_45338_cast_fp16")]; + tensor var_45342_begin_0 = const()[name = tensor("op_45342_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_45342_end_0 = const()[name = tensor("op_45342_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_45342_end_mask_0 = const()[name = tensor("op_45342_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45342_cast_fp16 = slice_by_index(begin = var_45342_begin_0, end = var_45342_end_0, end_mask = var_45342_end_mask_0, x = k_403_cast_fp16)[name = tensor("op_45342_cast_fp16")]; + tensor var_45344_begin_0 = const()[name = tensor("op_45344_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_45344_end_0 = const()[name = tensor("op_45344_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_45344_end_mask_0 = const()[name = tensor("op_45344_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45344_cast_fp16 = slice_by_index(begin = var_45344_begin_0, end = var_45344_end_0, end_mask = var_45344_end_mask_0, x = v_201_cast_fp16)[name = tensor("op_45344_cast_fp16")]; + tensor var_45348_begin_0 = const()[name = tensor("op_45348_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_45348_end_0 = const()[name = tensor("op_45348_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_45348_end_mask_0 = const()[name = tensor("op_45348_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45348_cast_fp16 = slice_by_index(begin = var_45348_begin_0, end = var_45348_end_0, end_mask = var_45348_end_mask_0, x = v_201_cast_fp16)[name = tensor("op_45348_cast_fp16")]; + tensor var_45352_begin_0 = const()[name = tensor("op_45352_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_45352_end_0 = const()[name = tensor("op_45352_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_45352_end_mask_0 = const()[name = tensor("op_45352_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45352_cast_fp16 = slice_by_index(begin = var_45352_begin_0, end = var_45352_end_0, end_mask = var_45352_end_mask_0, x = v_201_cast_fp16)[name = tensor("op_45352_cast_fp16")]; + tensor var_45356_begin_0 = const()[name = tensor("op_45356_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_45356_end_0 = const()[name = tensor("op_45356_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_45356_end_mask_0 = const()[name = tensor("op_45356_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45356_cast_fp16 = slice_by_index(begin = var_45356_begin_0, end = var_45356_end_0, end_mask = var_45356_end_mask_0, x = v_201_cast_fp16)[name = tensor("op_45356_cast_fp16")]; + tensor var_45360_begin_0 = const()[name = tensor("op_45360_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_45360_end_0 = const()[name = tensor("op_45360_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_45360_end_mask_0 = const()[name = tensor("op_45360_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45360_cast_fp16 = slice_by_index(begin = var_45360_begin_0, end = var_45360_end_0, end_mask = var_45360_end_mask_0, x = v_201_cast_fp16)[name = tensor("op_45360_cast_fp16")]; + tensor var_45364_begin_0 = const()[name = tensor("op_45364_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_45364_end_0 = const()[name = tensor("op_45364_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_45364_end_mask_0 = const()[name = tensor("op_45364_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45364_cast_fp16 = slice_by_index(begin = var_45364_begin_0, end = var_45364_end_0, end_mask = var_45364_end_mask_0, x = v_201_cast_fp16)[name = tensor("op_45364_cast_fp16")]; + tensor var_45368_begin_0 = const()[name = tensor("op_45368_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_45368_end_0 = const()[name = tensor("op_45368_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_45368_end_mask_0 = const()[name = tensor("op_45368_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45368_cast_fp16 = slice_by_index(begin = var_45368_begin_0, end = var_45368_end_0, end_mask = var_45368_end_mask_0, x = v_201_cast_fp16)[name = tensor("op_45368_cast_fp16")]; + tensor var_45372_begin_0 = const()[name = tensor("op_45372_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_45372_end_0 = const()[name = tensor("op_45372_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_45372_end_mask_0 = const()[name = tensor("op_45372_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45372_cast_fp16 = slice_by_index(begin = var_45372_begin_0, end = var_45372_end_0, end_mask = var_45372_end_mask_0, x = v_201_cast_fp16)[name = tensor("op_45372_cast_fp16")]; + tensor var_45376_begin_0 = const()[name = tensor("op_45376_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_45376_end_0 = const()[name = tensor("op_45376_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_45376_end_mask_0 = const()[name = tensor("op_45376_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45376_cast_fp16 = slice_by_index(begin = var_45376_begin_0, end = var_45376_end_0, end_mask = var_45376_end_mask_0, x = v_201_cast_fp16)[name = tensor("op_45376_cast_fp16")]; + tensor var_45380_begin_0 = const()[name = tensor("op_45380_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_45380_end_0 = const()[name = tensor("op_45380_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_45380_end_mask_0 = const()[name = tensor("op_45380_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45380_cast_fp16 = slice_by_index(begin = var_45380_begin_0, end = var_45380_end_0, end_mask = var_45380_end_mask_0, x = v_201_cast_fp16)[name = tensor("op_45380_cast_fp16")]; + tensor var_45384_begin_0 = const()[name = tensor("op_45384_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_45384_end_0 = const()[name = tensor("op_45384_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_45384_end_mask_0 = const()[name = tensor("op_45384_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45384_cast_fp16 = slice_by_index(begin = var_45384_begin_0, end = var_45384_end_0, end_mask = var_45384_end_mask_0, x = v_201_cast_fp16)[name = tensor("op_45384_cast_fp16")]; + tensor var_45388_begin_0 = const()[name = tensor("op_45388_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_45388_end_0 = const()[name = tensor("op_45388_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_45388_end_mask_0 = const()[name = tensor("op_45388_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45388_cast_fp16 = slice_by_index(begin = var_45388_begin_0, end = var_45388_end_0, end_mask = var_45388_end_mask_0, x = v_201_cast_fp16)[name = tensor("op_45388_cast_fp16")]; + tensor var_45392_begin_0 = const()[name = tensor("op_45392_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_45392_end_0 = const()[name = tensor("op_45392_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_45392_end_mask_0 = const()[name = tensor("op_45392_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45392_cast_fp16 = slice_by_index(begin = var_45392_begin_0, end = var_45392_end_0, end_mask = var_45392_end_mask_0, x = v_201_cast_fp16)[name = tensor("op_45392_cast_fp16")]; + tensor var_45396_begin_0 = const()[name = tensor("op_45396_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_45396_end_0 = const()[name = tensor("op_45396_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_45396_end_mask_0 = const()[name = tensor("op_45396_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45396_cast_fp16 = slice_by_index(begin = var_45396_begin_0, end = var_45396_end_0, end_mask = var_45396_end_mask_0, x = v_201_cast_fp16)[name = tensor("op_45396_cast_fp16")]; + tensor var_45400_begin_0 = const()[name = tensor("op_45400_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_45400_end_0 = const()[name = tensor("op_45400_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_45400_end_mask_0 = const()[name = tensor("op_45400_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45400_cast_fp16 = slice_by_index(begin = var_45400_begin_0, end = var_45400_end_0, end_mask = var_45400_end_mask_0, x = v_201_cast_fp16)[name = tensor("op_45400_cast_fp16")]; + tensor var_45404_begin_0 = const()[name = tensor("op_45404_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_45404_end_0 = const()[name = tensor("op_45404_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_45404_end_mask_0 = const()[name = tensor("op_45404_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45404_cast_fp16 = slice_by_index(begin = var_45404_begin_0, end = var_45404_end_0, end_mask = var_45404_end_mask_0, x = v_201_cast_fp16)[name = tensor("op_45404_cast_fp16")]; + tensor var_45408_begin_0 = const()[name = tensor("op_45408_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_45408_end_0 = const()[name = tensor("op_45408_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_45408_end_mask_0 = const()[name = tensor("op_45408_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45408_cast_fp16 = slice_by_index(begin = var_45408_begin_0, end = var_45408_end_0, end_mask = var_45408_end_mask_0, x = v_201_cast_fp16)[name = tensor("op_45408_cast_fp16")]; + tensor var_45412_begin_0 = const()[name = tensor("op_45412_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_45412_end_0 = const()[name = tensor("op_45412_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_45412_end_mask_0 = const()[name = tensor("op_45412_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45412_cast_fp16 = slice_by_index(begin = var_45412_begin_0, end = var_45412_end_0, end_mask = var_45412_end_mask_0, x = v_201_cast_fp16)[name = tensor("op_45412_cast_fp16")]; + tensor var_45416_begin_0 = const()[name = tensor("op_45416_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_45416_end_0 = const()[name = tensor("op_45416_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_45416_end_mask_0 = const()[name = tensor("op_45416_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45416_cast_fp16 = slice_by_index(begin = var_45416_begin_0, end = var_45416_end_0, end_mask = var_45416_end_mask_0, x = v_201_cast_fp16)[name = tensor("op_45416_cast_fp16")]; + tensor var_45420_begin_0 = const()[name = tensor("op_45420_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_45420_end_0 = const()[name = tensor("op_45420_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_45420_end_mask_0 = const()[name = tensor("op_45420_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45420_cast_fp16 = slice_by_index(begin = var_45420_begin_0, end = var_45420_end_0, end_mask = var_45420_end_mask_0, x = v_201_cast_fp16)[name = tensor("op_45420_cast_fp16")]; + tensor var_45424_equation_0 = const()[name = tensor("op_45424_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45424_cast_fp16 = einsum(equation = var_45424_equation_0, values = (var_45266_cast_fp16, var_45183_cast_fp16))[name = tensor("op_45424_cast_fp16")]; + tensor var_45425_to_fp16 = const()[name = tensor("op_45425_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3841_cast_fp16 = mul(x = var_45424_cast_fp16, y = var_45425_to_fp16)[name = tensor("aw_3841_cast_fp16")]; + tensor var_45428_equation_0 = const()[name = tensor("op_45428_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45428_cast_fp16 = einsum(equation = var_45428_equation_0, values = (var_45270_cast_fp16, var_45187_cast_fp16))[name = tensor("op_45428_cast_fp16")]; + tensor var_45429_to_fp16 = const()[name = tensor("op_45429_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3843_cast_fp16 = mul(x = var_45428_cast_fp16, y = var_45429_to_fp16)[name = tensor("aw_3843_cast_fp16")]; + tensor var_45432_equation_0 = const()[name = tensor("op_45432_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45432_cast_fp16 = einsum(equation = var_45432_equation_0, values = (var_45274_cast_fp16, var_45191_cast_fp16))[name = tensor("op_45432_cast_fp16")]; + tensor var_45433_to_fp16 = const()[name = tensor("op_45433_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3845_cast_fp16 = mul(x = var_45432_cast_fp16, y = var_45433_to_fp16)[name = tensor("aw_3845_cast_fp16")]; + tensor var_45436_equation_0 = const()[name = tensor("op_45436_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45436_cast_fp16 = einsum(equation = var_45436_equation_0, values = (var_45278_cast_fp16, var_45195_cast_fp16))[name = tensor("op_45436_cast_fp16")]; + tensor var_45437_to_fp16 = const()[name = tensor("op_45437_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3847_cast_fp16 = mul(x = var_45436_cast_fp16, y = var_45437_to_fp16)[name = tensor("aw_3847_cast_fp16")]; + tensor var_45440_equation_0 = const()[name = tensor("op_45440_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45440_cast_fp16 = einsum(equation = var_45440_equation_0, values = (var_45282_cast_fp16, var_45199_cast_fp16))[name = tensor("op_45440_cast_fp16")]; + tensor var_45441_to_fp16 = const()[name = tensor("op_45441_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3849_cast_fp16 = mul(x = var_45440_cast_fp16, y = var_45441_to_fp16)[name = tensor("aw_3849_cast_fp16")]; + tensor var_45444_equation_0 = const()[name = tensor("op_45444_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45444_cast_fp16 = einsum(equation = var_45444_equation_0, values = (var_45286_cast_fp16, var_45203_cast_fp16))[name = tensor("op_45444_cast_fp16")]; + tensor var_45445_to_fp16 = const()[name = tensor("op_45445_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3851_cast_fp16 = mul(x = var_45444_cast_fp16, y = var_45445_to_fp16)[name = tensor("aw_3851_cast_fp16")]; + tensor var_45448_equation_0 = const()[name = tensor("op_45448_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45448_cast_fp16 = einsum(equation = var_45448_equation_0, values = (var_45290_cast_fp16, var_45207_cast_fp16))[name = tensor("op_45448_cast_fp16")]; + tensor var_45449_to_fp16 = const()[name = tensor("op_45449_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3853_cast_fp16 = mul(x = var_45448_cast_fp16, y = var_45449_to_fp16)[name = tensor("aw_3853_cast_fp16")]; + tensor var_45452_equation_0 = const()[name = tensor("op_45452_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45452_cast_fp16 = einsum(equation = var_45452_equation_0, values = (var_45294_cast_fp16, var_45211_cast_fp16))[name = tensor("op_45452_cast_fp16")]; + tensor var_45453_to_fp16 = const()[name = tensor("op_45453_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3855_cast_fp16 = mul(x = var_45452_cast_fp16, y = var_45453_to_fp16)[name = tensor("aw_3855_cast_fp16")]; + tensor var_45456_equation_0 = const()[name = tensor("op_45456_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45456_cast_fp16 = einsum(equation = var_45456_equation_0, values = (var_45298_cast_fp16, var_45215_cast_fp16))[name = tensor("op_45456_cast_fp16")]; + tensor var_45457_to_fp16 = const()[name = tensor("op_45457_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3857_cast_fp16 = mul(x = var_45456_cast_fp16, y = var_45457_to_fp16)[name = tensor("aw_3857_cast_fp16")]; + tensor var_45460_equation_0 = const()[name = tensor("op_45460_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45460_cast_fp16 = einsum(equation = var_45460_equation_0, values = (var_45302_cast_fp16, var_45219_cast_fp16))[name = tensor("op_45460_cast_fp16")]; + tensor var_45461_to_fp16 = const()[name = tensor("op_45461_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3859_cast_fp16 = mul(x = var_45460_cast_fp16, y = var_45461_to_fp16)[name = tensor("aw_3859_cast_fp16")]; + tensor var_45464_equation_0 = const()[name = tensor("op_45464_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45464_cast_fp16 = einsum(equation = var_45464_equation_0, values = (var_45306_cast_fp16, var_45223_cast_fp16))[name = tensor("op_45464_cast_fp16")]; + tensor var_45465_to_fp16 = const()[name = tensor("op_45465_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3861_cast_fp16 = mul(x = var_45464_cast_fp16, y = var_45465_to_fp16)[name = tensor("aw_3861_cast_fp16")]; + tensor var_45468_equation_0 = const()[name = tensor("op_45468_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45468_cast_fp16 = einsum(equation = var_45468_equation_0, values = (var_45310_cast_fp16, var_45227_cast_fp16))[name = tensor("op_45468_cast_fp16")]; + tensor var_45469_to_fp16 = const()[name = tensor("op_45469_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3863_cast_fp16 = mul(x = var_45468_cast_fp16, y = var_45469_to_fp16)[name = tensor("aw_3863_cast_fp16")]; + tensor var_45472_equation_0 = const()[name = tensor("op_45472_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45472_cast_fp16 = einsum(equation = var_45472_equation_0, values = (var_45314_cast_fp16, var_45231_cast_fp16))[name = tensor("op_45472_cast_fp16")]; + tensor var_45473_to_fp16 = const()[name = tensor("op_45473_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3865_cast_fp16 = mul(x = var_45472_cast_fp16, y = var_45473_to_fp16)[name = tensor("aw_3865_cast_fp16")]; + tensor var_45476_equation_0 = const()[name = tensor("op_45476_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45476_cast_fp16 = einsum(equation = var_45476_equation_0, values = (var_45318_cast_fp16, var_45235_cast_fp16))[name = tensor("op_45476_cast_fp16")]; + tensor var_45477_to_fp16 = const()[name = tensor("op_45477_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3867_cast_fp16 = mul(x = var_45476_cast_fp16, y = var_45477_to_fp16)[name = tensor("aw_3867_cast_fp16")]; + tensor var_45480_equation_0 = const()[name = tensor("op_45480_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45480_cast_fp16 = einsum(equation = var_45480_equation_0, values = (var_45322_cast_fp16, var_45239_cast_fp16))[name = tensor("op_45480_cast_fp16")]; + tensor var_45481_to_fp16 = const()[name = tensor("op_45481_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3869_cast_fp16 = mul(x = var_45480_cast_fp16, y = var_45481_to_fp16)[name = tensor("aw_3869_cast_fp16")]; + tensor var_45484_equation_0 = const()[name = tensor("op_45484_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45484_cast_fp16 = einsum(equation = var_45484_equation_0, values = (var_45326_cast_fp16, var_45243_cast_fp16))[name = tensor("op_45484_cast_fp16")]; + tensor var_45485_to_fp16 = const()[name = tensor("op_45485_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3871_cast_fp16 = mul(x = var_45484_cast_fp16, y = var_45485_to_fp16)[name = tensor("aw_3871_cast_fp16")]; + tensor var_45488_equation_0 = const()[name = tensor("op_45488_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45488_cast_fp16 = einsum(equation = var_45488_equation_0, values = (var_45330_cast_fp16, var_45247_cast_fp16))[name = tensor("op_45488_cast_fp16")]; + tensor var_45489_to_fp16 = const()[name = tensor("op_45489_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3873_cast_fp16 = mul(x = var_45488_cast_fp16, y = var_45489_to_fp16)[name = tensor("aw_3873_cast_fp16")]; + tensor var_45492_equation_0 = const()[name = tensor("op_45492_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45492_cast_fp16 = einsum(equation = var_45492_equation_0, values = (var_45334_cast_fp16, var_45251_cast_fp16))[name = tensor("op_45492_cast_fp16")]; + tensor var_45493_to_fp16 = const()[name = tensor("op_45493_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3875_cast_fp16 = mul(x = var_45492_cast_fp16, y = var_45493_to_fp16)[name = tensor("aw_3875_cast_fp16")]; + tensor var_45496_equation_0 = const()[name = tensor("op_45496_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45496_cast_fp16 = einsum(equation = var_45496_equation_0, values = (var_45338_cast_fp16, var_45255_cast_fp16))[name = tensor("op_45496_cast_fp16")]; + tensor var_45497_to_fp16 = const()[name = tensor("op_45497_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3877_cast_fp16 = mul(x = var_45496_cast_fp16, y = var_45497_to_fp16)[name = tensor("aw_3877_cast_fp16")]; + tensor var_45500_equation_0 = const()[name = tensor("op_45500_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45500_cast_fp16 = einsum(equation = var_45500_equation_0, values = (var_45342_cast_fp16, var_45259_cast_fp16))[name = tensor("op_45500_cast_fp16")]; + tensor var_45501_to_fp16 = const()[name = tensor("op_45501_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3879_cast_fp16 = mul(x = var_45500_cast_fp16, y = var_45501_to_fp16)[name = tensor("aw_3879_cast_fp16")]; + tensor var_45503_cast_fp16 = softmax(axis = var_30355, x = aw_3841_cast_fp16)[name = tensor("op_45503_cast_fp16")]; + tensor var_45504_cast_fp16 = softmax(axis = var_30355, x = aw_3843_cast_fp16)[name = tensor("op_45504_cast_fp16")]; + tensor var_45505_cast_fp16 = softmax(axis = var_30355, x = aw_3845_cast_fp16)[name = tensor("op_45505_cast_fp16")]; + tensor var_45506_cast_fp16 = softmax(axis = var_30355, x = aw_3847_cast_fp16)[name = tensor("op_45506_cast_fp16")]; + tensor var_45507_cast_fp16 = softmax(axis = var_30355, x = aw_3849_cast_fp16)[name = tensor("op_45507_cast_fp16")]; + tensor var_45508_cast_fp16 = softmax(axis = var_30355, x = aw_3851_cast_fp16)[name = tensor("op_45508_cast_fp16")]; + tensor var_45509_cast_fp16 = softmax(axis = var_30355, x = aw_3853_cast_fp16)[name = tensor("op_45509_cast_fp16")]; + tensor var_45510_cast_fp16 = softmax(axis = var_30355, x = aw_3855_cast_fp16)[name = tensor("op_45510_cast_fp16")]; + tensor var_45511_cast_fp16 = softmax(axis = var_30355, x = aw_3857_cast_fp16)[name = tensor("op_45511_cast_fp16")]; + tensor var_45512_cast_fp16 = softmax(axis = var_30355, x = aw_3859_cast_fp16)[name = tensor("op_45512_cast_fp16")]; + tensor var_45513_cast_fp16 = softmax(axis = var_30355, x = aw_3861_cast_fp16)[name = tensor("op_45513_cast_fp16")]; + tensor var_45514_cast_fp16 = softmax(axis = var_30355, x = aw_3863_cast_fp16)[name = tensor("op_45514_cast_fp16")]; + tensor var_45515_cast_fp16 = softmax(axis = var_30355, x = aw_3865_cast_fp16)[name = tensor("op_45515_cast_fp16")]; + tensor var_45516_cast_fp16 = softmax(axis = var_30355, x = aw_3867_cast_fp16)[name = tensor("op_45516_cast_fp16")]; + tensor var_45517_cast_fp16 = softmax(axis = var_30355, x = aw_3869_cast_fp16)[name = tensor("op_45517_cast_fp16")]; + tensor var_45518_cast_fp16 = softmax(axis = var_30355, x = aw_3871_cast_fp16)[name = tensor("op_45518_cast_fp16")]; + tensor var_45519_cast_fp16 = softmax(axis = var_30355, x = aw_3873_cast_fp16)[name = tensor("op_45519_cast_fp16")]; + tensor var_45520_cast_fp16 = softmax(axis = var_30355, x = aw_3875_cast_fp16)[name = tensor("op_45520_cast_fp16")]; + tensor var_45521_cast_fp16 = softmax(axis = var_30355, x = aw_3877_cast_fp16)[name = tensor("op_45521_cast_fp16")]; + tensor var_45522_cast_fp16 = softmax(axis = var_30355, x = aw_3879_cast_fp16)[name = tensor("op_45522_cast_fp16")]; + tensor var_45524_equation_0 = const()[name = tensor("op_45524_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45524_cast_fp16 = einsum(equation = var_45524_equation_0, values = (var_45344_cast_fp16, var_45503_cast_fp16))[name = tensor("op_45524_cast_fp16")]; + tensor var_45526_equation_0 = const()[name = tensor("op_45526_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45526_cast_fp16 = einsum(equation = var_45526_equation_0, values = (var_45348_cast_fp16, var_45504_cast_fp16))[name = tensor("op_45526_cast_fp16")]; + tensor var_45528_equation_0 = const()[name = tensor("op_45528_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45528_cast_fp16 = einsum(equation = var_45528_equation_0, values = (var_45352_cast_fp16, var_45505_cast_fp16))[name = tensor("op_45528_cast_fp16")]; + tensor var_45530_equation_0 = const()[name = tensor("op_45530_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45530_cast_fp16 = einsum(equation = var_45530_equation_0, values = (var_45356_cast_fp16, var_45506_cast_fp16))[name = tensor("op_45530_cast_fp16")]; + tensor var_45532_equation_0 = const()[name = tensor("op_45532_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45532_cast_fp16 = einsum(equation = var_45532_equation_0, values = (var_45360_cast_fp16, var_45507_cast_fp16))[name = tensor("op_45532_cast_fp16")]; + tensor var_45534_equation_0 = const()[name = tensor("op_45534_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45534_cast_fp16 = einsum(equation = var_45534_equation_0, values = (var_45364_cast_fp16, var_45508_cast_fp16))[name = tensor("op_45534_cast_fp16")]; + tensor var_45536_equation_0 = const()[name = tensor("op_45536_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45536_cast_fp16 = einsum(equation = var_45536_equation_0, values = (var_45368_cast_fp16, var_45509_cast_fp16))[name = tensor("op_45536_cast_fp16")]; + tensor var_45538_equation_0 = const()[name = tensor("op_45538_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45538_cast_fp16 = einsum(equation = var_45538_equation_0, values = (var_45372_cast_fp16, var_45510_cast_fp16))[name = tensor("op_45538_cast_fp16")]; + tensor var_45540_equation_0 = const()[name = tensor("op_45540_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45540_cast_fp16 = einsum(equation = var_45540_equation_0, values = (var_45376_cast_fp16, var_45511_cast_fp16))[name = tensor("op_45540_cast_fp16")]; + tensor var_45542_equation_0 = const()[name = tensor("op_45542_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45542_cast_fp16 = einsum(equation = var_45542_equation_0, values = (var_45380_cast_fp16, var_45512_cast_fp16))[name = tensor("op_45542_cast_fp16")]; + tensor var_45544_equation_0 = const()[name = tensor("op_45544_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45544_cast_fp16 = einsum(equation = var_45544_equation_0, values = (var_45384_cast_fp16, var_45513_cast_fp16))[name = tensor("op_45544_cast_fp16")]; + tensor var_45546_equation_0 = const()[name = tensor("op_45546_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45546_cast_fp16 = einsum(equation = var_45546_equation_0, values = (var_45388_cast_fp16, var_45514_cast_fp16))[name = tensor("op_45546_cast_fp16")]; + tensor var_45548_equation_0 = const()[name = tensor("op_45548_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45548_cast_fp16 = einsum(equation = var_45548_equation_0, values = (var_45392_cast_fp16, var_45515_cast_fp16))[name = tensor("op_45548_cast_fp16")]; + tensor var_45550_equation_0 = const()[name = tensor("op_45550_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45550_cast_fp16 = einsum(equation = var_45550_equation_0, values = (var_45396_cast_fp16, var_45516_cast_fp16))[name = tensor("op_45550_cast_fp16")]; + tensor var_45552_equation_0 = const()[name = tensor("op_45552_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45552_cast_fp16 = einsum(equation = var_45552_equation_0, values = (var_45400_cast_fp16, var_45517_cast_fp16))[name = tensor("op_45552_cast_fp16")]; + tensor var_45554_equation_0 = const()[name = tensor("op_45554_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45554_cast_fp16 = einsum(equation = var_45554_equation_0, values = (var_45404_cast_fp16, var_45518_cast_fp16))[name = tensor("op_45554_cast_fp16")]; + tensor var_45556_equation_0 = const()[name = tensor("op_45556_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45556_cast_fp16 = einsum(equation = var_45556_equation_0, values = (var_45408_cast_fp16, var_45519_cast_fp16))[name = tensor("op_45556_cast_fp16")]; + tensor var_45558_equation_0 = const()[name = tensor("op_45558_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45558_cast_fp16 = einsum(equation = var_45558_equation_0, values = (var_45412_cast_fp16, var_45520_cast_fp16))[name = tensor("op_45558_cast_fp16")]; + tensor var_45560_equation_0 = const()[name = tensor("op_45560_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45560_cast_fp16 = einsum(equation = var_45560_equation_0, values = (var_45416_cast_fp16, var_45521_cast_fp16))[name = tensor("op_45560_cast_fp16")]; + tensor var_45562_equation_0 = const()[name = tensor("op_45562_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45562_cast_fp16 = einsum(equation = var_45562_equation_0, values = (var_45420_cast_fp16, var_45522_cast_fp16))[name = tensor("op_45562_cast_fp16")]; + tensor input_585_interleave_0 = const()[name = tensor("input_585_interleave_0"), val = tensor(false)]; + tensor input_585_cast_fp16 = concat(axis = var_30355, interleave = input_585_interleave_0, values = (var_45524_cast_fp16, var_45526_cast_fp16, var_45528_cast_fp16, var_45530_cast_fp16, var_45532_cast_fp16, var_45534_cast_fp16, var_45536_cast_fp16, var_45538_cast_fp16, var_45540_cast_fp16, var_45542_cast_fp16, var_45544_cast_fp16, var_45546_cast_fp16, var_45548_cast_fp16, var_45550_cast_fp16, var_45552_cast_fp16, var_45554_cast_fp16, var_45556_cast_fp16, var_45558_cast_fp16, var_45560_cast_fp16, var_45562_cast_fp16))[name = tensor("input_585_cast_fp16")]; + tensor var_45568 = const()[name = tensor("op_45568"), val = tensor([1, 1])]; + tensor var_45570 = const()[name = tensor("op_45570"), val = tensor([1, 1])]; + tensor var_45572_pad_type_0 = const()[name = tensor("op_45572_pad_type_0"), val = tensor("custom")]; + tensor var_45572_pad_0 = const()[name = tensor("op_45572_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_6_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_6_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3815359168)))]; + tensor up_blocks_0_attentions_1_transformer_blocks_6_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_6_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3818636032)))]; + tensor var_45572_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_6_attn1_to_out_0_bias_to_fp16, dilations = var_45570, groups = var_30355, pad = var_45572_pad_0, pad_type = var_45572_pad_type_0, strides = var_45568, weight = up_blocks_0_attentions_1_transformer_blocks_6_attn1_to_out_0_weight_to_fp16, x = input_585_cast_fp16)[name = tensor("op_45572_cast_fp16")]; + tensor inputs_303_cast_fp16 = add(x = var_45572_cast_fp16, y = inputs_301_cast_fp16)[name = tensor("inputs_303_cast_fp16")]; + tensor hidden_states_399_axes_0 = const()[name = tensor("hidden_states_399_axes_0"), val = tensor([1])]; + tensor hidden_states_399_gamma_0_to_fp16 = const()[name = tensor("hidden_states_399_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3818638656)))]; + tensor hidden_states_399_beta_0_to_fp16 = const()[name = tensor("hidden_states_399_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3818641280)))]; + tensor var_45582_to_fp16 = const()[name = tensor("op_45582_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_399_cast_fp16 = layer_norm(axes = hidden_states_399_axes_0, beta = hidden_states_399_beta_0_to_fp16, epsilon = var_45582_to_fp16, gamma = hidden_states_399_gamma_0_to_fp16, x = inputs_303_cast_fp16)[name = tensor("hidden_states_399_cast_fp16")]; + tensor var_45597 = const()[name = tensor("op_45597"), val = tensor([1, 1])]; + tensor var_45599 = const()[name = tensor("op_45599"), val = tensor([1, 1])]; + tensor q_203_pad_type_0 = const()[name = tensor("q_203_pad_type_0"), val = tensor("custom")]; + tensor q_203_pad_0 = const()[name = tensor("q_203_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_6_attn2_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_6_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3818643904)))]; + tensor q_203_cast_fp16 = conv(dilations = var_45599, groups = var_30355, pad = q_203_pad_0, pad_type = q_203_pad_type_0, strides = var_45597, weight = up_blocks_0_attentions_1_transformer_blocks_6_attn2_to_q_weight_to_fp16, x = hidden_states_399_cast_fp16)[name = tensor("q_203_cast_fp16")]; + tensor var_45603 = const()[name = tensor("op_45603"), val = tensor([1, 1])]; + tensor var_45605 = const()[name = tensor("op_45605"), val = tensor([1, 1])]; + tensor k_405_pad_type_0 = const()[name = tensor("k_405_pad_type_0"), val = tensor("custom")]; + tensor k_405_pad_0 = const()[name = tensor("k_405_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_6_attn2_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_6_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3821920768)))]; + tensor k_405_cast_fp16 = conv(dilations = var_45605, groups = var_30355, pad = k_405_pad_0, pad_type = k_405_pad_type_0, strides = var_45603, weight = up_blocks_0_attentions_1_transformer_blocks_6_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_405_cast_fp16")]; + tensor var_45609 = const()[name = tensor("op_45609"), val = tensor([1, 1])]; + tensor var_45611 = const()[name = tensor("op_45611"), val = tensor([1, 1])]; + tensor v_203_pad_type_0 = const()[name = tensor("v_203_pad_type_0"), val = tensor("custom")]; + tensor v_203_pad_0 = const()[name = tensor("v_203_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_6_attn2_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_6_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3827163712)))]; + tensor v_203_cast_fp16 = conv(dilations = var_45611, groups = var_30355, pad = v_203_pad_0, pad_type = v_203_pad_type_0, strides = var_45609, weight = up_blocks_0_attentions_1_transformer_blocks_6_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_203_cast_fp16")]; + tensor var_45615_begin_0 = const()[name = tensor("op_45615_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_45615_end_0 = const()[name = tensor("op_45615_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_45615_end_mask_0 = const()[name = tensor("op_45615_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45615_cast_fp16 = slice_by_index(begin = var_45615_begin_0, end = var_45615_end_0, end_mask = var_45615_end_mask_0, x = q_203_cast_fp16)[name = tensor("op_45615_cast_fp16")]; + tensor var_45619_begin_0 = const()[name = tensor("op_45619_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_45619_end_0 = const()[name = tensor("op_45619_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_45619_end_mask_0 = const()[name = tensor("op_45619_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45619_cast_fp16 = slice_by_index(begin = var_45619_begin_0, end = var_45619_end_0, end_mask = var_45619_end_mask_0, x = q_203_cast_fp16)[name = tensor("op_45619_cast_fp16")]; + tensor var_45623_begin_0 = const()[name = tensor("op_45623_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_45623_end_0 = const()[name = tensor("op_45623_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_45623_end_mask_0 = const()[name = tensor("op_45623_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45623_cast_fp16 = slice_by_index(begin = var_45623_begin_0, end = var_45623_end_0, end_mask = var_45623_end_mask_0, x = q_203_cast_fp16)[name = tensor("op_45623_cast_fp16")]; + tensor var_45627_begin_0 = const()[name = tensor("op_45627_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_45627_end_0 = const()[name = tensor("op_45627_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_45627_end_mask_0 = const()[name = tensor("op_45627_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45627_cast_fp16 = slice_by_index(begin = var_45627_begin_0, end = var_45627_end_0, end_mask = var_45627_end_mask_0, x = q_203_cast_fp16)[name = tensor("op_45627_cast_fp16")]; + tensor var_45631_begin_0 = const()[name = tensor("op_45631_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_45631_end_0 = const()[name = tensor("op_45631_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_45631_end_mask_0 = const()[name = tensor("op_45631_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45631_cast_fp16 = slice_by_index(begin = var_45631_begin_0, end = var_45631_end_0, end_mask = var_45631_end_mask_0, x = q_203_cast_fp16)[name = tensor("op_45631_cast_fp16")]; + tensor var_45635_begin_0 = const()[name = tensor("op_45635_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_45635_end_0 = const()[name = tensor("op_45635_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_45635_end_mask_0 = const()[name = tensor("op_45635_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45635_cast_fp16 = slice_by_index(begin = var_45635_begin_0, end = var_45635_end_0, end_mask = var_45635_end_mask_0, x = q_203_cast_fp16)[name = tensor("op_45635_cast_fp16")]; + tensor var_45639_begin_0 = const()[name = tensor("op_45639_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_45639_end_0 = const()[name = tensor("op_45639_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_45639_end_mask_0 = const()[name = tensor("op_45639_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45639_cast_fp16 = slice_by_index(begin = var_45639_begin_0, end = var_45639_end_0, end_mask = var_45639_end_mask_0, x = q_203_cast_fp16)[name = tensor("op_45639_cast_fp16")]; + tensor var_45643_begin_0 = const()[name = tensor("op_45643_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_45643_end_0 = const()[name = tensor("op_45643_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_45643_end_mask_0 = const()[name = tensor("op_45643_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45643_cast_fp16 = slice_by_index(begin = var_45643_begin_0, end = var_45643_end_0, end_mask = var_45643_end_mask_0, x = q_203_cast_fp16)[name = tensor("op_45643_cast_fp16")]; + tensor var_45647_begin_0 = const()[name = tensor("op_45647_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_45647_end_0 = const()[name = tensor("op_45647_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_45647_end_mask_0 = const()[name = tensor("op_45647_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45647_cast_fp16 = slice_by_index(begin = var_45647_begin_0, end = var_45647_end_0, end_mask = var_45647_end_mask_0, x = q_203_cast_fp16)[name = tensor("op_45647_cast_fp16")]; + tensor var_45651_begin_0 = const()[name = tensor("op_45651_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_45651_end_0 = const()[name = tensor("op_45651_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_45651_end_mask_0 = const()[name = tensor("op_45651_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45651_cast_fp16 = slice_by_index(begin = var_45651_begin_0, end = var_45651_end_0, end_mask = var_45651_end_mask_0, x = q_203_cast_fp16)[name = tensor("op_45651_cast_fp16")]; + tensor var_45655_begin_0 = const()[name = tensor("op_45655_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_45655_end_0 = const()[name = tensor("op_45655_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_45655_end_mask_0 = const()[name = tensor("op_45655_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45655_cast_fp16 = slice_by_index(begin = var_45655_begin_0, end = var_45655_end_0, end_mask = var_45655_end_mask_0, x = q_203_cast_fp16)[name = tensor("op_45655_cast_fp16")]; + tensor var_45659_begin_0 = const()[name = tensor("op_45659_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_45659_end_0 = const()[name = tensor("op_45659_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_45659_end_mask_0 = const()[name = tensor("op_45659_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45659_cast_fp16 = slice_by_index(begin = var_45659_begin_0, end = var_45659_end_0, end_mask = var_45659_end_mask_0, x = q_203_cast_fp16)[name = tensor("op_45659_cast_fp16")]; + tensor var_45663_begin_0 = const()[name = tensor("op_45663_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_45663_end_0 = const()[name = tensor("op_45663_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_45663_end_mask_0 = const()[name = tensor("op_45663_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45663_cast_fp16 = slice_by_index(begin = var_45663_begin_0, end = var_45663_end_0, end_mask = var_45663_end_mask_0, x = q_203_cast_fp16)[name = tensor("op_45663_cast_fp16")]; + tensor var_45667_begin_0 = const()[name = tensor("op_45667_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_45667_end_0 = const()[name = tensor("op_45667_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_45667_end_mask_0 = const()[name = tensor("op_45667_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45667_cast_fp16 = slice_by_index(begin = var_45667_begin_0, end = var_45667_end_0, end_mask = var_45667_end_mask_0, x = q_203_cast_fp16)[name = tensor("op_45667_cast_fp16")]; + tensor var_45671_begin_0 = const()[name = tensor("op_45671_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_45671_end_0 = const()[name = tensor("op_45671_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_45671_end_mask_0 = const()[name = tensor("op_45671_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45671_cast_fp16 = slice_by_index(begin = var_45671_begin_0, end = var_45671_end_0, end_mask = var_45671_end_mask_0, x = q_203_cast_fp16)[name = tensor("op_45671_cast_fp16")]; + tensor var_45675_begin_0 = const()[name = tensor("op_45675_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_45675_end_0 = const()[name = tensor("op_45675_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_45675_end_mask_0 = const()[name = tensor("op_45675_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45675_cast_fp16 = slice_by_index(begin = var_45675_begin_0, end = var_45675_end_0, end_mask = var_45675_end_mask_0, x = q_203_cast_fp16)[name = tensor("op_45675_cast_fp16")]; + tensor var_45679_begin_0 = const()[name = tensor("op_45679_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_45679_end_0 = const()[name = tensor("op_45679_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_45679_end_mask_0 = const()[name = tensor("op_45679_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45679_cast_fp16 = slice_by_index(begin = var_45679_begin_0, end = var_45679_end_0, end_mask = var_45679_end_mask_0, x = q_203_cast_fp16)[name = tensor("op_45679_cast_fp16")]; + tensor var_45683_begin_0 = const()[name = tensor("op_45683_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_45683_end_0 = const()[name = tensor("op_45683_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_45683_end_mask_0 = const()[name = tensor("op_45683_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45683_cast_fp16 = slice_by_index(begin = var_45683_begin_0, end = var_45683_end_0, end_mask = var_45683_end_mask_0, x = q_203_cast_fp16)[name = tensor("op_45683_cast_fp16")]; + tensor var_45687_begin_0 = const()[name = tensor("op_45687_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_45687_end_0 = const()[name = tensor("op_45687_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_45687_end_mask_0 = const()[name = tensor("op_45687_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45687_cast_fp16 = slice_by_index(begin = var_45687_begin_0, end = var_45687_end_0, end_mask = var_45687_end_mask_0, x = q_203_cast_fp16)[name = tensor("op_45687_cast_fp16")]; + tensor var_45691_begin_0 = const()[name = tensor("op_45691_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_45691_end_0 = const()[name = tensor("op_45691_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_45691_end_mask_0 = const()[name = tensor("op_45691_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45691_cast_fp16 = slice_by_index(begin = var_45691_begin_0, end = var_45691_end_0, end_mask = var_45691_end_mask_0, x = q_203_cast_fp16)[name = tensor("op_45691_cast_fp16")]; + tensor k_407_perm_0 = const()[name = tensor("k_407_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_45698_begin_0 = const()[name = tensor("op_45698_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_45698_end_0 = const()[name = tensor("op_45698_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_45698_end_mask_0 = const()[name = tensor("op_45698_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_407_cast_fp16 = transpose(perm = k_407_perm_0, x = k_405_cast_fp16)[name = tensor("transpose_38")]; + tensor var_45698_cast_fp16 = slice_by_index(begin = var_45698_begin_0, end = var_45698_end_0, end_mask = var_45698_end_mask_0, x = k_407_cast_fp16)[name = tensor("op_45698_cast_fp16")]; + tensor var_45702_begin_0 = const()[name = tensor("op_45702_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_45702_end_0 = const()[name = tensor("op_45702_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_45702_end_mask_0 = const()[name = tensor("op_45702_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45702_cast_fp16 = slice_by_index(begin = var_45702_begin_0, end = var_45702_end_0, end_mask = var_45702_end_mask_0, x = k_407_cast_fp16)[name = tensor("op_45702_cast_fp16")]; + tensor var_45706_begin_0 = const()[name = tensor("op_45706_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_45706_end_0 = const()[name = tensor("op_45706_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_45706_end_mask_0 = const()[name = tensor("op_45706_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45706_cast_fp16 = slice_by_index(begin = var_45706_begin_0, end = var_45706_end_0, end_mask = var_45706_end_mask_0, x = k_407_cast_fp16)[name = tensor("op_45706_cast_fp16")]; + tensor var_45710_begin_0 = const()[name = tensor("op_45710_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_45710_end_0 = const()[name = tensor("op_45710_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_45710_end_mask_0 = const()[name = tensor("op_45710_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45710_cast_fp16 = slice_by_index(begin = var_45710_begin_0, end = var_45710_end_0, end_mask = var_45710_end_mask_0, x = k_407_cast_fp16)[name = tensor("op_45710_cast_fp16")]; + tensor var_45714_begin_0 = const()[name = tensor("op_45714_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_45714_end_0 = const()[name = tensor("op_45714_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_45714_end_mask_0 = const()[name = tensor("op_45714_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45714_cast_fp16 = slice_by_index(begin = var_45714_begin_0, end = var_45714_end_0, end_mask = var_45714_end_mask_0, x = k_407_cast_fp16)[name = tensor("op_45714_cast_fp16")]; + tensor var_45718_begin_0 = const()[name = tensor("op_45718_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_45718_end_0 = const()[name = tensor("op_45718_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_45718_end_mask_0 = const()[name = tensor("op_45718_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45718_cast_fp16 = slice_by_index(begin = var_45718_begin_0, end = var_45718_end_0, end_mask = var_45718_end_mask_0, x = k_407_cast_fp16)[name = tensor("op_45718_cast_fp16")]; + tensor var_45722_begin_0 = const()[name = tensor("op_45722_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_45722_end_0 = const()[name = tensor("op_45722_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_45722_end_mask_0 = const()[name = tensor("op_45722_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45722_cast_fp16 = slice_by_index(begin = var_45722_begin_0, end = var_45722_end_0, end_mask = var_45722_end_mask_0, x = k_407_cast_fp16)[name = tensor("op_45722_cast_fp16")]; + tensor var_45726_begin_0 = const()[name = tensor("op_45726_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_45726_end_0 = const()[name = tensor("op_45726_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_45726_end_mask_0 = const()[name = tensor("op_45726_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45726_cast_fp16 = slice_by_index(begin = var_45726_begin_0, end = var_45726_end_0, end_mask = var_45726_end_mask_0, x = k_407_cast_fp16)[name = tensor("op_45726_cast_fp16")]; + tensor var_45730_begin_0 = const()[name = tensor("op_45730_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_45730_end_0 = const()[name = tensor("op_45730_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_45730_end_mask_0 = const()[name = tensor("op_45730_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45730_cast_fp16 = slice_by_index(begin = var_45730_begin_0, end = var_45730_end_0, end_mask = var_45730_end_mask_0, x = k_407_cast_fp16)[name = tensor("op_45730_cast_fp16")]; + tensor var_45734_begin_0 = const()[name = tensor("op_45734_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_45734_end_0 = const()[name = tensor("op_45734_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_45734_end_mask_0 = const()[name = tensor("op_45734_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45734_cast_fp16 = slice_by_index(begin = var_45734_begin_0, end = var_45734_end_0, end_mask = var_45734_end_mask_0, x = k_407_cast_fp16)[name = tensor("op_45734_cast_fp16")]; + tensor var_45738_begin_0 = const()[name = tensor("op_45738_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_45738_end_0 = const()[name = tensor("op_45738_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_45738_end_mask_0 = const()[name = tensor("op_45738_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45738_cast_fp16 = slice_by_index(begin = var_45738_begin_0, end = var_45738_end_0, end_mask = var_45738_end_mask_0, x = k_407_cast_fp16)[name = tensor("op_45738_cast_fp16")]; + tensor var_45742_begin_0 = const()[name = tensor("op_45742_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_45742_end_0 = const()[name = tensor("op_45742_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_45742_end_mask_0 = const()[name = tensor("op_45742_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45742_cast_fp16 = slice_by_index(begin = var_45742_begin_0, end = var_45742_end_0, end_mask = var_45742_end_mask_0, x = k_407_cast_fp16)[name = tensor("op_45742_cast_fp16")]; + tensor var_45746_begin_0 = const()[name = tensor("op_45746_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_45746_end_0 = const()[name = tensor("op_45746_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_45746_end_mask_0 = const()[name = tensor("op_45746_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45746_cast_fp16 = slice_by_index(begin = var_45746_begin_0, end = var_45746_end_0, end_mask = var_45746_end_mask_0, x = k_407_cast_fp16)[name = tensor("op_45746_cast_fp16")]; + tensor var_45750_begin_0 = const()[name = tensor("op_45750_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_45750_end_0 = const()[name = tensor("op_45750_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_45750_end_mask_0 = const()[name = tensor("op_45750_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45750_cast_fp16 = slice_by_index(begin = var_45750_begin_0, end = var_45750_end_0, end_mask = var_45750_end_mask_0, x = k_407_cast_fp16)[name = tensor("op_45750_cast_fp16")]; + tensor var_45754_begin_0 = const()[name = tensor("op_45754_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_45754_end_0 = const()[name = tensor("op_45754_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_45754_end_mask_0 = const()[name = tensor("op_45754_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45754_cast_fp16 = slice_by_index(begin = var_45754_begin_0, end = var_45754_end_0, end_mask = var_45754_end_mask_0, x = k_407_cast_fp16)[name = tensor("op_45754_cast_fp16")]; + tensor var_45758_begin_0 = const()[name = tensor("op_45758_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_45758_end_0 = const()[name = tensor("op_45758_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_45758_end_mask_0 = const()[name = tensor("op_45758_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45758_cast_fp16 = slice_by_index(begin = var_45758_begin_0, end = var_45758_end_0, end_mask = var_45758_end_mask_0, x = k_407_cast_fp16)[name = tensor("op_45758_cast_fp16")]; + tensor var_45762_begin_0 = const()[name = tensor("op_45762_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_45762_end_0 = const()[name = tensor("op_45762_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_45762_end_mask_0 = const()[name = tensor("op_45762_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45762_cast_fp16 = slice_by_index(begin = var_45762_begin_0, end = var_45762_end_0, end_mask = var_45762_end_mask_0, x = k_407_cast_fp16)[name = tensor("op_45762_cast_fp16")]; + tensor var_45766_begin_0 = const()[name = tensor("op_45766_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_45766_end_0 = const()[name = tensor("op_45766_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_45766_end_mask_0 = const()[name = tensor("op_45766_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45766_cast_fp16 = slice_by_index(begin = var_45766_begin_0, end = var_45766_end_0, end_mask = var_45766_end_mask_0, x = k_407_cast_fp16)[name = tensor("op_45766_cast_fp16")]; + tensor var_45770_begin_0 = const()[name = tensor("op_45770_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_45770_end_0 = const()[name = tensor("op_45770_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_45770_end_mask_0 = const()[name = tensor("op_45770_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45770_cast_fp16 = slice_by_index(begin = var_45770_begin_0, end = var_45770_end_0, end_mask = var_45770_end_mask_0, x = k_407_cast_fp16)[name = tensor("op_45770_cast_fp16")]; + tensor var_45774_begin_0 = const()[name = tensor("op_45774_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_45774_end_0 = const()[name = tensor("op_45774_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_45774_end_mask_0 = const()[name = tensor("op_45774_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45774_cast_fp16 = slice_by_index(begin = var_45774_begin_0, end = var_45774_end_0, end_mask = var_45774_end_mask_0, x = k_407_cast_fp16)[name = tensor("op_45774_cast_fp16")]; + tensor var_45776_begin_0 = const()[name = tensor("op_45776_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_45776_end_0 = const()[name = tensor("op_45776_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_45776_end_mask_0 = const()[name = tensor("op_45776_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45776_cast_fp16 = slice_by_index(begin = var_45776_begin_0, end = var_45776_end_0, end_mask = var_45776_end_mask_0, x = v_203_cast_fp16)[name = tensor("op_45776_cast_fp16")]; + tensor var_45780_begin_0 = const()[name = tensor("op_45780_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_45780_end_0 = const()[name = tensor("op_45780_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_45780_end_mask_0 = const()[name = tensor("op_45780_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45780_cast_fp16 = slice_by_index(begin = var_45780_begin_0, end = var_45780_end_0, end_mask = var_45780_end_mask_0, x = v_203_cast_fp16)[name = tensor("op_45780_cast_fp16")]; + tensor var_45784_begin_0 = const()[name = tensor("op_45784_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_45784_end_0 = const()[name = tensor("op_45784_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_45784_end_mask_0 = const()[name = tensor("op_45784_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45784_cast_fp16 = slice_by_index(begin = var_45784_begin_0, end = var_45784_end_0, end_mask = var_45784_end_mask_0, x = v_203_cast_fp16)[name = tensor("op_45784_cast_fp16")]; + tensor var_45788_begin_0 = const()[name = tensor("op_45788_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_45788_end_0 = const()[name = tensor("op_45788_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_45788_end_mask_0 = const()[name = tensor("op_45788_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45788_cast_fp16 = slice_by_index(begin = var_45788_begin_0, end = var_45788_end_0, end_mask = var_45788_end_mask_0, x = v_203_cast_fp16)[name = tensor("op_45788_cast_fp16")]; + tensor var_45792_begin_0 = const()[name = tensor("op_45792_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_45792_end_0 = const()[name = tensor("op_45792_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_45792_end_mask_0 = const()[name = tensor("op_45792_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45792_cast_fp16 = slice_by_index(begin = var_45792_begin_0, end = var_45792_end_0, end_mask = var_45792_end_mask_0, x = v_203_cast_fp16)[name = tensor("op_45792_cast_fp16")]; + tensor var_45796_begin_0 = const()[name = tensor("op_45796_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_45796_end_0 = const()[name = tensor("op_45796_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_45796_end_mask_0 = const()[name = tensor("op_45796_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45796_cast_fp16 = slice_by_index(begin = var_45796_begin_0, end = var_45796_end_0, end_mask = var_45796_end_mask_0, x = v_203_cast_fp16)[name = tensor("op_45796_cast_fp16")]; + tensor var_45800_begin_0 = const()[name = tensor("op_45800_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_45800_end_0 = const()[name = tensor("op_45800_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_45800_end_mask_0 = const()[name = tensor("op_45800_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45800_cast_fp16 = slice_by_index(begin = var_45800_begin_0, end = var_45800_end_0, end_mask = var_45800_end_mask_0, x = v_203_cast_fp16)[name = tensor("op_45800_cast_fp16")]; + tensor var_45804_begin_0 = const()[name = tensor("op_45804_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_45804_end_0 = const()[name = tensor("op_45804_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_45804_end_mask_0 = const()[name = tensor("op_45804_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45804_cast_fp16 = slice_by_index(begin = var_45804_begin_0, end = var_45804_end_0, end_mask = var_45804_end_mask_0, x = v_203_cast_fp16)[name = tensor("op_45804_cast_fp16")]; + tensor var_45808_begin_0 = const()[name = tensor("op_45808_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_45808_end_0 = const()[name = tensor("op_45808_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_45808_end_mask_0 = const()[name = tensor("op_45808_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45808_cast_fp16 = slice_by_index(begin = var_45808_begin_0, end = var_45808_end_0, end_mask = var_45808_end_mask_0, x = v_203_cast_fp16)[name = tensor("op_45808_cast_fp16")]; + tensor var_45812_begin_0 = const()[name = tensor("op_45812_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_45812_end_0 = const()[name = tensor("op_45812_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_45812_end_mask_0 = const()[name = tensor("op_45812_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45812_cast_fp16 = slice_by_index(begin = var_45812_begin_0, end = var_45812_end_0, end_mask = var_45812_end_mask_0, x = v_203_cast_fp16)[name = tensor("op_45812_cast_fp16")]; + tensor var_45816_begin_0 = const()[name = tensor("op_45816_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_45816_end_0 = const()[name = tensor("op_45816_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_45816_end_mask_0 = const()[name = tensor("op_45816_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45816_cast_fp16 = slice_by_index(begin = var_45816_begin_0, end = var_45816_end_0, end_mask = var_45816_end_mask_0, x = v_203_cast_fp16)[name = tensor("op_45816_cast_fp16")]; + tensor var_45820_begin_0 = const()[name = tensor("op_45820_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_45820_end_0 = const()[name = tensor("op_45820_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_45820_end_mask_0 = const()[name = tensor("op_45820_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45820_cast_fp16 = slice_by_index(begin = var_45820_begin_0, end = var_45820_end_0, end_mask = var_45820_end_mask_0, x = v_203_cast_fp16)[name = tensor("op_45820_cast_fp16")]; + tensor var_45824_begin_0 = const()[name = tensor("op_45824_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_45824_end_0 = const()[name = tensor("op_45824_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_45824_end_mask_0 = const()[name = tensor("op_45824_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45824_cast_fp16 = slice_by_index(begin = var_45824_begin_0, end = var_45824_end_0, end_mask = var_45824_end_mask_0, x = v_203_cast_fp16)[name = tensor("op_45824_cast_fp16")]; + tensor var_45828_begin_0 = const()[name = tensor("op_45828_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_45828_end_0 = const()[name = tensor("op_45828_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_45828_end_mask_0 = const()[name = tensor("op_45828_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45828_cast_fp16 = slice_by_index(begin = var_45828_begin_0, end = var_45828_end_0, end_mask = var_45828_end_mask_0, x = v_203_cast_fp16)[name = tensor("op_45828_cast_fp16")]; + tensor var_45832_begin_0 = const()[name = tensor("op_45832_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_45832_end_0 = const()[name = tensor("op_45832_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_45832_end_mask_0 = const()[name = tensor("op_45832_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45832_cast_fp16 = slice_by_index(begin = var_45832_begin_0, end = var_45832_end_0, end_mask = var_45832_end_mask_0, x = v_203_cast_fp16)[name = tensor("op_45832_cast_fp16")]; + tensor var_45836_begin_0 = const()[name = tensor("op_45836_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_45836_end_0 = const()[name = tensor("op_45836_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_45836_end_mask_0 = const()[name = tensor("op_45836_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45836_cast_fp16 = slice_by_index(begin = var_45836_begin_0, end = var_45836_end_0, end_mask = var_45836_end_mask_0, x = v_203_cast_fp16)[name = tensor("op_45836_cast_fp16")]; + tensor var_45840_begin_0 = const()[name = tensor("op_45840_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_45840_end_0 = const()[name = tensor("op_45840_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_45840_end_mask_0 = const()[name = tensor("op_45840_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45840_cast_fp16 = slice_by_index(begin = var_45840_begin_0, end = var_45840_end_0, end_mask = var_45840_end_mask_0, x = v_203_cast_fp16)[name = tensor("op_45840_cast_fp16")]; + tensor var_45844_begin_0 = const()[name = tensor("op_45844_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_45844_end_0 = const()[name = tensor("op_45844_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_45844_end_mask_0 = const()[name = tensor("op_45844_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45844_cast_fp16 = slice_by_index(begin = var_45844_begin_0, end = var_45844_end_0, end_mask = var_45844_end_mask_0, x = v_203_cast_fp16)[name = tensor("op_45844_cast_fp16")]; + tensor var_45848_begin_0 = const()[name = tensor("op_45848_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_45848_end_0 = const()[name = tensor("op_45848_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_45848_end_mask_0 = const()[name = tensor("op_45848_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45848_cast_fp16 = slice_by_index(begin = var_45848_begin_0, end = var_45848_end_0, end_mask = var_45848_end_mask_0, x = v_203_cast_fp16)[name = tensor("op_45848_cast_fp16")]; + tensor var_45852_begin_0 = const()[name = tensor("op_45852_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_45852_end_0 = const()[name = tensor("op_45852_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_45852_end_mask_0 = const()[name = tensor("op_45852_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45852_cast_fp16 = slice_by_index(begin = var_45852_begin_0, end = var_45852_end_0, end_mask = var_45852_end_mask_0, x = v_203_cast_fp16)[name = tensor("op_45852_cast_fp16")]; + tensor var_45856_equation_0 = const()[name = tensor("op_45856_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45856_cast_fp16 = einsum(equation = var_45856_equation_0, values = (var_45698_cast_fp16, var_45615_cast_fp16))[name = tensor("op_45856_cast_fp16")]; + tensor var_45857_to_fp16 = const()[name = tensor("op_45857_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3881_cast_fp16 = mul(x = var_45856_cast_fp16, y = var_45857_to_fp16)[name = tensor("aw_3881_cast_fp16")]; + tensor var_45860_equation_0 = const()[name = tensor("op_45860_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45860_cast_fp16 = einsum(equation = var_45860_equation_0, values = (var_45702_cast_fp16, var_45619_cast_fp16))[name = tensor("op_45860_cast_fp16")]; + tensor var_45861_to_fp16 = const()[name = tensor("op_45861_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3883_cast_fp16 = mul(x = var_45860_cast_fp16, y = var_45861_to_fp16)[name = tensor("aw_3883_cast_fp16")]; + tensor var_45864_equation_0 = const()[name = tensor("op_45864_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45864_cast_fp16 = einsum(equation = var_45864_equation_0, values = (var_45706_cast_fp16, var_45623_cast_fp16))[name = tensor("op_45864_cast_fp16")]; + tensor var_45865_to_fp16 = const()[name = tensor("op_45865_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3885_cast_fp16 = mul(x = var_45864_cast_fp16, y = var_45865_to_fp16)[name = tensor("aw_3885_cast_fp16")]; + tensor var_45868_equation_0 = const()[name = tensor("op_45868_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45868_cast_fp16 = einsum(equation = var_45868_equation_0, values = (var_45710_cast_fp16, var_45627_cast_fp16))[name = tensor("op_45868_cast_fp16")]; + tensor var_45869_to_fp16 = const()[name = tensor("op_45869_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3887_cast_fp16 = mul(x = var_45868_cast_fp16, y = var_45869_to_fp16)[name = tensor("aw_3887_cast_fp16")]; + tensor var_45872_equation_0 = const()[name = tensor("op_45872_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45872_cast_fp16 = einsum(equation = var_45872_equation_0, values = (var_45714_cast_fp16, var_45631_cast_fp16))[name = tensor("op_45872_cast_fp16")]; + tensor var_45873_to_fp16 = const()[name = tensor("op_45873_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3889_cast_fp16 = mul(x = var_45872_cast_fp16, y = var_45873_to_fp16)[name = tensor("aw_3889_cast_fp16")]; + tensor var_45876_equation_0 = const()[name = tensor("op_45876_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45876_cast_fp16 = einsum(equation = var_45876_equation_0, values = (var_45718_cast_fp16, var_45635_cast_fp16))[name = tensor("op_45876_cast_fp16")]; + tensor var_45877_to_fp16 = const()[name = tensor("op_45877_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3891_cast_fp16 = mul(x = var_45876_cast_fp16, y = var_45877_to_fp16)[name = tensor("aw_3891_cast_fp16")]; + tensor var_45880_equation_0 = const()[name = tensor("op_45880_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45880_cast_fp16 = einsum(equation = var_45880_equation_0, values = (var_45722_cast_fp16, var_45639_cast_fp16))[name = tensor("op_45880_cast_fp16")]; + tensor var_45881_to_fp16 = const()[name = tensor("op_45881_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3893_cast_fp16 = mul(x = var_45880_cast_fp16, y = var_45881_to_fp16)[name = tensor("aw_3893_cast_fp16")]; + tensor var_45884_equation_0 = const()[name = tensor("op_45884_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45884_cast_fp16 = einsum(equation = var_45884_equation_0, values = (var_45726_cast_fp16, var_45643_cast_fp16))[name = tensor("op_45884_cast_fp16")]; + tensor var_45885_to_fp16 = const()[name = tensor("op_45885_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3895_cast_fp16 = mul(x = var_45884_cast_fp16, y = var_45885_to_fp16)[name = tensor("aw_3895_cast_fp16")]; + tensor var_45888_equation_0 = const()[name = tensor("op_45888_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45888_cast_fp16 = einsum(equation = var_45888_equation_0, values = (var_45730_cast_fp16, var_45647_cast_fp16))[name = tensor("op_45888_cast_fp16")]; + tensor var_45889_to_fp16 = const()[name = tensor("op_45889_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3897_cast_fp16 = mul(x = var_45888_cast_fp16, y = var_45889_to_fp16)[name = tensor("aw_3897_cast_fp16")]; + tensor var_45892_equation_0 = const()[name = tensor("op_45892_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45892_cast_fp16 = einsum(equation = var_45892_equation_0, values = (var_45734_cast_fp16, var_45651_cast_fp16))[name = tensor("op_45892_cast_fp16")]; + tensor var_45893_to_fp16 = const()[name = tensor("op_45893_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3899_cast_fp16 = mul(x = var_45892_cast_fp16, y = var_45893_to_fp16)[name = tensor("aw_3899_cast_fp16")]; + tensor var_45896_equation_0 = const()[name = tensor("op_45896_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45896_cast_fp16 = einsum(equation = var_45896_equation_0, values = (var_45738_cast_fp16, var_45655_cast_fp16))[name = tensor("op_45896_cast_fp16")]; + tensor var_45897_to_fp16 = const()[name = tensor("op_45897_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3901_cast_fp16 = mul(x = var_45896_cast_fp16, y = var_45897_to_fp16)[name = tensor("aw_3901_cast_fp16")]; + tensor var_45900_equation_0 = const()[name = tensor("op_45900_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45900_cast_fp16 = einsum(equation = var_45900_equation_0, values = (var_45742_cast_fp16, var_45659_cast_fp16))[name = tensor("op_45900_cast_fp16")]; + tensor var_45901_to_fp16 = const()[name = tensor("op_45901_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3903_cast_fp16 = mul(x = var_45900_cast_fp16, y = var_45901_to_fp16)[name = tensor("aw_3903_cast_fp16")]; + tensor var_45904_equation_0 = const()[name = tensor("op_45904_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45904_cast_fp16 = einsum(equation = var_45904_equation_0, values = (var_45746_cast_fp16, var_45663_cast_fp16))[name = tensor("op_45904_cast_fp16")]; + tensor var_45905_to_fp16 = const()[name = tensor("op_45905_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3905_cast_fp16 = mul(x = var_45904_cast_fp16, y = var_45905_to_fp16)[name = tensor("aw_3905_cast_fp16")]; + tensor var_45908_equation_0 = const()[name = tensor("op_45908_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45908_cast_fp16 = einsum(equation = var_45908_equation_0, values = (var_45750_cast_fp16, var_45667_cast_fp16))[name = tensor("op_45908_cast_fp16")]; + tensor var_45909_to_fp16 = const()[name = tensor("op_45909_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3907_cast_fp16 = mul(x = var_45908_cast_fp16, y = var_45909_to_fp16)[name = tensor("aw_3907_cast_fp16")]; + tensor var_45912_equation_0 = const()[name = tensor("op_45912_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45912_cast_fp16 = einsum(equation = var_45912_equation_0, values = (var_45754_cast_fp16, var_45671_cast_fp16))[name = tensor("op_45912_cast_fp16")]; + tensor var_45913_to_fp16 = const()[name = tensor("op_45913_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3909_cast_fp16 = mul(x = var_45912_cast_fp16, y = var_45913_to_fp16)[name = tensor("aw_3909_cast_fp16")]; + tensor var_45916_equation_0 = const()[name = tensor("op_45916_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45916_cast_fp16 = einsum(equation = var_45916_equation_0, values = (var_45758_cast_fp16, var_45675_cast_fp16))[name = tensor("op_45916_cast_fp16")]; + tensor var_45917_to_fp16 = const()[name = tensor("op_45917_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3911_cast_fp16 = mul(x = var_45916_cast_fp16, y = var_45917_to_fp16)[name = tensor("aw_3911_cast_fp16")]; + tensor var_45920_equation_0 = const()[name = tensor("op_45920_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45920_cast_fp16 = einsum(equation = var_45920_equation_0, values = (var_45762_cast_fp16, var_45679_cast_fp16))[name = tensor("op_45920_cast_fp16")]; + tensor var_45921_to_fp16 = const()[name = tensor("op_45921_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3913_cast_fp16 = mul(x = var_45920_cast_fp16, y = var_45921_to_fp16)[name = tensor("aw_3913_cast_fp16")]; + tensor var_45924_equation_0 = const()[name = tensor("op_45924_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45924_cast_fp16 = einsum(equation = var_45924_equation_0, values = (var_45766_cast_fp16, var_45683_cast_fp16))[name = tensor("op_45924_cast_fp16")]; + tensor var_45925_to_fp16 = const()[name = tensor("op_45925_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3915_cast_fp16 = mul(x = var_45924_cast_fp16, y = var_45925_to_fp16)[name = tensor("aw_3915_cast_fp16")]; + tensor var_45928_equation_0 = const()[name = tensor("op_45928_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45928_cast_fp16 = einsum(equation = var_45928_equation_0, values = (var_45770_cast_fp16, var_45687_cast_fp16))[name = tensor("op_45928_cast_fp16")]; + tensor var_45929_to_fp16 = const()[name = tensor("op_45929_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3917_cast_fp16 = mul(x = var_45928_cast_fp16, y = var_45929_to_fp16)[name = tensor("aw_3917_cast_fp16")]; + tensor var_45932_equation_0 = const()[name = tensor("op_45932_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45932_cast_fp16 = einsum(equation = var_45932_equation_0, values = (var_45774_cast_fp16, var_45691_cast_fp16))[name = tensor("op_45932_cast_fp16")]; + tensor var_45933_to_fp16 = const()[name = tensor("op_45933_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3919_cast_fp16 = mul(x = var_45932_cast_fp16, y = var_45933_to_fp16)[name = tensor("aw_3919_cast_fp16")]; + tensor var_45935_cast_fp16 = softmax(axis = var_30355, x = aw_3881_cast_fp16)[name = tensor("op_45935_cast_fp16")]; + tensor var_45936_cast_fp16 = softmax(axis = var_30355, x = aw_3883_cast_fp16)[name = tensor("op_45936_cast_fp16")]; + tensor var_45937_cast_fp16 = softmax(axis = var_30355, x = aw_3885_cast_fp16)[name = tensor("op_45937_cast_fp16")]; + tensor var_45938_cast_fp16 = softmax(axis = var_30355, x = aw_3887_cast_fp16)[name = tensor("op_45938_cast_fp16")]; + tensor var_45939_cast_fp16 = softmax(axis = var_30355, x = aw_3889_cast_fp16)[name = tensor("op_45939_cast_fp16")]; + tensor var_45940_cast_fp16 = softmax(axis = var_30355, x = aw_3891_cast_fp16)[name = tensor("op_45940_cast_fp16")]; + tensor var_45941_cast_fp16 = softmax(axis = var_30355, x = aw_3893_cast_fp16)[name = tensor("op_45941_cast_fp16")]; + tensor var_45942_cast_fp16 = softmax(axis = var_30355, x = aw_3895_cast_fp16)[name = tensor("op_45942_cast_fp16")]; + tensor var_45943_cast_fp16 = softmax(axis = var_30355, x = aw_3897_cast_fp16)[name = tensor("op_45943_cast_fp16")]; + tensor var_45944_cast_fp16 = softmax(axis = var_30355, x = aw_3899_cast_fp16)[name = tensor("op_45944_cast_fp16")]; + tensor var_45945_cast_fp16 = softmax(axis = var_30355, x = aw_3901_cast_fp16)[name = tensor("op_45945_cast_fp16")]; + tensor var_45946_cast_fp16 = softmax(axis = var_30355, x = aw_3903_cast_fp16)[name = tensor("op_45946_cast_fp16")]; + tensor var_45947_cast_fp16 = softmax(axis = var_30355, x = aw_3905_cast_fp16)[name = tensor("op_45947_cast_fp16")]; + tensor var_45948_cast_fp16 = softmax(axis = var_30355, x = aw_3907_cast_fp16)[name = tensor("op_45948_cast_fp16")]; + tensor var_45949_cast_fp16 = softmax(axis = var_30355, x = aw_3909_cast_fp16)[name = tensor("op_45949_cast_fp16")]; + tensor var_45950_cast_fp16 = softmax(axis = var_30355, x = aw_3911_cast_fp16)[name = tensor("op_45950_cast_fp16")]; + tensor var_45951_cast_fp16 = softmax(axis = var_30355, x = aw_3913_cast_fp16)[name = tensor("op_45951_cast_fp16")]; + tensor var_45952_cast_fp16 = softmax(axis = var_30355, x = aw_3915_cast_fp16)[name = tensor("op_45952_cast_fp16")]; + tensor var_45953_cast_fp16 = softmax(axis = var_30355, x = aw_3917_cast_fp16)[name = tensor("op_45953_cast_fp16")]; + tensor var_45954_cast_fp16 = softmax(axis = var_30355, x = aw_3919_cast_fp16)[name = tensor("op_45954_cast_fp16")]; + tensor var_45956_equation_0 = const()[name = tensor("op_45956_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45956_cast_fp16 = einsum(equation = var_45956_equation_0, values = (var_45776_cast_fp16, var_45935_cast_fp16))[name = tensor("op_45956_cast_fp16")]; + tensor var_45958_equation_0 = const()[name = tensor("op_45958_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45958_cast_fp16 = einsum(equation = var_45958_equation_0, values = (var_45780_cast_fp16, var_45936_cast_fp16))[name = tensor("op_45958_cast_fp16")]; + tensor var_45960_equation_0 = const()[name = tensor("op_45960_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45960_cast_fp16 = einsum(equation = var_45960_equation_0, values = (var_45784_cast_fp16, var_45937_cast_fp16))[name = tensor("op_45960_cast_fp16")]; + tensor var_45962_equation_0 = const()[name = tensor("op_45962_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45962_cast_fp16 = einsum(equation = var_45962_equation_0, values = (var_45788_cast_fp16, var_45938_cast_fp16))[name = tensor("op_45962_cast_fp16")]; + tensor var_45964_equation_0 = const()[name = tensor("op_45964_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45964_cast_fp16 = einsum(equation = var_45964_equation_0, values = (var_45792_cast_fp16, var_45939_cast_fp16))[name = tensor("op_45964_cast_fp16")]; + tensor var_45966_equation_0 = const()[name = tensor("op_45966_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45966_cast_fp16 = einsum(equation = var_45966_equation_0, values = (var_45796_cast_fp16, var_45940_cast_fp16))[name = tensor("op_45966_cast_fp16")]; + tensor var_45968_equation_0 = const()[name = tensor("op_45968_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45968_cast_fp16 = einsum(equation = var_45968_equation_0, values = (var_45800_cast_fp16, var_45941_cast_fp16))[name = tensor("op_45968_cast_fp16")]; + tensor var_45970_equation_0 = const()[name = tensor("op_45970_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45970_cast_fp16 = einsum(equation = var_45970_equation_0, values = (var_45804_cast_fp16, var_45942_cast_fp16))[name = tensor("op_45970_cast_fp16")]; + tensor var_45972_equation_0 = const()[name = tensor("op_45972_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45972_cast_fp16 = einsum(equation = var_45972_equation_0, values = (var_45808_cast_fp16, var_45943_cast_fp16))[name = tensor("op_45972_cast_fp16")]; + tensor var_45974_equation_0 = const()[name = tensor("op_45974_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45974_cast_fp16 = einsum(equation = var_45974_equation_0, values = (var_45812_cast_fp16, var_45944_cast_fp16))[name = tensor("op_45974_cast_fp16")]; + tensor var_45976_equation_0 = const()[name = tensor("op_45976_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45976_cast_fp16 = einsum(equation = var_45976_equation_0, values = (var_45816_cast_fp16, var_45945_cast_fp16))[name = tensor("op_45976_cast_fp16")]; + tensor var_45978_equation_0 = const()[name = tensor("op_45978_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45978_cast_fp16 = einsum(equation = var_45978_equation_0, values = (var_45820_cast_fp16, var_45946_cast_fp16))[name = tensor("op_45978_cast_fp16")]; + tensor var_45980_equation_0 = const()[name = tensor("op_45980_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45980_cast_fp16 = einsum(equation = var_45980_equation_0, values = (var_45824_cast_fp16, var_45947_cast_fp16))[name = tensor("op_45980_cast_fp16")]; + tensor var_45982_equation_0 = const()[name = tensor("op_45982_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45982_cast_fp16 = einsum(equation = var_45982_equation_0, values = (var_45828_cast_fp16, var_45948_cast_fp16))[name = tensor("op_45982_cast_fp16")]; + tensor var_45984_equation_0 = const()[name = tensor("op_45984_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45984_cast_fp16 = einsum(equation = var_45984_equation_0, values = (var_45832_cast_fp16, var_45949_cast_fp16))[name = tensor("op_45984_cast_fp16")]; + tensor var_45986_equation_0 = const()[name = tensor("op_45986_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45986_cast_fp16 = einsum(equation = var_45986_equation_0, values = (var_45836_cast_fp16, var_45950_cast_fp16))[name = tensor("op_45986_cast_fp16")]; + tensor var_45988_equation_0 = const()[name = tensor("op_45988_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45988_cast_fp16 = einsum(equation = var_45988_equation_0, values = (var_45840_cast_fp16, var_45951_cast_fp16))[name = tensor("op_45988_cast_fp16")]; + tensor var_45990_equation_0 = const()[name = tensor("op_45990_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45990_cast_fp16 = einsum(equation = var_45990_equation_0, values = (var_45844_cast_fp16, var_45952_cast_fp16))[name = tensor("op_45990_cast_fp16")]; + tensor var_45992_equation_0 = const()[name = tensor("op_45992_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45992_cast_fp16 = einsum(equation = var_45992_equation_0, values = (var_45848_cast_fp16, var_45953_cast_fp16))[name = tensor("op_45992_cast_fp16")]; + tensor var_45994_equation_0 = const()[name = tensor("op_45994_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45994_cast_fp16 = einsum(equation = var_45994_equation_0, values = (var_45852_cast_fp16, var_45954_cast_fp16))[name = tensor("op_45994_cast_fp16")]; + tensor input_587_interleave_0 = const()[name = tensor("input_587_interleave_0"), val = tensor(false)]; + tensor input_587_cast_fp16 = concat(axis = var_30355, interleave = input_587_interleave_0, values = (var_45956_cast_fp16, var_45958_cast_fp16, var_45960_cast_fp16, var_45962_cast_fp16, var_45964_cast_fp16, var_45966_cast_fp16, var_45968_cast_fp16, var_45970_cast_fp16, var_45972_cast_fp16, var_45974_cast_fp16, var_45976_cast_fp16, var_45978_cast_fp16, var_45980_cast_fp16, var_45982_cast_fp16, var_45984_cast_fp16, var_45986_cast_fp16, var_45988_cast_fp16, var_45990_cast_fp16, var_45992_cast_fp16, var_45994_cast_fp16))[name = tensor("input_587_cast_fp16")]; + tensor var_46000 = const()[name = tensor("op_46000"), val = tensor([1, 1])]; + tensor var_46002 = const()[name = tensor("op_46002"), val = tensor([1, 1])]; + tensor var_46004_pad_type_0 = const()[name = tensor("op_46004_pad_type_0"), val = tensor("custom")]; + tensor var_46004_pad_0 = const()[name = tensor("op_46004_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_6_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_6_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3832406656)))]; + tensor up_blocks_0_attentions_1_transformer_blocks_6_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_6_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3835683520)))]; + tensor var_46004_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_6_attn2_to_out_0_bias_to_fp16, dilations = var_46002, groups = var_30355, pad = var_46004_pad_0, pad_type = var_46004_pad_type_0, strides = var_46000, weight = up_blocks_0_attentions_1_transformer_blocks_6_attn2_to_out_0_weight_to_fp16, x = input_587_cast_fp16)[name = tensor("op_46004_cast_fp16")]; + tensor inputs_305_cast_fp16 = add(x = var_46004_cast_fp16, y = inputs_303_cast_fp16)[name = tensor("inputs_305_cast_fp16")]; + tensor input_589_axes_0 = const()[name = tensor("input_589_axes_0"), val = tensor([1])]; + tensor input_589_gamma_0_to_fp16 = const()[name = tensor("input_589_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3835686144)))]; + tensor input_589_beta_0_to_fp16 = const()[name = tensor("input_589_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3835688768)))]; + tensor var_46014_to_fp16 = const()[name = tensor("op_46014_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_589_cast_fp16 = layer_norm(axes = input_589_axes_0, beta = input_589_beta_0_to_fp16, epsilon = var_46014_to_fp16, gamma = input_589_gamma_0_to_fp16, x = inputs_305_cast_fp16)[name = tensor("input_589_cast_fp16")]; + tensor var_46030 = const()[name = tensor("op_46030"), val = tensor([1, 1])]; + tensor var_46032 = const()[name = tensor("op_46032"), val = tensor([1, 1])]; + tensor var_46034_pad_type_0 = const()[name = tensor("op_46034_pad_type_0"), val = tensor("custom")]; + tensor var_46034_pad_0 = const()[name = tensor("op_46034_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_6_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_6_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3835691392)))]; + tensor up_blocks_0_attentions_1_transformer_blocks_6_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_6_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3861905856)))]; + tensor var_46034_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_6_ff_net_0_proj_bias_to_fp16, dilations = var_46032, groups = var_30355, pad = var_46034_pad_0, pad_type = var_46034_pad_type_0, strides = var_46030, weight = up_blocks_0_attentions_1_transformer_blocks_6_ff_net_0_proj_weight_to_fp16, x = input_589_cast_fp16)[name = tensor("op_46034_cast_fp16")]; + tensor var_46035_split_sizes_0 = const()[name = tensor("op_46035_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_46035_axis_0 = const()[name = tensor("op_46035_axis_0"), val = tensor(1)]; + tensor var_46035_cast_fp16_0, tensor var_46035_cast_fp16_1 = split(axis = var_46035_axis_0, split_sizes = var_46035_split_sizes_0, x = var_46034_cast_fp16)[name = tensor("op_46035_cast_fp16")]; + tensor var_46037_mode_0 = const()[name = tensor("op_46037_mode_0"), val = tensor("EXACT")]; + tensor var_46037_cast_fp16 = gelu(mode = var_46037_mode_0, x = var_46035_cast_fp16_1)[name = tensor("op_46037_cast_fp16")]; + tensor input_591_cast_fp16 = mul(x = var_46035_cast_fp16_0, y = var_46037_cast_fp16)[name = tensor("input_591_cast_fp16")]; + tensor var_46041 = const()[name = tensor("op_46041"), val = tensor([1, 1])]; + tensor var_46043 = const()[name = tensor("op_46043"), val = tensor([1, 1])]; + tensor var_46045_pad_type_0 = const()[name = tensor("op_46045_pad_type_0"), val = tensor("custom")]; + tensor var_46045_pad_0 = const()[name = tensor("op_46045_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_6_ff_net_2_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_6_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3861926400)))]; + tensor up_blocks_0_attentions_1_transformer_blocks_6_ff_net_2_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_6_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3875033664)))]; + tensor var_46045_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_6_ff_net_2_bias_to_fp16, dilations = var_46043, groups = var_30355, pad = var_46045_pad_0, pad_type = var_46045_pad_type_0, strides = var_46041, weight = up_blocks_0_attentions_1_transformer_blocks_6_ff_net_2_weight_to_fp16, x = input_591_cast_fp16)[name = tensor("op_46045_cast_fp16")]; + tensor inputs_307_cast_fp16 = add(x = var_46045_cast_fp16, y = inputs_305_cast_fp16)[name = tensor("inputs_307_cast_fp16")]; + tensor hidden_states_403_axes_0 = const()[name = tensor("hidden_states_403_axes_0"), val = tensor([1])]; + tensor hidden_states_403_gamma_0_to_fp16 = const()[name = tensor("hidden_states_403_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3875036288)))]; + tensor hidden_states_403_beta_0_to_fp16 = const()[name = tensor("hidden_states_403_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3875038912)))]; + tensor var_46061_to_fp16 = const()[name = tensor("op_46061_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_403_cast_fp16 = layer_norm(axes = hidden_states_403_axes_0, beta = hidden_states_403_beta_0_to_fp16, epsilon = var_46061_to_fp16, gamma = hidden_states_403_gamma_0_to_fp16, x = inputs_307_cast_fp16)[name = tensor("hidden_states_403_cast_fp16")]; + tensor var_46076 = const()[name = tensor("op_46076"), val = tensor([1, 1])]; + tensor var_46078 = const()[name = tensor("op_46078"), val = tensor([1, 1])]; + tensor q_205_pad_type_0 = const()[name = tensor("q_205_pad_type_0"), val = tensor("custom")]; + tensor q_205_pad_0 = const()[name = tensor("q_205_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_7_attn1_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_7_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3875041536)))]; + tensor q_205_cast_fp16 = conv(dilations = var_46078, groups = var_30355, pad = q_205_pad_0, pad_type = q_205_pad_type_0, strides = var_46076, weight = up_blocks_0_attentions_1_transformer_blocks_7_attn1_to_q_weight_to_fp16, x = hidden_states_403_cast_fp16)[name = tensor("q_205_cast_fp16")]; + tensor var_46082 = const()[name = tensor("op_46082"), val = tensor([1, 1])]; + tensor var_46084 = const()[name = tensor("op_46084"), val = tensor([1, 1])]; + tensor k_409_pad_type_0 = const()[name = tensor("k_409_pad_type_0"), val = tensor("custom")]; + tensor k_409_pad_0 = const()[name = tensor("k_409_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_7_attn1_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_7_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3878318400)))]; + tensor k_409_cast_fp16 = conv(dilations = var_46084, groups = var_30355, pad = k_409_pad_0, pad_type = k_409_pad_type_0, strides = var_46082, weight = up_blocks_0_attentions_1_transformer_blocks_7_attn1_to_k_weight_to_fp16, x = hidden_states_403_cast_fp16)[name = tensor("k_409_cast_fp16")]; + tensor var_46088 = const()[name = tensor("op_46088"), val = tensor([1, 1])]; + tensor var_46090 = const()[name = tensor("op_46090"), val = tensor([1, 1])]; + tensor v_205_pad_type_0 = const()[name = tensor("v_205_pad_type_0"), val = tensor("custom")]; + tensor v_205_pad_0 = const()[name = tensor("v_205_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_7_attn1_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_7_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3881595264)))]; + tensor v_205_cast_fp16 = conv(dilations = var_46090, groups = var_30355, pad = v_205_pad_0, pad_type = v_205_pad_type_0, strides = var_46088, weight = up_blocks_0_attentions_1_transformer_blocks_7_attn1_to_v_weight_to_fp16, x = hidden_states_403_cast_fp16)[name = tensor("v_205_cast_fp16")]; + tensor var_46094_begin_0 = const()[name = tensor("op_46094_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46094_end_0 = const()[name = tensor("op_46094_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_46094_end_mask_0 = const()[name = tensor("op_46094_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46094_cast_fp16 = slice_by_index(begin = var_46094_begin_0, end = var_46094_end_0, end_mask = var_46094_end_mask_0, x = q_205_cast_fp16)[name = tensor("op_46094_cast_fp16")]; + tensor var_46098_begin_0 = const()[name = tensor("op_46098_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_46098_end_0 = const()[name = tensor("op_46098_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_46098_end_mask_0 = const()[name = tensor("op_46098_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46098_cast_fp16 = slice_by_index(begin = var_46098_begin_0, end = var_46098_end_0, end_mask = var_46098_end_mask_0, x = q_205_cast_fp16)[name = tensor("op_46098_cast_fp16")]; + tensor var_46102_begin_0 = const()[name = tensor("op_46102_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_46102_end_0 = const()[name = tensor("op_46102_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_46102_end_mask_0 = const()[name = tensor("op_46102_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46102_cast_fp16 = slice_by_index(begin = var_46102_begin_0, end = var_46102_end_0, end_mask = var_46102_end_mask_0, x = q_205_cast_fp16)[name = tensor("op_46102_cast_fp16")]; + tensor var_46106_begin_0 = const()[name = tensor("op_46106_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_46106_end_0 = const()[name = tensor("op_46106_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_46106_end_mask_0 = const()[name = tensor("op_46106_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46106_cast_fp16 = slice_by_index(begin = var_46106_begin_0, end = var_46106_end_0, end_mask = var_46106_end_mask_0, x = q_205_cast_fp16)[name = tensor("op_46106_cast_fp16")]; + tensor var_46110_begin_0 = const()[name = tensor("op_46110_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_46110_end_0 = const()[name = tensor("op_46110_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_46110_end_mask_0 = const()[name = tensor("op_46110_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46110_cast_fp16 = slice_by_index(begin = var_46110_begin_0, end = var_46110_end_0, end_mask = var_46110_end_mask_0, x = q_205_cast_fp16)[name = tensor("op_46110_cast_fp16")]; + tensor var_46114_begin_0 = const()[name = tensor("op_46114_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_46114_end_0 = const()[name = tensor("op_46114_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_46114_end_mask_0 = const()[name = tensor("op_46114_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46114_cast_fp16 = slice_by_index(begin = var_46114_begin_0, end = var_46114_end_0, end_mask = var_46114_end_mask_0, x = q_205_cast_fp16)[name = tensor("op_46114_cast_fp16")]; + tensor var_46118_begin_0 = const()[name = tensor("op_46118_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_46118_end_0 = const()[name = tensor("op_46118_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_46118_end_mask_0 = const()[name = tensor("op_46118_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46118_cast_fp16 = slice_by_index(begin = var_46118_begin_0, end = var_46118_end_0, end_mask = var_46118_end_mask_0, x = q_205_cast_fp16)[name = tensor("op_46118_cast_fp16")]; + tensor var_46122_begin_0 = const()[name = tensor("op_46122_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_46122_end_0 = const()[name = tensor("op_46122_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_46122_end_mask_0 = const()[name = tensor("op_46122_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46122_cast_fp16 = slice_by_index(begin = var_46122_begin_0, end = var_46122_end_0, end_mask = var_46122_end_mask_0, x = q_205_cast_fp16)[name = tensor("op_46122_cast_fp16")]; + tensor var_46126_begin_0 = const()[name = tensor("op_46126_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_46126_end_0 = const()[name = tensor("op_46126_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_46126_end_mask_0 = const()[name = tensor("op_46126_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46126_cast_fp16 = slice_by_index(begin = var_46126_begin_0, end = var_46126_end_0, end_mask = var_46126_end_mask_0, x = q_205_cast_fp16)[name = tensor("op_46126_cast_fp16")]; + tensor var_46130_begin_0 = const()[name = tensor("op_46130_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_46130_end_0 = const()[name = tensor("op_46130_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_46130_end_mask_0 = const()[name = tensor("op_46130_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46130_cast_fp16 = slice_by_index(begin = var_46130_begin_0, end = var_46130_end_0, end_mask = var_46130_end_mask_0, x = q_205_cast_fp16)[name = tensor("op_46130_cast_fp16")]; + tensor var_46134_begin_0 = const()[name = tensor("op_46134_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_46134_end_0 = const()[name = tensor("op_46134_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_46134_end_mask_0 = const()[name = tensor("op_46134_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46134_cast_fp16 = slice_by_index(begin = var_46134_begin_0, end = var_46134_end_0, end_mask = var_46134_end_mask_0, x = q_205_cast_fp16)[name = tensor("op_46134_cast_fp16")]; + tensor var_46138_begin_0 = const()[name = tensor("op_46138_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_46138_end_0 = const()[name = tensor("op_46138_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_46138_end_mask_0 = const()[name = tensor("op_46138_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46138_cast_fp16 = slice_by_index(begin = var_46138_begin_0, end = var_46138_end_0, end_mask = var_46138_end_mask_0, x = q_205_cast_fp16)[name = tensor("op_46138_cast_fp16")]; + tensor var_46142_begin_0 = const()[name = tensor("op_46142_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_46142_end_0 = const()[name = tensor("op_46142_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_46142_end_mask_0 = const()[name = tensor("op_46142_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46142_cast_fp16 = slice_by_index(begin = var_46142_begin_0, end = var_46142_end_0, end_mask = var_46142_end_mask_0, x = q_205_cast_fp16)[name = tensor("op_46142_cast_fp16")]; + tensor var_46146_begin_0 = const()[name = tensor("op_46146_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_46146_end_0 = const()[name = tensor("op_46146_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_46146_end_mask_0 = const()[name = tensor("op_46146_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46146_cast_fp16 = slice_by_index(begin = var_46146_begin_0, end = var_46146_end_0, end_mask = var_46146_end_mask_0, x = q_205_cast_fp16)[name = tensor("op_46146_cast_fp16")]; + tensor var_46150_begin_0 = const()[name = tensor("op_46150_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_46150_end_0 = const()[name = tensor("op_46150_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_46150_end_mask_0 = const()[name = tensor("op_46150_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46150_cast_fp16 = slice_by_index(begin = var_46150_begin_0, end = var_46150_end_0, end_mask = var_46150_end_mask_0, x = q_205_cast_fp16)[name = tensor("op_46150_cast_fp16")]; + tensor var_46154_begin_0 = const()[name = tensor("op_46154_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_46154_end_0 = const()[name = tensor("op_46154_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_46154_end_mask_0 = const()[name = tensor("op_46154_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46154_cast_fp16 = slice_by_index(begin = var_46154_begin_0, end = var_46154_end_0, end_mask = var_46154_end_mask_0, x = q_205_cast_fp16)[name = tensor("op_46154_cast_fp16")]; + tensor var_46158_begin_0 = const()[name = tensor("op_46158_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_46158_end_0 = const()[name = tensor("op_46158_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_46158_end_mask_0 = const()[name = tensor("op_46158_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46158_cast_fp16 = slice_by_index(begin = var_46158_begin_0, end = var_46158_end_0, end_mask = var_46158_end_mask_0, x = q_205_cast_fp16)[name = tensor("op_46158_cast_fp16")]; + tensor var_46162_begin_0 = const()[name = tensor("op_46162_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_46162_end_0 = const()[name = tensor("op_46162_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_46162_end_mask_0 = const()[name = tensor("op_46162_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46162_cast_fp16 = slice_by_index(begin = var_46162_begin_0, end = var_46162_end_0, end_mask = var_46162_end_mask_0, x = q_205_cast_fp16)[name = tensor("op_46162_cast_fp16")]; + tensor var_46166_begin_0 = const()[name = tensor("op_46166_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_46166_end_0 = const()[name = tensor("op_46166_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_46166_end_mask_0 = const()[name = tensor("op_46166_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46166_cast_fp16 = slice_by_index(begin = var_46166_begin_0, end = var_46166_end_0, end_mask = var_46166_end_mask_0, x = q_205_cast_fp16)[name = tensor("op_46166_cast_fp16")]; + tensor var_46170_begin_0 = const()[name = tensor("op_46170_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_46170_end_0 = const()[name = tensor("op_46170_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_46170_end_mask_0 = const()[name = tensor("op_46170_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46170_cast_fp16 = slice_by_index(begin = var_46170_begin_0, end = var_46170_end_0, end_mask = var_46170_end_mask_0, x = q_205_cast_fp16)[name = tensor("op_46170_cast_fp16")]; + tensor k_411_perm_0 = const()[name = tensor("k_411_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_46177_begin_0 = const()[name = tensor("op_46177_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46177_end_0 = const()[name = tensor("op_46177_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_46177_end_mask_0 = const()[name = tensor("op_46177_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_411_cast_fp16 = transpose(perm = k_411_perm_0, x = k_409_cast_fp16)[name = tensor("transpose_37")]; + tensor var_46177_cast_fp16 = slice_by_index(begin = var_46177_begin_0, end = var_46177_end_0, end_mask = var_46177_end_mask_0, x = k_411_cast_fp16)[name = tensor("op_46177_cast_fp16")]; + tensor var_46181_begin_0 = const()[name = tensor("op_46181_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_46181_end_0 = const()[name = tensor("op_46181_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_46181_end_mask_0 = const()[name = tensor("op_46181_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46181_cast_fp16 = slice_by_index(begin = var_46181_begin_0, end = var_46181_end_0, end_mask = var_46181_end_mask_0, x = k_411_cast_fp16)[name = tensor("op_46181_cast_fp16")]; + tensor var_46185_begin_0 = const()[name = tensor("op_46185_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_46185_end_0 = const()[name = tensor("op_46185_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_46185_end_mask_0 = const()[name = tensor("op_46185_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46185_cast_fp16 = slice_by_index(begin = var_46185_begin_0, end = var_46185_end_0, end_mask = var_46185_end_mask_0, x = k_411_cast_fp16)[name = tensor("op_46185_cast_fp16")]; + tensor var_46189_begin_0 = const()[name = tensor("op_46189_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_46189_end_0 = const()[name = tensor("op_46189_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_46189_end_mask_0 = const()[name = tensor("op_46189_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46189_cast_fp16 = slice_by_index(begin = var_46189_begin_0, end = var_46189_end_0, end_mask = var_46189_end_mask_0, x = k_411_cast_fp16)[name = tensor("op_46189_cast_fp16")]; + tensor var_46193_begin_0 = const()[name = tensor("op_46193_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_46193_end_0 = const()[name = tensor("op_46193_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_46193_end_mask_0 = const()[name = tensor("op_46193_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46193_cast_fp16 = slice_by_index(begin = var_46193_begin_0, end = var_46193_end_0, end_mask = var_46193_end_mask_0, x = k_411_cast_fp16)[name = tensor("op_46193_cast_fp16")]; + tensor var_46197_begin_0 = const()[name = tensor("op_46197_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_46197_end_0 = const()[name = tensor("op_46197_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_46197_end_mask_0 = const()[name = tensor("op_46197_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46197_cast_fp16 = slice_by_index(begin = var_46197_begin_0, end = var_46197_end_0, end_mask = var_46197_end_mask_0, x = k_411_cast_fp16)[name = tensor("op_46197_cast_fp16")]; + tensor var_46201_begin_0 = const()[name = tensor("op_46201_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_46201_end_0 = const()[name = tensor("op_46201_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_46201_end_mask_0 = const()[name = tensor("op_46201_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46201_cast_fp16 = slice_by_index(begin = var_46201_begin_0, end = var_46201_end_0, end_mask = var_46201_end_mask_0, x = k_411_cast_fp16)[name = tensor("op_46201_cast_fp16")]; + tensor var_46205_begin_0 = const()[name = tensor("op_46205_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_46205_end_0 = const()[name = tensor("op_46205_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_46205_end_mask_0 = const()[name = tensor("op_46205_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46205_cast_fp16 = slice_by_index(begin = var_46205_begin_0, end = var_46205_end_0, end_mask = var_46205_end_mask_0, x = k_411_cast_fp16)[name = tensor("op_46205_cast_fp16")]; + tensor var_46209_begin_0 = const()[name = tensor("op_46209_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_46209_end_0 = const()[name = tensor("op_46209_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_46209_end_mask_0 = const()[name = tensor("op_46209_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46209_cast_fp16 = slice_by_index(begin = var_46209_begin_0, end = var_46209_end_0, end_mask = var_46209_end_mask_0, x = k_411_cast_fp16)[name = tensor("op_46209_cast_fp16")]; + tensor var_46213_begin_0 = const()[name = tensor("op_46213_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_46213_end_0 = const()[name = tensor("op_46213_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_46213_end_mask_0 = const()[name = tensor("op_46213_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46213_cast_fp16 = slice_by_index(begin = var_46213_begin_0, end = var_46213_end_0, end_mask = var_46213_end_mask_0, x = k_411_cast_fp16)[name = tensor("op_46213_cast_fp16")]; + tensor var_46217_begin_0 = const()[name = tensor("op_46217_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_46217_end_0 = const()[name = tensor("op_46217_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_46217_end_mask_0 = const()[name = tensor("op_46217_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46217_cast_fp16 = slice_by_index(begin = var_46217_begin_0, end = var_46217_end_0, end_mask = var_46217_end_mask_0, x = k_411_cast_fp16)[name = tensor("op_46217_cast_fp16")]; + tensor var_46221_begin_0 = const()[name = tensor("op_46221_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_46221_end_0 = const()[name = tensor("op_46221_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_46221_end_mask_0 = const()[name = tensor("op_46221_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46221_cast_fp16 = slice_by_index(begin = var_46221_begin_0, end = var_46221_end_0, end_mask = var_46221_end_mask_0, x = k_411_cast_fp16)[name = tensor("op_46221_cast_fp16")]; + tensor var_46225_begin_0 = const()[name = tensor("op_46225_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_46225_end_0 = const()[name = tensor("op_46225_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_46225_end_mask_0 = const()[name = tensor("op_46225_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46225_cast_fp16 = slice_by_index(begin = var_46225_begin_0, end = var_46225_end_0, end_mask = var_46225_end_mask_0, x = k_411_cast_fp16)[name = tensor("op_46225_cast_fp16")]; + tensor var_46229_begin_0 = const()[name = tensor("op_46229_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_46229_end_0 = const()[name = tensor("op_46229_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_46229_end_mask_0 = const()[name = tensor("op_46229_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46229_cast_fp16 = slice_by_index(begin = var_46229_begin_0, end = var_46229_end_0, end_mask = var_46229_end_mask_0, x = k_411_cast_fp16)[name = tensor("op_46229_cast_fp16")]; + tensor var_46233_begin_0 = const()[name = tensor("op_46233_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_46233_end_0 = const()[name = tensor("op_46233_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_46233_end_mask_0 = const()[name = tensor("op_46233_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46233_cast_fp16 = slice_by_index(begin = var_46233_begin_0, end = var_46233_end_0, end_mask = var_46233_end_mask_0, x = k_411_cast_fp16)[name = tensor("op_46233_cast_fp16")]; + tensor var_46237_begin_0 = const()[name = tensor("op_46237_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_46237_end_0 = const()[name = tensor("op_46237_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_46237_end_mask_0 = const()[name = tensor("op_46237_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46237_cast_fp16 = slice_by_index(begin = var_46237_begin_0, end = var_46237_end_0, end_mask = var_46237_end_mask_0, x = k_411_cast_fp16)[name = tensor("op_46237_cast_fp16")]; + tensor var_46241_begin_0 = const()[name = tensor("op_46241_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_46241_end_0 = const()[name = tensor("op_46241_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_46241_end_mask_0 = const()[name = tensor("op_46241_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46241_cast_fp16 = slice_by_index(begin = var_46241_begin_0, end = var_46241_end_0, end_mask = var_46241_end_mask_0, x = k_411_cast_fp16)[name = tensor("op_46241_cast_fp16")]; + tensor var_46245_begin_0 = const()[name = tensor("op_46245_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_46245_end_0 = const()[name = tensor("op_46245_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_46245_end_mask_0 = const()[name = tensor("op_46245_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46245_cast_fp16 = slice_by_index(begin = var_46245_begin_0, end = var_46245_end_0, end_mask = var_46245_end_mask_0, x = k_411_cast_fp16)[name = tensor("op_46245_cast_fp16")]; + tensor var_46249_begin_0 = const()[name = tensor("op_46249_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_46249_end_0 = const()[name = tensor("op_46249_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_46249_end_mask_0 = const()[name = tensor("op_46249_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46249_cast_fp16 = slice_by_index(begin = var_46249_begin_0, end = var_46249_end_0, end_mask = var_46249_end_mask_0, x = k_411_cast_fp16)[name = tensor("op_46249_cast_fp16")]; + tensor var_46253_begin_0 = const()[name = tensor("op_46253_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_46253_end_0 = const()[name = tensor("op_46253_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_46253_end_mask_0 = const()[name = tensor("op_46253_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46253_cast_fp16 = slice_by_index(begin = var_46253_begin_0, end = var_46253_end_0, end_mask = var_46253_end_mask_0, x = k_411_cast_fp16)[name = tensor("op_46253_cast_fp16")]; + tensor var_46255_begin_0 = const()[name = tensor("op_46255_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46255_end_0 = const()[name = tensor("op_46255_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_46255_end_mask_0 = const()[name = tensor("op_46255_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46255_cast_fp16 = slice_by_index(begin = var_46255_begin_0, end = var_46255_end_0, end_mask = var_46255_end_mask_0, x = v_205_cast_fp16)[name = tensor("op_46255_cast_fp16")]; + tensor var_46259_begin_0 = const()[name = tensor("op_46259_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_46259_end_0 = const()[name = tensor("op_46259_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_46259_end_mask_0 = const()[name = tensor("op_46259_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46259_cast_fp16 = slice_by_index(begin = var_46259_begin_0, end = var_46259_end_0, end_mask = var_46259_end_mask_0, x = v_205_cast_fp16)[name = tensor("op_46259_cast_fp16")]; + tensor var_46263_begin_0 = const()[name = tensor("op_46263_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_46263_end_0 = const()[name = tensor("op_46263_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_46263_end_mask_0 = const()[name = tensor("op_46263_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46263_cast_fp16 = slice_by_index(begin = var_46263_begin_0, end = var_46263_end_0, end_mask = var_46263_end_mask_0, x = v_205_cast_fp16)[name = tensor("op_46263_cast_fp16")]; + tensor var_46267_begin_0 = const()[name = tensor("op_46267_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_46267_end_0 = const()[name = tensor("op_46267_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_46267_end_mask_0 = const()[name = tensor("op_46267_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46267_cast_fp16 = slice_by_index(begin = var_46267_begin_0, end = var_46267_end_0, end_mask = var_46267_end_mask_0, x = v_205_cast_fp16)[name = tensor("op_46267_cast_fp16")]; + tensor var_46271_begin_0 = const()[name = tensor("op_46271_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_46271_end_0 = const()[name = tensor("op_46271_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_46271_end_mask_0 = const()[name = tensor("op_46271_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46271_cast_fp16 = slice_by_index(begin = var_46271_begin_0, end = var_46271_end_0, end_mask = var_46271_end_mask_0, x = v_205_cast_fp16)[name = tensor("op_46271_cast_fp16")]; + tensor var_46275_begin_0 = const()[name = tensor("op_46275_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_46275_end_0 = const()[name = tensor("op_46275_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_46275_end_mask_0 = const()[name = tensor("op_46275_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46275_cast_fp16 = slice_by_index(begin = var_46275_begin_0, end = var_46275_end_0, end_mask = var_46275_end_mask_0, x = v_205_cast_fp16)[name = tensor("op_46275_cast_fp16")]; + tensor var_46279_begin_0 = const()[name = tensor("op_46279_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_46279_end_0 = const()[name = tensor("op_46279_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_46279_end_mask_0 = const()[name = tensor("op_46279_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46279_cast_fp16 = slice_by_index(begin = var_46279_begin_0, end = var_46279_end_0, end_mask = var_46279_end_mask_0, x = v_205_cast_fp16)[name = tensor("op_46279_cast_fp16")]; + tensor var_46283_begin_0 = const()[name = tensor("op_46283_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_46283_end_0 = const()[name = tensor("op_46283_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_46283_end_mask_0 = const()[name = tensor("op_46283_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46283_cast_fp16 = slice_by_index(begin = var_46283_begin_0, end = var_46283_end_0, end_mask = var_46283_end_mask_0, x = v_205_cast_fp16)[name = tensor("op_46283_cast_fp16")]; + tensor var_46287_begin_0 = const()[name = tensor("op_46287_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_46287_end_0 = const()[name = tensor("op_46287_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_46287_end_mask_0 = const()[name = tensor("op_46287_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46287_cast_fp16 = slice_by_index(begin = var_46287_begin_0, end = var_46287_end_0, end_mask = var_46287_end_mask_0, x = v_205_cast_fp16)[name = tensor("op_46287_cast_fp16")]; + tensor var_46291_begin_0 = const()[name = tensor("op_46291_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_46291_end_0 = const()[name = tensor("op_46291_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_46291_end_mask_0 = const()[name = tensor("op_46291_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46291_cast_fp16 = slice_by_index(begin = var_46291_begin_0, end = var_46291_end_0, end_mask = var_46291_end_mask_0, x = v_205_cast_fp16)[name = tensor("op_46291_cast_fp16")]; + tensor var_46295_begin_0 = const()[name = tensor("op_46295_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_46295_end_0 = const()[name = tensor("op_46295_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_46295_end_mask_0 = const()[name = tensor("op_46295_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46295_cast_fp16 = slice_by_index(begin = var_46295_begin_0, end = var_46295_end_0, end_mask = var_46295_end_mask_0, x = v_205_cast_fp16)[name = tensor("op_46295_cast_fp16")]; + tensor var_46299_begin_0 = const()[name = tensor("op_46299_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_46299_end_0 = const()[name = tensor("op_46299_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_46299_end_mask_0 = const()[name = tensor("op_46299_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46299_cast_fp16 = slice_by_index(begin = var_46299_begin_0, end = var_46299_end_0, end_mask = var_46299_end_mask_0, x = v_205_cast_fp16)[name = tensor("op_46299_cast_fp16")]; + tensor var_46303_begin_0 = const()[name = tensor("op_46303_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_46303_end_0 = const()[name = tensor("op_46303_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_46303_end_mask_0 = const()[name = tensor("op_46303_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46303_cast_fp16 = slice_by_index(begin = var_46303_begin_0, end = var_46303_end_0, end_mask = var_46303_end_mask_0, x = v_205_cast_fp16)[name = tensor("op_46303_cast_fp16")]; + tensor var_46307_begin_0 = const()[name = tensor("op_46307_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_46307_end_0 = const()[name = tensor("op_46307_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_46307_end_mask_0 = const()[name = tensor("op_46307_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46307_cast_fp16 = slice_by_index(begin = var_46307_begin_0, end = var_46307_end_0, end_mask = var_46307_end_mask_0, x = v_205_cast_fp16)[name = tensor("op_46307_cast_fp16")]; + tensor var_46311_begin_0 = const()[name = tensor("op_46311_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_46311_end_0 = const()[name = tensor("op_46311_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_46311_end_mask_0 = const()[name = tensor("op_46311_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46311_cast_fp16 = slice_by_index(begin = var_46311_begin_0, end = var_46311_end_0, end_mask = var_46311_end_mask_0, x = v_205_cast_fp16)[name = tensor("op_46311_cast_fp16")]; + tensor var_46315_begin_0 = const()[name = tensor("op_46315_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_46315_end_0 = const()[name = tensor("op_46315_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_46315_end_mask_0 = const()[name = tensor("op_46315_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46315_cast_fp16 = slice_by_index(begin = var_46315_begin_0, end = var_46315_end_0, end_mask = var_46315_end_mask_0, x = v_205_cast_fp16)[name = tensor("op_46315_cast_fp16")]; + tensor var_46319_begin_0 = const()[name = tensor("op_46319_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_46319_end_0 = const()[name = tensor("op_46319_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_46319_end_mask_0 = const()[name = tensor("op_46319_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46319_cast_fp16 = slice_by_index(begin = var_46319_begin_0, end = var_46319_end_0, end_mask = var_46319_end_mask_0, x = v_205_cast_fp16)[name = tensor("op_46319_cast_fp16")]; + tensor var_46323_begin_0 = const()[name = tensor("op_46323_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_46323_end_0 = const()[name = tensor("op_46323_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_46323_end_mask_0 = const()[name = tensor("op_46323_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46323_cast_fp16 = slice_by_index(begin = var_46323_begin_0, end = var_46323_end_0, end_mask = var_46323_end_mask_0, x = v_205_cast_fp16)[name = tensor("op_46323_cast_fp16")]; + tensor var_46327_begin_0 = const()[name = tensor("op_46327_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_46327_end_0 = const()[name = tensor("op_46327_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_46327_end_mask_0 = const()[name = tensor("op_46327_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46327_cast_fp16 = slice_by_index(begin = var_46327_begin_0, end = var_46327_end_0, end_mask = var_46327_end_mask_0, x = v_205_cast_fp16)[name = tensor("op_46327_cast_fp16")]; + tensor var_46331_begin_0 = const()[name = tensor("op_46331_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_46331_end_0 = const()[name = tensor("op_46331_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_46331_end_mask_0 = const()[name = tensor("op_46331_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46331_cast_fp16 = slice_by_index(begin = var_46331_begin_0, end = var_46331_end_0, end_mask = var_46331_end_mask_0, x = v_205_cast_fp16)[name = tensor("op_46331_cast_fp16")]; + tensor var_46335_equation_0 = const()[name = tensor("op_46335_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46335_cast_fp16 = einsum(equation = var_46335_equation_0, values = (var_46177_cast_fp16, var_46094_cast_fp16))[name = tensor("op_46335_cast_fp16")]; + tensor var_46336_to_fp16 = const()[name = tensor("op_46336_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3921_cast_fp16 = mul(x = var_46335_cast_fp16, y = var_46336_to_fp16)[name = tensor("aw_3921_cast_fp16")]; + tensor var_46339_equation_0 = const()[name = tensor("op_46339_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46339_cast_fp16 = einsum(equation = var_46339_equation_0, values = (var_46181_cast_fp16, var_46098_cast_fp16))[name = tensor("op_46339_cast_fp16")]; + tensor var_46340_to_fp16 = const()[name = tensor("op_46340_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3923_cast_fp16 = mul(x = var_46339_cast_fp16, y = var_46340_to_fp16)[name = tensor("aw_3923_cast_fp16")]; + tensor var_46343_equation_0 = const()[name = tensor("op_46343_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46343_cast_fp16 = einsum(equation = var_46343_equation_0, values = (var_46185_cast_fp16, var_46102_cast_fp16))[name = tensor("op_46343_cast_fp16")]; + tensor var_46344_to_fp16 = const()[name = tensor("op_46344_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3925_cast_fp16 = mul(x = var_46343_cast_fp16, y = var_46344_to_fp16)[name = tensor("aw_3925_cast_fp16")]; + tensor var_46347_equation_0 = const()[name = tensor("op_46347_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46347_cast_fp16 = einsum(equation = var_46347_equation_0, values = (var_46189_cast_fp16, var_46106_cast_fp16))[name = tensor("op_46347_cast_fp16")]; + tensor var_46348_to_fp16 = const()[name = tensor("op_46348_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3927_cast_fp16 = mul(x = var_46347_cast_fp16, y = var_46348_to_fp16)[name = tensor("aw_3927_cast_fp16")]; + tensor var_46351_equation_0 = const()[name = tensor("op_46351_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46351_cast_fp16 = einsum(equation = var_46351_equation_0, values = (var_46193_cast_fp16, var_46110_cast_fp16))[name = tensor("op_46351_cast_fp16")]; + tensor var_46352_to_fp16 = const()[name = tensor("op_46352_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3929_cast_fp16 = mul(x = var_46351_cast_fp16, y = var_46352_to_fp16)[name = tensor("aw_3929_cast_fp16")]; + tensor var_46355_equation_0 = const()[name = tensor("op_46355_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46355_cast_fp16 = einsum(equation = var_46355_equation_0, values = (var_46197_cast_fp16, var_46114_cast_fp16))[name = tensor("op_46355_cast_fp16")]; + tensor var_46356_to_fp16 = const()[name = tensor("op_46356_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3931_cast_fp16 = mul(x = var_46355_cast_fp16, y = var_46356_to_fp16)[name = tensor("aw_3931_cast_fp16")]; + tensor var_46359_equation_0 = const()[name = tensor("op_46359_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46359_cast_fp16 = einsum(equation = var_46359_equation_0, values = (var_46201_cast_fp16, var_46118_cast_fp16))[name = tensor("op_46359_cast_fp16")]; + tensor var_46360_to_fp16 = const()[name = tensor("op_46360_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3933_cast_fp16 = mul(x = var_46359_cast_fp16, y = var_46360_to_fp16)[name = tensor("aw_3933_cast_fp16")]; + tensor var_46363_equation_0 = const()[name = tensor("op_46363_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46363_cast_fp16 = einsum(equation = var_46363_equation_0, values = (var_46205_cast_fp16, var_46122_cast_fp16))[name = tensor("op_46363_cast_fp16")]; + tensor var_46364_to_fp16 = const()[name = tensor("op_46364_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3935_cast_fp16 = mul(x = var_46363_cast_fp16, y = var_46364_to_fp16)[name = tensor("aw_3935_cast_fp16")]; + tensor var_46367_equation_0 = const()[name = tensor("op_46367_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46367_cast_fp16 = einsum(equation = var_46367_equation_0, values = (var_46209_cast_fp16, var_46126_cast_fp16))[name = tensor("op_46367_cast_fp16")]; + tensor var_46368_to_fp16 = const()[name = tensor("op_46368_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3937_cast_fp16 = mul(x = var_46367_cast_fp16, y = var_46368_to_fp16)[name = tensor("aw_3937_cast_fp16")]; + tensor var_46371_equation_0 = const()[name = tensor("op_46371_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46371_cast_fp16 = einsum(equation = var_46371_equation_0, values = (var_46213_cast_fp16, var_46130_cast_fp16))[name = tensor("op_46371_cast_fp16")]; + tensor var_46372_to_fp16 = const()[name = tensor("op_46372_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3939_cast_fp16 = mul(x = var_46371_cast_fp16, y = var_46372_to_fp16)[name = tensor("aw_3939_cast_fp16")]; + tensor var_46375_equation_0 = const()[name = tensor("op_46375_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46375_cast_fp16 = einsum(equation = var_46375_equation_0, values = (var_46217_cast_fp16, var_46134_cast_fp16))[name = tensor("op_46375_cast_fp16")]; + tensor var_46376_to_fp16 = const()[name = tensor("op_46376_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3941_cast_fp16 = mul(x = var_46375_cast_fp16, y = var_46376_to_fp16)[name = tensor("aw_3941_cast_fp16")]; + tensor var_46379_equation_0 = const()[name = tensor("op_46379_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46379_cast_fp16 = einsum(equation = var_46379_equation_0, values = (var_46221_cast_fp16, var_46138_cast_fp16))[name = tensor("op_46379_cast_fp16")]; + tensor var_46380_to_fp16 = const()[name = tensor("op_46380_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3943_cast_fp16 = mul(x = var_46379_cast_fp16, y = var_46380_to_fp16)[name = tensor("aw_3943_cast_fp16")]; + tensor var_46383_equation_0 = const()[name = tensor("op_46383_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46383_cast_fp16 = einsum(equation = var_46383_equation_0, values = (var_46225_cast_fp16, var_46142_cast_fp16))[name = tensor("op_46383_cast_fp16")]; + tensor var_46384_to_fp16 = const()[name = tensor("op_46384_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3945_cast_fp16 = mul(x = var_46383_cast_fp16, y = var_46384_to_fp16)[name = tensor("aw_3945_cast_fp16")]; + tensor var_46387_equation_0 = const()[name = tensor("op_46387_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46387_cast_fp16 = einsum(equation = var_46387_equation_0, values = (var_46229_cast_fp16, var_46146_cast_fp16))[name = tensor("op_46387_cast_fp16")]; + tensor var_46388_to_fp16 = const()[name = tensor("op_46388_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3947_cast_fp16 = mul(x = var_46387_cast_fp16, y = var_46388_to_fp16)[name = tensor("aw_3947_cast_fp16")]; + tensor var_46391_equation_0 = const()[name = tensor("op_46391_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46391_cast_fp16 = einsum(equation = var_46391_equation_0, values = (var_46233_cast_fp16, var_46150_cast_fp16))[name = tensor("op_46391_cast_fp16")]; + tensor var_46392_to_fp16 = const()[name = tensor("op_46392_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3949_cast_fp16 = mul(x = var_46391_cast_fp16, y = var_46392_to_fp16)[name = tensor("aw_3949_cast_fp16")]; + tensor var_46395_equation_0 = const()[name = tensor("op_46395_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46395_cast_fp16 = einsum(equation = var_46395_equation_0, values = (var_46237_cast_fp16, var_46154_cast_fp16))[name = tensor("op_46395_cast_fp16")]; + tensor var_46396_to_fp16 = const()[name = tensor("op_46396_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3951_cast_fp16 = mul(x = var_46395_cast_fp16, y = var_46396_to_fp16)[name = tensor("aw_3951_cast_fp16")]; + tensor var_46399_equation_0 = const()[name = tensor("op_46399_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46399_cast_fp16 = einsum(equation = var_46399_equation_0, values = (var_46241_cast_fp16, var_46158_cast_fp16))[name = tensor("op_46399_cast_fp16")]; + tensor var_46400_to_fp16 = const()[name = tensor("op_46400_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3953_cast_fp16 = mul(x = var_46399_cast_fp16, y = var_46400_to_fp16)[name = tensor("aw_3953_cast_fp16")]; + tensor var_46403_equation_0 = const()[name = tensor("op_46403_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46403_cast_fp16 = einsum(equation = var_46403_equation_0, values = (var_46245_cast_fp16, var_46162_cast_fp16))[name = tensor("op_46403_cast_fp16")]; + tensor var_46404_to_fp16 = const()[name = tensor("op_46404_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3955_cast_fp16 = mul(x = var_46403_cast_fp16, y = var_46404_to_fp16)[name = tensor("aw_3955_cast_fp16")]; + tensor var_46407_equation_0 = const()[name = tensor("op_46407_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46407_cast_fp16 = einsum(equation = var_46407_equation_0, values = (var_46249_cast_fp16, var_46166_cast_fp16))[name = tensor("op_46407_cast_fp16")]; + tensor var_46408_to_fp16 = const()[name = tensor("op_46408_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3957_cast_fp16 = mul(x = var_46407_cast_fp16, y = var_46408_to_fp16)[name = tensor("aw_3957_cast_fp16")]; + tensor var_46411_equation_0 = const()[name = tensor("op_46411_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46411_cast_fp16 = einsum(equation = var_46411_equation_0, values = (var_46253_cast_fp16, var_46170_cast_fp16))[name = tensor("op_46411_cast_fp16")]; + tensor var_46412_to_fp16 = const()[name = tensor("op_46412_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3959_cast_fp16 = mul(x = var_46411_cast_fp16, y = var_46412_to_fp16)[name = tensor("aw_3959_cast_fp16")]; + tensor var_46414_cast_fp16 = softmax(axis = var_30355, x = aw_3921_cast_fp16)[name = tensor("op_46414_cast_fp16")]; + tensor var_46415_cast_fp16 = softmax(axis = var_30355, x = aw_3923_cast_fp16)[name = tensor("op_46415_cast_fp16")]; + tensor var_46416_cast_fp16 = softmax(axis = var_30355, x = aw_3925_cast_fp16)[name = tensor("op_46416_cast_fp16")]; + tensor var_46417_cast_fp16 = softmax(axis = var_30355, x = aw_3927_cast_fp16)[name = tensor("op_46417_cast_fp16")]; + tensor var_46418_cast_fp16 = softmax(axis = var_30355, x = aw_3929_cast_fp16)[name = tensor("op_46418_cast_fp16")]; + tensor var_46419_cast_fp16 = softmax(axis = var_30355, x = aw_3931_cast_fp16)[name = tensor("op_46419_cast_fp16")]; + tensor var_46420_cast_fp16 = softmax(axis = var_30355, x = aw_3933_cast_fp16)[name = tensor("op_46420_cast_fp16")]; + tensor var_46421_cast_fp16 = softmax(axis = var_30355, x = aw_3935_cast_fp16)[name = tensor("op_46421_cast_fp16")]; + tensor var_46422_cast_fp16 = softmax(axis = var_30355, x = aw_3937_cast_fp16)[name = tensor("op_46422_cast_fp16")]; + tensor var_46423_cast_fp16 = softmax(axis = var_30355, x = aw_3939_cast_fp16)[name = tensor("op_46423_cast_fp16")]; + tensor var_46424_cast_fp16 = softmax(axis = var_30355, x = aw_3941_cast_fp16)[name = tensor("op_46424_cast_fp16")]; + tensor var_46425_cast_fp16 = softmax(axis = var_30355, x = aw_3943_cast_fp16)[name = tensor("op_46425_cast_fp16")]; + tensor var_46426_cast_fp16 = softmax(axis = var_30355, x = aw_3945_cast_fp16)[name = tensor("op_46426_cast_fp16")]; + tensor var_46427_cast_fp16 = softmax(axis = var_30355, x = aw_3947_cast_fp16)[name = tensor("op_46427_cast_fp16")]; + tensor var_46428_cast_fp16 = softmax(axis = var_30355, x = aw_3949_cast_fp16)[name = tensor("op_46428_cast_fp16")]; + tensor var_46429_cast_fp16 = softmax(axis = var_30355, x = aw_3951_cast_fp16)[name = tensor("op_46429_cast_fp16")]; + tensor var_46430_cast_fp16 = softmax(axis = var_30355, x = aw_3953_cast_fp16)[name = tensor("op_46430_cast_fp16")]; + tensor var_46431_cast_fp16 = softmax(axis = var_30355, x = aw_3955_cast_fp16)[name = tensor("op_46431_cast_fp16")]; + tensor var_46432_cast_fp16 = softmax(axis = var_30355, x = aw_3957_cast_fp16)[name = tensor("op_46432_cast_fp16")]; + tensor var_46433_cast_fp16 = softmax(axis = var_30355, x = aw_3959_cast_fp16)[name = tensor("op_46433_cast_fp16")]; + tensor var_46435_equation_0 = const()[name = tensor("op_46435_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_46435_cast_fp16 = einsum(equation = var_46435_equation_0, values = (var_46255_cast_fp16, var_46414_cast_fp16))[name = tensor("op_46435_cast_fp16")]; + tensor var_46437_equation_0 = const()[name = tensor("op_46437_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_46437_cast_fp16 = einsum(equation = var_46437_equation_0, values = (var_46259_cast_fp16, var_46415_cast_fp16))[name = tensor("op_46437_cast_fp16")]; + tensor var_46439_equation_0 = const()[name = tensor("op_46439_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_46439_cast_fp16 = einsum(equation = var_46439_equation_0, values = (var_46263_cast_fp16, var_46416_cast_fp16))[name = tensor("op_46439_cast_fp16")]; + tensor var_46441_equation_0 = const()[name = tensor("op_46441_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_46441_cast_fp16 = einsum(equation = var_46441_equation_0, values = (var_46267_cast_fp16, var_46417_cast_fp16))[name = tensor("op_46441_cast_fp16")]; + tensor var_46443_equation_0 = const()[name = tensor("op_46443_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_46443_cast_fp16 = einsum(equation = var_46443_equation_0, values = (var_46271_cast_fp16, var_46418_cast_fp16))[name = tensor("op_46443_cast_fp16")]; + tensor var_46445_equation_0 = const()[name = tensor("op_46445_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_46445_cast_fp16 = einsum(equation = var_46445_equation_0, values = (var_46275_cast_fp16, var_46419_cast_fp16))[name = tensor("op_46445_cast_fp16")]; + tensor var_46447_equation_0 = const()[name = tensor("op_46447_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_46447_cast_fp16 = einsum(equation = var_46447_equation_0, values = (var_46279_cast_fp16, var_46420_cast_fp16))[name = tensor("op_46447_cast_fp16")]; + tensor var_46449_equation_0 = const()[name = tensor("op_46449_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_46449_cast_fp16 = einsum(equation = var_46449_equation_0, values = (var_46283_cast_fp16, var_46421_cast_fp16))[name = tensor("op_46449_cast_fp16")]; + tensor var_46451_equation_0 = const()[name = tensor("op_46451_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_46451_cast_fp16 = einsum(equation = var_46451_equation_0, values = (var_46287_cast_fp16, var_46422_cast_fp16))[name = tensor("op_46451_cast_fp16")]; + tensor var_46453_equation_0 = const()[name = tensor("op_46453_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_46453_cast_fp16 = einsum(equation = var_46453_equation_0, values = (var_46291_cast_fp16, var_46423_cast_fp16))[name = tensor("op_46453_cast_fp16")]; + tensor var_46455_equation_0 = const()[name = tensor("op_46455_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_46455_cast_fp16 = einsum(equation = var_46455_equation_0, values = (var_46295_cast_fp16, var_46424_cast_fp16))[name = tensor("op_46455_cast_fp16")]; + tensor var_46457_equation_0 = const()[name = tensor("op_46457_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_46457_cast_fp16 = einsum(equation = var_46457_equation_0, values = (var_46299_cast_fp16, var_46425_cast_fp16))[name = tensor("op_46457_cast_fp16")]; + tensor var_46459_equation_0 = const()[name = tensor("op_46459_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_46459_cast_fp16 = einsum(equation = var_46459_equation_0, values = (var_46303_cast_fp16, var_46426_cast_fp16))[name = tensor("op_46459_cast_fp16")]; + tensor var_46461_equation_0 = const()[name = tensor("op_46461_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_46461_cast_fp16 = einsum(equation = var_46461_equation_0, values = (var_46307_cast_fp16, var_46427_cast_fp16))[name = tensor("op_46461_cast_fp16")]; + tensor var_46463_equation_0 = const()[name = tensor("op_46463_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_46463_cast_fp16 = einsum(equation = var_46463_equation_0, values = (var_46311_cast_fp16, var_46428_cast_fp16))[name = tensor("op_46463_cast_fp16")]; + tensor var_46465_equation_0 = const()[name = tensor("op_46465_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_46465_cast_fp16 = einsum(equation = var_46465_equation_0, values = (var_46315_cast_fp16, var_46429_cast_fp16))[name = tensor("op_46465_cast_fp16")]; + tensor var_46467_equation_0 = const()[name = tensor("op_46467_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_46467_cast_fp16 = einsum(equation = var_46467_equation_0, values = (var_46319_cast_fp16, var_46430_cast_fp16))[name = tensor("op_46467_cast_fp16")]; + tensor var_46469_equation_0 = const()[name = tensor("op_46469_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_46469_cast_fp16 = einsum(equation = var_46469_equation_0, values = (var_46323_cast_fp16, var_46431_cast_fp16))[name = tensor("op_46469_cast_fp16")]; + tensor var_46471_equation_0 = const()[name = tensor("op_46471_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_46471_cast_fp16 = einsum(equation = var_46471_equation_0, values = (var_46327_cast_fp16, var_46432_cast_fp16))[name = tensor("op_46471_cast_fp16")]; + tensor var_46473_equation_0 = const()[name = tensor("op_46473_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_46473_cast_fp16 = einsum(equation = var_46473_equation_0, values = (var_46331_cast_fp16, var_46433_cast_fp16))[name = tensor("op_46473_cast_fp16")]; + tensor input_593_interleave_0 = const()[name = tensor("input_593_interleave_0"), val = tensor(false)]; + tensor input_593_cast_fp16 = concat(axis = var_30355, interleave = input_593_interleave_0, values = (var_46435_cast_fp16, var_46437_cast_fp16, var_46439_cast_fp16, var_46441_cast_fp16, var_46443_cast_fp16, var_46445_cast_fp16, var_46447_cast_fp16, var_46449_cast_fp16, var_46451_cast_fp16, var_46453_cast_fp16, var_46455_cast_fp16, var_46457_cast_fp16, var_46459_cast_fp16, var_46461_cast_fp16, var_46463_cast_fp16, var_46465_cast_fp16, var_46467_cast_fp16, var_46469_cast_fp16, var_46471_cast_fp16, var_46473_cast_fp16))[name = tensor("input_593_cast_fp16")]; + tensor var_46479 = const()[name = tensor("op_46479"), val = tensor([1, 1])]; + tensor var_46481 = const()[name = tensor("op_46481"), val = tensor([1, 1])]; + tensor var_46483_pad_type_0 = const()[name = tensor("op_46483_pad_type_0"), val = tensor("custom")]; + tensor var_46483_pad_0 = const()[name = tensor("op_46483_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_7_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_7_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3884872128)))]; + tensor up_blocks_0_attentions_1_transformer_blocks_7_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_7_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3888148992)))]; + tensor var_46483_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_7_attn1_to_out_0_bias_to_fp16, dilations = var_46481, groups = var_30355, pad = var_46483_pad_0, pad_type = var_46483_pad_type_0, strides = var_46479, weight = up_blocks_0_attentions_1_transformer_blocks_7_attn1_to_out_0_weight_to_fp16, x = input_593_cast_fp16)[name = tensor("op_46483_cast_fp16")]; + tensor inputs_309_cast_fp16 = add(x = var_46483_cast_fp16, y = inputs_307_cast_fp16)[name = tensor("inputs_309_cast_fp16")]; + tensor hidden_states_405_axes_0 = const()[name = tensor("hidden_states_405_axes_0"), val = tensor([1])]; + tensor hidden_states_405_gamma_0_to_fp16 = const()[name = tensor("hidden_states_405_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3888151616)))]; + tensor hidden_states_405_beta_0_to_fp16 = const()[name = tensor("hidden_states_405_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3888154240)))]; + tensor var_46493_to_fp16 = const()[name = tensor("op_46493_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_405_cast_fp16 = layer_norm(axes = hidden_states_405_axes_0, beta = hidden_states_405_beta_0_to_fp16, epsilon = var_46493_to_fp16, gamma = hidden_states_405_gamma_0_to_fp16, x = inputs_309_cast_fp16)[name = tensor("hidden_states_405_cast_fp16")]; + tensor var_46508 = const()[name = tensor("op_46508"), val = tensor([1, 1])]; + tensor var_46510 = const()[name = tensor("op_46510"), val = tensor([1, 1])]; + tensor q_207_pad_type_0 = const()[name = tensor("q_207_pad_type_0"), val = tensor("custom")]; + tensor q_207_pad_0 = const()[name = tensor("q_207_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_7_attn2_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_7_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3888156864)))]; + tensor q_207_cast_fp16 = conv(dilations = var_46510, groups = var_30355, pad = q_207_pad_0, pad_type = q_207_pad_type_0, strides = var_46508, weight = up_blocks_0_attentions_1_transformer_blocks_7_attn2_to_q_weight_to_fp16, x = hidden_states_405_cast_fp16)[name = tensor("q_207_cast_fp16")]; + tensor var_46514 = const()[name = tensor("op_46514"), val = tensor([1, 1])]; + tensor var_46516 = const()[name = tensor("op_46516"), val = tensor([1, 1])]; + tensor k_413_pad_type_0 = const()[name = tensor("k_413_pad_type_0"), val = tensor("custom")]; + tensor k_413_pad_0 = const()[name = tensor("k_413_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_7_attn2_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_7_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3891433728)))]; + tensor k_413_cast_fp16 = conv(dilations = var_46516, groups = var_30355, pad = k_413_pad_0, pad_type = k_413_pad_type_0, strides = var_46514, weight = up_blocks_0_attentions_1_transformer_blocks_7_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_413_cast_fp16")]; + tensor var_46520 = const()[name = tensor("op_46520"), val = tensor([1, 1])]; + tensor var_46522 = const()[name = tensor("op_46522"), val = tensor([1, 1])]; + tensor v_207_pad_type_0 = const()[name = tensor("v_207_pad_type_0"), val = tensor("custom")]; + tensor v_207_pad_0 = const()[name = tensor("v_207_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_7_attn2_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_7_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3896676672)))]; + tensor v_207_cast_fp16 = conv(dilations = var_46522, groups = var_30355, pad = v_207_pad_0, pad_type = v_207_pad_type_0, strides = var_46520, weight = up_blocks_0_attentions_1_transformer_blocks_7_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_207_cast_fp16")]; + tensor var_46526_begin_0 = const()[name = tensor("op_46526_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46526_end_0 = const()[name = tensor("op_46526_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_46526_end_mask_0 = const()[name = tensor("op_46526_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46526_cast_fp16 = slice_by_index(begin = var_46526_begin_0, end = var_46526_end_0, end_mask = var_46526_end_mask_0, x = q_207_cast_fp16)[name = tensor("op_46526_cast_fp16")]; + tensor var_46530_begin_0 = const()[name = tensor("op_46530_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_46530_end_0 = const()[name = tensor("op_46530_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_46530_end_mask_0 = const()[name = tensor("op_46530_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46530_cast_fp16 = slice_by_index(begin = var_46530_begin_0, end = var_46530_end_0, end_mask = var_46530_end_mask_0, x = q_207_cast_fp16)[name = tensor("op_46530_cast_fp16")]; + tensor var_46534_begin_0 = const()[name = tensor("op_46534_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_46534_end_0 = const()[name = tensor("op_46534_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_46534_end_mask_0 = const()[name = tensor("op_46534_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46534_cast_fp16 = slice_by_index(begin = var_46534_begin_0, end = var_46534_end_0, end_mask = var_46534_end_mask_0, x = q_207_cast_fp16)[name = tensor("op_46534_cast_fp16")]; + tensor var_46538_begin_0 = const()[name = tensor("op_46538_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_46538_end_0 = const()[name = tensor("op_46538_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_46538_end_mask_0 = const()[name = tensor("op_46538_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46538_cast_fp16 = slice_by_index(begin = var_46538_begin_0, end = var_46538_end_0, end_mask = var_46538_end_mask_0, x = q_207_cast_fp16)[name = tensor("op_46538_cast_fp16")]; + tensor var_46542_begin_0 = const()[name = tensor("op_46542_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_46542_end_0 = const()[name = tensor("op_46542_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_46542_end_mask_0 = const()[name = tensor("op_46542_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46542_cast_fp16 = slice_by_index(begin = var_46542_begin_0, end = var_46542_end_0, end_mask = var_46542_end_mask_0, x = q_207_cast_fp16)[name = tensor("op_46542_cast_fp16")]; + tensor var_46546_begin_0 = const()[name = tensor("op_46546_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_46546_end_0 = const()[name = tensor("op_46546_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_46546_end_mask_0 = const()[name = tensor("op_46546_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46546_cast_fp16 = slice_by_index(begin = var_46546_begin_0, end = var_46546_end_0, end_mask = var_46546_end_mask_0, x = q_207_cast_fp16)[name = tensor("op_46546_cast_fp16")]; + tensor var_46550_begin_0 = const()[name = tensor("op_46550_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_46550_end_0 = const()[name = tensor("op_46550_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_46550_end_mask_0 = const()[name = tensor("op_46550_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46550_cast_fp16 = slice_by_index(begin = var_46550_begin_0, end = var_46550_end_0, end_mask = var_46550_end_mask_0, x = q_207_cast_fp16)[name = tensor("op_46550_cast_fp16")]; + tensor var_46554_begin_0 = const()[name = tensor("op_46554_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_46554_end_0 = const()[name = tensor("op_46554_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_46554_end_mask_0 = const()[name = tensor("op_46554_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46554_cast_fp16 = slice_by_index(begin = var_46554_begin_0, end = var_46554_end_0, end_mask = var_46554_end_mask_0, x = q_207_cast_fp16)[name = tensor("op_46554_cast_fp16")]; + tensor var_46558_begin_0 = const()[name = tensor("op_46558_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_46558_end_0 = const()[name = tensor("op_46558_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_46558_end_mask_0 = const()[name = tensor("op_46558_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46558_cast_fp16 = slice_by_index(begin = var_46558_begin_0, end = var_46558_end_0, end_mask = var_46558_end_mask_0, x = q_207_cast_fp16)[name = tensor("op_46558_cast_fp16")]; + tensor var_46562_begin_0 = const()[name = tensor("op_46562_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_46562_end_0 = const()[name = tensor("op_46562_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_46562_end_mask_0 = const()[name = tensor("op_46562_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46562_cast_fp16 = slice_by_index(begin = var_46562_begin_0, end = var_46562_end_0, end_mask = var_46562_end_mask_0, x = q_207_cast_fp16)[name = tensor("op_46562_cast_fp16")]; + tensor var_46566_begin_0 = const()[name = tensor("op_46566_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_46566_end_0 = const()[name = tensor("op_46566_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_46566_end_mask_0 = const()[name = tensor("op_46566_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46566_cast_fp16 = slice_by_index(begin = var_46566_begin_0, end = var_46566_end_0, end_mask = var_46566_end_mask_0, x = q_207_cast_fp16)[name = tensor("op_46566_cast_fp16")]; + tensor var_46570_begin_0 = const()[name = tensor("op_46570_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_46570_end_0 = const()[name = tensor("op_46570_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_46570_end_mask_0 = const()[name = tensor("op_46570_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46570_cast_fp16 = slice_by_index(begin = var_46570_begin_0, end = var_46570_end_0, end_mask = var_46570_end_mask_0, x = q_207_cast_fp16)[name = tensor("op_46570_cast_fp16")]; + tensor var_46574_begin_0 = const()[name = tensor("op_46574_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_46574_end_0 = const()[name = tensor("op_46574_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_46574_end_mask_0 = const()[name = tensor("op_46574_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46574_cast_fp16 = slice_by_index(begin = var_46574_begin_0, end = var_46574_end_0, end_mask = var_46574_end_mask_0, x = q_207_cast_fp16)[name = tensor("op_46574_cast_fp16")]; + tensor var_46578_begin_0 = const()[name = tensor("op_46578_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_46578_end_0 = const()[name = tensor("op_46578_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_46578_end_mask_0 = const()[name = tensor("op_46578_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46578_cast_fp16 = slice_by_index(begin = var_46578_begin_0, end = var_46578_end_0, end_mask = var_46578_end_mask_0, x = q_207_cast_fp16)[name = tensor("op_46578_cast_fp16")]; + tensor var_46582_begin_0 = const()[name = tensor("op_46582_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_46582_end_0 = const()[name = tensor("op_46582_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_46582_end_mask_0 = const()[name = tensor("op_46582_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46582_cast_fp16 = slice_by_index(begin = var_46582_begin_0, end = var_46582_end_0, end_mask = var_46582_end_mask_0, x = q_207_cast_fp16)[name = tensor("op_46582_cast_fp16")]; + tensor var_46586_begin_0 = const()[name = tensor("op_46586_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_46586_end_0 = const()[name = tensor("op_46586_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_46586_end_mask_0 = const()[name = tensor("op_46586_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46586_cast_fp16 = slice_by_index(begin = var_46586_begin_0, end = var_46586_end_0, end_mask = var_46586_end_mask_0, x = q_207_cast_fp16)[name = tensor("op_46586_cast_fp16")]; + tensor var_46590_begin_0 = const()[name = tensor("op_46590_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_46590_end_0 = const()[name = tensor("op_46590_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_46590_end_mask_0 = const()[name = tensor("op_46590_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46590_cast_fp16 = slice_by_index(begin = var_46590_begin_0, end = var_46590_end_0, end_mask = var_46590_end_mask_0, x = q_207_cast_fp16)[name = tensor("op_46590_cast_fp16")]; + tensor var_46594_begin_0 = const()[name = tensor("op_46594_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_46594_end_0 = const()[name = tensor("op_46594_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_46594_end_mask_0 = const()[name = tensor("op_46594_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46594_cast_fp16 = slice_by_index(begin = var_46594_begin_0, end = var_46594_end_0, end_mask = var_46594_end_mask_0, x = q_207_cast_fp16)[name = tensor("op_46594_cast_fp16")]; + tensor var_46598_begin_0 = const()[name = tensor("op_46598_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_46598_end_0 = const()[name = tensor("op_46598_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_46598_end_mask_0 = const()[name = tensor("op_46598_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46598_cast_fp16 = slice_by_index(begin = var_46598_begin_0, end = var_46598_end_0, end_mask = var_46598_end_mask_0, x = q_207_cast_fp16)[name = tensor("op_46598_cast_fp16")]; + tensor var_46602_begin_0 = const()[name = tensor("op_46602_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_46602_end_0 = const()[name = tensor("op_46602_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_46602_end_mask_0 = const()[name = tensor("op_46602_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46602_cast_fp16 = slice_by_index(begin = var_46602_begin_0, end = var_46602_end_0, end_mask = var_46602_end_mask_0, x = q_207_cast_fp16)[name = tensor("op_46602_cast_fp16")]; + tensor k_415_perm_0 = const()[name = tensor("k_415_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_46609_begin_0 = const()[name = tensor("op_46609_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46609_end_0 = const()[name = tensor("op_46609_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_46609_end_mask_0 = const()[name = tensor("op_46609_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_415_cast_fp16 = transpose(perm = k_415_perm_0, x = k_413_cast_fp16)[name = tensor("transpose_36")]; + tensor var_46609_cast_fp16 = slice_by_index(begin = var_46609_begin_0, end = var_46609_end_0, end_mask = var_46609_end_mask_0, x = k_415_cast_fp16)[name = tensor("op_46609_cast_fp16")]; + tensor var_46613_begin_0 = const()[name = tensor("op_46613_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_46613_end_0 = const()[name = tensor("op_46613_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_46613_end_mask_0 = const()[name = tensor("op_46613_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46613_cast_fp16 = slice_by_index(begin = var_46613_begin_0, end = var_46613_end_0, end_mask = var_46613_end_mask_0, x = k_415_cast_fp16)[name = tensor("op_46613_cast_fp16")]; + tensor var_46617_begin_0 = const()[name = tensor("op_46617_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_46617_end_0 = const()[name = tensor("op_46617_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_46617_end_mask_0 = const()[name = tensor("op_46617_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46617_cast_fp16 = slice_by_index(begin = var_46617_begin_0, end = var_46617_end_0, end_mask = var_46617_end_mask_0, x = k_415_cast_fp16)[name = tensor("op_46617_cast_fp16")]; + tensor var_46621_begin_0 = const()[name = tensor("op_46621_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_46621_end_0 = const()[name = tensor("op_46621_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_46621_end_mask_0 = const()[name = tensor("op_46621_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46621_cast_fp16 = slice_by_index(begin = var_46621_begin_0, end = var_46621_end_0, end_mask = var_46621_end_mask_0, x = k_415_cast_fp16)[name = tensor("op_46621_cast_fp16")]; + tensor var_46625_begin_0 = const()[name = tensor("op_46625_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_46625_end_0 = const()[name = tensor("op_46625_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_46625_end_mask_0 = const()[name = tensor("op_46625_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46625_cast_fp16 = slice_by_index(begin = var_46625_begin_0, end = var_46625_end_0, end_mask = var_46625_end_mask_0, x = k_415_cast_fp16)[name = tensor("op_46625_cast_fp16")]; + tensor var_46629_begin_0 = const()[name = tensor("op_46629_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_46629_end_0 = const()[name = tensor("op_46629_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_46629_end_mask_0 = const()[name = tensor("op_46629_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46629_cast_fp16 = slice_by_index(begin = var_46629_begin_0, end = var_46629_end_0, end_mask = var_46629_end_mask_0, x = k_415_cast_fp16)[name = tensor("op_46629_cast_fp16")]; + tensor var_46633_begin_0 = const()[name = tensor("op_46633_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_46633_end_0 = const()[name = tensor("op_46633_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_46633_end_mask_0 = const()[name = tensor("op_46633_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46633_cast_fp16 = slice_by_index(begin = var_46633_begin_0, end = var_46633_end_0, end_mask = var_46633_end_mask_0, x = k_415_cast_fp16)[name = tensor("op_46633_cast_fp16")]; + tensor var_46637_begin_0 = const()[name = tensor("op_46637_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_46637_end_0 = const()[name = tensor("op_46637_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_46637_end_mask_0 = const()[name = tensor("op_46637_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46637_cast_fp16 = slice_by_index(begin = var_46637_begin_0, end = var_46637_end_0, end_mask = var_46637_end_mask_0, x = k_415_cast_fp16)[name = tensor("op_46637_cast_fp16")]; + tensor var_46641_begin_0 = const()[name = tensor("op_46641_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_46641_end_0 = const()[name = tensor("op_46641_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_46641_end_mask_0 = const()[name = tensor("op_46641_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46641_cast_fp16 = slice_by_index(begin = var_46641_begin_0, end = var_46641_end_0, end_mask = var_46641_end_mask_0, x = k_415_cast_fp16)[name = tensor("op_46641_cast_fp16")]; + tensor var_46645_begin_0 = const()[name = tensor("op_46645_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_46645_end_0 = const()[name = tensor("op_46645_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_46645_end_mask_0 = const()[name = tensor("op_46645_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46645_cast_fp16 = slice_by_index(begin = var_46645_begin_0, end = var_46645_end_0, end_mask = var_46645_end_mask_0, x = k_415_cast_fp16)[name = tensor("op_46645_cast_fp16")]; + tensor var_46649_begin_0 = const()[name = tensor("op_46649_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_46649_end_0 = const()[name = tensor("op_46649_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_46649_end_mask_0 = const()[name = tensor("op_46649_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46649_cast_fp16 = slice_by_index(begin = var_46649_begin_0, end = var_46649_end_0, end_mask = var_46649_end_mask_0, x = k_415_cast_fp16)[name = tensor("op_46649_cast_fp16")]; + tensor var_46653_begin_0 = const()[name = tensor("op_46653_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_46653_end_0 = const()[name = tensor("op_46653_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_46653_end_mask_0 = const()[name = tensor("op_46653_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46653_cast_fp16 = slice_by_index(begin = var_46653_begin_0, end = var_46653_end_0, end_mask = var_46653_end_mask_0, x = k_415_cast_fp16)[name = tensor("op_46653_cast_fp16")]; + tensor var_46657_begin_0 = const()[name = tensor("op_46657_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_46657_end_0 = const()[name = tensor("op_46657_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_46657_end_mask_0 = const()[name = tensor("op_46657_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46657_cast_fp16 = slice_by_index(begin = var_46657_begin_0, end = var_46657_end_0, end_mask = var_46657_end_mask_0, x = k_415_cast_fp16)[name = tensor("op_46657_cast_fp16")]; + tensor var_46661_begin_0 = const()[name = tensor("op_46661_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_46661_end_0 = const()[name = tensor("op_46661_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_46661_end_mask_0 = const()[name = tensor("op_46661_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46661_cast_fp16 = slice_by_index(begin = var_46661_begin_0, end = var_46661_end_0, end_mask = var_46661_end_mask_0, x = k_415_cast_fp16)[name = tensor("op_46661_cast_fp16")]; + tensor var_46665_begin_0 = const()[name = tensor("op_46665_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_46665_end_0 = const()[name = tensor("op_46665_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_46665_end_mask_0 = const()[name = tensor("op_46665_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46665_cast_fp16 = slice_by_index(begin = var_46665_begin_0, end = var_46665_end_0, end_mask = var_46665_end_mask_0, x = k_415_cast_fp16)[name = tensor("op_46665_cast_fp16")]; + tensor var_46669_begin_0 = const()[name = tensor("op_46669_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_46669_end_0 = const()[name = tensor("op_46669_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_46669_end_mask_0 = const()[name = tensor("op_46669_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46669_cast_fp16 = slice_by_index(begin = var_46669_begin_0, end = var_46669_end_0, end_mask = var_46669_end_mask_0, x = k_415_cast_fp16)[name = tensor("op_46669_cast_fp16")]; + tensor var_46673_begin_0 = const()[name = tensor("op_46673_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_46673_end_0 = const()[name = tensor("op_46673_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_46673_end_mask_0 = const()[name = tensor("op_46673_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46673_cast_fp16 = slice_by_index(begin = var_46673_begin_0, end = var_46673_end_0, end_mask = var_46673_end_mask_0, x = k_415_cast_fp16)[name = tensor("op_46673_cast_fp16")]; + tensor var_46677_begin_0 = const()[name = tensor("op_46677_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_46677_end_0 = const()[name = tensor("op_46677_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_46677_end_mask_0 = const()[name = tensor("op_46677_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46677_cast_fp16 = slice_by_index(begin = var_46677_begin_0, end = var_46677_end_0, end_mask = var_46677_end_mask_0, x = k_415_cast_fp16)[name = tensor("op_46677_cast_fp16")]; + tensor var_46681_begin_0 = const()[name = tensor("op_46681_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_46681_end_0 = const()[name = tensor("op_46681_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_46681_end_mask_0 = const()[name = tensor("op_46681_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46681_cast_fp16 = slice_by_index(begin = var_46681_begin_0, end = var_46681_end_0, end_mask = var_46681_end_mask_0, x = k_415_cast_fp16)[name = tensor("op_46681_cast_fp16")]; + tensor var_46685_begin_0 = const()[name = tensor("op_46685_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_46685_end_0 = const()[name = tensor("op_46685_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_46685_end_mask_0 = const()[name = tensor("op_46685_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46685_cast_fp16 = slice_by_index(begin = var_46685_begin_0, end = var_46685_end_0, end_mask = var_46685_end_mask_0, x = k_415_cast_fp16)[name = tensor("op_46685_cast_fp16")]; + tensor var_46687_begin_0 = const()[name = tensor("op_46687_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46687_end_0 = const()[name = tensor("op_46687_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_46687_end_mask_0 = const()[name = tensor("op_46687_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46687_cast_fp16 = slice_by_index(begin = var_46687_begin_0, end = var_46687_end_0, end_mask = var_46687_end_mask_0, x = v_207_cast_fp16)[name = tensor("op_46687_cast_fp16")]; + tensor var_46691_begin_0 = const()[name = tensor("op_46691_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_46691_end_0 = const()[name = tensor("op_46691_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_46691_end_mask_0 = const()[name = tensor("op_46691_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46691_cast_fp16 = slice_by_index(begin = var_46691_begin_0, end = var_46691_end_0, end_mask = var_46691_end_mask_0, x = v_207_cast_fp16)[name = tensor("op_46691_cast_fp16")]; + tensor var_46695_begin_0 = const()[name = tensor("op_46695_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_46695_end_0 = const()[name = tensor("op_46695_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_46695_end_mask_0 = const()[name = tensor("op_46695_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46695_cast_fp16 = slice_by_index(begin = var_46695_begin_0, end = var_46695_end_0, end_mask = var_46695_end_mask_0, x = v_207_cast_fp16)[name = tensor("op_46695_cast_fp16")]; + tensor var_46699_begin_0 = const()[name = tensor("op_46699_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_46699_end_0 = const()[name = tensor("op_46699_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_46699_end_mask_0 = const()[name = tensor("op_46699_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46699_cast_fp16 = slice_by_index(begin = var_46699_begin_0, end = var_46699_end_0, end_mask = var_46699_end_mask_0, x = v_207_cast_fp16)[name = tensor("op_46699_cast_fp16")]; + tensor var_46703_begin_0 = const()[name = tensor("op_46703_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_46703_end_0 = const()[name = tensor("op_46703_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_46703_end_mask_0 = const()[name = tensor("op_46703_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46703_cast_fp16 = slice_by_index(begin = var_46703_begin_0, end = var_46703_end_0, end_mask = var_46703_end_mask_0, x = v_207_cast_fp16)[name = tensor("op_46703_cast_fp16")]; + tensor var_46707_begin_0 = const()[name = tensor("op_46707_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_46707_end_0 = const()[name = tensor("op_46707_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_46707_end_mask_0 = const()[name = tensor("op_46707_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46707_cast_fp16 = slice_by_index(begin = var_46707_begin_0, end = var_46707_end_0, end_mask = var_46707_end_mask_0, x = v_207_cast_fp16)[name = tensor("op_46707_cast_fp16")]; + tensor var_46711_begin_0 = const()[name = tensor("op_46711_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_46711_end_0 = const()[name = tensor("op_46711_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_46711_end_mask_0 = const()[name = tensor("op_46711_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46711_cast_fp16 = slice_by_index(begin = var_46711_begin_0, end = var_46711_end_0, end_mask = var_46711_end_mask_0, x = v_207_cast_fp16)[name = tensor("op_46711_cast_fp16")]; + tensor var_46715_begin_0 = const()[name = tensor("op_46715_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_46715_end_0 = const()[name = tensor("op_46715_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_46715_end_mask_0 = const()[name = tensor("op_46715_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46715_cast_fp16 = slice_by_index(begin = var_46715_begin_0, end = var_46715_end_0, end_mask = var_46715_end_mask_0, x = v_207_cast_fp16)[name = tensor("op_46715_cast_fp16")]; + tensor var_46719_begin_0 = const()[name = tensor("op_46719_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_46719_end_0 = const()[name = tensor("op_46719_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_46719_end_mask_0 = const()[name = tensor("op_46719_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46719_cast_fp16 = slice_by_index(begin = var_46719_begin_0, end = var_46719_end_0, end_mask = var_46719_end_mask_0, x = v_207_cast_fp16)[name = tensor("op_46719_cast_fp16")]; + tensor var_46723_begin_0 = const()[name = tensor("op_46723_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_46723_end_0 = const()[name = tensor("op_46723_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_46723_end_mask_0 = const()[name = tensor("op_46723_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46723_cast_fp16 = slice_by_index(begin = var_46723_begin_0, end = var_46723_end_0, end_mask = var_46723_end_mask_0, x = v_207_cast_fp16)[name = tensor("op_46723_cast_fp16")]; + tensor var_46727_begin_0 = const()[name = tensor("op_46727_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_46727_end_0 = const()[name = tensor("op_46727_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_46727_end_mask_0 = const()[name = tensor("op_46727_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46727_cast_fp16 = slice_by_index(begin = var_46727_begin_0, end = var_46727_end_0, end_mask = var_46727_end_mask_0, x = v_207_cast_fp16)[name = tensor("op_46727_cast_fp16")]; + tensor var_46731_begin_0 = const()[name = tensor("op_46731_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_46731_end_0 = const()[name = tensor("op_46731_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_46731_end_mask_0 = const()[name = tensor("op_46731_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46731_cast_fp16 = slice_by_index(begin = var_46731_begin_0, end = var_46731_end_0, end_mask = var_46731_end_mask_0, x = v_207_cast_fp16)[name = tensor("op_46731_cast_fp16")]; + tensor var_46735_begin_0 = const()[name = tensor("op_46735_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_46735_end_0 = const()[name = tensor("op_46735_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_46735_end_mask_0 = const()[name = tensor("op_46735_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46735_cast_fp16 = slice_by_index(begin = var_46735_begin_0, end = var_46735_end_0, end_mask = var_46735_end_mask_0, x = v_207_cast_fp16)[name = tensor("op_46735_cast_fp16")]; + tensor var_46739_begin_0 = const()[name = tensor("op_46739_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_46739_end_0 = const()[name = tensor("op_46739_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_46739_end_mask_0 = const()[name = tensor("op_46739_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46739_cast_fp16 = slice_by_index(begin = var_46739_begin_0, end = var_46739_end_0, end_mask = var_46739_end_mask_0, x = v_207_cast_fp16)[name = tensor("op_46739_cast_fp16")]; + tensor var_46743_begin_0 = const()[name = tensor("op_46743_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_46743_end_0 = const()[name = tensor("op_46743_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_46743_end_mask_0 = const()[name = tensor("op_46743_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46743_cast_fp16 = slice_by_index(begin = var_46743_begin_0, end = var_46743_end_0, end_mask = var_46743_end_mask_0, x = v_207_cast_fp16)[name = tensor("op_46743_cast_fp16")]; + tensor var_46747_begin_0 = const()[name = tensor("op_46747_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_46747_end_0 = const()[name = tensor("op_46747_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_46747_end_mask_0 = const()[name = tensor("op_46747_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46747_cast_fp16 = slice_by_index(begin = var_46747_begin_0, end = var_46747_end_0, end_mask = var_46747_end_mask_0, x = v_207_cast_fp16)[name = tensor("op_46747_cast_fp16")]; + tensor var_46751_begin_0 = const()[name = tensor("op_46751_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_46751_end_0 = const()[name = tensor("op_46751_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_46751_end_mask_0 = const()[name = tensor("op_46751_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46751_cast_fp16 = slice_by_index(begin = var_46751_begin_0, end = var_46751_end_0, end_mask = var_46751_end_mask_0, x = v_207_cast_fp16)[name = tensor("op_46751_cast_fp16")]; + tensor var_46755_begin_0 = const()[name = tensor("op_46755_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_46755_end_0 = const()[name = tensor("op_46755_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_46755_end_mask_0 = const()[name = tensor("op_46755_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46755_cast_fp16 = slice_by_index(begin = var_46755_begin_0, end = var_46755_end_0, end_mask = var_46755_end_mask_0, x = v_207_cast_fp16)[name = tensor("op_46755_cast_fp16")]; + tensor var_46759_begin_0 = const()[name = tensor("op_46759_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_46759_end_0 = const()[name = tensor("op_46759_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_46759_end_mask_0 = const()[name = tensor("op_46759_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46759_cast_fp16 = slice_by_index(begin = var_46759_begin_0, end = var_46759_end_0, end_mask = var_46759_end_mask_0, x = v_207_cast_fp16)[name = tensor("op_46759_cast_fp16")]; + tensor var_46763_begin_0 = const()[name = tensor("op_46763_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_46763_end_0 = const()[name = tensor("op_46763_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_46763_end_mask_0 = const()[name = tensor("op_46763_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46763_cast_fp16 = slice_by_index(begin = var_46763_begin_0, end = var_46763_end_0, end_mask = var_46763_end_mask_0, x = v_207_cast_fp16)[name = tensor("op_46763_cast_fp16")]; + tensor var_46767_equation_0 = const()[name = tensor("op_46767_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46767_cast_fp16 = einsum(equation = var_46767_equation_0, values = (var_46609_cast_fp16, var_46526_cast_fp16))[name = tensor("op_46767_cast_fp16")]; + tensor var_46768_to_fp16 = const()[name = tensor("op_46768_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3961_cast_fp16 = mul(x = var_46767_cast_fp16, y = var_46768_to_fp16)[name = tensor("aw_3961_cast_fp16")]; + tensor var_46771_equation_0 = const()[name = tensor("op_46771_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46771_cast_fp16 = einsum(equation = var_46771_equation_0, values = (var_46613_cast_fp16, var_46530_cast_fp16))[name = tensor("op_46771_cast_fp16")]; + tensor var_46772_to_fp16 = const()[name = tensor("op_46772_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3963_cast_fp16 = mul(x = var_46771_cast_fp16, y = var_46772_to_fp16)[name = tensor("aw_3963_cast_fp16")]; + tensor var_46775_equation_0 = const()[name = tensor("op_46775_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46775_cast_fp16 = einsum(equation = var_46775_equation_0, values = (var_46617_cast_fp16, var_46534_cast_fp16))[name = tensor("op_46775_cast_fp16")]; + tensor var_46776_to_fp16 = const()[name = tensor("op_46776_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3965_cast_fp16 = mul(x = var_46775_cast_fp16, y = var_46776_to_fp16)[name = tensor("aw_3965_cast_fp16")]; + tensor var_46779_equation_0 = const()[name = tensor("op_46779_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46779_cast_fp16 = einsum(equation = var_46779_equation_0, values = (var_46621_cast_fp16, var_46538_cast_fp16))[name = tensor("op_46779_cast_fp16")]; + tensor var_46780_to_fp16 = const()[name = tensor("op_46780_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3967_cast_fp16 = mul(x = var_46779_cast_fp16, y = var_46780_to_fp16)[name = tensor("aw_3967_cast_fp16")]; + tensor var_46783_equation_0 = const()[name = tensor("op_46783_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46783_cast_fp16 = einsum(equation = var_46783_equation_0, values = (var_46625_cast_fp16, var_46542_cast_fp16))[name = tensor("op_46783_cast_fp16")]; + tensor var_46784_to_fp16 = const()[name = tensor("op_46784_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3969_cast_fp16 = mul(x = var_46783_cast_fp16, y = var_46784_to_fp16)[name = tensor("aw_3969_cast_fp16")]; + tensor var_46787_equation_0 = const()[name = tensor("op_46787_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46787_cast_fp16 = einsum(equation = var_46787_equation_0, values = (var_46629_cast_fp16, var_46546_cast_fp16))[name = tensor("op_46787_cast_fp16")]; + tensor var_46788_to_fp16 = const()[name = tensor("op_46788_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3971_cast_fp16 = mul(x = var_46787_cast_fp16, y = var_46788_to_fp16)[name = tensor("aw_3971_cast_fp16")]; + tensor var_46791_equation_0 = const()[name = tensor("op_46791_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46791_cast_fp16 = einsum(equation = var_46791_equation_0, values = (var_46633_cast_fp16, var_46550_cast_fp16))[name = tensor("op_46791_cast_fp16")]; + tensor var_46792_to_fp16 = const()[name = tensor("op_46792_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3973_cast_fp16 = mul(x = var_46791_cast_fp16, y = var_46792_to_fp16)[name = tensor("aw_3973_cast_fp16")]; + tensor var_46795_equation_0 = const()[name = tensor("op_46795_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46795_cast_fp16 = einsum(equation = var_46795_equation_0, values = (var_46637_cast_fp16, var_46554_cast_fp16))[name = tensor("op_46795_cast_fp16")]; + tensor var_46796_to_fp16 = const()[name = tensor("op_46796_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3975_cast_fp16 = mul(x = var_46795_cast_fp16, y = var_46796_to_fp16)[name = tensor("aw_3975_cast_fp16")]; + tensor var_46799_equation_0 = const()[name = tensor("op_46799_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46799_cast_fp16 = einsum(equation = var_46799_equation_0, values = (var_46641_cast_fp16, var_46558_cast_fp16))[name = tensor("op_46799_cast_fp16")]; + tensor var_46800_to_fp16 = const()[name = tensor("op_46800_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3977_cast_fp16 = mul(x = var_46799_cast_fp16, y = var_46800_to_fp16)[name = tensor("aw_3977_cast_fp16")]; + tensor var_46803_equation_0 = const()[name = tensor("op_46803_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46803_cast_fp16 = einsum(equation = var_46803_equation_0, values = (var_46645_cast_fp16, var_46562_cast_fp16))[name = tensor("op_46803_cast_fp16")]; + tensor var_46804_to_fp16 = const()[name = tensor("op_46804_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3979_cast_fp16 = mul(x = var_46803_cast_fp16, y = var_46804_to_fp16)[name = tensor("aw_3979_cast_fp16")]; + tensor var_46807_equation_0 = const()[name = tensor("op_46807_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46807_cast_fp16 = einsum(equation = var_46807_equation_0, values = (var_46649_cast_fp16, var_46566_cast_fp16))[name = tensor("op_46807_cast_fp16")]; + tensor var_46808_to_fp16 = const()[name = tensor("op_46808_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3981_cast_fp16 = mul(x = var_46807_cast_fp16, y = var_46808_to_fp16)[name = tensor("aw_3981_cast_fp16")]; + tensor var_46811_equation_0 = const()[name = tensor("op_46811_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46811_cast_fp16 = einsum(equation = var_46811_equation_0, values = (var_46653_cast_fp16, var_46570_cast_fp16))[name = tensor("op_46811_cast_fp16")]; + tensor var_46812_to_fp16 = const()[name = tensor("op_46812_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3983_cast_fp16 = mul(x = var_46811_cast_fp16, y = var_46812_to_fp16)[name = tensor("aw_3983_cast_fp16")]; + tensor var_46815_equation_0 = const()[name = tensor("op_46815_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46815_cast_fp16 = einsum(equation = var_46815_equation_0, values = (var_46657_cast_fp16, var_46574_cast_fp16))[name = tensor("op_46815_cast_fp16")]; + tensor var_46816_to_fp16 = const()[name = tensor("op_46816_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3985_cast_fp16 = mul(x = var_46815_cast_fp16, y = var_46816_to_fp16)[name = tensor("aw_3985_cast_fp16")]; + tensor var_46819_equation_0 = const()[name = tensor("op_46819_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46819_cast_fp16 = einsum(equation = var_46819_equation_0, values = (var_46661_cast_fp16, var_46578_cast_fp16))[name = tensor("op_46819_cast_fp16")]; + tensor var_46820_to_fp16 = const()[name = tensor("op_46820_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3987_cast_fp16 = mul(x = var_46819_cast_fp16, y = var_46820_to_fp16)[name = tensor("aw_3987_cast_fp16")]; + tensor var_46823_equation_0 = const()[name = tensor("op_46823_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46823_cast_fp16 = einsum(equation = var_46823_equation_0, values = (var_46665_cast_fp16, var_46582_cast_fp16))[name = tensor("op_46823_cast_fp16")]; + tensor var_46824_to_fp16 = const()[name = tensor("op_46824_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3989_cast_fp16 = mul(x = var_46823_cast_fp16, y = var_46824_to_fp16)[name = tensor("aw_3989_cast_fp16")]; + tensor var_46827_equation_0 = const()[name = tensor("op_46827_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46827_cast_fp16 = einsum(equation = var_46827_equation_0, values = (var_46669_cast_fp16, var_46586_cast_fp16))[name = tensor("op_46827_cast_fp16")]; + tensor var_46828_to_fp16 = const()[name = tensor("op_46828_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3991_cast_fp16 = mul(x = var_46827_cast_fp16, y = var_46828_to_fp16)[name = tensor("aw_3991_cast_fp16")]; + tensor var_46831_equation_0 = const()[name = tensor("op_46831_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46831_cast_fp16 = einsum(equation = var_46831_equation_0, values = (var_46673_cast_fp16, var_46590_cast_fp16))[name = tensor("op_46831_cast_fp16")]; + tensor var_46832_to_fp16 = const()[name = tensor("op_46832_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3993_cast_fp16 = mul(x = var_46831_cast_fp16, y = var_46832_to_fp16)[name = tensor("aw_3993_cast_fp16")]; + tensor var_46835_equation_0 = const()[name = tensor("op_46835_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46835_cast_fp16 = einsum(equation = var_46835_equation_0, values = (var_46677_cast_fp16, var_46594_cast_fp16))[name = tensor("op_46835_cast_fp16")]; + tensor var_46836_to_fp16 = const()[name = tensor("op_46836_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3995_cast_fp16 = mul(x = var_46835_cast_fp16, y = var_46836_to_fp16)[name = tensor("aw_3995_cast_fp16")]; + tensor var_46839_equation_0 = const()[name = tensor("op_46839_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46839_cast_fp16 = einsum(equation = var_46839_equation_0, values = (var_46681_cast_fp16, var_46598_cast_fp16))[name = tensor("op_46839_cast_fp16")]; + tensor var_46840_to_fp16 = const()[name = tensor("op_46840_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3997_cast_fp16 = mul(x = var_46839_cast_fp16, y = var_46840_to_fp16)[name = tensor("aw_3997_cast_fp16")]; + tensor var_46843_equation_0 = const()[name = tensor("op_46843_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46843_cast_fp16 = einsum(equation = var_46843_equation_0, values = (var_46685_cast_fp16, var_46602_cast_fp16))[name = tensor("op_46843_cast_fp16")]; + tensor var_46844_to_fp16 = const()[name = tensor("op_46844_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3999_cast_fp16 = mul(x = var_46843_cast_fp16, y = var_46844_to_fp16)[name = tensor("aw_3999_cast_fp16")]; + tensor var_46846_cast_fp16 = softmax(axis = var_30355, x = aw_3961_cast_fp16)[name = tensor("op_46846_cast_fp16")]; + tensor var_46847_cast_fp16 = softmax(axis = var_30355, x = aw_3963_cast_fp16)[name = tensor("op_46847_cast_fp16")]; + tensor var_46848_cast_fp16 = softmax(axis = var_30355, x = aw_3965_cast_fp16)[name = tensor("op_46848_cast_fp16")]; + tensor var_46849_cast_fp16 = softmax(axis = var_30355, x = aw_3967_cast_fp16)[name = tensor("op_46849_cast_fp16")]; + tensor var_46850_cast_fp16 = softmax(axis = var_30355, x = aw_3969_cast_fp16)[name = tensor("op_46850_cast_fp16")]; + tensor var_46851_cast_fp16 = softmax(axis = var_30355, x = aw_3971_cast_fp16)[name = tensor("op_46851_cast_fp16")]; + tensor var_46852_cast_fp16 = softmax(axis = var_30355, x = aw_3973_cast_fp16)[name = tensor("op_46852_cast_fp16")]; + tensor var_46853_cast_fp16 = softmax(axis = var_30355, x = aw_3975_cast_fp16)[name = tensor("op_46853_cast_fp16")]; + tensor var_46854_cast_fp16 = softmax(axis = var_30355, x = aw_3977_cast_fp16)[name = tensor("op_46854_cast_fp16")]; + tensor var_46855_cast_fp16 = softmax(axis = var_30355, x = aw_3979_cast_fp16)[name = tensor("op_46855_cast_fp16")]; + tensor var_46856_cast_fp16 = softmax(axis = var_30355, x = aw_3981_cast_fp16)[name = tensor("op_46856_cast_fp16")]; + tensor var_46857_cast_fp16 = softmax(axis = var_30355, x = aw_3983_cast_fp16)[name = tensor("op_46857_cast_fp16")]; + tensor var_46858_cast_fp16 = softmax(axis = var_30355, x = aw_3985_cast_fp16)[name = tensor("op_46858_cast_fp16")]; + tensor var_46859_cast_fp16 = softmax(axis = var_30355, x = aw_3987_cast_fp16)[name = tensor("op_46859_cast_fp16")]; + tensor var_46860_cast_fp16 = softmax(axis = var_30355, x = aw_3989_cast_fp16)[name = tensor("op_46860_cast_fp16")]; + tensor var_46861_cast_fp16 = softmax(axis = var_30355, x = aw_3991_cast_fp16)[name = tensor("op_46861_cast_fp16")]; + tensor var_46862_cast_fp16 = softmax(axis = var_30355, x = aw_3993_cast_fp16)[name = tensor("op_46862_cast_fp16")]; + tensor var_46863_cast_fp16 = softmax(axis = var_30355, x = aw_3995_cast_fp16)[name = tensor("op_46863_cast_fp16")]; + tensor var_46864_cast_fp16 = softmax(axis = var_30355, x = aw_3997_cast_fp16)[name = tensor("op_46864_cast_fp16")]; + tensor var_46865_cast_fp16 = softmax(axis = var_30355, x = aw_3999_cast_fp16)[name = tensor("op_46865_cast_fp16")]; + tensor var_46867_equation_0 = const()[name = tensor("op_46867_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_46867_cast_fp16 = einsum(equation = var_46867_equation_0, values = (var_46687_cast_fp16, var_46846_cast_fp16))[name = tensor("op_46867_cast_fp16")]; + tensor var_46869_equation_0 = const()[name = tensor("op_46869_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_46869_cast_fp16 = einsum(equation = var_46869_equation_0, values = (var_46691_cast_fp16, var_46847_cast_fp16))[name = tensor("op_46869_cast_fp16")]; + tensor var_46871_equation_0 = const()[name = tensor("op_46871_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_46871_cast_fp16 = einsum(equation = var_46871_equation_0, values = (var_46695_cast_fp16, var_46848_cast_fp16))[name = tensor("op_46871_cast_fp16")]; + tensor var_46873_equation_0 = const()[name = tensor("op_46873_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_46873_cast_fp16 = einsum(equation = var_46873_equation_0, values = (var_46699_cast_fp16, var_46849_cast_fp16))[name = tensor("op_46873_cast_fp16")]; + tensor var_46875_equation_0 = const()[name = tensor("op_46875_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_46875_cast_fp16 = einsum(equation = var_46875_equation_0, values = (var_46703_cast_fp16, var_46850_cast_fp16))[name = tensor("op_46875_cast_fp16")]; + tensor var_46877_equation_0 = const()[name = tensor("op_46877_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_46877_cast_fp16 = einsum(equation = var_46877_equation_0, values = (var_46707_cast_fp16, var_46851_cast_fp16))[name = tensor("op_46877_cast_fp16")]; + tensor var_46879_equation_0 = const()[name = tensor("op_46879_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_46879_cast_fp16 = einsum(equation = var_46879_equation_0, values = (var_46711_cast_fp16, var_46852_cast_fp16))[name = tensor("op_46879_cast_fp16")]; + tensor var_46881_equation_0 = const()[name = tensor("op_46881_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_46881_cast_fp16 = einsum(equation = var_46881_equation_0, values = (var_46715_cast_fp16, var_46853_cast_fp16))[name = tensor("op_46881_cast_fp16")]; + tensor var_46883_equation_0 = const()[name = tensor("op_46883_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_46883_cast_fp16 = einsum(equation = var_46883_equation_0, values = (var_46719_cast_fp16, var_46854_cast_fp16))[name = tensor("op_46883_cast_fp16")]; + tensor var_46885_equation_0 = const()[name = tensor("op_46885_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_46885_cast_fp16 = einsum(equation = var_46885_equation_0, values = (var_46723_cast_fp16, var_46855_cast_fp16))[name = tensor("op_46885_cast_fp16")]; + tensor var_46887_equation_0 = const()[name = tensor("op_46887_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_46887_cast_fp16 = einsum(equation = var_46887_equation_0, values = (var_46727_cast_fp16, var_46856_cast_fp16))[name = tensor("op_46887_cast_fp16")]; + tensor var_46889_equation_0 = const()[name = tensor("op_46889_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_46889_cast_fp16 = einsum(equation = var_46889_equation_0, values = (var_46731_cast_fp16, var_46857_cast_fp16))[name = tensor("op_46889_cast_fp16")]; + tensor var_46891_equation_0 = const()[name = tensor("op_46891_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_46891_cast_fp16 = einsum(equation = var_46891_equation_0, values = (var_46735_cast_fp16, var_46858_cast_fp16))[name = tensor("op_46891_cast_fp16")]; + tensor var_46893_equation_0 = const()[name = tensor("op_46893_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_46893_cast_fp16 = einsum(equation = var_46893_equation_0, values = (var_46739_cast_fp16, var_46859_cast_fp16))[name = tensor("op_46893_cast_fp16")]; + tensor var_46895_equation_0 = const()[name = tensor("op_46895_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_46895_cast_fp16 = einsum(equation = var_46895_equation_0, values = (var_46743_cast_fp16, var_46860_cast_fp16))[name = tensor("op_46895_cast_fp16")]; + tensor var_46897_equation_0 = const()[name = tensor("op_46897_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_46897_cast_fp16 = einsum(equation = var_46897_equation_0, values = (var_46747_cast_fp16, var_46861_cast_fp16))[name = tensor("op_46897_cast_fp16")]; + tensor var_46899_equation_0 = const()[name = tensor("op_46899_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_46899_cast_fp16 = einsum(equation = var_46899_equation_0, values = (var_46751_cast_fp16, var_46862_cast_fp16))[name = tensor("op_46899_cast_fp16")]; + tensor var_46901_equation_0 = const()[name = tensor("op_46901_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_46901_cast_fp16 = einsum(equation = var_46901_equation_0, values = (var_46755_cast_fp16, var_46863_cast_fp16))[name = tensor("op_46901_cast_fp16")]; + tensor var_46903_equation_0 = const()[name = tensor("op_46903_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_46903_cast_fp16 = einsum(equation = var_46903_equation_0, values = (var_46759_cast_fp16, var_46864_cast_fp16))[name = tensor("op_46903_cast_fp16")]; + tensor var_46905_equation_0 = const()[name = tensor("op_46905_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_46905_cast_fp16 = einsum(equation = var_46905_equation_0, values = (var_46763_cast_fp16, var_46865_cast_fp16))[name = tensor("op_46905_cast_fp16")]; + tensor input_595_interleave_0 = const()[name = tensor("input_595_interleave_0"), val = tensor(false)]; + tensor input_595_cast_fp16 = concat(axis = var_30355, interleave = input_595_interleave_0, values = (var_46867_cast_fp16, var_46869_cast_fp16, var_46871_cast_fp16, var_46873_cast_fp16, var_46875_cast_fp16, var_46877_cast_fp16, var_46879_cast_fp16, var_46881_cast_fp16, var_46883_cast_fp16, var_46885_cast_fp16, var_46887_cast_fp16, var_46889_cast_fp16, var_46891_cast_fp16, var_46893_cast_fp16, var_46895_cast_fp16, var_46897_cast_fp16, var_46899_cast_fp16, var_46901_cast_fp16, var_46903_cast_fp16, var_46905_cast_fp16))[name = tensor("input_595_cast_fp16")]; + tensor var_46911 = const()[name = tensor("op_46911"), val = tensor([1, 1])]; + tensor var_46913 = const()[name = tensor("op_46913"), val = tensor([1, 1])]; + tensor var_46915_pad_type_0 = const()[name = tensor("op_46915_pad_type_0"), val = tensor("custom")]; + tensor var_46915_pad_0 = const()[name = tensor("op_46915_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_7_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_7_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3901919616)))]; + tensor up_blocks_0_attentions_1_transformer_blocks_7_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_7_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3905196480)))]; + tensor var_46915_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_7_attn2_to_out_0_bias_to_fp16, dilations = var_46913, groups = var_30355, pad = var_46915_pad_0, pad_type = var_46915_pad_type_0, strides = var_46911, weight = up_blocks_0_attentions_1_transformer_blocks_7_attn2_to_out_0_weight_to_fp16, x = input_595_cast_fp16)[name = tensor("op_46915_cast_fp16")]; + tensor inputs_311_cast_fp16 = add(x = var_46915_cast_fp16, y = inputs_309_cast_fp16)[name = tensor("inputs_311_cast_fp16")]; + tensor input_597_axes_0 = const()[name = tensor("input_597_axes_0"), val = tensor([1])]; + tensor input_597_gamma_0_to_fp16 = const()[name = tensor("input_597_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3905199104)))]; + tensor input_597_beta_0_to_fp16 = const()[name = tensor("input_597_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3905201728)))]; + tensor var_46925_to_fp16 = const()[name = tensor("op_46925_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_597_cast_fp16 = layer_norm(axes = input_597_axes_0, beta = input_597_beta_0_to_fp16, epsilon = var_46925_to_fp16, gamma = input_597_gamma_0_to_fp16, x = inputs_311_cast_fp16)[name = tensor("input_597_cast_fp16")]; + tensor var_46941 = const()[name = tensor("op_46941"), val = tensor([1, 1])]; + tensor var_46943 = const()[name = tensor("op_46943"), val = tensor([1, 1])]; + tensor var_46945_pad_type_0 = const()[name = tensor("op_46945_pad_type_0"), val = tensor("custom")]; + tensor var_46945_pad_0 = const()[name = tensor("op_46945_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_7_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_7_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3905204352)))]; + tensor up_blocks_0_attentions_1_transformer_blocks_7_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_7_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3931418816)))]; + tensor var_46945_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_7_ff_net_0_proj_bias_to_fp16, dilations = var_46943, groups = var_30355, pad = var_46945_pad_0, pad_type = var_46945_pad_type_0, strides = var_46941, weight = up_blocks_0_attentions_1_transformer_blocks_7_ff_net_0_proj_weight_to_fp16, x = input_597_cast_fp16)[name = tensor("op_46945_cast_fp16")]; + tensor var_46946_split_sizes_0 = const()[name = tensor("op_46946_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_46946_axis_0 = const()[name = tensor("op_46946_axis_0"), val = tensor(1)]; + tensor var_46946_cast_fp16_0, tensor var_46946_cast_fp16_1 = split(axis = var_46946_axis_0, split_sizes = var_46946_split_sizes_0, x = var_46945_cast_fp16)[name = tensor("op_46946_cast_fp16")]; + tensor var_46948_mode_0 = const()[name = tensor("op_46948_mode_0"), val = tensor("EXACT")]; + tensor var_46948_cast_fp16 = gelu(mode = var_46948_mode_0, x = var_46946_cast_fp16_1)[name = tensor("op_46948_cast_fp16")]; + tensor input_599_cast_fp16 = mul(x = var_46946_cast_fp16_0, y = var_46948_cast_fp16)[name = tensor("input_599_cast_fp16")]; + tensor var_46952 = const()[name = tensor("op_46952"), val = tensor([1, 1])]; + tensor var_46954 = const()[name = tensor("op_46954"), val = tensor([1, 1])]; + tensor var_46956_pad_type_0 = const()[name = tensor("op_46956_pad_type_0"), val = tensor("custom")]; + tensor var_46956_pad_0 = const()[name = tensor("op_46956_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_7_ff_net_2_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_7_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3931439360)))]; + tensor up_blocks_0_attentions_1_transformer_blocks_7_ff_net_2_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_7_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3944546624)))]; + tensor var_46956_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_7_ff_net_2_bias_to_fp16, dilations = var_46954, groups = var_30355, pad = var_46956_pad_0, pad_type = var_46956_pad_type_0, strides = var_46952, weight = up_blocks_0_attentions_1_transformer_blocks_7_ff_net_2_weight_to_fp16, x = input_599_cast_fp16)[name = tensor("op_46956_cast_fp16")]; + tensor inputs_313_cast_fp16 = add(x = var_46956_cast_fp16, y = inputs_311_cast_fp16)[name = tensor("inputs_313_cast_fp16")]; + tensor hidden_states_409_axes_0 = const()[name = tensor("hidden_states_409_axes_0"), val = tensor([1])]; + tensor hidden_states_409_gamma_0_to_fp16 = const()[name = tensor("hidden_states_409_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3944549248)))]; + tensor hidden_states_409_beta_0_to_fp16 = const()[name = tensor("hidden_states_409_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3944551872)))]; + tensor var_46972_to_fp16 = const()[name = tensor("op_46972_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_409_cast_fp16 = layer_norm(axes = hidden_states_409_axes_0, beta = hidden_states_409_beta_0_to_fp16, epsilon = var_46972_to_fp16, gamma = hidden_states_409_gamma_0_to_fp16, x = inputs_313_cast_fp16)[name = tensor("hidden_states_409_cast_fp16")]; + tensor var_46987 = const()[name = tensor("op_46987"), val = tensor([1, 1])]; + tensor var_46989 = const()[name = tensor("op_46989"), val = tensor([1, 1])]; + tensor q_209_pad_type_0 = const()[name = tensor("q_209_pad_type_0"), val = tensor("custom")]; + tensor q_209_pad_0 = const()[name = tensor("q_209_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_8_attn1_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_8_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3944554496)))]; + tensor q_209_cast_fp16 = conv(dilations = var_46989, groups = var_30355, pad = q_209_pad_0, pad_type = q_209_pad_type_0, strides = var_46987, weight = up_blocks_0_attentions_1_transformer_blocks_8_attn1_to_q_weight_to_fp16, x = hidden_states_409_cast_fp16)[name = tensor("q_209_cast_fp16")]; + tensor var_46993 = const()[name = tensor("op_46993"), val = tensor([1, 1])]; + tensor var_46995 = const()[name = tensor("op_46995"), val = tensor([1, 1])]; + tensor k_417_pad_type_0 = const()[name = tensor("k_417_pad_type_0"), val = tensor("custom")]; + tensor k_417_pad_0 = const()[name = tensor("k_417_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_8_attn1_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_8_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3947831360)))]; + tensor k_417_cast_fp16 = conv(dilations = var_46995, groups = var_30355, pad = k_417_pad_0, pad_type = k_417_pad_type_0, strides = var_46993, weight = up_blocks_0_attentions_1_transformer_blocks_8_attn1_to_k_weight_to_fp16, x = hidden_states_409_cast_fp16)[name = tensor("k_417_cast_fp16")]; + tensor var_46999 = const()[name = tensor("op_46999"), val = tensor([1, 1])]; + tensor var_47001 = const()[name = tensor("op_47001"), val = tensor([1, 1])]; + tensor v_209_pad_type_0 = const()[name = tensor("v_209_pad_type_0"), val = tensor("custom")]; + tensor v_209_pad_0 = const()[name = tensor("v_209_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_8_attn1_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_8_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3951108224)))]; + tensor v_209_cast_fp16 = conv(dilations = var_47001, groups = var_30355, pad = v_209_pad_0, pad_type = v_209_pad_type_0, strides = var_46999, weight = up_blocks_0_attentions_1_transformer_blocks_8_attn1_to_v_weight_to_fp16, x = hidden_states_409_cast_fp16)[name = tensor("v_209_cast_fp16")]; + tensor var_47005_begin_0 = const()[name = tensor("op_47005_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_47005_end_0 = const()[name = tensor("op_47005_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_47005_end_mask_0 = const()[name = tensor("op_47005_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47005_cast_fp16 = slice_by_index(begin = var_47005_begin_0, end = var_47005_end_0, end_mask = var_47005_end_mask_0, x = q_209_cast_fp16)[name = tensor("op_47005_cast_fp16")]; + tensor var_47009_begin_0 = const()[name = tensor("op_47009_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_47009_end_0 = const()[name = tensor("op_47009_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_47009_end_mask_0 = const()[name = tensor("op_47009_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47009_cast_fp16 = slice_by_index(begin = var_47009_begin_0, end = var_47009_end_0, end_mask = var_47009_end_mask_0, x = q_209_cast_fp16)[name = tensor("op_47009_cast_fp16")]; + tensor var_47013_begin_0 = const()[name = tensor("op_47013_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_47013_end_0 = const()[name = tensor("op_47013_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_47013_end_mask_0 = const()[name = tensor("op_47013_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47013_cast_fp16 = slice_by_index(begin = var_47013_begin_0, end = var_47013_end_0, end_mask = var_47013_end_mask_0, x = q_209_cast_fp16)[name = tensor("op_47013_cast_fp16")]; + tensor var_47017_begin_0 = const()[name = tensor("op_47017_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_47017_end_0 = const()[name = tensor("op_47017_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_47017_end_mask_0 = const()[name = tensor("op_47017_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47017_cast_fp16 = slice_by_index(begin = var_47017_begin_0, end = var_47017_end_0, end_mask = var_47017_end_mask_0, x = q_209_cast_fp16)[name = tensor("op_47017_cast_fp16")]; + tensor var_47021_begin_0 = const()[name = tensor("op_47021_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_47021_end_0 = const()[name = tensor("op_47021_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_47021_end_mask_0 = const()[name = tensor("op_47021_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47021_cast_fp16 = slice_by_index(begin = var_47021_begin_0, end = var_47021_end_0, end_mask = var_47021_end_mask_0, x = q_209_cast_fp16)[name = tensor("op_47021_cast_fp16")]; + tensor var_47025_begin_0 = const()[name = tensor("op_47025_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_47025_end_0 = const()[name = tensor("op_47025_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_47025_end_mask_0 = const()[name = tensor("op_47025_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47025_cast_fp16 = slice_by_index(begin = var_47025_begin_0, end = var_47025_end_0, end_mask = var_47025_end_mask_0, x = q_209_cast_fp16)[name = tensor("op_47025_cast_fp16")]; + tensor var_47029_begin_0 = const()[name = tensor("op_47029_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_47029_end_0 = const()[name = tensor("op_47029_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_47029_end_mask_0 = const()[name = tensor("op_47029_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47029_cast_fp16 = slice_by_index(begin = var_47029_begin_0, end = var_47029_end_0, end_mask = var_47029_end_mask_0, x = q_209_cast_fp16)[name = tensor("op_47029_cast_fp16")]; + tensor var_47033_begin_0 = const()[name = tensor("op_47033_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_47033_end_0 = const()[name = tensor("op_47033_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_47033_end_mask_0 = const()[name = tensor("op_47033_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47033_cast_fp16 = slice_by_index(begin = var_47033_begin_0, end = var_47033_end_0, end_mask = var_47033_end_mask_0, x = q_209_cast_fp16)[name = tensor("op_47033_cast_fp16")]; + tensor var_47037_begin_0 = const()[name = tensor("op_47037_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_47037_end_0 = const()[name = tensor("op_47037_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_47037_end_mask_0 = const()[name = tensor("op_47037_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47037_cast_fp16 = slice_by_index(begin = var_47037_begin_0, end = var_47037_end_0, end_mask = var_47037_end_mask_0, x = q_209_cast_fp16)[name = tensor("op_47037_cast_fp16")]; + tensor var_47041_begin_0 = const()[name = tensor("op_47041_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_47041_end_0 = const()[name = tensor("op_47041_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_47041_end_mask_0 = const()[name = tensor("op_47041_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47041_cast_fp16 = slice_by_index(begin = var_47041_begin_0, end = var_47041_end_0, end_mask = var_47041_end_mask_0, x = q_209_cast_fp16)[name = tensor("op_47041_cast_fp16")]; + tensor var_47045_begin_0 = const()[name = tensor("op_47045_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_47045_end_0 = const()[name = tensor("op_47045_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_47045_end_mask_0 = const()[name = tensor("op_47045_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47045_cast_fp16 = slice_by_index(begin = var_47045_begin_0, end = var_47045_end_0, end_mask = var_47045_end_mask_0, x = q_209_cast_fp16)[name = tensor("op_47045_cast_fp16")]; + tensor var_47049_begin_0 = const()[name = tensor("op_47049_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_47049_end_0 = const()[name = tensor("op_47049_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_47049_end_mask_0 = const()[name = tensor("op_47049_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47049_cast_fp16 = slice_by_index(begin = var_47049_begin_0, end = var_47049_end_0, end_mask = var_47049_end_mask_0, x = q_209_cast_fp16)[name = tensor("op_47049_cast_fp16")]; + tensor var_47053_begin_0 = const()[name = tensor("op_47053_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_47053_end_0 = const()[name = tensor("op_47053_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_47053_end_mask_0 = const()[name = tensor("op_47053_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47053_cast_fp16 = slice_by_index(begin = var_47053_begin_0, end = var_47053_end_0, end_mask = var_47053_end_mask_0, x = q_209_cast_fp16)[name = tensor("op_47053_cast_fp16")]; + tensor var_47057_begin_0 = const()[name = tensor("op_47057_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_47057_end_0 = const()[name = tensor("op_47057_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_47057_end_mask_0 = const()[name = tensor("op_47057_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47057_cast_fp16 = slice_by_index(begin = var_47057_begin_0, end = var_47057_end_0, end_mask = var_47057_end_mask_0, x = q_209_cast_fp16)[name = tensor("op_47057_cast_fp16")]; + tensor var_47061_begin_0 = const()[name = tensor("op_47061_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_47061_end_0 = const()[name = tensor("op_47061_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_47061_end_mask_0 = const()[name = tensor("op_47061_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47061_cast_fp16 = slice_by_index(begin = var_47061_begin_0, end = var_47061_end_0, end_mask = var_47061_end_mask_0, x = q_209_cast_fp16)[name = tensor("op_47061_cast_fp16")]; + tensor var_47065_begin_0 = const()[name = tensor("op_47065_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_47065_end_0 = const()[name = tensor("op_47065_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_47065_end_mask_0 = const()[name = tensor("op_47065_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47065_cast_fp16 = slice_by_index(begin = var_47065_begin_0, end = var_47065_end_0, end_mask = var_47065_end_mask_0, x = q_209_cast_fp16)[name = tensor("op_47065_cast_fp16")]; + tensor var_47069_begin_0 = const()[name = tensor("op_47069_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_47069_end_0 = const()[name = tensor("op_47069_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_47069_end_mask_0 = const()[name = tensor("op_47069_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47069_cast_fp16 = slice_by_index(begin = var_47069_begin_0, end = var_47069_end_0, end_mask = var_47069_end_mask_0, x = q_209_cast_fp16)[name = tensor("op_47069_cast_fp16")]; + tensor var_47073_begin_0 = const()[name = tensor("op_47073_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_47073_end_0 = const()[name = tensor("op_47073_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_47073_end_mask_0 = const()[name = tensor("op_47073_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47073_cast_fp16 = slice_by_index(begin = var_47073_begin_0, end = var_47073_end_0, end_mask = var_47073_end_mask_0, x = q_209_cast_fp16)[name = tensor("op_47073_cast_fp16")]; + tensor var_47077_begin_0 = const()[name = tensor("op_47077_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_47077_end_0 = const()[name = tensor("op_47077_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_47077_end_mask_0 = const()[name = tensor("op_47077_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47077_cast_fp16 = slice_by_index(begin = var_47077_begin_0, end = var_47077_end_0, end_mask = var_47077_end_mask_0, x = q_209_cast_fp16)[name = tensor("op_47077_cast_fp16")]; + tensor var_47081_begin_0 = const()[name = tensor("op_47081_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_47081_end_0 = const()[name = tensor("op_47081_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_47081_end_mask_0 = const()[name = tensor("op_47081_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47081_cast_fp16 = slice_by_index(begin = var_47081_begin_0, end = var_47081_end_0, end_mask = var_47081_end_mask_0, x = q_209_cast_fp16)[name = tensor("op_47081_cast_fp16")]; + tensor k_419_perm_0 = const()[name = tensor("k_419_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_47088_begin_0 = const()[name = tensor("op_47088_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_47088_end_0 = const()[name = tensor("op_47088_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_47088_end_mask_0 = const()[name = tensor("op_47088_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_419_cast_fp16 = transpose(perm = k_419_perm_0, x = k_417_cast_fp16)[name = tensor("transpose_35")]; + tensor var_47088_cast_fp16 = slice_by_index(begin = var_47088_begin_0, end = var_47088_end_0, end_mask = var_47088_end_mask_0, x = k_419_cast_fp16)[name = tensor("op_47088_cast_fp16")]; + tensor var_47092_begin_0 = const()[name = tensor("op_47092_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_47092_end_0 = const()[name = tensor("op_47092_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_47092_end_mask_0 = const()[name = tensor("op_47092_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47092_cast_fp16 = slice_by_index(begin = var_47092_begin_0, end = var_47092_end_0, end_mask = var_47092_end_mask_0, x = k_419_cast_fp16)[name = tensor("op_47092_cast_fp16")]; + tensor var_47096_begin_0 = const()[name = tensor("op_47096_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_47096_end_0 = const()[name = tensor("op_47096_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_47096_end_mask_0 = const()[name = tensor("op_47096_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47096_cast_fp16 = slice_by_index(begin = var_47096_begin_0, end = var_47096_end_0, end_mask = var_47096_end_mask_0, x = k_419_cast_fp16)[name = tensor("op_47096_cast_fp16")]; + tensor var_47100_begin_0 = const()[name = tensor("op_47100_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_47100_end_0 = const()[name = tensor("op_47100_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_47100_end_mask_0 = const()[name = tensor("op_47100_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47100_cast_fp16 = slice_by_index(begin = var_47100_begin_0, end = var_47100_end_0, end_mask = var_47100_end_mask_0, x = k_419_cast_fp16)[name = tensor("op_47100_cast_fp16")]; + tensor var_47104_begin_0 = const()[name = tensor("op_47104_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_47104_end_0 = const()[name = tensor("op_47104_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_47104_end_mask_0 = const()[name = tensor("op_47104_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47104_cast_fp16 = slice_by_index(begin = var_47104_begin_0, end = var_47104_end_0, end_mask = var_47104_end_mask_0, x = k_419_cast_fp16)[name = tensor("op_47104_cast_fp16")]; + tensor var_47108_begin_0 = const()[name = tensor("op_47108_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_47108_end_0 = const()[name = tensor("op_47108_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_47108_end_mask_0 = const()[name = tensor("op_47108_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47108_cast_fp16 = slice_by_index(begin = var_47108_begin_0, end = var_47108_end_0, end_mask = var_47108_end_mask_0, x = k_419_cast_fp16)[name = tensor("op_47108_cast_fp16")]; + tensor var_47112_begin_0 = const()[name = tensor("op_47112_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_47112_end_0 = const()[name = tensor("op_47112_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_47112_end_mask_0 = const()[name = tensor("op_47112_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47112_cast_fp16 = slice_by_index(begin = var_47112_begin_0, end = var_47112_end_0, end_mask = var_47112_end_mask_0, x = k_419_cast_fp16)[name = tensor("op_47112_cast_fp16")]; + tensor var_47116_begin_0 = const()[name = tensor("op_47116_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_47116_end_0 = const()[name = tensor("op_47116_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_47116_end_mask_0 = const()[name = tensor("op_47116_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47116_cast_fp16 = slice_by_index(begin = var_47116_begin_0, end = var_47116_end_0, end_mask = var_47116_end_mask_0, x = k_419_cast_fp16)[name = tensor("op_47116_cast_fp16")]; + tensor var_47120_begin_0 = const()[name = tensor("op_47120_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_47120_end_0 = const()[name = tensor("op_47120_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_47120_end_mask_0 = const()[name = tensor("op_47120_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47120_cast_fp16 = slice_by_index(begin = var_47120_begin_0, end = var_47120_end_0, end_mask = var_47120_end_mask_0, x = k_419_cast_fp16)[name = tensor("op_47120_cast_fp16")]; + tensor var_47124_begin_0 = const()[name = tensor("op_47124_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_47124_end_0 = const()[name = tensor("op_47124_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_47124_end_mask_0 = const()[name = tensor("op_47124_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47124_cast_fp16 = slice_by_index(begin = var_47124_begin_0, end = var_47124_end_0, end_mask = var_47124_end_mask_0, x = k_419_cast_fp16)[name = tensor("op_47124_cast_fp16")]; + tensor var_47128_begin_0 = const()[name = tensor("op_47128_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_47128_end_0 = const()[name = tensor("op_47128_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_47128_end_mask_0 = const()[name = tensor("op_47128_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47128_cast_fp16 = slice_by_index(begin = var_47128_begin_0, end = var_47128_end_0, end_mask = var_47128_end_mask_0, x = k_419_cast_fp16)[name = tensor("op_47128_cast_fp16")]; + tensor var_47132_begin_0 = const()[name = tensor("op_47132_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_47132_end_0 = const()[name = tensor("op_47132_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_47132_end_mask_0 = const()[name = tensor("op_47132_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47132_cast_fp16 = slice_by_index(begin = var_47132_begin_0, end = var_47132_end_0, end_mask = var_47132_end_mask_0, x = k_419_cast_fp16)[name = tensor("op_47132_cast_fp16")]; + tensor var_47136_begin_0 = const()[name = tensor("op_47136_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_47136_end_0 = const()[name = tensor("op_47136_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_47136_end_mask_0 = const()[name = tensor("op_47136_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47136_cast_fp16 = slice_by_index(begin = var_47136_begin_0, end = var_47136_end_0, end_mask = var_47136_end_mask_0, x = k_419_cast_fp16)[name = tensor("op_47136_cast_fp16")]; + tensor var_47140_begin_0 = const()[name = tensor("op_47140_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_47140_end_0 = const()[name = tensor("op_47140_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_47140_end_mask_0 = const()[name = tensor("op_47140_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47140_cast_fp16 = slice_by_index(begin = var_47140_begin_0, end = var_47140_end_0, end_mask = var_47140_end_mask_0, x = k_419_cast_fp16)[name = tensor("op_47140_cast_fp16")]; + tensor var_47144_begin_0 = const()[name = tensor("op_47144_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_47144_end_0 = const()[name = tensor("op_47144_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_47144_end_mask_0 = const()[name = tensor("op_47144_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47144_cast_fp16 = slice_by_index(begin = var_47144_begin_0, end = var_47144_end_0, end_mask = var_47144_end_mask_0, x = k_419_cast_fp16)[name = tensor("op_47144_cast_fp16")]; + tensor var_47148_begin_0 = const()[name = tensor("op_47148_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_47148_end_0 = const()[name = tensor("op_47148_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_47148_end_mask_0 = const()[name = tensor("op_47148_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47148_cast_fp16 = slice_by_index(begin = var_47148_begin_0, end = var_47148_end_0, end_mask = var_47148_end_mask_0, x = k_419_cast_fp16)[name = tensor("op_47148_cast_fp16")]; + tensor var_47152_begin_0 = const()[name = tensor("op_47152_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_47152_end_0 = const()[name = tensor("op_47152_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_47152_end_mask_0 = const()[name = tensor("op_47152_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47152_cast_fp16 = slice_by_index(begin = var_47152_begin_0, end = var_47152_end_0, end_mask = var_47152_end_mask_0, x = k_419_cast_fp16)[name = tensor("op_47152_cast_fp16")]; + tensor var_47156_begin_0 = const()[name = tensor("op_47156_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_47156_end_0 = const()[name = tensor("op_47156_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_47156_end_mask_0 = const()[name = tensor("op_47156_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47156_cast_fp16 = slice_by_index(begin = var_47156_begin_0, end = var_47156_end_0, end_mask = var_47156_end_mask_0, x = k_419_cast_fp16)[name = tensor("op_47156_cast_fp16")]; + tensor var_47160_begin_0 = const()[name = tensor("op_47160_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_47160_end_0 = const()[name = tensor("op_47160_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_47160_end_mask_0 = const()[name = tensor("op_47160_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47160_cast_fp16 = slice_by_index(begin = var_47160_begin_0, end = var_47160_end_0, end_mask = var_47160_end_mask_0, x = k_419_cast_fp16)[name = tensor("op_47160_cast_fp16")]; + tensor var_47164_begin_0 = const()[name = tensor("op_47164_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_47164_end_0 = const()[name = tensor("op_47164_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_47164_end_mask_0 = const()[name = tensor("op_47164_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47164_cast_fp16 = slice_by_index(begin = var_47164_begin_0, end = var_47164_end_0, end_mask = var_47164_end_mask_0, x = k_419_cast_fp16)[name = tensor("op_47164_cast_fp16")]; + tensor var_47166_begin_0 = const()[name = tensor("op_47166_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_47166_end_0 = const()[name = tensor("op_47166_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_47166_end_mask_0 = const()[name = tensor("op_47166_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47166_cast_fp16 = slice_by_index(begin = var_47166_begin_0, end = var_47166_end_0, end_mask = var_47166_end_mask_0, x = v_209_cast_fp16)[name = tensor("op_47166_cast_fp16")]; + tensor var_47170_begin_0 = const()[name = tensor("op_47170_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_47170_end_0 = const()[name = tensor("op_47170_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_47170_end_mask_0 = const()[name = tensor("op_47170_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47170_cast_fp16 = slice_by_index(begin = var_47170_begin_0, end = var_47170_end_0, end_mask = var_47170_end_mask_0, x = v_209_cast_fp16)[name = tensor("op_47170_cast_fp16")]; + tensor var_47174_begin_0 = const()[name = tensor("op_47174_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_47174_end_0 = const()[name = tensor("op_47174_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_47174_end_mask_0 = const()[name = tensor("op_47174_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47174_cast_fp16 = slice_by_index(begin = var_47174_begin_0, end = var_47174_end_0, end_mask = var_47174_end_mask_0, x = v_209_cast_fp16)[name = tensor("op_47174_cast_fp16")]; + tensor var_47178_begin_0 = const()[name = tensor("op_47178_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_47178_end_0 = const()[name = tensor("op_47178_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_47178_end_mask_0 = const()[name = tensor("op_47178_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47178_cast_fp16 = slice_by_index(begin = var_47178_begin_0, end = var_47178_end_0, end_mask = var_47178_end_mask_0, x = v_209_cast_fp16)[name = tensor("op_47178_cast_fp16")]; + tensor var_47182_begin_0 = const()[name = tensor("op_47182_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_47182_end_0 = const()[name = tensor("op_47182_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_47182_end_mask_0 = const()[name = tensor("op_47182_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47182_cast_fp16 = slice_by_index(begin = var_47182_begin_0, end = var_47182_end_0, end_mask = var_47182_end_mask_0, x = v_209_cast_fp16)[name = tensor("op_47182_cast_fp16")]; + tensor var_47186_begin_0 = const()[name = tensor("op_47186_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_47186_end_0 = const()[name = tensor("op_47186_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_47186_end_mask_0 = const()[name = tensor("op_47186_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47186_cast_fp16 = slice_by_index(begin = var_47186_begin_0, end = var_47186_end_0, end_mask = var_47186_end_mask_0, x = v_209_cast_fp16)[name = tensor("op_47186_cast_fp16")]; + tensor var_47190_begin_0 = const()[name = tensor("op_47190_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_47190_end_0 = const()[name = tensor("op_47190_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_47190_end_mask_0 = const()[name = tensor("op_47190_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47190_cast_fp16 = slice_by_index(begin = var_47190_begin_0, end = var_47190_end_0, end_mask = var_47190_end_mask_0, x = v_209_cast_fp16)[name = tensor("op_47190_cast_fp16")]; + tensor var_47194_begin_0 = const()[name = tensor("op_47194_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_47194_end_0 = const()[name = tensor("op_47194_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_47194_end_mask_0 = const()[name = tensor("op_47194_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47194_cast_fp16 = slice_by_index(begin = var_47194_begin_0, end = var_47194_end_0, end_mask = var_47194_end_mask_0, x = v_209_cast_fp16)[name = tensor("op_47194_cast_fp16")]; + tensor var_47198_begin_0 = const()[name = tensor("op_47198_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_47198_end_0 = const()[name = tensor("op_47198_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_47198_end_mask_0 = const()[name = tensor("op_47198_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47198_cast_fp16 = slice_by_index(begin = var_47198_begin_0, end = var_47198_end_0, end_mask = var_47198_end_mask_0, x = v_209_cast_fp16)[name = tensor("op_47198_cast_fp16")]; + tensor var_47202_begin_0 = const()[name = tensor("op_47202_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_47202_end_0 = const()[name = tensor("op_47202_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_47202_end_mask_0 = const()[name = tensor("op_47202_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47202_cast_fp16 = slice_by_index(begin = var_47202_begin_0, end = var_47202_end_0, end_mask = var_47202_end_mask_0, x = v_209_cast_fp16)[name = tensor("op_47202_cast_fp16")]; + tensor var_47206_begin_0 = const()[name = tensor("op_47206_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_47206_end_0 = const()[name = tensor("op_47206_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_47206_end_mask_0 = const()[name = tensor("op_47206_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47206_cast_fp16 = slice_by_index(begin = var_47206_begin_0, end = var_47206_end_0, end_mask = var_47206_end_mask_0, x = v_209_cast_fp16)[name = tensor("op_47206_cast_fp16")]; + tensor var_47210_begin_0 = const()[name = tensor("op_47210_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_47210_end_0 = const()[name = tensor("op_47210_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_47210_end_mask_0 = const()[name = tensor("op_47210_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47210_cast_fp16 = slice_by_index(begin = var_47210_begin_0, end = var_47210_end_0, end_mask = var_47210_end_mask_0, x = v_209_cast_fp16)[name = tensor("op_47210_cast_fp16")]; + tensor var_47214_begin_0 = const()[name = tensor("op_47214_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_47214_end_0 = const()[name = tensor("op_47214_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_47214_end_mask_0 = const()[name = tensor("op_47214_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47214_cast_fp16 = slice_by_index(begin = var_47214_begin_0, end = var_47214_end_0, end_mask = var_47214_end_mask_0, x = v_209_cast_fp16)[name = tensor("op_47214_cast_fp16")]; + tensor var_47218_begin_0 = const()[name = tensor("op_47218_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_47218_end_0 = const()[name = tensor("op_47218_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_47218_end_mask_0 = const()[name = tensor("op_47218_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47218_cast_fp16 = slice_by_index(begin = var_47218_begin_0, end = var_47218_end_0, end_mask = var_47218_end_mask_0, x = v_209_cast_fp16)[name = tensor("op_47218_cast_fp16")]; + tensor var_47222_begin_0 = const()[name = tensor("op_47222_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_47222_end_0 = const()[name = tensor("op_47222_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_47222_end_mask_0 = const()[name = tensor("op_47222_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47222_cast_fp16 = slice_by_index(begin = var_47222_begin_0, end = var_47222_end_0, end_mask = var_47222_end_mask_0, x = v_209_cast_fp16)[name = tensor("op_47222_cast_fp16")]; + tensor var_47226_begin_0 = const()[name = tensor("op_47226_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_47226_end_0 = const()[name = tensor("op_47226_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_47226_end_mask_0 = const()[name = tensor("op_47226_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47226_cast_fp16 = slice_by_index(begin = var_47226_begin_0, end = var_47226_end_0, end_mask = var_47226_end_mask_0, x = v_209_cast_fp16)[name = tensor("op_47226_cast_fp16")]; + tensor var_47230_begin_0 = const()[name = tensor("op_47230_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_47230_end_0 = const()[name = tensor("op_47230_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_47230_end_mask_0 = const()[name = tensor("op_47230_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47230_cast_fp16 = slice_by_index(begin = var_47230_begin_0, end = var_47230_end_0, end_mask = var_47230_end_mask_0, x = v_209_cast_fp16)[name = tensor("op_47230_cast_fp16")]; + tensor var_47234_begin_0 = const()[name = tensor("op_47234_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_47234_end_0 = const()[name = tensor("op_47234_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_47234_end_mask_0 = const()[name = tensor("op_47234_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47234_cast_fp16 = slice_by_index(begin = var_47234_begin_0, end = var_47234_end_0, end_mask = var_47234_end_mask_0, x = v_209_cast_fp16)[name = tensor("op_47234_cast_fp16")]; + tensor var_47238_begin_0 = const()[name = tensor("op_47238_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_47238_end_0 = const()[name = tensor("op_47238_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_47238_end_mask_0 = const()[name = tensor("op_47238_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47238_cast_fp16 = slice_by_index(begin = var_47238_begin_0, end = var_47238_end_0, end_mask = var_47238_end_mask_0, x = v_209_cast_fp16)[name = tensor("op_47238_cast_fp16")]; + tensor var_47242_begin_0 = const()[name = tensor("op_47242_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_47242_end_0 = const()[name = tensor("op_47242_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_47242_end_mask_0 = const()[name = tensor("op_47242_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47242_cast_fp16 = slice_by_index(begin = var_47242_begin_0, end = var_47242_end_0, end_mask = var_47242_end_mask_0, x = v_209_cast_fp16)[name = tensor("op_47242_cast_fp16")]; + tensor var_47246_equation_0 = const()[name = tensor("op_47246_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47246_cast_fp16 = einsum(equation = var_47246_equation_0, values = (var_47088_cast_fp16, var_47005_cast_fp16))[name = tensor("op_47246_cast_fp16")]; + tensor var_47247_to_fp16 = const()[name = tensor("op_47247_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4001_cast_fp16 = mul(x = var_47246_cast_fp16, y = var_47247_to_fp16)[name = tensor("aw_4001_cast_fp16")]; + tensor var_47250_equation_0 = const()[name = tensor("op_47250_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47250_cast_fp16 = einsum(equation = var_47250_equation_0, values = (var_47092_cast_fp16, var_47009_cast_fp16))[name = tensor("op_47250_cast_fp16")]; + tensor var_47251_to_fp16 = const()[name = tensor("op_47251_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4003_cast_fp16 = mul(x = var_47250_cast_fp16, y = var_47251_to_fp16)[name = tensor("aw_4003_cast_fp16")]; + tensor var_47254_equation_0 = const()[name = tensor("op_47254_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47254_cast_fp16 = einsum(equation = var_47254_equation_0, values = (var_47096_cast_fp16, var_47013_cast_fp16))[name = tensor("op_47254_cast_fp16")]; + tensor var_47255_to_fp16 = const()[name = tensor("op_47255_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4005_cast_fp16 = mul(x = var_47254_cast_fp16, y = var_47255_to_fp16)[name = tensor("aw_4005_cast_fp16")]; + tensor var_47258_equation_0 = const()[name = tensor("op_47258_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47258_cast_fp16 = einsum(equation = var_47258_equation_0, values = (var_47100_cast_fp16, var_47017_cast_fp16))[name = tensor("op_47258_cast_fp16")]; + tensor var_47259_to_fp16 = const()[name = tensor("op_47259_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4007_cast_fp16 = mul(x = var_47258_cast_fp16, y = var_47259_to_fp16)[name = tensor("aw_4007_cast_fp16")]; + tensor var_47262_equation_0 = const()[name = tensor("op_47262_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47262_cast_fp16 = einsum(equation = var_47262_equation_0, values = (var_47104_cast_fp16, var_47021_cast_fp16))[name = tensor("op_47262_cast_fp16")]; + tensor var_47263_to_fp16 = const()[name = tensor("op_47263_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4009_cast_fp16 = mul(x = var_47262_cast_fp16, y = var_47263_to_fp16)[name = tensor("aw_4009_cast_fp16")]; + tensor var_47266_equation_0 = const()[name = tensor("op_47266_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47266_cast_fp16 = einsum(equation = var_47266_equation_0, values = (var_47108_cast_fp16, var_47025_cast_fp16))[name = tensor("op_47266_cast_fp16")]; + tensor var_47267_to_fp16 = const()[name = tensor("op_47267_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4011_cast_fp16 = mul(x = var_47266_cast_fp16, y = var_47267_to_fp16)[name = tensor("aw_4011_cast_fp16")]; + tensor var_47270_equation_0 = const()[name = tensor("op_47270_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47270_cast_fp16 = einsum(equation = var_47270_equation_0, values = (var_47112_cast_fp16, var_47029_cast_fp16))[name = tensor("op_47270_cast_fp16")]; + tensor var_47271_to_fp16 = const()[name = tensor("op_47271_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4013_cast_fp16 = mul(x = var_47270_cast_fp16, y = var_47271_to_fp16)[name = tensor("aw_4013_cast_fp16")]; + tensor var_47274_equation_0 = const()[name = tensor("op_47274_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47274_cast_fp16 = einsum(equation = var_47274_equation_0, values = (var_47116_cast_fp16, var_47033_cast_fp16))[name = tensor("op_47274_cast_fp16")]; + tensor var_47275_to_fp16 = const()[name = tensor("op_47275_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4015_cast_fp16 = mul(x = var_47274_cast_fp16, y = var_47275_to_fp16)[name = tensor("aw_4015_cast_fp16")]; + tensor var_47278_equation_0 = const()[name = tensor("op_47278_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47278_cast_fp16 = einsum(equation = var_47278_equation_0, values = (var_47120_cast_fp16, var_47037_cast_fp16))[name = tensor("op_47278_cast_fp16")]; + tensor var_47279_to_fp16 = const()[name = tensor("op_47279_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4017_cast_fp16 = mul(x = var_47278_cast_fp16, y = var_47279_to_fp16)[name = tensor("aw_4017_cast_fp16")]; + tensor var_47282_equation_0 = const()[name = tensor("op_47282_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47282_cast_fp16 = einsum(equation = var_47282_equation_0, values = (var_47124_cast_fp16, var_47041_cast_fp16))[name = tensor("op_47282_cast_fp16")]; + tensor var_47283_to_fp16 = const()[name = tensor("op_47283_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4019_cast_fp16 = mul(x = var_47282_cast_fp16, y = var_47283_to_fp16)[name = tensor("aw_4019_cast_fp16")]; + tensor var_47286_equation_0 = const()[name = tensor("op_47286_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47286_cast_fp16 = einsum(equation = var_47286_equation_0, values = (var_47128_cast_fp16, var_47045_cast_fp16))[name = tensor("op_47286_cast_fp16")]; + tensor var_47287_to_fp16 = const()[name = tensor("op_47287_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4021_cast_fp16 = mul(x = var_47286_cast_fp16, y = var_47287_to_fp16)[name = tensor("aw_4021_cast_fp16")]; + tensor var_47290_equation_0 = const()[name = tensor("op_47290_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47290_cast_fp16 = einsum(equation = var_47290_equation_0, values = (var_47132_cast_fp16, var_47049_cast_fp16))[name = tensor("op_47290_cast_fp16")]; + tensor var_47291_to_fp16 = const()[name = tensor("op_47291_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4023_cast_fp16 = mul(x = var_47290_cast_fp16, y = var_47291_to_fp16)[name = tensor("aw_4023_cast_fp16")]; + tensor var_47294_equation_0 = const()[name = tensor("op_47294_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47294_cast_fp16 = einsum(equation = var_47294_equation_0, values = (var_47136_cast_fp16, var_47053_cast_fp16))[name = tensor("op_47294_cast_fp16")]; + tensor var_47295_to_fp16 = const()[name = tensor("op_47295_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4025_cast_fp16 = mul(x = var_47294_cast_fp16, y = var_47295_to_fp16)[name = tensor("aw_4025_cast_fp16")]; + tensor var_47298_equation_0 = const()[name = tensor("op_47298_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47298_cast_fp16 = einsum(equation = var_47298_equation_0, values = (var_47140_cast_fp16, var_47057_cast_fp16))[name = tensor("op_47298_cast_fp16")]; + tensor var_47299_to_fp16 = const()[name = tensor("op_47299_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4027_cast_fp16 = mul(x = var_47298_cast_fp16, y = var_47299_to_fp16)[name = tensor("aw_4027_cast_fp16")]; + tensor var_47302_equation_0 = const()[name = tensor("op_47302_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47302_cast_fp16 = einsum(equation = var_47302_equation_0, values = (var_47144_cast_fp16, var_47061_cast_fp16))[name = tensor("op_47302_cast_fp16")]; + tensor var_47303_to_fp16 = const()[name = tensor("op_47303_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4029_cast_fp16 = mul(x = var_47302_cast_fp16, y = var_47303_to_fp16)[name = tensor("aw_4029_cast_fp16")]; + tensor var_47306_equation_0 = const()[name = tensor("op_47306_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47306_cast_fp16 = einsum(equation = var_47306_equation_0, values = (var_47148_cast_fp16, var_47065_cast_fp16))[name = tensor("op_47306_cast_fp16")]; + tensor var_47307_to_fp16 = const()[name = tensor("op_47307_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4031_cast_fp16 = mul(x = var_47306_cast_fp16, y = var_47307_to_fp16)[name = tensor("aw_4031_cast_fp16")]; + tensor var_47310_equation_0 = const()[name = tensor("op_47310_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47310_cast_fp16 = einsum(equation = var_47310_equation_0, values = (var_47152_cast_fp16, var_47069_cast_fp16))[name = tensor("op_47310_cast_fp16")]; + tensor var_47311_to_fp16 = const()[name = tensor("op_47311_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4033_cast_fp16 = mul(x = var_47310_cast_fp16, y = var_47311_to_fp16)[name = tensor("aw_4033_cast_fp16")]; + tensor var_47314_equation_0 = const()[name = tensor("op_47314_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47314_cast_fp16 = einsum(equation = var_47314_equation_0, values = (var_47156_cast_fp16, var_47073_cast_fp16))[name = tensor("op_47314_cast_fp16")]; + tensor var_47315_to_fp16 = const()[name = tensor("op_47315_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4035_cast_fp16 = mul(x = var_47314_cast_fp16, y = var_47315_to_fp16)[name = tensor("aw_4035_cast_fp16")]; + tensor var_47318_equation_0 = const()[name = tensor("op_47318_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47318_cast_fp16 = einsum(equation = var_47318_equation_0, values = (var_47160_cast_fp16, var_47077_cast_fp16))[name = tensor("op_47318_cast_fp16")]; + tensor var_47319_to_fp16 = const()[name = tensor("op_47319_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4037_cast_fp16 = mul(x = var_47318_cast_fp16, y = var_47319_to_fp16)[name = tensor("aw_4037_cast_fp16")]; + tensor var_47322_equation_0 = const()[name = tensor("op_47322_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47322_cast_fp16 = einsum(equation = var_47322_equation_0, values = (var_47164_cast_fp16, var_47081_cast_fp16))[name = tensor("op_47322_cast_fp16")]; + tensor var_47323_to_fp16 = const()[name = tensor("op_47323_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4039_cast_fp16 = mul(x = var_47322_cast_fp16, y = var_47323_to_fp16)[name = tensor("aw_4039_cast_fp16")]; + tensor var_47325_cast_fp16 = softmax(axis = var_30355, x = aw_4001_cast_fp16)[name = tensor("op_47325_cast_fp16")]; + tensor var_47326_cast_fp16 = softmax(axis = var_30355, x = aw_4003_cast_fp16)[name = tensor("op_47326_cast_fp16")]; + tensor var_47327_cast_fp16 = softmax(axis = var_30355, x = aw_4005_cast_fp16)[name = tensor("op_47327_cast_fp16")]; + tensor var_47328_cast_fp16 = softmax(axis = var_30355, x = aw_4007_cast_fp16)[name = tensor("op_47328_cast_fp16")]; + tensor var_47329_cast_fp16 = softmax(axis = var_30355, x = aw_4009_cast_fp16)[name = tensor("op_47329_cast_fp16")]; + tensor var_47330_cast_fp16 = softmax(axis = var_30355, x = aw_4011_cast_fp16)[name = tensor("op_47330_cast_fp16")]; + tensor var_47331_cast_fp16 = softmax(axis = var_30355, x = aw_4013_cast_fp16)[name = tensor("op_47331_cast_fp16")]; + tensor var_47332_cast_fp16 = softmax(axis = var_30355, x = aw_4015_cast_fp16)[name = tensor("op_47332_cast_fp16")]; + tensor var_47333_cast_fp16 = softmax(axis = var_30355, x = aw_4017_cast_fp16)[name = tensor("op_47333_cast_fp16")]; + tensor var_47334_cast_fp16 = softmax(axis = var_30355, x = aw_4019_cast_fp16)[name = tensor("op_47334_cast_fp16")]; + tensor var_47335_cast_fp16 = softmax(axis = var_30355, x = aw_4021_cast_fp16)[name = tensor("op_47335_cast_fp16")]; + tensor var_47336_cast_fp16 = softmax(axis = var_30355, x = aw_4023_cast_fp16)[name = tensor("op_47336_cast_fp16")]; + tensor var_47337_cast_fp16 = softmax(axis = var_30355, x = aw_4025_cast_fp16)[name = tensor("op_47337_cast_fp16")]; + tensor var_47338_cast_fp16 = softmax(axis = var_30355, x = aw_4027_cast_fp16)[name = tensor("op_47338_cast_fp16")]; + tensor var_47339_cast_fp16 = softmax(axis = var_30355, x = aw_4029_cast_fp16)[name = tensor("op_47339_cast_fp16")]; + tensor var_47340_cast_fp16 = softmax(axis = var_30355, x = aw_4031_cast_fp16)[name = tensor("op_47340_cast_fp16")]; + tensor var_47341_cast_fp16 = softmax(axis = var_30355, x = aw_4033_cast_fp16)[name = tensor("op_47341_cast_fp16")]; + tensor var_47342_cast_fp16 = softmax(axis = var_30355, x = aw_4035_cast_fp16)[name = tensor("op_47342_cast_fp16")]; + tensor var_47343_cast_fp16 = softmax(axis = var_30355, x = aw_4037_cast_fp16)[name = tensor("op_47343_cast_fp16")]; + tensor var_47344_cast_fp16 = softmax(axis = var_30355, x = aw_4039_cast_fp16)[name = tensor("op_47344_cast_fp16")]; + tensor var_47346_equation_0 = const()[name = tensor("op_47346_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47346_cast_fp16 = einsum(equation = var_47346_equation_0, values = (var_47166_cast_fp16, var_47325_cast_fp16))[name = tensor("op_47346_cast_fp16")]; + tensor var_47348_equation_0 = const()[name = tensor("op_47348_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47348_cast_fp16 = einsum(equation = var_47348_equation_0, values = (var_47170_cast_fp16, var_47326_cast_fp16))[name = tensor("op_47348_cast_fp16")]; + tensor var_47350_equation_0 = const()[name = tensor("op_47350_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47350_cast_fp16 = einsum(equation = var_47350_equation_0, values = (var_47174_cast_fp16, var_47327_cast_fp16))[name = tensor("op_47350_cast_fp16")]; + tensor var_47352_equation_0 = const()[name = tensor("op_47352_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47352_cast_fp16 = einsum(equation = var_47352_equation_0, values = (var_47178_cast_fp16, var_47328_cast_fp16))[name = tensor("op_47352_cast_fp16")]; + tensor var_47354_equation_0 = const()[name = tensor("op_47354_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47354_cast_fp16 = einsum(equation = var_47354_equation_0, values = (var_47182_cast_fp16, var_47329_cast_fp16))[name = tensor("op_47354_cast_fp16")]; + tensor var_47356_equation_0 = const()[name = tensor("op_47356_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47356_cast_fp16 = einsum(equation = var_47356_equation_0, values = (var_47186_cast_fp16, var_47330_cast_fp16))[name = tensor("op_47356_cast_fp16")]; + tensor var_47358_equation_0 = const()[name = tensor("op_47358_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47358_cast_fp16 = einsum(equation = var_47358_equation_0, values = (var_47190_cast_fp16, var_47331_cast_fp16))[name = tensor("op_47358_cast_fp16")]; + tensor var_47360_equation_0 = const()[name = tensor("op_47360_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47360_cast_fp16 = einsum(equation = var_47360_equation_0, values = (var_47194_cast_fp16, var_47332_cast_fp16))[name = tensor("op_47360_cast_fp16")]; + tensor var_47362_equation_0 = const()[name = tensor("op_47362_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47362_cast_fp16 = einsum(equation = var_47362_equation_0, values = (var_47198_cast_fp16, var_47333_cast_fp16))[name = tensor("op_47362_cast_fp16")]; + tensor var_47364_equation_0 = const()[name = tensor("op_47364_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47364_cast_fp16 = einsum(equation = var_47364_equation_0, values = (var_47202_cast_fp16, var_47334_cast_fp16))[name = tensor("op_47364_cast_fp16")]; + tensor var_47366_equation_0 = const()[name = tensor("op_47366_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47366_cast_fp16 = einsum(equation = var_47366_equation_0, values = (var_47206_cast_fp16, var_47335_cast_fp16))[name = tensor("op_47366_cast_fp16")]; + tensor var_47368_equation_0 = const()[name = tensor("op_47368_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47368_cast_fp16 = einsum(equation = var_47368_equation_0, values = (var_47210_cast_fp16, var_47336_cast_fp16))[name = tensor("op_47368_cast_fp16")]; + tensor var_47370_equation_0 = const()[name = tensor("op_47370_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47370_cast_fp16 = einsum(equation = var_47370_equation_0, values = (var_47214_cast_fp16, var_47337_cast_fp16))[name = tensor("op_47370_cast_fp16")]; + tensor var_47372_equation_0 = const()[name = tensor("op_47372_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47372_cast_fp16 = einsum(equation = var_47372_equation_0, values = (var_47218_cast_fp16, var_47338_cast_fp16))[name = tensor("op_47372_cast_fp16")]; + tensor var_47374_equation_0 = const()[name = tensor("op_47374_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47374_cast_fp16 = einsum(equation = var_47374_equation_0, values = (var_47222_cast_fp16, var_47339_cast_fp16))[name = tensor("op_47374_cast_fp16")]; + tensor var_47376_equation_0 = const()[name = tensor("op_47376_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47376_cast_fp16 = einsum(equation = var_47376_equation_0, values = (var_47226_cast_fp16, var_47340_cast_fp16))[name = tensor("op_47376_cast_fp16")]; + tensor var_47378_equation_0 = const()[name = tensor("op_47378_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47378_cast_fp16 = einsum(equation = var_47378_equation_0, values = (var_47230_cast_fp16, var_47341_cast_fp16))[name = tensor("op_47378_cast_fp16")]; + tensor var_47380_equation_0 = const()[name = tensor("op_47380_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47380_cast_fp16 = einsum(equation = var_47380_equation_0, values = (var_47234_cast_fp16, var_47342_cast_fp16))[name = tensor("op_47380_cast_fp16")]; + tensor var_47382_equation_0 = const()[name = tensor("op_47382_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47382_cast_fp16 = einsum(equation = var_47382_equation_0, values = (var_47238_cast_fp16, var_47343_cast_fp16))[name = tensor("op_47382_cast_fp16")]; + tensor var_47384_equation_0 = const()[name = tensor("op_47384_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47384_cast_fp16 = einsum(equation = var_47384_equation_0, values = (var_47242_cast_fp16, var_47344_cast_fp16))[name = tensor("op_47384_cast_fp16")]; + tensor input_601_interleave_0 = const()[name = tensor("input_601_interleave_0"), val = tensor(false)]; + tensor input_601_cast_fp16 = concat(axis = var_30355, interleave = input_601_interleave_0, values = (var_47346_cast_fp16, var_47348_cast_fp16, var_47350_cast_fp16, var_47352_cast_fp16, var_47354_cast_fp16, var_47356_cast_fp16, var_47358_cast_fp16, var_47360_cast_fp16, var_47362_cast_fp16, var_47364_cast_fp16, var_47366_cast_fp16, var_47368_cast_fp16, var_47370_cast_fp16, var_47372_cast_fp16, var_47374_cast_fp16, var_47376_cast_fp16, var_47378_cast_fp16, var_47380_cast_fp16, var_47382_cast_fp16, var_47384_cast_fp16))[name = tensor("input_601_cast_fp16")]; + tensor var_47390 = const()[name = tensor("op_47390"), val = tensor([1, 1])]; + tensor var_47392 = const()[name = tensor("op_47392"), val = tensor([1, 1])]; + tensor var_47394_pad_type_0 = const()[name = tensor("op_47394_pad_type_0"), val = tensor("custom")]; + tensor var_47394_pad_0 = const()[name = tensor("op_47394_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_8_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_8_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3954385088)))]; + tensor up_blocks_0_attentions_1_transformer_blocks_8_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_8_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3957661952)))]; + tensor var_47394_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_8_attn1_to_out_0_bias_to_fp16, dilations = var_47392, groups = var_30355, pad = var_47394_pad_0, pad_type = var_47394_pad_type_0, strides = var_47390, weight = up_blocks_0_attentions_1_transformer_blocks_8_attn1_to_out_0_weight_to_fp16, x = input_601_cast_fp16)[name = tensor("op_47394_cast_fp16")]; + tensor inputs_315_cast_fp16 = add(x = var_47394_cast_fp16, y = inputs_313_cast_fp16)[name = tensor("inputs_315_cast_fp16")]; + tensor hidden_states_411_axes_0 = const()[name = tensor("hidden_states_411_axes_0"), val = tensor([1])]; + tensor hidden_states_411_gamma_0_to_fp16 = const()[name = tensor("hidden_states_411_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3957664576)))]; + tensor hidden_states_411_beta_0_to_fp16 = const()[name = tensor("hidden_states_411_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3957667200)))]; + tensor var_47404_to_fp16 = const()[name = tensor("op_47404_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_411_cast_fp16 = layer_norm(axes = hidden_states_411_axes_0, beta = hidden_states_411_beta_0_to_fp16, epsilon = var_47404_to_fp16, gamma = hidden_states_411_gamma_0_to_fp16, x = inputs_315_cast_fp16)[name = tensor("hidden_states_411_cast_fp16")]; + tensor var_47419 = const()[name = tensor("op_47419"), val = tensor([1, 1])]; + tensor var_47421 = const()[name = tensor("op_47421"), val = tensor([1, 1])]; + tensor q_211_pad_type_0 = const()[name = tensor("q_211_pad_type_0"), val = tensor("custom")]; + tensor q_211_pad_0 = const()[name = tensor("q_211_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_8_attn2_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_8_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3957669824)))]; + tensor q_211_cast_fp16 = conv(dilations = var_47421, groups = var_30355, pad = q_211_pad_0, pad_type = q_211_pad_type_0, strides = var_47419, weight = up_blocks_0_attentions_1_transformer_blocks_8_attn2_to_q_weight_to_fp16, x = hidden_states_411_cast_fp16)[name = tensor("q_211_cast_fp16")]; + tensor var_47425 = const()[name = tensor("op_47425"), val = tensor([1, 1])]; + tensor var_47427 = const()[name = tensor("op_47427"), val = tensor([1, 1])]; + tensor k_421_pad_type_0 = const()[name = tensor("k_421_pad_type_0"), val = tensor("custom")]; + tensor k_421_pad_0 = const()[name = tensor("k_421_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_8_attn2_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_8_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3960946688)))]; + tensor k_421_cast_fp16 = conv(dilations = var_47427, groups = var_30355, pad = k_421_pad_0, pad_type = k_421_pad_type_0, strides = var_47425, weight = up_blocks_0_attentions_1_transformer_blocks_8_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_421_cast_fp16")]; + tensor var_47431 = const()[name = tensor("op_47431"), val = tensor([1, 1])]; + tensor var_47433 = const()[name = tensor("op_47433"), val = tensor([1, 1])]; + tensor v_211_pad_type_0 = const()[name = tensor("v_211_pad_type_0"), val = tensor("custom")]; + tensor v_211_pad_0 = const()[name = tensor("v_211_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_8_attn2_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_8_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3966189632)))]; + tensor v_211_cast_fp16 = conv(dilations = var_47433, groups = var_30355, pad = v_211_pad_0, pad_type = v_211_pad_type_0, strides = var_47431, weight = up_blocks_0_attentions_1_transformer_blocks_8_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_211_cast_fp16")]; + tensor var_47437_begin_0 = const()[name = tensor("op_47437_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_47437_end_0 = const()[name = tensor("op_47437_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_47437_end_mask_0 = const()[name = tensor("op_47437_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47437_cast_fp16 = slice_by_index(begin = var_47437_begin_0, end = var_47437_end_0, end_mask = var_47437_end_mask_0, x = q_211_cast_fp16)[name = tensor("op_47437_cast_fp16")]; + tensor var_47441_begin_0 = const()[name = tensor("op_47441_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_47441_end_0 = const()[name = tensor("op_47441_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_47441_end_mask_0 = const()[name = tensor("op_47441_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47441_cast_fp16 = slice_by_index(begin = var_47441_begin_0, end = var_47441_end_0, end_mask = var_47441_end_mask_0, x = q_211_cast_fp16)[name = tensor("op_47441_cast_fp16")]; + tensor var_47445_begin_0 = const()[name = tensor("op_47445_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_47445_end_0 = const()[name = tensor("op_47445_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_47445_end_mask_0 = const()[name = tensor("op_47445_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47445_cast_fp16 = slice_by_index(begin = var_47445_begin_0, end = var_47445_end_0, end_mask = var_47445_end_mask_0, x = q_211_cast_fp16)[name = tensor("op_47445_cast_fp16")]; + tensor var_47449_begin_0 = const()[name = tensor("op_47449_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_47449_end_0 = const()[name = tensor("op_47449_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_47449_end_mask_0 = const()[name = tensor("op_47449_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47449_cast_fp16 = slice_by_index(begin = var_47449_begin_0, end = var_47449_end_0, end_mask = var_47449_end_mask_0, x = q_211_cast_fp16)[name = tensor("op_47449_cast_fp16")]; + tensor var_47453_begin_0 = const()[name = tensor("op_47453_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_47453_end_0 = const()[name = tensor("op_47453_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_47453_end_mask_0 = const()[name = tensor("op_47453_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47453_cast_fp16 = slice_by_index(begin = var_47453_begin_0, end = var_47453_end_0, end_mask = var_47453_end_mask_0, x = q_211_cast_fp16)[name = tensor("op_47453_cast_fp16")]; + tensor var_47457_begin_0 = const()[name = tensor("op_47457_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_47457_end_0 = const()[name = tensor("op_47457_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_47457_end_mask_0 = const()[name = tensor("op_47457_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47457_cast_fp16 = slice_by_index(begin = var_47457_begin_0, end = var_47457_end_0, end_mask = var_47457_end_mask_0, x = q_211_cast_fp16)[name = tensor("op_47457_cast_fp16")]; + tensor var_47461_begin_0 = const()[name = tensor("op_47461_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_47461_end_0 = const()[name = tensor("op_47461_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_47461_end_mask_0 = const()[name = tensor("op_47461_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47461_cast_fp16 = slice_by_index(begin = var_47461_begin_0, end = var_47461_end_0, end_mask = var_47461_end_mask_0, x = q_211_cast_fp16)[name = tensor("op_47461_cast_fp16")]; + tensor var_47465_begin_0 = const()[name = tensor("op_47465_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_47465_end_0 = const()[name = tensor("op_47465_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_47465_end_mask_0 = const()[name = tensor("op_47465_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47465_cast_fp16 = slice_by_index(begin = var_47465_begin_0, end = var_47465_end_0, end_mask = var_47465_end_mask_0, x = q_211_cast_fp16)[name = tensor("op_47465_cast_fp16")]; + tensor var_47469_begin_0 = const()[name = tensor("op_47469_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_47469_end_0 = const()[name = tensor("op_47469_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_47469_end_mask_0 = const()[name = tensor("op_47469_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47469_cast_fp16 = slice_by_index(begin = var_47469_begin_0, end = var_47469_end_0, end_mask = var_47469_end_mask_0, x = q_211_cast_fp16)[name = tensor("op_47469_cast_fp16")]; + tensor var_47473_begin_0 = const()[name = tensor("op_47473_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_47473_end_0 = const()[name = tensor("op_47473_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_47473_end_mask_0 = const()[name = tensor("op_47473_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47473_cast_fp16 = slice_by_index(begin = var_47473_begin_0, end = var_47473_end_0, end_mask = var_47473_end_mask_0, x = q_211_cast_fp16)[name = tensor("op_47473_cast_fp16")]; + tensor var_47477_begin_0 = const()[name = tensor("op_47477_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_47477_end_0 = const()[name = tensor("op_47477_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_47477_end_mask_0 = const()[name = tensor("op_47477_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47477_cast_fp16 = slice_by_index(begin = var_47477_begin_0, end = var_47477_end_0, end_mask = var_47477_end_mask_0, x = q_211_cast_fp16)[name = tensor("op_47477_cast_fp16")]; + tensor var_47481_begin_0 = const()[name = tensor("op_47481_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_47481_end_0 = const()[name = tensor("op_47481_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_47481_end_mask_0 = const()[name = tensor("op_47481_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47481_cast_fp16 = slice_by_index(begin = var_47481_begin_0, end = var_47481_end_0, end_mask = var_47481_end_mask_0, x = q_211_cast_fp16)[name = tensor("op_47481_cast_fp16")]; + tensor var_47485_begin_0 = const()[name = tensor("op_47485_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_47485_end_0 = const()[name = tensor("op_47485_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_47485_end_mask_0 = const()[name = tensor("op_47485_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47485_cast_fp16 = slice_by_index(begin = var_47485_begin_0, end = var_47485_end_0, end_mask = var_47485_end_mask_0, x = q_211_cast_fp16)[name = tensor("op_47485_cast_fp16")]; + tensor var_47489_begin_0 = const()[name = tensor("op_47489_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_47489_end_0 = const()[name = tensor("op_47489_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_47489_end_mask_0 = const()[name = tensor("op_47489_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47489_cast_fp16 = slice_by_index(begin = var_47489_begin_0, end = var_47489_end_0, end_mask = var_47489_end_mask_0, x = q_211_cast_fp16)[name = tensor("op_47489_cast_fp16")]; + tensor var_47493_begin_0 = const()[name = tensor("op_47493_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_47493_end_0 = const()[name = tensor("op_47493_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_47493_end_mask_0 = const()[name = tensor("op_47493_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47493_cast_fp16 = slice_by_index(begin = var_47493_begin_0, end = var_47493_end_0, end_mask = var_47493_end_mask_0, x = q_211_cast_fp16)[name = tensor("op_47493_cast_fp16")]; + tensor var_47497_begin_0 = const()[name = tensor("op_47497_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_47497_end_0 = const()[name = tensor("op_47497_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_47497_end_mask_0 = const()[name = tensor("op_47497_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47497_cast_fp16 = slice_by_index(begin = var_47497_begin_0, end = var_47497_end_0, end_mask = var_47497_end_mask_0, x = q_211_cast_fp16)[name = tensor("op_47497_cast_fp16")]; + tensor var_47501_begin_0 = const()[name = tensor("op_47501_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_47501_end_0 = const()[name = tensor("op_47501_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_47501_end_mask_0 = const()[name = tensor("op_47501_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47501_cast_fp16 = slice_by_index(begin = var_47501_begin_0, end = var_47501_end_0, end_mask = var_47501_end_mask_0, x = q_211_cast_fp16)[name = tensor("op_47501_cast_fp16")]; + tensor var_47505_begin_0 = const()[name = tensor("op_47505_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_47505_end_0 = const()[name = tensor("op_47505_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_47505_end_mask_0 = const()[name = tensor("op_47505_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47505_cast_fp16 = slice_by_index(begin = var_47505_begin_0, end = var_47505_end_0, end_mask = var_47505_end_mask_0, x = q_211_cast_fp16)[name = tensor("op_47505_cast_fp16")]; + tensor var_47509_begin_0 = const()[name = tensor("op_47509_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_47509_end_0 = const()[name = tensor("op_47509_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_47509_end_mask_0 = const()[name = tensor("op_47509_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47509_cast_fp16 = slice_by_index(begin = var_47509_begin_0, end = var_47509_end_0, end_mask = var_47509_end_mask_0, x = q_211_cast_fp16)[name = tensor("op_47509_cast_fp16")]; + tensor var_47513_begin_0 = const()[name = tensor("op_47513_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_47513_end_0 = const()[name = tensor("op_47513_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_47513_end_mask_0 = const()[name = tensor("op_47513_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47513_cast_fp16 = slice_by_index(begin = var_47513_begin_0, end = var_47513_end_0, end_mask = var_47513_end_mask_0, x = q_211_cast_fp16)[name = tensor("op_47513_cast_fp16")]; + tensor k_423_perm_0 = const()[name = tensor("k_423_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_47520_begin_0 = const()[name = tensor("op_47520_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_47520_end_0 = const()[name = tensor("op_47520_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_47520_end_mask_0 = const()[name = tensor("op_47520_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_423_cast_fp16 = transpose(perm = k_423_perm_0, x = k_421_cast_fp16)[name = tensor("transpose_34")]; + tensor var_47520_cast_fp16 = slice_by_index(begin = var_47520_begin_0, end = var_47520_end_0, end_mask = var_47520_end_mask_0, x = k_423_cast_fp16)[name = tensor("op_47520_cast_fp16")]; + tensor var_47524_begin_0 = const()[name = tensor("op_47524_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_47524_end_0 = const()[name = tensor("op_47524_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_47524_end_mask_0 = const()[name = tensor("op_47524_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47524_cast_fp16 = slice_by_index(begin = var_47524_begin_0, end = var_47524_end_0, end_mask = var_47524_end_mask_0, x = k_423_cast_fp16)[name = tensor("op_47524_cast_fp16")]; + tensor var_47528_begin_0 = const()[name = tensor("op_47528_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_47528_end_0 = const()[name = tensor("op_47528_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_47528_end_mask_0 = const()[name = tensor("op_47528_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47528_cast_fp16 = slice_by_index(begin = var_47528_begin_0, end = var_47528_end_0, end_mask = var_47528_end_mask_0, x = k_423_cast_fp16)[name = tensor("op_47528_cast_fp16")]; + tensor var_47532_begin_0 = const()[name = tensor("op_47532_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_47532_end_0 = const()[name = tensor("op_47532_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_47532_end_mask_0 = const()[name = tensor("op_47532_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47532_cast_fp16 = slice_by_index(begin = var_47532_begin_0, end = var_47532_end_0, end_mask = var_47532_end_mask_0, x = k_423_cast_fp16)[name = tensor("op_47532_cast_fp16")]; + tensor var_47536_begin_0 = const()[name = tensor("op_47536_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_47536_end_0 = const()[name = tensor("op_47536_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_47536_end_mask_0 = const()[name = tensor("op_47536_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47536_cast_fp16 = slice_by_index(begin = var_47536_begin_0, end = var_47536_end_0, end_mask = var_47536_end_mask_0, x = k_423_cast_fp16)[name = tensor("op_47536_cast_fp16")]; + tensor var_47540_begin_0 = const()[name = tensor("op_47540_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_47540_end_0 = const()[name = tensor("op_47540_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_47540_end_mask_0 = const()[name = tensor("op_47540_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47540_cast_fp16 = slice_by_index(begin = var_47540_begin_0, end = var_47540_end_0, end_mask = var_47540_end_mask_0, x = k_423_cast_fp16)[name = tensor("op_47540_cast_fp16")]; + tensor var_47544_begin_0 = const()[name = tensor("op_47544_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_47544_end_0 = const()[name = tensor("op_47544_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_47544_end_mask_0 = const()[name = tensor("op_47544_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47544_cast_fp16 = slice_by_index(begin = var_47544_begin_0, end = var_47544_end_0, end_mask = var_47544_end_mask_0, x = k_423_cast_fp16)[name = tensor("op_47544_cast_fp16")]; + tensor var_47548_begin_0 = const()[name = tensor("op_47548_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_47548_end_0 = const()[name = tensor("op_47548_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_47548_end_mask_0 = const()[name = tensor("op_47548_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47548_cast_fp16 = slice_by_index(begin = var_47548_begin_0, end = var_47548_end_0, end_mask = var_47548_end_mask_0, x = k_423_cast_fp16)[name = tensor("op_47548_cast_fp16")]; + tensor var_47552_begin_0 = const()[name = tensor("op_47552_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_47552_end_0 = const()[name = tensor("op_47552_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_47552_end_mask_0 = const()[name = tensor("op_47552_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47552_cast_fp16 = slice_by_index(begin = var_47552_begin_0, end = var_47552_end_0, end_mask = var_47552_end_mask_0, x = k_423_cast_fp16)[name = tensor("op_47552_cast_fp16")]; + tensor var_47556_begin_0 = const()[name = tensor("op_47556_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_47556_end_0 = const()[name = tensor("op_47556_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_47556_end_mask_0 = const()[name = tensor("op_47556_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47556_cast_fp16 = slice_by_index(begin = var_47556_begin_0, end = var_47556_end_0, end_mask = var_47556_end_mask_0, x = k_423_cast_fp16)[name = tensor("op_47556_cast_fp16")]; + tensor var_47560_begin_0 = const()[name = tensor("op_47560_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_47560_end_0 = const()[name = tensor("op_47560_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_47560_end_mask_0 = const()[name = tensor("op_47560_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47560_cast_fp16 = slice_by_index(begin = var_47560_begin_0, end = var_47560_end_0, end_mask = var_47560_end_mask_0, x = k_423_cast_fp16)[name = tensor("op_47560_cast_fp16")]; + tensor var_47564_begin_0 = const()[name = tensor("op_47564_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_47564_end_0 = const()[name = tensor("op_47564_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_47564_end_mask_0 = const()[name = tensor("op_47564_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47564_cast_fp16 = slice_by_index(begin = var_47564_begin_0, end = var_47564_end_0, end_mask = var_47564_end_mask_0, x = k_423_cast_fp16)[name = tensor("op_47564_cast_fp16")]; + tensor var_47568_begin_0 = const()[name = tensor("op_47568_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_47568_end_0 = const()[name = tensor("op_47568_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_47568_end_mask_0 = const()[name = tensor("op_47568_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47568_cast_fp16 = slice_by_index(begin = var_47568_begin_0, end = var_47568_end_0, end_mask = var_47568_end_mask_0, x = k_423_cast_fp16)[name = tensor("op_47568_cast_fp16")]; + tensor var_47572_begin_0 = const()[name = tensor("op_47572_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_47572_end_0 = const()[name = tensor("op_47572_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_47572_end_mask_0 = const()[name = tensor("op_47572_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47572_cast_fp16 = slice_by_index(begin = var_47572_begin_0, end = var_47572_end_0, end_mask = var_47572_end_mask_0, x = k_423_cast_fp16)[name = tensor("op_47572_cast_fp16")]; + tensor var_47576_begin_0 = const()[name = tensor("op_47576_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_47576_end_0 = const()[name = tensor("op_47576_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_47576_end_mask_0 = const()[name = tensor("op_47576_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47576_cast_fp16 = slice_by_index(begin = var_47576_begin_0, end = var_47576_end_0, end_mask = var_47576_end_mask_0, x = k_423_cast_fp16)[name = tensor("op_47576_cast_fp16")]; + tensor var_47580_begin_0 = const()[name = tensor("op_47580_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_47580_end_0 = const()[name = tensor("op_47580_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_47580_end_mask_0 = const()[name = tensor("op_47580_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47580_cast_fp16 = slice_by_index(begin = var_47580_begin_0, end = var_47580_end_0, end_mask = var_47580_end_mask_0, x = k_423_cast_fp16)[name = tensor("op_47580_cast_fp16")]; + tensor var_47584_begin_0 = const()[name = tensor("op_47584_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_47584_end_0 = const()[name = tensor("op_47584_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_47584_end_mask_0 = const()[name = tensor("op_47584_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47584_cast_fp16 = slice_by_index(begin = var_47584_begin_0, end = var_47584_end_0, end_mask = var_47584_end_mask_0, x = k_423_cast_fp16)[name = tensor("op_47584_cast_fp16")]; + tensor var_47588_begin_0 = const()[name = tensor("op_47588_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_47588_end_0 = const()[name = tensor("op_47588_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_47588_end_mask_0 = const()[name = tensor("op_47588_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47588_cast_fp16 = slice_by_index(begin = var_47588_begin_0, end = var_47588_end_0, end_mask = var_47588_end_mask_0, x = k_423_cast_fp16)[name = tensor("op_47588_cast_fp16")]; + tensor var_47592_begin_0 = const()[name = tensor("op_47592_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_47592_end_0 = const()[name = tensor("op_47592_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_47592_end_mask_0 = const()[name = tensor("op_47592_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47592_cast_fp16 = slice_by_index(begin = var_47592_begin_0, end = var_47592_end_0, end_mask = var_47592_end_mask_0, x = k_423_cast_fp16)[name = tensor("op_47592_cast_fp16")]; + tensor var_47596_begin_0 = const()[name = tensor("op_47596_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_47596_end_0 = const()[name = tensor("op_47596_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_47596_end_mask_0 = const()[name = tensor("op_47596_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47596_cast_fp16 = slice_by_index(begin = var_47596_begin_0, end = var_47596_end_0, end_mask = var_47596_end_mask_0, x = k_423_cast_fp16)[name = tensor("op_47596_cast_fp16")]; + tensor var_47598_begin_0 = const()[name = tensor("op_47598_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_47598_end_0 = const()[name = tensor("op_47598_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_47598_end_mask_0 = const()[name = tensor("op_47598_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47598_cast_fp16 = slice_by_index(begin = var_47598_begin_0, end = var_47598_end_0, end_mask = var_47598_end_mask_0, x = v_211_cast_fp16)[name = tensor("op_47598_cast_fp16")]; + tensor var_47602_begin_0 = const()[name = tensor("op_47602_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_47602_end_0 = const()[name = tensor("op_47602_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_47602_end_mask_0 = const()[name = tensor("op_47602_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47602_cast_fp16 = slice_by_index(begin = var_47602_begin_0, end = var_47602_end_0, end_mask = var_47602_end_mask_0, x = v_211_cast_fp16)[name = tensor("op_47602_cast_fp16")]; + tensor var_47606_begin_0 = const()[name = tensor("op_47606_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_47606_end_0 = const()[name = tensor("op_47606_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_47606_end_mask_0 = const()[name = tensor("op_47606_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47606_cast_fp16 = slice_by_index(begin = var_47606_begin_0, end = var_47606_end_0, end_mask = var_47606_end_mask_0, x = v_211_cast_fp16)[name = tensor("op_47606_cast_fp16")]; + tensor var_47610_begin_0 = const()[name = tensor("op_47610_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_47610_end_0 = const()[name = tensor("op_47610_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_47610_end_mask_0 = const()[name = tensor("op_47610_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47610_cast_fp16 = slice_by_index(begin = var_47610_begin_0, end = var_47610_end_0, end_mask = var_47610_end_mask_0, x = v_211_cast_fp16)[name = tensor("op_47610_cast_fp16")]; + tensor var_47614_begin_0 = const()[name = tensor("op_47614_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_47614_end_0 = const()[name = tensor("op_47614_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_47614_end_mask_0 = const()[name = tensor("op_47614_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47614_cast_fp16 = slice_by_index(begin = var_47614_begin_0, end = var_47614_end_0, end_mask = var_47614_end_mask_0, x = v_211_cast_fp16)[name = tensor("op_47614_cast_fp16")]; + tensor var_47618_begin_0 = const()[name = tensor("op_47618_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_47618_end_0 = const()[name = tensor("op_47618_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_47618_end_mask_0 = const()[name = tensor("op_47618_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47618_cast_fp16 = slice_by_index(begin = var_47618_begin_0, end = var_47618_end_0, end_mask = var_47618_end_mask_0, x = v_211_cast_fp16)[name = tensor("op_47618_cast_fp16")]; + tensor var_47622_begin_0 = const()[name = tensor("op_47622_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_47622_end_0 = const()[name = tensor("op_47622_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_47622_end_mask_0 = const()[name = tensor("op_47622_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47622_cast_fp16 = slice_by_index(begin = var_47622_begin_0, end = var_47622_end_0, end_mask = var_47622_end_mask_0, x = v_211_cast_fp16)[name = tensor("op_47622_cast_fp16")]; + tensor var_47626_begin_0 = const()[name = tensor("op_47626_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_47626_end_0 = const()[name = tensor("op_47626_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_47626_end_mask_0 = const()[name = tensor("op_47626_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47626_cast_fp16 = slice_by_index(begin = var_47626_begin_0, end = var_47626_end_0, end_mask = var_47626_end_mask_0, x = v_211_cast_fp16)[name = tensor("op_47626_cast_fp16")]; + tensor var_47630_begin_0 = const()[name = tensor("op_47630_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_47630_end_0 = const()[name = tensor("op_47630_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_47630_end_mask_0 = const()[name = tensor("op_47630_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47630_cast_fp16 = slice_by_index(begin = var_47630_begin_0, end = var_47630_end_0, end_mask = var_47630_end_mask_0, x = v_211_cast_fp16)[name = tensor("op_47630_cast_fp16")]; + tensor var_47634_begin_0 = const()[name = tensor("op_47634_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_47634_end_0 = const()[name = tensor("op_47634_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_47634_end_mask_0 = const()[name = tensor("op_47634_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47634_cast_fp16 = slice_by_index(begin = var_47634_begin_0, end = var_47634_end_0, end_mask = var_47634_end_mask_0, x = v_211_cast_fp16)[name = tensor("op_47634_cast_fp16")]; + tensor var_47638_begin_0 = const()[name = tensor("op_47638_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_47638_end_0 = const()[name = tensor("op_47638_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_47638_end_mask_0 = const()[name = tensor("op_47638_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47638_cast_fp16 = slice_by_index(begin = var_47638_begin_0, end = var_47638_end_0, end_mask = var_47638_end_mask_0, x = v_211_cast_fp16)[name = tensor("op_47638_cast_fp16")]; + tensor var_47642_begin_0 = const()[name = tensor("op_47642_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_47642_end_0 = const()[name = tensor("op_47642_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_47642_end_mask_0 = const()[name = tensor("op_47642_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47642_cast_fp16 = slice_by_index(begin = var_47642_begin_0, end = var_47642_end_0, end_mask = var_47642_end_mask_0, x = v_211_cast_fp16)[name = tensor("op_47642_cast_fp16")]; + tensor var_47646_begin_0 = const()[name = tensor("op_47646_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_47646_end_0 = const()[name = tensor("op_47646_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_47646_end_mask_0 = const()[name = tensor("op_47646_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47646_cast_fp16 = slice_by_index(begin = var_47646_begin_0, end = var_47646_end_0, end_mask = var_47646_end_mask_0, x = v_211_cast_fp16)[name = tensor("op_47646_cast_fp16")]; + tensor var_47650_begin_0 = const()[name = tensor("op_47650_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_47650_end_0 = const()[name = tensor("op_47650_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_47650_end_mask_0 = const()[name = tensor("op_47650_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47650_cast_fp16 = slice_by_index(begin = var_47650_begin_0, end = var_47650_end_0, end_mask = var_47650_end_mask_0, x = v_211_cast_fp16)[name = tensor("op_47650_cast_fp16")]; + tensor var_47654_begin_0 = const()[name = tensor("op_47654_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_47654_end_0 = const()[name = tensor("op_47654_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_47654_end_mask_0 = const()[name = tensor("op_47654_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47654_cast_fp16 = slice_by_index(begin = var_47654_begin_0, end = var_47654_end_0, end_mask = var_47654_end_mask_0, x = v_211_cast_fp16)[name = tensor("op_47654_cast_fp16")]; + tensor var_47658_begin_0 = const()[name = tensor("op_47658_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_47658_end_0 = const()[name = tensor("op_47658_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_47658_end_mask_0 = const()[name = tensor("op_47658_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47658_cast_fp16 = slice_by_index(begin = var_47658_begin_0, end = var_47658_end_0, end_mask = var_47658_end_mask_0, x = v_211_cast_fp16)[name = tensor("op_47658_cast_fp16")]; + tensor var_47662_begin_0 = const()[name = tensor("op_47662_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_47662_end_0 = const()[name = tensor("op_47662_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_47662_end_mask_0 = const()[name = tensor("op_47662_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47662_cast_fp16 = slice_by_index(begin = var_47662_begin_0, end = var_47662_end_0, end_mask = var_47662_end_mask_0, x = v_211_cast_fp16)[name = tensor("op_47662_cast_fp16")]; + tensor var_47666_begin_0 = const()[name = tensor("op_47666_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_47666_end_0 = const()[name = tensor("op_47666_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_47666_end_mask_0 = const()[name = tensor("op_47666_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47666_cast_fp16 = slice_by_index(begin = var_47666_begin_0, end = var_47666_end_0, end_mask = var_47666_end_mask_0, x = v_211_cast_fp16)[name = tensor("op_47666_cast_fp16")]; + tensor var_47670_begin_0 = const()[name = tensor("op_47670_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_47670_end_0 = const()[name = tensor("op_47670_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_47670_end_mask_0 = const()[name = tensor("op_47670_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47670_cast_fp16 = slice_by_index(begin = var_47670_begin_0, end = var_47670_end_0, end_mask = var_47670_end_mask_0, x = v_211_cast_fp16)[name = tensor("op_47670_cast_fp16")]; + tensor var_47674_begin_0 = const()[name = tensor("op_47674_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_47674_end_0 = const()[name = tensor("op_47674_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_47674_end_mask_0 = const()[name = tensor("op_47674_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47674_cast_fp16 = slice_by_index(begin = var_47674_begin_0, end = var_47674_end_0, end_mask = var_47674_end_mask_0, x = v_211_cast_fp16)[name = tensor("op_47674_cast_fp16")]; + tensor var_47678_equation_0 = const()[name = tensor("op_47678_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47678_cast_fp16 = einsum(equation = var_47678_equation_0, values = (var_47520_cast_fp16, var_47437_cast_fp16))[name = tensor("op_47678_cast_fp16")]; + tensor var_47679_to_fp16 = const()[name = tensor("op_47679_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4041_cast_fp16 = mul(x = var_47678_cast_fp16, y = var_47679_to_fp16)[name = tensor("aw_4041_cast_fp16")]; + tensor var_47682_equation_0 = const()[name = tensor("op_47682_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47682_cast_fp16 = einsum(equation = var_47682_equation_0, values = (var_47524_cast_fp16, var_47441_cast_fp16))[name = tensor("op_47682_cast_fp16")]; + tensor var_47683_to_fp16 = const()[name = tensor("op_47683_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4043_cast_fp16 = mul(x = var_47682_cast_fp16, y = var_47683_to_fp16)[name = tensor("aw_4043_cast_fp16")]; + tensor var_47686_equation_0 = const()[name = tensor("op_47686_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47686_cast_fp16 = einsum(equation = var_47686_equation_0, values = (var_47528_cast_fp16, var_47445_cast_fp16))[name = tensor("op_47686_cast_fp16")]; + tensor var_47687_to_fp16 = const()[name = tensor("op_47687_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4045_cast_fp16 = mul(x = var_47686_cast_fp16, y = var_47687_to_fp16)[name = tensor("aw_4045_cast_fp16")]; + tensor var_47690_equation_0 = const()[name = tensor("op_47690_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47690_cast_fp16 = einsum(equation = var_47690_equation_0, values = (var_47532_cast_fp16, var_47449_cast_fp16))[name = tensor("op_47690_cast_fp16")]; + tensor var_47691_to_fp16 = const()[name = tensor("op_47691_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4047_cast_fp16 = mul(x = var_47690_cast_fp16, y = var_47691_to_fp16)[name = tensor("aw_4047_cast_fp16")]; + tensor var_47694_equation_0 = const()[name = tensor("op_47694_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47694_cast_fp16 = einsum(equation = var_47694_equation_0, values = (var_47536_cast_fp16, var_47453_cast_fp16))[name = tensor("op_47694_cast_fp16")]; + tensor var_47695_to_fp16 = const()[name = tensor("op_47695_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4049_cast_fp16 = mul(x = var_47694_cast_fp16, y = var_47695_to_fp16)[name = tensor("aw_4049_cast_fp16")]; + tensor var_47698_equation_0 = const()[name = tensor("op_47698_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47698_cast_fp16 = einsum(equation = var_47698_equation_0, values = (var_47540_cast_fp16, var_47457_cast_fp16))[name = tensor("op_47698_cast_fp16")]; + tensor var_47699_to_fp16 = const()[name = tensor("op_47699_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4051_cast_fp16 = mul(x = var_47698_cast_fp16, y = var_47699_to_fp16)[name = tensor("aw_4051_cast_fp16")]; + tensor var_47702_equation_0 = const()[name = tensor("op_47702_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47702_cast_fp16 = einsum(equation = var_47702_equation_0, values = (var_47544_cast_fp16, var_47461_cast_fp16))[name = tensor("op_47702_cast_fp16")]; + tensor var_47703_to_fp16 = const()[name = tensor("op_47703_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4053_cast_fp16 = mul(x = var_47702_cast_fp16, y = var_47703_to_fp16)[name = tensor("aw_4053_cast_fp16")]; + tensor var_47706_equation_0 = const()[name = tensor("op_47706_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47706_cast_fp16 = einsum(equation = var_47706_equation_0, values = (var_47548_cast_fp16, var_47465_cast_fp16))[name = tensor("op_47706_cast_fp16")]; + tensor var_47707_to_fp16 = const()[name = tensor("op_47707_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4055_cast_fp16 = mul(x = var_47706_cast_fp16, y = var_47707_to_fp16)[name = tensor("aw_4055_cast_fp16")]; + tensor var_47710_equation_0 = const()[name = tensor("op_47710_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47710_cast_fp16 = einsum(equation = var_47710_equation_0, values = (var_47552_cast_fp16, var_47469_cast_fp16))[name = tensor("op_47710_cast_fp16")]; + tensor var_47711_to_fp16 = const()[name = tensor("op_47711_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4057_cast_fp16 = mul(x = var_47710_cast_fp16, y = var_47711_to_fp16)[name = tensor("aw_4057_cast_fp16")]; + tensor var_47714_equation_0 = const()[name = tensor("op_47714_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47714_cast_fp16 = einsum(equation = var_47714_equation_0, values = (var_47556_cast_fp16, var_47473_cast_fp16))[name = tensor("op_47714_cast_fp16")]; + tensor var_47715_to_fp16 = const()[name = tensor("op_47715_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4059_cast_fp16 = mul(x = var_47714_cast_fp16, y = var_47715_to_fp16)[name = tensor("aw_4059_cast_fp16")]; + tensor var_47718_equation_0 = const()[name = tensor("op_47718_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47718_cast_fp16 = einsum(equation = var_47718_equation_0, values = (var_47560_cast_fp16, var_47477_cast_fp16))[name = tensor("op_47718_cast_fp16")]; + tensor var_47719_to_fp16 = const()[name = tensor("op_47719_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4061_cast_fp16 = mul(x = var_47718_cast_fp16, y = var_47719_to_fp16)[name = tensor("aw_4061_cast_fp16")]; + tensor var_47722_equation_0 = const()[name = tensor("op_47722_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47722_cast_fp16 = einsum(equation = var_47722_equation_0, values = (var_47564_cast_fp16, var_47481_cast_fp16))[name = tensor("op_47722_cast_fp16")]; + tensor var_47723_to_fp16 = const()[name = tensor("op_47723_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4063_cast_fp16 = mul(x = var_47722_cast_fp16, y = var_47723_to_fp16)[name = tensor("aw_4063_cast_fp16")]; + tensor var_47726_equation_0 = const()[name = tensor("op_47726_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47726_cast_fp16 = einsum(equation = var_47726_equation_0, values = (var_47568_cast_fp16, var_47485_cast_fp16))[name = tensor("op_47726_cast_fp16")]; + tensor var_47727_to_fp16 = const()[name = tensor("op_47727_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4065_cast_fp16 = mul(x = var_47726_cast_fp16, y = var_47727_to_fp16)[name = tensor("aw_4065_cast_fp16")]; + tensor var_47730_equation_0 = const()[name = tensor("op_47730_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47730_cast_fp16 = einsum(equation = var_47730_equation_0, values = (var_47572_cast_fp16, var_47489_cast_fp16))[name = tensor("op_47730_cast_fp16")]; + tensor var_47731_to_fp16 = const()[name = tensor("op_47731_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4067_cast_fp16 = mul(x = var_47730_cast_fp16, y = var_47731_to_fp16)[name = tensor("aw_4067_cast_fp16")]; + tensor var_47734_equation_0 = const()[name = tensor("op_47734_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47734_cast_fp16 = einsum(equation = var_47734_equation_0, values = (var_47576_cast_fp16, var_47493_cast_fp16))[name = tensor("op_47734_cast_fp16")]; + tensor var_47735_to_fp16 = const()[name = tensor("op_47735_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4069_cast_fp16 = mul(x = var_47734_cast_fp16, y = var_47735_to_fp16)[name = tensor("aw_4069_cast_fp16")]; + tensor var_47738_equation_0 = const()[name = tensor("op_47738_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47738_cast_fp16 = einsum(equation = var_47738_equation_0, values = (var_47580_cast_fp16, var_47497_cast_fp16))[name = tensor("op_47738_cast_fp16")]; + tensor var_47739_to_fp16 = const()[name = tensor("op_47739_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4071_cast_fp16 = mul(x = var_47738_cast_fp16, y = var_47739_to_fp16)[name = tensor("aw_4071_cast_fp16")]; + tensor var_47742_equation_0 = const()[name = tensor("op_47742_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47742_cast_fp16 = einsum(equation = var_47742_equation_0, values = (var_47584_cast_fp16, var_47501_cast_fp16))[name = tensor("op_47742_cast_fp16")]; + tensor var_47743_to_fp16 = const()[name = tensor("op_47743_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4073_cast_fp16 = mul(x = var_47742_cast_fp16, y = var_47743_to_fp16)[name = tensor("aw_4073_cast_fp16")]; + tensor var_47746_equation_0 = const()[name = tensor("op_47746_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47746_cast_fp16 = einsum(equation = var_47746_equation_0, values = (var_47588_cast_fp16, var_47505_cast_fp16))[name = tensor("op_47746_cast_fp16")]; + tensor var_47747_to_fp16 = const()[name = tensor("op_47747_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4075_cast_fp16 = mul(x = var_47746_cast_fp16, y = var_47747_to_fp16)[name = tensor("aw_4075_cast_fp16")]; + tensor var_47750_equation_0 = const()[name = tensor("op_47750_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47750_cast_fp16 = einsum(equation = var_47750_equation_0, values = (var_47592_cast_fp16, var_47509_cast_fp16))[name = tensor("op_47750_cast_fp16")]; + tensor var_47751_to_fp16 = const()[name = tensor("op_47751_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4077_cast_fp16 = mul(x = var_47750_cast_fp16, y = var_47751_to_fp16)[name = tensor("aw_4077_cast_fp16")]; + tensor var_47754_equation_0 = const()[name = tensor("op_47754_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47754_cast_fp16 = einsum(equation = var_47754_equation_0, values = (var_47596_cast_fp16, var_47513_cast_fp16))[name = tensor("op_47754_cast_fp16")]; + tensor var_47755_to_fp16 = const()[name = tensor("op_47755_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4079_cast_fp16 = mul(x = var_47754_cast_fp16, y = var_47755_to_fp16)[name = tensor("aw_4079_cast_fp16")]; + tensor var_47757_cast_fp16 = softmax(axis = var_30355, x = aw_4041_cast_fp16)[name = tensor("op_47757_cast_fp16")]; + tensor var_47758_cast_fp16 = softmax(axis = var_30355, x = aw_4043_cast_fp16)[name = tensor("op_47758_cast_fp16")]; + tensor var_47759_cast_fp16 = softmax(axis = var_30355, x = aw_4045_cast_fp16)[name = tensor("op_47759_cast_fp16")]; + tensor var_47760_cast_fp16 = softmax(axis = var_30355, x = aw_4047_cast_fp16)[name = tensor("op_47760_cast_fp16")]; + tensor var_47761_cast_fp16 = softmax(axis = var_30355, x = aw_4049_cast_fp16)[name = tensor("op_47761_cast_fp16")]; + tensor var_47762_cast_fp16 = softmax(axis = var_30355, x = aw_4051_cast_fp16)[name = tensor("op_47762_cast_fp16")]; + tensor var_47763_cast_fp16 = softmax(axis = var_30355, x = aw_4053_cast_fp16)[name = tensor("op_47763_cast_fp16")]; + tensor var_47764_cast_fp16 = softmax(axis = var_30355, x = aw_4055_cast_fp16)[name = tensor("op_47764_cast_fp16")]; + tensor var_47765_cast_fp16 = softmax(axis = var_30355, x = aw_4057_cast_fp16)[name = tensor("op_47765_cast_fp16")]; + tensor var_47766_cast_fp16 = softmax(axis = var_30355, x = aw_4059_cast_fp16)[name = tensor("op_47766_cast_fp16")]; + tensor var_47767_cast_fp16 = softmax(axis = var_30355, x = aw_4061_cast_fp16)[name = tensor("op_47767_cast_fp16")]; + tensor var_47768_cast_fp16 = softmax(axis = var_30355, x = aw_4063_cast_fp16)[name = tensor("op_47768_cast_fp16")]; + tensor var_47769_cast_fp16 = softmax(axis = var_30355, x = aw_4065_cast_fp16)[name = tensor("op_47769_cast_fp16")]; + tensor var_47770_cast_fp16 = softmax(axis = var_30355, x = aw_4067_cast_fp16)[name = tensor("op_47770_cast_fp16")]; + tensor var_47771_cast_fp16 = softmax(axis = var_30355, x = aw_4069_cast_fp16)[name = tensor("op_47771_cast_fp16")]; + tensor var_47772_cast_fp16 = softmax(axis = var_30355, x = aw_4071_cast_fp16)[name = tensor("op_47772_cast_fp16")]; + tensor var_47773_cast_fp16 = softmax(axis = var_30355, x = aw_4073_cast_fp16)[name = tensor("op_47773_cast_fp16")]; + tensor var_47774_cast_fp16 = softmax(axis = var_30355, x = aw_4075_cast_fp16)[name = tensor("op_47774_cast_fp16")]; + tensor var_47775_cast_fp16 = softmax(axis = var_30355, x = aw_4077_cast_fp16)[name = tensor("op_47775_cast_fp16")]; + tensor var_47776_cast_fp16 = softmax(axis = var_30355, x = aw_4079_cast_fp16)[name = tensor("op_47776_cast_fp16")]; + tensor var_47778_equation_0 = const()[name = tensor("op_47778_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47778_cast_fp16 = einsum(equation = var_47778_equation_0, values = (var_47598_cast_fp16, var_47757_cast_fp16))[name = tensor("op_47778_cast_fp16")]; + tensor var_47780_equation_0 = const()[name = tensor("op_47780_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47780_cast_fp16 = einsum(equation = var_47780_equation_0, values = (var_47602_cast_fp16, var_47758_cast_fp16))[name = tensor("op_47780_cast_fp16")]; + tensor var_47782_equation_0 = const()[name = tensor("op_47782_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47782_cast_fp16 = einsum(equation = var_47782_equation_0, values = (var_47606_cast_fp16, var_47759_cast_fp16))[name = tensor("op_47782_cast_fp16")]; + tensor var_47784_equation_0 = const()[name = tensor("op_47784_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47784_cast_fp16 = einsum(equation = var_47784_equation_0, values = (var_47610_cast_fp16, var_47760_cast_fp16))[name = tensor("op_47784_cast_fp16")]; + tensor var_47786_equation_0 = const()[name = tensor("op_47786_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47786_cast_fp16 = einsum(equation = var_47786_equation_0, values = (var_47614_cast_fp16, var_47761_cast_fp16))[name = tensor("op_47786_cast_fp16")]; + tensor var_47788_equation_0 = const()[name = tensor("op_47788_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47788_cast_fp16 = einsum(equation = var_47788_equation_0, values = (var_47618_cast_fp16, var_47762_cast_fp16))[name = tensor("op_47788_cast_fp16")]; + tensor var_47790_equation_0 = const()[name = tensor("op_47790_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47790_cast_fp16 = einsum(equation = var_47790_equation_0, values = (var_47622_cast_fp16, var_47763_cast_fp16))[name = tensor("op_47790_cast_fp16")]; + tensor var_47792_equation_0 = const()[name = tensor("op_47792_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47792_cast_fp16 = einsum(equation = var_47792_equation_0, values = (var_47626_cast_fp16, var_47764_cast_fp16))[name = tensor("op_47792_cast_fp16")]; + tensor var_47794_equation_0 = const()[name = tensor("op_47794_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47794_cast_fp16 = einsum(equation = var_47794_equation_0, values = (var_47630_cast_fp16, var_47765_cast_fp16))[name = tensor("op_47794_cast_fp16")]; + tensor var_47796_equation_0 = const()[name = tensor("op_47796_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47796_cast_fp16 = einsum(equation = var_47796_equation_0, values = (var_47634_cast_fp16, var_47766_cast_fp16))[name = tensor("op_47796_cast_fp16")]; + tensor var_47798_equation_0 = const()[name = tensor("op_47798_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47798_cast_fp16 = einsum(equation = var_47798_equation_0, values = (var_47638_cast_fp16, var_47767_cast_fp16))[name = tensor("op_47798_cast_fp16")]; + tensor var_47800_equation_0 = const()[name = tensor("op_47800_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47800_cast_fp16 = einsum(equation = var_47800_equation_0, values = (var_47642_cast_fp16, var_47768_cast_fp16))[name = tensor("op_47800_cast_fp16")]; + tensor var_47802_equation_0 = const()[name = tensor("op_47802_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47802_cast_fp16 = einsum(equation = var_47802_equation_0, values = (var_47646_cast_fp16, var_47769_cast_fp16))[name = tensor("op_47802_cast_fp16")]; + tensor var_47804_equation_0 = const()[name = tensor("op_47804_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47804_cast_fp16 = einsum(equation = var_47804_equation_0, values = (var_47650_cast_fp16, var_47770_cast_fp16))[name = tensor("op_47804_cast_fp16")]; + tensor var_47806_equation_0 = const()[name = tensor("op_47806_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47806_cast_fp16 = einsum(equation = var_47806_equation_0, values = (var_47654_cast_fp16, var_47771_cast_fp16))[name = tensor("op_47806_cast_fp16")]; + tensor var_47808_equation_0 = const()[name = tensor("op_47808_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47808_cast_fp16 = einsum(equation = var_47808_equation_0, values = (var_47658_cast_fp16, var_47772_cast_fp16))[name = tensor("op_47808_cast_fp16")]; + tensor var_47810_equation_0 = const()[name = tensor("op_47810_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47810_cast_fp16 = einsum(equation = var_47810_equation_0, values = (var_47662_cast_fp16, var_47773_cast_fp16))[name = tensor("op_47810_cast_fp16")]; + tensor var_47812_equation_0 = const()[name = tensor("op_47812_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47812_cast_fp16 = einsum(equation = var_47812_equation_0, values = (var_47666_cast_fp16, var_47774_cast_fp16))[name = tensor("op_47812_cast_fp16")]; + tensor var_47814_equation_0 = const()[name = tensor("op_47814_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47814_cast_fp16 = einsum(equation = var_47814_equation_0, values = (var_47670_cast_fp16, var_47775_cast_fp16))[name = tensor("op_47814_cast_fp16")]; + tensor var_47816_equation_0 = const()[name = tensor("op_47816_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47816_cast_fp16 = einsum(equation = var_47816_equation_0, values = (var_47674_cast_fp16, var_47776_cast_fp16))[name = tensor("op_47816_cast_fp16")]; + tensor input_603_interleave_0 = const()[name = tensor("input_603_interleave_0"), val = tensor(false)]; + tensor input_603_cast_fp16 = concat(axis = var_30355, interleave = input_603_interleave_0, values = (var_47778_cast_fp16, var_47780_cast_fp16, var_47782_cast_fp16, var_47784_cast_fp16, var_47786_cast_fp16, var_47788_cast_fp16, var_47790_cast_fp16, var_47792_cast_fp16, var_47794_cast_fp16, var_47796_cast_fp16, var_47798_cast_fp16, var_47800_cast_fp16, var_47802_cast_fp16, var_47804_cast_fp16, var_47806_cast_fp16, var_47808_cast_fp16, var_47810_cast_fp16, var_47812_cast_fp16, var_47814_cast_fp16, var_47816_cast_fp16))[name = tensor("input_603_cast_fp16")]; + tensor var_47822 = const()[name = tensor("op_47822"), val = tensor([1, 1])]; + tensor var_47824 = const()[name = tensor("op_47824"), val = tensor([1, 1])]; + tensor var_47826_pad_type_0 = const()[name = tensor("op_47826_pad_type_0"), val = tensor("custom")]; + tensor var_47826_pad_0 = const()[name = tensor("op_47826_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_8_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_8_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3971432576)))]; + tensor up_blocks_0_attentions_1_transformer_blocks_8_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_8_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3974709440)))]; + tensor var_47826_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_8_attn2_to_out_0_bias_to_fp16, dilations = var_47824, groups = var_30355, pad = var_47826_pad_0, pad_type = var_47826_pad_type_0, strides = var_47822, weight = up_blocks_0_attentions_1_transformer_blocks_8_attn2_to_out_0_weight_to_fp16, x = input_603_cast_fp16)[name = tensor("op_47826_cast_fp16")]; + tensor inputs_317_cast_fp16 = add(x = var_47826_cast_fp16, y = inputs_315_cast_fp16)[name = tensor("inputs_317_cast_fp16")]; + tensor input_605_axes_0 = const()[name = tensor("input_605_axes_0"), val = tensor([1])]; + tensor input_605_gamma_0_to_fp16 = const()[name = tensor("input_605_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3974712064)))]; + tensor input_605_beta_0_to_fp16 = const()[name = tensor("input_605_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3974714688)))]; + tensor var_47836_to_fp16 = const()[name = tensor("op_47836_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_605_cast_fp16 = layer_norm(axes = input_605_axes_0, beta = input_605_beta_0_to_fp16, epsilon = var_47836_to_fp16, gamma = input_605_gamma_0_to_fp16, x = inputs_317_cast_fp16)[name = tensor("input_605_cast_fp16")]; + tensor var_47852 = const()[name = tensor("op_47852"), val = tensor([1, 1])]; + tensor var_47854 = const()[name = tensor("op_47854"), val = tensor([1, 1])]; + tensor var_47856_pad_type_0 = const()[name = tensor("op_47856_pad_type_0"), val = tensor("custom")]; + tensor var_47856_pad_0 = const()[name = tensor("op_47856_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_8_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_8_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3974717312)))]; + tensor up_blocks_0_attentions_1_transformer_blocks_8_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_8_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4000931776)))]; + tensor var_47856_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_8_ff_net_0_proj_bias_to_fp16, dilations = var_47854, groups = var_30355, pad = var_47856_pad_0, pad_type = var_47856_pad_type_0, strides = var_47852, weight = up_blocks_0_attentions_1_transformer_blocks_8_ff_net_0_proj_weight_to_fp16, x = input_605_cast_fp16)[name = tensor("op_47856_cast_fp16")]; + tensor var_47857_split_sizes_0 = const()[name = tensor("op_47857_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_47857_axis_0 = const()[name = tensor("op_47857_axis_0"), val = tensor(1)]; + tensor var_47857_cast_fp16_0, tensor var_47857_cast_fp16_1 = split(axis = var_47857_axis_0, split_sizes = var_47857_split_sizes_0, x = var_47856_cast_fp16)[name = tensor("op_47857_cast_fp16")]; + tensor var_47859_mode_0 = const()[name = tensor("op_47859_mode_0"), val = tensor("EXACT")]; + tensor var_47859_cast_fp16 = gelu(mode = var_47859_mode_0, x = var_47857_cast_fp16_1)[name = tensor("op_47859_cast_fp16")]; + tensor input_607_cast_fp16 = mul(x = var_47857_cast_fp16_0, y = var_47859_cast_fp16)[name = tensor("input_607_cast_fp16")]; + tensor var_47863 = const()[name = tensor("op_47863"), val = tensor([1, 1])]; + tensor var_47865 = const()[name = tensor("op_47865"), val = tensor([1, 1])]; + tensor var_47867_pad_type_0 = const()[name = tensor("op_47867_pad_type_0"), val = tensor("custom")]; + tensor var_47867_pad_0 = const()[name = tensor("op_47867_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_8_ff_net_2_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_8_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4000952320)))]; + tensor up_blocks_0_attentions_1_transformer_blocks_8_ff_net_2_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_8_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4014059584)))]; + tensor var_47867_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_8_ff_net_2_bias_to_fp16, dilations = var_47865, groups = var_30355, pad = var_47867_pad_0, pad_type = var_47867_pad_type_0, strides = var_47863, weight = up_blocks_0_attentions_1_transformer_blocks_8_ff_net_2_weight_to_fp16, x = input_607_cast_fp16)[name = tensor("op_47867_cast_fp16")]; + tensor inputs_319_cast_fp16 = add(x = var_47867_cast_fp16, y = inputs_317_cast_fp16)[name = tensor("inputs_319_cast_fp16")]; + tensor hidden_states_415_axes_0 = const()[name = tensor("hidden_states_415_axes_0"), val = tensor([1])]; + tensor hidden_states_415_gamma_0_to_fp16 = const()[name = tensor("hidden_states_415_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4014062208)))]; + tensor hidden_states_415_beta_0_to_fp16 = const()[name = tensor("hidden_states_415_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4014064832)))]; + tensor var_47883_to_fp16 = const()[name = tensor("op_47883_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_415_cast_fp16 = layer_norm(axes = hidden_states_415_axes_0, beta = hidden_states_415_beta_0_to_fp16, epsilon = var_47883_to_fp16, gamma = hidden_states_415_gamma_0_to_fp16, x = inputs_319_cast_fp16)[name = tensor("hidden_states_415_cast_fp16")]; + tensor var_47898 = const()[name = tensor("op_47898"), val = tensor([1, 1])]; + tensor var_47900 = const()[name = tensor("op_47900"), val = tensor([1, 1])]; + tensor q_213_pad_type_0 = const()[name = tensor("q_213_pad_type_0"), val = tensor("custom")]; + tensor q_213_pad_0 = const()[name = tensor("q_213_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_9_attn1_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_9_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4014067456)))]; + tensor q_213_cast_fp16 = conv(dilations = var_47900, groups = var_30355, pad = q_213_pad_0, pad_type = q_213_pad_type_0, strides = var_47898, weight = up_blocks_0_attentions_1_transformer_blocks_9_attn1_to_q_weight_to_fp16, x = hidden_states_415_cast_fp16)[name = tensor("q_213_cast_fp16")]; + tensor var_47904 = const()[name = tensor("op_47904"), val = tensor([1, 1])]; + tensor var_47906 = const()[name = tensor("op_47906"), val = tensor([1, 1])]; + tensor k_425_pad_type_0 = const()[name = tensor("k_425_pad_type_0"), val = tensor("custom")]; + tensor k_425_pad_0 = const()[name = tensor("k_425_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_9_attn1_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_9_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4017344320)))]; + tensor k_425_cast_fp16 = conv(dilations = var_47906, groups = var_30355, pad = k_425_pad_0, pad_type = k_425_pad_type_0, strides = var_47904, weight = up_blocks_0_attentions_1_transformer_blocks_9_attn1_to_k_weight_to_fp16, x = hidden_states_415_cast_fp16)[name = tensor("k_425_cast_fp16")]; + tensor var_47910 = const()[name = tensor("op_47910"), val = tensor([1, 1])]; + tensor var_47912 = const()[name = tensor("op_47912"), val = tensor([1, 1])]; + tensor v_213_pad_type_0 = const()[name = tensor("v_213_pad_type_0"), val = tensor("custom")]; + tensor v_213_pad_0 = const()[name = tensor("v_213_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_9_attn1_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_9_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4020621184)))]; + tensor v_213_cast_fp16 = conv(dilations = var_47912, groups = var_30355, pad = v_213_pad_0, pad_type = v_213_pad_type_0, strides = var_47910, weight = up_blocks_0_attentions_1_transformer_blocks_9_attn1_to_v_weight_to_fp16, x = hidden_states_415_cast_fp16)[name = tensor("v_213_cast_fp16")]; + tensor var_47916_begin_0 = const()[name = tensor("op_47916_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_47916_end_0 = const()[name = tensor("op_47916_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_47916_end_mask_0 = const()[name = tensor("op_47916_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47916_cast_fp16 = slice_by_index(begin = var_47916_begin_0, end = var_47916_end_0, end_mask = var_47916_end_mask_0, x = q_213_cast_fp16)[name = tensor("op_47916_cast_fp16")]; + tensor var_47920_begin_0 = const()[name = tensor("op_47920_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_47920_end_0 = const()[name = tensor("op_47920_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_47920_end_mask_0 = const()[name = tensor("op_47920_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47920_cast_fp16 = slice_by_index(begin = var_47920_begin_0, end = var_47920_end_0, end_mask = var_47920_end_mask_0, x = q_213_cast_fp16)[name = tensor("op_47920_cast_fp16")]; + tensor var_47924_begin_0 = const()[name = tensor("op_47924_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_47924_end_0 = const()[name = tensor("op_47924_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_47924_end_mask_0 = const()[name = tensor("op_47924_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47924_cast_fp16 = slice_by_index(begin = var_47924_begin_0, end = var_47924_end_0, end_mask = var_47924_end_mask_0, x = q_213_cast_fp16)[name = tensor("op_47924_cast_fp16")]; + tensor var_47928_begin_0 = const()[name = tensor("op_47928_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_47928_end_0 = const()[name = tensor("op_47928_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_47928_end_mask_0 = const()[name = tensor("op_47928_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47928_cast_fp16 = slice_by_index(begin = var_47928_begin_0, end = var_47928_end_0, end_mask = var_47928_end_mask_0, x = q_213_cast_fp16)[name = tensor("op_47928_cast_fp16")]; + tensor var_47932_begin_0 = const()[name = tensor("op_47932_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_47932_end_0 = const()[name = tensor("op_47932_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_47932_end_mask_0 = const()[name = tensor("op_47932_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47932_cast_fp16 = slice_by_index(begin = var_47932_begin_0, end = var_47932_end_0, end_mask = var_47932_end_mask_0, x = q_213_cast_fp16)[name = tensor("op_47932_cast_fp16")]; + tensor var_47936_begin_0 = const()[name = tensor("op_47936_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_47936_end_0 = const()[name = tensor("op_47936_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_47936_end_mask_0 = const()[name = tensor("op_47936_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47936_cast_fp16 = slice_by_index(begin = var_47936_begin_0, end = var_47936_end_0, end_mask = var_47936_end_mask_0, x = q_213_cast_fp16)[name = tensor("op_47936_cast_fp16")]; + tensor var_47940_begin_0 = const()[name = tensor("op_47940_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_47940_end_0 = const()[name = tensor("op_47940_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_47940_end_mask_0 = const()[name = tensor("op_47940_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47940_cast_fp16 = slice_by_index(begin = var_47940_begin_0, end = var_47940_end_0, end_mask = var_47940_end_mask_0, x = q_213_cast_fp16)[name = tensor("op_47940_cast_fp16")]; + tensor var_47944_begin_0 = const()[name = tensor("op_47944_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_47944_end_0 = const()[name = tensor("op_47944_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_47944_end_mask_0 = const()[name = tensor("op_47944_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47944_cast_fp16 = slice_by_index(begin = var_47944_begin_0, end = var_47944_end_0, end_mask = var_47944_end_mask_0, x = q_213_cast_fp16)[name = tensor("op_47944_cast_fp16")]; + tensor var_47948_begin_0 = const()[name = tensor("op_47948_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_47948_end_0 = const()[name = tensor("op_47948_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_47948_end_mask_0 = const()[name = tensor("op_47948_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47948_cast_fp16 = slice_by_index(begin = var_47948_begin_0, end = var_47948_end_0, end_mask = var_47948_end_mask_0, x = q_213_cast_fp16)[name = tensor("op_47948_cast_fp16")]; + tensor var_47952_begin_0 = const()[name = tensor("op_47952_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_47952_end_0 = const()[name = tensor("op_47952_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_47952_end_mask_0 = const()[name = tensor("op_47952_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47952_cast_fp16 = slice_by_index(begin = var_47952_begin_0, end = var_47952_end_0, end_mask = var_47952_end_mask_0, x = q_213_cast_fp16)[name = tensor("op_47952_cast_fp16")]; + tensor var_47956_begin_0 = const()[name = tensor("op_47956_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_47956_end_0 = const()[name = tensor("op_47956_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_47956_end_mask_0 = const()[name = tensor("op_47956_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47956_cast_fp16 = slice_by_index(begin = var_47956_begin_0, end = var_47956_end_0, end_mask = var_47956_end_mask_0, x = q_213_cast_fp16)[name = tensor("op_47956_cast_fp16")]; + tensor var_47960_begin_0 = const()[name = tensor("op_47960_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_47960_end_0 = const()[name = tensor("op_47960_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_47960_end_mask_0 = const()[name = tensor("op_47960_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47960_cast_fp16 = slice_by_index(begin = var_47960_begin_0, end = var_47960_end_0, end_mask = var_47960_end_mask_0, x = q_213_cast_fp16)[name = tensor("op_47960_cast_fp16")]; + tensor var_47964_begin_0 = const()[name = tensor("op_47964_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_47964_end_0 = const()[name = tensor("op_47964_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_47964_end_mask_0 = const()[name = tensor("op_47964_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47964_cast_fp16 = slice_by_index(begin = var_47964_begin_0, end = var_47964_end_0, end_mask = var_47964_end_mask_0, x = q_213_cast_fp16)[name = tensor("op_47964_cast_fp16")]; + tensor var_47968_begin_0 = const()[name = tensor("op_47968_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_47968_end_0 = const()[name = tensor("op_47968_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_47968_end_mask_0 = const()[name = tensor("op_47968_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47968_cast_fp16 = slice_by_index(begin = var_47968_begin_0, end = var_47968_end_0, end_mask = var_47968_end_mask_0, x = q_213_cast_fp16)[name = tensor("op_47968_cast_fp16")]; + tensor var_47972_begin_0 = const()[name = tensor("op_47972_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_47972_end_0 = const()[name = tensor("op_47972_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_47972_end_mask_0 = const()[name = tensor("op_47972_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47972_cast_fp16 = slice_by_index(begin = var_47972_begin_0, end = var_47972_end_0, end_mask = var_47972_end_mask_0, x = q_213_cast_fp16)[name = tensor("op_47972_cast_fp16")]; + tensor var_47976_begin_0 = const()[name = tensor("op_47976_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_47976_end_0 = const()[name = tensor("op_47976_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_47976_end_mask_0 = const()[name = tensor("op_47976_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47976_cast_fp16 = slice_by_index(begin = var_47976_begin_0, end = var_47976_end_0, end_mask = var_47976_end_mask_0, x = q_213_cast_fp16)[name = tensor("op_47976_cast_fp16")]; + tensor var_47980_begin_0 = const()[name = tensor("op_47980_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_47980_end_0 = const()[name = tensor("op_47980_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_47980_end_mask_0 = const()[name = tensor("op_47980_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47980_cast_fp16 = slice_by_index(begin = var_47980_begin_0, end = var_47980_end_0, end_mask = var_47980_end_mask_0, x = q_213_cast_fp16)[name = tensor("op_47980_cast_fp16")]; + tensor var_47984_begin_0 = const()[name = tensor("op_47984_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_47984_end_0 = const()[name = tensor("op_47984_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_47984_end_mask_0 = const()[name = tensor("op_47984_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47984_cast_fp16 = slice_by_index(begin = var_47984_begin_0, end = var_47984_end_0, end_mask = var_47984_end_mask_0, x = q_213_cast_fp16)[name = tensor("op_47984_cast_fp16")]; + tensor var_47988_begin_0 = const()[name = tensor("op_47988_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_47988_end_0 = const()[name = tensor("op_47988_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_47988_end_mask_0 = const()[name = tensor("op_47988_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47988_cast_fp16 = slice_by_index(begin = var_47988_begin_0, end = var_47988_end_0, end_mask = var_47988_end_mask_0, x = q_213_cast_fp16)[name = tensor("op_47988_cast_fp16")]; + tensor var_47992_begin_0 = const()[name = tensor("op_47992_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_47992_end_0 = const()[name = tensor("op_47992_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_47992_end_mask_0 = const()[name = tensor("op_47992_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47992_cast_fp16 = slice_by_index(begin = var_47992_begin_0, end = var_47992_end_0, end_mask = var_47992_end_mask_0, x = q_213_cast_fp16)[name = tensor("op_47992_cast_fp16")]; + tensor k_427_perm_0 = const()[name = tensor("k_427_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_47999_begin_0 = const()[name = tensor("op_47999_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_47999_end_0 = const()[name = tensor("op_47999_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_47999_end_mask_0 = const()[name = tensor("op_47999_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_427_cast_fp16 = transpose(perm = k_427_perm_0, x = k_425_cast_fp16)[name = tensor("transpose_33")]; + tensor var_47999_cast_fp16 = slice_by_index(begin = var_47999_begin_0, end = var_47999_end_0, end_mask = var_47999_end_mask_0, x = k_427_cast_fp16)[name = tensor("op_47999_cast_fp16")]; + tensor var_48003_begin_0 = const()[name = tensor("op_48003_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_48003_end_0 = const()[name = tensor("op_48003_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_48003_end_mask_0 = const()[name = tensor("op_48003_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48003_cast_fp16 = slice_by_index(begin = var_48003_begin_0, end = var_48003_end_0, end_mask = var_48003_end_mask_0, x = k_427_cast_fp16)[name = tensor("op_48003_cast_fp16")]; + tensor var_48007_begin_0 = const()[name = tensor("op_48007_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_48007_end_0 = const()[name = tensor("op_48007_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_48007_end_mask_0 = const()[name = tensor("op_48007_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48007_cast_fp16 = slice_by_index(begin = var_48007_begin_0, end = var_48007_end_0, end_mask = var_48007_end_mask_0, x = k_427_cast_fp16)[name = tensor("op_48007_cast_fp16")]; + tensor var_48011_begin_0 = const()[name = tensor("op_48011_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_48011_end_0 = const()[name = tensor("op_48011_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_48011_end_mask_0 = const()[name = tensor("op_48011_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48011_cast_fp16 = slice_by_index(begin = var_48011_begin_0, end = var_48011_end_0, end_mask = var_48011_end_mask_0, x = k_427_cast_fp16)[name = tensor("op_48011_cast_fp16")]; + tensor var_48015_begin_0 = const()[name = tensor("op_48015_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_48015_end_0 = const()[name = tensor("op_48015_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_48015_end_mask_0 = const()[name = tensor("op_48015_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48015_cast_fp16 = slice_by_index(begin = var_48015_begin_0, end = var_48015_end_0, end_mask = var_48015_end_mask_0, x = k_427_cast_fp16)[name = tensor("op_48015_cast_fp16")]; + tensor var_48019_begin_0 = const()[name = tensor("op_48019_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_48019_end_0 = const()[name = tensor("op_48019_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_48019_end_mask_0 = const()[name = tensor("op_48019_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48019_cast_fp16 = slice_by_index(begin = var_48019_begin_0, end = var_48019_end_0, end_mask = var_48019_end_mask_0, x = k_427_cast_fp16)[name = tensor("op_48019_cast_fp16")]; + tensor var_48023_begin_0 = const()[name = tensor("op_48023_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_48023_end_0 = const()[name = tensor("op_48023_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_48023_end_mask_0 = const()[name = tensor("op_48023_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48023_cast_fp16 = slice_by_index(begin = var_48023_begin_0, end = var_48023_end_0, end_mask = var_48023_end_mask_0, x = k_427_cast_fp16)[name = tensor("op_48023_cast_fp16")]; + tensor var_48027_begin_0 = const()[name = tensor("op_48027_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_48027_end_0 = const()[name = tensor("op_48027_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_48027_end_mask_0 = const()[name = tensor("op_48027_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48027_cast_fp16 = slice_by_index(begin = var_48027_begin_0, end = var_48027_end_0, end_mask = var_48027_end_mask_0, x = k_427_cast_fp16)[name = tensor("op_48027_cast_fp16")]; + tensor var_48031_begin_0 = const()[name = tensor("op_48031_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_48031_end_0 = const()[name = tensor("op_48031_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_48031_end_mask_0 = const()[name = tensor("op_48031_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48031_cast_fp16 = slice_by_index(begin = var_48031_begin_0, end = var_48031_end_0, end_mask = var_48031_end_mask_0, x = k_427_cast_fp16)[name = tensor("op_48031_cast_fp16")]; + tensor var_48035_begin_0 = const()[name = tensor("op_48035_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_48035_end_0 = const()[name = tensor("op_48035_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_48035_end_mask_0 = const()[name = tensor("op_48035_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48035_cast_fp16 = slice_by_index(begin = var_48035_begin_0, end = var_48035_end_0, end_mask = var_48035_end_mask_0, x = k_427_cast_fp16)[name = tensor("op_48035_cast_fp16")]; + tensor var_48039_begin_0 = const()[name = tensor("op_48039_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_48039_end_0 = const()[name = tensor("op_48039_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_48039_end_mask_0 = const()[name = tensor("op_48039_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48039_cast_fp16 = slice_by_index(begin = var_48039_begin_0, end = var_48039_end_0, end_mask = var_48039_end_mask_0, x = k_427_cast_fp16)[name = tensor("op_48039_cast_fp16")]; + tensor var_48043_begin_0 = const()[name = tensor("op_48043_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_48043_end_0 = const()[name = tensor("op_48043_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_48043_end_mask_0 = const()[name = tensor("op_48043_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48043_cast_fp16 = slice_by_index(begin = var_48043_begin_0, end = var_48043_end_0, end_mask = var_48043_end_mask_0, x = k_427_cast_fp16)[name = tensor("op_48043_cast_fp16")]; + tensor var_48047_begin_0 = const()[name = tensor("op_48047_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_48047_end_0 = const()[name = tensor("op_48047_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_48047_end_mask_0 = const()[name = tensor("op_48047_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48047_cast_fp16 = slice_by_index(begin = var_48047_begin_0, end = var_48047_end_0, end_mask = var_48047_end_mask_0, x = k_427_cast_fp16)[name = tensor("op_48047_cast_fp16")]; + tensor var_48051_begin_0 = const()[name = tensor("op_48051_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_48051_end_0 = const()[name = tensor("op_48051_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_48051_end_mask_0 = const()[name = tensor("op_48051_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48051_cast_fp16 = slice_by_index(begin = var_48051_begin_0, end = var_48051_end_0, end_mask = var_48051_end_mask_0, x = k_427_cast_fp16)[name = tensor("op_48051_cast_fp16")]; + tensor var_48055_begin_0 = const()[name = tensor("op_48055_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_48055_end_0 = const()[name = tensor("op_48055_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_48055_end_mask_0 = const()[name = tensor("op_48055_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48055_cast_fp16 = slice_by_index(begin = var_48055_begin_0, end = var_48055_end_0, end_mask = var_48055_end_mask_0, x = k_427_cast_fp16)[name = tensor("op_48055_cast_fp16")]; + tensor var_48059_begin_0 = const()[name = tensor("op_48059_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_48059_end_0 = const()[name = tensor("op_48059_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_48059_end_mask_0 = const()[name = tensor("op_48059_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48059_cast_fp16 = slice_by_index(begin = var_48059_begin_0, end = var_48059_end_0, end_mask = var_48059_end_mask_0, x = k_427_cast_fp16)[name = tensor("op_48059_cast_fp16")]; + tensor var_48063_begin_0 = const()[name = tensor("op_48063_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_48063_end_0 = const()[name = tensor("op_48063_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_48063_end_mask_0 = const()[name = tensor("op_48063_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48063_cast_fp16 = slice_by_index(begin = var_48063_begin_0, end = var_48063_end_0, end_mask = var_48063_end_mask_0, x = k_427_cast_fp16)[name = tensor("op_48063_cast_fp16")]; + tensor var_48067_begin_0 = const()[name = tensor("op_48067_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_48067_end_0 = const()[name = tensor("op_48067_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_48067_end_mask_0 = const()[name = tensor("op_48067_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48067_cast_fp16 = slice_by_index(begin = var_48067_begin_0, end = var_48067_end_0, end_mask = var_48067_end_mask_0, x = k_427_cast_fp16)[name = tensor("op_48067_cast_fp16")]; + tensor var_48071_begin_0 = const()[name = tensor("op_48071_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_48071_end_0 = const()[name = tensor("op_48071_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_48071_end_mask_0 = const()[name = tensor("op_48071_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48071_cast_fp16 = slice_by_index(begin = var_48071_begin_0, end = var_48071_end_0, end_mask = var_48071_end_mask_0, x = k_427_cast_fp16)[name = tensor("op_48071_cast_fp16")]; + tensor var_48075_begin_0 = const()[name = tensor("op_48075_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_48075_end_0 = const()[name = tensor("op_48075_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_48075_end_mask_0 = const()[name = tensor("op_48075_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48075_cast_fp16 = slice_by_index(begin = var_48075_begin_0, end = var_48075_end_0, end_mask = var_48075_end_mask_0, x = k_427_cast_fp16)[name = tensor("op_48075_cast_fp16")]; + tensor var_48077_begin_0 = const()[name = tensor("op_48077_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_48077_end_0 = const()[name = tensor("op_48077_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_48077_end_mask_0 = const()[name = tensor("op_48077_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48077_cast_fp16 = slice_by_index(begin = var_48077_begin_0, end = var_48077_end_0, end_mask = var_48077_end_mask_0, x = v_213_cast_fp16)[name = tensor("op_48077_cast_fp16")]; + tensor var_48081_begin_0 = const()[name = tensor("op_48081_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_48081_end_0 = const()[name = tensor("op_48081_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_48081_end_mask_0 = const()[name = tensor("op_48081_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48081_cast_fp16 = slice_by_index(begin = var_48081_begin_0, end = var_48081_end_0, end_mask = var_48081_end_mask_0, x = v_213_cast_fp16)[name = tensor("op_48081_cast_fp16")]; + tensor var_48085_begin_0 = const()[name = tensor("op_48085_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_48085_end_0 = const()[name = tensor("op_48085_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_48085_end_mask_0 = const()[name = tensor("op_48085_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48085_cast_fp16 = slice_by_index(begin = var_48085_begin_0, end = var_48085_end_0, end_mask = var_48085_end_mask_0, x = v_213_cast_fp16)[name = tensor("op_48085_cast_fp16")]; + tensor var_48089_begin_0 = const()[name = tensor("op_48089_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_48089_end_0 = const()[name = tensor("op_48089_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_48089_end_mask_0 = const()[name = tensor("op_48089_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48089_cast_fp16 = slice_by_index(begin = var_48089_begin_0, end = var_48089_end_0, end_mask = var_48089_end_mask_0, x = v_213_cast_fp16)[name = tensor("op_48089_cast_fp16")]; + tensor var_48093_begin_0 = const()[name = tensor("op_48093_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_48093_end_0 = const()[name = tensor("op_48093_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_48093_end_mask_0 = const()[name = tensor("op_48093_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48093_cast_fp16 = slice_by_index(begin = var_48093_begin_0, end = var_48093_end_0, end_mask = var_48093_end_mask_0, x = v_213_cast_fp16)[name = tensor("op_48093_cast_fp16")]; + tensor var_48097_begin_0 = const()[name = tensor("op_48097_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_48097_end_0 = const()[name = tensor("op_48097_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_48097_end_mask_0 = const()[name = tensor("op_48097_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48097_cast_fp16 = slice_by_index(begin = var_48097_begin_0, end = var_48097_end_0, end_mask = var_48097_end_mask_0, x = v_213_cast_fp16)[name = tensor("op_48097_cast_fp16")]; + tensor var_48101_begin_0 = const()[name = tensor("op_48101_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_48101_end_0 = const()[name = tensor("op_48101_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_48101_end_mask_0 = const()[name = tensor("op_48101_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48101_cast_fp16 = slice_by_index(begin = var_48101_begin_0, end = var_48101_end_0, end_mask = var_48101_end_mask_0, x = v_213_cast_fp16)[name = tensor("op_48101_cast_fp16")]; + tensor var_48105_begin_0 = const()[name = tensor("op_48105_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_48105_end_0 = const()[name = tensor("op_48105_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_48105_end_mask_0 = const()[name = tensor("op_48105_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48105_cast_fp16 = slice_by_index(begin = var_48105_begin_0, end = var_48105_end_0, end_mask = var_48105_end_mask_0, x = v_213_cast_fp16)[name = tensor("op_48105_cast_fp16")]; + tensor var_48109_begin_0 = const()[name = tensor("op_48109_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_48109_end_0 = const()[name = tensor("op_48109_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_48109_end_mask_0 = const()[name = tensor("op_48109_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48109_cast_fp16 = slice_by_index(begin = var_48109_begin_0, end = var_48109_end_0, end_mask = var_48109_end_mask_0, x = v_213_cast_fp16)[name = tensor("op_48109_cast_fp16")]; + tensor var_48113_begin_0 = const()[name = tensor("op_48113_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_48113_end_0 = const()[name = tensor("op_48113_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_48113_end_mask_0 = const()[name = tensor("op_48113_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48113_cast_fp16 = slice_by_index(begin = var_48113_begin_0, end = var_48113_end_0, end_mask = var_48113_end_mask_0, x = v_213_cast_fp16)[name = tensor("op_48113_cast_fp16")]; + tensor var_48117_begin_0 = const()[name = tensor("op_48117_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_48117_end_0 = const()[name = tensor("op_48117_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_48117_end_mask_0 = const()[name = tensor("op_48117_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48117_cast_fp16 = slice_by_index(begin = var_48117_begin_0, end = var_48117_end_0, end_mask = var_48117_end_mask_0, x = v_213_cast_fp16)[name = tensor("op_48117_cast_fp16")]; + tensor var_48121_begin_0 = const()[name = tensor("op_48121_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_48121_end_0 = const()[name = tensor("op_48121_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_48121_end_mask_0 = const()[name = tensor("op_48121_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48121_cast_fp16 = slice_by_index(begin = var_48121_begin_0, end = var_48121_end_0, end_mask = var_48121_end_mask_0, x = v_213_cast_fp16)[name = tensor("op_48121_cast_fp16")]; + tensor var_48125_begin_0 = const()[name = tensor("op_48125_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_48125_end_0 = const()[name = tensor("op_48125_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_48125_end_mask_0 = const()[name = tensor("op_48125_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48125_cast_fp16 = slice_by_index(begin = var_48125_begin_0, end = var_48125_end_0, end_mask = var_48125_end_mask_0, x = v_213_cast_fp16)[name = tensor("op_48125_cast_fp16")]; + tensor var_48129_begin_0 = const()[name = tensor("op_48129_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_48129_end_0 = const()[name = tensor("op_48129_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_48129_end_mask_0 = const()[name = tensor("op_48129_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48129_cast_fp16 = slice_by_index(begin = var_48129_begin_0, end = var_48129_end_0, end_mask = var_48129_end_mask_0, x = v_213_cast_fp16)[name = tensor("op_48129_cast_fp16")]; + tensor var_48133_begin_0 = const()[name = tensor("op_48133_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_48133_end_0 = const()[name = tensor("op_48133_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_48133_end_mask_0 = const()[name = tensor("op_48133_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48133_cast_fp16 = slice_by_index(begin = var_48133_begin_0, end = var_48133_end_0, end_mask = var_48133_end_mask_0, x = v_213_cast_fp16)[name = tensor("op_48133_cast_fp16")]; + tensor var_48137_begin_0 = const()[name = tensor("op_48137_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_48137_end_0 = const()[name = tensor("op_48137_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_48137_end_mask_0 = const()[name = tensor("op_48137_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48137_cast_fp16 = slice_by_index(begin = var_48137_begin_0, end = var_48137_end_0, end_mask = var_48137_end_mask_0, x = v_213_cast_fp16)[name = tensor("op_48137_cast_fp16")]; + tensor var_48141_begin_0 = const()[name = tensor("op_48141_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_48141_end_0 = const()[name = tensor("op_48141_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_48141_end_mask_0 = const()[name = tensor("op_48141_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48141_cast_fp16 = slice_by_index(begin = var_48141_begin_0, end = var_48141_end_0, end_mask = var_48141_end_mask_0, x = v_213_cast_fp16)[name = tensor("op_48141_cast_fp16")]; + tensor var_48145_begin_0 = const()[name = tensor("op_48145_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_48145_end_0 = const()[name = tensor("op_48145_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_48145_end_mask_0 = const()[name = tensor("op_48145_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48145_cast_fp16 = slice_by_index(begin = var_48145_begin_0, end = var_48145_end_0, end_mask = var_48145_end_mask_0, x = v_213_cast_fp16)[name = tensor("op_48145_cast_fp16")]; + tensor var_48149_begin_0 = const()[name = tensor("op_48149_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_48149_end_0 = const()[name = tensor("op_48149_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_48149_end_mask_0 = const()[name = tensor("op_48149_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48149_cast_fp16 = slice_by_index(begin = var_48149_begin_0, end = var_48149_end_0, end_mask = var_48149_end_mask_0, x = v_213_cast_fp16)[name = tensor("op_48149_cast_fp16")]; + tensor var_48153_begin_0 = const()[name = tensor("op_48153_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_48153_end_0 = const()[name = tensor("op_48153_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_48153_end_mask_0 = const()[name = tensor("op_48153_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48153_cast_fp16 = slice_by_index(begin = var_48153_begin_0, end = var_48153_end_0, end_mask = var_48153_end_mask_0, x = v_213_cast_fp16)[name = tensor("op_48153_cast_fp16")]; + tensor var_48157_equation_0 = const()[name = tensor("op_48157_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48157_cast_fp16 = einsum(equation = var_48157_equation_0, values = (var_47999_cast_fp16, var_47916_cast_fp16))[name = tensor("op_48157_cast_fp16")]; + tensor var_48158_to_fp16 = const()[name = tensor("op_48158_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4081_cast_fp16 = mul(x = var_48157_cast_fp16, y = var_48158_to_fp16)[name = tensor("aw_4081_cast_fp16")]; + tensor var_48161_equation_0 = const()[name = tensor("op_48161_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48161_cast_fp16 = einsum(equation = var_48161_equation_0, values = (var_48003_cast_fp16, var_47920_cast_fp16))[name = tensor("op_48161_cast_fp16")]; + tensor var_48162_to_fp16 = const()[name = tensor("op_48162_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4083_cast_fp16 = mul(x = var_48161_cast_fp16, y = var_48162_to_fp16)[name = tensor("aw_4083_cast_fp16")]; + tensor var_48165_equation_0 = const()[name = tensor("op_48165_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48165_cast_fp16 = einsum(equation = var_48165_equation_0, values = (var_48007_cast_fp16, var_47924_cast_fp16))[name = tensor("op_48165_cast_fp16")]; + tensor var_48166_to_fp16 = const()[name = tensor("op_48166_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4085_cast_fp16 = mul(x = var_48165_cast_fp16, y = var_48166_to_fp16)[name = tensor("aw_4085_cast_fp16")]; + tensor var_48169_equation_0 = const()[name = tensor("op_48169_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48169_cast_fp16 = einsum(equation = var_48169_equation_0, values = (var_48011_cast_fp16, var_47928_cast_fp16))[name = tensor("op_48169_cast_fp16")]; + tensor var_48170_to_fp16 = const()[name = tensor("op_48170_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4087_cast_fp16 = mul(x = var_48169_cast_fp16, y = var_48170_to_fp16)[name = tensor("aw_4087_cast_fp16")]; + tensor var_48173_equation_0 = const()[name = tensor("op_48173_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48173_cast_fp16 = einsum(equation = var_48173_equation_0, values = (var_48015_cast_fp16, var_47932_cast_fp16))[name = tensor("op_48173_cast_fp16")]; + tensor var_48174_to_fp16 = const()[name = tensor("op_48174_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4089_cast_fp16 = mul(x = var_48173_cast_fp16, y = var_48174_to_fp16)[name = tensor("aw_4089_cast_fp16")]; + tensor var_48177_equation_0 = const()[name = tensor("op_48177_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48177_cast_fp16 = einsum(equation = var_48177_equation_0, values = (var_48019_cast_fp16, var_47936_cast_fp16))[name = tensor("op_48177_cast_fp16")]; + tensor var_48178_to_fp16 = const()[name = tensor("op_48178_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4091_cast_fp16 = mul(x = var_48177_cast_fp16, y = var_48178_to_fp16)[name = tensor("aw_4091_cast_fp16")]; + tensor var_48181_equation_0 = const()[name = tensor("op_48181_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48181_cast_fp16 = einsum(equation = var_48181_equation_0, values = (var_48023_cast_fp16, var_47940_cast_fp16))[name = tensor("op_48181_cast_fp16")]; + tensor var_48182_to_fp16 = const()[name = tensor("op_48182_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4093_cast_fp16 = mul(x = var_48181_cast_fp16, y = var_48182_to_fp16)[name = tensor("aw_4093_cast_fp16")]; + tensor var_48185_equation_0 = const()[name = tensor("op_48185_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48185_cast_fp16 = einsum(equation = var_48185_equation_0, values = (var_48027_cast_fp16, var_47944_cast_fp16))[name = tensor("op_48185_cast_fp16")]; + tensor var_48186_to_fp16 = const()[name = tensor("op_48186_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4095_cast_fp16 = mul(x = var_48185_cast_fp16, y = var_48186_to_fp16)[name = tensor("aw_4095_cast_fp16")]; + tensor var_48189_equation_0 = const()[name = tensor("op_48189_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48189_cast_fp16 = einsum(equation = var_48189_equation_0, values = (var_48031_cast_fp16, var_47948_cast_fp16))[name = tensor("op_48189_cast_fp16")]; + tensor var_48190_to_fp16 = const()[name = tensor("op_48190_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4097_cast_fp16 = mul(x = var_48189_cast_fp16, y = var_48190_to_fp16)[name = tensor("aw_4097_cast_fp16")]; + tensor var_48193_equation_0 = const()[name = tensor("op_48193_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48193_cast_fp16 = einsum(equation = var_48193_equation_0, values = (var_48035_cast_fp16, var_47952_cast_fp16))[name = tensor("op_48193_cast_fp16")]; + tensor var_48194_to_fp16 = const()[name = tensor("op_48194_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4099_cast_fp16 = mul(x = var_48193_cast_fp16, y = var_48194_to_fp16)[name = tensor("aw_4099_cast_fp16")]; + tensor var_48197_equation_0 = const()[name = tensor("op_48197_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48197_cast_fp16 = einsum(equation = var_48197_equation_0, values = (var_48039_cast_fp16, var_47956_cast_fp16))[name = tensor("op_48197_cast_fp16")]; + tensor var_48198_to_fp16 = const()[name = tensor("op_48198_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4101_cast_fp16 = mul(x = var_48197_cast_fp16, y = var_48198_to_fp16)[name = tensor("aw_4101_cast_fp16")]; + tensor var_48201_equation_0 = const()[name = tensor("op_48201_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48201_cast_fp16 = einsum(equation = var_48201_equation_0, values = (var_48043_cast_fp16, var_47960_cast_fp16))[name = tensor("op_48201_cast_fp16")]; + tensor var_48202_to_fp16 = const()[name = tensor("op_48202_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4103_cast_fp16 = mul(x = var_48201_cast_fp16, y = var_48202_to_fp16)[name = tensor("aw_4103_cast_fp16")]; + tensor var_48205_equation_0 = const()[name = tensor("op_48205_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48205_cast_fp16 = einsum(equation = var_48205_equation_0, values = (var_48047_cast_fp16, var_47964_cast_fp16))[name = tensor("op_48205_cast_fp16")]; + tensor var_48206_to_fp16 = const()[name = tensor("op_48206_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4105_cast_fp16 = mul(x = var_48205_cast_fp16, y = var_48206_to_fp16)[name = tensor("aw_4105_cast_fp16")]; + tensor var_48209_equation_0 = const()[name = tensor("op_48209_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48209_cast_fp16 = einsum(equation = var_48209_equation_0, values = (var_48051_cast_fp16, var_47968_cast_fp16))[name = tensor("op_48209_cast_fp16")]; + tensor var_48210_to_fp16 = const()[name = tensor("op_48210_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4107_cast_fp16 = mul(x = var_48209_cast_fp16, y = var_48210_to_fp16)[name = tensor("aw_4107_cast_fp16")]; + tensor var_48213_equation_0 = const()[name = tensor("op_48213_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48213_cast_fp16 = einsum(equation = var_48213_equation_0, values = (var_48055_cast_fp16, var_47972_cast_fp16))[name = tensor("op_48213_cast_fp16")]; + tensor var_48214_to_fp16 = const()[name = tensor("op_48214_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4109_cast_fp16 = mul(x = var_48213_cast_fp16, y = var_48214_to_fp16)[name = tensor("aw_4109_cast_fp16")]; + tensor var_48217_equation_0 = const()[name = tensor("op_48217_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48217_cast_fp16 = einsum(equation = var_48217_equation_0, values = (var_48059_cast_fp16, var_47976_cast_fp16))[name = tensor("op_48217_cast_fp16")]; + tensor var_48218_to_fp16 = const()[name = tensor("op_48218_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4111_cast_fp16 = mul(x = var_48217_cast_fp16, y = var_48218_to_fp16)[name = tensor("aw_4111_cast_fp16")]; + tensor var_48221_equation_0 = const()[name = tensor("op_48221_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48221_cast_fp16 = einsum(equation = var_48221_equation_0, values = (var_48063_cast_fp16, var_47980_cast_fp16))[name = tensor("op_48221_cast_fp16")]; + tensor var_48222_to_fp16 = const()[name = tensor("op_48222_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4113_cast_fp16 = mul(x = var_48221_cast_fp16, y = var_48222_to_fp16)[name = tensor("aw_4113_cast_fp16")]; + tensor var_48225_equation_0 = const()[name = tensor("op_48225_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48225_cast_fp16 = einsum(equation = var_48225_equation_0, values = (var_48067_cast_fp16, var_47984_cast_fp16))[name = tensor("op_48225_cast_fp16")]; + tensor var_48226_to_fp16 = const()[name = tensor("op_48226_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4115_cast_fp16 = mul(x = var_48225_cast_fp16, y = var_48226_to_fp16)[name = tensor("aw_4115_cast_fp16")]; + tensor var_48229_equation_0 = const()[name = tensor("op_48229_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48229_cast_fp16 = einsum(equation = var_48229_equation_0, values = (var_48071_cast_fp16, var_47988_cast_fp16))[name = tensor("op_48229_cast_fp16")]; + tensor var_48230_to_fp16 = const()[name = tensor("op_48230_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4117_cast_fp16 = mul(x = var_48229_cast_fp16, y = var_48230_to_fp16)[name = tensor("aw_4117_cast_fp16")]; + tensor var_48233_equation_0 = const()[name = tensor("op_48233_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48233_cast_fp16 = einsum(equation = var_48233_equation_0, values = (var_48075_cast_fp16, var_47992_cast_fp16))[name = tensor("op_48233_cast_fp16")]; + tensor var_48234_to_fp16 = const()[name = tensor("op_48234_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4119_cast_fp16 = mul(x = var_48233_cast_fp16, y = var_48234_to_fp16)[name = tensor("aw_4119_cast_fp16")]; + tensor var_48236_cast_fp16 = softmax(axis = var_30355, x = aw_4081_cast_fp16)[name = tensor("op_48236_cast_fp16")]; + tensor var_48237_cast_fp16 = softmax(axis = var_30355, x = aw_4083_cast_fp16)[name = tensor("op_48237_cast_fp16")]; + tensor var_48238_cast_fp16 = softmax(axis = var_30355, x = aw_4085_cast_fp16)[name = tensor("op_48238_cast_fp16")]; + tensor var_48239_cast_fp16 = softmax(axis = var_30355, x = aw_4087_cast_fp16)[name = tensor("op_48239_cast_fp16")]; + tensor var_48240_cast_fp16 = softmax(axis = var_30355, x = aw_4089_cast_fp16)[name = tensor("op_48240_cast_fp16")]; + tensor var_48241_cast_fp16 = softmax(axis = var_30355, x = aw_4091_cast_fp16)[name = tensor("op_48241_cast_fp16")]; + tensor var_48242_cast_fp16 = softmax(axis = var_30355, x = aw_4093_cast_fp16)[name = tensor("op_48242_cast_fp16")]; + tensor var_48243_cast_fp16 = softmax(axis = var_30355, x = aw_4095_cast_fp16)[name = tensor("op_48243_cast_fp16")]; + tensor var_48244_cast_fp16 = softmax(axis = var_30355, x = aw_4097_cast_fp16)[name = tensor("op_48244_cast_fp16")]; + tensor var_48245_cast_fp16 = softmax(axis = var_30355, x = aw_4099_cast_fp16)[name = tensor("op_48245_cast_fp16")]; + tensor var_48246_cast_fp16 = softmax(axis = var_30355, x = aw_4101_cast_fp16)[name = tensor("op_48246_cast_fp16")]; + tensor var_48247_cast_fp16 = softmax(axis = var_30355, x = aw_4103_cast_fp16)[name = tensor("op_48247_cast_fp16")]; + tensor var_48248_cast_fp16 = softmax(axis = var_30355, x = aw_4105_cast_fp16)[name = tensor("op_48248_cast_fp16")]; + tensor var_48249_cast_fp16 = softmax(axis = var_30355, x = aw_4107_cast_fp16)[name = tensor("op_48249_cast_fp16")]; + tensor var_48250_cast_fp16 = softmax(axis = var_30355, x = aw_4109_cast_fp16)[name = tensor("op_48250_cast_fp16")]; + tensor var_48251_cast_fp16 = softmax(axis = var_30355, x = aw_4111_cast_fp16)[name = tensor("op_48251_cast_fp16")]; + tensor var_48252_cast_fp16 = softmax(axis = var_30355, x = aw_4113_cast_fp16)[name = tensor("op_48252_cast_fp16")]; + tensor var_48253_cast_fp16 = softmax(axis = var_30355, x = aw_4115_cast_fp16)[name = tensor("op_48253_cast_fp16")]; + tensor var_48254_cast_fp16 = softmax(axis = var_30355, x = aw_4117_cast_fp16)[name = tensor("op_48254_cast_fp16")]; + tensor var_48255_cast_fp16 = softmax(axis = var_30355, x = aw_4119_cast_fp16)[name = tensor("op_48255_cast_fp16")]; + tensor var_48257_equation_0 = const()[name = tensor("op_48257_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48257_cast_fp16 = einsum(equation = var_48257_equation_0, values = (var_48077_cast_fp16, var_48236_cast_fp16))[name = tensor("op_48257_cast_fp16")]; + tensor var_48259_equation_0 = const()[name = tensor("op_48259_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48259_cast_fp16 = einsum(equation = var_48259_equation_0, values = (var_48081_cast_fp16, var_48237_cast_fp16))[name = tensor("op_48259_cast_fp16")]; + tensor var_48261_equation_0 = const()[name = tensor("op_48261_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48261_cast_fp16 = einsum(equation = var_48261_equation_0, values = (var_48085_cast_fp16, var_48238_cast_fp16))[name = tensor("op_48261_cast_fp16")]; + tensor var_48263_equation_0 = const()[name = tensor("op_48263_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48263_cast_fp16 = einsum(equation = var_48263_equation_0, values = (var_48089_cast_fp16, var_48239_cast_fp16))[name = tensor("op_48263_cast_fp16")]; + tensor var_48265_equation_0 = const()[name = tensor("op_48265_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48265_cast_fp16 = einsum(equation = var_48265_equation_0, values = (var_48093_cast_fp16, var_48240_cast_fp16))[name = tensor("op_48265_cast_fp16")]; + tensor var_48267_equation_0 = const()[name = tensor("op_48267_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48267_cast_fp16 = einsum(equation = var_48267_equation_0, values = (var_48097_cast_fp16, var_48241_cast_fp16))[name = tensor("op_48267_cast_fp16")]; + tensor var_48269_equation_0 = const()[name = tensor("op_48269_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48269_cast_fp16 = einsum(equation = var_48269_equation_0, values = (var_48101_cast_fp16, var_48242_cast_fp16))[name = tensor("op_48269_cast_fp16")]; + tensor var_48271_equation_0 = const()[name = tensor("op_48271_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48271_cast_fp16 = einsum(equation = var_48271_equation_0, values = (var_48105_cast_fp16, var_48243_cast_fp16))[name = tensor("op_48271_cast_fp16")]; + tensor var_48273_equation_0 = const()[name = tensor("op_48273_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48273_cast_fp16 = einsum(equation = var_48273_equation_0, values = (var_48109_cast_fp16, var_48244_cast_fp16))[name = tensor("op_48273_cast_fp16")]; + tensor var_48275_equation_0 = const()[name = tensor("op_48275_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48275_cast_fp16 = einsum(equation = var_48275_equation_0, values = (var_48113_cast_fp16, var_48245_cast_fp16))[name = tensor("op_48275_cast_fp16")]; + tensor var_48277_equation_0 = const()[name = tensor("op_48277_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48277_cast_fp16 = einsum(equation = var_48277_equation_0, values = (var_48117_cast_fp16, var_48246_cast_fp16))[name = tensor("op_48277_cast_fp16")]; + tensor var_48279_equation_0 = const()[name = tensor("op_48279_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48279_cast_fp16 = einsum(equation = var_48279_equation_0, values = (var_48121_cast_fp16, var_48247_cast_fp16))[name = tensor("op_48279_cast_fp16")]; + tensor var_48281_equation_0 = const()[name = tensor("op_48281_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48281_cast_fp16 = einsum(equation = var_48281_equation_0, values = (var_48125_cast_fp16, var_48248_cast_fp16))[name = tensor("op_48281_cast_fp16")]; + tensor var_48283_equation_0 = const()[name = tensor("op_48283_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48283_cast_fp16 = einsum(equation = var_48283_equation_0, values = (var_48129_cast_fp16, var_48249_cast_fp16))[name = tensor("op_48283_cast_fp16")]; + tensor var_48285_equation_0 = const()[name = tensor("op_48285_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48285_cast_fp16 = einsum(equation = var_48285_equation_0, values = (var_48133_cast_fp16, var_48250_cast_fp16))[name = tensor("op_48285_cast_fp16")]; + tensor var_48287_equation_0 = const()[name = tensor("op_48287_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48287_cast_fp16 = einsum(equation = var_48287_equation_0, values = (var_48137_cast_fp16, var_48251_cast_fp16))[name = tensor("op_48287_cast_fp16")]; + tensor var_48289_equation_0 = const()[name = tensor("op_48289_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48289_cast_fp16 = einsum(equation = var_48289_equation_0, values = (var_48141_cast_fp16, var_48252_cast_fp16))[name = tensor("op_48289_cast_fp16")]; + tensor var_48291_equation_0 = const()[name = tensor("op_48291_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48291_cast_fp16 = einsum(equation = var_48291_equation_0, values = (var_48145_cast_fp16, var_48253_cast_fp16))[name = tensor("op_48291_cast_fp16")]; + tensor var_48293_equation_0 = const()[name = tensor("op_48293_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48293_cast_fp16 = einsum(equation = var_48293_equation_0, values = (var_48149_cast_fp16, var_48254_cast_fp16))[name = tensor("op_48293_cast_fp16")]; + tensor var_48295_equation_0 = const()[name = tensor("op_48295_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48295_cast_fp16 = einsum(equation = var_48295_equation_0, values = (var_48153_cast_fp16, var_48255_cast_fp16))[name = tensor("op_48295_cast_fp16")]; + tensor input_609_interleave_0 = const()[name = tensor("input_609_interleave_0"), val = tensor(false)]; + tensor input_609_cast_fp16 = concat(axis = var_30355, interleave = input_609_interleave_0, values = (var_48257_cast_fp16, var_48259_cast_fp16, var_48261_cast_fp16, var_48263_cast_fp16, var_48265_cast_fp16, var_48267_cast_fp16, var_48269_cast_fp16, var_48271_cast_fp16, var_48273_cast_fp16, var_48275_cast_fp16, var_48277_cast_fp16, var_48279_cast_fp16, var_48281_cast_fp16, var_48283_cast_fp16, var_48285_cast_fp16, var_48287_cast_fp16, var_48289_cast_fp16, var_48291_cast_fp16, var_48293_cast_fp16, var_48295_cast_fp16))[name = tensor("input_609_cast_fp16")]; + tensor var_48301 = const()[name = tensor("op_48301"), val = tensor([1, 1])]; + tensor var_48303 = const()[name = tensor("op_48303"), val = tensor([1, 1])]; + tensor var_48305_pad_type_0 = const()[name = tensor("op_48305_pad_type_0"), val = tensor("custom")]; + tensor var_48305_pad_0 = const()[name = tensor("op_48305_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_9_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_9_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4023898048)))]; + tensor up_blocks_0_attentions_1_transformer_blocks_9_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_9_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4027174912)))]; + tensor var_48305_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_9_attn1_to_out_0_bias_to_fp16, dilations = var_48303, groups = var_30355, pad = var_48305_pad_0, pad_type = var_48305_pad_type_0, strides = var_48301, weight = up_blocks_0_attentions_1_transformer_blocks_9_attn1_to_out_0_weight_to_fp16, x = input_609_cast_fp16)[name = tensor("op_48305_cast_fp16")]; + tensor inputs_321_cast_fp16 = add(x = var_48305_cast_fp16, y = inputs_319_cast_fp16)[name = tensor("inputs_321_cast_fp16")]; + tensor hidden_states_417_axes_0 = const()[name = tensor("hidden_states_417_axes_0"), val = tensor([1])]; + tensor hidden_states_417_gamma_0_to_fp16 = const()[name = tensor("hidden_states_417_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4027177536)))]; + tensor hidden_states_417_beta_0_to_fp16 = const()[name = tensor("hidden_states_417_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4027180160)))]; + tensor var_48315_to_fp16 = const()[name = tensor("op_48315_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_417_cast_fp16 = layer_norm(axes = hidden_states_417_axes_0, beta = hidden_states_417_beta_0_to_fp16, epsilon = var_48315_to_fp16, gamma = hidden_states_417_gamma_0_to_fp16, x = inputs_321_cast_fp16)[name = tensor("hidden_states_417_cast_fp16")]; + tensor var_48330 = const()[name = tensor("op_48330"), val = tensor([1, 1])]; + tensor var_48332 = const()[name = tensor("op_48332"), val = tensor([1, 1])]; + tensor q_215_pad_type_0 = const()[name = tensor("q_215_pad_type_0"), val = tensor("custom")]; + tensor q_215_pad_0 = const()[name = tensor("q_215_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_9_attn2_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_9_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4027182784)))]; + tensor q_215_cast_fp16 = conv(dilations = var_48332, groups = var_30355, pad = q_215_pad_0, pad_type = q_215_pad_type_0, strides = var_48330, weight = up_blocks_0_attentions_1_transformer_blocks_9_attn2_to_q_weight_to_fp16, x = hidden_states_417_cast_fp16)[name = tensor("q_215_cast_fp16")]; + tensor var_48336 = const()[name = tensor("op_48336"), val = tensor([1, 1])]; + tensor var_48338 = const()[name = tensor("op_48338"), val = tensor([1, 1])]; + tensor k_429_pad_type_0 = const()[name = tensor("k_429_pad_type_0"), val = tensor("custom")]; + tensor k_429_pad_0 = const()[name = tensor("k_429_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_9_attn2_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_9_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4030459648)))]; + tensor k_429_cast_fp16 = conv(dilations = var_48338, groups = var_30355, pad = k_429_pad_0, pad_type = k_429_pad_type_0, strides = var_48336, weight = up_blocks_0_attentions_1_transformer_blocks_9_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_429_cast_fp16")]; + tensor var_48342 = const()[name = tensor("op_48342"), val = tensor([1, 1])]; + tensor var_48344 = const()[name = tensor("op_48344"), val = tensor([1, 1])]; + tensor v_215_pad_type_0 = const()[name = tensor("v_215_pad_type_0"), val = tensor("custom")]; + tensor v_215_pad_0 = const()[name = tensor("v_215_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_9_attn2_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_9_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4035702592)))]; + tensor v_215_cast_fp16 = conv(dilations = var_48344, groups = var_30355, pad = v_215_pad_0, pad_type = v_215_pad_type_0, strides = var_48342, weight = up_blocks_0_attentions_1_transformer_blocks_9_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_215_cast_fp16")]; + tensor var_48348_begin_0 = const()[name = tensor("op_48348_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_48348_end_0 = const()[name = tensor("op_48348_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_48348_end_mask_0 = const()[name = tensor("op_48348_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48348_cast_fp16 = slice_by_index(begin = var_48348_begin_0, end = var_48348_end_0, end_mask = var_48348_end_mask_0, x = q_215_cast_fp16)[name = tensor("op_48348_cast_fp16")]; + tensor var_48352_begin_0 = const()[name = tensor("op_48352_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_48352_end_0 = const()[name = tensor("op_48352_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_48352_end_mask_0 = const()[name = tensor("op_48352_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48352_cast_fp16 = slice_by_index(begin = var_48352_begin_0, end = var_48352_end_0, end_mask = var_48352_end_mask_0, x = q_215_cast_fp16)[name = tensor("op_48352_cast_fp16")]; + tensor var_48356_begin_0 = const()[name = tensor("op_48356_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_48356_end_0 = const()[name = tensor("op_48356_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_48356_end_mask_0 = const()[name = tensor("op_48356_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48356_cast_fp16 = slice_by_index(begin = var_48356_begin_0, end = var_48356_end_0, end_mask = var_48356_end_mask_0, x = q_215_cast_fp16)[name = tensor("op_48356_cast_fp16")]; + tensor var_48360_begin_0 = const()[name = tensor("op_48360_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_48360_end_0 = const()[name = tensor("op_48360_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_48360_end_mask_0 = const()[name = tensor("op_48360_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48360_cast_fp16 = slice_by_index(begin = var_48360_begin_0, end = var_48360_end_0, end_mask = var_48360_end_mask_0, x = q_215_cast_fp16)[name = tensor("op_48360_cast_fp16")]; + tensor var_48364_begin_0 = const()[name = tensor("op_48364_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_48364_end_0 = const()[name = tensor("op_48364_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_48364_end_mask_0 = const()[name = tensor("op_48364_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48364_cast_fp16 = slice_by_index(begin = var_48364_begin_0, end = var_48364_end_0, end_mask = var_48364_end_mask_0, x = q_215_cast_fp16)[name = tensor("op_48364_cast_fp16")]; + tensor var_48368_begin_0 = const()[name = tensor("op_48368_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_48368_end_0 = const()[name = tensor("op_48368_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_48368_end_mask_0 = const()[name = tensor("op_48368_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48368_cast_fp16 = slice_by_index(begin = var_48368_begin_0, end = var_48368_end_0, end_mask = var_48368_end_mask_0, x = q_215_cast_fp16)[name = tensor("op_48368_cast_fp16")]; + tensor var_48372_begin_0 = const()[name = tensor("op_48372_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_48372_end_0 = const()[name = tensor("op_48372_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_48372_end_mask_0 = const()[name = tensor("op_48372_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48372_cast_fp16 = slice_by_index(begin = var_48372_begin_0, end = var_48372_end_0, end_mask = var_48372_end_mask_0, x = q_215_cast_fp16)[name = tensor("op_48372_cast_fp16")]; + tensor var_48376_begin_0 = const()[name = tensor("op_48376_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_48376_end_0 = const()[name = tensor("op_48376_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_48376_end_mask_0 = const()[name = tensor("op_48376_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48376_cast_fp16 = slice_by_index(begin = var_48376_begin_0, end = var_48376_end_0, end_mask = var_48376_end_mask_0, x = q_215_cast_fp16)[name = tensor("op_48376_cast_fp16")]; + tensor var_48380_begin_0 = const()[name = tensor("op_48380_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_48380_end_0 = const()[name = tensor("op_48380_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_48380_end_mask_0 = const()[name = tensor("op_48380_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48380_cast_fp16 = slice_by_index(begin = var_48380_begin_0, end = var_48380_end_0, end_mask = var_48380_end_mask_0, x = q_215_cast_fp16)[name = tensor("op_48380_cast_fp16")]; + tensor var_48384_begin_0 = const()[name = tensor("op_48384_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_48384_end_0 = const()[name = tensor("op_48384_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_48384_end_mask_0 = const()[name = tensor("op_48384_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48384_cast_fp16 = slice_by_index(begin = var_48384_begin_0, end = var_48384_end_0, end_mask = var_48384_end_mask_0, x = q_215_cast_fp16)[name = tensor("op_48384_cast_fp16")]; + tensor var_48388_begin_0 = const()[name = tensor("op_48388_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_48388_end_0 = const()[name = tensor("op_48388_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_48388_end_mask_0 = const()[name = tensor("op_48388_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48388_cast_fp16 = slice_by_index(begin = var_48388_begin_0, end = var_48388_end_0, end_mask = var_48388_end_mask_0, x = q_215_cast_fp16)[name = tensor("op_48388_cast_fp16")]; + tensor var_48392_begin_0 = const()[name = tensor("op_48392_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_48392_end_0 = const()[name = tensor("op_48392_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_48392_end_mask_0 = const()[name = tensor("op_48392_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48392_cast_fp16 = slice_by_index(begin = var_48392_begin_0, end = var_48392_end_0, end_mask = var_48392_end_mask_0, x = q_215_cast_fp16)[name = tensor("op_48392_cast_fp16")]; + tensor var_48396_begin_0 = const()[name = tensor("op_48396_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_48396_end_0 = const()[name = tensor("op_48396_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_48396_end_mask_0 = const()[name = tensor("op_48396_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48396_cast_fp16 = slice_by_index(begin = var_48396_begin_0, end = var_48396_end_0, end_mask = var_48396_end_mask_0, x = q_215_cast_fp16)[name = tensor("op_48396_cast_fp16")]; + tensor var_48400_begin_0 = const()[name = tensor("op_48400_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_48400_end_0 = const()[name = tensor("op_48400_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_48400_end_mask_0 = const()[name = tensor("op_48400_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48400_cast_fp16 = slice_by_index(begin = var_48400_begin_0, end = var_48400_end_0, end_mask = var_48400_end_mask_0, x = q_215_cast_fp16)[name = tensor("op_48400_cast_fp16")]; + tensor var_48404_begin_0 = const()[name = tensor("op_48404_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_48404_end_0 = const()[name = tensor("op_48404_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_48404_end_mask_0 = const()[name = tensor("op_48404_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48404_cast_fp16 = slice_by_index(begin = var_48404_begin_0, end = var_48404_end_0, end_mask = var_48404_end_mask_0, x = q_215_cast_fp16)[name = tensor("op_48404_cast_fp16")]; + tensor var_48408_begin_0 = const()[name = tensor("op_48408_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_48408_end_0 = const()[name = tensor("op_48408_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_48408_end_mask_0 = const()[name = tensor("op_48408_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48408_cast_fp16 = slice_by_index(begin = var_48408_begin_0, end = var_48408_end_0, end_mask = var_48408_end_mask_0, x = q_215_cast_fp16)[name = tensor("op_48408_cast_fp16")]; + tensor var_48412_begin_0 = const()[name = tensor("op_48412_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_48412_end_0 = const()[name = tensor("op_48412_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_48412_end_mask_0 = const()[name = tensor("op_48412_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48412_cast_fp16 = slice_by_index(begin = var_48412_begin_0, end = var_48412_end_0, end_mask = var_48412_end_mask_0, x = q_215_cast_fp16)[name = tensor("op_48412_cast_fp16")]; + tensor var_48416_begin_0 = const()[name = tensor("op_48416_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_48416_end_0 = const()[name = tensor("op_48416_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_48416_end_mask_0 = const()[name = tensor("op_48416_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48416_cast_fp16 = slice_by_index(begin = var_48416_begin_0, end = var_48416_end_0, end_mask = var_48416_end_mask_0, x = q_215_cast_fp16)[name = tensor("op_48416_cast_fp16")]; + tensor var_48420_begin_0 = const()[name = tensor("op_48420_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_48420_end_0 = const()[name = tensor("op_48420_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_48420_end_mask_0 = const()[name = tensor("op_48420_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48420_cast_fp16 = slice_by_index(begin = var_48420_begin_0, end = var_48420_end_0, end_mask = var_48420_end_mask_0, x = q_215_cast_fp16)[name = tensor("op_48420_cast_fp16")]; + tensor var_48424_begin_0 = const()[name = tensor("op_48424_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_48424_end_0 = const()[name = tensor("op_48424_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_48424_end_mask_0 = const()[name = tensor("op_48424_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48424_cast_fp16 = slice_by_index(begin = var_48424_begin_0, end = var_48424_end_0, end_mask = var_48424_end_mask_0, x = q_215_cast_fp16)[name = tensor("op_48424_cast_fp16")]; + tensor k_431_perm_0 = const()[name = tensor("k_431_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_48431_begin_0 = const()[name = tensor("op_48431_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_48431_end_0 = const()[name = tensor("op_48431_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_48431_end_mask_0 = const()[name = tensor("op_48431_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_431_cast_fp16 = transpose(perm = k_431_perm_0, x = k_429_cast_fp16)[name = tensor("transpose_32")]; + tensor var_48431_cast_fp16 = slice_by_index(begin = var_48431_begin_0, end = var_48431_end_0, end_mask = var_48431_end_mask_0, x = k_431_cast_fp16)[name = tensor("op_48431_cast_fp16")]; + tensor var_48435_begin_0 = const()[name = tensor("op_48435_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_48435_end_0 = const()[name = tensor("op_48435_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_48435_end_mask_0 = const()[name = tensor("op_48435_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48435_cast_fp16 = slice_by_index(begin = var_48435_begin_0, end = var_48435_end_0, end_mask = var_48435_end_mask_0, x = k_431_cast_fp16)[name = tensor("op_48435_cast_fp16")]; + tensor var_48439_begin_0 = const()[name = tensor("op_48439_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_48439_end_0 = const()[name = tensor("op_48439_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_48439_end_mask_0 = const()[name = tensor("op_48439_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48439_cast_fp16 = slice_by_index(begin = var_48439_begin_0, end = var_48439_end_0, end_mask = var_48439_end_mask_0, x = k_431_cast_fp16)[name = tensor("op_48439_cast_fp16")]; + tensor var_48443_begin_0 = const()[name = tensor("op_48443_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_48443_end_0 = const()[name = tensor("op_48443_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_48443_end_mask_0 = const()[name = tensor("op_48443_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48443_cast_fp16 = slice_by_index(begin = var_48443_begin_0, end = var_48443_end_0, end_mask = var_48443_end_mask_0, x = k_431_cast_fp16)[name = tensor("op_48443_cast_fp16")]; + tensor var_48447_begin_0 = const()[name = tensor("op_48447_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_48447_end_0 = const()[name = tensor("op_48447_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_48447_end_mask_0 = const()[name = tensor("op_48447_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48447_cast_fp16 = slice_by_index(begin = var_48447_begin_0, end = var_48447_end_0, end_mask = var_48447_end_mask_0, x = k_431_cast_fp16)[name = tensor("op_48447_cast_fp16")]; + tensor var_48451_begin_0 = const()[name = tensor("op_48451_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_48451_end_0 = const()[name = tensor("op_48451_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_48451_end_mask_0 = const()[name = tensor("op_48451_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48451_cast_fp16 = slice_by_index(begin = var_48451_begin_0, end = var_48451_end_0, end_mask = var_48451_end_mask_0, x = k_431_cast_fp16)[name = tensor("op_48451_cast_fp16")]; + tensor var_48455_begin_0 = const()[name = tensor("op_48455_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_48455_end_0 = const()[name = tensor("op_48455_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_48455_end_mask_0 = const()[name = tensor("op_48455_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48455_cast_fp16 = slice_by_index(begin = var_48455_begin_0, end = var_48455_end_0, end_mask = var_48455_end_mask_0, x = k_431_cast_fp16)[name = tensor("op_48455_cast_fp16")]; + tensor var_48459_begin_0 = const()[name = tensor("op_48459_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_48459_end_0 = const()[name = tensor("op_48459_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_48459_end_mask_0 = const()[name = tensor("op_48459_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48459_cast_fp16 = slice_by_index(begin = var_48459_begin_0, end = var_48459_end_0, end_mask = var_48459_end_mask_0, x = k_431_cast_fp16)[name = tensor("op_48459_cast_fp16")]; + tensor var_48463_begin_0 = const()[name = tensor("op_48463_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_48463_end_0 = const()[name = tensor("op_48463_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_48463_end_mask_0 = const()[name = tensor("op_48463_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48463_cast_fp16 = slice_by_index(begin = var_48463_begin_0, end = var_48463_end_0, end_mask = var_48463_end_mask_0, x = k_431_cast_fp16)[name = tensor("op_48463_cast_fp16")]; + tensor var_48467_begin_0 = const()[name = tensor("op_48467_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_48467_end_0 = const()[name = tensor("op_48467_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_48467_end_mask_0 = const()[name = tensor("op_48467_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48467_cast_fp16 = slice_by_index(begin = var_48467_begin_0, end = var_48467_end_0, end_mask = var_48467_end_mask_0, x = k_431_cast_fp16)[name = tensor("op_48467_cast_fp16")]; + tensor var_48471_begin_0 = const()[name = tensor("op_48471_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_48471_end_0 = const()[name = tensor("op_48471_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_48471_end_mask_0 = const()[name = tensor("op_48471_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48471_cast_fp16 = slice_by_index(begin = var_48471_begin_0, end = var_48471_end_0, end_mask = var_48471_end_mask_0, x = k_431_cast_fp16)[name = tensor("op_48471_cast_fp16")]; + tensor var_48475_begin_0 = const()[name = tensor("op_48475_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_48475_end_0 = const()[name = tensor("op_48475_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_48475_end_mask_0 = const()[name = tensor("op_48475_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48475_cast_fp16 = slice_by_index(begin = var_48475_begin_0, end = var_48475_end_0, end_mask = var_48475_end_mask_0, x = k_431_cast_fp16)[name = tensor("op_48475_cast_fp16")]; + tensor var_48479_begin_0 = const()[name = tensor("op_48479_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_48479_end_0 = const()[name = tensor("op_48479_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_48479_end_mask_0 = const()[name = tensor("op_48479_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48479_cast_fp16 = slice_by_index(begin = var_48479_begin_0, end = var_48479_end_0, end_mask = var_48479_end_mask_0, x = k_431_cast_fp16)[name = tensor("op_48479_cast_fp16")]; + tensor var_48483_begin_0 = const()[name = tensor("op_48483_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_48483_end_0 = const()[name = tensor("op_48483_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_48483_end_mask_0 = const()[name = tensor("op_48483_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48483_cast_fp16 = slice_by_index(begin = var_48483_begin_0, end = var_48483_end_0, end_mask = var_48483_end_mask_0, x = k_431_cast_fp16)[name = tensor("op_48483_cast_fp16")]; + tensor var_48487_begin_0 = const()[name = tensor("op_48487_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_48487_end_0 = const()[name = tensor("op_48487_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_48487_end_mask_0 = const()[name = tensor("op_48487_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48487_cast_fp16 = slice_by_index(begin = var_48487_begin_0, end = var_48487_end_0, end_mask = var_48487_end_mask_0, x = k_431_cast_fp16)[name = tensor("op_48487_cast_fp16")]; + tensor var_48491_begin_0 = const()[name = tensor("op_48491_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_48491_end_0 = const()[name = tensor("op_48491_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_48491_end_mask_0 = const()[name = tensor("op_48491_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48491_cast_fp16 = slice_by_index(begin = var_48491_begin_0, end = var_48491_end_0, end_mask = var_48491_end_mask_0, x = k_431_cast_fp16)[name = tensor("op_48491_cast_fp16")]; + tensor var_48495_begin_0 = const()[name = tensor("op_48495_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_48495_end_0 = const()[name = tensor("op_48495_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_48495_end_mask_0 = const()[name = tensor("op_48495_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48495_cast_fp16 = slice_by_index(begin = var_48495_begin_0, end = var_48495_end_0, end_mask = var_48495_end_mask_0, x = k_431_cast_fp16)[name = tensor("op_48495_cast_fp16")]; + tensor var_48499_begin_0 = const()[name = tensor("op_48499_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_48499_end_0 = const()[name = tensor("op_48499_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_48499_end_mask_0 = const()[name = tensor("op_48499_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48499_cast_fp16 = slice_by_index(begin = var_48499_begin_0, end = var_48499_end_0, end_mask = var_48499_end_mask_0, x = k_431_cast_fp16)[name = tensor("op_48499_cast_fp16")]; + tensor var_48503_begin_0 = const()[name = tensor("op_48503_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_48503_end_0 = const()[name = tensor("op_48503_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_48503_end_mask_0 = const()[name = tensor("op_48503_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48503_cast_fp16 = slice_by_index(begin = var_48503_begin_0, end = var_48503_end_0, end_mask = var_48503_end_mask_0, x = k_431_cast_fp16)[name = tensor("op_48503_cast_fp16")]; + tensor var_48507_begin_0 = const()[name = tensor("op_48507_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_48507_end_0 = const()[name = tensor("op_48507_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_48507_end_mask_0 = const()[name = tensor("op_48507_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48507_cast_fp16 = slice_by_index(begin = var_48507_begin_0, end = var_48507_end_0, end_mask = var_48507_end_mask_0, x = k_431_cast_fp16)[name = tensor("op_48507_cast_fp16")]; + tensor var_48509_begin_0 = const()[name = tensor("op_48509_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_48509_end_0 = const()[name = tensor("op_48509_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_48509_end_mask_0 = const()[name = tensor("op_48509_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48509_cast_fp16 = slice_by_index(begin = var_48509_begin_0, end = var_48509_end_0, end_mask = var_48509_end_mask_0, x = v_215_cast_fp16)[name = tensor("op_48509_cast_fp16")]; + tensor var_48513_begin_0 = const()[name = tensor("op_48513_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_48513_end_0 = const()[name = tensor("op_48513_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_48513_end_mask_0 = const()[name = tensor("op_48513_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48513_cast_fp16 = slice_by_index(begin = var_48513_begin_0, end = var_48513_end_0, end_mask = var_48513_end_mask_0, x = v_215_cast_fp16)[name = tensor("op_48513_cast_fp16")]; + tensor var_48517_begin_0 = const()[name = tensor("op_48517_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_48517_end_0 = const()[name = tensor("op_48517_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_48517_end_mask_0 = const()[name = tensor("op_48517_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48517_cast_fp16 = slice_by_index(begin = var_48517_begin_0, end = var_48517_end_0, end_mask = var_48517_end_mask_0, x = v_215_cast_fp16)[name = tensor("op_48517_cast_fp16")]; + tensor var_48521_begin_0 = const()[name = tensor("op_48521_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_48521_end_0 = const()[name = tensor("op_48521_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_48521_end_mask_0 = const()[name = tensor("op_48521_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48521_cast_fp16 = slice_by_index(begin = var_48521_begin_0, end = var_48521_end_0, end_mask = var_48521_end_mask_0, x = v_215_cast_fp16)[name = tensor("op_48521_cast_fp16")]; + tensor var_48525_begin_0 = const()[name = tensor("op_48525_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_48525_end_0 = const()[name = tensor("op_48525_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_48525_end_mask_0 = const()[name = tensor("op_48525_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48525_cast_fp16 = slice_by_index(begin = var_48525_begin_0, end = var_48525_end_0, end_mask = var_48525_end_mask_0, x = v_215_cast_fp16)[name = tensor("op_48525_cast_fp16")]; + tensor var_48529_begin_0 = const()[name = tensor("op_48529_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_48529_end_0 = const()[name = tensor("op_48529_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_48529_end_mask_0 = const()[name = tensor("op_48529_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48529_cast_fp16 = slice_by_index(begin = var_48529_begin_0, end = var_48529_end_0, end_mask = var_48529_end_mask_0, x = v_215_cast_fp16)[name = tensor("op_48529_cast_fp16")]; + tensor var_48533_begin_0 = const()[name = tensor("op_48533_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_48533_end_0 = const()[name = tensor("op_48533_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_48533_end_mask_0 = const()[name = tensor("op_48533_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48533_cast_fp16 = slice_by_index(begin = var_48533_begin_0, end = var_48533_end_0, end_mask = var_48533_end_mask_0, x = v_215_cast_fp16)[name = tensor("op_48533_cast_fp16")]; + tensor var_48537_begin_0 = const()[name = tensor("op_48537_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_48537_end_0 = const()[name = tensor("op_48537_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_48537_end_mask_0 = const()[name = tensor("op_48537_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48537_cast_fp16 = slice_by_index(begin = var_48537_begin_0, end = var_48537_end_0, end_mask = var_48537_end_mask_0, x = v_215_cast_fp16)[name = tensor("op_48537_cast_fp16")]; + tensor var_48541_begin_0 = const()[name = tensor("op_48541_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_48541_end_0 = const()[name = tensor("op_48541_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_48541_end_mask_0 = const()[name = tensor("op_48541_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48541_cast_fp16 = slice_by_index(begin = var_48541_begin_0, end = var_48541_end_0, end_mask = var_48541_end_mask_0, x = v_215_cast_fp16)[name = tensor("op_48541_cast_fp16")]; + tensor var_48545_begin_0 = const()[name = tensor("op_48545_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_48545_end_0 = const()[name = tensor("op_48545_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_48545_end_mask_0 = const()[name = tensor("op_48545_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48545_cast_fp16 = slice_by_index(begin = var_48545_begin_0, end = var_48545_end_0, end_mask = var_48545_end_mask_0, x = v_215_cast_fp16)[name = tensor("op_48545_cast_fp16")]; + tensor var_48549_begin_0 = const()[name = tensor("op_48549_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_48549_end_0 = const()[name = tensor("op_48549_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_48549_end_mask_0 = const()[name = tensor("op_48549_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48549_cast_fp16 = slice_by_index(begin = var_48549_begin_0, end = var_48549_end_0, end_mask = var_48549_end_mask_0, x = v_215_cast_fp16)[name = tensor("op_48549_cast_fp16")]; + tensor var_48553_begin_0 = const()[name = tensor("op_48553_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_48553_end_0 = const()[name = tensor("op_48553_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_48553_end_mask_0 = const()[name = tensor("op_48553_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48553_cast_fp16 = slice_by_index(begin = var_48553_begin_0, end = var_48553_end_0, end_mask = var_48553_end_mask_0, x = v_215_cast_fp16)[name = tensor("op_48553_cast_fp16")]; + tensor var_48557_begin_0 = const()[name = tensor("op_48557_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_48557_end_0 = const()[name = tensor("op_48557_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_48557_end_mask_0 = const()[name = tensor("op_48557_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48557_cast_fp16 = slice_by_index(begin = var_48557_begin_0, end = var_48557_end_0, end_mask = var_48557_end_mask_0, x = v_215_cast_fp16)[name = tensor("op_48557_cast_fp16")]; + tensor var_48561_begin_0 = const()[name = tensor("op_48561_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_48561_end_0 = const()[name = tensor("op_48561_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_48561_end_mask_0 = const()[name = tensor("op_48561_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48561_cast_fp16 = slice_by_index(begin = var_48561_begin_0, end = var_48561_end_0, end_mask = var_48561_end_mask_0, x = v_215_cast_fp16)[name = tensor("op_48561_cast_fp16")]; + tensor var_48565_begin_0 = const()[name = tensor("op_48565_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_48565_end_0 = const()[name = tensor("op_48565_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_48565_end_mask_0 = const()[name = tensor("op_48565_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48565_cast_fp16 = slice_by_index(begin = var_48565_begin_0, end = var_48565_end_0, end_mask = var_48565_end_mask_0, x = v_215_cast_fp16)[name = tensor("op_48565_cast_fp16")]; + tensor var_48569_begin_0 = const()[name = tensor("op_48569_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_48569_end_0 = const()[name = tensor("op_48569_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_48569_end_mask_0 = const()[name = tensor("op_48569_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48569_cast_fp16 = slice_by_index(begin = var_48569_begin_0, end = var_48569_end_0, end_mask = var_48569_end_mask_0, x = v_215_cast_fp16)[name = tensor("op_48569_cast_fp16")]; + tensor var_48573_begin_0 = const()[name = tensor("op_48573_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_48573_end_0 = const()[name = tensor("op_48573_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_48573_end_mask_0 = const()[name = tensor("op_48573_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48573_cast_fp16 = slice_by_index(begin = var_48573_begin_0, end = var_48573_end_0, end_mask = var_48573_end_mask_0, x = v_215_cast_fp16)[name = tensor("op_48573_cast_fp16")]; + tensor var_48577_begin_0 = const()[name = tensor("op_48577_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_48577_end_0 = const()[name = tensor("op_48577_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_48577_end_mask_0 = const()[name = tensor("op_48577_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48577_cast_fp16 = slice_by_index(begin = var_48577_begin_0, end = var_48577_end_0, end_mask = var_48577_end_mask_0, x = v_215_cast_fp16)[name = tensor("op_48577_cast_fp16")]; + tensor var_48581_begin_0 = const()[name = tensor("op_48581_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_48581_end_0 = const()[name = tensor("op_48581_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_48581_end_mask_0 = const()[name = tensor("op_48581_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48581_cast_fp16 = slice_by_index(begin = var_48581_begin_0, end = var_48581_end_0, end_mask = var_48581_end_mask_0, x = v_215_cast_fp16)[name = tensor("op_48581_cast_fp16")]; + tensor var_48585_begin_0 = const()[name = tensor("op_48585_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_48585_end_0 = const()[name = tensor("op_48585_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_48585_end_mask_0 = const()[name = tensor("op_48585_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48585_cast_fp16 = slice_by_index(begin = var_48585_begin_0, end = var_48585_end_0, end_mask = var_48585_end_mask_0, x = v_215_cast_fp16)[name = tensor("op_48585_cast_fp16")]; + tensor var_48589_equation_0 = const()[name = tensor("op_48589_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48589_cast_fp16 = einsum(equation = var_48589_equation_0, values = (var_48431_cast_fp16, var_48348_cast_fp16))[name = tensor("op_48589_cast_fp16")]; + tensor var_48590_to_fp16 = const()[name = tensor("op_48590_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4121_cast_fp16 = mul(x = var_48589_cast_fp16, y = var_48590_to_fp16)[name = tensor("aw_4121_cast_fp16")]; + tensor var_48593_equation_0 = const()[name = tensor("op_48593_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48593_cast_fp16 = einsum(equation = var_48593_equation_0, values = (var_48435_cast_fp16, var_48352_cast_fp16))[name = tensor("op_48593_cast_fp16")]; + tensor var_48594_to_fp16 = const()[name = tensor("op_48594_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4123_cast_fp16 = mul(x = var_48593_cast_fp16, y = var_48594_to_fp16)[name = tensor("aw_4123_cast_fp16")]; + tensor var_48597_equation_0 = const()[name = tensor("op_48597_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48597_cast_fp16 = einsum(equation = var_48597_equation_0, values = (var_48439_cast_fp16, var_48356_cast_fp16))[name = tensor("op_48597_cast_fp16")]; + tensor var_48598_to_fp16 = const()[name = tensor("op_48598_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4125_cast_fp16 = mul(x = var_48597_cast_fp16, y = var_48598_to_fp16)[name = tensor("aw_4125_cast_fp16")]; + tensor var_48601_equation_0 = const()[name = tensor("op_48601_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48601_cast_fp16 = einsum(equation = var_48601_equation_0, values = (var_48443_cast_fp16, var_48360_cast_fp16))[name = tensor("op_48601_cast_fp16")]; + tensor var_48602_to_fp16 = const()[name = tensor("op_48602_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4127_cast_fp16 = mul(x = var_48601_cast_fp16, y = var_48602_to_fp16)[name = tensor("aw_4127_cast_fp16")]; + tensor var_48605_equation_0 = const()[name = tensor("op_48605_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48605_cast_fp16 = einsum(equation = var_48605_equation_0, values = (var_48447_cast_fp16, var_48364_cast_fp16))[name = tensor("op_48605_cast_fp16")]; + tensor var_48606_to_fp16 = const()[name = tensor("op_48606_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4129_cast_fp16 = mul(x = var_48605_cast_fp16, y = var_48606_to_fp16)[name = tensor("aw_4129_cast_fp16")]; + tensor var_48609_equation_0 = const()[name = tensor("op_48609_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48609_cast_fp16 = einsum(equation = var_48609_equation_0, values = (var_48451_cast_fp16, var_48368_cast_fp16))[name = tensor("op_48609_cast_fp16")]; + tensor var_48610_to_fp16 = const()[name = tensor("op_48610_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4131_cast_fp16 = mul(x = var_48609_cast_fp16, y = var_48610_to_fp16)[name = tensor("aw_4131_cast_fp16")]; + tensor var_48613_equation_0 = const()[name = tensor("op_48613_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48613_cast_fp16 = einsum(equation = var_48613_equation_0, values = (var_48455_cast_fp16, var_48372_cast_fp16))[name = tensor("op_48613_cast_fp16")]; + tensor var_48614_to_fp16 = const()[name = tensor("op_48614_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4133_cast_fp16 = mul(x = var_48613_cast_fp16, y = var_48614_to_fp16)[name = tensor("aw_4133_cast_fp16")]; + tensor var_48617_equation_0 = const()[name = tensor("op_48617_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48617_cast_fp16 = einsum(equation = var_48617_equation_0, values = (var_48459_cast_fp16, var_48376_cast_fp16))[name = tensor("op_48617_cast_fp16")]; + tensor var_48618_to_fp16 = const()[name = tensor("op_48618_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4135_cast_fp16 = mul(x = var_48617_cast_fp16, y = var_48618_to_fp16)[name = tensor("aw_4135_cast_fp16")]; + tensor var_48621_equation_0 = const()[name = tensor("op_48621_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48621_cast_fp16 = einsum(equation = var_48621_equation_0, values = (var_48463_cast_fp16, var_48380_cast_fp16))[name = tensor("op_48621_cast_fp16")]; + tensor var_48622_to_fp16 = const()[name = tensor("op_48622_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4137_cast_fp16 = mul(x = var_48621_cast_fp16, y = var_48622_to_fp16)[name = tensor("aw_4137_cast_fp16")]; + tensor var_48625_equation_0 = const()[name = tensor("op_48625_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48625_cast_fp16 = einsum(equation = var_48625_equation_0, values = (var_48467_cast_fp16, var_48384_cast_fp16))[name = tensor("op_48625_cast_fp16")]; + tensor var_48626_to_fp16 = const()[name = tensor("op_48626_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4139_cast_fp16 = mul(x = var_48625_cast_fp16, y = var_48626_to_fp16)[name = tensor("aw_4139_cast_fp16")]; + tensor var_48629_equation_0 = const()[name = tensor("op_48629_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48629_cast_fp16 = einsum(equation = var_48629_equation_0, values = (var_48471_cast_fp16, var_48388_cast_fp16))[name = tensor("op_48629_cast_fp16")]; + tensor var_48630_to_fp16 = const()[name = tensor("op_48630_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4141_cast_fp16 = mul(x = var_48629_cast_fp16, y = var_48630_to_fp16)[name = tensor("aw_4141_cast_fp16")]; + tensor var_48633_equation_0 = const()[name = tensor("op_48633_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48633_cast_fp16 = einsum(equation = var_48633_equation_0, values = (var_48475_cast_fp16, var_48392_cast_fp16))[name = tensor("op_48633_cast_fp16")]; + tensor var_48634_to_fp16 = const()[name = tensor("op_48634_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4143_cast_fp16 = mul(x = var_48633_cast_fp16, y = var_48634_to_fp16)[name = tensor("aw_4143_cast_fp16")]; + tensor var_48637_equation_0 = const()[name = tensor("op_48637_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48637_cast_fp16 = einsum(equation = var_48637_equation_0, values = (var_48479_cast_fp16, var_48396_cast_fp16))[name = tensor("op_48637_cast_fp16")]; + tensor var_48638_to_fp16 = const()[name = tensor("op_48638_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4145_cast_fp16 = mul(x = var_48637_cast_fp16, y = var_48638_to_fp16)[name = tensor("aw_4145_cast_fp16")]; + tensor var_48641_equation_0 = const()[name = tensor("op_48641_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48641_cast_fp16 = einsum(equation = var_48641_equation_0, values = (var_48483_cast_fp16, var_48400_cast_fp16))[name = tensor("op_48641_cast_fp16")]; + tensor var_48642_to_fp16 = const()[name = tensor("op_48642_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4147_cast_fp16 = mul(x = var_48641_cast_fp16, y = var_48642_to_fp16)[name = tensor("aw_4147_cast_fp16")]; + tensor var_48645_equation_0 = const()[name = tensor("op_48645_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48645_cast_fp16 = einsum(equation = var_48645_equation_0, values = (var_48487_cast_fp16, var_48404_cast_fp16))[name = tensor("op_48645_cast_fp16")]; + tensor var_48646_to_fp16 = const()[name = tensor("op_48646_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4149_cast_fp16 = mul(x = var_48645_cast_fp16, y = var_48646_to_fp16)[name = tensor("aw_4149_cast_fp16")]; + tensor var_48649_equation_0 = const()[name = tensor("op_48649_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48649_cast_fp16 = einsum(equation = var_48649_equation_0, values = (var_48491_cast_fp16, var_48408_cast_fp16))[name = tensor("op_48649_cast_fp16")]; + tensor var_48650_to_fp16 = const()[name = tensor("op_48650_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4151_cast_fp16 = mul(x = var_48649_cast_fp16, y = var_48650_to_fp16)[name = tensor("aw_4151_cast_fp16")]; + tensor var_48653_equation_0 = const()[name = tensor("op_48653_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48653_cast_fp16 = einsum(equation = var_48653_equation_0, values = (var_48495_cast_fp16, var_48412_cast_fp16))[name = tensor("op_48653_cast_fp16")]; + tensor var_48654_to_fp16 = const()[name = tensor("op_48654_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4153_cast_fp16 = mul(x = var_48653_cast_fp16, y = var_48654_to_fp16)[name = tensor("aw_4153_cast_fp16")]; + tensor var_48657_equation_0 = const()[name = tensor("op_48657_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48657_cast_fp16 = einsum(equation = var_48657_equation_0, values = (var_48499_cast_fp16, var_48416_cast_fp16))[name = tensor("op_48657_cast_fp16")]; + tensor var_48658_to_fp16 = const()[name = tensor("op_48658_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4155_cast_fp16 = mul(x = var_48657_cast_fp16, y = var_48658_to_fp16)[name = tensor("aw_4155_cast_fp16")]; + tensor var_48661_equation_0 = const()[name = tensor("op_48661_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48661_cast_fp16 = einsum(equation = var_48661_equation_0, values = (var_48503_cast_fp16, var_48420_cast_fp16))[name = tensor("op_48661_cast_fp16")]; + tensor var_48662_to_fp16 = const()[name = tensor("op_48662_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4157_cast_fp16 = mul(x = var_48661_cast_fp16, y = var_48662_to_fp16)[name = tensor("aw_4157_cast_fp16")]; + tensor var_48665_equation_0 = const()[name = tensor("op_48665_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48665_cast_fp16 = einsum(equation = var_48665_equation_0, values = (var_48507_cast_fp16, var_48424_cast_fp16))[name = tensor("op_48665_cast_fp16")]; + tensor var_48666_to_fp16 = const()[name = tensor("op_48666_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4159_cast_fp16 = mul(x = var_48665_cast_fp16, y = var_48666_to_fp16)[name = tensor("aw_4159_cast_fp16")]; + tensor var_48668_cast_fp16 = softmax(axis = var_30355, x = aw_4121_cast_fp16)[name = tensor("op_48668_cast_fp16")]; + tensor var_48669_cast_fp16 = softmax(axis = var_30355, x = aw_4123_cast_fp16)[name = tensor("op_48669_cast_fp16")]; + tensor var_48670_cast_fp16 = softmax(axis = var_30355, x = aw_4125_cast_fp16)[name = tensor("op_48670_cast_fp16")]; + tensor var_48671_cast_fp16 = softmax(axis = var_30355, x = aw_4127_cast_fp16)[name = tensor("op_48671_cast_fp16")]; + tensor var_48672_cast_fp16 = softmax(axis = var_30355, x = aw_4129_cast_fp16)[name = tensor("op_48672_cast_fp16")]; + tensor var_48673_cast_fp16 = softmax(axis = var_30355, x = aw_4131_cast_fp16)[name = tensor("op_48673_cast_fp16")]; + tensor var_48674_cast_fp16 = softmax(axis = var_30355, x = aw_4133_cast_fp16)[name = tensor("op_48674_cast_fp16")]; + tensor var_48675_cast_fp16 = softmax(axis = var_30355, x = aw_4135_cast_fp16)[name = tensor("op_48675_cast_fp16")]; + tensor var_48676_cast_fp16 = softmax(axis = var_30355, x = aw_4137_cast_fp16)[name = tensor("op_48676_cast_fp16")]; + tensor var_48677_cast_fp16 = softmax(axis = var_30355, x = aw_4139_cast_fp16)[name = tensor("op_48677_cast_fp16")]; + tensor var_48678_cast_fp16 = softmax(axis = var_30355, x = aw_4141_cast_fp16)[name = tensor("op_48678_cast_fp16")]; + tensor var_48679_cast_fp16 = softmax(axis = var_30355, x = aw_4143_cast_fp16)[name = tensor("op_48679_cast_fp16")]; + tensor var_48680_cast_fp16 = softmax(axis = var_30355, x = aw_4145_cast_fp16)[name = tensor("op_48680_cast_fp16")]; + tensor var_48681_cast_fp16 = softmax(axis = var_30355, x = aw_4147_cast_fp16)[name = tensor("op_48681_cast_fp16")]; + tensor var_48682_cast_fp16 = softmax(axis = var_30355, x = aw_4149_cast_fp16)[name = tensor("op_48682_cast_fp16")]; + tensor var_48683_cast_fp16 = softmax(axis = var_30355, x = aw_4151_cast_fp16)[name = tensor("op_48683_cast_fp16")]; + tensor var_48684_cast_fp16 = softmax(axis = var_30355, x = aw_4153_cast_fp16)[name = tensor("op_48684_cast_fp16")]; + tensor var_48685_cast_fp16 = softmax(axis = var_30355, x = aw_4155_cast_fp16)[name = tensor("op_48685_cast_fp16")]; + tensor var_48686_cast_fp16 = softmax(axis = var_30355, x = aw_4157_cast_fp16)[name = tensor("op_48686_cast_fp16")]; + tensor var_48687_cast_fp16 = softmax(axis = var_30355, x = aw_4159_cast_fp16)[name = tensor("op_48687_cast_fp16")]; + tensor var_48689_equation_0 = const()[name = tensor("op_48689_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48689_cast_fp16 = einsum(equation = var_48689_equation_0, values = (var_48509_cast_fp16, var_48668_cast_fp16))[name = tensor("op_48689_cast_fp16")]; + tensor var_48691_equation_0 = const()[name = tensor("op_48691_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48691_cast_fp16 = einsum(equation = var_48691_equation_0, values = (var_48513_cast_fp16, var_48669_cast_fp16))[name = tensor("op_48691_cast_fp16")]; + tensor var_48693_equation_0 = const()[name = tensor("op_48693_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48693_cast_fp16 = einsum(equation = var_48693_equation_0, values = (var_48517_cast_fp16, var_48670_cast_fp16))[name = tensor("op_48693_cast_fp16")]; + tensor var_48695_equation_0 = const()[name = tensor("op_48695_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48695_cast_fp16 = einsum(equation = var_48695_equation_0, values = (var_48521_cast_fp16, var_48671_cast_fp16))[name = tensor("op_48695_cast_fp16")]; + tensor var_48697_equation_0 = const()[name = tensor("op_48697_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48697_cast_fp16 = einsum(equation = var_48697_equation_0, values = (var_48525_cast_fp16, var_48672_cast_fp16))[name = tensor("op_48697_cast_fp16")]; + tensor var_48699_equation_0 = const()[name = tensor("op_48699_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48699_cast_fp16 = einsum(equation = var_48699_equation_0, values = (var_48529_cast_fp16, var_48673_cast_fp16))[name = tensor("op_48699_cast_fp16")]; + tensor var_48701_equation_0 = const()[name = tensor("op_48701_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48701_cast_fp16 = einsum(equation = var_48701_equation_0, values = (var_48533_cast_fp16, var_48674_cast_fp16))[name = tensor("op_48701_cast_fp16")]; + tensor var_48703_equation_0 = const()[name = tensor("op_48703_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48703_cast_fp16 = einsum(equation = var_48703_equation_0, values = (var_48537_cast_fp16, var_48675_cast_fp16))[name = tensor("op_48703_cast_fp16")]; + tensor var_48705_equation_0 = const()[name = tensor("op_48705_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48705_cast_fp16 = einsum(equation = var_48705_equation_0, values = (var_48541_cast_fp16, var_48676_cast_fp16))[name = tensor("op_48705_cast_fp16")]; + tensor var_48707_equation_0 = const()[name = tensor("op_48707_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48707_cast_fp16 = einsum(equation = var_48707_equation_0, values = (var_48545_cast_fp16, var_48677_cast_fp16))[name = tensor("op_48707_cast_fp16")]; + tensor var_48709_equation_0 = const()[name = tensor("op_48709_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48709_cast_fp16 = einsum(equation = var_48709_equation_0, values = (var_48549_cast_fp16, var_48678_cast_fp16))[name = tensor("op_48709_cast_fp16")]; + tensor var_48711_equation_0 = const()[name = tensor("op_48711_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48711_cast_fp16 = einsum(equation = var_48711_equation_0, values = (var_48553_cast_fp16, var_48679_cast_fp16))[name = tensor("op_48711_cast_fp16")]; + tensor var_48713_equation_0 = const()[name = tensor("op_48713_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48713_cast_fp16 = einsum(equation = var_48713_equation_0, values = (var_48557_cast_fp16, var_48680_cast_fp16))[name = tensor("op_48713_cast_fp16")]; + tensor var_48715_equation_0 = const()[name = tensor("op_48715_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48715_cast_fp16 = einsum(equation = var_48715_equation_0, values = (var_48561_cast_fp16, var_48681_cast_fp16))[name = tensor("op_48715_cast_fp16")]; + tensor var_48717_equation_0 = const()[name = tensor("op_48717_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48717_cast_fp16 = einsum(equation = var_48717_equation_0, values = (var_48565_cast_fp16, var_48682_cast_fp16))[name = tensor("op_48717_cast_fp16")]; + tensor var_48719_equation_0 = const()[name = tensor("op_48719_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48719_cast_fp16 = einsum(equation = var_48719_equation_0, values = (var_48569_cast_fp16, var_48683_cast_fp16))[name = tensor("op_48719_cast_fp16")]; + tensor var_48721_equation_0 = const()[name = tensor("op_48721_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48721_cast_fp16 = einsum(equation = var_48721_equation_0, values = (var_48573_cast_fp16, var_48684_cast_fp16))[name = tensor("op_48721_cast_fp16")]; + tensor var_48723_equation_0 = const()[name = tensor("op_48723_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48723_cast_fp16 = einsum(equation = var_48723_equation_0, values = (var_48577_cast_fp16, var_48685_cast_fp16))[name = tensor("op_48723_cast_fp16")]; + tensor var_48725_equation_0 = const()[name = tensor("op_48725_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48725_cast_fp16 = einsum(equation = var_48725_equation_0, values = (var_48581_cast_fp16, var_48686_cast_fp16))[name = tensor("op_48725_cast_fp16")]; + tensor var_48727_equation_0 = const()[name = tensor("op_48727_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48727_cast_fp16 = einsum(equation = var_48727_equation_0, values = (var_48585_cast_fp16, var_48687_cast_fp16))[name = tensor("op_48727_cast_fp16")]; + tensor input_611_interleave_0 = const()[name = tensor("input_611_interleave_0"), val = tensor(false)]; + tensor input_611_cast_fp16 = concat(axis = var_30355, interleave = input_611_interleave_0, values = (var_48689_cast_fp16, var_48691_cast_fp16, var_48693_cast_fp16, var_48695_cast_fp16, var_48697_cast_fp16, var_48699_cast_fp16, var_48701_cast_fp16, var_48703_cast_fp16, var_48705_cast_fp16, var_48707_cast_fp16, var_48709_cast_fp16, var_48711_cast_fp16, var_48713_cast_fp16, var_48715_cast_fp16, var_48717_cast_fp16, var_48719_cast_fp16, var_48721_cast_fp16, var_48723_cast_fp16, var_48725_cast_fp16, var_48727_cast_fp16))[name = tensor("input_611_cast_fp16")]; + tensor var_48733 = const()[name = tensor("op_48733"), val = tensor([1, 1])]; + tensor var_48735 = const()[name = tensor("op_48735"), val = tensor([1, 1])]; + tensor var_48737_pad_type_0 = const()[name = tensor("op_48737_pad_type_0"), val = tensor("custom")]; + tensor var_48737_pad_0 = const()[name = tensor("op_48737_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_9_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_9_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4040945536)))]; + tensor up_blocks_0_attentions_1_transformer_blocks_9_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_9_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4044222400)))]; + tensor var_48737_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_9_attn2_to_out_0_bias_to_fp16, dilations = var_48735, groups = var_30355, pad = var_48737_pad_0, pad_type = var_48737_pad_type_0, strides = var_48733, weight = up_blocks_0_attentions_1_transformer_blocks_9_attn2_to_out_0_weight_to_fp16, x = input_611_cast_fp16)[name = tensor("op_48737_cast_fp16")]; + tensor inputs_323_cast_fp16 = add(x = var_48737_cast_fp16, y = inputs_321_cast_fp16)[name = tensor("inputs_323_cast_fp16")]; + tensor input_613_axes_0 = const()[name = tensor("input_613_axes_0"), val = tensor([1])]; + tensor input_613_gamma_0_to_fp16 = const()[name = tensor("input_613_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4044225024)))]; + tensor input_613_beta_0_to_fp16 = const()[name = tensor("input_613_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4044227648)))]; + tensor var_48747_to_fp16 = const()[name = tensor("op_48747_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_613_cast_fp16 = layer_norm(axes = input_613_axes_0, beta = input_613_beta_0_to_fp16, epsilon = var_48747_to_fp16, gamma = input_613_gamma_0_to_fp16, x = inputs_323_cast_fp16)[name = tensor("input_613_cast_fp16")]; + tensor var_48763 = const()[name = tensor("op_48763"), val = tensor([1, 1])]; + tensor var_48765 = const()[name = tensor("op_48765"), val = tensor([1, 1])]; + tensor var_48767_pad_type_0 = const()[name = tensor("op_48767_pad_type_0"), val = tensor("custom")]; + tensor var_48767_pad_0 = const()[name = tensor("op_48767_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_9_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_9_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4044230272)))]; + tensor up_blocks_0_attentions_1_transformer_blocks_9_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_9_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4070444736)))]; + tensor var_48767_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_9_ff_net_0_proj_bias_to_fp16, dilations = var_48765, groups = var_30355, pad = var_48767_pad_0, pad_type = var_48767_pad_type_0, strides = var_48763, weight = up_blocks_0_attentions_1_transformer_blocks_9_ff_net_0_proj_weight_to_fp16, x = input_613_cast_fp16)[name = tensor("op_48767_cast_fp16")]; + tensor var_48768_split_sizes_0 = const()[name = tensor("op_48768_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_48768_axis_0 = const()[name = tensor("op_48768_axis_0"), val = tensor(1)]; + tensor var_48768_cast_fp16_0, tensor var_48768_cast_fp16_1 = split(axis = var_48768_axis_0, split_sizes = var_48768_split_sizes_0, x = var_48767_cast_fp16)[name = tensor("op_48768_cast_fp16")]; + tensor var_48770_mode_0 = const()[name = tensor("op_48770_mode_0"), val = tensor("EXACT")]; + tensor var_48770_cast_fp16 = gelu(mode = var_48770_mode_0, x = var_48768_cast_fp16_1)[name = tensor("op_48770_cast_fp16")]; + tensor input_615_cast_fp16 = mul(x = var_48768_cast_fp16_0, y = var_48770_cast_fp16)[name = tensor("input_615_cast_fp16")]; + tensor var_48774 = const()[name = tensor("op_48774"), val = tensor([1, 1])]; + tensor var_48776 = const()[name = tensor("op_48776"), val = tensor([1, 1])]; + tensor var_48778_pad_type_0 = const()[name = tensor("op_48778_pad_type_0"), val = tensor("custom")]; + tensor var_48778_pad_0 = const()[name = tensor("op_48778_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_transformer_blocks_9_ff_net_2_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_9_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4070465280)))]; + tensor up_blocks_0_attentions_1_transformer_blocks_9_ff_net_2_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_transformer_blocks_9_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4083572544)))]; + tensor var_48778_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_9_ff_net_2_bias_to_fp16, dilations = var_48776, groups = var_30355, pad = var_48778_pad_0, pad_type = var_48778_pad_type_0, strides = var_48774, weight = up_blocks_0_attentions_1_transformer_blocks_9_ff_net_2_weight_to_fp16, x = input_615_cast_fp16)[name = tensor("op_48778_cast_fp16")]; + tensor hidden_states_421_cast_fp16 = add(x = var_48778_cast_fp16, y = inputs_323_cast_fp16)[name = tensor("hidden_states_421_cast_fp16")]; + tensor var_48780 = const()[name = tensor("op_48780"), val = tensor([2, 1280, 32, 32])]; + tensor input_617_cast_fp16 = reshape(shape = var_48780, x = hidden_states_421_cast_fp16)[name = tensor("input_617_cast_fp16")]; + tensor var_48784 = const()[name = tensor("op_48784"), val = tensor([1, 1])]; + tensor var_48786 = const()[name = tensor("op_48786"), val = tensor([1, 1])]; + tensor hidden_states_423_pad_type_0 = const()[name = tensor("hidden_states_423_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_423_pad_0 = const()[name = tensor("hidden_states_423_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_1_proj_out_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_proj_out_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4083575168)))]; + tensor up_blocks_0_attentions_1_proj_out_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_proj_out_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4086852032)))]; + tensor hidden_states_423_cast_fp16 = conv(bias = up_blocks_0_attentions_1_proj_out_bias_to_fp16, dilations = var_48786, groups = var_30355, pad = hidden_states_423_pad_0, pad_type = hidden_states_423_pad_type_0, strides = var_48784, weight = up_blocks_0_attentions_1_proj_out_weight_to_fp16, x = input_617_cast_fp16)[name = tensor("hidden_states_423_cast_fp16")]; + tensor hidden_states_425_cast_fp16 = add(x = hidden_states_423_cast_fp16, y = hidden_states_357_cast_fp16)[name = tensor("hidden_states_425_cast_fp16")]; + tensor input_619_interleave_0 = const()[name = tensor("input_619_interleave_0"), val = tensor(false)]; + tensor input_619_cast_fp16 = concat(axis = var_30355, interleave = input_619_interleave_0, values = (hidden_states_425_cast_fp16, input_115_cast_fp16))[name = tensor("input_619_cast_fp16")]; + tensor reshape_108_shape_0 = const()[name = tensor("reshape_108_shape_0"), val = tensor([2, 32, 60, 32, 32])]; + tensor reshape_108_cast_fp16 = reshape(shape = reshape_108_shape_0, x = input_619_cast_fp16)[name = tensor("reshape_108_cast_fp16")]; + tensor reduce_mean_81_axes_0 = const()[name = tensor("reduce_mean_81_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_81_keep_dims_0 = const()[name = tensor("reduce_mean_81_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_81_cast_fp16 = reduce_mean(axes = reduce_mean_81_axes_0, keep_dims = reduce_mean_81_keep_dims_0, x = reshape_108_cast_fp16)[name = tensor("reduce_mean_81_cast_fp16")]; + tensor sub_54_cast_fp16 = sub(x = reshape_108_cast_fp16, y = reduce_mean_81_cast_fp16)[name = tensor("sub_54_cast_fp16")]; + tensor square_27_cast_fp16 = square(x = sub_54_cast_fp16)[name = tensor("square_27_cast_fp16")]; + tensor reduce_mean_83_axes_0 = const()[name = tensor("reduce_mean_83_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_83_keep_dims_0 = const()[name = tensor("reduce_mean_83_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_83_cast_fp16 = reduce_mean(axes = reduce_mean_83_axes_0, keep_dims = reduce_mean_83_keep_dims_0, x = square_27_cast_fp16)[name = tensor("reduce_mean_83_cast_fp16")]; + tensor add_54_y_0_to_fp16 = const()[name = tensor("add_54_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_54_cast_fp16 = add(x = reduce_mean_83_cast_fp16, y = add_54_y_0_to_fp16)[name = tensor("add_54_cast_fp16")]; + tensor sqrt_27_cast_fp16 = sqrt(x = add_54_cast_fp16)[name = tensor("sqrt_27_cast_fp16")]; + tensor real_div_27_cast_fp16 = real_div(x = sub_54_cast_fp16, y = sqrt_27_cast_fp16)[name = tensor("real_div_27_cast_fp16")]; + tensor reshape_109_shape_0 = const()[name = tensor("reshape_109_shape_0"), val = tensor([2, 1920, 32, 32])]; + tensor reshape_109_cast_fp16 = reshape(shape = reshape_109_shape_0, x = real_div_27_cast_fp16)[name = tensor("reshape_109_cast_fp16")]; + tensor add_55_mean_0_to_fp16 = const()[name = tensor("add_55_mean_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4086854656)))]; + tensor add_55_variance_0_to_fp16 = const()[name = tensor("add_55_variance_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4086858560)))]; + tensor add_55_gamma_0_to_fp16 = const()[name = tensor("add_55_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4086862464)))]; + tensor add_55_beta_0_to_fp16 = const()[name = tensor("add_55_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4086866368)))]; + tensor add_55_epsilon_0_to_fp16 = const()[name = tensor("add_55_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_55_cast_fp16 = batch_norm(beta = add_55_beta_0_to_fp16, epsilon = add_55_epsilon_0_to_fp16, gamma = add_55_gamma_0_to_fp16, mean = add_55_mean_0_to_fp16, variance = add_55_variance_0_to_fp16, x = reshape_109_cast_fp16)[name = tensor("add_55_cast_fp16")]; + tensor input_623_cast_fp16 = silu(x = add_55_cast_fp16)[name = tensor("input_623_cast_fp16")]; + tensor var_48804 = const()[name = tensor("op_48804"), val = tensor([1, 1])]; + tensor var_48806 = const()[name = tensor("op_48806"), val = tensor([1, 1])]; + tensor hidden_states_427_pad_type_0 = const()[name = tensor("hidden_states_427_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_427_pad_0 = const()[name = tensor("hidden_states_427_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor up_blocks_0_resnets_2_conv1_weight_to_fp16 = const()[name = tensor("up_blocks_0_resnets_2_conv1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4086870272)))]; + tensor up_blocks_0_resnets_2_conv1_bias_to_fp16 = const()[name = tensor("up_blocks_0_resnets_2_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4131107136)))]; + tensor hidden_states_427_cast_fp16 = conv(bias = up_blocks_0_resnets_2_conv1_bias_to_fp16, dilations = var_48806, groups = var_30355, pad = hidden_states_427_pad_0, pad_type = hidden_states_427_pad_type_0, strides = var_48804, weight = up_blocks_0_resnets_2_conv1_weight_to_fp16, x = input_623_cast_fp16)[name = tensor("hidden_states_427_cast_fp16")]; + tensor var_48812 = const()[name = tensor("op_48812"), val = tensor([1, 1])]; + tensor var_48814 = const()[name = tensor("op_48814"), val = tensor([1, 1])]; + tensor temb_21_pad_type_0 = const()[name = tensor("temb_21_pad_type_0"), val = tensor("custom")]; + tensor temb_21_pad_0 = const()[name = tensor("temb_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_resnets_2_time_emb_proj_weight_to_fp16 = const()[name = tensor("up_blocks_0_resnets_2_time_emb_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4131109760)))]; + tensor up_blocks_0_resnets_2_time_emb_proj_bias_to_fp16 = const()[name = tensor("up_blocks_0_resnets_2_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4134386624)))]; + tensor temb_21_cast_fp16 = conv(bias = up_blocks_0_resnets_2_time_emb_proj_bias_to_fp16, dilations = var_48814, groups = var_30355, pad = temb_21_pad_0, pad_type = temb_21_pad_type_0, strides = var_48812, weight = up_blocks_0_resnets_2_time_emb_proj_weight_to_fp16, x = input_21_cast_fp16)[name = tensor("temb_21_cast_fp16")]; + tensor input_627_cast_fp16 = add(x = hidden_states_427_cast_fp16, y = temb_21_cast_fp16)[name = tensor("input_627_cast_fp16")]; + tensor reshape_112_shape_0 = const()[name = tensor("reshape_112_shape_0"), val = tensor([2, 32, 40, 32, 32])]; + tensor reshape_112_cast_fp16 = reshape(shape = reshape_112_shape_0, x = input_627_cast_fp16)[name = tensor("reshape_112_cast_fp16")]; + tensor reduce_mean_84_axes_0 = const()[name = tensor("reduce_mean_84_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_84_keep_dims_0 = const()[name = tensor("reduce_mean_84_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_84_cast_fp16 = reduce_mean(axes = reduce_mean_84_axes_0, keep_dims = reduce_mean_84_keep_dims_0, x = reshape_112_cast_fp16)[name = tensor("reduce_mean_84_cast_fp16")]; + tensor sub_56_cast_fp16 = sub(x = reshape_112_cast_fp16, y = reduce_mean_84_cast_fp16)[name = tensor("sub_56_cast_fp16")]; + tensor square_28_cast_fp16 = square(x = sub_56_cast_fp16)[name = tensor("square_28_cast_fp16")]; + tensor reduce_mean_86_axes_0 = const()[name = tensor("reduce_mean_86_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_86_keep_dims_0 = const()[name = tensor("reduce_mean_86_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_86_cast_fp16 = reduce_mean(axes = reduce_mean_86_axes_0, keep_dims = reduce_mean_86_keep_dims_0, x = square_28_cast_fp16)[name = tensor("reduce_mean_86_cast_fp16")]; + tensor add_56_y_0_to_fp16 = const()[name = tensor("add_56_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_56_cast_fp16 = add(x = reduce_mean_86_cast_fp16, y = add_56_y_0_to_fp16)[name = tensor("add_56_cast_fp16")]; + tensor sqrt_28_cast_fp16 = sqrt(x = add_56_cast_fp16)[name = tensor("sqrt_28_cast_fp16")]; + tensor real_div_28_cast_fp16 = real_div(x = sub_56_cast_fp16, y = sqrt_28_cast_fp16)[name = tensor("real_div_28_cast_fp16")]; + tensor reshape_113_shape_0 = const()[name = tensor("reshape_113_shape_0"), val = tensor([2, 1280, 32, 32])]; + tensor reshape_113_cast_fp16 = reshape(shape = reshape_113_shape_0, x = real_div_28_cast_fp16)[name = tensor("reshape_113_cast_fp16")]; + tensor add_57_gamma_0_to_fp16 = const()[name = tensor("add_57_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4134389248)))]; + tensor add_57_beta_0_to_fp16 = const()[name = tensor("add_57_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4134391872)))]; + tensor add_57_epsilon_0_to_fp16 = const()[name = tensor("add_57_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_57_cast_fp16 = batch_norm(beta = add_57_beta_0_to_fp16, epsilon = add_57_epsilon_0_to_fp16, gamma = add_57_gamma_0_to_fp16, mean = add_23_mean_0_to_fp16, variance = add_23_variance_0_to_fp16, x = reshape_113_cast_fp16)[name = tensor("add_57_cast_fp16")]; + tensor input_631_cast_fp16 = silu(x = add_57_cast_fp16)[name = tensor("input_631_cast_fp16")]; + tensor var_48824 = const()[name = tensor("op_48824"), val = tensor([1, 1])]; + tensor var_48826 = const()[name = tensor("op_48826"), val = tensor([1, 1])]; + tensor hidden_states_429_pad_type_0 = const()[name = tensor("hidden_states_429_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_429_pad_0 = const()[name = tensor("hidden_states_429_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor up_blocks_0_resnets_2_conv2_weight_to_fp16 = const()[name = tensor("up_blocks_0_resnets_2_conv2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4134394496)))]; + tensor up_blocks_0_resnets_2_conv2_bias_to_fp16 = const()[name = tensor("up_blocks_0_resnets_2_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4163885760)))]; + tensor hidden_states_429_cast_fp16 = conv(bias = up_blocks_0_resnets_2_conv2_bias_to_fp16, dilations = var_48826, groups = var_30355, pad = hidden_states_429_pad_0, pad_type = hidden_states_429_pad_type_0, strides = var_48824, weight = up_blocks_0_resnets_2_conv2_weight_to_fp16, x = input_631_cast_fp16)[name = tensor("hidden_states_429_cast_fp16")]; + tensor var_48831 = const()[name = tensor("op_48831"), val = tensor([1, 1])]; + tensor var_48833 = const()[name = tensor("op_48833"), val = tensor([1, 1])]; + tensor x_9_pad_type_0 = const()[name = tensor("x_9_pad_type_0"), val = tensor("custom")]; + tensor x_9_pad_0 = const()[name = tensor("x_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_resnets_2_conv_shortcut_weight_to_fp16 = const()[name = tensor("up_blocks_0_resnets_2_conv_shortcut_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4163888384)))]; + tensor up_blocks_0_resnets_2_conv_shortcut_bias_to_fp16 = const()[name = tensor("up_blocks_0_resnets_2_conv_shortcut_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4168803648)))]; + tensor x_9_cast_fp16 = conv(bias = up_blocks_0_resnets_2_conv_shortcut_bias_to_fp16, dilations = var_48833, groups = var_30355, pad = x_9_pad_0, pad_type = x_9_pad_type_0, strides = var_48831, weight = up_blocks_0_resnets_2_conv_shortcut_weight_to_fp16, x = input_619_cast_fp16)[name = tensor("x_9_cast_fp16")]; + tensor hidden_states_431_cast_fp16 = add(x = x_9_cast_fp16, y = hidden_states_429_cast_fp16)[name = tensor("hidden_states_431_cast_fp16")]; + tensor reshape_116_shape_0 = const()[name = tensor("reshape_116_shape_0"), val = tensor([2, 32, 40, 32, 32])]; + tensor reshape_116_cast_fp16 = reshape(shape = reshape_116_shape_0, x = hidden_states_431_cast_fp16)[name = tensor("reshape_116_cast_fp16")]; + tensor reduce_mean_87_axes_0 = const()[name = tensor("reduce_mean_87_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_87_keep_dims_0 = const()[name = tensor("reduce_mean_87_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_87_cast_fp16 = reduce_mean(axes = reduce_mean_87_axes_0, keep_dims = reduce_mean_87_keep_dims_0, x = reshape_116_cast_fp16)[name = tensor("reduce_mean_87_cast_fp16")]; + tensor sub_58_cast_fp16 = sub(x = reshape_116_cast_fp16, y = reduce_mean_87_cast_fp16)[name = tensor("sub_58_cast_fp16")]; + tensor square_29_cast_fp16 = square(x = sub_58_cast_fp16)[name = tensor("square_29_cast_fp16")]; + tensor reduce_mean_89_axes_0 = const()[name = tensor("reduce_mean_89_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_89_keep_dims_0 = const()[name = tensor("reduce_mean_89_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_89_cast_fp16 = reduce_mean(axes = reduce_mean_89_axes_0, keep_dims = reduce_mean_89_keep_dims_0, x = square_29_cast_fp16)[name = tensor("reduce_mean_89_cast_fp16")]; + tensor add_58_y_0_to_fp16 = const()[name = tensor("add_58_y_0_to_fp16"), val = tensor(0x1.1p-20)]; + tensor add_58_cast_fp16 = add(x = reduce_mean_89_cast_fp16, y = add_58_y_0_to_fp16)[name = tensor("add_58_cast_fp16")]; + tensor sqrt_29_cast_fp16 = sqrt(x = add_58_cast_fp16)[name = tensor("sqrt_29_cast_fp16")]; + tensor real_div_29_cast_fp16 = real_div(x = sub_58_cast_fp16, y = sqrt_29_cast_fp16)[name = tensor("real_div_29_cast_fp16")]; + tensor reshape_117_shape_0 = const()[name = tensor("reshape_117_shape_0"), val = tensor([2, 1280, 32, 32])]; + tensor reshape_117_cast_fp16 = reshape(shape = reshape_117_shape_0, x = real_div_29_cast_fp16)[name = tensor("reshape_117_cast_fp16")]; + tensor add_59_gamma_0_to_fp16 = const()[name = tensor("add_59_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4168806272)))]; + tensor add_59_beta_0_to_fp16 = const()[name = tensor("add_59_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4168808896)))]; + tensor add_59_epsilon_0_to_fp16 = const()[name = tensor("add_59_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_59_cast_fp16 = batch_norm(beta = add_59_beta_0_to_fp16, epsilon = add_59_epsilon_0_to_fp16, gamma = add_59_gamma_0_to_fp16, mean = add_23_mean_0_to_fp16, variance = add_23_variance_0_to_fp16, x = reshape_117_cast_fp16)[name = tensor("add_59_cast_fp16")]; + tensor var_48871 = const()[name = tensor("op_48871"), val = tensor([1, 1])]; + tensor var_48873 = const()[name = tensor("op_48873"), val = tensor([1, 1])]; + tensor hidden_states_433_pad_type_0 = const()[name = tensor("hidden_states_433_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_433_pad_0 = const()[name = tensor("hidden_states_433_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_proj_in_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_proj_in_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4168811520)))]; + tensor up_blocks_0_attentions_2_proj_in_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_proj_in_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4172088384)))]; + tensor hidden_states_433_cast_fp16 = conv(bias = up_blocks_0_attentions_2_proj_in_bias_to_fp16, dilations = var_48873, groups = var_30355, pad = hidden_states_433_pad_0, pad_type = hidden_states_433_pad_type_0, strides = var_48871, weight = up_blocks_0_attentions_2_proj_in_weight_to_fp16, x = add_59_cast_fp16)[name = tensor("hidden_states_433_cast_fp16")]; + tensor var_48878 = const()[name = tensor("op_48878"), val = tensor([2, 1280, 1, 1024])]; + tensor inputs_325_cast_fp16 = reshape(shape = var_48878, x = hidden_states_433_cast_fp16)[name = tensor("inputs_325_cast_fp16")]; + tensor hidden_states_435_axes_0 = const()[name = tensor("hidden_states_435_axes_0"), val = tensor([1])]; + tensor hidden_states_435_gamma_0_to_fp16 = const()[name = tensor("hidden_states_435_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4172091008)))]; + tensor hidden_states_435_beta_0_to_fp16 = const()[name = tensor("hidden_states_435_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4172093632)))]; + tensor var_48894_to_fp16 = const()[name = tensor("op_48894_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_435_cast_fp16 = layer_norm(axes = hidden_states_435_axes_0, beta = hidden_states_435_beta_0_to_fp16, epsilon = var_48894_to_fp16, gamma = hidden_states_435_gamma_0_to_fp16, x = inputs_325_cast_fp16)[name = tensor("hidden_states_435_cast_fp16")]; + tensor var_48909 = const()[name = tensor("op_48909"), val = tensor([1, 1])]; + tensor var_48911 = const()[name = tensor("op_48911"), val = tensor([1, 1])]; + tensor q_217_pad_type_0 = const()[name = tensor("q_217_pad_type_0"), val = tensor("custom")]; + tensor q_217_pad_0 = const()[name = tensor("q_217_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_0_attn1_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_0_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4172096256)))]; + tensor q_217_cast_fp16 = conv(dilations = var_48911, groups = var_30355, pad = q_217_pad_0, pad_type = q_217_pad_type_0, strides = var_48909, weight = up_blocks_0_attentions_2_transformer_blocks_0_attn1_to_q_weight_to_fp16, x = hidden_states_435_cast_fp16)[name = tensor("q_217_cast_fp16")]; + tensor var_48915 = const()[name = tensor("op_48915"), val = tensor([1, 1])]; + tensor var_48917 = const()[name = tensor("op_48917"), val = tensor([1, 1])]; + tensor k_433_pad_type_0 = const()[name = tensor("k_433_pad_type_0"), val = tensor("custom")]; + tensor k_433_pad_0 = const()[name = tensor("k_433_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_0_attn1_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_0_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4175373120)))]; + tensor k_433_cast_fp16 = conv(dilations = var_48917, groups = var_30355, pad = k_433_pad_0, pad_type = k_433_pad_type_0, strides = var_48915, weight = up_blocks_0_attentions_2_transformer_blocks_0_attn1_to_k_weight_to_fp16, x = hidden_states_435_cast_fp16)[name = tensor("k_433_cast_fp16")]; + tensor var_48921 = const()[name = tensor("op_48921"), val = tensor([1, 1])]; + tensor var_48923 = const()[name = tensor("op_48923"), val = tensor([1, 1])]; + tensor v_217_pad_type_0 = const()[name = tensor("v_217_pad_type_0"), val = tensor("custom")]; + tensor v_217_pad_0 = const()[name = tensor("v_217_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_0_attn1_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_0_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4178649984)))]; + tensor v_217_cast_fp16 = conv(dilations = var_48923, groups = var_30355, pad = v_217_pad_0, pad_type = v_217_pad_type_0, strides = var_48921, weight = up_blocks_0_attentions_2_transformer_blocks_0_attn1_to_v_weight_to_fp16, x = hidden_states_435_cast_fp16)[name = tensor("v_217_cast_fp16")]; + tensor var_48927_begin_0 = const()[name = tensor("op_48927_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_48927_end_0 = const()[name = tensor("op_48927_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_48927_end_mask_0 = const()[name = tensor("op_48927_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48927_cast_fp16 = slice_by_index(begin = var_48927_begin_0, end = var_48927_end_0, end_mask = var_48927_end_mask_0, x = q_217_cast_fp16)[name = tensor("op_48927_cast_fp16")]; + tensor var_48931_begin_0 = const()[name = tensor("op_48931_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_48931_end_0 = const()[name = tensor("op_48931_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_48931_end_mask_0 = const()[name = tensor("op_48931_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48931_cast_fp16 = slice_by_index(begin = var_48931_begin_0, end = var_48931_end_0, end_mask = var_48931_end_mask_0, x = q_217_cast_fp16)[name = tensor("op_48931_cast_fp16")]; + tensor var_48935_begin_0 = const()[name = tensor("op_48935_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_48935_end_0 = const()[name = tensor("op_48935_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_48935_end_mask_0 = const()[name = tensor("op_48935_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48935_cast_fp16 = slice_by_index(begin = var_48935_begin_0, end = var_48935_end_0, end_mask = var_48935_end_mask_0, x = q_217_cast_fp16)[name = tensor("op_48935_cast_fp16")]; + tensor var_48939_begin_0 = const()[name = tensor("op_48939_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_48939_end_0 = const()[name = tensor("op_48939_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_48939_end_mask_0 = const()[name = tensor("op_48939_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48939_cast_fp16 = slice_by_index(begin = var_48939_begin_0, end = var_48939_end_0, end_mask = var_48939_end_mask_0, x = q_217_cast_fp16)[name = tensor("op_48939_cast_fp16")]; + tensor var_48943_begin_0 = const()[name = tensor("op_48943_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_48943_end_0 = const()[name = tensor("op_48943_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_48943_end_mask_0 = const()[name = tensor("op_48943_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48943_cast_fp16 = slice_by_index(begin = var_48943_begin_0, end = var_48943_end_0, end_mask = var_48943_end_mask_0, x = q_217_cast_fp16)[name = tensor("op_48943_cast_fp16")]; + tensor var_48947_begin_0 = const()[name = tensor("op_48947_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_48947_end_0 = const()[name = tensor("op_48947_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_48947_end_mask_0 = const()[name = tensor("op_48947_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48947_cast_fp16 = slice_by_index(begin = var_48947_begin_0, end = var_48947_end_0, end_mask = var_48947_end_mask_0, x = q_217_cast_fp16)[name = tensor("op_48947_cast_fp16")]; + tensor var_48951_begin_0 = const()[name = tensor("op_48951_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_48951_end_0 = const()[name = tensor("op_48951_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_48951_end_mask_0 = const()[name = tensor("op_48951_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48951_cast_fp16 = slice_by_index(begin = var_48951_begin_0, end = var_48951_end_0, end_mask = var_48951_end_mask_0, x = q_217_cast_fp16)[name = tensor("op_48951_cast_fp16")]; + tensor var_48955_begin_0 = const()[name = tensor("op_48955_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_48955_end_0 = const()[name = tensor("op_48955_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_48955_end_mask_0 = const()[name = tensor("op_48955_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48955_cast_fp16 = slice_by_index(begin = var_48955_begin_0, end = var_48955_end_0, end_mask = var_48955_end_mask_0, x = q_217_cast_fp16)[name = tensor("op_48955_cast_fp16")]; + tensor var_48959_begin_0 = const()[name = tensor("op_48959_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_48959_end_0 = const()[name = tensor("op_48959_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_48959_end_mask_0 = const()[name = tensor("op_48959_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48959_cast_fp16 = slice_by_index(begin = var_48959_begin_0, end = var_48959_end_0, end_mask = var_48959_end_mask_0, x = q_217_cast_fp16)[name = tensor("op_48959_cast_fp16")]; + tensor var_48963_begin_0 = const()[name = tensor("op_48963_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_48963_end_0 = const()[name = tensor("op_48963_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_48963_end_mask_0 = const()[name = tensor("op_48963_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48963_cast_fp16 = slice_by_index(begin = var_48963_begin_0, end = var_48963_end_0, end_mask = var_48963_end_mask_0, x = q_217_cast_fp16)[name = tensor("op_48963_cast_fp16")]; + tensor var_48967_begin_0 = const()[name = tensor("op_48967_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_48967_end_0 = const()[name = tensor("op_48967_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_48967_end_mask_0 = const()[name = tensor("op_48967_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48967_cast_fp16 = slice_by_index(begin = var_48967_begin_0, end = var_48967_end_0, end_mask = var_48967_end_mask_0, x = q_217_cast_fp16)[name = tensor("op_48967_cast_fp16")]; + tensor var_48971_begin_0 = const()[name = tensor("op_48971_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_48971_end_0 = const()[name = tensor("op_48971_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_48971_end_mask_0 = const()[name = tensor("op_48971_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48971_cast_fp16 = slice_by_index(begin = var_48971_begin_0, end = var_48971_end_0, end_mask = var_48971_end_mask_0, x = q_217_cast_fp16)[name = tensor("op_48971_cast_fp16")]; + tensor var_48975_begin_0 = const()[name = tensor("op_48975_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_48975_end_0 = const()[name = tensor("op_48975_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_48975_end_mask_0 = const()[name = tensor("op_48975_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48975_cast_fp16 = slice_by_index(begin = var_48975_begin_0, end = var_48975_end_0, end_mask = var_48975_end_mask_0, x = q_217_cast_fp16)[name = tensor("op_48975_cast_fp16")]; + tensor var_48979_begin_0 = const()[name = tensor("op_48979_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_48979_end_0 = const()[name = tensor("op_48979_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_48979_end_mask_0 = const()[name = tensor("op_48979_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48979_cast_fp16 = slice_by_index(begin = var_48979_begin_0, end = var_48979_end_0, end_mask = var_48979_end_mask_0, x = q_217_cast_fp16)[name = tensor("op_48979_cast_fp16")]; + tensor var_48983_begin_0 = const()[name = tensor("op_48983_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_48983_end_0 = const()[name = tensor("op_48983_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_48983_end_mask_0 = const()[name = tensor("op_48983_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48983_cast_fp16 = slice_by_index(begin = var_48983_begin_0, end = var_48983_end_0, end_mask = var_48983_end_mask_0, x = q_217_cast_fp16)[name = tensor("op_48983_cast_fp16")]; + tensor var_48987_begin_0 = const()[name = tensor("op_48987_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_48987_end_0 = const()[name = tensor("op_48987_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_48987_end_mask_0 = const()[name = tensor("op_48987_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48987_cast_fp16 = slice_by_index(begin = var_48987_begin_0, end = var_48987_end_0, end_mask = var_48987_end_mask_0, x = q_217_cast_fp16)[name = tensor("op_48987_cast_fp16")]; + tensor var_48991_begin_0 = const()[name = tensor("op_48991_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_48991_end_0 = const()[name = tensor("op_48991_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_48991_end_mask_0 = const()[name = tensor("op_48991_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48991_cast_fp16 = slice_by_index(begin = var_48991_begin_0, end = var_48991_end_0, end_mask = var_48991_end_mask_0, x = q_217_cast_fp16)[name = tensor("op_48991_cast_fp16")]; + tensor var_48995_begin_0 = const()[name = tensor("op_48995_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_48995_end_0 = const()[name = tensor("op_48995_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_48995_end_mask_0 = const()[name = tensor("op_48995_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48995_cast_fp16 = slice_by_index(begin = var_48995_begin_0, end = var_48995_end_0, end_mask = var_48995_end_mask_0, x = q_217_cast_fp16)[name = tensor("op_48995_cast_fp16")]; + tensor var_48999_begin_0 = const()[name = tensor("op_48999_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_48999_end_0 = const()[name = tensor("op_48999_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_48999_end_mask_0 = const()[name = tensor("op_48999_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48999_cast_fp16 = slice_by_index(begin = var_48999_begin_0, end = var_48999_end_0, end_mask = var_48999_end_mask_0, x = q_217_cast_fp16)[name = tensor("op_48999_cast_fp16")]; + tensor var_49003_begin_0 = const()[name = tensor("op_49003_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_49003_end_0 = const()[name = tensor("op_49003_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_49003_end_mask_0 = const()[name = tensor("op_49003_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49003_cast_fp16 = slice_by_index(begin = var_49003_begin_0, end = var_49003_end_0, end_mask = var_49003_end_mask_0, x = q_217_cast_fp16)[name = tensor("op_49003_cast_fp16")]; + tensor k_435_perm_0 = const()[name = tensor("k_435_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_49010_begin_0 = const()[name = tensor("op_49010_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_49010_end_0 = const()[name = tensor("op_49010_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_49010_end_mask_0 = const()[name = tensor("op_49010_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_435_cast_fp16 = transpose(perm = k_435_perm_0, x = k_433_cast_fp16)[name = tensor("transpose_31")]; + tensor var_49010_cast_fp16 = slice_by_index(begin = var_49010_begin_0, end = var_49010_end_0, end_mask = var_49010_end_mask_0, x = k_435_cast_fp16)[name = tensor("op_49010_cast_fp16")]; + tensor var_49014_begin_0 = const()[name = tensor("op_49014_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_49014_end_0 = const()[name = tensor("op_49014_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_49014_end_mask_0 = const()[name = tensor("op_49014_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49014_cast_fp16 = slice_by_index(begin = var_49014_begin_0, end = var_49014_end_0, end_mask = var_49014_end_mask_0, x = k_435_cast_fp16)[name = tensor("op_49014_cast_fp16")]; + tensor var_49018_begin_0 = const()[name = tensor("op_49018_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_49018_end_0 = const()[name = tensor("op_49018_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_49018_end_mask_0 = const()[name = tensor("op_49018_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49018_cast_fp16 = slice_by_index(begin = var_49018_begin_0, end = var_49018_end_0, end_mask = var_49018_end_mask_0, x = k_435_cast_fp16)[name = tensor("op_49018_cast_fp16")]; + tensor var_49022_begin_0 = const()[name = tensor("op_49022_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_49022_end_0 = const()[name = tensor("op_49022_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_49022_end_mask_0 = const()[name = tensor("op_49022_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49022_cast_fp16 = slice_by_index(begin = var_49022_begin_0, end = var_49022_end_0, end_mask = var_49022_end_mask_0, x = k_435_cast_fp16)[name = tensor("op_49022_cast_fp16")]; + tensor var_49026_begin_0 = const()[name = tensor("op_49026_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_49026_end_0 = const()[name = tensor("op_49026_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_49026_end_mask_0 = const()[name = tensor("op_49026_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49026_cast_fp16 = slice_by_index(begin = var_49026_begin_0, end = var_49026_end_0, end_mask = var_49026_end_mask_0, x = k_435_cast_fp16)[name = tensor("op_49026_cast_fp16")]; + tensor var_49030_begin_0 = const()[name = tensor("op_49030_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_49030_end_0 = const()[name = tensor("op_49030_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_49030_end_mask_0 = const()[name = tensor("op_49030_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49030_cast_fp16 = slice_by_index(begin = var_49030_begin_0, end = var_49030_end_0, end_mask = var_49030_end_mask_0, x = k_435_cast_fp16)[name = tensor("op_49030_cast_fp16")]; + tensor var_49034_begin_0 = const()[name = tensor("op_49034_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_49034_end_0 = const()[name = tensor("op_49034_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_49034_end_mask_0 = const()[name = tensor("op_49034_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49034_cast_fp16 = slice_by_index(begin = var_49034_begin_0, end = var_49034_end_0, end_mask = var_49034_end_mask_0, x = k_435_cast_fp16)[name = tensor("op_49034_cast_fp16")]; + tensor var_49038_begin_0 = const()[name = tensor("op_49038_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_49038_end_0 = const()[name = tensor("op_49038_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_49038_end_mask_0 = const()[name = tensor("op_49038_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49038_cast_fp16 = slice_by_index(begin = var_49038_begin_0, end = var_49038_end_0, end_mask = var_49038_end_mask_0, x = k_435_cast_fp16)[name = tensor("op_49038_cast_fp16")]; + tensor var_49042_begin_0 = const()[name = tensor("op_49042_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_49042_end_0 = const()[name = tensor("op_49042_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_49042_end_mask_0 = const()[name = tensor("op_49042_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49042_cast_fp16 = slice_by_index(begin = var_49042_begin_0, end = var_49042_end_0, end_mask = var_49042_end_mask_0, x = k_435_cast_fp16)[name = tensor("op_49042_cast_fp16")]; + tensor var_49046_begin_0 = const()[name = tensor("op_49046_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_49046_end_0 = const()[name = tensor("op_49046_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_49046_end_mask_0 = const()[name = tensor("op_49046_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49046_cast_fp16 = slice_by_index(begin = var_49046_begin_0, end = var_49046_end_0, end_mask = var_49046_end_mask_0, x = k_435_cast_fp16)[name = tensor("op_49046_cast_fp16")]; + tensor var_49050_begin_0 = const()[name = tensor("op_49050_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_49050_end_0 = const()[name = tensor("op_49050_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_49050_end_mask_0 = const()[name = tensor("op_49050_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49050_cast_fp16 = slice_by_index(begin = var_49050_begin_0, end = var_49050_end_0, end_mask = var_49050_end_mask_0, x = k_435_cast_fp16)[name = tensor("op_49050_cast_fp16")]; + tensor var_49054_begin_0 = const()[name = tensor("op_49054_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_49054_end_0 = const()[name = tensor("op_49054_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_49054_end_mask_0 = const()[name = tensor("op_49054_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49054_cast_fp16 = slice_by_index(begin = var_49054_begin_0, end = var_49054_end_0, end_mask = var_49054_end_mask_0, x = k_435_cast_fp16)[name = tensor("op_49054_cast_fp16")]; + tensor var_49058_begin_0 = const()[name = tensor("op_49058_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_49058_end_0 = const()[name = tensor("op_49058_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_49058_end_mask_0 = const()[name = tensor("op_49058_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49058_cast_fp16 = slice_by_index(begin = var_49058_begin_0, end = var_49058_end_0, end_mask = var_49058_end_mask_0, x = k_435_cast_fp16)[name = tensor("op_49058_cast_fp16")]; + tensor var_49062_begin_0 = const()[name = tensor("op_49062_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_49062_end_0 = const()[name = tensor("op_49062_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_49062_end_mask_0 = const()[name = tensor("op_49062_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49062_cast_fp16 = slice_by_index(begin = var_49062_begin_0, end = var_49062_end_0, end_mask = var_49062_end_mask_0, x = k_435_cast_fp16)[name = tensor("op_49062_cast_fp16")]; + tensor var_49066_begin_0 = const()[name = tensor("op_49066_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_49066_end_0 = const()[name = tensor("op_49066_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_49066_end_mask_0 = const()[name = tensor("op_49066_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49066_cast_fp16 = slice_by_index(begin = var_49066_begin_0, end = var_49066_end_0, end_mask = var_49066_end_mask_0, x = k_435_cast_fp16)[name = tensor("op_49066_cast_fp16")]; + tensor var_49070_begin_0 = const()[name = tensor("op_49070_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_49070_end_0 = const()[name = tensor("op_49070_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_49070_end_mask_0 = const()[name = tensor("op_49070_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49070_cast_fp16 = slice_by_index(begin = var_49070_begin_0, end = var_49070_end_0, end_mask = var_49070_end_mask_0, x = k_435_cast_fp16)[name = tensor("op_49070_cast_fp16")]; + tensor var_49074_begin_0 = const()[name = tensor("op_49074_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_49074_end_0 = const()[name = tensor("op_49074_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_49074_end_mask_0 = const()[name = tensor("op_49074_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49074_cast_fp16 = slice_by_index(begin = var_49074_begin_0, end = var_49074_end_0, end_mask = var_49074_end_mask_0, x = k_435_cast_fp16)[name = tensor("op_49074_cast_fp16")]; + tensor var_49078_begin_0 = const()[name = tensor("op_49078_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_49078_end_0 = const()[name = tensor("op_49078_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_49078_end_mask_0 = const()[name = tensor("op_49078_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49078_cast_fp16 = slice_by_index(begin = var_49078_begin_0, end = var_49078_end_0, end_mask = var_49078_end_mask_0, x = k_435_cast_fp16)[name = tensor("op_49078_cast_fp16")]; + tensor var_49082_begin_0 = const()[name = tensor("op_49082_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_49082_end_0 = const()[name = tensor("op_49082_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_49082_end_mask_0 = const()[name = tensor("op_49082_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49082_cast_fp16 = slice_by_index(begin = var_49082_begin_0, end = var_49082_end_0, end_mask = var_49082_end_mask_0, x = k_435_cast_fp16)[name = tensor("op_49082_cast_fp16")]; + tensor var_49086_begin_0 = const()[name = tensor("op_49086_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_49086_end_0 = const()[name = tensor("op_49086_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_49086_end_mask_0 = const()[name = tensor("op_49086_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49086_cast_fp16 = slice_by_index(begin = var_49086_begin_0, end = var_49086_end_0, end_mask = var_49086_end_mask_0, x = k_435_cast_fp16)[name = tensor("op_49086_cast_fp16")]; + tensor var_49088_begin_0 = const()[name = tensor("op_49088_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_49088_end_0 = const()[name = tensor("op_49088_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_49088_end_mask_0 = const()[name = tensor("op_49088_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49088_cast_fp16 = slice_by_index(begin = var_49088_begin_0, end = var_49088_end_0, end_mask = var_49088_end_mask_0, x = v_217_cast_fp16)[name = tensor("op_49088_cast_fp16")]; + tensor var_49092_begin_0 = const()[name = tensor("op_49092_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_49092_end_0 = const()[name = tensor("op_49092_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_49092_end_mask_0 = const()[name = tensor("op_49092_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49092_cast_fp16 = slice_by_index(begin = var_49092_begin_0, end = var_49092_end_0, end_mask = var_49092_end_mask_0, x = v_217_cast_fp16)[name = tensor("op_49092_cast_fp16")]; + tensor var_49096_begin_0 = const()[name = tensor("op_49096_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_49096_end_0 = const()[name = tensor("op_49096_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_49096_end_mask_0 = const()[name = tensor("op_49096_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49096_cast_fp16 = slice_by_index(begin = var_49096_begin_0, end = var_49096_end_0, end_mask = var_49096_end_mask_0, x = v_217_cast_fp16)[name = tensor("op_49096_cast_fp16")]; + tensor var_49100_begin_0 = const()[name = tensor("op_49100_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_49100_end_0 = const()[name = tensor("op_49100_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_49100_end_mask_0 = const()[name = tensor("op_49100_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49100_cast_fp16 = slice_by_index(begin = var_49100_begin_0, end = var_49100_end_0, end_mask = var_49100_end_mask_0, x = v_217_cast_fp16)[name = tensor("op_49100_cast_fp16")]; + tensor var_49104_begin_0 = const()[name = tensor("op_49104_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_49104_end_0 = const()[name = tensor("op_49104_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_49104_end_mask_0 = const()[name = tensor("op_49104_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49104_cast_fp16 = slice_by_index(begin = var_49104_begin_0, end = var_49104_end_0, end_mask = var_49104_end_mask_0, x = v_217_cast_fp16)[name = tensor("op_49104_cast_fp16")]; + tensor var_49108_begin_0 = const()[name = tensor("op_49108_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_49108_end_0 = const()[name = tensor("op_49108_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_49108_end_mask_0 = const()[name = tensor("op_49108_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49108_cast_fp16 = slice_by_index(begin = var_49108_begin_0, end = var_49108_end_0, end_mask = var_49108_end_mask_0, x = v_217_cast_fp16)[name = tensor("op_49108_cast_fp16")]; + tensor var_49112_begin_0 = const()[name = tensor("op_49112_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_49112_end_0 = const()[name = tensor("op_49112_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_49112_end_mask_0 = const()[name = tensor("op_49112_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49112_cast_fp16 = slice_by_index(begin = var_49112_begin_0, end = var_49112_end_0, end_mask = var_49112_end_mask_0, x = v_217_cast_fp16)[name = tensor("op_49112_cast_fp16")]; + tensor var_49116_begin_0 = const()[name = tensor("op_49116_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_49116_end_0 = const()[name = tensor("op_49116_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_49116_end_mask_0 = const()[name = tensor("op_49116_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49116_cast_fp16 = slice_by_index(begin = var_49116_begin_0, end = var_49116_end_0, end_mask = var_49116_end_mask_0, x = v_217_cast_fp16)[name = tensor("op_49116_cast_fp16")]; + tensor var_49120_begin_0 = const()[name = tensor("op_49120_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_49120_end_0 = const()[name = tensor("op_49120_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_49120_end_mask_0 = const()[name = tensor("op_49120_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49120_cast_fp16 = slice_by_index(begin = var_49120_begin_0, end = var_49120_end_0, end_mask = var_49120_end_mask_0, x = v_217_cast_fp16)[name = tensor("op_49120_cast_fp16")]; + tensor var_49124_begin_0 = const()[name = tensor("op_49124_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_49124_end_0 = const()[name = tensor("op_49124_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_49124_end_mask_0 = const()[name = tensor("op_49124_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49124_cast_fp16 = slice_by_index(begin = var_49124_begin_0, end = var_49124_end_0, end_mask = var_49124_end_mask_0, x = v_217_cast_fp16)[name = tensor("op_49124_cast_fp16")]; + tensor var_49128_begin_0 = const()[name = tensor("op_49128_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_49128_end_0 = const()[name = tensor("op_49128_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_49128_end_mask_0 = const()[name = tensor("op_49128_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49128_cast_fp16 = slice_by_index(begin = var_49128_begin_0, end = var_49128_end_0, end_mask = var_49128_end_mask_0, x = v_217_cast_fp16)[name = tensor("op_49128_cast_fp16")]; + tensor var_49132_begin_0 = const()[name = tensor("op_49132_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_49132_end_0 = const()[name = tensor("op_49132_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_49132_end_mask_0 = const()[name = tensor("op_49132_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49132_cast_fp16 = slice_by_index(begin = var_49132_begin_0, end = var_49132_end_0, end_mask = var_49132_end_mask_0, x = v_217_cast_fp16)[name = tensor("op_49132_cast_fp16")]; + tensor var_49136_begin_0 = const()[name = tensor("op_49136_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_49136_end_0 = const()[name = tensor("op_49136_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_49136_end_mask_0 = const()[name = tensor("op_49136_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49136_cast_fp16 = slice_by_index(begin = var_49136_begin_0, end = var_49136_end_0, end_mask = var_49136_end_mask_0, x = v_217_cast_fp16)[name = tensor("op_49136_cast_fp16")]; + tensor var_49140_begin_0 = const()[name = tensor("op_49140_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_49140_end_0 = const()[name = tensor("op_49140_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_49140_end_mask_0 = const()[name = tensor("op_49140_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49140_cast_fp16 = slice_by_index(begin = var_49140_begin_0, end = var_49140_end_0, end_mask = var_49140_end_mask_0, x = v_217_cast_fp16)[name = tensor("op_49140_cast_fp16")]; + tensor var_49144_begin_0 = const()[name = tensor("op_49144_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_49144_end_0 = const()[name = tensor("op_49144_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_49144_end_mask_0 = const()[name = tensor("op_49144_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49144_cast_fp16 = slice_by_index(begin = var_49144_begin_0, end = var_49144_end_0, end_mask = var_49144_end_mask_0, x = v_217_cast_fp16)[name = tensor("op_49144_cast_fp16")]; + tensor var_49148_begin_0 = const()[name = tensor("op_49148_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_49148_end_0 = const()[name = tensor("op_49148_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_49148_end_mask_0 = const()[name = tensor("op_49148_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49148_cast_fp16 = slice_by_index(begin = var_49148_begin_0, end = var_49148_end_0, end_mask = var_49148_end_mask_0, x = v_217_cast_fp16)[name = tensor("op_49148_cast_fp16")]; + tensor var_49152_begin_0 = const()[name = tensor("op_49152_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_49152_end_0 = const()[name = tensor("op_49152_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_49152_end_mask_0 = const()[name = tensor("op_49152_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49152_cast_fp16 = slice_by_index(begin = var_49152_begin_0, end = var_49152_end_0, end_mask = var_49152_end_mask_0, x = v_217_cast_fp16)[name = tensor("op_49152_cast_fp16")]; + tensor var_49156_begin_0 = const()[name = tensor("op_49156_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_49156_end_0 = const()[name = tensor("op_49156_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_49156_end_mask_0 = const()[name = tensor("op_49156_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49156_cast_fp16 = slice_by_index(begin = var_49156_begin_0, end = var_49156_end_0, end_mask = var_49156_end_mask_0, x = v_217_cast_fp16)[name = tensor("op_49156_cast_fp16")]; + tensor var_49160_begin_0 = const()[name = tensor("op_49160_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_49160_end_0 = const()[name = tensor("op_49160_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_49160_end_mask_0 = const()[name = tensor("op_49160_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49160_cast_fp16 = slice_by_index(begin = var_49160_begin_0, end = var_49160_end_0, end_mask = var_49160_end_mask_0, x = v_217_cast_fp16)[name = tensor("op_49160_cast_fp16")]; + tensor var_49164_begin_0 = const()[name = tensor("op_49164_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_49164_end_0 = const()[name = tensor("op_49164_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_49164_end_mask_0 = const()[name = tensor("op_49164_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49164_cast_fp16 = slice_by_index(begin = var_49164_begin_0, end = var_49164_end_0, end_mask = var_49164_end_mask_0, x = v_217_cast_fp16)[name = tensor("op_49164_cast_fp16")]; + tensor var_49168_equation_0 = const()[name = tensor("op_49168_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_49168_cast_fp16 = einsum(equation = var_49168_equation_0, values = (var_49010_cast_fp16, var_48927_cast_fp16))[name = tensor("op_49168_cast_fp16")]; + tensor var_49169_to_fp16 = const()[name = tensor("op_49169_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4161_cast_fp16 = mul(x = var_49168_cast_fp16, y = var_49169_to_fp16)[name = tensor("aw_4161_cast_fp16")]; + tensor var_49172_equation_0 = const()[name = tensor("op_49172_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_49172_cast_fp16 = einsum(equation = var_49172_equation_0, values = (var_49014_cast_fp16, var_48931_cast_fp16))[name = tensor("op_49172_cast_fp16")]; + tensor var_49173_to_fp16 = const()[name = tensor("op_49173_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4163_cast_fp16 = mul(x = var_49172_cast_fp16, y = var_49173_to_fp16)[name = tensor("aw_4163_cast_fp16")]; + tensor var_49176_equation_0 = const()[name = tensor("op_49176_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_49176_cast_fp16 = einsum(equation = var_49176_equation_0, values = (var_49018_cast_fp16, var_48935_cast_fp16))[name = tensor("op_49176_cast_fp16")]; + tensor var_49177_to_fp16 = const()[name = tensor("op_49177_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4165_cast_fp16 = mul(x = var_49176_cast_fp16, y = var_49177_to_fp16)[name = tensor("aw_4165_cast_fp16")]; + tensor var_49180_equation_0 = const()[name = tensor("op_49180_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_49180_cast_fp16 = einsum(equation = var_49180_equation_0, values = (var_49022_cast_fp16, var_48939_cast_fp16))[name = tensor("op_49180_cast_fp16")]; + tensor var_49181_to_fp16 = const()[name = tensor("op_49181_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4167_cast_fp16 = mul(x = var_49180_cast_fp16, y = var_49181_to_fp16)[name = tensor("aw_4167_cast_fp16")]; + tensor var_49184_equation_0 = const()[name = tensor("op_49184_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_49184_cast_fp16 = einsum(equation = var_49184_equation_0, values = (var_49026_cast_fp16, var_48943_cast_fp16))[name = tensor("op_49184_cast_fp16")]; + tensor var_49185_to_fp16 = const()[name = tensor("op_49185_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4169_cast_fp16 = mul(x = var_49184_cast_fp16, y = var_49185_to_fp16)[name = tensor("aw_4169_cast_fp16")]; + tensor var_49188_equation_0 = const()[name = tensor("op_49188_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_49188_cast_fp16 = einsum(equation = var_49188_equation_0, values = (var_49030_cast_fp16, var_48947_cast_fp16))[name = tensor("op_49188_cast_fp16")]; + tensor var_49189_to_fp16 = const()[name = tensor("op_49189_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4171_cast_fp16 = mul(x = var_49188_cast_fp16, y = var_49189_to_fp16)[name = tensor("aw_4171_cast_fp16")]; + tensor var_49192_equation_0 = const()[name = tensor("op_49192_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_49192_cast_fp16 = einsum(equation = var_49192_equation_0, values = (var_49034_cast_fp16, var_48951_cast_fp16))[name = tensor("op_49192_cast_fp16")]; + tensor var_49193_to_fp16 = const()[name = tensor("op_49193_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4173_cast_fp16 = mul(x = var_49192_cast_fp16, y = var_49193_to_fp16)[name = tensor("aw_4173_cast_fp16")]; + tensor var_49196_equation_0 = const()[name = tensor("op_49196_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_49196_cast_fp16 = einsum(equation = var_49196_equation_0, values = (var_49038_cast_fp16, var_48955_cast_fp16))[name = tensor("op_49196_cast_fp16")]; + tensor var_49197_to_fp16 = const()[name = tensor("op_49197_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4175_cast_fp16 = mul(x = var_49196_cast_fp16, y = var_49197_to_fp16)[name = tensor("aw_4175_cast_fp16")]; + tensor var_49200_equation_0 = const()[name = tensor("op_49200_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_49200_cast_fp16 = einsum(equation = var_49200_equation_0, values = (var_49042_cast_fp16, var_48959_cast_fp16))[name = tensor("op_49200_cast_fp16")]; + tensor var_49201_to_fp16 = const()[name = tensor("op_49201_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4177_cast_fp16 = mul(x = var_49200_cast_fp16, y = var_49201_to_fp16)[name = tensor("aw_4177_cast_fp16")]; + tensor var_49204_equation_0 = const()[name = tensor("op_49204_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_49204_cast_fp16 = einsum(equation = var_49204_equation_0, values = (var_49046_cast_fp16, var_48963_cast_fp16))[name = tensor("op_49204_cast_fp16")]; + tensor var_49205_to_fp16 = const()[name = tensor("op_49205_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4179_cast_fp16 = mul(x = var_49204_cast_fp16, y = var_49205_to_fp16)[name = tensor("aw_4179_cast_fp16")]; + tensor var_49208_equation_0 = const()[name = tensor("op_49208_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_49208_cast_fp16 = einsum(equation = var_49208_equation_0, values = (var_49050_cast_fp16, var_48967_cast_fp16))[name = tensor("op_49208_cast_fp16")]; + tensor var_49209_to_fp16 = const()[name = tensor("op_49209_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4181_cast_fp16 = mul(x = var_49208_cast_fp16, y = var_49209_to_fp16)[name = tensor("aw_4181_cast_fp16")]; + tensor var_49212_equation_0 = const()[name = tensor("op_49212_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_49212_cast_fp16 = einsum(equation = var_49212_equation_0, values = (var_49054_cast_fp16, var_48971_cast_fp16))[name = tensor("op_49212_cast_fp16")]; + tensor var_49213_to_fp16 = const()[name = tensor("op_49213_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4183_cast_fp16 = mul(x = var_49212_cast_fp16, y = var_49213_to_fp16)[name = tensor("aw_4183_cast_fp16")]; + tensor var_49216_equation_0 = const()[name = tensor("op_49216_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_49216_cast_fp16 = einsum(equation = var_49216_equation_0, values = (var_49058_cast_fp16, var_48975_cast_fp16))[name = tensor("op_49216_cast_fp16")]; + tensor var_49217_to_fp16 = const()[name = tensor("op_49217_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4185_cast_fp16 = mul(x = var_49216_cast_fp16, y = var_49217_to_fp16)[name = tensor("aw_4185_cast_fp16")]; + tensor var_49220_equation_0 = const()[name = tensor("op_49220_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_49220_cast_fp16 = einsum(equation = var_49220_equation_0, values = (var_49062_cast_fp16, var_48979_cast_fp16))[name = tensor("op_49220_cast_fp16")]; + tensor var_49221_to_fp16 = const()[name = tensor("op_49221_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4187_cast_fp16 = mul(x = var_49220_cast_fp16, y = var_49221_to_fp16)[name = tensor("aw_4187_cast_fp16")]; + tensor var_49224_equation_0 = const()[name = tensor("op_49224_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_49224_cast_fp16 = einsum(equation = var_49224_equation_0, values = (var_49066_cast_fp16, var_48983_cast_fp16))[name = tensor("op_49224_cast_fp16")]; + tensor var_49225_to_fp16 = const()[name = tensor("op_49225_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4189_cast_fp16 = mul(x = var_49224_cast_fp16, y = var_49225_to_fp16)[name = tensor("aw_4189_cast_fp16")]; + tensor var_49228_equation_0 = const()[name = tensor("op_49228_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_49228_cast_fp16 = einsum(equation = var_49228_equation_0, values = (var_49070_cast_fp16, var_48987_cast_fp16))[name = tensor("op_49228_cast_fp16")]; + tensor var_49229_to_fp16 = const()[name = tensor("op_49229_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4191_cast_fp16 = mul(x = var_49228_cast_fp16, y = var_49229_to_fp16)[name = tensor("aw_4191_cast_fp16")]; + tensor var_49232_equation_0 = const()[name = tensor("op_49232_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_49232_cast_fp16 = einsum(equation = var_49232_equation_0, values = (var_49074_cast_fp16, var_48991_cast_fp16))[name = tensor("op_49232_cast_fp16")]; + tensor var_49233_to_fp16 = const()[name = tensor("op_49233_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4193_cast_fp16 = mul(x = var_49232_cast_fp16, y = var_49233_to_fp16)[name = tensor("aw_4193_cast_fp16")]; + tensor var_49236_equation_0 = const()[name = tensor("op_49236_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_49236_cast_fp16 = einsum(equation = var_49236_equation_0, values = (var_49078_cast_fp16, var_48995_cast_fp16))[name = tensor("op_49236_cast_fp16")]; + tensor var_49237_to_fp16 = const()[name = tensor("op_49237_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4195_cast_fp16 = mul(x = var_49236_cast_fp16, y = var_49237_to_fp16)[name = tensor("aw_4195_cast_fp16")]; + tensor var_49240_equation_0 = const()[name = tensor("op_49240_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_49240_cast_fp16 = einsum(equation = var_49240_equation_0, values = (var_49082_cast_fp16, var_48999_cast_fp16))[name = tensor("op_49240_cast_fp16")]; + tensor var_49241_to_fp16 = const()[name = tensor("op_49241_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4197_cast_fp16 = mul(x = var_49240_cast_fp16, y = var_49241_to_fp16)[name = tensor("aw_4197_cast_fp16")]; + tensor var_49244_equation_0 = const()[name = tensor("op_49244_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_49244_cast_fp16 = einsum(equation = var_49244_equation_0, values = (var_49086_cast_fp16, var_49003_cast_fp16))[name = tensor("op_49244_cast_fp16")]; + tensor var_49245_to_fp16 = const()[name = tensor("op_49245_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4199_cast_fp16 = mul(x = var_49244_cast_fp16, y = var_49245_to_fp16)[name = tensor("aw_4199_cast_fp16")]; + tensor var_49247_cast_fp16 = softmax(axis = var_30355, x = aw_4161_cast_fp16)[name = tensor("op_49247_cast_fp16")]; + tensor var_49248_cast_fp16 = softmax(axis = var_30355, x = aw_4163_cast_fp16)[name = tensor("op_49248_cast_fp16")]; + tensor var_49249_cast_fp16 = softmax(axis = var_30355, x = aw_4165_cast_fp16)[name = tensor("op_49249_cast_fp16")]; + tensor var_49250_cast_fp16 = softmax(axis = var_30355, x = aw_4167_cast_fp16)[name = tensor("op_49250_cast_fp16")]; + tensor var_49251_cast_fp16 = softmax(axis = var_30355, x = aw_4169_cast_fp16)[name = tensor("op_49251_cast_fp16")]; + tensor var_49252_cast_fp16 = softmax(axis = var_30355, x = aw_4171_cast_fp16)[name = tensor("op_49252_cast_fp16")]; + tensor var_49253_cast_fp16 = softmax(axis = var_30355, x = aw_4173_cast_fp16)[name = tensor("op_49253_cast_fp16")]; + tensor var_49254_cast_fp16 = softmax(axis = var_30355, x = aw_4175_cast_fp16)[name = tensor("op_49254_cast_fp16")]; + tensor var_49255_cast_fp16 = softmax(axis = var_30355, x = aw_4177_cast_fp16)[name = tensor("op_49255_cast_fp16")]; + tensor var_49256_cast_fp16 = softmax(axis = var_30355, x = aw_4179_cast_fp16)[name = tensor("op_49256_cast_fp16")]; + tensor var_49257_cast_fp16 = softmax(axis = var_30355, x = aw_4181_cast_fp16)[name = tensor("op_49257_cast_fp16")]; + tensor var_49258_cast_fp16 = softmax(axis = var_30355, x = aw_4183_cast_fp16)[name = tensor("op_49258_cast_fp16")]; + tensor var_49259_cast_fp16 = softmax(axis = var_30355, x = aw_4185_cast_fp16)[name = tensor("op_49259_cast_fp16")]; + tensor var_49260_cast_fp16 = softmax(axis = var_30355, x = aw_4187_cast_fp16)[name = tensor("op_49260_cast_fp16")]; + tensor var_49261_cast_fp16 = softmax(axis = var_30355, x = aw_4189_cast_fp16)[name = tensor("op_49261_cast_fp16")]; + tensor var_49262_cast_fp16 = softmax(axis = var_30355, x = aw_4191_cast_fp16)[name = tensor("op_49262_cast_fp16")]; + tensor var_49263_cast_fp16 = softmax(axis = var_30355, x = aw_4193_cast_fp16)[name = tensor("op_49263_cast_fp16")]; + tensor var_49264_cast_fp16 = softmax(axis = var_30355, x = aw_4195_cast_fp16)[name = tensor("op_49264_cast_fp16")]; + tensor var_49265_cast_fp16 = softmax(axis = var_30355, x = aw_4197_cast_fp16)[name = tensor("op_49265_cast_fp16")]; + tensor var_49266_cast_fp16 = softmax(axis = var_30355, x = aw_4199_cast_fp16)[name = tensor("op_49266_cast_fp16")]; + tensor var_49268_equation_0 = const()[name = tensor("op_49268_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49268_cast_fp16 = einsum(equation = var_49268_equation_0, values = (var_49088_cast_fp16, var_49247_cast_fp16))[name = tensor("op_49268_cast_fp16")]; + tensor var_49270_equation_0 = const()[name = tensor("op_49270_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49270_cast_fp16 = einsum(equation = var_49270_equation_0, values = (var_49092_cast_fp16, var_49248_cast_fp16))[name = tensor("op_49270_cast_fp16")]; + tensor var_49272_equation_0 = const()[name = tensor("op_49272_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49272_cast_fp16 = einsum(equation = var_49272_equation_0, values = (var_49096_cast_fp16, var_49249_cast_fp16))[name = tensor("op_49272_cast_fp16")]; + tensor var_49274_equation_0 = const()[name = tensor("op_49274_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49274_cast_fp16 = einsum(equation = var_49274_equation_0, values = (var_49100_cast_fp16, var_49250_cast_fp16))[name = tensor("op_49274_cast_fp16")]; + tensor var_49276_equation_0 = const()[name = tensor("op_49276_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49276_cast_fp16 = einsum(equation = var_49276_equation_0, values = (var_49104_cast_fp16, var_49251_cast_fp16))[name = tensor("op_49276_cast_fp16")]; + tensor var_49278_equation_0 = const()[name = tensor("op_49278_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49278_cast_fp16 = einsum(equation = var_49278_equation_0, values = (var_49108_cast_fp16, var_49252_cast_fp16))[name = tensor("op_49278_cast_fp16")]; + tensor var_49280_equation_0 = const()[name = tensor("op_49280_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49280_cast_fp16 = einsum(equation = var_49280_equation_0, values = (var_49112_cast_fp16, var_49253_cast_fp16))[name = tensor("op_49280_cast_fp16")]; + tensor var_49282_equation_0 = const()[name = tensor("op_49282_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49282_cast_fp16 = einsum(equation = var_49282_equation_0, values = (var_49116_cast_fp16, var_49254_cast_fp16))[name = tensor("op_49282_cast_fp16")]; + tensor var_49284_equation_0 = const()[name = tensor("op_49284_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49284_cast_fp16 = einsum(equation = var_49284_equation_0, values = (var_49120_cast_fp16, var_49255_cast_fp16))[name = tensor("op_49284_cast_fp16")]; + tensor var_49286_equation_0 = const()[name = tensor("op_49286_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49286_cast_fp16 = einsum(equation = var_49286_equation_0, values = (var_49124_cast_fp16, var_49256_cast_fp16))[name = tensor("op_49286_cast_fp16")]; + tensor var_49288_equation_0 = const()[name = tensor("op_49288_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49288_cast_fp16 = einsum(equation = var_49288_equation_0, values = (var_49128_cast_fp16, var_49257_cast_fp16))[name = tensor("op_49288_cast_fp16")]; + tensor var_49290_equation_0 = const()[name = tensor("op_49290_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49290_cast_fp16 = einsum(equation = var_49290_equation_0, values = (var_49132_cast_fp16, var_49258_cast_fp16))[name = tensor("op_49290_cast_fp16")]; + tensor var_49292_equation_0 = const()[name = tensor("op_49292_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49292_cast_fp16 = einsum(equation = var_49292_equation_0, values = (var_49136_cast_fp16, var_49259_cast_fp16))[name = tensor("op_49292_cast_fp16")]; + tensor var_49294_equation_0 = const()[name = tensor("op_49294_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49294_cast_fp16 = einsum(equation = var_49294_equation_0, values = (var_49140_cast_fp16, var_49260_cast_fp16))[name = tensor("op_49294_cast_fp16")]; + tensor var_49296_equation_0 = const()[name = tensor("op_49296_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49296_cast_fp16 = einsum(equation = var_49296_equation_0, values = (var_49144_cast_fp16, var_49261_cast_fp16))[name = tensor("op_49296_cast_fp16")]; + tensor var_49298_equation_0 = const()[name = tensor("op_49298_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49298_cast_fp16 = einsum(equation = var_49298_equation_0, values = (var_49148_cast_fp16, var_49262_cast_fp16))[name = tensor("op_49298_cast_fp16")]; + tensor var_49300_equation_0 = const()[name = tensor("op_49300_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49300_cast_fp16 = einsum(equation = var_49300_equation_0, values = (var_49152_cast_fp16, var_49263_cast_fp16))[name = tensor("op_49300_cast_fp16")]; + tensor var_49302_equation_0 = const()[name = tensor("op_49302_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49302_cast_fp16 = einsum(equation = var_49302_equation_0, values = (var_49156_cast_fp16, var_49264_cast_fp16))[name = tensor("op_49302_cast_fp16")]; + tensor var_49304_equation_0 = const()[name = tensor("op_49304_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49304_cast_fp16 = einsum(equation = var_49304_equation_0, values = (var_49160_cast_fp16, var_49265_cast_fp16))[name = tensor("op_49304_cast_fp16")]; + tensor var_49306_equation_0 = const()[name = tensor("op_49306_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49306_cast_fp16 = einsum(equation = var_49306_equation_0, values = (var_49164_cast_fp16, var_49266_cast_fp16))[name = tensor("op_49306_cast_fp16")]; + tensor input_635_interleave_0 = const()[name = tensor("input_635_interleave_0"), val = tensor(false)]; + tensor input_635_cast_fp16 = concat(axis = var_30355, interleave = input_635_interleave_0, values = (var_49268_cast_fp16, var_49270_cast_fp16, var_49272_cast_fp16, var_49274_cast_fp16, var_49276_cast_fp16, var_49278_cast_fp16, var_49280_cast_fp16, var_49282_cast_fp16, var_49284_cast_fp16, var_49286_cast_fp16, var_49288_cast_fp16, var_49290_cast_fp16, var_49292_cast_fp16, var_49294_cast_fp16, var_49296_cast_fp16, var_49298_cast_fp16, var_49300_cast_fp16, var_49302_cast_fp16, var_49304_cast_fp16, var_49306_cast_fp16))[name = tensor("input_635_cast_fp16")]; + tensor var_49312 = const()[name = tensor("op_49312"), val = tensor([1, 1])]; + tensor var_49314 = const()[name = tensor("op_49314"), val = tensor([1, 1])]; + tensor var_49316_pad_type_0 = const()[name = tensor("op_49316_pad_type_0"), val = tensor("custom")]; + tensor var_49316_pad_0 = const()[name = tensor("op_49316_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_0_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_0_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4181926848)))]; + tensor up_blocks_0_attentions_2_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4185203712)))]; + tensor var_49316_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_49314, groups = var_30355, pad = var_49316_pad_0, pad_type = var_49316_pad_type_0, strides = var_49312, weight = up_blocks_0_attentions_2_transformer_blocks_0_attn1_to_out_0_weight_to_fp16, x = input_635_cast_fp16)[name = tensor("op_49316_cast_fp16")]; + tensor inputs_327_cast_fp16 = add(x = var_49316_cast_fp16, y = inputs_325_cast_fp16)[name = tensor("inputs_327_cast_fp16")]; + tensor hidden_states_437_axes_0 = const()[name = tensor("hidden_states_437_axes_0"), val = tensor([1])]; + tensor hidden_states_437_gamma_0_to_fp16 = const()[name = tensor("hidden_states_437_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4185206336)))]; + tensor hidden_states_437_beta_0_to_fp16 = const()[name = tensor("hidden_states_437_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4185208960)))]; + tensor var_49326_to_fp16 = const()[name = tensor("op_49326_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_437_cast_fp16 = layer_norm(axes = hidden_states_437_axes_0, beta = hidden_states_437_beta_0_to_fp16, epsilon = var_49326_to_fp16, gamma = hidden_states_437_gamma_0_to_fp16, x = inputs_327_cast_fp16)[name = tensor("hidden_states_437_cast_fp16")]; + tensor var_49341 = const()[name = tensor("op_49341"), val = tensor([1, 1])]; + tensor var_49343 = const()[name = tensor("op_49343"), val = tensor([1, 1])]; + tensor q_219_pad_type_0 = const()[name = tensor("q_219_pad_type_0"), val = tensor("custom")]; + tensor q_219_pad_0 = const()[name = tensor("q_219_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_0_attn2_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_0_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4185211584)))]; + tensor q_219_cast_fp16 = conv(dilations = var_49343, groups = var_30355, pad = q_219_pad_0, pad_type = q_219_pad_type_0, strides = var_49341, weight = up_blocks_0_attentions_2_transformer_blocks_0_attn2_to_q_weight_to_fp16, x = hidden_states_437_cast_fp16)[name = tensor("q_219_cast_fp16")]; + tensor var_49347 = const()[name = tensor("op_49347"), val = tensor([1, 1])]; + tensor var_49349 = const()[name = tensor("op_49349"), val = tensor([1, 1])]; + tensor k_437_pad_type_0 = const()[name = tensor("k_437_pad_type_0"), val = tensor("custom")]; + tensor k_437_pad_0 = const()[name = tensor("k_437_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4188488448)))]; + tensor k_437_cast_fp16 = conv(dilations = var_49349, groups = var_30355, pad = k_437_pad_0, pad_type = k_437_pad_type_0, strides = var_49347, weight = up_blocks_0_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_437_cast_fp16")]; + tensor var_49353 = const()[name = tensor("op_49353"), val = tensor([1, 1])]; + tensor var_49355 = const()[name = tensor("op_49355"), val = tensor([1, 1])]; + tensor v_219_pad_type_0 = const()[name = tensor("v_219_pad_type_0"), val = tensor("custom")]; + tensor v_219_pad_0 = const()[name = tensor("v_219_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4193731392)))]; + tensor v_219_cast_fp16 = conv(dilations = var_49355, groups = var_30355, pad = v_219_pad_0, pad_type = v_219_pad_type_0, strides = var_49353, weight = up_blocks_0_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_219_cast_fp16")]; + tensor var_49359_begin_0 = const()[name = tensor("op_49359_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_49359_end_0 = const()[name = tensor("op_49359_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_49359_end_mask_0 = const()[name = tensor("op_49359_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49359_cast_fp16 = slice_by_index(begin = var_49359_begin_0, end = var_49359_end_0, end_mask = var_49359_end_mask_0, x = q_219_cast_fp16)[name = tensor("op_49359_cast_fp16")]; + tensor var_49363_begin_0 = const()[name = tensor("op_49363_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_49363_end_0 = const()[name = tensor("op_49363_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_49363_end_mask_0 = const()[name = tensor("op_49363_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49363_cast_fp16 = slice_by_index(begin = var_49363_begin_0, end = var_49363_end_0, end_mask = var_49363_end_mask_0, x = q_219_cast_fp16)[name = tensor("op_49363_cast_fp16")]; + tensor var_49367_begin_0 = const()[name = tensor("op_49367_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_49367_end_0 = const()[name = tensor("op_49367_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_49367_end_mask_0 = const()[name = tensor("op_49367_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49367_cast_fp16 = slice_by_index(begin = var_49367_begin_0, end = var_49367_end_0, end_mask = var_49367_end_mask_0, x = q_219_cast_fp16)[name = tensor("op_49367_cast_fp16")]; + tensor var_49371_begin_0 = const()[name = tensor("op_49371_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_49371_end_0 = const()[name = tensor("op_49371_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_49371_end_mask_0 = const()[name = tensor("op_49371_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49371_cast_fp16 = slice_by_index(begin = var_49371_begin_0, end = var_49371_end_0, end_mask = var_49371_end_mask_0, x = q_219_cast_fp16)[name = tensor("op_49371_cast_fp16")]; + tensor var_49375_begin_0 = const()[name = tensor("op_49375_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_49375_end_0 = const()[name = tensor("op_49375_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_49375_end_mask_0 = const()[name = tensor("op_49375_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49375_cast_fp16 = slice_by_index(begin = var_49375_begin_0, end = var_49375_end_0, end_mask = var_49375_end_mask_0, x = q_219_cast_fp16)[name = tensor("op_49375_cast_fp16")]; + tensor var_49379_begin_0 = const()[name = tensor("op_49379_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_49379_end_0 = const()[name = tensor("op_49379_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_49379_end_mask_0 = const()[name = tensor("op_49379_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49379_cast_fp16 = slice_by_index(begin = var_49379_begin_0, end = var_49379_end_0, end_mask = var_49379_end_mask_0, x = q_219_cast_fp16)[name = tensor("op_49379_cast_fp16")]; + tensor var_49383_begin_0 = const()[name = tensor("op_49383_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_49383_end_0 = const()[name = tensor("op_49383_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_49383_end_mask_0 = const()[name = tensor("op_49383_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49383_cast_fp16 = slice_by_index(begin = var_49383_begin_0, end = var_49383_end_0, end_mask = var_49383_end_mask_0, x = q_219_cast_fp16)[name = tensor("op_49383_cast_fp16")]; + tensor var_49387_begin_0 = const()[name = tensor("op_49387_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_49387_end_0 = const()[name = tensor("op_49387_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_49387_end_mask_0 = const()[name = tensor("op_49387_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49387_cast_fp16 = slice_by_index(begin = var_49387_begin_0, end = var_49387_end_0, end_mask = var_49387_end_mask_0, x = q_219_cast_fp16)[name = tensor("op_49387_cast_fp16")]; + tensor var_49391_begin_0 = const()[name = tensor("op_49391_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_49391_end_0 = const()[name = tensor("op_49391_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_49391_end_mask_0 = const()[name = tensor("op_49391_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49391_cast_fp16 = slice_by_index(begin = var_49391_begin_0, end = var_49391_end_0, end_mask = var_49391_end_mask_0, x = q_219_cast_fp16)[name = tensor("op_49391_cast_fp16")]; + tensor var_49395_begin_0 = const()[name = tensor("op_49395_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_49395_end_0 = const()[name = tensor("op_49395_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_49395_end_mask_0 = const()[name = tensor("op_49395_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49395_cast_fp16 = slice_by_index(begin = var_49395_begin_0, end = var_49395_end_0, end_mask = var_49395_end_mask_0, x = q_219_cast_fp16)[name = tensor("op_49395_cast_fp16")]; + tensor var_49399_begin_0 = const()[name = tensor("op_49399_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_49399_end_0 = const()[name = tensor("op_49399_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_49399_end_mask_0 = const()[name = tensor("op_49399_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49399_cast_fp16 = slice_by_index(begin = var_49399_begin_0, end = var_49399_end_0, end_mask = var_49399_end_mask_0, x = q_219_cast_fp16)[name = tensor("op_49399_cast_fp16")]; + tensor var_49403_begin_0 = const()[name = tensor("op_49403_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_49403_end_0 = const()[name = tensor("op_49403_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_49403_end_mask_0 = const()[name = tensor("op_49403_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49403_cast_fp16 = slice_by_index(begin = var_49403_begin_0, end = var_49403_end_0, end_mask = var_49403_end_mask_0, x = q_219_cast_fp16)[name = tensor("op_49403_cast_fp16")]; + tensor var_49407_begin_0 = const()[name = tensor("op_49407_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_49407_end_0 = const()[name = tensor("op_49407_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_49407_end_mask_0 = const()[name = tensor("op_49407_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49407_cast_fp16 = slice_by_index(begin = var_49407_begin_0, end = var_49407_end_0, end_mask = var_49407_end_mask_0, x = q_219_cast_fp16)[name = tensor("op_49407_cast_fp16")]; + tensor var_49411_begin_0 = const()[name = tensor("op_49411_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_49411_end_0 = const()[name = tensor("op_49411_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_49411_end_mask_0 = const()[name = tensor("op_49411_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49411_cast_fp16 = slice_by_index(begin = var_49411_begin_0, end = var_49411_end_0, end_mask = var_49411_end_mask_0, x = q_219_cast_fp16)[name = tensor("op_49411_cast_fp16")]; + tensor var_49415_begin_0 = const()[name = tensor("op_49415_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_49415_end_0 = const()[name = tensor("op_49415_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_49415_end_mask_0 = const()[name = tensor("op_49415_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49415_cast_fp16 = slice_by_index(begin = var_49415_begin_0, end = var_49415_end_0, end_mask = var_49415_end_mask_0, x = q_219_cast_fp16)[name = tensor("op_49415_cast_fp16")]; + tensor var_49419_begin_0 = const()[name = tensor("op_49419_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_49419_end_0 = const()[name = tensor("op_49419_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_49419_end_mask_0 = const()[name = tensor("op_49419_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49419_cast_fp16 = slice_by_index(begin = var_49419_begin_0, end = var_49419_end_0, end_mask = var_49419_end_mask_0, x = q_219_cast_fp16)[name = tensor("op_49419_cast_fp16")]; + tensor var_49423_begin_0 = const()[name = tensor("op_49423_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_49423_end_0 = const()[name = tensor("op_49423_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_49423_end_mask_0 = const()[name = tensor("op_49423_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49423_cast_fp16 = slice_by_index(begin = var_49423_begin_0, end = var_49423_end_0, end_mask = var_49423_end_mask_0, x = q_219_cast_fp16)[name = tensor("op_49423_cast_fp16")]; + tensor var_49427_begin_0 = const()[name = tensor("op_49427_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_49427_end_0 = const()[name = tensor("op_49427_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_49427_end_mask_0 = const()[name = tensor("op_49427_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49427_cast_fp16 = slice_by_index(begin = var_49427_begin_0, end = var_49427_end_0, end_mask = var_49427_end_mask_0, x = q_219_cast_fp16)[name = tensor("op_49427_cast_fp16")]; + tensor var_49431_begin_0 = const()[name = tensor("op_49431_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_49431_end_0 = const()[name = tensor("op_49431_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_49431_end_mask_0 = const()[name = tensor("op_49431_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49431_cast_fp16 = slice_by_index(begin = var_49431_begin_0, end = var_49431_end_0, end_mask = var_49431_end_mask_0, x = q_219_cast_fp16)[name = tensor("op_49431_cast_fp16")]; + tensor var_49435_begin_0 = const()[name = tensor("op_49435_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_49435_end_0 = const()[name = tensor("op_49435_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_49435_end_mask_0 = const()[name = tensor("op_49435_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49435_cast_fp16 = slice_by_index(begin = var_49435_begin_0, end = var_49435_end_0, end_mask = var_49435_end_mask_0, x = q_219_cast_fp16)[name = tensor("op_49435_cast_fp16")]; + tensor k_439_perm_0 = const()[name = tensor("k_439_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_49442_begin_0 = const()[name = tensor("op_49442_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_49442_end_0 = const()[name = tensor("op_49442_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_49442_end_mask_0 = const()[name = tensor("op_49442_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_439_cast_fp16 = transpose(perm = k_439_perm_0, x = k_437_cast_fp16)[name = tensor("transpose_30")]; + tensor var_49442_cast_fp16 = slice_by_index(begin = var_49442_begin_0, end = var_49442_end_0, end_mask = var_49442_end_mask_0, x = k_439_cast_fp16)[name = tensor("op_49442_cast_fp16")]; + tensor var_49446_begin_0 = const()[name = tensor("op_49446_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_49446_end_0 = const()[name = tensor("op_49446_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_49446_end_mask_0 = const()[name = tensor("op_49446_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49446_cast_fp16 = slice_by_index(begin = var_49446_begin_0, end = var_49446_end_0, end_mask = var_49446_end_mask_0, x = k_439_cast_fp16)[name = tensor("op_49446_cast_fp16")]; + tensor var_49450_begin_0 = const()[name = tensor("op_49450_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_49450_end_0 = const()[name = tensor("op_49450_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_49450_end_mask_0 = const()[name = tensor("op_49450_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49450_cast_fp16 = slice_by_index(begin = var_49450_begin_0, end = var_49450_end_0, end_mask = var_49450_end_mask_0, x = k_439_cast_fp16)[name = tensor("op_49450_cast_fp16")]; + tensor var_49454_begin_0 = const()[name = tensor("op_49454_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_49454_end_0 = const()[name = tensor("op_49454_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_49454_end_mask_0 = const()[name = tensor("op_49454_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49454_cast_fp16 = slice_by_index(begin = var_49454_begin_0, end = var_49454_end_0, end_mask = var_49454_end_mask_0, x = k_439_cast_fp16)[name = tensor("op_49454_cast_fp16")]; + tensor var_49458_begin_0 = const()[name = tensor("op_49458_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_49458_end_0 = const()[name = tensor("op_49458_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_49458_end_mask_0 = const()[name = tensor("op_49458_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49458_cast_fp16 = slice_by_index(begin = var_49458_begin_0, end = var_49458_end_0, end_mask = var_49458_end_mask_0, x = k_439_cast_fp16)[name = tensor("op_49458_cast_fp16")]; + tensor var_49462_begin_0 = const()[name = tensor("op_49462_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_49462_end_0 = const()[name = tensor("op_49462_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_49462_end_mask_0 = const()[name = tensor("op_49462_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49462_cast_fp16 = slice_by_index(begin = var_49462_begin_0, end = var_49462_end_0, end_mask = var_49462_end_mask_0, x = k_439_cast_fp16)[name = tensor("op_49462_cast_fp16")]; + tensor var_49466_begin_0 = const()[name = tensor("op_49466_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_49466_end_0 = const()[name = tensor("op_49466_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_49466_end_mask_0 = const()[name = tensor("op_49466_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49466_cast_fp16 = slice_by_index(begin = var_49466_begin_0, end = var_49466_end_0, end_mask = var_49466_end_mask_0, x = k_439_cast_fp16)[name = tensor("op_49466_cast_fp16")]; + tensor var_49470_begin_0 = const()[name = tensor("op_49470_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_49470_end_0 = const()[name = tensor("op_49470_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_49470_end_mask_0 = const()[name = tensor("op_49470_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49470_cast_fp16 = slice_by_index(begin = var_49470_begin_0, end = var_49470_end_0, end_mask = var_49470_end_mask_0, x = k_439_cast_fp16)[name = tensor("op_49470_cast_fp16")]; + tensor var_49474_begin_0 = const()[name = tensor("op_49474_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_49474_end_0 = const()[name = tensor("op_49474_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_49474_end_mask_0 = const()[name = tensor("op_49474_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49474_cast_fp16 = slice_by_index(begin = var_49474_begin_0, end = var_49474_end_0, end_mask = var_49474_end_mask_0, x = k_439_cast_fp16)[name = tensor("op_49474_cast_fp16")]; + tensor var_49478_begin_0 = const()[name = tensor("op_49478_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_49478_end_0 = const()[name = tensor("op_49478_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_49478_end_mask_0 = const()[name = tensor("op_49478_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49478_cast_fp16 = slice_by_index(begin = var_49478_begin_0, end = var_49478_end_0, end_mask = var_49478_end_mask_0, x = k_439_cast_fp16)[name = tensor("op_49478_cast_fp16")]; + tensor var_49482_begin_0 = const()[name = tensor("op_49482_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_49482_end_0 = const()[name = tensor("op_49482_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_49482_end_mask_0 = const()[name = tensor("op_49482_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49482_cast_fp16 = slice_by_index(begin = var_49482_begin_0, end = var_49482_end_0, end_mask = var_49482_end_mask_0, x = k_439_cast_fp16)[name = tensor("op_49482_cast_fp16")]; + tensor var_49486_begin_0 = const()[name = tensor("op_49486_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_49486_end_0 = const()[name = tensor("op_49486_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_49486_end_mask_0 = const()[name = tensor("op_49486_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49486_cast_fp16 = slice_by_index(begin = var_49486_begin_0, end = var_49486_end_0, end_mask = var_49486_end_mask_0, x = k_439_cast_fp16)[name = tensor("op_49486_cast_fp16")]; + tensor var_49490_begin_0 = const()[name = tensor("op_49490_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_49490_end_0 = const()[name = tensor("op_49490_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_49490_end_mask_0 = const()[name = tensor("op_49490_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49490_cast_fp16 = slice_by_index(begin = var_49490_begin_0, end = var_49490_end_0, end_mask = var_49490_end_mask_0, x = k_439_cast_fp16)[name = tensor("op_49490_cast_fp16")]; + tensor var_49494_begin_0 = const()[name = tensor("op_49494_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_49494_end_0 = const()[name = tensor("op_49494_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_49494_end_mask_0 = const()[name = tensor("op_49494_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49494_cast_fp16 = slice_by_index(begin = var_49494_begin_0, end = var_49494_end_0, end_mask = var_49494_end_mask_0, x = k_439_cast_fp16)[name = tensor("op_49494_cast_fp16")]; + tensor var_49498_begin_0 = const()[name = tensor("op_49498_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_49498_end_0 = const()[name = tensor("op_49498_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_49498_end_mask_0 = const()[name = tensor("op_49498_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49498_cast_fp16 = slice_by_index(begin = var_49498_begin_0, end = var_49498_end_0, end_mask = var_49498_end_mask_0, x = k_439_cast_fp16)[name = tensor("op_49498_cast_fp16")]; + tensor var_49502_begin_0 = const()[name = tensor("op_49502_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_49502_end_0 = const()[name = tensor("op_49502_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_49502_end_mask_0 = const()[name = tensor("op_49502_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49502_cast_fp16 = slice_by_index(begin = var_49502_begin_0, end = var_49502_end_0, end_mask = var_49502_end_mask_0, x = k_439_cast_fp16)[name = tensor("op_49502_cast_fp16")]; + tensor var_49506_begin_0 = const()[name = tensor("op_49506_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_49506_end_0 = const()[name = tensor("op_49506_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_49506_end_mask_0 = const()[name = tensor("op_49506_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49506_cast_fp16 = slice_by_index(begin = var_49506_begin_0, end = var_49506_end_0, end_mask = var_49506_end_mask_0, x = k_439_cast_fp16)[name = tensor("op_49506_cast_fp16")]; + tensor var_49510_begin_0 = const()[name = tensor("op_49510_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_49510_end_0 = const()[name = tensor("op_49510_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_49510_end_mask_0 = const()[name = tensor("op_49510_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49510_cast_fp16 = slice_by_index(begin = var_49510_begin_0, end = var_49510_end_0, end_mask = var_49510_end_mask_0, x = k_439_cast_fp16)[name = tensor("op_49510_cast_fp16")]; + tensor var_49514_begin_0 = const()[name = tensor("op_49514_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_49514_end_0 = const()[name = tensor("op_49514_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_49514_end_mask_0 = const()[name = tensor("op_49514_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49514_cast_fp16 = slice_by_index(begin = var_49514_begin_0, end = var_49514_end_0, end_mask = var_49514_end_mask_0, x = k_439_cast_fp16)[name = tensor("op_49514_cast_fp16")]; + tensor var_49518_begin_0 = const()[name = tensor("op_49518_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_49518_end_0 = const()[name = tensor("op_49518_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_49518_end_mask_0 = const()[name = tensor("op_49518_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49518_cast_fp16 = slice_by_index(begin = var_49518_begin_0, end = var_49518_end_0, end_mask = var_49518_end_mask_0, x = k_439_cast_fp16)[name = tensor("op_49518_cast_fp16")]; + tensor var_49520_begin_0 = const()[name = tensor("op_49520_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_49520_end_0 = const()[name = tensor("op_49520_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_49520_end_mask_0 = const()[name = tensor("op_49520_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49520_cast_fp16 = slice_by_index(begin = var_49520_begin_0, end = var_49520_end_0, end_mask = var_49520_end_mask_0, x = v_219_cast_fp16)[name = tensor("op_49520_cast_fp16")]; + tensor var_49524_begin_0 = const()[name = tensor("op_49524_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_49524_end_0 = const()[name = tensor("op_49524_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_49524_end_mask_0 = const()[name = tensor("op_49524_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49524_cast_fp16 = slice_by_index(begin = var_49524_begin_0, end = var_49524_end_0, end_mask = var_49524_end_mask_0, x = v_219_cast_fp16)[name = tensor("op_49524_cast_fp16")]; + tensor var_49528_begin_0 = const()[name = tensor("op_49528_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_49528_end_0 = const()[name = tensor("op_49528_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_49528_end_mask_0 = const()[name = tensor("op_49528_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49528_cast_fp16 = slice_by_index(begin = var_49528_begin_0, end = var_49528_end_0, end_mask = var_49528_end_mask_0, x = v_219_cast_fp16)[name = tensor("op_49528_cast_fp16")]; + tensor var_49532_begin_0 = const()[name = tensor("op_49532_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_49532_end_0 = const()[name = tensor("op_49532_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_49532_end_mask_0 = const()[name = tensor("op_49532_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49532_cast_fp16 = slice_by_index(begin = var_49532_begin_0, end = var_49532_end_0, end_mask = var_49532_end_mask_0, x = v_219_cast_fp16)[name = tensor("op_49532_cast_fp16")]; + tensor var_49536_begin_0 = const()[name = tensor("op_49536_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_49536_end_0 = const()[name = tensor("op_49536_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_49536_end_mask_0 = const()[name = tensor("op_49536_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49536_cast_fp16 = slice_by_index(begin = var_49536_begin_0, end = var_49536_end_0, end_mask = var_49536_end_mask_0, x = v_219_cast_fp16)[name = tensor("op_49536_cast_fp16")]; + tensor var_49540_begin_0 = const()[name = tensor("op_49540_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_49540_end_0 = const()[name = tensor("op_49540_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_49540_end_mask_0 = const()[name = tensor("op_49540_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49540_cast_fp16 = slice_by_index(begin = var_49540_begin_0, end = var_49540_end_0, end_mask = var_49540_end_mask_0, x = v_219_cast_fp16)[name = tensor("op_49540_cast_fp16")]; + tensor var_49544_begin_0 = const()[name = tensor("op_49544_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_49544_end_0 = const()[name = tensor("op_49544_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_49544_end_mask_0 = const()[name = tensor("op_49544_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49544_cast_fp16 = slice_by_index(begin = var_49544_begin_0, end = var_49544_end_0, end_mask = var_49544_end_mask_0, x = v_219_cast_fp16)[name = tensor("op_49544_cast_fp16")]; + tensor var_49548_begin_0 = const()[name = tensor("op_49548_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_49548_end_0 = const()[name = tensor("op_49548_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_49548_end_mask_0 = const()[name = tensor("op_49548_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49548_cast_fp16 = slice_by_index(begin = var_49548_begin_0, end = var_49548_end_0, end_mask = var_49548_end_mask_0, x = v_219_cast_fp16)[name = tensor("op_49548_cast_fp16")]; + tensor var_49552_begin_0 = const()[name = tensor("op_49552_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_49552_end_0 = const()[name = tensor("op_49552_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_49552_end_mask_0 = const()[name = tensor("op_49552_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49552_cast_fp16 = slice_by_index(begin = var_49552_begin_0, end = var_49552_end_0, end_mask = var_49552_end_mask_0, x = v_219_cast_fp16)[name = tensor("op_49552_cast_fp16")]; + tensor var_49556_begin_0 = const()[name = tensor("op_49556_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_49556_end_0 = const()[name = tensor("op_49556_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_49556_end_mask_0 = const()[name = tensor("op_49556_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49556_cast_fp16 = slice_by_index(begin = var_49556_begin_0, end = var_49556_end_0, end_mask = var_49556_end_mask_0, x = v_219_cast_fp16)[name = tensor("op_49556_cast_fp16")]; + tensor var_49560_begin_0 = const()[name = tensor("op_49560_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_49560_end_0 = const()[name = tensor("op_49560_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_49560_end_mask_0 = const()[name = tensor("op_49560_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49560_cast_fp16 = slice_by_index(begin = var_49560_begin_0, end = var_49560_end_0, end_mask = var_49560_end_mask_0, x = v_219_cast_fp16)[name = tensor("op_49560_cast_fp16")]; + tensor var_49564_begin_0 = const()[name = tensor("op_49564_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_49564_end_0 = const()[name = tensor("op_49564_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_49564_end_mask_0 = const()[name = tensor("op_49564_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49564_cast_fp16 = slice_by_index(begin = var_49564_begin_0, end = var_49564_end_0, end_mask = var_49564_end_mask_0, x = v_219_cast_fp16)[name = tensor("op_49564_cast_fp16")]; + tensor var_49568_begin_0 = const()[name = tensor("op_49568_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_49568_end_0 = const()[name = tensor("op_49568_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_49568_end_mask_0 = const()[name = tensor("op_49568_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49568_cast_fp16 = slice_by_index(begin = var_49568_begin_0, end = var_49568_end_0, end_mask = var_49568_end_mask_0, x = v_219_cast_fp16)[name = tensor("op_49568_cast_fp16")]; + tensor var_49572_begin_0 = const()[name = tensor("op_49572_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_49572_end_0 = const()[name = tensor("op_49572_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_49572_end_mask_0 = const()[name = tensor("op_49572_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49572_cast_fp16 = slice_by_index(begin = var_49572_begin_0, end = var_49572_end_0, end_mask = var_49572_end_mask_0, x = v_219_cast_fp16)[name = tensor("op_49572_cast_fp16")]; + tensor var_49576_begin_0 = const()[name = tensor("op_49576_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_49576_end_0 = const()[name = tensor("op_49576_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_49576_end_mask_0 = const()[name = tensor("op_49576_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49576_cast_fp16 = slice_by_index(begin = var_49576_begin_0, end = var_49576_end_0, end_mask = var_49576_end_mask_0, x = v_219_cast_fp16)[name = tensor("op_49576_cast_fp16")]; + tensor var_49580_begin_0 = const()[name = tensor("op_49580_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_49580_end_0 = const()[name = tensor("op_49580_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_49580_end_mask_0 = const()[name = tensor("op_49580_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49580_cast_fp16 = slice_by_index(begin = var_49580_begin_0, end = var_49580_end_0, end_mask = var_49580_end_mask_0, x = v_219_cast_fp16)[name = tensor("op_49580_cast_fp16")]; + tensor var_49584_begin_0 = const()[name = tensor("op_49584_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_49584_end_0 = const()[name = tensor("op_49584_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_49584_end_mask_0 = const()[name = tensor("op_49584_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49584_cast_fp16 = slice_by_index(begin = var_49584_begin_0, end = var_49584_end_0, end_mask = var_49584_end_mask_0, x = v_219_cast_fp16)[name = tensor("op_49584_cast_fp16")]; + tensor var_49588_begin_0 = const()[name = tensor("op_49588_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_49588_end_0 = const()[name = tensor("op_49588_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_49588_end_mask_0 = const()[name = tensor("op_49588_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49588_cast_fp16 = slice_by_index(begin = var_49588_begin_0, end = var_49588_end_0, end_mask = var_49588_end_mask_0, x = v_219_cast_fp16)[name = tensor("op_49588_cast_fp16")]; + tensor var_49592_begin_0 = const()[name = tensor("op_49592_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_49592_end_0 = const()[name = tensor("op_49592_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_49592_end_mask_0 = const()[name = tensor("op_49592_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49592_cast_fp16 = slice_by_index(begin = var_49592_begin_0, end = var_49592_end_0, end_mask = var_49592_end_mask_0, x = v_219_cast_fp16)[name = tensor("op_49592_cast_fp16")]; + tensor var_49596_begin_0 = const()[name = tensor("op_49596_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_49596_end_0 = const()[name = tensor("op_49596_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_49596_end_mask_0 = const()[name = tensor("op_49596_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49596_cast_fp16 = slice_by_index(begin = var_49596_begin_0, end = var_49596_end_0, end_mask = var_49596_end_mask_0, x = v_219_cast_fp16)[name = tensor("op_49596_cast_fp16")]; + tensor var_49600_equation_0 = const()[name = tensor("op_49600_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_49600_cast_fp16 = einsum(equation = var_49600_equation_0, values = (var_49442_cast_fp16, var_49359_cast_fp16))[name = tensor("op_49600_cast_fp16")]; + tensor var_49601_to_fp16 = const()[name = tensor("op_49601_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4201_cast_fp16 = mul(x = var_49600_cast_fp16, y = var_49601_to_fp16)[name = tensor("aw_4201_cast_fp16")]; + tensor var_49604_equation_0 = const()[name = tensor("op_49604_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_49604_cast_fp16 = einsum(equation = var_49604_equation_0, values = (var_49446_cast_fp16, var_49363_cast_fp16))[name = tensor("op_49604_cast_fp16")]; + tensor var_49605_to_fp16 = const()[name = tensor("op_49605_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4203_cast_fp16 = mul(x = var_49604_cast_fp16, y = var_49605_to_fp16)[name = tensor("aw_4203_cast_fp16")]; + tensor var_49608_equation_0 = const()[name = tensor("op_49608_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_49608_cast_fp16 = einsum(equation = var_49608_equation_0, values = (var_49450_cast_fp16, var_49367_cast_fp16))[name = tensor("op_49608_cast_fp16")]; + tensor var_49609_to_fp16 = const()[name = tensor("op_49609_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4205_cast_fp16 = mul(x = var_49608_cast_fp16, y = var_49609_to_fp16)[name = tensor("aw_4205_cast_fp16")]; + tensor var_49612_equation_0 = const()[name = tensor("op_49612_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_49612_cast_fp16 = einsum(equation = var_49612_equation_0, values = (var_49454_cast_fp16, var_49371_cast_fp16))[name = tensor("op_49612_cast_fp16")]; + tensor var_49613_to_fp16 = const()[name = tensor("op_49613_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4207_cast_fp16 = mul(x = var_49612_cast_fp16, y = var_49613_to_fp16)[name = tensor("aw_4207_cast_fp16")]; + tensor var_49616_equation_0 = const()[name = tensor("op_49616_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_49616_cast_fp16 = einsum(equation = var_49616_equation_0, values = (var_49458_cast_fp16, var_49375_cast_fp16))[name = tensor("op_49616_cast_fp16")]; + tensor var_49617_to_fp16 = const()[name = tensor("op_49617_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4209_cast_fp16 = mul(x = var_49616_cast_fp16, y = var_49617_to_fp16)[name = tensor("aw_4209_cast_fp16")]; + tensor var_49620_equation_0 = const()[name = tensor("op_49620_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_49620_cast_fp16 = einsum(equation = var_49620_equation_0, values = (var_49462_cast_fp16, var_49379_cast_fp16))[name = tensor("op_49620_cast_fp16")]; + tensor var_49621_to_fp16 = const()[name = tensor("op_49621_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4211_cast_fp16 = mul(x = var_49620_cast_fp16, y = var_49621_to_fp16)[name = tensor("aw_4211_cast_fp16")]; + tensor var_49624_equation_0 = const()[name = tensor("op_49624_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_49624_cast_fp16 = einsum(equation = var_49624_equation_0, values = (var_49466_cast_fp16, var_49383_cast_fp16))[name = tensor("op_49624_cast_fp16")]; + tensor var_49625_to_fp16 = const()[name = tensor("op_49625_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4213_cast_fp16 = mul(x = var_49624_cast_fp16, y = var_49625_to_fp16)[name = tensor("aw_4213_cast_fp16")]; + tensor var_49628_equation_0 = const()[name = tensor("op_49628_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_49628_cast_fp16 = einsum(equation = var_49628_equation_0, values = (var_49470_cast_fp16, var_49387_cast_fp16))[name = tensor("op_49628_cast_fp16")]; + tensor var_49629_to_fp16 = const()[name = tensor("op_49629_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4215_cast_fp16 = mul(x = var_49628_cast_fp16, y = var_49629_to_fp16)[name = tensor("aw_4215_cast_fp16")]; + tensor var_49632_equation_0 = const()[name = tensor("op_49632_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_49632_cast_fp16 = einsum(equation = var_49632_equation_0, values = (var_49474_cast_fp16, var_49391_cast_fp16))[name = tensor("op_49632_cast_fp16")]; + tensor var_49633_to_fp16 = const()[name = tensor("op_49633_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4217_cast_fp16 = mul(x = var_49632_cast_fp16, y = var_49633_to_fp16)[name = tensor("aw_4217_cast_fp16")]; + tensor var_49636_equation_0 = const()[name = tensor("op_49636_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_49636_cast_fp16 = einsum(equation = var_49636_equation_0, values = (var_49478_cast_fp16, var_49395_cast_fp16))[name = tensor("op_49636_cast_fp16")]; + tensor var_49637_to_fp16 = const()[name = tensor("op_49637_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4219_cast_fp16 = mul(x = var_49636_cast_fp16, y = var_49637_to_fp16)[name = tensor("aw_4219_cast_fp16")]; + tensor var_49640_equation_0 = const()[name = tensor("op_49640_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_49640_cast_fp16 = einsum(equation = var_49640_equation_0, values = (var_49482_cast_fp16, var_49399_cast_fp16))[name = tensor("op_49640_cast_fp16")]; + tensor var_49641_to_fp16 = const()[name = tensor("op_49641_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4221_cast_fp16 = mul(x = var_49640_cast_fp16, y = var_49641_to_fp16)[name = tensor("aw_4221_cast_fp16")]; + tensor var_49644_equation_0 = const()[name = tensor("op_49644_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_49644_cast_fp16 = einsum(equation = var_49644_equation_0, values = (var_49486_cast_fp16, var_49403_cast_fp16))[name = tensor("op_49644_cast_fp16")]; + tensor var_49645_to_fp16 = const()[name = tensor("op_49645_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4223_cast_fp16 = mul(x = var_49644_cast_fp16, y = var_49645_to_fp16)[name = tensor("aw_4223_cast_fp16")]; + tensor var_49648_equation_0 = const()[name = tensor("op_49648_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_49648_cast_fp16 = einsum(equation = var_49648_equation_0, values = (var_49490_cast_fp16, var_49407_cast_fp16))[name = tensor("op_49648_cast_fp16")]; + tensor var_49649_to_fp16 = const()[name = tensor("op_49649_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4225_cast_fp16 = mul(x = var_49648_cast_fp16, y = var_49649_to_fp16)[name = tensor("aw_4225_cast_fp16")]; + tensor var_49652_equation_0 = const()[name = tensor("op_49652_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_49652_cast_fp16 = einsum(equation = var_49652_equation_0, values = (var_49494_cast_fp16, var_49411_cast_fp16))[name = tensor("op_49652_cast_fp16")]; + tensor var_49653_to_fp16 = const()[name = tensor("op_49653_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4227_cast_fp16 = mul(x = var_49652_cast_fp16, y = var_49653_to_fp16)[name = tensor("aw_4227_cast_fp16")]; + tensor var_49656_equation_0 = const()[name = tensor("op_49656_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_49656_cast_fp16 = einsum(equation = var_49656_equation_0, values = (var_49498_cast_fp16, var_49415_cast_fp16))[name = tensor("op_49656_cast_fp16")]; + tensor var_49657_to_fp16 = const()[name = tensor("op_49657_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4229_cast_fp16 = mul(x = var_49656_cast_fp16, y = var_49657_to_fp16)[name = tensor("aw_4229_cast_fp16")]; + tensor var_49660_equation_0 = const()[name = tensor("op_49660_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_49660_cast_fp16 = einsum(equation = var_49660_equation_0, values = (var_49502_cast_fp16, var_49419_cast_fp16))[name = tensor("op_49660_cast_fp16")]; + tensor var_49661_to_fp16 = const()[name = tensor("op_49661_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4231_cast_fp16 = mul(x = var_49660_cast_fp16, y = var_49661_to_fp16)[name = tensor("aw_4231_cast_fp16")]; + tensor var_49664_equation_0 = const()[name = tensor("op_49664_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_49664_cast_fp16 = einsum(equation = var_49664_equation_0, values = (var_49506_cast_fp16, var_49423_cast_fp16))[name = tensor("op_49664_cast_fp16")]; + tensor var_49665_to_fp16 = const()[name = tensor("op_49665_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4233_cast_fp16 = mul(x = var_49664_cast_fp16, y = var_49665_to_fp16)[name = tensor("aw_4233_cast_fp16")]; + tensor var_49668_equation_0 = const()[name = tensor("op_49668_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_49668_cast_fp16 = einsum(equation = var_49668_equation_0, values = (var_49510_cast_fp16, var_49427_cast_fp16))[name = tensor("op_49668_cast_fp16")]; + tensor var_49669_to_fp16 = const()[name = tensor("op_49669_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4235_cast_fp16 = mul(x = var_49668_cast_fp16, y = var_49669_to_fp16)[name = tensor("aw_4235_cast_fp16")]; + tensor var_49672_equation_0 = const()[name = tensor("op_49672_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_49672_cast_fp16 = einsum(equation = var_49672_equation_0, values = (var_49514_cast_fp16, var_49431_cast_fp16))[name = tensor("op_49672_cast_fp16")]; + tensor var_49673_to_fp16 = const()[name = tensor("op_49673_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4237_cast_fp16 = mul(x = var_49672_cast_fp16, y = var_49673_to_fp16)[name = tensor("aw_4237_cast_fp16")]; + tensor var_49676_equation_0 = const()[name = tensor("op_49676_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_49676_cast_fp16 = einsum(equation = var_49676_equation_0, values = (var_49518_cast_fp16, var_49435_cast_fp16))[name = tensor("op_49676_cast_fp16")]; + tensor var_49677_to_fp16 = const()[name = tensor("op_49677_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4239_cast_fp16 = mul(x = var_49676_cast_fp16, y = var_49677_to_fp16)[name = tensor("aw_4239_cast_fp16")]; + tensor var_49679_cast_fp16 = softmax(axis = var_30355, x = aw_4201_cast_fp16)[name = tensor("op_49679_cast_fp16")]; + tensor var_49680_cast_fp16 = softmax(axis = var_30355, x = aw_4203_cast_fp16)[name = tensor("op_49680_cast_fp16")]; + tensor var_49681_cast_fp16 = softmax(axis = var_30355, x = aw_4205_cast_fp16)[name = tensor("op_49681_cast_fp16")]; + tensor var_49682_cast_fp16 = softmax(axis = var_30355, x = aw_4207_cast_fp16)[name = tensor("op_49682_cast_fp16")]; + tensor var_49683_cast_fp16 = softmax(axis = var_30355, x = aw_4209_cast_fp16)[name = tensor("op_49683_cast_fp16")]; + tensor var_49684_cast_fp16 = softmax(axis = var_30355, x = aw_4211_cast_fp16)[name = tensor("op_49684_cast_fp16")]; + tensor var_49685_cast_fp16 = softmax(axis = var_30355, x = aw_4213_cast_fp16)[name = tensor("op_49685_cast_fp16")]; + tensor var_49686_cast_fp16 = softmax(axis = var_30355, x = aw_4215_cast_fp16)[name = tensor("op_49686_cast_fp16")]; + tensor var_49687_cast_fp16 = softmax(axis = var_30355, x = aw_4217_cast_fp16)[name = tensor("op_49687_cast_fp16")]; + tensor var_49688_cast_fp16 = softmax(axis = var_30355, x = aw_4219_cast_fp16)[name = tensor("op_49688_cast_fp16")]; + tensor var_49689_cast_fp16 = softmax(axis = var_30355, x = aw_4221_cast_fp16)[name = tensor("op_49689_cast_fp16")]; + tensor var_49690_cast_fp16 = softmax(axis = var_30355, x = aw_4223_cast_fp16)[name = tensor("op_49690_cast_fp16")]; + tensor var_49691_cast_fp16 = softmax(axis = var_30355, x = aw_4225_cast_fp16)[name = tensor("op_49691_cast_fp16")]; + tensor var_49692_cast_fp16 = softmax(axis = var_30355, x = aw_4227_cast_fp16)[name = tensor("op_49692_cast_fp16")]; + tensor var_49693_cast_fp16 = softmax(axis = var_30355, x = aw_4229_cast_fp16)[name = tensor("op_49693_cast_fp16")]; + tensor var_49694_cast_fp16 = softmax(axis = var_30355, x = aw_4231_cast_fp16)[name = tensor("op_49694_cast_fp16")]; + tensor var_49695_cast_fp16 = softmax(axis = var_30355, x = aw_4233_cast_fp16)[name = tensor("op_49695_cast_fp16")]; + tensor var_49696_cast_fp16 = softmax(axis = var_30355, x = aw_4235_cast_fp16)[name = tensor("op_49696_cast_fp16")]; + tensor var_49697_cast_fp16 = softmax(axis = var_30355, x = aw_4237_cast_fp16)[name = tensor("op_49697_cast_fp16")]; + tensor var_49698_cast_fp16 = softmax(axis = var_30355, x = aw_4239_cast_fp16)[name = tensor("op_49698_cast_fp16")]; + tensor var_49700_equation_0 = const()[name = tensor("op_49700_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49700_cast_fp16 = einsum(equation = var_49700_equation_0, values = (var_49520_cast_fp16, var_49679_cast_fp16))[name = tensor("op_49700_cast_fp16")]; + tensor var_49702_equation_0 = const()[name = tensor("op_49702_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49702_cast_fp16 = einsum(equation = var_49702_equation_0, values = (var_49524_cast_fp16, var_49680_cast_fp16))[name = tensor("op_49702_cast_fp16")]; + tensor var_49704_equation_0 = const()[name = tensor("op_49704_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49704_cast_fp16 = einsum(equation = var_49704_equation_0, values = (var_49528_cast_fp16, var_49681_cast_fp16))[name = tensor("op_49704_cast_fp16")]; + tensor var_49706_equation_0 = const()[name = tensor("op_49706_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49706_cast_fp16 = einsum(equation = var_49706_equation_0, values = (var_49532_cast_fp16, var_49682_cast_fp16))[name = tensor("op_49706_cast_fp16")]; + tensor var_49708_equation_0 = const()[name = tensor("op_49708_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49708_cast_fp16 = einsum(equation = var_49708_equation_0, values = (var_49536_cast_fp16, var_49683_cast_fp16))[name = tensor("op_49708_cast_fp16")]; + tensor var_49710_equation_0 = const()[name = tensor("op_49710_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49710_cast_fp16 = einsum(equation = var_49710_equation_0, values = (var_49540_cast_fp16, var_49684_cast_fp16))[name = tensor("op_49710_cast_fp16")]; + tensor var_49712_equation_0 = const()[name = tensor("op_49712_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49712_cast_fp16 = einsum(equation = var_49712_equation_0, values = (var_49544_cast_fp16, var_49685_cast_fp16))[name = tensor("op_49712_cast_fp16")]; + tensor var_49714_equation_0 = const()[name = tensor("op_49714_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49714_cast_fp16 = einsum(equation = var_49714_equation_0, values = (var_49548_cast_fp16, var_49686_cast_fp16))[name = tensor("op_49714_cast_fp16")]; + tensor var_49716_equation_0 = const()[name = tensor("op_49716_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49716_cast_fp16 = einsum(equation = var_49716_equation_0, values = (var_49552_cast_fp16, var_49687_cast_fp16))[name = tensor("op_49716_cast_fp16")]; + tensor var_49718_equation_0 = const()[name = tensor("op_49718_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49718_cast_fp16 = einsum(equation = var_49718_equation_0, values = (var_49556_cast_fp16, var_49688_cast_fp16))[name = tensor("op_49718_cast_fp16")]; + tensor var_49720_equation_0 = const()[name = tensor("op_49720_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49720_cast_fp16 = einsum(equation = var_49720_equation_0, values = (var_49560_cast_fp16, var_49689_cast_fp16))[name = tensor("op_49720_cast_fp16")]; + tensor var_49722_equation_0 = const()[name = tensor("op_49722_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49722_cast_fp16 = einsum(equation = var_49722_equation_0, values = (var_49564_cast_fp16, var_49690_cast_fp16))[name = tensor("op_49722_cast_fp16")]; + tensor var_49724_equation_0 = const()[name = tensor("op_49724_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49724_cast_fp16 = einsum(equation = var_49724_equation_0, values = (var_49568_cast_fp16, var_49691_cast_fp16))[name = tensor("op_49724_cast_fp16")]; + tensor var_49726_equation_0 = const()[name = tensor("op_49726_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49726_cast_fp16 = einsum(equation = var_49726_equation_0, values = (var_49572_cast_fp16, var_49692_cast_fp16))[name = tensor("op_49726_cast_fp16")]; + tensor var_49728_equation_0 = const()[name = tensor("op_49728_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49728_cast_fp16 = einsum(equation = var_49728_equation_0, values = (var_49576_cast_fp16, var_49693_cast_fp16))[name = tensor("op_49728_cast_fp16")]; + tensor var_49730_equation_0 = const()[name = tensor("op_49730_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49730_cast_fp16 = einsum(equation = var_49730_equation_0, values = (var_49580_cast_fp16, var_49694_cast_fp16))[name = tensor("op_49730_cast_fp16")]; + tensor var_49732_equation_0 = const()[name = tensor("op_49732_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49732_cast_fp16 = einsum(equation = var_49732_equation_0, values = (var_49584_cast_fp16, var_49695_cast_fp16))[name = tensor("op_49732_cast_fp16")]; + tensor var_49734_equation_0 = const()[name = tensor("op_49734_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49734_cast_fp16 = einsum(equation = var_49734_equation_0, values = (var_49588_cast_fp16, var_49696_cast_fp16))[name = tensor("op_49734_cast_fp16")]; + tensor var_49736_equation_0 = const()[name = tensor("op_49736_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49736_cast_fp16 = einsum(equation = var_49736_equation_0, values = (var_49592_cast_fp16, var_49697_cast_fp16))[name = tensor("op_49736_cast_fp16")]; + tensor var_49738_equation_0 = const()[name = tensor("op_49738_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49738_cast_fp16 = einsum(equation = var_49738_equation_0, values = (var_49596_cast_fp16, var_49698_cast_fp16))[name = tensor("op_49738_cast_fp16")]; + tensor input_637_interleave_0 = const()[name = tensor("input_637_interleave_0"), val = tensor(false)]; + tensor input_637_cast_fp16 = concat(axis = var_30355, interleave = input_637_interleave_0, values = (var_49700_cast_fp16, var_49702_cast_fp16, var_49704_cast_fp16, var_49706_cast_fp16, var_49708_cast_fp16, var_49710_cast_fp16, var_49712_cast_fp16, var_49714_cast_fp16, var_49716_cast_fp16, var_49718_cast_fp16, var_49720_cast_fp16, var_49722_cast_fp16, var_49724_cast_fp16, var_49726_cast_fp16, var_49728_cast_fp16, var_49730_cast_fp16, var_49732_cast_fp16, var_49734_cast_fp16, var_49736_cast_fp16, var_49738_cast_fp16))[name = tensor("input_637_cast_fp16")]; + tensor var_49744 = const()[name = tensor("op_49744"), val = tensor([1, 1])]; + tensor var_49746 = const()[name = tensor("op_49746"), val = tensor([1, 1])]; + tensor var_49748_pad_type_0 = const()[name = tensor("op_49748_pad_type_0"), val = tensor("custom")]; + tensor var_49748_pad_0 = const()[name = tensor("op_49748_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_0_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_0_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4198974336)))]; + tensor up_blocks_0_attentions_2_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4202251200)))]; + tensor var_49748_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_49746, groups = var_30355, pad = var_49748_pad_0, pad_type = var_49748_pad_type_0, strides = var_49744, weight = up_blocks_0_attentions_2_transformer_blocks_0_attn2_to_out_0_weight_to_fp16, x = input_637_cast_fp16)[name = tensor("op_49748_cast_fp16")]; + tensor inputs_329_cast_fp16 = add(x = var_49748_cast_fp16, y = inputs_327_cast_fp16)[name = tensor("inputs_329_cast_fp16")]; + tensor input_639_axes_0 = const()[name = tensor("input_639_axes_0"), val = tensor([1])]; + tensor input_639_gamma_0_to_fp16 = const()[name = tensor("input_639_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4202253824)))]; + tensor input_639_beta_0_to_fp16 = const()[name = tensor("input_639_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4202256448)))]; + tensor var_49758_to_fp16 = const()[name = tensor("op_49758_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_639_cast_fp16 = layer_norm(axes = input_639_axes_0, beta = input_639_beta_0_to_fp16, epsilon = var_49758_to_fp16, gamma = input_639_gamma_0_to_fp16, x = inputs_329_cast_fp16)[name = tensor("input_639_cast_fp16")]; + tensor var_49774 = const()[name = tensor("op_49774"), val = tensor([1, 1])]; + tensor var_49776 = const()[name = tensor("op_49776"), val = tensor([1, 1])]; + tensor var_49778_pad_type_0 = const()[name = tensor("op_49778_pad_type_0"), val = tensor("custom")]; + tensor var_49778_pad_0 = const()[name = tensor("op_49778_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_0_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_0_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4202259072)))]; + tensor up_blocks_0_attentions_2_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4228473536)))]; + tensor var_49778_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_49776, groups = var_30355, pad = var_49778_pad_0, pad_type = var_49778_pad_type_0, strides = var_49774, weight = up_blocks_0_attentions_2_transformer_blocks_0_ff_net_0_proj_weight_to_fp16, x = input_639_cast_fp16)[name = tensor("op_49778_cast_fp16")]; + tensor var_49779_split_sizes_0 = const()[name = tensor("op_49779_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_49779_axis_0 = const()[name = tensor("op_49779_axis_0"), val = tensor(1)]; + tensor var_49779_cast_fp16_0, tensor var_49779_cast_fp16_1 = split(axis = var_49779_axis_0, split_sizes = var_49779_split_sizes_0, x = var_49778_cast_fp16)[name = tensor("op_49779_cast_fp16")]; + tensor var_49781_mode_0 = const()[name = tensor("op_49781_mode_0"), val = tensor("EXACT")]; + tensor var_49781_cast_fp16 = gelu(mode = var_49781_mode_0, x = var_49779_cast_fp16_1)[name = tensor("op_49781_cast_fp16")]; + tensor input_641_cast_fp16 = mul(x = var_49779_cast_fp16_0, y = var_49781_cast_fp16)[name = tensor("input_641_cast_fp16")]; + tensor var_49785 = const()[name = tensor("op_49785"), val = tensor([1, 1])]; + tensor var_49787 = const()[name = tensor("op_49787"), val = tensor([1, 1])]; + tensor var_49789_pad_type_0 = const()[name = tensor("op_49789_pad_type_0"), val = tensor("custom")]; + tensor var_49789_pad_0 = const()[name = tensor("op_49789_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_0_ff_net_2_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_0_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4228494080)))]; + tensor up_blocks_0_attentions_2_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4241601344)))]; + tensor var_49789_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_49787, groups = var_30355, pad = var_49789_pad_0, pad_type = var_49789_pad_type_0, strides = var_49785, weight = up_blocks_0_attentions_2_transformer_blocks_0_ff_net_2_weight_to_fp16, x = input_641_cast_fp16)[name = tensor("op_49789_cast_fp16")]; + tensor inputs_331_cast_fp16 = add(x = var_49789_cast_fp16, y = inputs_329_cast_fp16)[name = tensor("inputs_331_cast_fp16")]; + tensor hidden_states_441_axes_0 = const()[name = tensor("hidden_states_441_axes_0"), val = tensor([1])]; + tensor hidden_states_441_gamma_0_to_fp16 = const()[name = tensor("hidden_states_441_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4241603968)))]; + tensor hidden_states_441_beta_0_to_fp16 = const()[name = tensor("hidden_states_441_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4241606592)))]; + tensor var_49805_to_fp16 = const()[name = tensor("op_49805_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_441_cast_fp16 = layer_norm(axes = hidden_states_441_axes_0, beta = hidden_states_441_beta_0_to_fp16, epsilon = var_49805_to_fp16, gamma = hidden_states_441_gamma_0_to_fp16, x = inputs_331_cast_fp16)[name = tensor("hidden_states_441_cast_fp16")]; + tensor var_49820 = const()[name = tensor("op_49820"), val = tensor([1, 1])]; + tensor var_49822 = const()[name = tensor("op_49822"), val = tensor([1, 1])]; + tensor q_221_pad_type_0 = const()[name = tensor("q_221_pad_type_0"), val = tensor("custom")]; + tensor q_221_pad_0 = const()[name = tensor("q_221_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_1_attn1_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_1_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4241609216)))]; + tensor q_221_cast_fp16 = conv(dilations = var_49822, groups = var_30355, pad = q_221_pad_0, pad_type = q_221_pad_type_0, strides = var_49820, weight = up_blocks_0_attentions_2_transformer_blocks_1_attn1_to_q_weight_to_fp16, x = hidden_states_441_cast_fp16)[name = tensor("q_221_cast_fp16")]; + tensor var_49826 = const()[name = tensor("op_49826"), val = tensor([1, 1])]; + tensor var_49828 = const()[name = tensor("op_49828"), val = tensor([1, 1])]; + tensor k_441_pad_type_0 = const()[name = tensor("k_441_pad_type_0"), val = tensor("custom")]; + tensor k_441_pad_0 = const()[name = tensor("k_441_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_1_attn1_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_1_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4244886080)))]; + tensor k_441_cast_fp16 = conv(dilations = var_49828, groups = var_30355, pad = k_441_pad_0, pad_type = k_441_pad_type_0, strides = var_49826, weight = up_blocks_0_attentions_2_transformer_blocks_1_attn1_to_k_weight_to_fp16, x = hidden_states_441_cast_fp16)[name = tensor("k_441_cast_fp16")]; + tensor var_49832 = const()[name = tensor("op_49832"), val = tensor([1, 1])]; + tensor var_49834 = const()[name = tensor("op_49834"), val = tensor([1, 1])]; + tensor v_221_pad_type_0 = const()[name = tensor("v_221_pad_type_0"), val = tensor("custom")]; + tensor v_221_pad_0 = const()[name = tensor("v_221_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_1_attn1_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_1_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4248162944)))]; + tensor v_221_cast_fp16 = conv(dilations = var_49834, groups = var_30355, pad = v_221_pad_0, pad_type = v_221_pad_type_0, strides = var_49832, weight = up_blocks_0_attentions_2_transformer_blocks_1_attn1_to_v_weight_to_fp16, x = hidden_states_441_cast_fp16)[name = tensor("v_221_cast_fp16")]; + tensor var_49838_begin_0 = const()[name = tensor("op_49838_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_49838_end_0 = const()[name = tensor("op_49838_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_49838_end_mask_0 = const()[name = tensor("op_49838_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49838_cast_fp16 = slice_by_index(begin = var_49838_begin_0, end = var_49838_end_0, end_mask = var_49838_end_mask_0, x = q_221_cast_fp16)[name = tensor("op_49838_cast_fp16")]; + tensor var_49842_begin_0 = const()[name = tensor("op_49842_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_49842_end_0 = const()[name = tensor("op_49842_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_49842_end_mask_0 = const()[name = tensor("op_49842_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49842_cast_fp16 = slice_by_index(begin = var_49842_begin_0, end = var_49842_end_0, end_mask = var_49842_end_mask_0, x = q_221_cast_fp16)[name = tensor("op_49842_cast_fp16")]; + tensor var_49846_begin_0 = const()[name = tensor("op_49846_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_49846_end_0 = const()[name = tensor("op_49846_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_49846_end_mask_0 = const()[name = tensor("op_49846_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49846_cast_fp16 = slice_by_index(begin = var_49846_begin_0, end = var_49846_end_0, end_mask = var_49846_end_mask_0, x = q_221_cast_fp16)[name = tensor("op_49846_cast_fp16")]; + tensor var_49850_begin_0 = const()[name = tensor("op_49850_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_49850_end_0 = const()[name = tensor("op_49850_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_49850_end_mask_0 = const()[name = tensor("op_49850_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49850_cast_fp16 = slice_by_index(begin = var_49850_begin_0, end = var_49850_end_0, end_mask = var_49850_end_mask_0, x = q_221_cast_fp16)[name = tensor("op_49850_cast_fp16")]; + tensor var_49854_begin_0 = const()[name = tensor("op_49854_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_49854_end_0 = const()[name = tensor("op_49854_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_49854_end_mask_0 = const()[name = tensor("op_49854_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49854_cast_fp16 = slice_by_index(begin = var_49854_begin_0, end = var_49854_end_0, end_mask = var_49854_end_mask_0, x = q_221_cast_fp16)[name = tensor("op_49854_cast_fp16")]; + tensor var_49858_begin_0 = const()[name = tensor("op_49858_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_49858_end_0 = const()[name = tensor("op_49858_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_49858_end_mask_0 = const()[name = tensor("op_49858_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49858_cast_fp16 = slice_by_index(begin = var_49858_begin_0, end = var_49858_end_0, end_mask = var_49858_end_mask_0, x = q_221_cast_fp16)[name = tensor("op_49858_cast_fp16")]; + tensor var_49862_begin_0 = const()[name = tensor("op_49862_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_49862_end_0 = const()[name = tensor("op_49862_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_49862_end_mask_0 = const()[name = tensor("op_49862_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49862_cast_fp16 = slice_by_index(begin = var_49862_begin_0, end = var_49862_end_0, end_mask = var_49862_end_mask_0, x = q_221_cast_fp16)[name = tensor("op_49862_cast_fp16")]; + tensor var_49866_begin_0 = const()[name = tensor("op_49866_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_49866_end_0 = const()[name = tensor("op_49866_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_49866_end_mask_0 = const()[name = tensor("op_49866_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49866_cast_fp16 = slice_by_index(begin = var_49866_begin_0, end = var_49866_end_0, end_mask = var_49866_end_mask_0, x = q_221_cast_fp16)[name = tensor("op_49866_cast_fp16")]; + tensor var_49870_begin_0 = const()[name = tensor("op_49870_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_49870_end_0 = const()[name = tensor("op_49870_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_49870_end_mask_0 = const()[name = tensor("op_49870_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49870_cast_fp16 = slice_by_index(begin = var_49870_begin_0, end = var_49870_end_0, end_mask = var_49870_end_mask_0, x = q_221_cast_fp16)[name = tensor("op_49870_cast_fp16")]; + tensor var_49874_begin_0 = const()[name = tensor("op_49874_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_49874_end_0 = const()[name = tensor("op_49874_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_49874_end_mask_0 = const()[name = tensor("op_49874_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49874_cast_fp16 = slice_by_index(begin = var_49874_begin_0, end = var_49874_end_0, end_mask = var_49874_end_mask_0, x = q_221_cast_fp16)[name = tensor("op_49874_cast_fp16")]; + tensor var_49878_begin_0 = const()[name = tensor("op_49878_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_49878_end_0 = const()[name = tensor("op_49878_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_49878_end_mask_0 = const()[name = tensor("op_49878_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49878_cast_fp16 = slice_by_index(begin = var_49878_begin_0, end = var_49878_end_0, end_mask = var_49878_end_mask_0, x = q_221_cast_fp16)[name = tensor("op_49878_cast_fp16")]; + tensor var_49882_begin_0 = const()[name = tensor("op_49882_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_49882_end_0 = const()[name = tensor("op_49882_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_49882_end_mask_0 = const()[name = tensor("op_49882_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49882_cast_fp16 = slice_by_index(begin = var_49882_begin_0, end = var_49882_end_0, end_mask = var_49882_end_mask_0, x = q_221_cast_fp16)[name = tensor("op_49882_cast_fp16")]; + tensor var_49886_begin_0 = const()[name = tensor("op_49886_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_49886_end_0 = const()[name = tensor("op_49886_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_49886_end_mask_0 = const()[name = tensor("op_49886_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49886_cast_fp16 = slice_by_index(begin = var_49886_begin_0, end = var_49886_end_0, end_mask = var_49886_end_mask_0, x = q_221_cast_fp16)[name = tensor("op_49886_cast_fp16")]; + tensor var_49890_begin_0 = const()[name = tensor("op_49890_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_49890_end_0 = const()[name = tensor("op_49890_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_49890_end_mask_0 = const()[name = tensor("op_49890_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49890_cast_fp16 = slice_by_index(begin = var_49890_begin_0, end = var_49890_end_0, end_mask = var_49890_end_mask_0, x = q_221_cast_fp16)[name = tensor("op_49890_cast_fp16")]; + tensor var_49894_begin_0 = const()[name = tensor("op_49894_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_49894_end_0 = const()[name = tensor("op_49894_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_49894_end_mask_0 = const()[name = tensor("op_49894_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49894_cast_fp16 = slice_by_index(begin = var_49894_begin_0, end = var_49894_end_0, end_mask = var_49894_end_mask_0, x = q_221_cast_fp16)[name = tensor("op_49894_cast_fp16")]; + tensor var_49898_begin_0 = const()[name = tensor("op_49898_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_49898_end_0 = const()[name = tensor("op_49898_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_49898_end_mask_0 = const()[name = tensor("op_49898_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49898_cast_fp16 = slice_by_index(begin = var_49898_begin_0, end = var_49898_end_0, end_mask = var_49898_end_mask_0, x = q_221_cast_fp16)[name = tensor("op_49898_cast_fp16")]; + tensor var_49902_begin_0 = const()[name = tensor("op_49902_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_49902_end_0 = const()[name = tensor("op_49902_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_49902_end_mask_0 = const()[name = tensor("op_49902_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49902_cast_fp16 = slice_by_index(begin = var_49902_begin_0, end = var_49902_end_0, end_mask = var_49902_end_mask_0, x = q_221_cast_fp16)[name = tensor("op_49902_cast_fp16")]; + tensor var_49906_begin_0 = const()[name = tensor("op_49906_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_49906_end_0 = const()[name = tensor("op_49906_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_49906_end_mask_0 = const()[name = tensor("op_49906_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49906_cast_fp16 = slice_by_index(begin = var_49906_begin_0, end = var_49906_end_0, end_mask = var_49906_end_mask_0, x = q_221_cast_fp16)[name = tensor("op_49906_cast_fp16")]; + tensor var_49910_begin_0 = const()[name = tensor("op_49910_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_49910_end_0 = const()[name = tensor("op_49910_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_49910_end_mask_0 = const()[name = tensor("op_49910_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49910_cast_fp16 = slice_by_index(begin = var_49910_begin_0, end = var_49910_end_0, end_mask = var_49910_end_mask_0, x = q_221_cast_fp16)[name = tensor("op_49910_cast_fp16")]; + tensor var_49914_begin_0 = const()[name = tensor("op_49914_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_49914_end_0 = const()[name = tensor("op_49914_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_49914_end_mask_0 = const()[name = tensor("op_49914_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49914_cast_fp16 = slice_by_index(begin = var_49914_begin_0, end = var_49914_end_0, end_mask = var_49914_end_mask_0, x = q_221_cast_fp16)[name = tensor("op_49914_cast_fp16")]; + tensor k_443_perm_0 = const()[name = tensor("k_443_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_49921_begin_0 = const()[name = tensor("op_49921_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_49921_end_0 = const()[name = tensor("op_49921_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_49921_end_mask_0 = const()[name = tensor("op_49921_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_443_cast_fp16 = transpose(perm = k_443_perm_0, x = k_441_cast_fp16)[name = tensor("transpose_29")]; + tensor var_49921_cast_fp16 = slice_by_index(begin = var_49921_begin_0, end = var_49921_end_0, end_mask = var_49921_end_mask_0, x = k_443_cast_fp16)[name = tensor("op_49921_cast_fp16")]; + tensor var_49925_begin_0 = const()[name = tensor("op_49925_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_49925_end_0 = const()[name = tensor("op_49925_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_49925_end_mask_0 = const()[name = tensor("op_49925_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49925_cast_fp16 = slice_by_index(begin = var_49925_begin_0, end = var_49925_end_0, end_mask = var_49925_end_mask_0, x = k_443_cast_fp16)[name = tensor("op_49925_cast_fp16")]; + tensor var_49929_begin_0 = const()[name = tensor("op_49929_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_49929_end_0 = const()[name = tensor("op_49929_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_49929_end_mask_0 = const()[name = tensor("op_49929_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49929_cast_fp16 = slice_by_index(begin = var_49929_begin_0, end = var_49929_end_0, end_mask = var_49929_end_mask_0, x = k_443_cast_fp16)[name = tensor("op_49929_cast_fp16")]; + tensor var_49933_begin_0 = const()[name = tensor("op_49933_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_49933_end_0 = const()[name = tensor("op_49933_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_49933_end_mask_0 = const()[name = tensor("op_49933_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49933_cast_fp16 = slice_by_index(begin = var_49933_begin_0, end = var_49933_end_0, end_mask = var_49933_end_mask_0, x = k_443_cast_fp16)[name = tensor("op_49933_cast_fp16")]; + tensor var_49937_begin_0 = const()[name = tensor("op_49937_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_49937_end_0 = const()[name = tensor("op_49937_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_49937_end_mask_0 = const()[name = tensor("op_49937_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49937_cast_fp16 = slice_by_index(begin = var_49937_begin_0, end = var_49937_end_0, end_mask = var_49937_end_mask_0, x = k_443_cast_fp16)[name = tensor("op_49937_cast_fp16")]; + tensor var_49941_begin_0 = const()[name = tensor("op_49941_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_49941_end_0 = const()[name = tensor("op_49941_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_49941_end_mask_0 = const()[name = tensor("op_49941_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49941_cast_fp16 = slice_by_index(begin = var_49941_begin_0, end = var_49941_end_0, end_mask = var_49941_end_mask_0, x = k_443_cast_fp16)[name = tensor("op_49941_cast_fp16")]; + tensor var_49945_begin_0 = const()[name = tensor("op_49945_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_49945_end_0 = const()[name = tensor("op_49945_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_49945_end_mask_0 = const()[name = tensor("op_49945_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49945_cast_fp16 = slice_by_index(begin = var_49945_begin_0, end = var_49945_end_0, end_mask = var_49945_end_mask_0, x = k_443_cast_fp16)[name = tensor("op_49945_cast_fp16")]; + tensor var_49949_begin_0 = const()[name = tensor("op_49949_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_49949_end_0 = const()[name = tensor("op_49949_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_49949_end_mask_0 = const()[name = tensor("op_49949_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49949_cast_fp16 = slice_by_index(begin = var_49949_begin_0, end = var_49949_end_0, end_mask = var_49949_end_mask_0, x = k_443_cast_fp16)[name = tensor("op_49949_cast_fp16")]; + tensor var_49953_begin_0 = const()[name = tensor("op_49953_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_49953_end_0 = const()[name = tensor("op_49953_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_49953_end_mask_0 = const()[name = tensor("op_49953_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49953_cast_fp16 = slice_by_index(begin = var_49953_begin_0, end = var_49953_end_0, end_mask = var_49953_end_mask_0, x = k_443_cast_fp16)[name = tensor("op_49953_cast_fp16")]; + tensor var_49957_begin_0 = const()[name = tensor("op_49957_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_49957_end_0 = const()[name = tensor("op_49957_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_49957_end_mask_0 = const()[name = tensor("op_49957_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49957_cast_fp16 = slice_by_index(begin = var_49957_begin_0, end = var_49957_end_0, end_mask = var_49957_end_mask_0, x = k_443_cast_fp16)[name = tensor("op_49957_cast_fp16")]; + tensor var_49961_begin_0 = const()[name = tensor("op_49961_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_49961_end_0 = const()[name = tensor("op_49961_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_49961_end_mask_0 = const()[name = tensor("op_49961_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49961_cast_fp16 = slice_by_index(begin = var_49961_begin_0, end = var_49961_end_0, end_mask = var_49961_end_mask_0, x = k_443_cast_fp16)[name = tensor("op_49961_cast_fp16")]; + tensor var_49965_begin_0 = const()[name = tensor("op_49965_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_49965_end_0 = const()[name = tensor("op_49965_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_49965_end_mask_0 = const()[name = tensor("op_49965_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49965_cast_fp16 = slice_by_index(begin = var_49965_begin_0, end = var_49965_end_0, end_mask = var_49965_end_mask_0, x = k_443_cast_fp16)[name = tensor("op_49965_cast_fp16")]; + tensor var_49969_begin_0 = const()[name = tensor("op_49969_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_49969_end_0 = const()[name = tensor("op_49969_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_49969_end_mask_0 = const()[name = tensor("op_49969_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49969_cast_fp16 = slice_by_index(begin = var_49969_begin_0, end = var_49969_end_0, end_mask = var_49969_end_mask_0, x = k_443_cast_fp16)[name = tensor("op_49969_cast_fp16")]; + tensor var_49973_begin_0 = const()[name = tensor("op_49973_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_49973_end_0 = const()[name = tensor("op_49973_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_49973_end_mask_0 = const()[name = tensor("op_49973_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49973_cast_fp16 = slice_by_index(begin = var_49973_begin_0, end = var_49973_end_0, end_mask = var_49973_end_mask_0, x = k_443_cast_fp16)[name = tensor("op_49973_cast_fp16")]; + tensor var_49977_begin_0 = const()[name = tensor("op_49977_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_49977_end_0 = const()[name = tensor("op_49977_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_49977_end_mask_0 = const()[name = tensor("op_49977_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49977_cast_fp16 = slice_by_index(begin = var_49977_begin_0, end = var_49977_end_0, end_mask = var_49977_end_mask_0, x = k_443_cast_fp16)[name = tensor("op_49977_cast_fp16")]; + tensor var_49981_begin_0 = const()[name = tensor("op_49981_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_49981_end_0 = const()[name = tensor("op_49981_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_49981_end_mask_0 = const()[name = tensor("op_49981_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49981_cast_fp16 = slice_by_index(begin = var_49981_begin_0, end = var_49981_end_0, end_mask = var_49981_end_mask_0, x = k_443_cast_fp16)[name = tensor("op_49981_cast_fp16")]; + tensor var_49985_begin_0 = const()[name = tensor("op_49985_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_49985_end_0 = const()[name = tensor("op_49985_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_49985_end_mask_0 = const()[name = tensor("op_49985_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49985_cast_fp16 = slice_by_index(begin = var_49985_begin_0, end = var_49985_end_0, end_mask = var_49985_end_mask_0, x = k_443_cast_fp16)[name = tensor("op_49985_cast_fp16")]; + tensor var_49989_begin_0 = const()[name = tensor("op_49989_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_49989_end_0 = const()[name = tensor("op_49989_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_49989_end_mask_0 = const()[name = tensor("op_49989_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49989_cast_fp16 = slice_by_index(begin = var_49989_begin_0, end = var_49989_end_0, end_mask = var_49989_end_mask_0, x = k_443_cast_fp16)[name = tensor("op_49989_cast_fp16")]; + tensor var_49993_begin_0 = const()[name = tensor("op_49993_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_49993_end_0 = const()[name = tensor("op_49993_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_49993_end_mask_0 = const()[name = tensor("op_49993_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49993_cast_fp16 = slice_by_index(begin = var_49993_begin_0, end = var_49993_end_0, end_mask = var_49993_end_mask_0, x = k_443_cast_fp16)[name = tensor("op_49993_cast_fp16")]; + tensor var_49997_begin_0 = const()[name = tensor("op_49997_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_49997_end_0 = const()[name = tensor("op_49997_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_49997_end_mask_0 = const()[name = tensor("op_49997_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49997_cast_fp16 = slice_by_index(begin = var_49997_begin_0, end = var_49997_end_0, end_mask = var_49997_end_mask_0, x = k_443_cast_fp16)[name = tensor("op_49997_cast_fp16")]; + tensor var_49999_begin_0 = const()[name = tensor("op_49999_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_49999_end_0 = const()[name = tensor("op_49999_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_49999_end_mask_0 = const()[name = tensor("op_49999_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49999_cast_fp16 = slice_by_index(begin = var_49999_begin_0, end = var_49999_end_0, end_mask = var_49999_end_mask_0, x = v_221_cast_fp16)[name = tensor("op_49999_cast_fp16")]; + tensor var_50003_begin_0 = const()[name = tensor("op_50003_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_50003_end_0 = const()[name = tensor("op_50003_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_50003_end_mask_0 = const()[name = tensor("op_50003_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50003_cast_fp16 = slice_by_index(begin = var_50003_begin_0, end = var_50003_end_0, end_mask = var_50003_end_mask_0, x = v_221_cast_fp16)[name = tensor("op_50003_cast_fp16")]; + tensor var_50007_begin_0 = const()[name = tensor("op_50007_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_50007_end_0 = const()[name = tensor("op_50007_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_50007_end_mask_0 = const()[name = tensor("op_50007_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50007_cast_fp16 = slice_by_index(begin = var_50007_begin_0, end = var_50007_end_0, end_mask = var_50007_end_mask_0, x = v_221_cast_fp16)[name = tensor("op_50007_cast_fp16")]; + tensor var_50011_begin_0 = const()[name = tensor("op_50011_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_50011_end_0 = const()[name = tensor("op_50011_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_50011_end_mask_0 = const()[name = tensor("op_50011_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50011_cast_fp16 = slice_by_index(begin = var_50011_begin_0, end = var_50011_end_0, end_mask = var_50011_end_mask_0, x = v_221_cast_fp16)[name = tensor("op_50011_cast_fp16")]; + tensor var_50015_begin_0 = const()[name = tensor("op_50015_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_50015_end_0 = const()[name = tensor("op_50015_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_50015_end_mask_0 = const()[name = tensor("op_50015_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50015_cast_fp16 = slice_by_index(begin = var_50015_begin_0, end = var_50015_end_0, end_mask = var_50015_end_mask_0, x = v_221_cast_fp16)[name = tensor("op_50015_cast_fp16")]; + tensor var_50019_begin_0 = const()[name = tensor("op_50019_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_50019_end_0 = const()[name = tensor("op_50019_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_50019_end_mask_0 = const()[name = tensor("op_50019_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50019_cast_fp16 = slice_by_index(begin = var_50019_begin_0, end = var_50019_end_0, end_mask = var_50019_end_mask_0, x = v_221_cast_fp16)[name = tensor("op_50019_cast_fp16")]; + tensor var_50023_begin_0 = const()[name = tensor("op_50023_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_50023_end_0 = const()[name = tensor("op_50023_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_50023_end_mask_0 = const()[name = tensor("op_50023_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50023_cast_fp16 = slice_by_index(begin = var_50023_begin_0, end = var_50023_end_0, end_mask = var_50023_end_mask_0, x = v_221_cast_fp16)[name = tensor("op_50023_cast_fp16")]; + tensor var_50027_begin_0 = const()[name = tensor("op_50027_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_50027_end_0 = const()[name = tensor("op_50027_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_50027_end_mask_0 = const()[name = tensor("op_50027_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50027_cast_fp16 = slice_by_index(begin = var_50027_begin_0, end = var_50027_end_0, end_mask = var_50027_end_mask_0, x = v_221_cast_fp16)[name = tensor("op_50027_cast_fp16")]; + tensor var_50031_begin_0 = const()[name = tensor("op_50031_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_50031_end_0 = const()[name = tensor("op_50031_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_50031_end_mask_0 = const()[name = tensor("op_50031_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50031_cast_fp16 = slice_by_index(begin = var_50031_begin_0, end = var_50031_end_0, end_mask = var_50031_end_mask_0, x = v_221_cast_fp16)[name = tensor("op_50031_cast_fp16")]; + tensor var_50035_begin_0 = const()[name = tensor("op_50035_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_50035_end_0 = const()[name = tensor("op_50035_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_50035_end_mask_0 = const()[name = tensor("op_50035_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50035_cast_fp16 = slice_by_index(begin = var_50035_begin_0, end = var_50035_end_0, end_mask = var_50035_end_mask_0, x = v_221_cast_fp16)[name = tensor("op_50035_cast_fp16")]; + tensor var_50039_begin_0 = const()[name = tensor("op_50039_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_50039_end_0 = const()[name = tensor("op_50039_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_50039_end_mask_0 = const()[name = tensor("op_50039_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50039_cast_fp16 = slice_by_index(begin = var_50039_begin_0, end = var_50039_end_0, end_mask = var_50039_end_mask_0, x = v_221_cast_fp16)[name = tensor("op_50039_cast_fp16")]; + tensor var_50043_begin_0 = const()[name = tensor("op_50043_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_50043_end_0 = const()[name = tensor("op_50043_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_50043_end_mask_0 = const()[name = tensor("op_50043_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50043_cast_fp16 = slice_by_index(begin = var_50043_begin_0, end = var_50043_end_0, end_mask = var_50043_end_mask_0, x = v_221_cast_fp16)[name = tensor("op_50043_cast_fp16")]; + tensor var_50047_begin_0 = const()[name = tensor("op_50047_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_50047_end_0 = const()[name = tensor("op_50047_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_50047_end_mask_0 = const()[name = tensor("op_50047_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50047_cast_fp16 = slice_by_index(begin = var_50047_begin_0, end = var_50047_end_0, end_mask = var_50047_end_mask_0, x = v_221_cast_fp16)[name = tensor("op_50047_cast_fp16")]; + tensor var_50051_begin_0 = const()[name = tensor("op_50051_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_50051_end_0 = const()[name = tensor("op_50051_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_50051_end_mask_0 = const()[name = tensor("op_50051_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50051_cast_fp16 = slice_by_index(begin = var_50051_begin_0, end = var_50051_end_0, end_mask = var_50051_end_mask_0, x = v_221_cast_fp16)[name = tensor("op_50051_cast_fp16")]; + tensor var_50055_begin_0 = const()[name = tensor("op_50055_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_50055_end_0 = const()[name = tensor("op_50055_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_50055_end_mask_0 = const()[name = tensor("op_50055_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50055_cast_fp16 = slice_by_index(begin = var_50055_begin_0, end = var_50055_end_0, end_mask = var_50055_end_mask_0, x = v_221_cast_fp16)[name = tensor("op_50055_cast_fp16")]; + tensor var_50059_begin_0 = const()[name = tensor("op_50059_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_50059_end_0 = const()[name = tensor("op_50059_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_50059_end_mask_0 = const()[name = tensor("op_50059_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50059_cast_fp16 = slice_by_index(begin = var_50059_begin_0, end = var_50059_end_0, end_mask = var_50059_end_mask_0, x = v_221_cast_fp16)[name = tensor("op_50059_cast_fp16")]; + tensor var_50063_begin_0 = const()[name = tensor("op_50063_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_50063_end_0 = const()[name = tensor("op_50063_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_50063_end_mask_0 = const()[name = tensor("op_50063_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50063_cast_fp16 = slice_by_index(begin = var_50063_begin_0, end = var_50063_end_0, end_mask = var_50063_end_mask_0, x = v_221_cast_fp16)[name = tensor("op_50063_cast_fp16")]; + tensor var_50067_begin_0 = const()[name = tensor("op_50067_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_50067_end_0 = const()[name = tensor("op_50067_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_50067_end_mask_0 = const()[name = tensor("op_50067_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50067_cast_fp16 = slice_by_index(begin = var_50067_begin_0, end = var_50067_end_0, end_mask = var_50067_end_mask_0, x = v_221_cast_fp16)[name = tensor("op_50067_cast_fp16")]; + tensor var_50071_begin_0 = const()[name = tensor("op_50071_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_50071_end_0 = const()[name = tensor("op_50071_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_50071_end_mask_0 = const()[name = tensor("op_50071_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50071_cast_fp16 = slice_by_index(begin = var_50071_begin_0, end = var_50071_end_0, end_mask = var_50071_end_mask_0, x = v_221_cast_fp16)[name = tensor("op_50071_cast_fp16")]; + tensor var_50075_begin_0 = const()[name = tensor("op_50075_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_50075_end_0 = const()[name = tensor("op_50075_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_50075_end_mask_0 = const()[name = tensor("op_50075_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50075_cast_fp16 = slice_by_index(begin = var_50075_begin_0, end = var_50075_end_0, end_mask = var_50075_end_mask_0, x = v_221_cast_fp16)[name = tensor("op_50075_cast_fp16")]; + tensor var_50079_equation_0 = const()[name = tensor("op_50079_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50079_cast_fp16 = einsum(equation = var_50079_equation_0, values = (var_49921_cast_fp16, var_49838_cast_fp16))[name = tensor("op_50079_cast_fp16")]; + tensor var_50080_to_fp16 = const()[name = tensor("op_50080_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4241_cast_fp16 = mul(x = var_50079_cast_fp16, y = var_50080_to_fp16)[name = tensor("aw_4241_cast_fp16")]; + tensor var_50083_equation_0 = const()[name = tensor("op_50083_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50083_cast_fp16 = einsum(equation = var_50083_equation_0, values = (var_49925_cast_fp16, var_49842_cast_fp16))[name = tensor("op_50083_cast_fp16")]; + tensor var_50084_to_fp16 = const()[name = tensor("op_50084_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4243_cast_fp16 = mul(x = var_50083_cast_fp16, y = var_50084_to_fp16)[name = tensor("aw_4243_cast_fp16")]; + tensor var_50087_equation_0 = const()[name = tensor("op_50087_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50087_cast_fp16 = einsum(equation = var_50087_equation_0, values = (var_49929_cast_fp16, var_49846_cast_fp16))[name = tensor("op_50087_cast_fp16")]; + tensor var_50088_to_fp16 = const()[name = tensor("op_50088_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4245_cast_fp16 = mul(x = var_50087_cast_fp16, y = var_50088_to_fp16)[name = tensor("aw_4245_cast_fp16")]; + tensor var_50091_equation_0 = const()[name = tensor("op_50091_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50091_cast_fp16 = einsum(equation = var_50091_equation_0, values = (var_49933_cast_fp16, var_49850_cast_fp16))[name = tensor("op_50091_cast_fp16")]; + tensor var_50092_to_fp16 = const()[name = tensor("op_50092_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4247_cast_fp16 = mul(x = var_50091_cast_fp16, y = var_50092_to_fp16)[name = tensor("aw_4247_cast_fp16")]; + tensor var_50095_equation_0 = const()[name = tensor("op_50095_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50095_cast_fp16 = einsum(equation = var_50095_equation_0, values = (var_49937_cast_fp16, var_49854_cast_fp16))[name = tensor("op_50095_cast_fp16")]; + tensor var_50096_to_fp16 = const()[name = tensor("op_50096_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4249_cast_fp16 = mul(x = var_50095_cast_fp16, y = var_50096_to_fp16)[name = tensor("aw_4249_cast_fp16")]; + tensor var_50099_equation_0 = const()[name = tensor("op_50099_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50099_cast_fp16 = einsum(equation = var_50099_equation_0, values = (var_49941_cast_fp16, var_49858_cast_fp16))[name = tensor("op_50099_cast_fp16")]; + tensor var_50100_to_fp16 = const()[name = tensor("op_50100_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4251_cast_fp16 = mul(x = var_50099_cast_fp16, y = var_50100_to_fp16)[name = tensor("aw_4251_cast_fp16")]; + tensor var_50103_equation_0 = const()[name = tensor("op_50103_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50103_cast_fp16 = einsum(equation = var_50103_equation_0, values = (var_49945_cast_fp16, var_49862_cast_fp16))[name = tensor("op_50103_cast_fp16")]; + tensor var_50104_to_fp16 = const()[name = tensor("op_50104_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4253_cast_fp16 = mul(x = var_50103_cast_fp16, y = var_50104_to_fp16)[name = tensor("aw_4253_cast_fp16")]; + tensor var_50107_equation_0 = const()[name = tensor("op_50107_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50107_cast_fp16 = einsum(equation = var_50107_equation_0, values = (var_49949_cast_fp16, var_49866_cast_fp16))[name = tensor("op_50107_cast_fp16")]; + tensor var_50108_to_fp16 = const()[name = tensor("op_50108_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4255_cast_fp16 = mul(x = var_50107_cast_fp16, y = var_50108_to_fp16)[name = tensor("aw_4255_cast_fp16")]; + tensor var_50111_equation_0 = const()[name = tensor("op_50111_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50111_cast_fp16 = einsum(equation = var_50111_equation_0, values = (var_49953_cast_fp16, var_49870_cast_fp16))[name = tensor("op_50111_cast_fp16")]; + tensor var_50112_to_fp16 = const()[name = tensor("op_50112_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4257_cast_fp16 = mul(x = var_50111_cast_fp16, y = var_50112_to_fp16)[name = tensor("aw_4257_cast_fp16")]; + tensor var_50115_equation_0 = const()[name = tensor("op_50115_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50115_cast_fp16 = einsum(equation = var_50115_equation_0, values = (var_49957_cast_fp16, var_49874_cast_fp16))[name = tensor("op_50115_cast_fp16")]; + tensor var_50116_to_fp16 = const()[name = tensor("op_50116_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4259_cast_fp16 = mul(x = var_50115_cast_fp16, y = var_50116_to_fp16)[name = tensor("aw_4259_cast_fp16")]; + tensor var_50119_equation_0 = const()[name = tensor("op_50119_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50119_cast_fp16 = einsum(equation = var_50119_equation_0, values = (var_49961_cast_fp16, var_49878_cast_fp16))[name = tensor("op_50119_cast_fp16")]; + tensor var_50120_to_fp16 = const()[name = tensor("op_50120_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4261_cast_fp16 = mul(x = var_50119_cast_fp16, y = var_50120_to_fp16)[name = tensor("aw_4261_cast_fp16")]; + tensor var_50123_equation_0 = const()[name = tensor("op_50123_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50123_cast_fp16 = einsum(equation = var_50123_equation_0, values = (var_49965_cast_fp16, var_49882_cast_fp16))[name = tensor("op_50123_cast_fp16")]; + tensor var_50124_to_fp16 = const()[name = tensor("op_50124_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4263_cast_fp16 = mul(x = var_50123_cast_fp16, y = var_50124_to_fp16)[name = tensor("aw_4263_cast_fp16")]; + tensor var_50127_equation_0 = const()[name = tensor("op_50127_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50127_cast_fp16 = einsum(equation = var_50127_equation_0, values = (var_49969_cast_fp16, var_49886_cast_fp16))[name = tensor("op_50127_cast_fp16")]; + tensor var_50128_to_fp16 = const()[name = tensor("op_50128_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4265_cast_fp16 = mul(x = var_50127_cast_fp16, y = var_50128_to_fp16)[name = tensor("aw_4265_cast_fp16")]; + tensor var_50131_equation_0 = const()[name = tensor("op_50131_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50131_cast_fp16 = einsum(equation = var_50131_equation_0, values = (var_49973_cast_fp16, var_49890_cast_fp16))[name = tensor("op_50131_cast_fp16")]; + tensor var_50132_to_fp16 = const()[name = tensor("op_50132_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4267_cast_fp16 = mul(x = var_50131_cast_fp16, y = var_50132_to_fp16)[name = tensor("aw_4267_cast_fp16")]; + tensor var_50135_equation_0 = const()[name = tensor("op_50135_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50135_cast_fp16 = einsum(equation = var_50135_equation_0, values = (var_49977_cast_fp16, var_49894_cast_fp16))[name = tensor("op_50135_cast_fp16")]; + tensor var_50136_to_fp16 = const()[name = tensor("op_50136_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4269_cast_fp16 = mul(x = var_50135_cast_fp16, y = var_50136_to_fp16)[name = tensor("aw_4269_cast_fp16")]; + tensor var_50139_equation_0 = const()[name = tensor("op_50139_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50139_cast_fp16 = einsum(equation = var_50139_equation_0, values = (var_49981_cast_fp16, var_49898_cast_fp16))[name = tensor("op_50139_cast_fp16")]; + tensor var_50140_to_fp16 = const()[name = tensor("op_50140_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4271_cast_fp16 = mul(x = var_50139_cast_fp16, y = var_50140_to_fp16)[name = tensor("aw_4271_cast_fp16")]; + tensor var_50143_equation_0 = const()[name = tensor("op_50143_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50143_cast_fp16 = einsum(equation = var_50143_equation_0, values = (var_49985_cast_fp16, var_49902_cast_fp16))[name = tensor("op_50143_cast_fp16")]; + tensor var_50144_to_fp16 = const()[name = tensor("op_50144_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4273_cast_fp16 = mul(x = var_50143_cast_fp16, y = var_50144_to_fp16)[name = tensor("aw_4273_cast_fp16")]; + tensor var_50147_equation_0 = const()[name = tensor("op_50147_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50147_cast_fp16 = einsum(equation = var_50147_equation_0, values = (var_49989_cast_fp16, var_49906_cast_fp16))[name = tensor("op_50147_cast_fp16")]; + tensor var_50148_to_fp16 = const()[name = tensor("op_50148_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4275_cast_fp16 = mul(x = var_50147_cast_fp16, y = var_50148_to_fp16)[name = tensor("aw_4275_cast_fp16")]; + tensor var_50151_equation_0 = const()[name = tensor("op_50151_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50151_cast_fp16 = einsum(equation = var_50151_equation_0, values = (var_49993_cast_fp16, var_49910_cast_fp16))[name = tensor("op_50151_cast_fp16")]; + tensor var_50152_to_fp16 = const()[name = tensor("op_50152_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4277_cast_fp16 = mul(x = var_50151_cast_fp16, y = var_50152_to_fp16)[name = tensor("aw_4277_cast_fp16")]; + tensor var_50155_equation_0 = const()[name = tensor("op_50155_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50155_cast_fp16 = einsum(equation = var_50155_equation_0, values = (var_49997_cast_fp16, var_49914_cast_fp16))[name = tensor("op_50155_cast_fp16")]; + tensor var_50156_to_fp16 = const()[name = tensor("op_50156_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4279_cast_fp16 = mul(x = var_50155_cast_fp16, y = var_50156_to_fp16)[name = tensor("aw_4279_cast_fp16")]; + tensor var_50158_cast_fp16 = softmax(axis = var_30355, x = aw_4241_cast_fp16)[name = tensor("op_50158_cast_fp16")]; + tensor var_50159_cast_fp16 = softmax(axis = var_30355, x = aw_4243_cast_fp16)[name = tensor("op_50159_cast_fp16")]; + tensor var_50160_cast_fp16 = softmax(axis = var_30355, x = aw_4245_cast_fp16)[name = tensor("op_50160_cast_fp16")]; + tensor var_50161_cast_fp16 = softmax(axis = var_30355, x = aw_4247_cast_fp16)[name = tensor("op_50161_cast_fp16")]; + tensor var_50162_cast_fp16 = softmax(axis = var_30355, x = aw_4249_cast_fp16)[name = tensor("op_50162_cast_fp16")]; + tensor var_50163_cast_fp16 = softmax(axis = var_30355, x = aw_4251_cast_fp16)[name = tensor("op_50163_cast_fp16")]; + tensor var_50164_cast_fp16 = softmax(axis = var_30355, x = aw_4253_cast_fp16)[name = tensor("op_50164_cast_fp16")]; + tensor var_50165_cast_fp16 = softmax(axis = var_30355, x = aw_4255_cast_fp16)[name = tensor("op_50165_cast_fp16")]; + tensor var_50166_cast_fp16 = softmax(axis = var_30355, x = aw_4257_cast_fp16)[name = tensor("op_50166_cast_fp16")]; + tensor var_50167_cast_fp16 = softmax(axis = var_30355, x = aw_4259_cast_fp16)[name = tensor("op_50167_cast_fp16")]; + tensor var_50168_cast_fp16 = softmax(axis = var_30355, x = aw_4261_cast_fp16)[name = tensor("op_50168_cast_fp16")]; + tensor var_50169_cast_fp16 = softmax(axis = var_30355, x = aw_4263_cast_fp16)[name = tensor("op_50169_cast_fp16")]; + tensor var_50170_cast_fp16 = softmax(axis = var_30355, x = aw_4265_cast_fp16)[name = tensor("op_50170_cast_fp16")]; + tensor var_50171_cast_fp16 = softmax(axis = var_30355, x = aw_4267_cast_fp16)[name = tensor("op_50171_cast_fp16")]; + tensor var_50172_cast_fp16 = softmax(axis = var_30355, x = aw_4269_cast_fp16)[name = tensor("op_50172_cast_fp16")]; + tensor var_50173_cast_fp16 = softmax(axis = var_30355, x = aw_4271_cast_fp16)[name = tensor("op_50173_cast_fp16")]; + tensor var_50174_cast_fp16 = softmax(axis = var_30355, x = aw_4273_cast_fp16)[name = tensor("op_50174_cast_fp16")]; + tensor var_50175_cast_fp16 = softmax(axis = var_30355, x = aw_4275_cast_fp16)[name = tensor("op_50175_cast_fp16")]; + tensor var_50176_cast_fp16 = softmax(axis = var_30355, x = aw_4277_cast_fp16)[name = tensor("op_50176_cast_fp16")]; + tensor var_50177_cast_fp16 = softmax(axis = var_30355, x = aw_4279_cast_fp16)[name = tensor("op_50177_cast_fp16")]; + tensor var_50179_equation_0 = const()[name = tensor("op_50179_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50179_cast_fp16 = einsum(equation = var_50179_equation_0, values = (var_49999_cast_fp16, var_50158_cast_fp16))[name = tensor("op_50179_cast_fp16")]; + tensor var_50181_equation_0 = const()[name = tensor("op_50181_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50181_cast_fp16 = einsum(equation = var_50181_equation_0, values = (var_50003_cast_fp16, var_50159_cast_fp16))[name = tensor("op_50181_cast_fp16")]; + tensor var_50183_equation_0 = const()[name = tensor("op_50183_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50183_cast_fp16 = einsum(equation = var_50183_equation_0, values = (var_50007_cast_fp16, var_50160_cast_fp16))[name = tensor("op_50183_cast_fp16")]; + tensor var_50185_equation_0 = const()[name = tensor("op_50185_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50185_cast_fp16 = einsum(equation = var_50185_equation_0, values = (var_50011_cast_fp16, var_50161_cast_fp16))[name = tensor("op_50185_cast_fp16")]; + tensor var_50187_equation_0 = const()[name = tensor("op_50187_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50187_cast_fp16 = einsum(equation = var_50187_equation_0, values = (var_50015_cast_fp16, var_50162_cast_fp16))[name = tensor("op_50187_cast_fp16")]; + tensor var_50189_equation_0 = const()[name = tensor("op_50189_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50189_cast_fp16 = einsum(equation = var_50189_equation_0, values = (var_50019_cast_fp16, var_50163_cast_fp16))[name = tensor("op_50189_cast_fp16")]; + tensor var_50191_equation_0 = const()[name = tensor("op_50191_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50191_cast_fp16 = einsum(equation = var_50191_equation_0, values = (var_50023_cast_fp16, var_50164_cast_fp16))[name = tensor("op_50191_cast_fp16")]; + tensor var_50193_equation_0 = const()[name = tensor("op_50193_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50193_cast_fp16 = einsum(equation = var_50193_equation_0, values = (var_50027_cast_fp16, var_50165_cast_fp16))[name = tensor("op_50193_cast_fp16")]; + tensor var_50195_equation_0 = const()[name = tensor("op_50195_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50195_cast_fp16 = einsum(equation = var_50195_equation_0, values = (var_50031_cast_fp16, var_50166_cast_fp16))[name = tensor("op_50195_cast_fp16")]; + tensor var_50197_equation_0 = const()[name = tensor("op_50197_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50197_cast_fp16 = einsum(equation = var_50197_equation_0, values = (var_50035_cast_fp16, var_50167_cast_fp16))[name = tensor("op_50197_cast_fp16")]; + tensor var_50199_equation_0 = const()[name = tensor("op_50199_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50199_cast_fp16 = einsum(equation = var_50199_equation_0, values = (var_50039_cast_fp16, var_50168_cast_fp16))[name = tensor("op_50199_cast_fp16")]; + tensor var_50201_equation_0 = const()[name = tensor("op_50201_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50201_cast_fp16 = einsum(equation = var_50201_equation_0, values = (var_50043_cast_fp16, var_50169_cast_fp16))[name = tensor("op_50201_cast_fp16")]; + tensor var_50203_equation_0 = const()[name = tensor("op_50203_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50203_cast_fp16 = einsum(equation = var_50203_equation_0, values = (var_50047_cast_fp16, var_50170_cast_fp16))[name = tensor("op_50203_cast_fp16")]; + tensor var_50205_equation_0 = const()[name = tensor("op_50205_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50205_cast_fp16 = einsum(equation = var_50205_equation_0, values = (var_50051_cast_fp16, var_50171_cast_fp16))[name = tensor("op_50205_cast_fp16")]; + tensor var_50207_equation_0 = const()[name = tensor("op_50207_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50207_cast_fp16 = einsum(equation = var_50207_equation_0, values = (var_50055_cast_fp16, var_50172_cast_fp16))[name = tensor("op_50207_cast_fp16")]; + tensor var_50209_equation_0 = const()[name = tensor("op_50209_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50209_cast_fp16 = einsum(equation = var_50209_equation_0, values = (var_50059_cast_fp16, var_50173_cast_fp16))[name = tensor("op_50209_cast_fp16")]; + tensor var_50211_equation_0 = const()[name = tensor("op_50211_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50211_cast_fp16 = einsum(equation = var_50211_equation_0, values = (var_50063_cast_fp16, var_50174_cast_fp16))[name = tensor("op_50211_cast_fp16")]; + tensor var_50213_equation_0 = const()[name = tensor("op_50213_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50213_cast_fp16 = einsum(equation = var_50213_equation_0, values = (var_50067_cast_fp16, var_50175_cast_fp16))[name = tensor("op_50213_cast_fp16")]; + tensor var_50215_equation_0 = const()[name = tensor("op_50215_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50215_cast_fp16 = einsum(equation = var_50215_equation_0, values = (var_50071_cast_fp16, var_50176_cast_fp16))[name = tensor("op_50215_cast_fp16")]; + tensor var_50217_equation_0 = const()[name = tensor("op_50217_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50217_cast_fp16 = einsum(equation = var_50217_equation_0, values = (var_50075_cast_fp16, var_50177_cast_fp16))[name = tensor("op_50217_cast_fp16")]; + tensor input_643_interleave_0 = const()[name = tensor("input_643_interleave_0"), val = tensor(false)]; + tensor input_643_cast_fp16 = concat(axis = var_30355, interleave = input_643_interleave_0, values = (var_50179_cast_fp16, var_50181_cast_fp16, var_50183_cast_fp16, var_50185_cast_fp16, var_50187_cast_fp16, var_50189_cast_fp16, var_50191_cast_fp16, var_50193_cast_fp16, var_50195_cast_fp16, var_50197_cast_fp16, var_50199_cast_fp16, var_50201_cast_fp16, var_50203_cast_fp16, var_50205_cast_fp16, var_50207_cast_fp16, var_50209_cast_fp16, var_50211_cast_fp16, var_50213_cast_fp16, var_50215_cast_fp16, var_50217_cast_fp16))[name = tensor("input_643_cast_fp16")]; + tensor var_50223 = const()[name = tensor("op_50223"), val = tensor([1, 1])]; + tensor var_50225 = const()[name = tensor("op_50225"), val = tensor([1, 1])]; + tensor var_50227_pad_type_0 = const()[name = tensor("op_50227_pad_type_0"), val = tensor("custom")]; + tensor var_50227_pad_0 = const()[name = tensor("op_50227_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_1_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_1_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4251439808)))]; + tensor up_blocks_0_attentions_2_transformer_blocks_1_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_1_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4254716672)))]; + tensor var_50227_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_1_attn1_to_out_0_bias_to_fp16, dilations = var_50225, groups = var_30355, pad = var_50227_pad_0, pad_type = var_50227_pad_type_0, strides = var_50223, weight = up_blocks_0_attentions_2_transformer_blocks_1_attn1_to_out_0_weight_to_fp16, x = input_643_cast_fp16)[name = tensor("op_50227_cast_fp16")]; + tensor inputs_333_cast_fp16 = add(x = var_50227_cast_fp16, y = inputs_331_cast_fp16)[name = tensor("inputs_333_cast_fp16")]; + tensor hidden_states_443_axes_0 = const()[name = tensor("hidden_states_443_axes_0"), val = tensor([1])]; + tensor hidden_states_443_gamma_0_to_fp16 = const()[name = tensor("hidden_states_443_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4254719296)))]; + tensor hidden_states_443_beta_0_to_fp16 = const()[name = tensor("hidden_states_443_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4254721920)))]; + tensor var_50237_to_fp16 = const()[name = tensor("op_50237_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_443_cast_fp16 = layer_norm(axes = hidden_states_443_axes_0, beta = hidden_states_443_beta_0_to_fp16, epsilon = var_50237_to_fp16, gamma = hidden_states_443_gamma_0_to_fp16, x = inputs_333_cast_fp16)[name = tensor("hidden_states_443_cast_fp16")]; + tensor var_50252 = const()[name = tensor("op_50252"), val = tensor([1, 1])]; + tensor var_50254 = const()[name = tensor("op_50254"), val = tensor([1, 1])]; + tensor q_223_pad_type_0 = const()[name = tensor("q_223_pad_type_0"), val = tensor("custom")]; + tensor q_223_pad_0 = const()[name = tensor("q_223_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_1_attn2_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_1_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4254724544)))]; + tensor q_223_cast_fp16 = conv(dilations = var_50254, groups = var_30355, pad = q_223_pad_0, pad_type = q_223_pad_type_0, strides = var_50252, weight = up_blocks_0_attentions_2_transformer_blocks_1_attn2_to_q_weight_to_fp16, x = hidden_states_443_cast_fp16)[name = tensor("q_223_cast_fp16")]; + tensor var_50258 = const()[name = tensor("op_50258"), val = tensor([1, 1])]; + tensor var_50260 = const()[name = tensor("op_50260"), val = tensor([1, 1])]; + tensor k_445_pad_type_0 = const()[name = tensor("k_445_pad_type_0"), val = tensor("custom")]; + tensor k_445_pad_0 = const()[name = tensor("k_445_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_1_attn2_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_1_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4258001408)))]; + tensor k_445_cast_fp16 = conv(dilations = var_50260, groups = var_30355, pad = k_445_pad_0, pad_type = k_445_pad_type_0, strides = var_50258, weight = up_blocks_0_attentions_2_transformer_blocks_1_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_445_cast_fp16")]; + tensor var_50264 = const()[name = tensor("op_50264"), val = tensor([1, 1])]; + tensor var_50266 = const()[name = tensor("op_50266"), val = tensor([1, 1])]; + tensor v_223_pad_type_0 = const()[name = tensor("v_223_pad_type_0"), val = tensor("custom")]; + tensor v_223_pad_0 = const()[name = tensor("v_223_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_1_attn2_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_1_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4263244352)))]; + tensor v_223_cast_fp16 = conv(dilations = var_50266, groups = var_30355, pad = v_223_pad_0, pad_type = v_223_pad_type_0, strides = var_50264, weight = up_blocks_0_attentions_2_transformer_blocks_1_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_223_cast_fp16")]; + tensor var_50270_begin_0 = const()[name = tensor("op_50270_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_50270_end_0 = const()[name = tensor("op_50270_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_50270_end_mask_0 = const()[name = tensor("op_50270_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50270_cast_fp16 = slice_by_index(begin = var_50270_begin_0, end = var_50270_end_0, end_mask = var_50270_end_mask_0, x = q_223_cast_fp16)[name = tensor("op_50270_cast_fp16")]; + tensor var_50274_begin_0 = const()[name = tensor("op_50274_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_50274_end_0 = const()[name = tensor("op_50274_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_50274_end_mask_0 = const()[name = tensor("op_50274_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50274_cast_fp16 = slice_by_index(begin = var_50274_begin_0, end = var_50274_end_0, end_mask = var_50274_end_mask_0, x = q_223_cast_fp16)[name = tensor("op_50274_cast_fp16")]; + tensor var_50278_begin_0 = const()[name = tensor("op_50278_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_50278_end_0 = const()[name = tensor("op_50278_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_50278_end_mask_0 = const()[name = tensor("op_50278_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50278_cast_fp16 = slice_by_index(begin = var_50278_begin_0, end = var_50278_end_0, end_mask = var_50278_end_mask_0, x = q_223_cast_fp16)[name = tensor("op_50278_cast_fp16")]; + tensor var_50282_begin_0 = const()[name = tensor("op_50282_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_50282_end_0 = const()[name = tensor("op_50282_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_50282_end_mask_0 = const()[name = tensor("op_50282_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50282_cast_fp16 = slice_by_index(begin = var_50282_begin_0, end = var_50282_end_0, end_mask = var_50282_end_mask_0, x = q_223_cast_fp16)[name = tensor("op_50282_cast_fp16")]; + tensor var_50286_begin_0 = const()[name = tensor("op_50286_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_50286_end_0 = const()[name = tensor("op_50286_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_50286_end_mask_0 = const()[name = tensor("op_50286_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50286_cast_fp16 = slice_by_index(begin = var_50286_begin_0, end = var_50286_end_0, end_mask = var_50286_end_mask_0, x = q_223_cast_fp16)[name = tensor("op_50286_cast_fp16")]; + tensor var_50290_begin_0 = const()[name = tensor("op_50290_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_50290_end_0 = const()[name = tensor("op_50290_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_50290_end_mask_0 = const()[name = tensor("op_50290_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50290_cast_fp16 = slice_by_index(begin = var_50290_begin_0, end = var_50290_end_0, end_mask = var_50290_end_mask_0, x = q_223_cast_fp16)[name = tensor("op_50290_cast_fp16")]; + tensor var_50294_begin_0 = const()[name = tensor("op_50294_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_50294_end_0 = const()[name = tensor("op_50294_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_50294_end_mask_0 = const()[name = tensor("op_50294_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50294_cast_fp16 = slice_by_index(begin = var_50294_begin_0, end = var_50294_end_0, end_mask = var_50294_end_mask_0, x = q_223_cast_fp16)[name = tensor("op_50294_cast_fp16")]; + tensor var_50298_begin_0 = const()[name = tensor("op_50298_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_50298_end_0 = const()[name = tensor("op_50298_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_50298_end_mask_0 = const()[name = tensor("op_50298_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50298_cast_fp16 = slice_by_index(begin = var_50298_begin_0, end = var_50298_end_0, end_mask = var_50298_end_mask_0, x = q_223_cast_fp16)[name = tensor("op_50298_cast_fp16")]; + tensor var_50302_begin_0 = const()[name = tensor("op_50302_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_50302_end_0 = const()[name = tensor("op_50302_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_50302_end_mask_0 = const()[name = tensor("op_50302_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50302_cast_fp16 = slice_by_index(begin = var_50302_begin_0, end = var_50302_end_0, end_mask = var_50302_end_mask_0, x = q_223_cast_fp16)[name = tensor("op_50302_cast_fp16")]; + tensor var_50306_begin_0 = const()[name = tensor("op_50306_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_50306_end_0 = const()[name = tensor("op_50306_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_50306_end_mask_0 = const()[name = tensor("op_50306_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50306_cast_fp16 = slice_by_index(begin = var_50306_begin_0, end = var_50306_end_0, end_mask = var_50306_end_mask_0, x = q_223_cast_fp16)[name = tensor("op_50306_cast_fp16")]; + tensor var_50310_begin_0 = const()[name = tensor("op_50310_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_50310_end_0 = const()[name = tensor("op_50310_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_50310_end_mask_0 = const()[name = tensor("op_50310_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50310_cast_fp16 = slice_by_index(begin = var_50310_begin_0, end = var_50310_end_0, end_mask = var_50310_end_mask_0, x = q_223_cast_fp16)[name = tensor("op_50310_cast_fp16")]; + tensor var_50314_begin_0 = const()[name = tensor("op_50314_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_50314_end_0 = const()[name = tensor("op_50314_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_50314_end_mask_0 = const()[name = tensor("op_50314_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50314_cast_fp16 = slice_by_index(begin = var_50314_begin_0, end = var_50314_end_0, end_mask = var_50314_end_mask_0, x = q_223_cast_fp16)[name = tensor("op_50314_cast_fp16")]; + tensor var_50318_begin_0 = const()[name = tensor("op_50318_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_50318_end_0 = const()[name = tensor("op_50318_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_50318_end_mask_0 = const()[name = tensor("op_50318_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50318_cast_fp16 = slice_by_index(begin = var_50318_begin_0, end = var_50318_end_0, end_mask = var_50318_end_mask_0, x = q_223_cast_fp16)[name = tensor("op_50318_cast_fp16")]; + tensor var_50322_begin_0 = const()[name = tensor("op_50322_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_50322_end_0 = const()[name = tensor("op_50322_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_50322_end_mask_0 = const()[name = tensor("op_50322_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50322_cast_fp16 = slice_by_index(begin = var_50322_begin_0, end = var_50322_end_0, end_mask = var_50322_end_mask_0, x = q_223_cast_fp16)[name = tensor("op_50322_cast_fp16")]; + tensor var_50326_begin_0 = const()[name = tensor("op_50326_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_50326_end_0 = const()[name = tensor("op_50326_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_50326_end_mask_0 = const()[name = tensor("op_50326_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50326_cast_fp16 = slice_by_index(begin = var_50326_begin_0, end = var_50326_end_0, end_mask = var_50326_end_mask_0, x = q_223_cast_fp16)[name = tensor("op_50326_cast_fp16")]; + tensor var_50330_begin_0 = const()[name = tensor("op_50330_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_50330_end_0 = const()[name = tensor("op_50330_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_50330_end_mask_0 = const()[name = tensor("op_50330_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50330_cast_fp16 = slice_by_index(begin = var_50330_begin_0, end = var_50330_end_0, end_mask = var_50330_end_mask_0, x = q_223_cast_fp16)[name = tensor("op_50330_cast_fp16")]; + tensor var_50334_begin_0 = const()[name = tensor("op_50334_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_50334_end_0 = const()[name = tensor("op_50334_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_50334_end_mask_0 = const()[name = tensor("op_50334_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50334_cast_fp16 = slice_by_index(begin = var_50334_begin_0, end = var_50334_end_0, end_mask = var_50334_end_mask_0, x = q_223_cast_fp16)[name = tensor("op_50334_cast_fp16")]; + tensor var_50338_begin_0 = const()[name = tensor("op_50338_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_50338_end_0 = const()[name = tensor("op_50338_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_50338_end_mask_0 = const()[name = tensor("op_50338_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50338_cast_fp16 = slice_by_index(begin = var_50338_begin_0, end = var_50338_end_0, end_mask = var_50338_end_mask_0, x = q_223_cast_fp16)[name = tensor("op_50338_cast_fp16")]; + tensor var_50342_begin_0 = const()[name = tensor("op_50342_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_50342_end_0 = const()[name = tensor("op_50342_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_50342_end_mask_0 = const()[name = tensor("op_50342_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50342_cast_fp16 = slice_by_index(begin = var_50342_begin_0, end = var_50342_end_0, end_mask = var_50342_end_mask_0, x = q_223_cast_fp16)[name = tensor("op_50342_cast_fp16")]; + tensor var_50346_begin_0 = const()[name = tensor("op_50346_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_50346_end_0 = const()[name = tensor("op_50346_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_50346_end_mask_0 = const()[name = tensor("op_50346_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50346_cast_fp16 = slice_by_index(begin = var_50346_begin_0, end = var_50346_end_0, end_mask = var_50346_end_mask_0, x = q_223_cast_fp16)[name = tensor("op_50346_cast_fp16")]; + tensor k_447_perm_0 = const()[name = tensor("k_447_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_50353_begin_0 = const()[name = tensor("op_50353_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_50353_end_0 = const()[name = tensor("op_50353_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_50353_end_mask_0 = const()[name = tensor("op_50353_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_447_cast_fp16 = transpose(perm = k_447_perm_0, x = k_445_cast_fp16)[name = tensor("transpose_28")]; + tensor var_50353_cast_fp16 = slice_by_index(begin = var_50353_begin_0, end = var_50353_end_0, end_mask = var_50353_end_mask_0, x = k_447_cast_fp16)[name = tensor("op_50353_cast_fp16")]; + tensor var_50357_begin_0 = const()[name = tensor("op_50357_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_50357_end_0 = const()[name = tensor("op_50357_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_50357_end_mask_0 = const()[name = tensor("op_50357_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_50357_cast_fp16 = slice_by_index(begin = var_50357_begin_0, end = var_50357_end_0, end_mask = var_50357_end_mask_0, x = k_447_cast_fp16)[name = tensor("op_50357_cast_fp16")]; + tensor var_50361_begin_0 = const()[name = tensor("op_50361_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_50361_end_0 = const()[name = tensor("op_50361_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_50361_end_mask_0 = const()[name = tensor("op_50361_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_50361_cast_fp16 = slice_by_index(begin = var_50361_begin_0, end = var_50361_end_0, end_mask = var_50361_end_mask_0, x = k_447_cast_fp16)[name = tensor("op_50361_cast_fp16")]; + tensor var_50365_begin_0 = const()[name = tensor("op_50365_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_50365_end_0 = const()[name = tensor("op_50365_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_50365_end_mask_0 = const()[name = tensor("op_50365_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_50365_cast_fp16 = slice_by_index(begin = var_50365_begin_0, end = var_50365_end_0, end_mask = var_50365_end_mask_0, x = k_447_cast_fp16)[name = tensor("op_50365_cast_fp16")]; + tensor var_50369_begin_0 = const()[name = tensor("op_50369_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_50369_end_0 = const()[name = tensor("op_50369_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_50369_end_mask_0 = const()[name = tensor("op_50369_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_50369_cast_fp16 = slice_by_index(begin = var_50369_begin_0, end = var_50369_end_0, end_mask = var_50369_end_mask_0, x = k_447_cast_fp16)[name = tensor("op_50369_cast_fp16")]; + tensor var_50373_begin_0 = const()[name = tensor("op_50373_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_50373_end_0 = const()[name = tensor("op_50373_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_50373_end_mask_0 = const()[name = tensor("op_50373_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_50373_cast_fp16 = slice_by_index(begin = var_50373_begin_0, end = var_50373_end_0, end_mask = var_50373_end_mask_0, x = k_447_cast_fp16)[name = tensor("op_50373_cast_fp16")]; + tensor var_50377_begin_0 = const()[name = tensor("op_50377_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_50377_end_0 = const()[name = tensor("op_50377_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_50377_end_mask_0 = const()[name = tensor("op_50377_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_50377_cast_fp16 = slice_by_index(begin = var_50377_begin_0, end = var_50377_end_0, end_mask = var_50377_end_mask_0, x = k_447_cast_fp16)[name = tensor("op_50377_cast_fp16")]; + tensor var_50381_begin_0 = const()[name = tensor("op_50381_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_50381_end_0 = const()[name = tensor("op_50381_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_50381_end_mask_0 = const()[name = tensor("op_50381_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_50381_cast_fp16 = slice_by_index(begin = var_50381_begin_0, end = var_50381_end_0, end_mask = var_50381_end_mask_0, x = k_447_cast_fp16)[name = tensor("op_50381_cast_fp16")]; + tensor var_50385_begin_0 = const()[name = tensor("op_50385_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_50385_end_0 = const()[name = tensor("op_50385_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_50385_end_mask_0 = const()[name = tensor("op_50385_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_50385_cast_fp16 = slice_by_index(begin = var_50385_begin_0, end = var_50385_end_0, end_mask = var_50385_end_mask_0, x = k_447_cast_fp16)[name = tensor("op_50385_cast_fp16")]; + tensor var_50389_begin_0 = const()[name = tensor("op_50389_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_50389_end_0 = const()[name = tensor("op_50389_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_50389_end_mask_0 = const()[name = tensor("op_50389_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_50389_cast_fp16 = slice_by_index(begin = var_50389_begin_0, end = var_50389_end_0, end_mask = var_50389_end_mask_0, x = k_447_cast_fp16)[name = tensor("op_50389_cast_fp16")]; + tensor var_50393_begin_0 = const()[name = tensor("op_50393_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_50393_end_0 = const()[name = tensor("op_50393_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_50393_end_mask_0 = const()[name = tensor("op_50393_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_50393_cast_fp16 = slice_by_index(begin = var_50393_begin_0, end = var_50393_end_0, end_mask = var_50393_end_mask_0, x = k_447_cast_fp16)[name = tensor("op_50393_cast_fp16")]; + tensor var_50397_begin_0 = const()[name = tensor("op_50397_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_50397_end_0 = const()[name = tensor("op_50397_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_50397_end_mask_0 = const()[name = tensor("op_50397_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_50397_cast_fp16 = slice_by_index(begin = var_50397_begin_0, end = var_50397_end_0, end_mask = var_50397_end_mask_0, x = k_447_cast_fp16)[name = tensor("op_50397_cast_fp16")]; + tensor var_50401_begin_0 = const()[name = tensor("op_50401_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_50401_end_0 = const()[name = tensor("op_50401_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_50401_end_mask_0 = const()[name = tensor("op_50401_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_50401_cast_fp16 = slice_by_index(begin = var_50401_begin_0, end = var_50401_end_0, end_mask = var_50401_end_mask_0, x = k_447_cast_fp16)[name = tensor("op_50401_cast_fp16")]; + tensor var_50405_begin_0 = const()[name = tensor("op_50405_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_50405_end_0 = const()[name = tensor("op_50405_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_50405_end_mask_0 = const()[name = tensor("op_50405_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_50405_cast_fp16 = slice_by_index(begin = var_50405_begin_0, end = var_50405_end_0, end_mask = var_50405_end_mask_0, x = k_447_cast_fp16)[name = tensor("op_50405_cast_fp16")]; + tensor var_50409_begin_0 = const()[name = tensor("op_50409_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_50409_end_0 = const()[name = tensor("op_50409_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_50409_end_mask_0 = const()[name = tensor("op_50409_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_50409_cast_fp16 = slice_by_index(begin = var_50409_begin_0, end = var_50409_end_0, end_mask = var_50409_end_mask_0, x = k_447_cast_fp16)[name = tensor("op_50409_cast_fp16")]; + tensor var_50413_begin_0 = const()[name = tensor("op_50413_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_50413_end_0 = const()[name = tensor("op_50413_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_50413_end_mask_0 = const()[name = tensor("op_50413_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_50413_cast_fp16 = slice_by_index(begin = var_50413_begin_0, end = var_50413_end_0, end_mask = var_50413_end_mask_0, x = k_447_cast_fp16)[name = tensor("op_50413_cast_fp16")]; + tensor var_50417_begin_0 = const()[name = tensor("op_50417_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_50417_end_0 = const()[name = tensor("op_50417_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_50417_end_mask_0 = const()[name = tensor("op_50417_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_50417_cast_fp16 = slice_by_index(begin = var_50417_begin_0, end = var_50417_end_0, end_mask = var_50417_end_mask_0, x = k_447_cast_fp16)[name = tensor("op_50417_cast_fp16")]; + tensor var_50421_begin_0 = const()[name = tensor("op_50421_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_50421_end_0 = const()[name = tensor("op_50421_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_50421_end_mask_0 = const()[name = tensor("op_50421_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_50421_cast_fp16 = slice_by_index(begin = var_50421_begin_0, end = var_50421_end_0, end_mask = var_50421_end_mask_0, x = k_447_cast_fp16)[name = tensor("op_50421_cast_fp16")]; + tensor var_50425_begin_0 = const()[name = tensor("op_50425_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_50425_end_0 = const()[name = tensor("op_50425_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_50425_end_mask_0 = const()[name = tensor("op_50425_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_50425_cast_fp16 = slice_by_index(begin = var_50425_begin_0, end = var_50425_end_0, end_mask = var_50425_end_mask_0, x = k_447_cast_fp16)[name = tensor("op_50425_cast_fp16")]; + tensor var_50429_begin_0 = const()[name = tensor("op_50429_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_50429_end_0 = const()[name = tensor("op_50429_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_50429_end_mask_0 = const()[name = tensor("op_50429_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_50429_cast_fp16 = slice_by_index(begin = var_50429_begin_0, end = var_50429_end_0, end_mask = var_50429_end_mask_0, x = k_447_cast_fp16)[name = tensor("op_50429_cast_fp16")]; + tensor var_50431_begin_0 = const()[name = tensor("op_50431_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_50431_end_0 = const()[name = tensor("op_50431_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_50431_end_mask_0 = const()[name = tensor("op_50431_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50431_cast_fp16 = slice_by_index(begin = var_50431_begin_0, end = var_50431_end_0, end_mask = var_50431_end_mask_0, x = v_223_cast_fp16)[name = tensor("op_50431_cast_fp16")]; + tensor var_50435_begin_0 = const()[name = tensor("op_50435_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_50435_end_0 = const()[name = tensor("op_50435_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_50435_end_mask_0 = const()[name = tensor("op_50435_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50435_cast_fp16 = slice_by_index(begin = var_50435_begin_0, end = var_50435_end_0, end_mask = var_50435_end_mask_0, x = v_223_cast_fp16)[name = tensor("op_50435_cast_fp16")]; + tensor var_50439_begin_0 = const()[name = tensor("op_50439_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_50439_end_0 = const()[name = tensor("op_50439_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_50439_end_mask_0 = const()[name = tensor("op_50439_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50439_cast_fp16 = slice_by_index(begin = var_50439_begin_0, end = var_50439_end_0, end_mask = var_50439_end_mask_0, x = v_223_cast_fp16)[name = tensor("op_50439_cast_fp16")]; + tensor var_50443_begin_0 = const()[name = tensor("op_50443_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_50443_end_0 = const()[name = tensor("op_50443_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_50443_end_mask_0 = const()[name = tensor("op_50443_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50443_cast_fp16 = slice_by_index(begin = var_50443_begin_0, end = var_50443_end_0, end_mask = var_50443_end_mask_0, x = v_223_cast_fp16)[name = tensor("op_50443_cast_fp16")]; + tensor var_50447_begin_0 = const()[name = tensor("op_50447_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_50447_end_0 = const()[name = tensor("op_50447_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_50447_end_mask_0 = const()[name = tensor("op_50447_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50447_cast_fp16 = slice_by_index(begin = var_50447_begin_0, end = var_50447_end_0, end_mask = var_50447_end_mask_0, x = v_223_cast_fp16)[name = tensor("op_50447_cast_fp16")]; + tensor var_50451_begin_0 = const()[name = tensor("op_50451_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_50451_end_0 = const()[name = tensor("op_50451_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_50451_end_mask_0 = const()[name = tensor("op_50451_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50451_cast_fp16 = slice_by_index(begin = var_50451_begin_0, end = var_50451_end_0, end_mask = var_50451_end_mask_0, x = v_223_cast_fp16)[name = tensor("op_50451_cast_fp16")]; + tensor var_50455_begin_0 = const()[name = tensor("op_50455_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_50455_end_0 = const()[name = tensor("op_50455_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_50455_end_mask_0 = const()[name = tensor("op_50455_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50455_cast_fp16 = slice_by_index(begin = var_50455_begin_0, end = var_50455_end_0, end_mask = var_50455_end_mask_0, x = v_223_cast_fp16)[name = tensor("op_50455_cast_fp16")]; + tensor var_50459_begin_0 = const()[name = tensor("op_50459_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_50459_end_0 = const()[name = tensor("op_50459_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_50459_end_mask_0 = const()[name = tensor("op_50459_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50459_cast_fp16 = slice_by_index(begin = var_50459_begin_0, end = var_50459_end_0, end_mask = var_50459_end_mask_0, x = v_223_cast_fp16)[name = tensor("op_50459_cast_fp16")]; + tensor var_50463_begin_0 = const()[name = tensor("op_50463_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_50463_end_0 = const()[name = tensor("op_50463_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_50463_end_mask_0 = const()[name = tensor("op_50463_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50463_cast_fp16 = slice_by_index(begin = var_50463_begin_0, end = var_50463_end_0, end_mask = var_50463_end_mask_0, x = v_223_cast_fp16)[name = tensor("op_50463_cast_fp16")]; + tensor var_50467_begin_0 = const()[name = tensor("op_50467_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_50467_end_0 = const()[name = tensor("op_50467_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_50467_end_mask_0 = const()[name = tensor("op_50467_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50467_cast_fp16 = slice_by_index(begin = var_50467_begin_0, end = var_50467_end_0, end_mask = var_50467_end_mask_0, x = v_223_cast_fp16)[name = tensor("op_50467_cast_fp16")]; + tensor var_50471_begin_0 = const()[name = tensor("op_50471_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_50471_end_0 = const()[name = tensor("op_50471_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_50471_end_mask_0 = const()[name = tensor("op_50471_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50471_cast_fp16 = slice_by_index(begin = var_50471_begin_0, end = var_50471_end_0, end_mask = var_50471_end_mask_0, x = v_223_cast_fp16)[name = tensor("op_50471_cast_fp16")]; + tensor var_50475_begin_0 = const()[name = tensor("op_50475_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_50475_end_0 = const()[name = tensor("op_50475_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_50475_end_mask_0 = const()[name = tensor("op_50475_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50475_cast_fp16 = slice_by_index(begin = var_50475_begin_0, end = var_50475_end_0, end_mask = var_50475_end_mask_0, x = v_223_cast_fp16)[name = tensor("op_50475_cast_fp16")]; + tensor var_50479_begin_0 = const()[name = tensor("op_50479_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_50479_end_0 = const()[name = tensor("op_50479_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_50479_end_mask_0 = const()[name = tensor("op_50479_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50479_cast_fp16 = slice_by_index(begin = var_50479_begin_0, end = var_50479_end_0, end_mask = var_50479_end_mask_0, x = v_223_cast_fp16)[name = tensor("op_50479_cast_fp16")]; + tensor var_50483_begin_0 = const()[name = tensor("op_50483_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_50483_end_0 = const()[name = tensor("op_50483_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_50483_end_mask_0 = const()[name = tensor("op_50483_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50483_cast_fp16 = slice_by_index(begin = var_50483_begin_0, end = var_50483_end_0, end_mask = var_50483_end_mask_0, x = v_223_cast_fp16)[name = tensor("op_50483_cast_fp16")]; + tensor var_50487_begin_0 = const()[name = tensor("op_50487_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_50487_end_0 = const()[name = tensor("op_50487_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_50487_end_mask_0 = const()[name = tensor("op_50487_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50487_cast_fp16 = slice_by_index(begin = var_50487_begin_0, end = var_50487_end_0, end_mask = var_50487_end_mask_0, x = v_223_cast_fp16)[name = tensor("op_50487_cast_fp16")]; + tensor var_50491_begin_0 = const()[name = tensor("op_50491_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_50491_end_0 = const()[name = tensor("op_50491_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_50491_end_mask_0 = const()[name = tensor("op_50491_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50491_cast_fp16 = slice_by_index(begin = var_50491_begin_0, end = var_50491_end_0, end_mask = var_50491_end_mask_0, x = v_223_cast_fp16)[name = tensor("op_50491_cast_fp16")]; + tensor var_50495_begin_0 = const()[name = tensor("op_50495_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_50495_end_0 = const()[name = tensor("op_50495_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_50495_end_mask_0 = const()[name = tensor("op_50495_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50495_cast_fp16 = slice_by_index(begin = var_50495_begin_0, end = var_50495_end_0, end_mask = var_50495_end_mask_0, x = v_223_cast_fp16)[name = tensor("op_50495_cast_fp16")]; + tensor var_50499_begin_0 = const()[name = tensor("op_50499_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_50499_end_0 = const()[name = tensor("op_50499_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_50499_end_mask_0 = const()[name = tensor("op_50499_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50499_cast_fp16 = slice_by_index(begin = var_50499_begin_0, end = var_50499_end_0, end_mask = var_50499_end_mask_0, x = v_223_cast_fp16)[name = tensor("op_50499_cast_fp16")]; + tensor var_50503_begin_0 = const()[name = tensor("op_50503_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_50503_end_0 = const()[name = tensor("op_50503_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_50503_end_mask_0 = const()[name = tensor("op_50503_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50503_cast_fp16 = slice_by_index(begin = var_50503_begin_0, end = var_50503_end_0, end_mask = var_50503_end_mask_0, x = v_223_cast_fp16)[name = tensor("op_50503_cast_fp16")]; + tensor var_50507_begin_0 = const()[name = tensor("op_50507_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_50507_end_0 = const()[name = tensor("op_50507_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_50507_end_mask_0 = const()[name = tensor("op_50507_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50507_cast_fp16 = slice_by_index(begin = var_50507_begin_0, end = var_50507_end_0, end_mask = var_50507_end_mask_0, x = v_223_cast_fp16)[name = tensor("op_50507_cast_fp16")]; + tensor var_50511_equation_0 = const()[name = tensor("op_50511_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50511_cast_fp16 = einsum(equation = var_50511_equation_0, values = (var_50353_cast_fp16, var_50270_cast_fp16))[name = tensor("op_50511_cast_fp16")]; + tensor var_50512_to_fp16 = const()[name = tensor("op_50512_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4281_cast_fp16 = mul(x = var_50511_cast_fp16, y = var_50512_to_fp16)[name = tensor("aw_4281_cast_fp16")]; + tensor var_50515_equation_0 = const()[name = tensor("op_50515_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50515_cast_fp16 = einsum(equation = var_50515_equation_0, values = (var_50357_cast_fp16, var_50274_cast_fp16))[name = tensor("op_50515_cast_fp16")]; + tensor var_50516_to_fp16 = const()[name = tensor("op_50516_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4283_cast_fp16 = mul(x = var_50515_cast_fp16, y = var_50516_to_fp16)[name = tensor("aw_4283_cast_fp16")]; + tensor var_50519_equation_0 = const()[name = tensor("op_50519_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50519_cast_fp16 = einsum(equation = var_50519_equation_0, values = (var_50361_cast_fp16, var_50278_cast_fp16))[name = tensor("op_50519_cast_fp16")]; + tensor var_50520_to_fp16 = const()[name = tensor("op_50520_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4285_cast_fp16 = mul(x = var_50519_cast_fp16, y = var_50520_to_fp16)[name = tensor("aw_4285_cast_fp16")]; + tensor var_50523_equation_0 = const()[name = tensor("op_50523_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50523_cast_fp16 = einsum(equation = var_50523_equation_0, values = (var_50365_cast_fp16, var_50282_cast_fp16))[name = tensor("op_50523_cast_fp16")]; + tensor var_50524_to_fp16 = const()[name = tensor("op_50524_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4287_cast_fp16 = mul(x = var_50523_cast_fp16, y = var_50524_to_fp16)[name = tensor("aw_4287_cast_fp16")]; + tensor var_50527_equation_0 = const()[name = tensor("op_50527_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50527_cast_fp16 = einsum(equation = var_50527_equation_0, values = (var_50369_cast_fp16, var_50286_cast_fp16))[name = tensor("op_50527_cast_fp16")]; + tensor var_50528_to_fp16 = const()[name = tensor("op_50528_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4289_cast_fp16 = mul(x = var_50527_cast_fp16, y = var_50528_to_fp16)[name = tensor("aw_4289_cast_fp16")]; + tensor var_50531_equation_0 = const()[name = tensor("op_50531_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50531_cast_fp16 = einsum(equation = var_50531_equation_0, values = (var_50373_cast_fp16, var_50290_cast_fp16))[name = tensor("op_50531_cast_fp16")]; + tensor var_50532_to_fp16 = const()[name = tensor("op_50532_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4291_cast_fp16 = mul(x = var_50531_cast_fp16, y = var_50532_to_fp16)[name = tensor("aw_4291_cast_fp16")]; + tensor var_50535_equation_0 = const()[name = tensor("op_50535_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50535_cast_fp16 = einsum(equation = var_50535_equation_0, values = (var_50377_cast_fp16, var_50294_cast_fp16))[name = tensor("op_50535_cast_fp16")]; + tensor var_50536_to_fp16 = const()[name = tensor("op_50536_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4293_cast_fp16 = mul(x = var_50535_cast_fp16, y = var_50536_to_fp16)[name = tensor("aw_4293_cast_fp16")]; + tensor var_50539_equation_0 = const()[name = tensor("op_50539_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50539_cast_fp16 = einsum(equation = var_50539_equation_0, values = (var_50381_cast_fp16, var_50298_cast_fp16))[name = tensor("op_50539_cast_fp16")]; + tensor var_50540_to_fp16 = const()[name = tensor("op_50540_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4295_cast_fp16 = mul(x = var_50539_cast_fp16, y = var_50540_to_fp16)[name = tensor("aw_4295_cast_fp16")]; + tensor var_50543_equation_0 = const()[name = tensor("op_50543_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50543_cast_fp16 = einsum(equation = var_50543_equation_0, values = (var_50385_cast_fp16, var_50302_cast_fp16))[name = tensor("op_50543_cast_fp16")]; + tensor var_50544_to_fp16 = const()[name = tensor("op_50544_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4297_cast_fp16 = mul(x = var_50543_cast_fp16, y = var_50544_to_fp16)[name = tensor("aw_4297_cast_fp16")]; + tensor var_50547_equation_0 = const()[name = tensor("op_50547_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50547_cast_fp16 = einsum(equation = var_50547_equation_0, values = (var_50389_cast_fp16, var_50306_cast_fp16))[name = tensor("op_50547_cast_fp16")]; + tensor var_50548_to_fp16 = const()[name = tensor("op_50548_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4299_cast_fp16 = mul(x = var_50547_cast_fp16, y = var_50548_to_fp16)[name = tensor("aw_4299_cast_fp16")]; + tensor var_50551_equation_0 = const()[name = tensor("op_50551_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50551_cast_fp16 = einsum(equation = var_50551_equation_0, values = (var_50393_cast_fp16, var_50310_cast_fp16))[name = tensor("op_50551_cast_fp16")]; + tensor var_50552_to_fp16 = const()[name = tensor("op_50552_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4301_cast_fp16 = mul(x = var_50551_cast_fp16, y = var_50552_to_fp16)[name = tensor("aw_4301_cast_fp16")]; + tensor var_50555_equation_0 = const()[name = tensor("op_50555_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50555_cast_fp16 = einsum(equation = var_50555_equation_0, values = (var_50397_cast_fp16, var_50314_cast_fp16))[name = tensor("op_50555_cast_fp16")]; + tensor var_50556_to_fp16 = const()[name = tensor("op_50556_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4303_cast_fp16 = mul(x = var_50555_cast_fp16, y = var_50556_to_fp16)[name = tensor("aw_4303_cast_fp16")]; + tensor var_50559_equation_0 = const()[name = tensor("op_50559_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50559_cast_fp16 = einsum(equation = var_50559_equation_0, values = (var_50401_cast_fp16, var_50318_cast_fp16))[name = tensor("op_50559_cast_fp16")]; + tensor var_50560_to_fp16 = const()[name = tensor("op_50560_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4305_cast_fp16 = mul(x = var_50559_cast_fp16, y = var_50560_to_fp16)[name = tensor("aw_4305_cast_fp16")]; + tensor var_50563_equation_0 = const()[name = tensor("op_50563_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50563_cast_fp16 = einsum(equation = var_50563_equation_0, values = (var_50405_cast_fp16, var_50322_cast_fp16))[name = tensor("op_50563_cast_fp16")]; + tensor var_50564_to_fp16 = const()[name = tensor("op_50564_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4307_cast_fp16 = mul(x = var_50563_cast_fp16, y = var_50564_to_fp16)[name = tensor("aw_4307_cast_fp16")]; + tensor var_50567_equation_0 = const()[name = tensor("op_50567_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50567_cast_fp16 = einsum(equation = var_50567_equation_0, values = (var_50409_cast_fp16, var_50326_cast_fp16))[name = tensor("op_50567_cast_fp16")]; + tensor var_50568_to_fp16 = const()[name = tensor("op_50568_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4309_cast_fp16 = mul(x = var_50567_cast_fp16, y = var_50568_to_fp16)[name = tensor("aw_4309_cast_fp16")]; + tensor var_50571_equation_0 = const()[name = tensor("op_50571_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50571_cast_fp16 = einsum(equation = var_50571_equation_0, values = (var_50413_cast_fp16, var_50330_cast_fp16))[name = tensor("op_50571_cast_fp16")]; + tensor var_50572_to_fp16 = const()[name = tensor("op_50572_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4311_cast_fp16 = mul(x = var_50571_cast_fp16, y = var_50572_to_fp16)[name = tensor("aw_4311_cast_fp16")]; + tensor var_50575_equation_0 = const()[name = tensor("op_50575_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50575_cast_fp16 = einsum(equation = var_50575_equation_0, values = (var_50417_cast_fp16, var_50334_cast_fp16))[name = tensor("op_50575_cast_fp16")]; + tensor var_50576_to_fp16 = const()[name = tensor("op_50576_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4313_cast_fp16 = mul(x = var_50575_cast_fp16, y = var_50576_to_fp16)[name = tensor("aw_4313_cast_fp16")]; + tensor var_50579_equation_0 = const()[name = tensor("op_50579_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50579_cast_fp16 = einsum(equation = var_50579_equation_0, values = (var_50421_cast_fp16, var_50338_cast_fp16))[name = tensor("op_50579_cast_fp16")]; + tensor var_50580_to_fp16 = const()[name = tensor("op_50580_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4315_cast_fp16 = mul(x = var_50579_cast_fp16, y = var_50580_to_fp16)[name = tensor("aw_4315_cast_fp16")]; + tensor var_50583_equation_0 = const()[name = tensor("op_50583_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50583_cast_fp16 = einsum(equation = var_50583_equation_0, values = (var_50425_cast_fp16, var_50342_cast_fp16))[name = tensor("op_50583_cast_fp16")]; + tensor var_50584_to_fp16 = const()[name = tensor("op_50584_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4317_cast_fp16 = mul(x = var_50583_cast_fp16, y = var_50584_to_fp16)[name = tensor("aw_4317_cast_fp16")]; + tensor var_50587_equation_0 = const()[name = tensor("op_50587_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50587_cast_fp16 = einsum(equation = var_50587_equation_0, values = (var_50429_cast_fp16, var_50346_cast_fp16))[name = tensor("op_50587_cast_fp16")]; + tensor var_50588_to_fp16 = const()[name = tensor("op_50588_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4319_cast_fp16 = mul(x = var_50587_cast_fp16, y = var_50588_to_fp16)[name = tensor("aw_4319_cast_fp16")]; + tensor var_50590_cast_fp16 = softmax(axis = var_30355, x = aw_4281_cast_fp16)[name = tensor("op_50590_cast_fp16")]; + tensor var_50591_cast_fp16 = softmax(axis = var_30355, x = aw_4283_cast_fp16)[name = tensor("op_50591_cast_fp16")]; + tensor var_50592_cast_fp16 = softmax(axis = var_30355, x = aw_4285_cast_fp16)[name = tensor("op_50592_cast_fp16")]; + tensor var_50593_cast_fp16 = softmax(axis = var_30355, x = aw_4287_cast_fp16)[name = tensor("op_50593_cast_fp16")]; + tensor var_50594_cast_fp16 = softmax(axis = var_30355, x = aw_4289_cast_fp16)[name = tensor("op_50594_cast_fp16")]; + tensor var_50595_cast_fp16 = softmax(axis = var_30355, x = aw_4291_cast_fp16)[name = tensor("op_50595_cast_fp16")]; + tensor var_50596_cast_fp16 = softmax(axis = var_30355, x = aw_4293_cast_fp16)[name = tensor("op_50596_cast_fp16")]; + tensor var_50597_cast_fp16 = softmax(axis = var_30355, x = aw_4295_cast_fp16)[name = tensor("op_50597_cast_fp16")]; + tensor var_50598_cast_fp16 = softmax(axis = var_30355, x = aw_4297_cast_fp16)[name = tensor("op_50598_cast_fp16")]; + tensor var_50599_cast_fp16 = softmax(axis = var_30355, x = aw_4299_cast_fp16)[name = tensor("op_50599_cast_fp16")]; + tensor var_50600_cast_fp16 = softmax(axis = var_30355, x = aw_4301_cast_fp16)[name = tensor("op_50600_cast_fp16")]; + tensor var_50601_cast_fp16 = softmax(axis = var_30355, x = aw_4303_cast_fp16)[name = tensor("op_50601_cast_fp16")]; + tensor var_50602_cast_fp16 = softmax(axis = var_30355, x = aw_4305_cast_fp16)[name = tensor("op_50602_cast_fp16")]; + tensor var_50603_cast_fp16 = softmax(axis = var_30355, x = aw_4307_cast_fp16)[name = tensor("op_50603_cast_fp16")]; + tensor var_50604_cast_fp16 = softmax(axis = var_30355, x = aw_4309_cast_fp16)[name = tensor("op_50604_cast_fp16")]; + tensor var_50605_cast_fp16 = softmax(axis = var_30355, x = aw_4311_cast_fp16)[name = tensor("op_50605_cast_fp16")]; + tensor var_50606_cast_fp16 = softmax(axis = var_30355, x = aw_4313_cast_fp16)[name = tensor("op_50606_cast_fp16")]; + tensor var_50607_cast_fp16 = softmax(axis = var_30355, x = aw_4315_cast_fp16)[name = tensor("op_50607_cast_fp16")]; + tensor var_50608_cast_fp16 = softmax(axis = var_30355, x = aw_4317_cast_fp16)[name = tensor("op_50608_cast_fp16")]; + tensor var_50609_cast_fp16 = softmax(axis = var_30355, x = aw_4319_cast_fp16)[name = tensor("op_50609_cast_fp16")]; + tensor var_50611_equation_0 = const()[name = tensor("op_50611_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50611_cast_fp16 = einsum(equation = var_50611_equation_0, values = (var_50431_cast_fp16, var_50590_cast_fp16))[name = tensor("op_50611_cast_fp16")]; + tensor var_50613_equation_0 = const()[name = tensor("op_50613_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50613_cast_fp16 = einsum(equation = var_50613_equation_0, values = (var_50435_cast_fp16, var_50591_cast_fp16))[name = tensor("op_50613_cast_fp16")]; + tensor var_50615_equation_0 = const()[name = tensor("op_50615_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50615_cast_fp16 = einsum(equation = var_50615_equation_0, values = (var_50439_cast_fp16, var_50592_cast_fp16))[name = tensor("op_50615_cast_fp16")]; + tensor var_50617_equation_0 = const()[name = tensor("op_50617_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50617_cast_fp16 = einsum(equation = var_50617_equation_0, values = (var_50443_cast_fp16, var_50593_cast_fp16))[name = tensor("op_50617_cast_fp16")]; + tensor var_50619_equation_0 = const()[name = tensor("op_50619_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50619_cast_fp16 = einsum(equation = var_50619_equation_0, values = (var_50447_cast_fp16, var_50594_cast_fp16))[name = tensor("op_50619_cast_fp16")]; + tensor var_50621_equation_0 = const()[name = tensor("op_50621_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50621_cast_fp16 = einsum(equation = var_50621_equation_0, values = (var_50451_cast_fp16, var_50595_cast_fp16))[name = tensor("op_50621_cast_fp16")]; + tensor var_50623_equation_0 = const()[name = tensor("op_50623_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50623_cast_fp16 = einsum(equation = var_50623_equation_0, values = (var_50455_cast_fp16, var_50596_cast_fp16))[name = tensor("op_50623_cast_fp16")]; + tensor var_50625_equation_0 = const()[name = tensor("op_50625_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50625_cast_fp16 = einsum(equation = var_50625_equation_0, values = (var_50459_cast_fp16, var_50597_cast_fp16))[name = tensor("op_50625_cast_fp16")]; + tensor var_50627_equation_0 = const()[name = tensor("op_50627_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50627_cast_fp16 = einsum(equation = var_50627_equation_0, values = (var_50463_cast_fp16, var_50598_cast_fp16))[name = tensor("op_50627_cast_fp16")]; + tensor var_50629_equation_0 = const()[name = tensor("op_50629_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50629_cast_fp16 = einsum(equation = var_50629_equation_0, values = (var_50467_cast_fp16, var_50599_cast_fp16))[name = tensor("op_50629_cast_fp16")]; + tensor var_50631_equation_0 = const()[name = tensor("op_50631_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50631_cast_fp16 = einsum(equation = var_50631_equation_0, values = (var_50471_cast_fp16, var_50600_cast_fp16))[name = tensor("op_50631_cast_fp16")]; + tensor var_50633_equation_0 = const()[name = tensor("op_50633_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50633_cast_fp16 = einsum(equation = var_50633_equation_0, values = (var_50475_cast_fp16, var_50601_cast_fp16))[name = tensor("op_50633_cast_fp16")]; + tensor var_50635_equation_0 = const()[name = tensor("op_50635_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50635_cast_fp16 = einsum(equation = var_50635_equation_0, values = (var_50479_cast_fp16, var_50602_cast_fp16))[name = tensor("op_50635_cast_fp16")]; + tensor var_50637_equation_0 = const()[name = tensor("op_50637_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50637_cast_fp16 = einsum(equation = var_50637_equation_0, values = (var_50483_cast_fp16, var_50603_cast_fp16))[name = tensor("op_50637_cast_fp16")]; + tensor var_50639_equation_0 = const()[name = tensor("op_50639_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50639_cast_fp16 = einsum(equation = var_50639_equation_0, values = (var_50487_cast_fp16, var_50604_cast_fp16))[name = tensor("op_50639_cast_fp16")]; + tensor var_50641_equation_0 = const()[name = tensor("op_50641_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50641_cast_fp16 = einsum(equation = var_50641_equation_0, values = (var_50491_cast_fp16, var_50605_cast_fp16))[name = tensor("op_50641_cast_fp16")]; + tensor var_50643_equation_0 = const()[name = tensor("op_50643_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50643_cast_fp16 = einsum(equation = var_50643_equation_0, values = (var_50495_cast_fp16, var_50606_cast_fp16))[name = tensor("op_50643_cast_fp16")]; + tensor var_50645_equation_0 = const()[name = tensor("op_50645_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50645_cast_fp16 = einsum(equation = var_50645_equation_0, values = (var_50499_cast_fp16, var_50607_cast_fp16))[name = tensor("op_50645_cast_fp16")]; + tensor var_50647_equation_0 = const()[name = tensor("op_50647_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50647_cast_fp16 = einsum(equation = var_50647_equation_0, values = (var_50503_cast_fp16, var_50608_cast_fp16))[name = tensor("op_50647_cast_fp16")]; + tensor var_50649_equation_0 = const()[name = tensor("op_50649_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50649_cast_fp16 = einsum(equation = var_50649_equation_0, values = (var_50507_cast_fp16, var_50609_cast_fp16))[name = tensor("op_50649_cast_fp16")]; + tensor input_645_interleave_0 = const()[name = tensor("input_645_interleave_0"), val = tensor(false)]; + tensor input_645_cast_fp16 = concat(axis = var_30355, interleave = input_645_interleave_0, values = (var_50611_cast_fp16, var_50613_cast_fp16, var_50615_cast_fp16, var_50617_cast_fp16, var_50619_cast_fp16, var_50621_cast_fp16, var_50623_cast_fp16, var_50625_cast_fp16, var_50627_cast_fp16, var_50629_cast_fp16, var_50631_cast_fp16, var_50633_cast_fp16, var_50635_cast_fp16, var_50637_cast_fp16, var_50639_cast_fp16, var_50641_cast_fp16, var_50643_cast_fp16, var_50645_cast_fp16, var_50647_cast_fp16, var_50649_cast_fp16))[name = tensor("input_645_cast_fp16")]; + tensor var_50655 = const()[name = tensor("op_50655"), val = tensor([1, 1])]; + tensor var_50657 = const()[name = tensor("op_50657"), val = tensor([1, 1])]; + tensor var_50659_pad_type_0 = const()[name = tensor("op_50659_pad_type_0"), val = tensor("custom")]; + tensor var_50659_pad_0 = const()[name = tensor("op_50659_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_1_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_1_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4268487296)))]; + tensor up_blocks_0_attentions_2_transformer_blocks_1_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_1_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4271764160)))]; + tensor var_50659_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_1_attn2_to_out_0_bias_to_fp16, dilations = var_50657, groups = var_30355, pad = var_50659_pad_0, pad_type = var_50659_pad_type_0, strides = var_50655, weight = up_blocks_0_attentions_2_transformer_blocks_1_attn2_to_out_0_weight_to_fp16, x = input_645_cast_fp16)[name = tensor("op_50659_cast_fp16")]; + tensor inputs_335_cast_fp16 = add(x = var_50659_cast_fp16, y = inputs_333_cast_fp16)[name = tensor("inputs_335_cast_fp16")]; + tensor input_647_axes_0 = const()[name = tensor("input_647_axes_0"), val = tensor([1])]; + tensor input_647_gamma_0_to_fp16 = const()[name = tensor("input_647_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4271766784)))]; + tensor input_647_beta_0_to_fp16 = const()[name = tensor("input_647_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4271769408)))]; + tensor var_50669_to_fp16 = const()[name = tensor("op_50669_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_647_cast_fp16 = layer_norm(axes = input_647_axes_0, beta = input_647_beta_0_to_fp16, epsilon = var_50669_to_fp16, gamma = input_647_gamma_0_to_fp16, x = inputs_335_cast_fp16)[name = tensor("input_647_cast_fp16")]; + tensor var_50685 = const()[name = tensor("op_50685"), val = tensor([1, 1])]; + tensor var_50687 = const()[name = tensor("op_50687"), val = tensor([1, 1])]; + tensor var_50689_pad_type_0 = const()[name = tensor("op_50689_pad_type_0"), val = tensor("custom")]; + tensor var_50689_pad_0 = const()[name = tensor("op_50689_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_1_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_1_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4271772032)))]; + tensor up_blocks_0_attentions_2_transformer_blocks_1_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_1_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4297986496)))]; + tensor var_50689_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_1_ff_net_0_proj_bias_to_fp16, dilations = var_50687, groups = var_30355, pad = var_50689_pad_0, pad_type = var_50689_pad_type_0, strides = var_50685, weight = up_blocks_0_attentions_2_transformer_blocks_1_ff_net_0_proj_weight_to_fp16, x = input_647_cast_fp16)[name = tensor("op_50689_cast_fp16")]; + tensor var_50690_split_sizes_0 = const()[name = tensor("op_50690_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_50690_axis_0 = const()[name = tensor("op_50690_axis_0"), val = tensor(1)]; + tensor var_50690_cast_fp16_0, tensor var_50690_cast_fp16_1 = split(axis = var_50690_axis_0, split_sizes = var_50690_split_sizes_0, x = var_50689_cast_fp16)[name = tensor("op_50690_cast_fp16")]; + tensor var_50692_mode_0 = const()[name = tensor("op_50692_mode_0"), val = tensor("EXACT")]; + tensor var_50692_cast_fp16 = gelu(mode = var_50692_mode_0, x = var_50690_cast_fp16_1)[name = tensor("op_50692_cast_fp16")]; + tensor input_649_cast_fp16 = mul(x = var_50690_cast_fp16_0, y = var_50692_cast_fp16)[name = tensor("input_649_cast_fp16")]; + tensor var_50696 = const()[name = tensor("op_50696"), val = tensor([1, 1])]; + tensor var_50698 = const()[name = tensor("op_50698"), val = tensor([1, 1])]; + tensor var_50700_pad_type_0 = const()[name = tensor("op_50700_pad_type_0"), val = tensor("custom")]; + tensor var_50700_pad_0 = const()[name = tensor("op_50700_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_1_ff_net_2_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_1_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4298007040)))]; + tensor up_blocks_0_attentions_2_transformer_blocks_1_ff_net_2_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_1_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4311114304)))]; + tensor var_50700_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_1_ff_net_2_bias_to_fp16, dilations = var_50698, groups = var_30355, pad = var_50700_pad_0, pad_type = var_50700_pad_type_0, strides = var_50696, weight = up_blocks_0_attentions_2_transformer_blocks_1_ff_net_2_weight_to_fp16, x = input_649_cast_fp16)[name = tensor("op_50700_cast_fp16")]; + tensor inputs_337_cast_fp16 = add(x = var_50700_cast_fp16, y = inputs_335_cast_fp16)[name = tensor("inputs_337_cast_fp16")]; + tensor hidden_states_447_axes_0 = const()[name = tensor("hidden_states_447_axes_0"), val = tensor([1])]; + tensor hidden_states_447_gamma_0_to_fp16 = const()[name = tensor("hidden_states_447_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4311116928)))]; + tensor hidden_states_447_beta_0_to_fp16 = const()[name = tensor("hidden_states_447_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4311119552)))]; + tensor var_50716_to_fp16 = const()[name = tensor("op_50716_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_447_cast_fp16 = layer_norm(axes = hidden_states_447_axes_0, beta = hidden_states_447_beta_0_to_fp16, epsilon = var_50716_to_fp16, gamma = hidden_states_447_gamma_0_to_fp16, x = inputs_337_cast_fp16)[name = tensor("hidden_states_447_cast_fp16")]; + tensor var_50731 = const()[name = tensor("op_50731"), val = tensor([1, 1])]; + tensor var_50733 = const()[name = tensor("op_50733"), val = tensor([1, 1])]; + tensor q_225_pad_type_0 = const()[name = tensor("q_225_pad_type_0"), val = tensor("custom")]; + tensor q_225_pad_0 = const()[name = tensor("q_225_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_2_attn1_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_2_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4311122176)))]; + tensor q_225_cast_fp16 = conv(dilations = var_50733, groups = var_30355, pad = q_225_pad_0, pad_type = q_225_pad_type_0, strides = var_50731, weight = up_blocks_0_attentions_2_transformer_blocks_2_attn1_to_q_weight_to_fp16, x = hidden_states_447_cast_fp16)[name = tensor("q_225_cast_fp16")]; + tensor var_50737 = const()[name = tensor("op_50737"), val = tensor([1, 1])]; + tensor var_50739 = const()[name = tensor("op_50739"), val = tensor([1, 1])]; + tensor k_449_pad_type_0 = const()[name = tensor("k_449_pad_type_0"), val = tensor("custom")]; + tensor k_449_pad_0 = const()[name = tensor("k_449_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_2_attn1_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_2_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4314399040)))]; + tensor k_449_cast_fp16 = conv(dilations = var_50739, groups = var_30355, pad = k_449_pad_0, pad_type = k_449_pad_type_0, strides = var_50737, weight = up_blocks_0_attentions_2_transformer_blocks_2_attn1_to_k_weight_to_fp16, x = hidden_states_447_cast_fp16)[name = tensor("k_449_cast_fp16")]; + tensor var_50743 = const()[name = tensor("op_50743"), val = tensor([1, 1])]; + tensor var_50745 = const()[name = tensor("op_50745"), val = tensor([1, 1])]; + tensor v_225_pad_type_0 = const()[name = tensor("v_225_pad_type_0"), val = tensor("custom")]; + tensor v_225_pad_0 = const()[name = tensor("v_225_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_2_attn1_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_2_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4317675904)))]; + tensor v_225_cast_fp16 = conv(dilations = var_50745, groups = var_30355, pad = v_225_pad_0, pad_type = v_225_pad_type_0, strides = var_50743, weight = up_blocks_0_attentions_2_transformer_blocks_2_attn1_to_v_weight_to_fp16, x = hidden_states_447_cast_fp16)[name = tensor("v_225_cast_fp16")]; + tensor var_50749_begin_0 = const()[name = tensor("op_50749_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_50749_end_0 = const()[name = tensor("op_50749_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_50749_end_mask_0 = const()[name = tensor("op_50749_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50749_cast_fp16 = slice_by_index(begin = var_50749_begin_0, end = var_50749_end_0, end_mask = var_50749_end_mask_0, x = q_225_cast_fp16)[name = tensor("op_50749_cast_fp16")]; + tensor var_50753_begin_0 = const()[name = tensor("op_50753_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_50753_end_0 = const()[name = tensor("op_50753_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_50753_end_mask_0 = const()[name = tensor("op_50753_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50753_cast_fp16 = slice_by_index(begin = var_50753_begin_0, end = var_50753_end_0, end_mask = var_50753_end_mask_0, x = q_225_cast_fp16)[name = tensor("op_50753_cast_fp16")]; + tensor var_50757_begin_0 = const()[name = tensor("op_50757_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_50757_end_0 = const()[name = tensor("op_50757_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_50757_end_mask_0 = const()[name = tensor("op_50757_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50757_cast_fp16 = slice_by_index(begin = var_50757_begin_0, end = var_50757_end_0, end_mask = var_50757_end_mask_0, x = q_225_cast_fp16)[name = tensor("op_50757_cast_fp16")]; + tensor var_50761_begin_0 = const()[name = tensor("op_50761_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_50761_end_0 = const()[name = tensor("op_50761_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_50761_end_mask_0 = const()[name = tensor("op_50761_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50761_cast_fp16 = slice_by_index(begin = var_50761_begin_0, end = var_50761_end_0, end_mask = var_50761_end_mask_0, x = q_225_cast_fp16)[name = tensor("op_50761_cast_fp16")]; + tensor var_50765_begin_0 = const()[name = tensor("op_50765_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_50765_end_0 = const()[name = tensor("op_50765_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_50765_end_mask_0 = const()[name = tensor("op_50765_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50765_cast_fp16 = slice_by_index(begin = var_50765_begin_0, end = var_50765_end_0, end_mask = var_50765_end_mask_0, x = q_225_cast_fp16)[name = tensor("op_50765_cast_fp16")]; + tensor var_50769_begin_0 = const()[name = tensor("op_50769_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_50769_end_0 = const()[name = tensor("op_50769_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_50769_end_mask_0 = const()[name = tensor("op_50769_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50769_cast_fp16 = slice_by_index(begin = var_50769_begin_0, end = var_50769_end_0, end_mask = var_50769_end_mask_0, x = q_225_cast_fp16)[name = tensor("op_50769_cast_fp16")]; + tensor var_50773_begin_0 = const()[name = tensor("op_50773_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_50773_end_0 = const()[name = tensor("op_50773_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_50773_end_mask_0 = const()[name = tensor("op_50773_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50773_cast_fp16 = slice_by_index(begin = var_50773_begin_0, end = var_50773_end_0, end_mask = var_50773_end_mask_0, x = q_225_cast_fp16)[name = tensor("op_50773_cast_fp16")]; + tensor var_50777_begin_0 = const()[name = tensor("op_50777_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_50777_end_0 = const()[name = tensor("op_50777_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_50777_end_mask_0 = const()[name = tensor("op_50777_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50777_cast_fp16 = slice_by_index(begin = var_50777_begin_0, end = var_50777_end_0, end_mask = var_50777_end_mask_0, x = q_225_cast_fp16)[name = tensor("op_50777_cast_fp16")]; + tensor var_50781_begin_0 = const()[name = tensor("op_50781_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_50781_end_0 = const()[name = tensor("op_50781_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_50781_end_mask_0 = const()[name = tensor("op_50781_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50781_cast_fp16 = slice_by_index(begin = var_50781_begin_0, end = var_50781_end_0, end_mask = var_50781_end_mask_0, x = q_225_cast_fp16)[name = tensor("op_50781_cast_fp16")]; + tensor var_50785_begin_0 = const()[name = tensor("op_50785_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_50785_end_0 = const()[name = tensor("op_50785_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_50785_end_mask_0 = const()[name = tensor("op_50785_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50785_cast_fp16 = slice_by_index(begin = var_50785_begin_0, end = var_50785_end_0, end_mask = var_50785_end_mask_0, x = q_225_cast_fp16)[name = tensor("op_50785_cast_fp16")]; + tensor var_50789_begin_0 = const()[name = tensor("op_50789_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_50789_end_0 = const()[name = tensor("op_50789_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_50789_end_mask_0 = const()[name = tensor("op_50789_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50789_cast_fp16 = slice_by_index(begin = var_50789_begin_0, end = var_50789_end_0, end_mask = var_50789_end_mask_0, x = q_225_cast_fp16)[name = tensor("op_50789_cast_fp16")]; + tensor var_50793_begin_0 = const()[name = tensor("op_50793_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_50793_end_0 = const()[name = tensor("op_50793_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_50793_end_mask_0 = const()[name = tensor("op_50793_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50793_cast_fp16 = slice_by_index(begin = var_50793_begin_0, end = var_50793_end_0, end_mask = var_50793_end_mask_0, x = q_225_cast_fp16)[name = tensor("op_50793_cast_fp16")]; + tensor var_50797_begin_0 = const()[name = tensor("op_50797_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_50797_end_0 = const()[name = tensor("op_50797_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_50797_end_mask_0 = const()[name = tensor("op_50797_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50797_cast_fp16 = slice_by_index(begin = var_50797_begin_0, end = var_50797_end_0, end_mask = var_50797_end_mask_0, x = q_225_cast_fp16)[name = tensor("op_50797_cast_fp16")]; + tensor var_50801_begin_0 = const()[name = tensor("op_50801_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_50801_end_0 = const()[name = tensor("op_50801_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_50801_end_mask_0 = const()[name = tensor("op_50801_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50801_cast_fp16 = slice_by_index(begin = var_50801_begin_0, end = var_50801_end_0, end_mask = var_50801_end_mask_0, x = q_225_cast_fp16)[name = tensor("op_50801_cast_fp16")]; + tensor var_50805_begin_0 = const()[name = tensor("op_50805_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_50805_end_0 = const()[name = tensor("op_50805_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_50805_end_mask_0 = const()[name = tensor("op_50805_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50805_cast_fp16 = slice_by_index(begin = var_50805_begin_0, end = var_50805_end_0, end_mask = var_50805_end_mask_0, x = q_225_cast_fp16)[name = tensor("op_50805_cast_fp16")]; + tensor var_50809_begin_0 = const()[name = tensor("op_50809_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_50809_end_0 = const()[name = tensor("op_50809_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_50809_end_mask_0 = const()[name = tensor("op_50809_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50809_cast_fp16 = slice_by_index(begin = var_50809_begin_0, end = var_50809_end_0, end_mask = var_50809_end_mask_0, x = q_225_cast_fp16)[name = tensor("op_50809_cast_fp16")]; + tensor var_50813_begin_0 = const()[name = tensor("op_50813_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_50813_end_0 = const()[name = tensor("op_50813_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_50813_end_mask_0 = const()[name = tensor("op_50813_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50813_cast_fp16 = slice_by_index(begin = var_50813_begin_0, end = var_50813_end_0, end_mask = var_50813_end_mask_0, x = q_225_cast_fp16)[name = tensor("op_50813_cast_fp16")]; + tensor var_50817_begin_0 = const()[name = tensor("op_50817_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_50817_end_0 = const()[name = tensor("op_50817_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_50817_end_mask_0 = const()[name = tensor("op_50817_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50817_cast_fp16 = slice_by_index(begin = var_50817_begin_0, end = var_50817_end_0, end_mask = var_50817_end_mask_0, x = q_225_cast_fp16)[name = tensor("op_50817_cast_fp16")]; + tensor var_50821_begin_0 = const()[name = tensor("op_50821_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_50821_end_0 = const()[name = tensor("op_50821_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_50821_end_mask_0 = const()[name = tensor("op_50821_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50821_cast_fp16 = slice_by_index(begin = var_50821_begin_0, end = var_50821_end_0, end_mask = var_50821_end_mask_0, x = q_225_cast_fp16)[name = tensor("op_50821_cast_fp16")]; + tensor var_50825_begin_0 = const()[name = tensor("op_50825_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_50825_end_0 = const()[name = tensor("op_50825_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_50825_end_mask_0 = const()[name = tensor("op_50825_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50825_cast_fp16 = slice_by_index(begin = var_50825_begin_0, end = var_50825_end_0, end_mask = var_50825_end_mask_0, x = q_225_cast_fp16)[name = tensor("op_50825_cast_fp16")]; + tensor k_451_perm_0 = const()[name = tensor("k_451_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_50832_begin_0 = const()[name = tensor("op_50832_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_50832_end_0 = const()[name = tensor("op_50832_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_50832_end_mask_0 = const()[name = tensor("op_50832_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_451_cast_fp16 = transpose(perm = k_451_perm_0, x = k_449_cast_fp16)[name = tensor("transpose_27")]; + tensor var_50832_cast_fp16 = slice_by_index(begin = var_50832_begin_0, end = var_50832_end_0, end_mask = var_50832_end_mask_0, x = k_451_cast_fp16)[name = tensor("op_50832_cast_fp16")]; + tensor var_50836_begin_0 = const()[name = tensor("op_50836_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_50836_end_0 = const()[name = tensor("op_50836_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_50836_end_mask_0 = const()[name = tensor("op_50836_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_50836_cast_fp16 = slice_by_index(begin = var_50836_begin_0, end = var_50836_end_0, end_mask = var_50836_end_mask_0, x = k_451_cast_fp16)[name = tensor("op_50836_cast_fp16")]; + tensor var_50840_begin_0 = const()[name = tensor("op_50840_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_50840_end_0 = const()[name = tensor("op_50840_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_50840_end_mask_0 = const()[name = tensor("op_50840_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_50840_cast_fp16 = slice_by_index(begin = var_50840_begin_0, end = var_50840_end_0, end_mask = var_50840_end_mask_0, x = k_451_cast_fp16)[name = tensor("op_50840_cast_fp16")]; + tensor var_50844_begin_0 = const()[name = tensor("op_50844_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_50844_end_0 = const()[name = tensor("op_50844_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_50844_end_mask_0 = const()[name = tensor("op_50844_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_50844_cast_fp16 = slice_by_index(begin = var_50844_begin_0, end = var_50844_end_0, end_mask = var_50844_end_mask_0, x = k_451_cast_fp16)[name = tensor("op_50844_cast_fp16")]; + tensor var_50848_begin_0 = const()[name = tensor("op_50848_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_50848_end_0 = const()[name = tensor("op_50848_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_50848_end_mask_0 = const()[name = tensor("op_50848_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_50848_cast_fp16 = slice_by_index(begin = var_50848_begin_0, end = var_50848_end_0, end_mask = var_50848_end_mask_0, x = k_451_cast_fp16)[name = tensor("op_50848_cast_fp16")]; + tensor var_50852_begin_0 = const()[name = tensor("op_50852_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_50852_end_0 = const()[name = tensor("op_50852_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_50852_end_mask_0 = const()[name = tensor("op_50852_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_50852_cast_fp16 = slice_by_index(begin = var_50852_begin_0, end = var_50852_end_0, end_mask = var_50852_end_mask_0, x = k_451_cast_fp16)[name = tensor("op_50852_cast_fp16")]; + tensor var_50856_begin_0 = const()[name = tensor("op_50856_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_50856_end_0 = const()[name = tensor("op_50856_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_50856_end_mask_0 = const()[name = tensor("op_50856_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_50856_cast_fp16 = slice_by_index(begin = var_50856_begin_0, end = var_50856_end_0, end_mask = var_50856_end_mask_0, x = k_451_cast_fp16)[name = tensor("op_50856_cast_fp16")]; + tensor var_50860_begin_0 = const()[name = tensor("op_50860_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_50860_end_0 = const()[name = tensor("op_50860_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_50860_end_mask_0 = const()[name = tensor("op_50860_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_50860_cast_fp16 = slice_by_index(begin = var_50860_begin_0, end = var_50860_end_0, end_mask = var_50860_end_mask_0, x = k_451_cast_fp16)[name = tensor("op_50860_cast_fp16")]; + tensor var_50864_begin_0 = const()[name = tensor("op_50864_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_50864_end_0 = const()[name = tensor("op_50864_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_50864_end_mask_0 = const()[name = tensor("op_50864_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_50864_cast_fp16 = slice_by_index(begin = var_50864_begin_0, end = var_50864_end_0, end_mask = var_50864_end_mask_0, x = k_451_cast_fp16)[name = tensor("op_50864_cast_fp16")]; + tensor var_50868_begin_0 = const()[name = tensor("op_50868_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_50868_end_0 = const()[name = tensor("op_50868_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_50868_end_mask_0 = const()[name = tensor("op_50868_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_50868_cast_fp16 = slice_by_index(begin = var_50868_begin_0, end = var_50868_end_0, end_mask = var_50868_end_mask_0, x = k_451_cast_fp16)[name = tensor("op_50868_cast_fp16")]; + tensor var_50872_begin_0 = const()[name = tensor("op_50872_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_50872_end_0 = const()[name = tensor("op_50872_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_50872_end_mask_0 = const()[name = tensor("op_50872_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_50872_cast_fp16 = slice_by_index(begin = var_50872_begin_0, end = var_50872_end_0, end_mask = var_50872_end_mask_0, x = k_451_cast_fp16)[name = tensor("op_50872_cast_fp16")]; + tensor var_50876_begin_0 = const()[name = tensor("op_50876_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_50876_end_0 = const()[name = tensor("op_50876_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_50876_end_mask_0 = const()[name = tensor("op_50876_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_50876_cast_fp16 = slice_by_index(begin = var_50876_begin_0, end = var_50876_end_0, end_mask = var_50876_end_mask_0, x = k_451_cast_fp16)[name = tensor("op_50876_cast_fp16")]; + tensor var_50880_begin_0 = const()[name = tensor("op_50880_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_50880_end_0 = const()[name = tensor("op_50880_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_50880_end_mask_0 = const()[name = tensor("op_50880_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_50880_cast_fp16 = slice_by_index(begin = var_50880_begin_0, end = var_50880_end_0, end_mask = var_50880_end_mask_0, x = k_451_cast_fp16)[name = tensor("op_50880_cast_fp16")]; + tensor var_50884_begin_0 = const()[name = tensor("op_50884_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_50884_end_0 = const()[name = tensor("op_50884_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_50884_end_mask_0 = const()[name = tensor("op_50884_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_50884_cast_fp16 = slice_by_index(begin = var_50884_begin_0, end = var_50884_end_0, end_mask = var_50884_end_mask_0, x = k_451_cast_fp16)[name = tensor("op_50884_cast_fp16")]; + tensor var_50888_begin_0 = const()[name = tensor("op_50888_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_50888_end_0 = const()[name = tensor("op_50888_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_50888_end_mask_0 = const()[name = tensor("op_50888_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_50888_cast_fp16 = slice_by_index(begin = var_50888_begin_0, end = var_50888_end_0, end_mask = var_50888_end_mask_0, x = k_451_cast_fp16)[name = tensor("op_50888_cast_fp16")]; + tensor var_50892_begin_0 = const()[name = tensor("op_50892_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_50892_end_0 = const()[name = tensor("op_50892_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_50892_end_mask_0 = const()[name = tensor("op_50892_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_50892_cast_fp16 = slice_by_index(begin = var_50892_begin_0, end = var_50892_end_0, end_mask = var_50892_end_mask_0, x = k_451_cast_fp16)[name = tensor("op_50892_cast_fp16")]; + tensor var_50896_begin_0 = const()[name = tensor("op_50896_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_50896_end_0 = const()[name = tensor("op_50896_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_50896_end_mask_0 = const()[name = tensor("op_50896_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_50896_cast_fp16 = slice_by_index(begin = var_50896_begin_0, end = var_50896_end_0, end_mask = var_50896_end_mask_0, x = k_451_cast_fp16)[name = tensor("op_50896_cast_fp16")]; + tensor var_50900_begin_0 = const()[name = tensor("op_50900_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_50900_end_0 = const()[name = tensor("op_50900_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_50900_end_mask_0 = const()[name = tensor("op_50900_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_50900_cast_fp16 = slice_by_index(begin = var_50900_begin_0, end = var_50900_end_0, end_mask = var_50900_end_mask_0, x = k_451_cast_fp16)[name = tensor("op_50900_cast_fp16")]; + tensor var_50904_begin_0 = const()[name = tensor("op_50904_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_50904_end_0 = const()[name = tensor("op_50904_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_50904_end_mask_0 = const()[name = tensor("op_50904_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_50904_cast_fp16 = slice_by_index(begin = var_50904_begin_0, end = var_50904_end_0, end_mask = var_50904_end_mask_0, x = k_451_cast_fp16)[name = tensor("op_50904_cast_fp16")]; + tensor var_50908_begin_0 = const()[name = tensor("op_50908_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_50908_end_0 = const()[name = tensor("op_50908_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_50908_end_mask_0 = const()[name = tensor("op_50908_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_50908_cast_fp16 = slice_by_index(begin = var_50908_begin_0, end = var_50908_end_0, end_mask = var_50908_end_mask_0, x = k_451_cast_fp16)[name = tensor("op_50908_cast_fp16")]; + tensor var_50910_begin_0 = const()[name = tensor("op_50910_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_50910_end_0 = const()[name = tensor("op_50910_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_50910_end_mask_0 = const()[name = tensor("op_50910_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50910_cast_fp16 = slice_by_index(begin = var_50910_begin_0, end = var_50910_end_0, end_mask = var_50910_end_mask_0, x = v_225_cast_fp16)[name = tensor("op_50910_cast_fp16")]; + tensor var_50914_begin_0 = const()[name = tensor("op_50914_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_50914_end_0 = const()[name = tensor("op_50914_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_50914_end_mask_0 = const()[name = tensor("op_50914_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50914_cast_fp16 = slice_by_index(begin = var_50914_begin_0, end = var_50914_end_0, end_mask = var_50914_end_mask_0, x = v_225_cast_fp16)[name = tensor("op_50914_cast_fp16")]; + tensor var_50918_begin_0 = const()[name = tensor("op_50918_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_50918_end_0 = const()[name = tensor("op_50918_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_50918_end_mask_0 = const()[name = tensor("op_50918_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50918_cast_fp16 = slice_by_index(begin = var_50918_begin_0, end = var_50918_end_0, end_mask = var_50918_end_mask_0, x = v_225_cast_fp16)[name = tensor("op_50918_cast_fp16")]; + tensor var_50922_begin_0 = const()[name = tensor("op_50922_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_50922_end_0 = const()[name = tensor("op_50922_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_50922_end_mask_0 = const()[name = tensor("op_50922_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50922_cast_fp16 = slice_by_index(begin = var_50922_begin_0, end = var_50922_end_0, end_mask = var_50922_end_mask_0, x = v_225_cast_fp16)[name = tensor("op_50922_cast_fp16")]; + tensor var_50926_begin_0 = const()[name = tensor("op_50926_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_50926_end_0 = const()[name = tensor("op_50926_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_50926_end_mask_0 = const()[name = tensor("op_50926_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50926_cast_fp16 = slice_by_index(begin = var_50926_begin_0, end = var_50926_end_0, end_mask = var_50926_end_mask_0, x = v_225_cast_fp16)[name = tensor("op_50926_cast_fp16")]; + tensor var_50930_begin_0 = const()[name = tensor("op_50930_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_50930_end_0 = const()[name = tensor("op_50930_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_50930_end_mask_0 = const()[name = tensor("op_50930_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50930_cast_fp16 = slice_by_index(begin = var_50930_begin_0, end = var_50930_end_0, end_mask = var_50930_end_mask_0, x = v_225_cast_fp16)[name = tensor("op_50930_cast_fp16")]; + tensor var_50934_begin_0 = const()[name = tensor("op_50934_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_50934_end_0 = const()[name = tensor("op_50934_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_50934_end_mask_0 = const()[name = tensor("op_50934_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50934_cast_fp16 = slice_by_index(begin = var_50934_begin_0, end = var_50934_end_0, end_mask = var_50934_end_mask_0, x = v_225_cast_fp16)[name = tensor("op_50934_cast_fp16")]; + tensor var_50938_begin_0 = const()[name = tensor("op_50938_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_50938_end_0 = const()[name = tensor("op_50938_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_50938_end_mask_0 = const()[name = tensor("op_50938_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50938_cast_fp16 = slice_by_index(begin = var_50938_begin_0, end = var_50938_end_0, end_mask = var_50938_end_mask_0, x = v_225_cast_fp16)[name = tensor("op_50938_cast_fp16")]; + tensor var_50942_begin_0 = const()[name = tensor("op_50942_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_50942_end_0 = const()[name = tensor("op_50942_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_50942_end_mask_0 = const()[name = tensor("op_50942_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50942_cast_fp16 = slice_by_index(begin = var_50942_begin_0, end = var_50942_end_0, end_mask = var_50942_end_mask_0, x = v_225_cast_fp16)[name = tensor("op_50942_cast_fp16")]; + tensor var_50946_begin_0 = const()[name = tensor("op_50946_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_50946_end_0 = const()[name = tensor("op_50946_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_50946_end_mask_0 = const()[name = tensor("op_50946_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50946_cast_fp16 = slice_by_index(begin = var_50946_begin_0, end = var_50946_end_0, end_mask = var_50946_end_mask_0, x = v_225_cast_fp16)[name = tensor("op_50946_cast_fp16")]; + tensor var_50950_begin_0 = const()[name = tensor("op_50950_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_50950_end_0 = const()[name = tensor("op_50950_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_50950_end_mask_0 = const()[name = tensor("op_50950_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50950_cast_fp16 = slice_by_index(begin = var_50950_begin_0, end = var_50950_end_0, end_mask = var_50950_end_mask_0, x = v_225_cast_fp16)[name = tensor("op_50950_cast_fp16")]; + tensor var_50954_begin_0 = const()[name = tensor("op_50954_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_50954_end_0 = const()[name = tensor("op_50954_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_50954_end_mask_0 = const()[name = tensor("op_50954_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50954_cast_fp16 = slice_by_index(begin = var_50954_begin_0, end = var_50954_end_0, end_mask = var_50954_end_mask_0, x = v_225_cast_fp16)[name = tensor("op_50954_cast_fp16")]; + tensor var_50958_begin_0 = const()[name = tensor("op_50958_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_50958_end_0 = const()[name = tensor("op_50958_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_50958_end_mask_0 = const()[name = tensor("op_50958_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50958_cast_fp16 = slice_by_index(begin = var_50958_begin_0, end = var_50958_end_0, end_mask = var_50958_end_mask_0, x = v_225_cast_fp16)[name = tensor("op_50958_cast_fp16")]; + tensor var_50962_begin_0 = const()[name = tensor("op_50962_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_50962_end_0 = const()[name = tensor("op_50962_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_50962_end_mask_0 = const()[name = tensor("op_50962_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50962_cast_fp16 = slice_by_index(begin = var_50962_begin_0, end = var_50962_end_0, end_mask = var_50962_end_mask_0, x = v_225_cast_fp16)[name = tensor("op_50962_cast_fp16")]; + tensor var_50966_begin_0 = const()[name = tensor("op_50966_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_50966_end_0 = const()[name = tensor("op_50966_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_50966_end_mask_0 = const()[name = tensor("op_50966_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50966_cast_fp16 = slice_by_index(begin = var_50966_begin_0, end = var_50966_end_0, end_mask = var_50966_end_mask_0, x = v_225_cast_fp16)[name = tensor("op_50966_cast_fp16")]; + tensor var_50970_begin_0 = const()[name = tensor("op_50970_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_50970_end_0 = const()[name = tensor("op_50970_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_50970_end_mask_0 = const()[name = tensor("op_50970_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50970_cast_fp16 = slice_by_index(begin = var_50970_begin_0, end = var_50970_end_0, end_mask = var_50970_end_mask_0, x = v_225_cast_fp16)[name = tensor("op_50970_cast_fp16")]; + tensor var_50974_begin_0 = const()[name = tensor("op_50974_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_50974_end_0 = const()[name = tensor("op_50974_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_50974_end_mask_0 = const()[name = tensor("op_50974_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50974_cast_fp16 = slice_by_index(begin = var_50974_begin_0, end = var_50974_end_0, end_mask = var_50974_end_mask_0, x = v_225_cast_fp16)[name = tensor("op_50974_cast_fp16")]; + tensor var_50978_begin_0 = const()[name = tensor("op_50978_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_50978_end_0 = const()[name = tensor("op_50978_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_50978_end_mask_0 = const()[name = tensor("op_50978_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50978_cast_fp16 = slice_by_index(begin = var_50978_begin_0, end = var_50978_end_0, end_mask = var_50978_end_mask_0, x = v_225_cast_fp16)[name = tensor("op_50978_cast_fp16")]; + tensor var_50982_begin_0 = const()[name = tensor("op_50982_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_50982_end_0 = const()[name = tensor("op_50982_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_50982_end_mask_0 = const()[name = tensor("op_50982_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50982_cast_fp16 = slice_by_index(begin = var_50982_begin_0, end = var_50982_end_0, end_mask = var_50982_end_mask_0, x = v_225_cast_fp16)[name = tensor("op_50982_cast_fp16")]; + tensor var_50986_begin_0 = const()[name = tensor("op_50986_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_50986_end_0 = const()[name = tensor("op_50986_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_50986_end_mask_0 = const()[name = tensor("op_50986_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50986_cast_fp16 = slice_by_index(begin = var_50986_begin_0, end = var_50986_end_0, end_mask = var_50986_end_mask_0, x = v_225_cast_fp16)[name = tensor("op_50986_cast_fp16")]; + tensor var_50990_equation_0 = const()[name = tensor("op_50990_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50990_cast_fp16 = einsum(equation = var_50990_equation_0, values = (var_50832_cast_fp16, var_50749_cast_fp16))[name = tensor("op_50990_cast_fp16")]; + tensor var_50991_to_fp16 = const()[name = tensor("op_50991_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4321_cast_fp16 = mul(x = var_50990_cast_fp16, y = var_50991_to_fp16)[name = tensor("aw_4321_cast_fp16")]; + tensor var_50994_equation_0 = const()[name = tensor("op_50994_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50994_cast_fp16 = einsum(equation = var_50994_equation_0, values = (var_50836_cast_fp16, var_50753_cast_fp16))[name = tensor("op_50994_cast_fp16")]; + tensor var_50995_to_fp16 = const()[name = tensor("op_50995_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4323_cast_fp16 = mul(x = var_50994_cast_fp16, y = var_50995_to_fp16)[name = tensor("aw_4323_cast_fp16")]; + tensor var_50998_equation_0 = const()[name = tensor("op_50998_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50998_cast_fp16 = einsum(equation = var_50998_equation_0, values = (var_50840_cast_fp16, var_50757_cast_fp16))[name = tensor("op_50998_cast_fp16")]; + tensor var_50999_to_fp16 = const()[name = tensor("op_50999_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4325_cast_fp16 = mul(x = var_50998_cast_fp16, y = var_50999_to_fp16)[name = tensor("aw_4325_cast_fp16")]; + tensor var_51002_equation_0 = const()[name = tensor("op_51002_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51002_cast_fp16 = einsum(equation = var_51002_equation_0, values = (var_50844_cast_fp16, var_50761_cast_fp16))[name = tensor("op_51002_cast_fp16")]; + tensor var_51003_to_fp16 = const()[name = tensor("op_51003_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4327_cast_fp16 = mul(x = var_51002_cast_fp16, y = var_51003_to_fp16)[name = tensor("aw_4327_cast_fp16")]; + tensor var_51006_equation_0 = const()[name = tensor("op_51006_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51006_cast_fp16 = einsum(equation = var_51006_equation_0, values = (var_50848_cast_fp16, var_50765_cast_fp16))[name = tensor("op_51006_cast_fp16")]; + tensor var_51007_to_fp16 = const()[name = tensor("op_51007_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4329_cast_fp16 = mul(x = var_51006_cast_fp16, y = var_51007_to_fp16)[name = tensor("aw_4329_cast_fp16")]; + tensor var_51010_equation_0 = const()[name = tensor("op_51010_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51010_cast_fp16 = einsum(equation = var_51010_equation_0, values = (var_50852_cast_fp16, var_50769_cast_fp16))[name = tensor("op_51010_cast_fp16")]; + tensor var_51011_to_fp16 = const()[name = tensor("op_51011_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4331_cast_fp16 = mul(x = var_51010_cast_fp16, y = var_51011_to_fp16)[name = tensor("aw_4331_cast_fp16")]; + tensor var_51014_equation_0 = const()[name = tensor("op_51014_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51014_cast_fp16 = einsum(equation = var_51014_equation_0, values = (var_50856_cast_fp16, var_50773_cast_fp16))[name = tensor("op_51014_cast_fp16")]; + tensor var_51015_to_fp16 = const()[name = tensor("op_51015_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4333_cast_fp16 = mul(x = var_51014_cast_fp16, y = var_51015_to_fp16)[name = tensor("aw_4333_cast_fp16")]; + tensor var_51018_equation_0 = const()[name = tensor("op_51018_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51018_cast_fp16 = einsum(equation = var_51018_equation_0, values = (var_50860_cast_fp16, var_50777_cast_fp16))[name = tensor("op_51018_cast_fp16")]; + tensor var_51019_to_fp16 = const()[name = tensor("op_51019_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4335_cast_fp16 = mul(x = var_51018_cast_fp16, y = var_51019_to_fp16)[name = tensor("aw_4335_cast_fp16")]; + tensor var_51022_equation_0 = const()[name = tensor("op_51022_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51022_cast_fp16 = einsum(equation = var_51022_equation_0, values = (var_50864_cast_fp16, var_50781_cast_fp16))[name = tensor("op_51022_cast_fp16")]; + tensor var_51023_to_fp16 = const()[name = tensor("op_51023_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4337_cast_fp16 = mul(x = var_51022_cast_fp16, y = var_51023_to_fp16)[name = tensor("aw_4337_cast_fp16")]; + tensor var_51026_equation_0 = const()[name = tensor("op_51026_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51026_cast_fp16 = einsum(equation = var_51026_equation_0, values = (var_50868_cast_fp16, var_50785_cast_fp16))[name = tensor("op_51026_cast_fp16")]; + tensor var_51027_to_fp16 = const()[name = tensor("op_51027_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4339_cast_fp16 = mul(x = var_51026_cast_fp16, y = var_51027_to_fp16)[name = tensor("aw_4339_cast_fp16")]; + tensor var_51030_equation_0 = const()[name = tensor("op_51030_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51030_cast_fp16 = einsum(equation = var_51030_equation_0, values = (var_50872_cast_fp16, var_50789_cast_fp16))[name = tensor("op_51030_cast_fp16")]; + tensor var_51031_to_fp16 = const()[name = tensor("op_51031_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4341_cast_fp16 = mul(x = var_51030_cast_fp16, y = var_51031_to_fp16)[name = tensor("aw_4341_cast_fp16")]; + tensor var_51034_equation_0 = const()[name = tensor("op_51034_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51034_cast_fp16 = einsum(equation = var_51034_equation_0, values = (var_50876_cast_fp16, var_50793_cast_fp16))[name = tensor("op_51034_cast_fp16")]; + tensor var_51035_to_fp16 = const()[name = tensor("op_51035_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4343_cast_fp16 = mul(x = var_51034_cast_fp16, y = var_51035_to_fp16)[name = tensor("aw_4343_cast_fp16")]; + tensor var_51038_equation_0 = const()[name = tensor("op_51038_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51038_cast_fp16 = einsum(equation = var_51038_equation_0, values = (var_50880_cast_fp16, var_50797_cast_fp16))[name = tensor("op_51038_cast_fp16")]; + tensor var_51039_to_fp16 = const()[name = tensor("op_51039_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4345_cast_fp16 = mul(x = var_51038_cast_fp16, y = var_51039_to_fp16)[name = tensor("aw_4345_cast_fp16")]; + tensor var_51042_equation_0 = const()[name = tensor("op_51042_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51042_cast_fp16 = einsum(equation = var_51042_equation_0, values = (var_50884_cast_fp16, var_50801_cast_fp16))[name = tensor("op_51042_cast_fp16")]; + tensor var_51043_to_fp16 = const()[name = tensor("op_51043_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4347_cast_fp16 = mul(x = var_51042_cast_fp16, y = var_51043_to_fp16)[name = tensor("aw_4347_cast_fp16")]; + tensor var_51046_equation_0 = const()[name = tensor("op_51046_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51046_cast_fp16 = einsum(equation = var_51046_equation_0, values = (var_50888_cast_fp16, var_50805_cast_fp16))[name = tensor("op_51046_cast_fp16")]; + tensor var_51047_to_fp16 = const()[name = tensor("op_51047_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4349_cast_fp16 = mul(x = var_51046_cast_fp16, y = var_51047_to_fp16)[name = tensor("aw_4349_cast_fp16")]; + tensor var_51050_equation_0 = const()[name = tensor("op_51050_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51050_cast_fp16 = einsum(equation = var_51050_equation_0, values = (var_50892_cast_fp16, var_50809_cast_fp16))[name = tensor("op_51050_cast_fp16")]; + tensor var_51051_to_fp16 = const()[name = tensor("op_51051_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4351_cast_fp16 = mul(x = var_51050_cast_fp16, y = var_51051_to_fp16)[name = tensor("aw_4351_cast_fp16")]; + tensor var_51054_equation_0 = const()[name = tensor("op_51054_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51054_cast_fp16 = einsum(equation = var_51054_equation_0, values = (var_50896_cast_fp16, var_50813_cast_fp16))[name = tensor("op_51054_cast_fp16")]; + tensor var_51055_to_fp16 = const()[name = tensor("op_51055_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4353_cast_fp16 = mul(x = var_51054_cast_fp16, y = var_51055_to_fp16)[name = tensor("aw_4353_cast_fp16")]; + tensor var_51058_equation_0 = const()[name = tensor("op_51058_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51058_cast_fp16 = einsum(equation = var_51058_equation_0, values = (var_50900_cast_fp16, var_50817_cast_fp16))[name = tensor("op_51058_cast_fp16")]; + tensor var_51059_to_fp16 = const()[name = tensor("op_51059_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4355_cast_fp16 = mul(x = var_51058_cast_fp16, y = var_51059_to_fp16)[name = tensor("aw_4355_cast_fp16")]; + tensor var_51062_equation_0 = const()[name = tensor("op_51062_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51062_cast_fp16 = einsum(equation = var_51062_equation_0, values = (var_50904_cast_fp16, var_50821_cast_fp16))[name = tensor("op_51062_cast_fp16")]; + tensor var_51063_to_fp16 = const()[name = tensor("op_51063_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4357_cast_fp16 = mul(x = var_51062_cast_fp16, y = var_51063_to_fp16)[name = tensor("aw_4357_cast_fp16")]; + tensor var_51066_equation_0 = const()[name = tensor("op_51066_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51066_cast_fp16 = einsum(equation = var_51066_equation_0, values = (var_50908_cast_fp16, var_50825_cast_fp16))[name = tensor("op_51066_cast_fp16")]; + tensor var_51067_to_fp16 = const()[name = tensor("op_51067_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4359_cast_fp16 = mul(x = var_51066_cast_fp16, y = var_51067_to_fp16)[name = tensor("aw_4359_cast_fp16")]; + tensor var_51069_cast_fp16 = softmax(axis = var_30355, x = aw_4321_cast_fp16)[name = tensor("op_51069_cast_fp16")]; + tensor var_51070_cast_fp16 = softmax(axis = var_30355, x = aw_4323_cast_fp16)[name = tensor("op_51070_cast_fp16")]; + tensor var_51071_cast_fp16 = softmax(axis = var_30355, x = aw_4325_cast_fp16)[name = tensor("op_51071_cast_fp16")]; + tensor var_51072_cast_fp16 = softmax(axis = var_30355, x = aw_4327_cast_fp16)[name = tensor("op_51072_cast_fp16")]; + tensor var_51073_cast_fp16 = softmax(axis = var_30355, x = aw_4329_cast_fp16)[name = tensor("op_51073_cast_fp16")]; + tensor var_51074_cast_fp16 = softmax(axis = var_30355, x = aw_4331_cast_fp16)[name = tensor("op_51074_cast_fp16")]; + tensor var_51075_cast_fp16 = softmax(axis = var_30355, x = aw_4333_cast_fp16)[name = tensor("op_51075_cast_fp16")]; + tensor var_51076_cast_fp16 = softmax(axis = var_30355, x = aw_4335_cast_fp16)[name = tensor("op_51076_cast_fp16")]; + tensor var_51077_cast_fp16 = softmax(axis = var_30355, x = aw_4337_cast_fp16)[name = tensor("op_51077_cast_fp16")]; + tensor var_51078_cast_fp16 = softmax(axis = var_30355, x = aw_4339_cast_fp16)[name = tensor("op_51078_cast_fp16")]; + tensor var_51079_cast_fp16 = softmax(axis = var_30355, x = aw_4341_cast_fp16)[name = tensor("op_51079_cast_fp16")]; + tensor var_51080_cast_fp16 = softmax(axis = var_30355, x = aw_4343_cast_fp16)[name = tensor("op_51080_cast_fp16")]; + tensor var_51081_cast_fp16 = softmax(axis = var_30355, x = aw_4345_cast_fp16)[name = tensor("op_51081_cast_fp16")]; + tensor var_51082_cast_fp16 = softmax(axis = var_30355, x = aw_4347_cast_fp16)[name = tensor("op_51082_cast_fp16")]; + tensor var_51083_cast_fp16 = softmax(axis = var_30355, x = aw_4349_cast_fp16)[name = tensor("op_51083_cast_fp16")]; + tensor var_51084_cast_fp16 = softmax(axis = var_30355, x = aw_4351_cast_fp16)[name = tensor("op_51084_cast_fp16")]; + tensor var_51085_cast_fp16 = softmax(axis = var_30355, x = aw_4353_cast_fp16)[name = tensor("op_51085_cast_fp16")]; + tensor var_51086_cast_fp16 = softmax(axis = var_30355, x = aw_4355_cast_fp16)[name = tensor("op_51086_cast_fp16")]; + tensor var_51087_cast_fp16 = softmax(axis = var_30355, x = aw_4357_cast_fp16)[name = tensor("op_51087_cast_fp16")]; + tensor var_51088_cast_fp16 = softmax(axis = var_30355, x = aw_4359_cast_fp16)[name = tensor("op_51088_cast_fp16")]; + tensor var_51090_equation_0 = const()[name = tensor("op_51090_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_51090_cast_fp16 = einsum(equation = var_51090_equation_0, values = (var_50910_cast_fp16, var_51069_cast_fp16))[name = tensor("op_51090_cast_fp16")]; + tensor var_51092_equation_0 = const()[name = tensor("op_51092_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_51092_cast_fp16 = einsum(equation = var_51092_equation_0, values = (var_50914_cast_fp16, var_51070_cast_fp16))[name = tensor("op_51092_cast_fp16")]; + tensor var_51094_equation_0 = const()[name = tensor("op_51094_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_51094_cast_fp16 = einsum(equation = var_51094_equation_0, values = (var_50918_cast_fp16, var_51071_cast_fp16))[name = tensor("op_51094_cast_fp16")]; + tensor var_51096_equation_0 = const()[name = tensor("op_51096_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_51096_cast_fp16 = einsum(equation = var_51096_equation_0, values = (var_50922_cast_fp16, var_51072_cast_fp16))[name = tensor("op_51096_cast_fp16")]; + tensor var_51098_equation_0 = const()[name = tensor("op_51098_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_51098_cast_fp16 = einsum(equation = var_51098_equation_0, values = (var_50926_cast_fp16, var_51073_cast_fp16))[name = tensor("op_51098_cast_fp16")]; + tensor var_51100_equation_0 = const()[name = tensor("op_51100_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_51100_cast_fp16 = einsum(equation = var_51100_equation_0, values = (var_50930_cast_fp16, var_51074_cast_fp16))[name = tensor("op_51100_cast_fp16")]; + tensor var_51102_equation_0 = const()[name = tensor("op_51102_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_51102_cast_fp16 = einsum(equation = var_51102_equation_0, values = (var_50934_cast_fp16, var_51075_cast_fp16))[name = tensor("op_51102_cast_fp16")]; + tensor var_51104_equation_0 = const()[name = tensor("op_51104_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_51104_cast_fp16 = einsum(equation = var_51104_equation_0, values = (var_50938_cast_fp16, var_51076_cast_fp16))[name = tensor("op_51104_cast_fp16")]; + tensor var_51106_equation_0 = const()[name = tensor("op_51106_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_51106_cast_fp16 = einsum(equation = var_51106_equation_0, values = (var_50942_cast_fp16, var_51077_cast_fp16))[name = tensor("op_51106_cast_fp16")]; + tensor var_51108_equation_0 = const()[name = tensor("op_51108_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_51108_cast_fp16 = einsum(equation = var_51108_equation_0, values = (var_50946_cast_fp16, var_51078_cast_fp16))[name = tensor("op_51108_cast_fp16")]; + tensor var_51110_equation_0 = const()[name = tensor("op_51110_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_51110_cast_fp16 = einsum(equation = var_51110_equation_0, values = (var_50950_cast_fp16, var_51079_cast_fp16))[name = tensor("op_51110_cast_fp16")]; + tensor var_51112_equation_0 = const()[name = tensor("op_51112_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_51112_cast_fp16 = einsum(equation = var_51112_equation_0, values = (var_50954_cast_fp16, var_51080_cast_fp16))[name = tensor("op_51112_cast_fp16")]; + tensor var_51114_equation_0 = const()[name = tensor("op_51114_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_51114_cast_fp16 = einsum(equation = var_51114_equation_0, values = (var_50958_cast_fp16, var_51081_cast_fp16))[name = tensor("op_51114_cast_fp16")]; + tensor var_51116_equation_0 = const()[name = tensor("op_51116_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_51116_cast_fp16 = einsum(equation = var_51116_equation_0, values = (var_50962_cast_fp16, var_51082_cast_fp16))[name = tensor("op_51116_cast_fp16")]; + tensor var_51118_equation_0 = const()[name = tensor("op_51118_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_51118_cast_fp16 = einsum(equation = var_51118_equation_0, values = (var_50966_cast_fp16, var_51083_cast_fp16))[name = tensor("op_51118_cast_fp16")]; + tensor var_51120_equation_0 = const()[name = tensor("op_51120_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_51120_cast_fp16 = einsum(equation = var_51120_equation_0, values = (var_50970_cast_fp16, var_51084_cast_fp16))[name = tensor("op_51120_cast_fp16")]; + tensor var_51122_equation_0 = const()[name = tensor("op_51122_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_51122_cast_fp16 = einsum(equation = var_51122_equation_0, values = (var_50974_cast_fp16, var_51085_cast_fp16))[name = tensor("op_51122_cast_fp16")]; + tensor var_51124_equation_0 = const()[name = tensor("op_51124_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_51124_cast_fp16 = einsum(equation = var_51124_equation_0, values = (var_50978_cast_fp16, var_51086_cast_fp16))[name = tensor("op_51124_cast_fp16")]; + tensor var_51126_equation_0 = const()[name = tensor("op_51126_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_51126_cast_fp16 = einsum(equation = var_51126_equation_0, values = (var_50982_cast_fp16, var_51087_cast_fp16))[name = tensor("op_51126_cast_fp16")]; + tensor var_51128_equation_0 = const()[name = tensor("op_51128_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_51128_cast_fp16 = einsum(equation = var_51128_equation_0, values = (var_50986_cast_fp16, var_51088_cast_fp16))[name = tensor("op_51128_cast_fp16")]; + tensor input_651_interleave_0 = const()[name = tensor("input_651_interleave_0"), val = tensor(false)]; + tensor input_651_cast_fp16 = concat(axis = var_30355, interleave = input_651_interleave_0, values = (var_51090_cast_fp16, var_51092_cast_fp16, var_51094_cast_fp16, var_51096_cast_fp16, var_51098_cast_fp16, var_51100_cast_fp16, var_51102_cast_fp16, var_51104_cast_fp16, var_51106_cast_fp16, var_51108_cast_fp16, var_51110_cast_fp16, var_51112_cast_fp16, var_51114_cast_fp16, var_51116_cast_fp16, var_51118_cast_fp16, var_51120_cast_fp16, var_51122_cast_fp16, var_51124_cast_fp16, var_51126_cast_fp16, var_51128_cast_fp16))[name = tensor("input_651_cast_fp16")]; + tensor var_51134 = const()[name = tensor("op_51134"), val = tensor([1, 1])]; + tensor var_51136 = const()[name = tensor("op_51136"), val = tensor([1, 1])]; + tensor var_51138_pad_type_0 = const()[name = tensor("op_51138_pad_type_0"), val = tensor("custom")]; + tensor var_51138_pad_0 = const()[name = tensor("op_51138_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_2_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_2_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4320952768)))]; + tensor up_blocks_0_attentions_2_transformer_blocks_2_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_2_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4324229632)))]; + tensor var_51138_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_2_attn1_to_out_0_bias_to_fp16, dilations = var_51136, groups = var_30355, pad = var_51138_pad_0, pad_type = var_51138_pad_type_0, strides = var_51134, weight = up_blocks_0_attentions_2_transformer_blocks_2_attn1_to_out_0_weight_to_fp16, x = input_651_cast_fp16)[name = tensor("op_51138_cast_fp16")]; + tensor inputs_339_cast_fp16 = add(x = var_51138_cast_fp16, y = inputs_337_cast_fp16)[name = tensor("inputs_339_cast_fp16")]; + tensor hidden_states_449_axes_0 = const()[name = tensor("hidden_states_449_axes_0"), val = tensor([1])]; + tensor hidden_states_449_gamma_0_to_fp16 = const()[name = tensor("hidden_states_449_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4324232256)))]; + tensor hidden_states_449_beta_0_to_fp16 = const()[name = tensor("hidden_states_449_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4324234880)))]; + tensor var_51148_to_fp16 = const()[name = tensor("op_51148_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_449_cast_fp16 = layer_norm(axes = hidden_states_449_axes_0, beta = hidden_states_449_beta_0_to_fp16, epsilon = var_51148_to_fp16, gamma = hidden_states_449_gamma_0_to_fp16, x = inputs_339_cast_fp16)[name = tensor("hidden_states_449_cast_fp16")]; + tensor var_51163 = const()[name = tensor("op_51163"), val = tensor([1, 1])]; + tensor var_51165 = const()[name = tensor("op_51165"), val = tensor([1, 1])]; + tensor q_227_pad_type_0 = const()[name = tensor("q_227_pad_type_0"), val = tensor("custom")]; + tensor q_227_pad_0 = const()[name = tensor("q_227_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_2_attn2_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_2_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4324237504)))]; + tensor q_227_cast_fp16 = conv(dilations = var_51165, groups = var_30355, pad = q_227_pad_0, pad_type = q_227_pad_type_0, strides = var_51163, weight = up_blocks_0_attentions_2_transformer_blocks_2_attn2_to_q_weight_to_fp16, x = hidden_states_449_cast_fp16)[name = tensor("q_227_cast_fp16")]; + tensor var_51169 = const()[name = tensor("op_51169"), val = tensor([1, 1])]; + tensor var_51171 = const()[name = tensor("op_51171"), val = tensor([1, 1])]; + tensor k_453_pad_type_0 = const()[name = tensor("k_453_pad_type_0"), val = tensor("custom")]; + tensor k_453_pad_0 = const()[name = tensor("k_453_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_2_attn2_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_2_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4327514368)))]; + tensor k_453_cast_fp16 = conv(dilations = var_51171, groups = var_30355, pad = k_453_pad_0, pad_type = k_453_pad_type_0, strides = var_51169, weight = up_blocks_0_attentions_2_transformer_blocks_2_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_453_cast_fp16")]; + tensor var_51175 = const()[name = tensor("op_51175"), val = tensor([1, 1])]; + tensor var_51177 = const()[name = tensor("op_51177"), val = tensor([1, 1])]; + tensor v_227_pad_type_0 = const()[name = tensor("v_227_pad_type_0"), val = tensor("custom")]; + tensor v_227_pad_0 = const()[name = tensor("v_227_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_2_attn2_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_2_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4332757312)))]; + tensor v_227_cast_fp16 = conv(dilations = var_51177, groups = var_30355, pad = v_227_pad_0, pad_type = v_227_pad_type_0, strides = var_51175, weight = up_blocks_0_attentions_2_transformer_blocks_2_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_227_cast_fp16")]; + tensor var_51181_begin_0 = const()[name = tensor("op_51181_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_51181_end_0 = const()[name = tensor("op_51181_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_51181_end_mask_0 = const()[name = tensor("op_51181_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51181_cast_fp16 = slice_by_index(begin = var_51181_begin_0, end = var_51181_end_0, end_mask = var_51181_end_mask_0, x = q_227_cast_fp16)[name = tensor("op_51181_cast_fp16")]; + tensor var_51185_begin_0 = const()[name = tensor("op_51185_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_51185_end_0 = const()[name = tensor("op_51185_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_51185_end_mask_0 = const()[name = tensor("op_51185_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51185_cast_fp16 = slice_by_index(begin = var_51185_begin_0, end = var_51185_end_0, end_mask = var_51185_end_mask_0, x = q_227_cast_fp16)[name = tensor("op_51185_cast_fp16")]; + tensor var_51189_begin_0 = const()[name = tensor("op_51189_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_51189_end_0 = const()[name = tensor("op_51189_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_51189_end_mask_0 = const()[name = tensor("op_51189_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51189_cast_fp16 = slice_by_index(begin = var_51189_begin_0, end = var_51189_end_0, end_mask = var_51189_end_mask_0, x = q_227_cast_fp16)[name = tensor("op_51189_cast_fp16")]; + tensor var_51193_begin_0 = const()[name = tensor("op_51193_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_51193_end_0 = const()[name = tensor("op_51193_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_51193_end_mask_0 = const()[name = tensor("op_51193_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51193_cast_fp16 = slice_by_index(begin = var_51193_begin_0, end = var_51193_end_0, end_mask = var_51193_end_mask_0, x = q_227_cast_fp16)[name = tensor("op_51193_cast_fp16")]; + tensor var_51197_begin_0 = const()[name = tensor("op_51197_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_51197_end_0 = const()[name = tensor("op_51197_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_51197_end_mask_0 = const()[name = tensor("op_51197_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51197_cast_fp16 = slice_by_index(begin = var_51197_begin_0, end = var_51197_end_0, end_mask = var_51197_end_mask_0, x = q_227_cast_fp16)[name = tensor("op_51197_cast_fp16")]; + tensor var_51201_begin_0 = const()[name = tensor("op_51201_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_51201_end_0 = const()[name = tensor("op_51201_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_51201_end_mask_0 = const()[name = tensor("op_51201_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51201_cast_fp16 = slice_by_index(begin = var_51201_begin_0, end = var_51201_end_0, end_mask = var_51201_end_mask_0, x = q_227_cast_fp16)[name = tensor("op_51201_cast_fp16")]; + tensor var_51205_begin_0 = const()[name = tensor("op_51205_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_51205_end_0 = const()[name = tensor("op_51205_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_51205_end_mask_0 = const()[name = tensor("op_51205_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51205_cast_fp16 = slice_by_index(begin = var_51205_begin_0, end = var_51205_end_0, end_mask = var_51205_end_mask_0, x = q_227_cast_fp16)[name = tensor("op_51205_cast_fp16")]; + tensor var_51209_begin_0 = const()[name = tensor("op_51209_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_51209_end_0 = const()[name = tensor("op_51209_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_51209_end_mask_0 = const()[name = tensor("op_51209_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51209_cast_fp16 = slice_by_index(begin = var_51209_begin_0, end = var_51209_end_0, end_mask = var_51209_end_mask_0, x = q_227_cast_fp16)[name = tensor("op_51209_cast_fp16")]; + tensor var_51213_begin_0 = const()[name = tensor("op_51213_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_51213_end_0 = const()[name = tensor("op_51213_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_51213_end_mask_0 = const()[name = tensor("op_51213_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51213_cast_fp16 = slice_by_index(begin = var_51213_begin_0, end = var_51213_end_0, end_mask = var_51213_end_mask_0, x = q_227_cast_fp16)[name = tensor("op_51213_cast_fp16")]; + tensor var_51217_begin_0 = const()[name = tensor("op_51217_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_51217_end_0 = const()[name = tensor("op_51217_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_51217_end_mask_0 = const()[name = tensor("op_51217_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51217_cast_fp16 = slice_by_index(begin = var_51217_begin_0, end = var_51217_end_0, end_mask = var_51217_end_mask_0, x = q_227_cast_fp16)[name = tensor("op_51217_cast_fp16")]; + tensor var_51221_begin_0 = const()[name = tensor("op_51221_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_51221_end_0 = const()[name = tensor("op_51221_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_51221_end_mask_0 = const()[name = tensor("op_51221_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51221_cast_fp16 = slice_by_index(begin = var_51221_begin_0, end = var_51221_end_0, end_mask = var_51221_end_mask_0, x = q_227_cast_fp16)[name = tensor("op_51221_cast_fp16")]; + tensor var_51225_begin_0 = const()[name = tensor("op_51225_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_51225_end_0 = const()[name = tensor("op_51225_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_51225_end_mask_0 = const()[name = tensor("op_51225_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51225_cast_fp16 = slice_by_index(begin = var_51225_begin_0, end = var_51225_end_0, end_mask = var_51225_end_mask_0, x = q_227_cast_fp16)[name = tensor("op_51225_cast_fp16")]; + tensor var_51229_begin_0 = const()[name = tensor("op_51229_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_51229_end_0 = const()[name = tensor("op_51229_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_51229_end_mask_0 = const()[name = tensor("op_51229_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51229_cast_fp16 = slice_by_index(begin = var_51229_begin_0, end = var_51229_end_0, end_mask = var_51229_end_mask_0, x = q_227_cast_fp16)[name = tensor("op_51229_cast_fp16")]; + tensor var_51233_begin_0 = const()[name = tensor("op_51233_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_51233_end_0 = const()[name = tensor("op_51233_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_51233_end_mask_0 = const()[name = tensor("op_51233_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51233_cast_fp16 = slice_by_index(begin = var_51233_begin_0, end = var_51233_end_0, end_mask = var_51233_end_mask_0, x = q_227_cast_fp16)[name = tensor("op_51233_cast_fp16")]; + tensor var_51237_begin_0 = const()[name = tensor("op_51237_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_51237_end_0 = const()[name = tensor("op_51237_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_51237_end_mask_0 = const()[name = tensor("op_51237_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51237_cast_fp16 = slice_by_index(begin = var_51237_begin_0, end = var_51237_end_0, end_mask = var_51237_end_mask_0, x = q_227_cast_fp16)[name = tensor("op_51237_cast_fp16")]; + tensor var_51241_begin_0 = const()[name = tensor("op_51241_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_51241_end_0 = const()[name = tensor("op_51241_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_51241_end_mask_0 = const()[name = tensor("op_51241_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51241_cast_fp16 = slice_by_index(begin = var_51241_begin_0, end = var_51241_end_0, end_mask = var_51241_end_mask_0, x = q_227_cast_fp16)[name = tensor("op_51241_cast_fp16")]; + tensor var_51245_begin_0 = const()[name = tensor("op_51245_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_51245_end_0 = const()[name = tensor("op_51245_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_51245_end_mask_0 = const()[name = tensor("op_51245_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51245_cast_fp16 = slice_by_index(begin = var_51245_begin_0, end = var_51245_end_0, end_mask = var_51245_end_mask_0, x = q_227_cast_fp16)[name = tensor("op_51245_cast_fp16")]; + tensor var_51249_begin_0 = const()[name = tensor("op_51249_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_51249_end_0 = const()[name = tensor("op_51249_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_51249_end_mask_0 = const()[name = tensor("op_51249_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51249_cast_fp16 = slice_by_index(begin = var_51249_begin_0, end = var_51249_end_0, end_mask = var_51249_end_mask_0, x = q_227_cast_fp16)[name = tensor("op_51249_cast_fp16")]; + tensor var_51253_begin_0 = const()[name = tensor("op_51253_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_51253_end_0 = const()[name = tensor("op_51253_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_51253_end_mask_0 = const()[name = tensor("op_51253_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51253_cast_fp16 = slice_by_index(begin = var_51253_begin_0, end = var_51253_end_0, end_mask = var_51253_end_mask_0, x = q_227_cast_fp16)[name = tensor("op_51253_cast_fp16")]; + tensor var_51257_begin_0 = const()[name = tensor("op_51257_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_51257_end_0 = const()[name = tensor("op_51257_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_51257_end_mask_0 = const()[name = tensor("op_51257_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51257_cast_fp16 = slice_by_index(begin = var_51257_begin_0, end = var_51257_end_0, end_mask = var_51257_end_mask_0, x = q_227_cast_fp16)[name = tensor("op_51257_cast_fp16")]; + tensor k_455_perm_0 = const()[name = tensor("k_455_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_51264_begin_0 = const()[name = tensor("op_51264_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_51264_end_0 = const()[name = tensor("op_51264_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_51264_end_mask_0 = const()[name = tensor("op_51264_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_455_cast_fp16 = transpose(perm = k_455_perm_0, x = k_453_cast_fp16)[name = tensor("transpose_26")]; + tensor var_51264_cast_fp16 = slice_by_index(begin = var_51264_begin_0, end = var_51264_end_0, end_mask = var_51264_end_mask_0, x = k_455_cast_fp16)[name = tensor("op_51264_cast_fp16")]; + tensor var_51268_begin_0 = const()[name = tensor("op_51268_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_51268_end_0 = const()[name = tensor("op_51268_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_51268_end_mask_0 = const()[name = tensor("op_51268_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51268_cast_fp16 = slice_by_index(begin = var_51268_begin_0, end = var_51268_end_0, end_mask = var_51268_end_mask_0, x = k_455_cast_fp16)[name = tensor("op_51268_cast_fp16")]; + tensor var_51272_begin_0 = const()[name = tensor("op_51272_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_51272_end_0 = const()[name = tensor("op_51272_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_51272_end_mask_0 = const()[name = tensor("op_51272_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51272_cast_fp16 = slice_by_index(begin = var_51272_begin_0, end = var_51272_end_0, end_mask = var_51272_end_mask_0, x = k_455_cast_fp16)[name = tensor("op_51272_cast_fp16")]; + tensor var_51276_begin_0 = const()[name = tensor("op_51276_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_51276_end_0 = const()[name = tensor("op_51276_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_51276_end_mask_0 = const()[name = tensor("op_51276_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51276_cast_fp16 = slice_by_index(begin = var_51276_begin_0, end = var_51276_end_0, end_mask = var_51276_end_mask_0, x = k_455_cast_fp16)[name = tensor("op_51276_cast_fp16")]; + tensor var_51280_begin_0 = const()[name = tensor("op_51280_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_51280_end_0 = const()[name = tensor("op_51280_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_51280_end_mask_0 = const()[name = tensor("op_51280_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51280_cast_fp16 = slice_by_index(begin = var_51280_begin_0, end = var_51280_end_0, end_mask = var_51280_end_mask_0, x = k_455_cast_fp16)[name = tensor("op_51280_cast_fp16")]; + tensor var_51284_begin_0 = const()[name = tensor("op_51284_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_51284_end_0 = const()[name = tensor("op_51284_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_51284_end_mask_0 = const()[name = tensor("op_51284_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51284_cast_fp16 = slice_by_index(begin = var_51284_begin_0, end = var_51284_end_0, end_mask = var_51284_end_mask_0, x = k_455_cast_fp16)[name = tensor("op_51284_cast_fp16")]; + tensor var_51288_begin_0 = const()[name = tensor("op_51288_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_51288_end_0 = const()[name = tensor("op_51288_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_51288_end_mask_0 = const()[name = tensor("op_51288_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51288_cast_fp16 = slice_by_index(begin = var_51288_begin_0, end = var_51288_end_0, end_mask = var_51288_end_mask_0, x = k_455_cast_fp16)[name = tensor("op_51288_cast_fp16")]; + tensor var_51292_begin_0 = const()[name = tensor("op_51292_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_51292_end_0 = const()[name = tensor("op_51292_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_51292_end_mask_0 = const()[name = tensor("op_51292_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51292_cast_fp16 = slice_by_index(begin = var_51292_begin_0, end = var_51292_end_0, end_mask = var_51292_end_mask_0, x = k_455_cast_fp16)[name = tensor("op_51292_cast_fp16")]; + tensor var_51296_begin_0 = const()[name = tensor("op_51296_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_51296_end_0 = const()[name = tensor("op_51296_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_51296_end_mask_0 = const()[name = tensor("op_51296_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51296_cast_fp16 = slice_by_index(begin = var_51296_begin_0, end = var_51296_end_0, end_mask = var_51296_end_mask_0, x = k_455_cast_fp16)[name = tensor("op_51296_cast_fp16")]; + tensor var_51300_begin_0 = const()[name = tensor("op_51300_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_51300_end_0 = const()[name = tensor("op_51300_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_51300_end_mask_0 = const()[name = tensor("op_51300_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51300_cast_fp16 = slice_by_index(begin = var_51300_begin_0, end = var_51300_end_0, end_mask = var_51300_end_mask_0, x = k_455_cast_fp16)[name = tensor("op_51300_cast_fp16")]; + tensor var_51304_begin_0 = const()[name = tensor("op_51304_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_51304_end_0 = const()[name = tensor("op_51304_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_51304_end_mask_0 = const()[name = tensor("op_51304_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51304_cast_fp16 = slice_by_index(begin = var_51304_begin_0, end = var_51304_end_0, end_mask = var_51304_end_mask_0, x = k_455_cast_fp16)[name = tensor("op_51304_cast_fp16")]; + tensor var_51308_begin_0 = const()[name = tensor("op_51308_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_51308_end_0 = const()[name = tensor("op_51308_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_51308_end_mask_0 = const()[name = tensor("op_51308_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51308_cast_fp16 = slice_by_index(begin = var_51308_begin_0, end = var_51308_end_0, end_mask = var_51308_end_mask_0, x = k_455_cast_fp16)[name = tensor("op_51308_cast_fp16")]; + tensor var_51312_begin_0 = const()[name = tensor("op_51312_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_51312_end_0 = const()[name = tensor("op_51312_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_51312_end_mask_0 = const()[name = tensor("op_51312_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51312_cast_fp16 = slice_by_index(begin = var_51312_begin_0, end = var_51312_end_0, end_mask = var_51312_end_mask_0, x = k_455_cast_fp16)[name = tensor("op_51312_cast_fp16")]; + tensor var_51316_begin_0 = const()[name = tensor("op_51316_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_51316_end_0 = const()[name = tensor("op_51316_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_51316_end_mask_0 = const()[name = tensor("op_51316_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51316_cast_fp16 = slice_by_index(begin = var_51316_begin_0, end = var_51316_end_0, end_mask = var_51316_end_mask_0, x = k_455_cast_fp16)[name = tensor("op_51316_cast_fp16")]; + tensor var_51320_begin_0 = const()[name = tensor("op_51320_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_51320_end_0 = const()[name = tensor("op_51320_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_51320_end_mask_0 = const()[name = tensor("op_51320_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51320_cast_fp16 = slice_by_index(begin = var_51320_begin_0, end = var_51320_end_0, end_mask = var_51320_end_mask_0, x = k_455_cast_fp16)[name = tensor("op_51320_cast_fp16")]; + tensor var_51324_begin_0 = const()[name = tensor("op_51324_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_51324_end_0 = const()[name = tensor("op_51324_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_51324_end_mask_0 = const()[name = tensor("op_51324_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51324_cast_fp16 = slice_by_index(begin = var_51324_begin_0, end = var_51324_end_0, end_mask = var_51324_end_mask_0, x = k_455_cast_fp16)[name = tensor("op_51324_cast_fp16")]; + tensor var_51328_begin_0 = const()[name = tensor("op_51328_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_51328_end_0 = const()[name = tensor("op_51328_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_51328_end_mask_0 = const()[name = tensor("op_51328_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51328_cast_fp16 = slice_by_index(begin = var_51328_begin_0, end = var_51328_end_0, end_mask = var_51328_end_mask_0, x = k_455_cast_fp16)[name = tensor("op_51328_cast_fp16")]; + tensor var_51332_begin_0 = const()[name = tensor("op_51332_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_51332_end_0 = const()[name = tensor("op_51332_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_51332_end_mask_0 = const()[name = tensor("op_51332_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51332_cast_fp16 = slice_by_index(begin = var_51332_begin_0, end = var_51332_end_0, end_mask = var_51332_end_mask_0, x = k_455_cast_fp16)[name = tensor("op_51332_cast_fp16")]; + tensor var_51336_begin_0 = const()[name = tensor("op_51336_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_51336_end_0 = const()[name = tensor("op_51336_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_51336_end_mask_0 = const()[name = tensor("op_51336_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51336_cast_fp16 = slice_by_index(begin = var_51336_begin_0, end = var_51336_end_0, end_mask = var_51336_end_mask_0, x = k_455_cast_fp16)[name = tensor("op_51336_cast_fp16")]; + tensor var_51340_begin_0 = const()[name = tensor("op_51340_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_51340_end_0 = const()[name = tensor("op_51340_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_51340_end_mask_0 = const()[name = tensor("op_51340_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51340_cast_fp16 = slice_by_index(begin = var_51340_begin_0, end = var_51340_end_0, end_mask = var_51340_end_mask_0, x = k_455_cast_fp16)[name = tensor("op_51340_cast_fp16")]; + tensor var_51342_begin_0 = const()[name = tensor("op_51342_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_51342_end_0 = const()[name = tensor("op_51342_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_51342_end_mask_0 = const()[name = tensor("op_51342_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51342_cast_fp16 = slice_by_index(begin = var_51342_begin_0, end = var_51342_end_0, end_mask = var_51342_end_mask_0, x = v_227_cast_fp16)[name = tensor("op_51342_cast_fp16")]; + tensor var_51346_begin_0 = const()[name = tensor("op_51346_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_51346_end_0 = const()[name = tensor("op_51346_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_51346_end_mask_0 = const()[name = tensor("op_51346_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51346_cast_fp16 = slice_by_index(begin = var_51346_begin_0, end = var_51346_end_0, end_mask = var_51346_end_mask_0, x = v_227_cast_fp16)[name = tensor("op_51346_cast_fp16")]; + tensor var_51350_begin_0 = const()[name = tensor("op_51350_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_51350_end_0 = const()[name = tensor("op_51350_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_51350_end_mask_0 = const()[name = tensor("op_51350_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51350_cast_fp16 = slice_by_index(begin = var_51350_begin_0, end = var_51350_end_0, end_mask = var_51350_end_mask_0, x = v_227_cast_fp16)[name = tensor("op_51350_cast_fp16")]; + tensor var_51354_begin_0 = const()[name = tensor("op_51354_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_51354_end_0 = const()[name = tensor("op_51354_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_51354_end_mask_0 = const()[name = tensor("op_51354_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51354_cast_fp16 = slice_by_index(begin = var_51354_begin_0, end = var_51354_end_0, end_mask = var_51354_end_mask_0, x = v_227_cast_fp16)[name = tensor("op_51354_cast_fp16")]; + tensor var_51358_begin_0 = const()[name = tensor("op_51358_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_51358_end_0 = const()[name = tensor("op_51358_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_51358_end_mask_0 = const()[name = tensor("op_51358_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51358_cast_fp16 = slice_by_index(begin = var_51358_begin_0, end = var_51358_end_0, end_mask = var_51358_end_mask_0, x = v_227_cast_fp16)[name = tensor("op_51358_cast_fp16")]; + tensor var_51362_begin_0 = const()[name = tensor("op_51362_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_51362_end_0 = const()[name = tensor("op_51362_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_51362_end_mask_0 = const()[name = tensor("op_51362_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51362_cast_fp16 = slice_by_index(begin = var_51362_begin_0, end = var_51362_end_0, end_mask = var_51362_end_mask_0, x = v_227_cast_fp16)[name = tensor("op_51362_cast_fp16")]; + tensor var_51366_begin_0 = const()[name = tensor("op_51366_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_51366_end_0 = const()[name = tensor("op_51366_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_51366_end_mask_0 = const()[name = tensor("op_51366_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51366_cast_fp16 = slice_by_index(begin = var_51366_begin_0, end = var_51366_end_0, end_mask = var_51366_end_mask_0, x = v_227_cast_fp16)[name = tensor("op_51366_cast_fp16")]; + tensor var_51370_begin_0 = const()[name = tensor("op_51370_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_51370_end_0 = const()[name = tensor("op_51370_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_51370_end_mask_0 = const()[name = tensor("op_51370_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51370_cast_fp16 = slice_by_index(begin = var_51370_begin_0, end = var_51370_end_0, end_mask = var_51370_end_mask_0, x = v_227_cast_fp16)[name = tensor("op_51370_cast_fp16")]; + tensor var_51374_begin_0 = const()[name = tensor("op_51374_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_51374_end_0 = const()[name = tensor("op_51374_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_51374_end_mask_0 = const()[name = tensor("op_51374_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51374_cast_fp16 = slice_by_index(begin = var_51374_begin_0, end = var_51374_end_0, end_mask = var_51374_end_mask_0, x = v_227_cast_fp16)[name = tensor("op_51374_cast_fp16")]; + tensor var_51378_begin_0 = const()[name = tensor("op_51378_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_51378_end_0 = const()[name = tensor("op_51378_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_51378_end_mask_0 = const()[name = tensor("op_51378_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51378_cast_fp16 = slice_by_index(begin = var_51378_begin_0, end = var_51378_end_0, end_mask = var_51378_end_mask_0, x = v_227_cast_fp16)[name = tensor("op_51378_cast_fp16")]; + tensor var_51382_begin_0 = const()[name = tensor("op_51382_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_51382_end_0 = const()[name = tensor("op_51382_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_51382_end_mask_0 = const()[name = tensor("op_51382_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51382_cast_fp16 = slice_by_index(begin = var_51382_begin_0, end = var_51382_end_0, end_mask = var_51382_end_mask_0, x = v_227_cast_fp16)[name = tensor("op_51382_cast_fp16")]; + tensor var_51386_begin_0 = const()[name = tensor("op_51386_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_51386_end_0 = const()[name = tensor("op_51386_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_51386_end_mask_0 = const()[name = tensor("op_51386_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51386_cast_fp16 = slice_by_index(begin = var_51386_begin_0, end = var_51386_end_0, end_mask = var_51386_end_mask_0, x = v_227_cast_fp16)[name = tensor("op_51386_cast_fp16")]; + tensor var_51390_begin_0 = const()[name = tensor("op_51390_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_51390_end_0 = const()[name = tensor("op_51390_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_51390_end_mask_0 = const()[name = tensor("op_51390_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51390_cast_fp16 = slice_by_index(begin = var_51390_begin_0, end = var_51390_end_0, end_mask = var_51390_end_mask_0, x = v_227_cast_fp16)[name = tensor("op_51390_cast_fp16")]; + tensor var_51394_begin_0 = const()[name = tensor("op_51394_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_51394_end_0 = const()[name = tensor("op_51394_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_51394_end_mask_0 = const()[name = tensor("op_51394_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51394_cast_fp16 = slice_by_index(begin = var_51394_begin_0, end = var_51394_end_0, end_mask = var_51394_end_mask_0, x = v_227_cast_fp16)[name = tensor("op_51394_cast_fp16")]; + tensor var_51398_begin_0 = const()[name = tensor("op_51398_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_51398_end_0 = const()[name = tensor("op_51398_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_51398_end_mask_0 = const()[name = tensor("op_51398_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51398_cast_fp16 = slice_by_index(begin = var_51398_begin_0, end = var_51398_end_0, end_mask = var_51398_end_mask_0, x = v_227_cast_fp16)[name = tensor("op_51398_cast_fp16")]; + tensor var_51402_begin_0 = const()[name = tensor("op_51402_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_51402_end_0 = const()[name = tensor("op_51402_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_51402_end_mask_0 = const()[name = tensor("op_51402_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51402_cast_fp16 = slice_by_index(begin = var_51402_begin_0, end = var_51402_end_0, end_mask = var_51402_end_mask_0, x = v_227_cast_fp16)[name = tensor("op_51402_cast_fp16")]; + tensor var_51406_begin_0 = const()[name = tensor("op_51406_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_51406_end_0 = const()[name = tensor("op_51406_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_51406_end_mask_0 = const()[name = tensor("op_51406_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51406_cast_fp16 = slice_by_index(begin = var_51406_begin_0, end = var_51406_end_0, end_mask = var_51406_end_mask_0, x = v_227_cast_fp16)[name = tensor("op_51406_cast_fp16")]; + tensor var_51410_begin_0 = const()[name = tensor("op_51410_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_51410_end_0 = const()[name = tensor("op_51410_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_51410_end_mask_0 = const()[name = tensor("op_51410_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51410_cast_fp16 = slice_by_index(begin = var_51410_begin_0, end = var_51410_end_0, end_mask = var_51410_end_mask_0, x = v_227_cast_fp16)[name = tensor("op_51410_cast_fp16")]; + tensor var_51414_begin_0 = const()[name = tensor("op_51414_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_51414_end_0 = const()[name = tensor("op_51414_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_51414_end_mask_0 = const()[name = tensor("op_51414_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51414_cast_fp16 = slice_by_index(begin = var_51414_begin_0, end = var_51414_end_0, end_mask = var_51414_end_mask_0, x = v_227_cast_fp16)[name = tensor("op_51414_cast_fp16")]; + tensor var_51418_begin_0 = const()[name = tensor("op_51418_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_51418_end_0 = const()[name = tensor("op_51418_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_51418_end_mask_0 = const()[name = tensor("op_51418_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51418_cast_fp16 = slice_by_index(begin = var_51418_begin_0, end = var_51418_end_0, end_mask = var_51418_end_mask_0, x = v_227_cast_fp16)[name = tensor("op_51418_cast_fp16")]; + tensor var_51422_equation_0 = const()[name = tensor("op_51422_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51422_cast_fp16 = einsum(equation = var_51422_equation_0, values = (var_51264_cast_fp16, var_51181_cast_fp16))[name = tensor("op_51422_cast_fp16")]; + tensor var_51423_to_fp16 = const()[name = tensor("op_51423_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4361_cast_fp16 = mul(x = var_51422_cast_fp16, y = var_51423_to_fp16)[name = tensor("aw_4361_cast_fp16")]; + tensor var_51426_equation_0 = const()[name = tensor("op_51426_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51426_cast_fp16 = einsum(equation = var_51426_equation_0, values = (var_51268_cast_fp16, var_51185_cast_fp16))[name = tensor("op_51426_cast_fp16")]; + tensor var_51427_to_fp16 = const()[name = tensor("op_51427_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4363_cast_fp16 = mul(x = var_51426_cast_fp16, y = var_51427_to_fp16)[name = tensor("aw_4363_cast_fp16")]; + tensor var_51430_equation_0 = const()[name = tensor("op_51430_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51430_cast_fp16 = einsum(equation = var_51430_equation_0, values = (var_51272_cast_fp16, var_51189_cast_fp16))[name = tensor("op_51430_cast_fp16")]; + tensor var_51431_to_fp16 = const()[name = tensor("op_51431_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4365_cast_fp16 = mul(x = var_51430_cast_fp16, y = var_51431_to_fp16)[name = tensor("aw_4365_cast_fp16")]; + tensor var_51434_equation_0 = const()[name = tensor("op_51434_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51434_cast_fp16 = einsum(equation = var_51434_equation_0, values = (var_51276_cast_fp16, var_51193_cast_fp16))[name = tensor("op_51434_cast_fp16")]; + tensor var_51435_to_fp16 = const()[name = tensor("op_51435_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4367_cast_fp16 = mul(x = var_51434_cast_fp16, y = var_51435_to_fp16)[name = tensor("aw_4367_cast_fp16")]; + tensor var_51438_equation_0 = const()[name = tensor("op_51438_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51438_cast_fp16 = einsum(equation = var_51438_equation_0, values = (var_51280_cast_fp16, var_51197_cast_fp16))[name = tensor("op_51438_cast_fp16")]; + tensor var_51439_to_fp16 = const()[name = tensor("op_51439_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4369_cast_fp16 = mul(x = var_51438_cast_fp16, y = var_51439_to_fp16)[name = tensor("aw_4369_cast_fp16")]; + tensor var_51442_equation_0 = const()[name = tensor("op_51442_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51442_cast_fp16 = einsum(equation = var_51442_equation_0, values = (var_51284_cast_fp16, var_51201_cast_fp16))[name = tensor("op_51442_cast_fp16")]; + tensor var_51443_to_fp16 = const()[name = tensor("op_51443_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4371_cast_fp16 = mul(x = var_51442_cast_fp16, y = var_51443_to_fp16)[name = tensor("aw_4371_cast_fp16")]; + tensor var_51446_equation_0 = const()[name = tensor("op_51446_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51446_cast_fp16 = einsum(equation = var_51446_equation_0, values = (var_51288_cast_fp16, var_51205_cast_fp16))[name = tensor("op_51446_cast_fp16")]; + tensor var_51447_to_fp16 = const()[name = tensor("op_51447_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4373_cast_fp16 = mul(x = var_51446_cast_fp16, y = var_51447_to_fp16)[name = tensor("aw_4373_cast_fp16")]; + tensor var_51450_equation_0 = const()[name = tensor("op_51450_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51450_cast_fp16 = einsum(equation = var_51450_equation_0, values = (var_51292_cast_fp16, var_51209_cast_fp16))[name = tensor("op_51450_cast_fp16")]; + tensor var_51451_to_fp16 = const()[name = tensor("op_51451_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4375_cast_fp16 = mul(x = var_51450_cast_fp16, y = var_51451_to_fp16)[name = tensor("aw_4375_cast_fp16")]; + tensor var_51454_equation_0 = const()[name = tensor("op_51454_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51454_cast_fp16 = einsum(equation = var_51454_equation_0, values = (var_51296_cast_fp16, var_51213_cast_fp16))[name = tensor("op_51454_cast_fp16")]; + tensor var_51455_to_fp16 = const()[name = tensor("op_51455_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4377_cast_fp16 = mul(x = var_51454_cast_fp16, y = var_51455_to_fp16)[name = tensor("aw_4377_cast_fp16")]; + tensor var_51458_equation_0 = const()[name = tensor("op_51458_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51458_cast_fp16 = einsum(equation = var_51458_equation_0, values = (var_51300_cast_fp16, var_51217_cast_fp16))[name = tensor("op_51458_cast_fp16")]; + tensor var_51459_to_fp16 = const()[name = tensor("op_51459_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4379_cast_fp16 = mul(x = var_51458_cast_fp16, y = var_51459_to_fp16)[name = tensor("aw_4379_cast_fp16")]; + tensor var_51462_equation_0 = const()[name = tensor("op_51462_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51462_cast_fp16 = einsum(equation = var_51462_equation_0, values = (var_51304_cast_fp16, var_51221_cast_fp16))[name = tensor("op_51462_cast_fp16")]; + tensor var_51463_to_fp16 = const()[name = tensor("op_51463_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4381_cast_fp16 = mul(x = var_51462_cast_fp16, y = var_51463_to_fp16)[name = tensor("aw_4381_cast_fp16")]; + tensor var_51466_equation_0 = const()[name = tensor("op_51466_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51466_cast_fp16 = einsum(equation = var_51466_equation_0, values = (var_51308_cast_fp16, var_51225_cast_fp16))[name = tensor("op_51466_cast_fp16")]; + tensor var_51467_to_fp16 = const()[name = tensor("op_51467_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4383_cast_fp16 = mul(x = var_51466_cast_fp16, y = var_51467_to_fp16)[name = tensor("aw_4383_cast_fp16")]; + tensor var_51470_equation_0 = const()[name = tensor("op_51470_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51470_cast_fp16 = einsum(equation = var_51470_equation_0, values = (var_51312_cast_fp16, var_51229_cast_fp16))[name = tensor("op_51470_cast_fp16")]; + tensor var_51471_to_fp16 = const()[name = tensor("op_51471_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4385_cast_fp16 = mul(x = var_51470_cast_fp16, y = var_51471_to_fp16)[name = tensor("aw_4385_cast_fp16")]; + tensor var_51474_equation_0 = const()[name = tensor("op_51474_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51474_cast_fp16 = einsum(equation = var_51474_equation_0, values = (var_51316_cast_fp16, var_51233_cast_fp16))[name = tensor("op_51474_cast_fp16")]; + tensor var_51475_to_fp16 = const()[name = tensor("op_51475_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4387_cast_fp16 = mul(x = var_51474_cast_fp16, y = var_51475_to_fp16)[name = tensor("aw_4387_cast_fp16")]; + tensor var_51478_equation_0 = const()[name = tensor("op_51478_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51478_cast_fp16 = einsum(equation = var_51478_equation_0, values = (var_51320_cast_fp16, var_51237_cast_fp16))[name = tensor("op_51478_cast_fp16")]; + tensor var_51479_to_fp16 = const()[name = tensor("op_51479_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4389_cast_fp16 = mul(x = var_51478_cast_fp16, y = var_51479_to_fp16)[name = tensor("aw_4389_cast_fp16")]; + tensor var_51482_equation_0 = const()[name = tensor("op_51482_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51482_cast_fp16 = einsum(equation = var_51482_equation_0, values = (var_51324_cast_fp16, var_51241_cast_fp16))[name = tensor("op_51482_cast_fp16")]; + tensor var_51483_to_fp16 = const()[name = tensor("op_51483_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4391_cast_fp16 = mul(x = var_51482_cast_fp16, y = var_51483_to_fp16)[name = tensor("aw_4391_cast_fp16")]; + tensor var_51486_equation_0 = const()[name = tensor("op_51486_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51486_cast_fp16 = einsum(equation = var_51486_equation_0, values = (var_51328_cast_fp16, var_51245_cast_fp16))[name = tensor("op_51486_cast_fp16")]; + tensor var_51487_to_fp16 = const()[name = tensor("op_51487_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4393_cast_fp16 = mul(x = var_51486_cast_fp16, y = var_51487_to_fp16)[name = tensor("aw_4393_cast_fp16")]; + tensor var_51490_equation_0 = const()[name = tensor("op_51490_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51490_cast_fp16 = einsum(equation = var_51490_equation_0, values = (var_51332_cast_fp16, var_51249_cast_fp16))[name = tensor("op_51490_cast_fp16")]; + tensor var_51491_to_fp16 = const()[name = tensor("op_51491_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4395_cast_fp16 = mul(x = var_51490_cast_fp16, y = var_51491_to_fp16)[name = tensor("aw_4395_cast_fp16")]; + tensor var_51494_equation_0 = const()[name = tensor("op_51494_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51494_cast_fp16 = einsum(equation = var_51494_equation_0, values = (var_51336_cast_fp16, var_51253_cast_fp16))[name = tensor("op_51494_cast_fp16")]; + tensor var_51495_to_fp16 = const()[name = tensor("op_51495_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4397_cast_fp16 = mul(x = var_51494_cast_fp16, y = var_51495_to_fp16)[name = tensor("aw_4397_cast_fp16")]; + tensor var_51498_equation_0 = const()[name = tensor("op_51498_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51498_cast_fp16 = einsum(equation = var_51498_equation_0, values = (var_51340_cast_fp16, var_51257_cast_fp16))[name = tensor("op_51498_cast_fp16")]; + tensor var_51499_to_fp16 = const()[name = tensor("op_51499_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4399_cast_fp16 = mul(x = var_51498_cast_fp16, y = var_51499_to_fp16)[name = tensor("aw_4399_cast_fp16")]; + tensor var_51501_cast_fp16 = softmax(axis = var_30355, x = aw_4361_cast_fp16)[name = tensor("op_51501_cast_fp16")]; + tensor var_51502_cast_fp16 = softmax(axis = var_30355, x = aw_4363_cast_fp16)[name = tensor("op_51502_cast_fp16")]; + tensor var_51503_cast_fp16 = softmax(axis = var_30355, x = aw_4365_cast_fp16)[name = tensor("op_51503_cast_fp16")]; + tensor var_51504_cast_fp16 = softmax(axis = var_30355, x = aw_4367_cast_fp16)[name = tensor("op_51504_cast_fp16")]; + tensor var_51505_cast_fp16 = softmax(axis = var_30355, x = aw_4369_cast_fp16)[name = tensor("op_51505_cast_fp16")]; + tensor var_51506_cast_fp16 = softmax(axis = var_30355, x = aw_4371_cast_fp16)[name = tensor("op_51506_cast_fp16")]; + tensor var_51507_cast_fp16 = softmax(axis = var_30355, x = aw_4373_cast_fp16)[name = tensor("op_51507_cast_fp16")]; + tensor var_51508_cast_fp16 = softmax(axis = var_30355, x = aw_4375_cast_fp16)[name = tensor("op_51508_cast_fp16")]; + tensor var_51509_cast_fp16 = softmax(axis = var_30355, x = aw_4377_cast_fp16)[name = tensor("op_51509_cast_fp16")]; + tensor var_51510_cast_fp16 = softmax(axis = var_30355, x = aw_4379_cast_fp16)[name = tensor("op_51510_cast_fp16")]; + tensor var_51511_cast_fp16 = softmax(axis = var_30355, x = aw_4381_cast_fp16)[name = tensor("op_51511_cast_fp16")]; + tensor var_51512_cast_fp16 = softmax(axis = var_30355, x = aw_4383_cast_fp16)[name = tensor("op_51512_cast_fp16")]; + tensor var_51513_cast_fp16 = softmax(axis = var_30355, x = aw_4385_cast_fp16)[name = tensor("op_51513_cast_fp16")]; + tensor var_51514_cast_fp16 = softmax(axis = var_30355, x = aw_4387_cast_fp16)[name = tensor("op_51514_cast_fp16")]; + tensor var_51515_cast_fp16 = softmax(axis = var_30355, x = aw_4389_cast_fp16)[name = tensor("op_51515_cast_fp16")]; + tensor var_51516_cast_fp16 = softmax(axis = var_30355, x = aw_4391_cast_fp16)[name = tensor("op_51516_cast_fp16")]; + tensor var_51517_cast_fp16 = softmax(axis = var_30355, x = aw_4393_cast_fp16)[name = tensor("op_51517_cast_fp16")]; + tensor var_51518_cast_fp16 = softmax(axis = var_30355, x = aw_4395_cast_fp16)[name = tensor("op_51518_cast_fp16")]; + tensor var_51519_cast_fp16 = softmax(axis = var_30355, x = aw_4397_cast_fp16)[name = tensor("op_51519_cast_fp16")]; + tensor var_51520_cast_fp16 = softmax(axis = var_30355, x = aw_4399_cast_fp16)[name = tensor("op_51520_cast_fp16")]; + tensor var_51522_equation_0 = const()[name = tensor("op_51522_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_51522_cast_fp16 = einsum(equation = var_51522_equation_0, values = (var_51342_cast_fp16, var_51501_cast_fp16))[name = tensor("op_51522_cast_fp16")]; + tensor var_51524_equation_0 = const()[name = tensor("op_51524_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_51524_cast_fp16 = einsum(equation = var_51524_equation_0, values = (var_51346_cast_fp16, var_51502_cast_fp16))[name = tensor("op_51524_cast_fp16")]; + tensor var_51526_equation_0 = const()[name = tensor("op_51526_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_51526_cast_fp16 = einsum(equation = var_51526_equation_0, values = (var_51350_cast_fp16, var_51503_cast_fp16))[name = tensor("op_51526_cast_fp16")]; + tensor var_51528_equation_0 = const()[name = tensor("op_51528_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_51528_cast_fp16 = einsum(equation = var_51528_equation_0, values = (var_51354_cast_fp16, var_51504_cast_fp16))[name = tensor("op_51528_cast_fp16")]; + tensor var_51530_equation_0 = const()[name = tensor("op_51530_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_51530_cast_fp16 = einsum(equation = var_51530_equation_0, values = (var_51358_cast_fp16, var_51505_cast_fp16))[name = tensor("op_51530_cast_fp16")]; + tensor var_51532_equation_0 = const()[name = tensor("op_51532_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_51532_cast_fp16 = einsum(equation = var_51532_equation_0, values = (var_51362_cast_fp16, var_51506_cast_fp16))[name = tensor("op_51532_cast_fp16")]; + tensor var_51534_equation_0 = const()[name = tensor("op_51534_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_51534_cast_fp16 = einsum(equation = var_51534_equation_0, values = (var_51366_cast_fp16, var_51507_cast_fp16))[name = tensor("op_51534_cast_fp16")]; + tensor var_51536_equation_0 = const()[name = tensor("op_51536_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_51536_cast_fp16 = einsum(equation = var_51536_equation_0, values = (var_51370_cast_fp16, var_51508_cast_fp16))[name = tensor("op_51536_cast_fp16")]; + tensor var_51538_equation_0 = const()[name = tensor("op_51538_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_51538_cast_fp16 = einsum(equation = var_51538_equation_0, values = (var_51374_cast_fp16, var_51509_cast_fp16))[name = tensor("op_51538_cast_fp16")]; + tensor var_51540_equation_0 = const()[name = tensor("op_51540_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_51540_cast_fp16 = einsum(equation = var_51540_equation_0, values = (var_51378_cast_fp16, var_51510_cast_fp16))[name = tensor("op_51540_cast_fp16")]; + tensor var_51542_equation_0 = const()[name = tensor("op_51542_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_51542_cast_fp16 = einsum(equation = var_51542_equation_0, values = (var_51382_cast_fp16, var_51511_cast_fp16))[name = tensor("op_51542_cast_fp16")]; + tensor var_51544_equation_0 = const()[name = tensor("op_51544_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_51544_cast_fp16 = einsum(equation = var_51544_equation_0, values = (var_51386_cast_fp16, var_51512_cast_fp16))[name = tensor("op_51544_cast_fp16")]; + tensor var_51546_equation_0 = const()[name = tensor("op_51546_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_51546_cast_fp16 = einsum(equation = var_51546_equation_0, values = (var_51390_cast_fp16, var_51513_cast_fp16))[name = tensor("op_51546_cast_fp16")]; + tensor var_51548_equation_0 = const()[name = tensor("op_51548_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_51548_cast_fp16 = einsum(equation = var_51548_equation_0, values = (var_51394_cast_fp16, var_51514_cast_fp16))[name = tensor("op_51548_cast_fp16")]; + tensor var_51550_equation_0 = const()[name = tensor("op_51550_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_51550_cast_fp16 = einsum(equation = var_51550_equation_0, values = (var_51398_cast_fp16, var_51515_cast_fp16))[name = tensor("op_51550_cast_fp16")]; + tensor var_51552_equation_0 = const()[name = tensor("op_51552_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_51552_cast_fp16 = einsum(equation = var_51552_equation_0, values = (var_51402_cast_fp16, var_51516_cast_fp16))[name = tensor("op_51552_cast_fp16")]; + tensor var_51554_equation_0 = const()[name = tensor("op_51554_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_51554_cast_fp16 = einsum(equation = var_51554_equation_0, values = (var_51406_cast_fp16, var_51517_cast_fp16))[name = tensor("op_51554_cast_fp16")]; + tensor var_51556_equation_0 = const()[name = tensor("op_51556_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_51556_cast_fp16 = einsum(equation = var_51556_equation_0, values = (var_51410_cast_fp16, var_51518_cast_fp16))[name = tensor("op_51556_cast_fp16")]; + tensor var_51558_equation_0 = const()[name = tensor("op_51558_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_51558_cast_fp16 = einsum(equation = var_51558_equation_0, values = (var_51414_cast_fp16, var_51519_cast_fp16))[name = tensor("op_51558_cast_fp16")]; + tensor var_51560_equation_0 = const()[name = tensor("op_51560_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_51560_cast_fp16 = einsum(equation = var_51560_equation_0, values = (var_51418_cast_fp16, var_51520_cast_fp16))[name = tensor("op_51560_cast_fp16")]; + tensor input_653_interleave_0 = const()[name = tensor("input_653_interleave_0"), val = tensor(false)]; + tensor input_653_cast_fp16 = concat(axis = var_30355, interleave = input_653_interleave_0, values = (var_51522_cast_fp16, var_51524_cast_fp16, var_51526_cast_fp16, var_51528_cast_fp16, var_51530_cast_fp16, var_51532_cast_fp16, var_51534_cast_fp16, var_51536_cast_fp16, var_51538_cast_fp16, var_51540_cast_fp16, var_51542_cast_fp16, var_51544_cast_fp16, var_51546_cast_fp16, var_51548_cast_fp16, var_51550_cast_fp16, var_51552_cast_fp16, var_51554_cast_fp16, var_51556_cast_fp16, var_51558_cast_fp16, var_51560_cast_fp16))[name = tensor("input_653_cast_fp16")]; + tensor var_51566 = const()[name = tensor("op_51566"), val = tensor([1, 1])]; + tensor var_51568 = const()[name = tensor("op_51568"), val = tensor([1, 1])]; + tensor var_51570_pad_type_0 = const()[name = tensor("op_51570_pad_type_0"), val = tensor("custom")]; + tensor var_51570_pad_0 = const()[name = tensor("op_51570_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_2_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_2_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4338000256)))]; + tensor up_blocks_0_attentions_2_transformer_blocks_2_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_2_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4341277120)))]; + tensor var_51570_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_2_attn2_to_out_0_bias_to_fp16, dilations = var_51568, groups = var_30355, pad = var_51570_pad_0, pad_type = var_51570_pad_type_0, strides = var_51566, weight = up_blocks_0_attentions_2_transformer_blocks_2_attn2_to_out_0_weight_to_fp16, x = input_653_cast_fp16)[name = tensor("op_51570_cast_fp16")]; + tensor inputs_341_cast_fp16 = add(x = var_51570_cast_fp16, y = inputs_339_cast_fp16)[name = tensor("inputs_341_cast_fp16")]; + tensor input_655_axes_0 = const()[name = tensor("input_655_axes_0"), val = tensor([1])]; + tensor input_655_gamma_0_to_fp16 = const()[name = tensor("input_655_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4341279744)))]; + tensor input_655_beta_0_to_fp16 = const()[name = tensor("input_655_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4341282368)))]; + tensor var_51580_to_fp16 = const()[name = tensor("op_51580_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_655_cast_fp16 = layer_norm(axes = input_655_axes_0, beta = input_655_beta_0_to_fp16, epsilon = var_51580_to_fp16, gamma = input_655_gamma_0_to_fp16, x = inputs_341_cast_fp16)[name = tensor("input_655_cast_fp16")]; + tensor var_51596 = const()[name = tensor("op_51596"), val = tensor([1, 1])]; + tensor var_51598 = const()[name = tensor("op_51598"), val = tensor([1, 1])]; + tensor var_51600_pad_type_0 = const()[name = tensor("op_51600_pad_type_0"), val = tensor("custom")]; + tensor var_51600_pad_0 = const()[name = tensor("op_51600_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_2_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_2_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4341284992)))]; + tensor up_blocks_0_attentions_2_transformer_blocks_2_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_2_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4367499456)))]; + tensor var_51600_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_2_ff_net_0_proj_bias_to_fp16, dilations = var_51598, groups = var_30355, pad = var_51600_pad_0, pad_type = var_51600_pad_type_0, strides = var_51596, weight = up_blocks_0_attentions_2_transformer_blocks_2_ff_net_0_proj_weight_to_fp16, x = input_655_cast_fp16)[name = tensor("op_51600_cast_fp16")]; + tensor var_51601_split_sizes_0 = const()[name = tensor("op_51601_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_51601_axis_0 = const()[name = tensor("op_51601_axis_0"), val = tensor(1)]; + tensor var_51601_cast_fp16_0, tensor var_51601_cast_fp16_1 = split(axis = var_51601_axis_0, split_sizes = var_51601_split_sizes_0, x = var_51600_cast_fp16)[name = tensor("op_51601_cast_fp16")]; + tensor var_51603_mode_0 = const()[name = tensor("op_51603_mode_0"), val = tensor("EXACT")]; + tensor var_51603_cast_fp16 = gelu(mode = var_51603_mode_0, x = var_51601_cast_fp16_1)[name = tensor("op_51603_cast_fp16")]; + tensor input_657_cast_fp16 = mul(x = var_51601_cast_fp16_0, y = var_51603_cast_fp16)[name = tensor("input_657_cast_fp16")]; + tensor var_51607 = const()[name = tensor("op_51607"), val = tensor([1, 1])]; + tensor var_51609 = const()[name = tensor("op_51609"), val = tensor([1, 1])]; + tensor var_51611_pad_type_0 = const()[name = tensor("op_51611_pad_type_0"), val = tensor("custom")]; + tensor var_51611_pad_0 = const()[name = tensor("op_51611_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_2_ff_net_2_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_2_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4367520000)))]; + tensor up_blocks_0_attentions_2_transformer_blocks_2_ff_net_2_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_2_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4380627264)))]; + tensor var_51611_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_2_ff_net_2_bias_to_fp16, dilations = var_51609, groups = var_30355, pad = var_51611_pad_0, pad_type = var_51611_pad_type_0, strides = var_51607, weight = up_blocks_0_attentions_2_transformer_blocks_2_ff_net_2_weight_to_fp16, x = input_657_cast_fp16)[name = tensor("op_51611_cast_fp16")]; + tensor inputs_343_cast_fp16 = add(x = var_51611_cast_fp16, y = inputs_341_cast_fp16)[name = tensor("inputs_343_cast_fp16")]; + tensor hidden_states_453_axes_0 = const()[name = tensor("hidden_states_453_axes_0"), val = tensor([1])]; + tensor hidden_states_453_gamma_0_to_fp16 = const()[name = tensor("hidden_states_453_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4380629888)))]; + tensor hidden_states_453_beta_0_to_fp16 = const()[name = tensor("hidden_states_453_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4380632512)))]; + tensor var_51627_to_fp16 = const()[name = tensor("op_51627_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_453_cast_fp16 = layer_norm(axes = hidden_states_453_axes_0, beta = hidden_states_453_beta_0_to_fp16, epsilon = var_51627_to_fp16, gamma = hidden_states_453_gamma_0_to_fp16, x = inputs_343_cast_fp16)[name = tensor("hidden_states_453_cast_fp16")]; + tensor var_51642 = const()[name = tensor("op_51642"), val = tensor([1, 1])]; + tensor var_51644 = const()[name = tensor("op_51644"), val = tensor([1, 1])]; + tensor q_229_pad_type_0 = const()[name = tensor("q_229_pad_type_0"), val = tensor("custom")]; + tensor q_229_pad_0 = const()[name = tensor("q_229_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_3_attn1_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_3_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4380635136)))]; + tensor q_229_cast_fp16 = conv(dilations = var_51644, groups = var_30355, pad = q_229_pad_0, pad_type = q_229_pad_type_0, strides = var_51642, weight = up_blocks_0_attentions_2_transformer_blocks_3_attn1_to_q_weight_to_fp16, x = hidden_states_453_cast_fp16)[name = tensor("q_229_cast_fp16")]; + tensor var_51648 = const()[name = tensor("op_51648"), val = tensor([1, 1])]; + tensor var_51650 = const()[name = tensor("op_51650"), val = tensor([1, 1])]; + tensor k_457_pad_type_0 = const()[name = tensor("k_457_pad_type_0"), val = tensor("custom")]; + tensor k_457_pad_0 = const()[name = tensor("k_457_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_3_attn1_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_3_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4383912000)))]; + tensor k_457_cast_fp16 = conv(dilations = var_51650, groups = var_30355, pad = k_457_pad_0, pad_type = k_457_pad_type_0, strides = var_51648, weight = up_blocks_0_attentions_2_transformer_blocks_3_attn1_to_k_weight_to_fp16, x = hidden_states_453_cast_fp16)[name = tensor("k_457_cast_fp16")]; + tensor var_51654 = const()[name = tensor("op_51654"), val = tensor([1, 1])]; + tensor var_51656 = const()[name = tensor("op_51656"), val = tensor([1, 1])]; + tensor v_229_pad_type_0 = const()[name = tensor("v_229_pad_type_0"), val = tensor("custom")]; + tensor v_229_pad_0 = const()[name = tensor("v_229_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_3_attn1_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_3_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4387188864)))]; + tensor v_229_cast_fp16 = conv(dilations = var_51656, groups = var_30355, pad = v_229_pad_0, pad_type = v_229_pad_type_0, strides = var_51654, weight = up_blocks_0_attentions_2_transformer_blocks_3_attn1_to_v_weight_to_fp16, x = hidden_states_453_cast_fp16)[name = tensor("v_229_cast_fp16")]; + tensor var_51660_begin_0 = const()[name = tensor("op_51660_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_51660_end_0 = const()[name = tensor("op_51660_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_51660_end_mask_0 = const()[name = tensor("op_51660_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51660_cast_fp16 = slice_by_index(begin = var_51660_begin_0, end = var_51660_end_0, end_mask = var_51660_end_mask_0, x = q_229_cast_fp16)[name = tensor("op_51660_cast_fp16")]; + tensor var_51664_begin_0 = const()[name = tensor("op_51664_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_51664_end_0 = const()[name = tensor("op_51664_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_51664_end_mask_0 = const()[name = tensor("op_51664_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51664_cast_fp16 = slice_by_index(begin = var_51664_begin_0, end = var_51664_end_0, end_mask = var_51664_end_mask_0, x = q_229_cast_fp16)[name = tensor("op_51664_cast_fp16")]; + tensor var_51668_begin_0 = const()[name = tensor("op_51668_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_51668_end_0 = const()[name = tensor("op_51668_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_51668_end_mask_0 = const()[name = tensor("op_51668_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51668_cast_fp16 = slice_by_index(begin = var_51668_begin_0, end = var_51668_end_0, end_mask = var_51668_end_mask_0, x = q_229_cast_fp16)[name = tensor("op_51668_cast_fp16")]; + tensor var_51672_begin_0 = const()[name = tensor("op_51672_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_51672_end_0 = const()[name = tensor("op_51672_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_51672_end_mask_0 = const()[name = tensor("op_51672_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51672_cast_fp16 = slice_by_index(begin = var_51672_begin_0, end = var_51672_end_0, end_mask = var_51672_end_mask_0, x = q_229_cast_fp16)[name = tensor("op_51672_cast_fp16")]; + tensor var_51676_begin_0 = const()[name = tensor("op_51676_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_51676_end_0 = const()[name = tensor("op_51676_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_51676_end_mask_0 = const()[name = tensor("op_51676_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51676_cast_fp16 = slice_by_index(begin = var_51676_begin_0, end = var_51676_end_0, end_mask = var_51676_end_mask_0, x = q_229_cast_fp16)[name = tensor("op_51676_cast_fp16")]; + tensor var_51680_begin_0 = const()[name = tensor("op_51680_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_51680_end_0 = const()[name = tensor("op_51680_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_51680_end_mask_0 = const()[name = tensor("op_51680_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51680_cast_fp16 = slice_by_index(begin = var_51680_begin_0, end = var_51680_end_0, end_mask = var_51680_end_mask_0, x = q_229_cast_fp16)[name = tensor("op_51680_cast_fp16")]; + tensor var_51684_begin_0 = const()[name = tensor("op_51684_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_51684_end_0 = const()[name = tensor("op_51684_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_51684_end_mask_0 = const()[name = tensor("op_51684_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51684_cast_fp16 = slice_by_index(begin = var_51684_begin_0, end = var_51684_end_0, end_mask = var_51684_end_mask_0, x = q_229_cast_fp16)[name = tensor("op_51684_cast_fp16")]; + tensor var_51688_begin_0 = const()[name = tensor("op_51688_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_51688_end_0 = const()[name = tensor("op_51688_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_51688_end_mask_0 = const()[name = tensor("op_51688_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51688_cast_fp16 = slice_by_index(begin = var_51688_begin_0, end = var_51688_end_0, end_mask = var_51688_end_mask_0, x = q_229_cast_fp16)[name = tensor("op_51688_cast_fp16")]; + tensor var_51692_begin_0 = const()[name = tensor("op_51692_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_51692_end_0 = const()[name = tensor("op_51692_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_51692_end_mask_0 = const()[name = tensor("op_51692_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51692_cast_fp16 = slice_by_index(begin = var_51692_begin_0, end = var_51692_end_0, end_mask = var_51692_end_mask_0, x = q_229_cast_fp16)[name = tensor("op_51692_cast_fp16")]; + tensor var_51696_begin_0 = const()[name = tensor("op_51696_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_51696_end_0 = const()[name = tensor("op_51696_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_51696_end_mask_0 = const()[name = tensor("op_51696_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51696_cast_fp16 = slice_by_index(begin = var_51696_begin_0, end = var_51696_end_0, end_mask = var_51696_end_mask_0, x = q_229_cast_fp16)[name = tensor("op_51696_cast_fp16")]; + tensor var_51700_begin_0 = const()[name = tensor("op_51700_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_51700_end_0 = const()[name = tensor("op_51700_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_51700_end_mask_0 = const()[name = tensor("op_51700_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51700_cast_fp16 = slice_by_index(begin = var_51700_begin_0, end = var_51700_end_0, end_mask = var_51700_end_mask_0, x = q_229_cast_fp16)[name = tensor("op_51700_cast_fp16")]; + tensor var_51704_begin_0 = const()[name = tensor("op_51704_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_51704_end_0 = const()[name = tensor("op_51704_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_51704_end_mask_0 = const()[name = tensor("op_51704_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51704_cast_fp16 = slice_by_index(begin = var_51704_begin_0, end = var_51704_end_0, end_mask = var_51704_end_mask_0, x = q_229_cast_fp16)[name = tensor("op_51704_cast_fp16")]; + tensor var_51708_begin_0 = const()[name = tensor("op_51708_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_51708_end_0 = const()[name = tensor("op_51708_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_51708_end_mask_0 = const()[name = tensor("op_51708_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51708_cast_fp16 = slice_by_index(begin = var_51708_begin_0, end = var_51708_end_0, end_mask = var_51708_end_mask_0, x = q_229_cast_fp16)[name = tensor("op_51708_cast_fp16")]; + tensor var_51712_begin_0 = const()[name = tensor("op_51712_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_51712_end_0 = const()[name = tensor("op_51712_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_51712_end_mask_0 = const()[name = tensor("op_51712_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51712_cast_fp16 = slice_by_index(begin = var_51712_begin_0, end = var_51712_end_0, end_mask = var_51712_end_mask_0, x = q_229_cast_fp16)[name = tensor("op_51712_cast_fp16")]; + tensor var_51716_begin_0 = const()[name = tensor("op_51716_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_51716_end_0 = const()[name = tensor("op_51716_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_51716_end_mask_0 = const()[name = tensor("op_51716_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51716_cast_fp16 = slice_by_index(begin = var_51716_begin_0, end = var_51716_end_0, end_mask = var_51716_end_mask_0, x = q_229_cast_fp16)[name = tensor("op_51716_cast_fp16")]; + tensor var_51720_begin_0 = const()[name = tensor("op_51720_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_51720_end_0 = const()[name = tensor("op_51720_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_51720_end_mask_0 = const()[name = tensor("op_51720_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51720_cast_fp16 = slice_by_index(begin = var_51720_begin_0, end = var_51720_end_0, end_mask = var_51720_end_mask_0, x = q_229_cast_fp16)[name = tensor("op_51720_cast_fp16")]; + tensor var_51724_begin_0 = const()[name = tensor("op_51724_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_51724_end_0 = const()[name = tensor("op_51724_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_51724_end_mask_0 = const()[name = tensor("op_51724_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51724_cast_fp16 = slice_by_index(begin = var_51724_begin_0, end = var_51724_end_0, end_mask = var_51724_end_mask_0, x = q_229_cast_fp16)[name = tensor("op_51724_cast_fp16")]; + tensor var_51728_begin_0 = const()[name = tensor("op_51728_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_51728_end_0 = const()[name = tensor("op_51728_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_51728_end_mask_0 = const()[name = tensor("op_51728_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51728_cast_fp16 = slice_by_index(begin = var_51728_begin_0, end = var_51728_end_0, end_mask = var_51728_end_mask_0, x = q_229_cast_fp16)[name = tensor("op_51728_cast_fp16")]; + tensor var_51732_begin_0 = const()[name = tensor("op_51732_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_51732_end_0 = const()[name = tensor("op_51732_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_51732_end_mask_0 = const()[name = tensor("op_51732_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51732_cast_fp16 = slice_by_index(begin = var_51732_begin_0, end = var_51732_end_0, end_mask = var_51732_end_mask_0, x = q_229_cast_fp16)[name = tensor("op_51732_cast_fp16")]; + tensor var_51736_begin_0 = const()[name = tensor("op_51736_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_51736_end_0 = const()[name = tensor("op_51736_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_51736_end_mask_0 = const()[name = tensor("op_51736_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51736_cast_fp16 = slice_by_index(begin = var_51736_begin_0, end = var_51736_end_0, end_mask = var_51736_end_mask_0, x = q_229_cast_fp16)[name = tensor("op_51736_cast_fp16")]; + tensor k_459_perm_0 = const()[name = tensor("k_459_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_51743_begin_0 = const()[name = tensor("op_51743_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_51743_end_0 = const()[name = tensor("op_51743_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_51743_end_mask_0 = const()[name = tensor("op_51743_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_459_cast_fp16 = transpose(perm = k_459_perm_0, x = k_457_cast_fp16)[name = tensor("transpose_25")]; + tensor var_51743_cast_fp16 = slice_by_index(begin = var_51743_begin_0, end = var_51743_end_0, end_mask = var_51743_end_mask_0, x = k_459_cast_fp16)[name = tensor("op_51743_cast_fp16")]; + tensor var_51747_begin_0 = const()[name = tensor("op_51747_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_51747_end_0 = const()[name = tensor("op_51747_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_51747_end_mask_0 = const()[name = tensor("op_51747_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51747_cast_fp16 = slice_by_index(begin = var_51747_begin_0, end = var_51747_end_0, end_mask = var_51747_end_mask_0, x = k_459_cast_fp16)[name = tensor("op_51747_cast_fp16")]; + tensor var_51751_begin_0 = const()[name = tensor("op_51751_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_51751_end_0 = const()[name = tensor("op_51751_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_51751_end_mask_0 = const()[name = tensor("op_51751_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51751_cast_fp16 = slice_by_index(begin = var_51751_begin_0, end = var_51751_end_0, end_mask = var_51751_end_mask_0, x = k_459_cast_fp16)[name = tensor("op_51751_cast_fp16")]; + tensor var_51755_begin_0 = const()[name = tensor("op_51755_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_51755_end_0 = const()[name = tensor("op_51755_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_51755_end_mask_0 = const()[name = tensor("op_51755_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51755_cast_fp16 = slice_by_index(begin = var_51755_begin_0, end = var_51755_end_0, end_mask = var_51755_end_mask_0, x = k_459_cast_fp16)[name = tensor("op_51755_cast_fp16")]; + tensor var_51759_begin_0 = const()[name = tensor("op_51759_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_51759_end_0 = const()[name = tensor("op_51759_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_51759_end_mask_0 = const()[name = tensor("op_51759_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51759_cast_fp16 = slice_by_index(begin = var_51759_begin_0, end = var_51759_end_0, end_mask = var_51759_end_mask_0, x = k_459_cast_fp16)[name = tensor("op_51759_cast_fp16")]; + tensor var_51763_begin_0 = const()[name = tensor("op_51763_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_51763_end_0 = const()[name = tensor("op_51763_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_51763_end_mask_0 = const()[name = tensor("op_51763_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51763_cast_fp16 = slice_by_index(begin = var_51763_begin_0, end = var_51763_end_0, end_mask = var_51763_end_mask_0, x = k_459_cast_fp16)[name = tensor("op_51763_cast_fp16")]; + tensor var_51767_begin_0 = const()[name = tensor("op_51767_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_51767_end_0 = const()[name = tensor("op_51767_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_51767_end_mask_0 = const()[name = tensor("op_51767_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51767_cast_fp16 = slice_by_index(begin = var_51767_begin_0, end = var_51767_end_0, end_mask = var_51767_end_mask_0, x = k_459_cast_fp16)[name = tensor("op_51767_cast_fp16")]; + tensor var_51771_begin_0 = const()[name = tensor("op_51771_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_51771_end_0 = const()[name = tensor("op_51771_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_51771_end_mask_0 = const()[name = tensor("op_51771_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51771_cast_fp16 = slice_by_index(begin = var_51771_begin_0, end = var_51771_end_0, end_mask = var_51771_end_mask_0, x = k_459_cast_fp16)[name = tensor("op_51771_cast_fp16")]; + tensor var_51775_begin_0 = const()[name = tensor("op_51775_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_51775_end_0 = const()[name = tensor("op_51775_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_51775_end_mask_0 = const()[name = tensor("op_51775_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51775_cast_fp16 = slice_by_index(begin = var_51775_begin_0, end = var_51775_end_0, end_mask = var_51775_end_mask_0, x = k_459_cast_fp16)[name = tensor("op_51775_cast_fp16")]; + tensor var_51779_begin_0 = const()[name = tensor("op_51779_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_51779_end_0 = const()[name = tensor("op_51779_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_51779_end_mask_0 = const()[name = tensor("op_51779_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51779_cast_fp16 = slice_by_index(begin = var_51779_begin_0, end = var_51779_end_0, end_mask = var_51779_end_mask_0, x = k_459_cast_fp16)[name = tensor("op_51779_cast_fp16")]; + tensor var_51783_begin_0 = const()[name = tensor("op_51783_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_51783_end_0 = const()[name = tensor("op_51783_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_51783_end_mask_0 = const()[name = tensor("op_51783_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51783_cast_fp16 = slice_by_index(begin = var_51783_begin_0, end = var_51783_end_0, end_mask = var_51783_end_mask_0, x = k_459_cast_fp16)[name = tensor("op_51783_cast_fp16")]; + tensor var_51787_begin_0 = const()[name = tensor("op_51787_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_51787_end_0 = const()[name = tensor("op_51787_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_51787_end_mask_0 = const()[name = tensor("op_51787_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51787_cast_fp16 = slice_by_index(begin = var_51787_begin_0, end = var_51787_end_0, end_mask = var_51787_end_mask_0, x = k_459_cast_fp16)[name = tensor("op_51787_cast_fp16")]; + tensor var_51791_begin_0 = const()[name = tensor("op_51791_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_51791_end_0 = const()[name = tensor("op_51791_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_51791_end_mask_0 = const()[name = tensor("op_51791_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51791_cast_fp16 = slice_by_index(begin = var_51791_begin_0, end = var_51791_end_0, end_mask = var_51791_end_mask_0, x = k_459_cast_fp16)[name = tensor("op_51791_cast_fp16")]; + tensor var_51795_begin_0 = const()[name = tensor("op_51795_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_51795_end_0 = const()[name = tensor("op_51795_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_51795_end_mask_0 = const()[name = tensor("op_51795_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51795_cast_fp16 = slice_by_index(begin = var_51795_begin_0, end = var_51795_end_0, end_mask = var_51795_end_mask_0, x = k_459_cast_fp16)[name = tensor("op_51795_cast_fp16")]; + tensor var_51799_begin_0 = const()[name = tensor("op_51799_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_51799_end_0 = const()[name = tensor("op_51799_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_51799_end_mask_0 = const()[name = tensor("op_51799_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51799_cast_fp16 = slice_by_index(begin = var_51799_begin_0, end = var_51799_end_0, end_mask = var_51799_end_mask_0, x = k_459_cast_fp16)[name = tensor("op_51799_cast_fp16")]; + tensor var_51803_begin_0 = const()[name = tensor("op_51803_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_51803_end_0 = const()[name = tensor("op_51803_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_51803_end_mask_0 = const()[name = tensor("op_51803_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51803_cast_fp16 = slice_by_index(begin = var_51803_begin_0, end = var_51803_end_0, end_mask = var_51803_end_mask_0, x = k_459_cast_fp16)[name = tensor("op_51803_cast_fp16")]; + tensor var_51807_begin_0 = const()[name = tensor("op_51807_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_51807_end_0 = const()[name = tensor("op_51807_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_51807_end_mask_0 = const()[name = tensor("op_51807_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51807_cast_fp16 = slice_by_index(begin = var_51807_begin_0, end = var_51807_end_0, end_mask = var_51807_end_mask_0, x = k_459_cast_fp16)[name = tensor("op_51807_cast_fp16")]; + tensor var_51811_begin_0 = const()[name = tensor("op_51811_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_51811_end_0 = const()[name = tensor("op_51811_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_51811_end_mask_0 = const()[name = tensor("op_51811_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51811_cast_fp16 = slice_by_index(begin = var_51811_begin_0, end = var_51811_end_0, end_mask = var_51811_end_mask_0, x = k_459_cast_fp16)[name = tensor("op_51811_cast_fp16")]; + tensor var_51815_begin_0 = const()[name = tensor("op_51815_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_51815_end_0 = const()[name = tensor("op_51815_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_51815_end_mask_0 = const()[name = tensor("op_51815_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51815_cast_fp16 = slice_by_index(begin = var_51815_begin_0, end = var_51815_end_0, end_mask = var_51815_end_mask_0, x = k_459_cast_fp16)[name = tensor("op_51815_cast_fp16")]; + tensor var_51819_begin_0 = const()[name = tensor("op_51819_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_51819_end_0 = const()[name = tensor("op_51819_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_51819_end_mask_0 = const()[name = tensor("op_51819_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51819_cast_fp16 = slice_by_index(begin = var_51819_begin_0, end = var_51819_end_0, end_mask = var_51819_end_mask_0, x = k_459_cast_fp16)[name = tensor("op_51819_cast_fp16")]; + tensor var_51821_begin_0 = const()[name = tensor("op_51821_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_51821_end_0 = const()[name = tensor("op_51821_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_51821_end_mask_0 = const()[name = tensor("op_51821_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51821_cast_fp16 = slice_by_index(begin = var_51821_begin_0, end = var_51821_end_0, end_mask = var_51821_end_mask_0, x = v_229_cast_fp16)[name = tensor("op_51821_cast_fp16")]; + tensor var_51825_begin_0 = const()[name = tensor("op_51825_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_51825_end_0 = const()[name = tensor("op_51825_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_51825_end_mask_0 = const()[name = tensor("op_51825_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51825_cast_fp16 = slice_by_index(begin = var_51825_begin_0, end = var_51825_end_0, end_mask = var_51825_end_mask_0, x = v_229_cast_fp16)[name = tensor("op_51825_cast_fp16")]; + tensor var_51829_begin_0 = const()[name = tensor("op_51829_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_51829_end_0 = const()[name = tensor("op_51829_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_51829_end_mask_0 = const()[name = tensor("op_51829_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51829_cast_fp16 = slice_by_index(begin = var_51829_begin_0, end = var_51829_end_0, end_mask = var_51829_end_mask_0, x = v_229_cast_fp16)[name = tensor("op_51829_cast_fp16")]; + tensor var_51833_begin_0 = const()[name = tensor("op_51833_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_51833_end_0 = const()[name = tensor("op_51833_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_51833_end_mask_0 = const()[name = tensor("op_51833_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51833_cast_fp16 = slice_by_index(begin = var_51833_begin_0, end = var_51833_end_0, end_mask = var_51833_end_mask_0, x = v_229_cast_fp16)[name = tensor("op_51833_cast_fp16")]; + tensor var_51837_begin_0 = const()[name = tensor("op_51837_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_51837_end_0 = const()[name = tensor("op_51837_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_51837_end_mask_0 = const()[name = tensor("op_51837_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51837_cast_fp16 = slice_by_index(begin = var_51837_begin_0, end = var_51837_end_0, end_mask = var_51837_end_mask_0, x = v_229_cast_fp16)[name = tensor("op_51837_cast_fp16")]; + tensor var_51841_begin_0 = const()[name = tensor("op_51841_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_51841_end_0 = const()[name = tensor("op_51841_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_51841_end_mask_0 = const()[name = tensor("op_51841_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51841_cast_fp16 = slice_by_index(begin = var_51841_begin_0, end = var_51841_end_0, end_mask = var_51841_end_mask_0, x = v_229_cast_fp16)[name = tensor("op_51841_cast_fp16")]; + tensor var_51845_begin_0 = const()[name = tensor("op_51845_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_51845_end_0 = const()[name = tensor("op_51845_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_51845_end_mask_0 = const()[name = tensor("op_51845_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51845_cast_fp16 = slice_by_index(begin = var_51845_begin_0, end = var_51845_end_0, end_mask = var_51845_end_mask_0, x = v_229_cast_fp16)[name = tensor("op_51845_cast_fp16")]; + tensor var_51849_begin_0 = const()[name = tensor("op_51849_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_51849_end_0 = const()[name = tensor("op_51849_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_51849_end_mask_0 = const()[name = tensor("op_51849_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51849_cast_fp16 = slice_by_index(begin = var_51849_begin_0, end = var_51849_end_0, end_mask = var_51849_end_mask_0, x = v_229_cast_fp16)[name = tensor("op_51849_cast_fp16")]; + tensor var_51853_begin_0 = const()[name = tensor("op_51853_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_51853_end_0 = const()[name = tensor("op_51853_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_51853_end_mask_0 = const()[name = tensor("op_51853_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51853_cast_fp16 = slice_by_index(begin = var_51853_begin_0, end = var_51853_end_0, end_mask = var_51853_end_mask_0, x = v_229_cast_fp16)[name = tensor("op_51853_cast_fp16")]; + tensor var_51857_begin_0 = const()[name = tensor("op_51857_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_51857_end_0 = const()[name = tensor("op_51857_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_51857_end_mask_0 = const()[name = tensor("op_51857_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51857_cast_fp16 = slice_by_index(begin = var_51857_begin_0, end = var_51857_end_0, end_mask = var_51857_end_mask_0, x = v_229_cast_fp16)[name = tensor("op_51857_cast_fp16")]; + tensor var_51861_begin_0 = const()[name = tensor("op_51861_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_51861_end_0 = const()[name = tensor("op_51861_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_51861_end_mask_0 = const()[name = tensor("op_51861_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51861_cast_fp16 = slice_by_index(begin = var_51861_begin_0, end = var_51861_end_0, end_mask = var_51861_end_mask_0, x = v_229_cast_fp16)[name = tensor("op_51861_cast_fp16")]; + tensor var_51865_begin_0 = const()[name = tensor("op_51865_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_51865_end_0 = const()[name = tensor("op_51865_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_51865_end_mask_0 = const()[name = tensor("op_51865_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51865_cast_fp16 = slice_by_index(begin = var_51865_begin_0, end = var_51865_end_0, end_mask = var_51865_end_mask_0, x = v_229_cast_fp16)[name = tensor("op_51865_cast_fp16")]; + tensor var_51869_begin_0 = const()[name = tensor("op_51869_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_51869_end_0 = const()[name = tensor("op_51869_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_51869_end_mask_0 = const()[name = tensor("op_51869_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51869_cast_fp16 = slice_by_index(begin = var_51869_begin_0, end = var_51869_end_0, end_mask = var_51869_end_mask_0, x = v_229_cast_fp16)[name = tensor("op_51869_cast_fp16")]; + tensor var_51873_begin_0 = const()[name = tensor("op_51873_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_51873_end_0 = const()[name = tensor("op_51873_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_51873_end_mask_0 = const()[name = tensor("op_51873_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51873_cast_fp16 = slice_by_index(begin = var_51873_begin_0, end = var_51873_end_0, end_mask = var_51873_end_mask_0, x = v_229_cast_fp16)[name = tensor("op_51873_cast_fp16")]; + tensor var_51877_begin_0 = const()[name = tensor("op_51877_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_51877_end_0 = const()[name = tensor("op_51877_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_51877_end_mask_0 = const()[name = tensor("op_51877_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51877_cast_fp16 = slice_by_index(begin = var_51877_begin_0, end = var_51877_end_0, end_mask = var_51877_end_mask_0, x = v_229_cast_fp16)[name = tensor("op_51877_cast_fp16")]; + tensor var_51881_begin_0 = const()[name = tensor("op_51881_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_51881_end_0 = const()[name = tensor("op_51881_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_51881_end_mask_0 = const()[name = tensor("op_51881_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51881_cast_fp16 = slice_by_index(begin = var_51881_begin_0, end = var_51881_end_0, end_mask = var_51881_end_mask_0, x = v_229_cast_fp16)[name = tensor("op_51881_cast_fp16")]; + tensor var_51885_begin_0 = const()[name = tensor("op_51885_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_51885_end_0 = const()[name = tensor("op_51885_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_51885_end_mask_0 = const()[name = tensor("op_51885_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51885_cast_fp16 = slice_by_index(begin = var_51885_begin_0, end = var_51885_end_0, end_mask = var_51885_end_mask_0, x = v_229_cast_fp16)[name = tensor("op_51885_cast_fp16")]; + tensor var_51889_begin_0 = const()[name = tensor("op_51889_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_51889_end_0 = const()[name = tensor("op_51889_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_51889_end_mask_0 = const()[name = tensor("op_51889_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51889_cast_fp16 = slice_by_index(begin = var_51889_begin_0, end = var_51889_end_0, end_mask = var_51889_end_mask_0, x = v_229_cast_fp16)[name = tensor("op_51889_cast_fp16")]; + tensor var_51893_begin_0 = const()[name = tensor("op_51893_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_51893_end_0 = const()[name = tensor("op_51893_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_51893_end_mask_0 = const()[name = tensor("op_51893_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51893_cast_fp16 = slice_by_index(begin = var_51893_begin_0, end = var_51893_end_0, end_mask = var_51893_end_mask_0, x = v_229_cast_fp16)[name = tensor("op_51893_cast_fp16")]; + tensor var_51897_begin_0 = const()[name = tensor("op_51897_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_51897_end_0 = const()[name = tensor("op_51897_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_51897_end_mask_0 = const()[name = tensor("op_51897_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51897_cast_fp16 = slice_by_index(begin = var_51897_begin_0, end = var_51897_end_0, end_mask = var_51897_end_mask_0, x = v_229_cast_fp16)[name = tensor("op_51897_cast_fp16")]; + tensor var_51901_equation_0 = const()[name = tensor("op_51901_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51901_cast_fp16 = einsum(equation = var_51901_equation_0, values = (var_51743_cast_fp16, var_51660_cast_fp16))[name = tensor("op_51901_cast_fp16")]; + tensor var_51902_to_fp16 = const()[name = tensor("op_51902_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4401_cast_fp16 = mul(x = var_51901_cast_fp16, y = var_51902_to_fp16)[name = tensor("aw_4401_cast_fp16")]; + tensor var_51905_equation_0 = const()[name = tensor("op_51905_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51905_cast_fp16 = einsum(equation = var_51905_equation_0, values = (var_51747_cast_fp16, var_51664_cast_fp16))[name = tensor("op_51905_cast_fp16")]; + tensor var_51906_to_fp16 = const()[name = tensor("op_51906_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4403_cast_fp16 = mul(x = var_51905_cast_fp16, y = var_51906_to_fp16)[name = tensor("aw_4403_cast_fp16")]; + tensor var_51909_equation_0 = const()[name = tensor("op_51909_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51909_cast_fp16 = einsum(equation = var_51909_equation_0, values = (var_51751_cast_fp16, var_51668_cast_fp16))[name = tensor("op_51909_cast_fp16")]; + tensor var_51910_to_fp16 = const()[name = tensor("op_51910_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4405_cast_fp16 = mul(x = var_51909_cast_fp16, y = var_51910_to_fp16)[name = tensor("aw_4405_cast_fp16")]; + tensor var_51913_equation_0 = const()[name = tensor("op_51913_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51913_cast_fp16 = einsum(equation = var_51913_equation_0, values = (var_51755_cast_fp16, var_51672_cast_fp16))[name = tensor("op_51913_cast_fp16")]; + tensor var_51914_to_fp16 = const()[name = tensor("op_51914_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4407_cast_fp16 = mul(x = var_51913_cast_fp16, y = var_51914_to_fp16)[name = tensor("aw_4407_cast_fp16")]; + tensor var_51917_equation_0 = const()[name = tensor("op_51917_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51917_cast_fp16 = einsum(equation = var_51917_equation_0, values = (var_51759_cast_fp16, var_51676_cast_fp16))[name = tensor("op_51917_cast_fp16")]; + tensor var_51918_to_fp16 = const()[name = tensor("op_51918_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4409_cast_fp16 = mul(x = var_51917_cast_fp16, y = var_51918_to_fp16)[name = tensor("aw_4409_cast_fp16")]; + tensor var_51921_equation_0 = const()[name = tensor("op_51921_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51921_cast_fp16 = einsum(equation = var_51921_equation_0, values = (var_51763_cast_fp16, var_51680_cast_fp16))[name = tensor("op_51921_cast_fp16")]; + tensor var_51922_to_fp16 = const()[name = tensor("op_51922_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4411_cast_fp16 = mul(x = var_51921_cast_fp16, y = var_51922_to_fp16)[name = tensor("aw_4411_cast_fp16")]; + tensor var_51925_equation_0 = const()[name = tensor("op_51925_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51925_cast_fp16 = einsum(equation = var_51925_equation_0, values = (var_51767_cast_fp16, var_51684_cast_fp16))[name = tensor("op_51925_cast_fp16")]; + tensor var_51926_to_fp16 = const()[name = tensor("op_51926_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4413_cast_fp16 = mul(x = var_51925_cast_fp16, y = var_51926_to_fp16)[name = tensor("aw_4413_cast_fp16")]; + tensor var_51929_equation_0 = const()[name = tensor("op_51929_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51929_cast_fp16 = einsum(equation = var_51929_equation_0, values = (var_51771_cast_fp16, var_51688_cast_fp16))[name = tensor("op_51929_cast_fp16")]; + tensor var_51930_to_fp16 = const()[name = tensor("op_51930_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4415_cast_fp16 = mul(x = var_51929_cast_fp16, y = var_51930_to_fp16)[name = tensor("aw_4415_cast_fp16")]; + tensor var_51933_equation_0 = const()[name = tensor("op_51933_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51933_cast_fp16 = einsum(equation = var_51933_equation_0, values = (var_51775_cast_fp16, var_51692_cast_fp16))[name = tensor("op_51933_cast_fp16")]; + tensor var_51934_to_fp16 = const()[name = tensor("op_51934_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4417_cast_fp16 = mul(x = var_51933_cast_fp16, y = var_51934_to_fp16)[name = tensor("aw_4417_cast_fp16")]; + tensor var_51937_equation_0 = const()[name = tensor("op_51937_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51937_cast_fp16 = einsum(equation = var_51937_equation_0, values = (var_51779_cast_fp16, var_51696_cast_fp16))[name = tensor("op_51937_cast_fp16")]; + tensor var_51938_to_fp16 = const()[name = tensor("op_51938_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4419_cast_fp16 = mul(x = var_51937_cast_fp16, y = var_51938_to_fp16)[name = tensor("aw_4419_cast_fp16")]; + tensor var_51941_equation_0 = const()[name = tensor("op_51941_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51941_cast_fp16 = einsum(equation = var_51941_equation_0, values = (var_51783_cast_fp16, var_51700_cast_fp16))[name = tensor("op_51941_cast_fp16")]; + tensor var_51942_to_fp16 = const()[name = tensor("op_51942_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4421_cast_fp16 = mul(x = var_51941_cast_fp16, y = var_51942_to_fp16)[name = tensor("aw_4421_cast_fp16")]; + tensor var_51945_equation_0 = const()[name = tensor("op_51945_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51945_cast_fp16 = einsum(equation = var_51945_equation_0, values = (var_51787_cast_fp16, var_51704_cast_fp16))[name = tensor("op_51945_cast_fp16")]; + tensor var_51946_to_fp16 = const()[name = tensor("op_51946_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4423_cast_fp16 = mul(x = var_51945_cast_fp16, y = var_51946_to_fp16)[name = tensor("aw_4423_cast_fp16")]; + tensor var_51949_equation_0 = const()[name = tensor("op_51949_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51949_cast_fp16 = einsum(equation = var_51949_equation_0, values = (var_51791_cast_fp16, var_51708_cast_fp16))[name = tensor("op_51949_cast_fp16")]; + tensor var_51950_to_fp16 = const()[name = tensor("op_51950_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4425_cast_fp16 = mul(x = var_51949_cast_fp16, y = var_51950_to_fp16)[name = tensor("aw_4425_cast_fp16")]; + tensor var_51953_equation_0 = const()[name = tensor("op_51953_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51953_cast_fp16 = einsum(equation = var_51953_equation_0, values = (var_51795_cast_fp16, var_51712_cast_fp16))[name = tensor("op_51953_cast_fp16")]; + tensor var_51954_to_fp16 = const()[name = tensor("op_51954_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4427_cast_fp16 = mul(x = var_51953_cast_fp16, y = var_51954_to_fp16)[name = tensor("aw_4427_cast_fp16")]; + tensor var_51957_equation_0 = const()[name = tensor("op_51957_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51957_cast_fp16 = einsum(equation = var_51957_equation_0, values = (var_51799_cast_fp16, var_51716_cast_fp16))[name = tensor("op_51957_cast_fp16")]; + tensor var_51958_to_fp16 = const()[name = tensor("op_51958_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4429_cast_fp16 = mul(x = var_51957_cast_fp16, y = var_51958_to_fp16)[name = tensor("aw_4429_cast_fp16")]; + tensor var_51961_equation_0 = const()[name = tensor("op_51961_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51961_cast_fp16 = einsum(equation = var_51961_equation_0, values = (var_51803_cast_fp16, var_51720_cast_fp16))[name = tensor("op_51961_cast_fp16")]; + tensor var_51962_to_fp16 = const()[name = tensor("op_51962_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4431_cast_fp16 = mul(x = var_51961_cast_fp16, y = var_51962_to_fp16)[name = tensor("aw_4431_cast_fp16")]; + tensor var_51965_equation_0 = const()[name = tensor("op_51965_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51965_cast_fp16 = einsum(equation = var_51965_equation_0, values = (var_51807_cast_fp16, var_51724_cast_fp16))[name = tensor("op_51965_cast_fp16")]; + tensor var_51966_to_fp16 = const()[name = tensor("op_51966_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4433_cast_fp16 = mul(x = var_51965_cast_fp16, y = var_51966_to_fp16)[name = tensor("aw_4433_cast_fp16")]; + tensor var_51969_equation_0 = const()[name = tensor("op_51969_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51969_cast_fp16 = einsum(equation = var_51969_equation_0, values = (var_51811_cast_fp16, var_51728_cast_fp16))[name = tensor("op_51969_cast_fp16")]; + tensor var_51970_to_fp16 = const()[name = tensor("op_51970_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4435_cast_fp16 = mul(x = var_51969_cast_fp16, y = var_51970_to_fp16)[name = tensor("aw_4435_cast_fp16")]; + tensor var_51973_equation_0 = const()[name = tensor("op_51973_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51973_cast_fp16 = einsum(equation = var_51973_equation_0, values = (var_51815_cast_fp16, var_51732_cast_fp16))[name = tensor("op_51973_cast_fp16")]; + tensor var_51974_to_fp16 = const()[name = tensor("op_51974_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4437_cast_fp16 = mul(x = var_51973_cast_fp16, y = var_51974_to_fp16)[name = tensor("aw_4437_cast_fp16")]; + tensor var_51977_equation_0 = const()[name = tensor("op_51977_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51977_cast_fp16 = einsum(equation = var_51977_equation_0, values = (var_51819_cast_fp16, var_51736_cast_fp16))[name = tensor("op_51977_cast_fp16")]; + tensor var_51978_to_fp16 = const()[name = tensor("op_51978_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4439_cast_fp16 = mul(x = var_51977_cast_fp16, y = var_51978_to_fp16)[name = tensor("aw_4439_cast_fp16")]; + tensor var_51980_cast_fp16 = softmax(axis = var_30355, x = aw_4401_cast_fp16)[name = tensor("op_51980_cast_fp16")]; + tensor var_51981_cast_fp16 = softmax(axis = var_30355, x = aw_4403_cast_fp16)[name = tensor("op_51981_cast_fp16")]; + tensor var_51982_cast_fp16 = softmax(axis = var_30355, x = aw_4405_cast_fp16)[name = tensor("op_51982_cast_fp16")]; + tensor var_51983_cast_fp16 = softmax(axis = var_30355, x = aw_4407_cast_fp16)[name = tensor("op_51983_cast_fp16")]; + tensor var_51984_cast_fp16 = softmax(axis = var_30355, x = aw_4409_cast_fp16)[name = tensor("op_51984_cast_fp16")]; + tensor var_51985_cast_fp16 = softmax(axis = var_30355, x = aw_4411_cast_fp16)[name = tensor("op_51985_cast_fp16")]; + tensor var_51986_cast_fp16 = softmax(axis = var_30355, x = aw_4413_cast_fp16)[name = tensor("op_51986_cast_fp16")]; + tensor var_51987_cast_fp16 = softmax(axis = var_30355, x = aw_4415_cast_fp16)[name = tensor("op_51987_cast_fp16")]; + tensor var_51988_cast_fp16 = softmax(axis = var_30355, x = aw_4417_cast_fp16)[name = tensor("op_51988_cast_fp16")]; + tensor var_51989_cast_fp16 = softmax(axis = var_30355, x = aw_4419_cast_fp16)[name = tensor("op_51989_cast_fp16")]; + tensor var_51990_cast_fp16 = softmax(axis = var_30355, x = aw_4421_cast_fp16)[name = tensor("op_51990_cast_fp16")]; + tensor var_51991_cast_fp16 = softmax(axis = var_30355, x = aw_4423_cast_fp16)[name = tensor("op_51991_cast_fp16")]; + tensor var_51992_cast_fp16 = softmax(axis = var_30355, x = aw_4425_cast_fp16)[name = tensor("op_51992_cast_fp16")]; + tensor var_51993_cast_fp16 = softmax(axis = var_30355, x = aw_4427_cast_fp16)[name = tensor("op_51993_cast_fp16")]; + tensor var_51994_cast_fp16 = softmax(axis = var_30355, x = aw_4429_cast_fp16)[name = tensor("op_51994_cast_fp16")]; + tensor var_51995_cast_fp16 = softmax(axis = var_30355, x = aw_4431_cast_fp16)[name = tensor("op_51995_cast_fp16")]; + tensor var_51996_cast_fp16 = softmax(axis = var_30355, x = aw_4433_cast_fp16)[name = tensor("op_51996_cast_fp16")]; + tensor var_51997_cast_fp16 = softmax(axis = var_30355, x = aw_4435_cast_fp16)[name = tensor("op_51997_cast_fp16")]; + tensor var_51998_cast_fp16 = softmax(axis = var_30355, x = aw_4437_cast_fp16)[name = tensor("op_51998_cast_fp16")]; + tensor var_51999_cast_fp16 = softmax(axis = var_30355, x = aw_4439_cast_fp16)[name = tensor("op_51999_cast_fp16")]; + tensor var_52001_equation_0 = const()[name = tensor("op_52001_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52001_cast_fp16 = einsum(equation = var_52001_equation_0, values = (var_51821_cast_fp16, var_51980_cast_fp16))[name = tensor("op_52001_cast_fp16")]; + tensor var_52003_equation_0 = const()[name = tensor("op_52003_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52003_cast_fp16 = einsum(equation = var_52003_equation_0, values = (var_51825_cast_fp16, var_51981_cast_fp16))[name = tensor("op_52003_cast_fp16")]; + tensor var_52005_equation_0 = const()[name = tensor("op_52005_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52005_cast_fp16 = einsum(equation = var_52005_equation_0, values = (var_51829_cast_fp16, var_51982_cast_fp16))[name = tensor("op_52005_cast_fp16")]; + tensor var_52007_equation_0 = const()[name = tensor("op_52007_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52007_cast_fp16 = einsum(equation = var_52007_equation_0, values = (var_51833_cast_fp16, var_51983_cast_fp16))[name = tensor("op_52007_cast_fp16")]; + tensor var_52009_equation_0 = const()[name = tensor("op_52009_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52009_cast_fp16 = einsum(equation = var_52009_equation_0, values = (var_51837_cast_fp16, var_51984_cast_fp16))[name = tensor("op_52009_cast_fp16")]; + tensor var_52011_equation_0 = const()[name = tensor("op_52011_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52011_cast_fp16 = einsum(equation = var_52011_equation_0, values = (var_51841_cast_fp16, var_51985_cast_fp16))[name = tensor("op_52011_cast_fp16")]; + tensor var_52013_equation_0 = const()[name = tensor("op_52013_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52013_cast_fp16 = einsum(equation = var_52013_equation_0, values = (var_51845_cast_fp16, var_51986_cast_fp16))[name = tensor("op_52013_cast_fp16")]; + tensor var_52015_equation_0 = const()[name = tensor("op_52015_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52015_cast_fp16 = einsum(equation = var_52015_equation_0, values = (var_51849_cast_fp16, var_51987_cast_fp16))[name = tensor("op_52015_cast_fp16")]; + tensor var_52017_equation_0 = const()[name = tensor("op_52017_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52017_cast_fp16 = einsum(equation = var_52017_equation_0, values = (var_51853_cast_fp16, var_51988_cast_fp16))[name = tensor("op_52017_cast_fp16")]; + tensor var_52019_equation_0 = const()[name = tensor("op_52019_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52019_cast_fp16 = einsum(equation = var_52019_equation_0, values = (var_51857_cast_fp16, var_51989_cast_fp16))[name = tensor("op_52019_cast_fp16")]; + tensor var_52021_equation_0 = const()[name = tensor("op_52021_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52021_cast_fp16 = einsum(equation = var_52021_equation_0, values = (var_51861_cast_fp16, var_51990_cast_fp16))[name = tensor("op_52021_cast_fp16")]; + tensor var_52023_equation_0 = const()[name = tensor("op_52023_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52023_cast_fp16 = einsum(equation = var_52023_equation_0, values = (var_51865_cast_fp16, var_51991_cast_fp16))[name = tensor("op_52023_cast_fp16")]; + tensor var_52025_equation_0 = const()[name = tensor("op_52025_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52025_cast_fp16 = einsum(equation = var_52025_equation_0, values = (var_51869_cast_fp16, var_51992_cast_fp16))[name = tensor("op_52025_cast_fp16")]; + tensor var_52027_equation_0 = const()[name = tensor("op_52027_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52027_cast_fp16 = einsum(equation = var_52027_equation_0, values = (var_51873_cast_fp16, var_51993_cast_fp16))[name = tensor("op_52027_cast_fp16")]; + tensor var_52029_equation_0 = const()[name = tensor("op_52029_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52029_cast_fp16 = einsum(equation = var_52029_equation_0, values = (var_51877_cast_fp16, var_51994_cast_fp16))[name = tensor("op_52029_cast_fp16")]; + tensor var_52031_equation_0 = const()[name = tensor("op_52031_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52031_cast_fp16 = einsum(equation = var_52031_equation_0, values = (var_51881_cast_fp16, var_51995_cast_fp16))[name = tensor("op_52031_cast_fp16")]; + tensor var_52033_equation_0 = const()[name = tensor("op_52033_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52033_cast_fp16 = einsum(equation = var_52033_equation_0, values = (var_51885_cast_fp16, var_51996_cast_fp16))[name = tensor("op_52033_cast_fp16")]; + tensor var_52035_equation_0 = const()[name = tensor("op_52035_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52035_cast_fp16 = einsum(equation = var_52035_equation_0, values = (var_51889_cast_fp16, var_51997_cast_fp16))[name = tensor("op_52035_cast_fp16")]; + tensor var_52037_equation_0 = const()[name = tensor("op_52037_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52037_cast_fp16 = einsum(equation = var_52037_equation_0, values = (var_51893_cast_fp16, var_51998_cast_fp16))[name = tensor("op_52037_cast_fp16")]; + tensor var_52039_equation_0 = const()[name = tensor("op_52039_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52039_cast_fp16 = einsum(equation = var_52039_equation_0, values = (var_51897_cast_fp16, var_51999_cast_fp16))[name = tensor("op_52039_cast_fp16")]; + tensor input_659_interleave_0 = const()[name = tensor("input_659_interleave_0"), val = tensor(false)]; + tensor input_659_cast_fp16 = concat(axis = var_30355, interleave = input_659_interleave_0, values = (var_52001_cast_fp16, var_52003_cast_fp16, var_52005_cast_fp16, var_52007_cast_fp16, var_52009_cast_fp16, var_52011_cast_fp16, var_52013_cast_fp16, var_52015_cast_fp16, var_52017_cast_fp16, var_52019_cast_fp16, var_52021_cast_fp16, var_52023_cast_fp16, var_52025_cast_fp16, var_52027_cast_fp16, var_52029_cast_fp16, var_52031_cast_fp16, var_52033_cast_fp16, var_52035_cast_fp16, var_52037_cast_fp16, var_52039_cast_fp16))[name = tensor("input_659_cast_fp16")]; + tensor var_52045 = const()[name = tensor("op_52045"), val = tensor([1, 1])]; + tensor var_52047 = const()[name = tensor("op_52047"), val = tensor([1, 1])]; + tensor var_52049_pad_type_0 = const()[name = tensor("op_52049_pad_type_0"), val = tensor("custom")]; + tensor var_52049_pad_0 = const()[name = tensor("op_52049_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_3_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_3_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4390465728)))]; + tensor up_blocks_0_attentions_2_transformer_blocks_3_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_3_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4393742592)))]; + tensor var_52049_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_3_attn1_to_out_0_bias_to_fp16, dilations = var_52047, groups = var_30355, pad = var_52049_pad_0, pad_type = var_52049_pad_type_0, strides = var_52045, weight = up_blocks_0_attentions_2_transformer_blocks_3_attn1_to_out_0_weight_to_fp16, x = input_659_cast_fp16)[name = tensor("op_52049_cast_fp16")]; + tensor inputs_345_cast_fp16 = add(x = var_52049_cast_fp16, y = inputs_343_cast_fp16)[name = tensor("inputs_345_cast_fp16")]; + tensor hidden_states_455_axes_0 = const()[name = tensor("hidden_states_455_axes_0"), val = tensor([1])]; + tensor hidden_states_455_gamma_0_to_fp16 = const()[name = tensor("hidden_states_455_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4393745216)))]; + tensor hidden_states_455_beta_0_to_fp16 = const()[name = tensor("hidden_states_455_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4393747840)))]; + tensor var_52059_to_fp16 = const()[name = tensor("op_52059_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_455_cast_fp16 = layer_norm(axes = hidden_states_455_axes_0, beta = hidden_states_455_beta_0_to_fp16, epsilon = var_52059_to_fp16, gamma = hidden_states_455_gamma_0_to_fp16, x = inputs_345_cast_fp16)[name = tensor("hidden_states_455_cast_fp16")]; + tensor var_52074 = const()[name = tensor("op_52074"), val = tensor([1, 1])]; + tensor var_52076 = const()[name = tensor("op_52076"), val = tensor([1, 1])]; + tensor q_231_pad_type_0 = const()[name = tensor("q_231_pad_type_0"), val = tensor("custom")]; + tensor q_231_pad_0 = const()[name = tensor("q_231_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_3_attn2_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_3_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4393750464)))]; + tensor q_231_cast_fp16 = conv(dilations = var_52076, groups = var_30355, pad = q_231_pad_0, pad_type = q_231_pad_type_0, strides = var_52074, weight = up_blocks_0_attentions_2_transformer_blocks_3_attn2_to_q_weight_to_fp16, x = hidden_states_455_cast_fp16)[name = tensor("q_231_cast_fp16")]; + tensor var_52080 = const()[name = tensor("op_52080"), val = tensor([1, 1])]; + tensor var_52082 = const()[name = tensor("op_52082"), val = tensor([1, 1])]; + tensor k_461_pad_type_0 = const()[name = tensor("k_461_pad_type_0"), val = tensor("custom")]; + tensor k_461_pad_0 = const()[name = tensor("k_461_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_3_attn2_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_3_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4397027328)))]; + tensor k_461_cast_fp16 = conv(dilations = var_52082, groups = var_30355, pad = k_461_pad_0, pad_type = k_461_pad_type_0, strides = var_52080, weight = up_blocks_0_attentions_2_transformer_blocks_3_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_461_cast_fp16")]; + tensor var_52086 = const()[name = tensor("op_52086"), val = tensor([1, 1])]; + tensor var_52088 = const()[name = tensor("op_52088"), val = tensor([1, 1])]; + tensor v_231_pad_type_0 = const()[name = tensor("v_231_pad_type_0"), val = tensor("custom")]; + tensor v_231_pad_0 = const()[name = tensor("v_231_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_3_attn2_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_3_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4402270272)))]; + tensor v_231_cast_fp16 = conv(dilations = var_52088, groups = var_30355, pad = v_231_pad_0, pad_type = v_231_pad_type_0, strides = var_52086, weight = up_blocks_0_attentions_2_transformer_blocks_3_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_231_cast_fp16")]; + tensor var_52092_begin_0 = const()[name = tensor("op_52092_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_52092_end_0 = const()[name = tensor("op_52092_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_52092_end_mask_0 = const()[name = tensor("op_52092_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52092_cast_fp16 = slice_by_index(begin = var_52092_begin_0, end = var_52092_end_0, end_mask = var_52092_end_mask_0, x = q_231_cast_fp16)[name = tensor("op_52092_cast_fp16")]; + tensor var_52096_begin_0 = const()[name = tensor("op_52096_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_52096_end_0 = const()[name = tensor("op_52096_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_52096_end_mask_0 = const()[name = tensor("op_52096_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52096_cast_fp16 = slice_by_index(begin = var_52096_begin_0, end = var_52096_end_0, end_mask = var_52096_end_mask_0, x = q_231_cast_fp16)[name = tensor("op_52096_cast_fp16")]; + tensor var_52100_begin_0 = const()[name = tensor("op_52100_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_52100_end_0 = const()[name = tensor("op_52100_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_52100_end_mask_0 = const()[name = tensor("op_52100_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52100_cast_fp16 = slice_by_index(begin = var_52100_begin_0, end = var_52100_end_0, end_mask = var_52100_end_mask_0, x = q_231_cast_fp16)[name = tensor("op_52100_cast_fp16")]; + tensor var_52104_begin_0 = const()[name = tensor("op_52104_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_52104_end_0 = const()[name = tensor("op_52104_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_52104_end_mask_0 = const()[name = tensor("op_52104_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52104_cast_fp16 = slice_by_index(begin = var_52104_begin_0, end = var_52104_end_0, end_mask = var_52104_end_mask_0, x = q_231_cast_fp16)[name = tensor("op_52104_cast_fp16")]; + tensor var_52108_begin_0 = const()[name = tensor("op_52108_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_52108_end_0 = const()[name = tensor("op_52108_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_52108_end_mask_0 = const()[name = tensor("op_52108_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52108_cast_fp16 = slice_by_index(begin = var_52108_begin_0, end = var_52108_end_0, end_mask = var_52108_end_mask_0, x = q_231_cast_fp16)[name = tensor("op_52108_cast_fp16")]; + tensor var_52112_begin_0 = const()[name = tensor("op_52112_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_52112_end_0 = const()[name = tensor("op_52112_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_52112_end_mask_0 = const()[name = tensor("op_52112_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52112_cast_fp16 = slice_by_index(begin = var_52112_begin_0, end = var_52112_end_0, end_mask = var_52112_end_mask_0, x = q_231_cast_fp16)[name = tensor("op_52112_cast_fp16")]; + tensor var_52116_begin_0 = const()[name = tensor("op_52116_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_52116_end_0 = const()[name = tensor("op_52116_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_52116_end_mask_0 = const()[name = tensor("op_52116_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52116_cast_fp16 = slice_by_index(begin = var_52116_begin_0, end = var_52116_end_0, end_mask = var_52116_end_mask_0, x = q_231_cast_fp16)[name = tensor("op_52116_cast_fp16")]; + tensor var_52120_begin_0 = const()[name = tensor("op_52120_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_52120_end_0 = const()[name = tensor("op_52120_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_52120_end_mask_0 = const()[name = tensor("op_52120_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52120_cast_fp16 = slice_by_index(begin = var_52120_begin_0, end = var_52120_end_0, end_mask = var_52120_end_mask_0, x = q_231_cast_fp16)[name = tensor("op_52120_cast_fp16")]; + tensor var_52124_begin_0 = const()[name = tensor("op_52124_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_52124_end_0 = const()[name = tensor("op_52124_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_52124_end_mask_0 = const()[name = tensor("op_52124_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52124_cast_fp16 = slice_by_index(begin = var_52124_begin_0, end = var_52124_end_0, end_mask = var_52124_end_mask_0, x = q_231_cast_fp16)[name = tensor("op_52124_cast_fp16")]; + tensor var_52128_begin_0 = const()[name = tensor("op_52128_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_52128_end_0 = const()[name = tensor("op_52128_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_52128_end_mask_0 = const()[name = tensor("op_52128_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52128_cast_fp16 = slice_by_index(begin = var_52128_begin_0, end = var_52128_end_0, end_mask = var_52128_end_mask_0, x = q_231_cast_fp16)[name = tensor("op_52128_cast_fp16")]; + tensor var_52132_begin_0 = const()[name = tensor("op_52132_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_52132_end_0 = const()[name = tensor("op_52132_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_52132_end_mask_0 = const()[name = tensor("op_52132_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52132_cast_fp16 = slice_by_index(begin = var_52132_begin_0, end = var_52132_end_0, end_mask = var_52132_end_mask_0, x = q_231_cast_fp16)[name = tensor("op_52132_cast_fp16")]; + tensor var_52136_begin_0 = const()[name = tensor("op_52136_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_52136_end_0 = const()[name = tensor("op_52136_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_52136_end_mask_0 = const()[name = tensor("op_52136_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52136_cast_fp16 = slice_by_index(begin = var_52136_begin_0, end = var_52136_end_0, end_mask = var_52136_end_mask_0, x = q_231_cast_fp16)[name = tensor("op_52136_cast_fp16")]; + tensor var_52140_begin_0 = const()[name = tensor("op_52140_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_52140_end_0 = const()[name = tensor("op_52140_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_52140_end_mask_0 = const()[name = tensor("op_52140_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52140_cast_fp16 = slice_by_index(begin = var_52140_begin_0, end = var_52140_end_0, end_mask = var_52140_end_mask_0, x = q_231_cast_fp16)[name = tensor("op_52140_cast_fp16")]; + tensor var_52144_begin_0 = const()[name = tensor("op_52144_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_52144_end_0 = const()[name = tensor("op_52144_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_52144_end_mask_0 = const()[name = tensor("op_52144_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52144_cast_fp16 = slice_by_index(begin = var_52144_begin_0, end = var_52144_end_0, end_mask = var_52144_end_mask_0, x = q_231_cast_fp16)[name = tensor("op_52144_cast_fp16")]; + tensor var_52148_begin_0 = const()[name = tensor("op_52148_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_52148_end_0 = const()[name = tensor("op_52148_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_52148_end_mask_0 = const()[name = tensor("op_52148_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52148_cast_fp16 = slice_by_index(begin = var_52148_begin_0, end = var_52148_end_0, end_mask = var_52148_end_mask_0, x = q_231_cast_fp16)[name = tensor("op_52148_cast_fp16")]; + tensor var_52152_begin_0 = const()[name = tensor("op_52152_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_52152_end_0 = const()[name = tensor("op_52152_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_52152_end_mask_0 = const()[name = tensor("op_52152_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52152_cast_fp16 = slice_by_index(begin = var_52152_begin_0, end = var_52152_end_0, end_mask = var_52152_end_mask_0, x = q_231_cast_fp16)[name = tensor("op_52152_cast_fp16")]; + tensor var_52156_begin_0 = const()[name = tensor("op_52156_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_52156_end_0 = const()[name = tensor("op_52156_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_52156_end_mask_0 = const()[name = tensor("op_52156_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52156_cast_fp16 = slice_by_index(begin = var_52156_begin_0, end = var_52156_end_0, end_mask = var_52156_end_mask_0, x = q_231_cast_fp16)[name = tensor("op_52156_cast_fp16")]; + tensor var_52160_begin_0 = const()[name = tensor("op_52160_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_52160_end_0 = const()[name = tensor("op_52160_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_52160_end_mask_0 = const()[name = tensor("op_52160_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52160_cast_fp16 = slice_by_index(begin = var_52160_begin_0, end = var_52160_end_0, end_mask = var_52160_end_mask_0, x = q_231_cast_fp16)[name = tensor("op_52160_cast_fp16")]; + tensor var_52164_begin_0 = const()[name = tensor("op_52164_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_52164_end_0 = const()[name = tensor("op_52164_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_52164_end_mask_0 = const()[name = tensor("op_52164_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52164_cast_fp16 = slice_by_index(begin = var_52164_begin_0, end = var_52164_end_0, end_mask = var_52164_end_mask_0, x = q_231_cast_fp16)[name = tensor("op_52164_cast_fp16")]; + tensor var_52168_begin_0 = const()[name = tensor("op_52168_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_52168_end_0 = const()[name = tensor("op_52168_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_52168_end_mask_0 = const()[name = tensor("op_52168_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52168_cast_fp16 = slice_by_index(begin = var_52168_begin_0, end = var_52168_end_0, end_mask = var_52168_end_mask_0, x = q_231_cast_fp16)[name = tensor("op_52168_cast_fp16")]; + tensor k_463_perm_0 = const()[name = tensor("k_463_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_52175_begin_0 = const()[name = tensor("op_52175_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_52175_end_0 = const()[name = tensor("op_52175_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_52175_end_mask_0 = const()[name = tensor("op_52175_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_463_cast_fp16 = transpose(perm = k_463_perm_0, x = k_461_cast_fp16)[name = tensor("transpose_24")]; + tensor var_52175_cast_fp16 = slice_by_index(begin = var_52175_begin_0, end = var_52175_end_0, end_mask = var_52175_end_mask_0, x = k_463_cast_fp16)[name = tensor("op_52175_cast_fp16")]; + tensor var_52179_begin_0 = const()[name = tensor("op_52179_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_52179_end_0 = const()[name = tensor("op_52179_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_52179_end_mask_0 = const()[name = tensor("op_52179_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_52179_cast_fp16 = slice_by_index(begin = var_52179_begin_0, end = var_52179_end_0, end_mask = var_52179_end_mask_0, x = k_463_cast_fp16)[name = tensor("op_52179_cast_fp16")]; + tensor var_52183_begin_0 = const()[name = tensor("op_52183_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_52183_end_0 = const()[name = tensor("op_52183_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_52183_end_mask_0 = const()[name = tensor("op_52183_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_52183_cast_fp16 = slice_by_index(begin = var_52183_begin_0, end = var_52183_end_0, end_mask = var_52183_end_mask_0, x = k_463_cast_fp16)[name = tensor("op_52183_cast_fp16")]; + tensor var_52187_begin_0 = const()[name = tensor("op_52187_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_52187_end_0 = const()[name = tensor("op_52187_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_52187_end_mask_0 = const()[name = tensor("op_52187_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_52187_cast_fp16 = slice_by_index(begin = var_52187_begin_0, end = var_52187_end_0, end_mask = var_52187_end_mask_0, x = k_463_cast_fp16)[name = tensor("op_52187_cast_fp16")]; + tensor var_52191_begin_0 = const()[name = tensor("op_52191_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_52191_end_0 = const()[name = tensor("op_52191_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_52191_end_mask_0 = const()[name = tensor("op_52191_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_52191_cast_fp16 = slice_by_index(begin = var_52191_begin_0, end = var_52191_end_0, end_mask = var_52191_end_mask_0, x = k_463_cast_fp16)[name = tensor("op_52191_cast_fp16")]; + tensor var_52195_begin_0 = const()[name = tensor("op_52195_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_52195_end_0 = const()[name = tensor("op_52195_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_52195_end_mask_0 = const()[name = tensor("op_52195_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_52195_cast_fp16 = slice_by_index(begin = var_52195_begin_0, end = var_52195_end_0, end_mask = var_52195_end_mask_0, x = k_463_cast_fp16)[name = tensor("op_52195_cast_fp16")]; + tensor var_52199_begin_0 = const()[name = tensor("op_52199_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_52199_end_0 = const()[name = tensor("op_52199_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_52199_end_mask_0 = const()[name = tensor("op_52199_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_52199_cast_fp16 = slice_by_index(begin = var_52199_begin_0, end = var_52199_end_0, end_mask = var_52199_end_mask_0, x = k_463_cast_fp16)[name = tensor("op_52199_cast_fp16")]; + tensor var_52203_begin_0 = const()[name = tensor("op_52203_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_52203_end_0 = const()[name = tensor("op_52203_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_52203_end_mask_0 = const()[name = tensor("op_52203_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_52203_cast_fp16 = slice_by_index(begin = var_52203_begin_0, end = var_52203_end_0, end_mask = var_52203_end_mask_0, x = k_463_cast_fp16)[name = tensor("op_52203_cast_fp16")]; + tensor var_52207_begin_0 = const()[name = tensor("op_52207_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_52207_end_0 = const()[name = tensor("op_52207_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_52207_end_mask_0 = const()[name = tensor("op_52207_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_52207_cast_fp16 = slice_by_index(begin = var_52207_begin_0, end = var_52207_end_0, end_mask = var_52207_end_mask_0, x = k_463_cast_fp16)[name = tensor("op_52207_cast_fp16")]; + tensor var_52211_begin_0 = const()[name = tensor("op_52211_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_52211_end_0 = const()[name = tensor("op_52211_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_52211_end_mask_0 = const()[name = tensor("op_52211_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_52211_cast_fp16 = slice_by_index(begin = var_52211_begin_0, end = var_52211_end_0, end_mask = var_52211_end_mask_0, x = k_463_cast_fp16)[name = tensor("op_52211_cast_fp16")]; + tensor var_52215_begin_0 = const()[name = tensor("op_52215_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_52215_end_0 = const()[name = tensor("op_52215_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_52215_end_mask_0 = const()[name = tensor("op_52215_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_52215_cast_fp16 = slice_by_index(begin = var_52215_begin_0, end = var_52215_end_0, end_mask = var_52215_end_mask_0, x = k_463_cast_fp16)[name = tensor("op_52215_cast_fp16")]; + tensor var_52219_begin_0 = const()[name = tensor("op_52219_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_52219_end_0 = const()[name = tensor("op_52219_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_52219_end_mask_0 = const()[name = tensor("op_52219_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_52219_cast_fp16 = slice_by_index(begin = var_52219_begin_0, end = var_52219_end_0, end_mask = var_52219_end_mask_0, x = k_463_cast_fp16)[name = tensor("op_52219_cast_fp16")]; + tensor var_52223_begin_0 = const()[name = tensor("op_52223_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_52223_end_0 = const()[name = tensor("op_52223_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_52223_end_mask_0 = const()[name = tensor("op_52223_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_52223_cast_fp16 = slice_by_index(begin = var_52223_begin_0, end = var_52223_end_0, end_mask = var_52223_end_mask_0, x = k_463_cast_fp16)[name = tensor("op_52223_cast_fp16")]; + tensor var_52227_begin_0 = const()[name = tensor("op_52227_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_52227_end_0 = const()[name = tensor("op_52227_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_52227_end_mask_0 = const()[name = tensor("op_52227_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_52227_cast_fp16 = slice_by_index(begin = var_52227_begin_0, end = var_52227_end_0, end_mask = var_52227_end_mask_0, x = k_463_cast_fp16)[name = tensor("op_52227_cast_fp16")]; + tensor var_52231_begin_0 = const()[name = tensor("op_52231_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_52231_end_0 = const()[name = tensor("op_52231_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_52231_end_mask_0 = const()[name = tensor("op_52231_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_52231_cast_fp16 = slice_by_index(begin = var_52231_begin_0, end = var_52231_end_0, end_mask = var_52231_end_mask_0, x = k_463_cast_fp16)[name = tensor("op_52231_cast_fp16")]; + tensor var_52235_begin_0 = const()[name = tensor("op_52235_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_52235_end_0 = const()[name = tensor("op_52235_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_52235_end_mask_0 = const()[name = tensor("op_52235_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_52235_cast_fp16 = slice_by_index(begin = var_52235_begin_0, end = var_52235_end_0, end_mask = var_52235_end_mask_0, x = k_463_cast_fp16)[name = tensor("op_52235_cast_fp16")]; + tensor var_52239_begin_0 = const()[name = tensor("op_52239_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_52239_end_0 = const()[name = tensor("op_52239_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_52239_end_mask_0 = const()[name = tensor("op_52239_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_52239_cast_fp16 = slice_by_index(begin = var_52239_begin_0, end = var_52239_end_0, end_mask = var_52239_end_mask_0, x = k_463_cast_fp16)[name = tensor("op_52239_cast_fp16")]; + tensor var_52243_begin_0 = const()[name = tensor("op_52243_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_52243_end_0 = const()[name = tensor("op_52243_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_52243_end_mask_0 = const()[name = tensor("op_52243_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_52243_cast_fp16 = slice_by_index(begin = var_52243_begin_0, end = var_52243_end_0, end_mask = var_52243_end_mask_0, x = k_463_cast_fp16)[name = tensor("op_52243_cast_fp16")]; + tensor var_52247_begin_0 = const()[name = tensor("op_52247_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_52247_end_0 = const()[name = tensor("op_52247_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_52247_end_mask_0 = const()[name = tensor("op_52247_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_52247_cast_fp16 = slice_by_index(begin = var_52247_begin_0, end = var_52247_end_0, end_mask = var_52247_end_mask_0, x = k_463_cast_fp16)[name = tensor("op_52247_cast_fp16")]; + tensor var_52251_begin_0 = const()[name = tensor("op_52251_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_52251_end_0 = const()[name = tensor("op_52251_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_52251_end_mask_0 = const()[name = tensor("op_52251_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_52251_cast_fp16 = slice_by_index(begin = var_52251_begin_0, end = var_52251_end_0, end_mask = var_52251_end_mask_0, x = k_463_cast_fp16)[name = tensor("op_52251_cast_fp16")]; + tensor var_52253_begin_0 = const()[name = tensor("op_52253_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_52253_end_0 = const()[name = tensor("op_52253_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_52253_end_mask_0 = const()[name = tensor("op_52253_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52253_cast_fp16 = slice_by_index(begin = var_52253_begin_0, end = var_52253_end_0, end_mask = var_52253_end_mask_0, x = v_231_cast_fp16)[name = tensor("op_52253_cast_fp16")]; + tensor var_52257_begin_0 = const()[name = tensor("op_52257_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_52257_end_0 = const()[name = tensor("op_52257_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_52257_end_mask_0 = const()[name = tensor("op_52257_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52257_cast_fp16 = slice_by_index(begin = var_52257_begin_0, end = var_52257_end_0, end_mask = var_52257_end_mask_0, x = v_231_cast_fp16)[name = tensor("op_52257_cast_fp16")]; + tensor var_52261_begin_0 = const()[name = tensor("op_52261_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_52261_end_0 = const()[name = tensor("op_52261_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_52261_end_mask_0 = const()[name = tensor("op_52261_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52261_cast_fp16 = slice_by_index(begin = var_52261_begin_0, end = var_52261_end_0, end_mask = var_52261_end_mask_0, x = v_231_cast_fp16)[name = tensor("op_52261_cast_fp16")]; + tensor var_52265_begin_0 = const()[name = tensor("op_52265_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_52265_end_0 = const()[name = tensor("op_52265_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_52265_end_mask_0 = const()[name = tensor("op_52265_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52265_cast_fp16 = slice_by_index(begin = var_52265_begin_0, end = var_52265_end_0, end_mask = var_52265_end_mask_0, x = v_231_cast_fp16)[name = tensor("op_52265_cast_fp16")]; + tensor var_52269_begin_0 = const()[name = tensor("op_52269_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_52269_end_0 = const()[name = tensor("op_52269_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_52269_end_mask_0 = const()[name = tensor("op_52269_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52269_cast_fp16 = slice_by_index(begin = var_52269_begin_0, end = var_52269_end_0, end_mask = var_52269_end_mask_0, x = v_231_cast_fp16)[name = tensor("op_52269_cast_fp16")]; + tensor var_52273_begin_0 = const()[name = tensor("op_52273_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_52273_end_0 = const()[name = tensor("op_52273_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_52273_end_mask_0 = const()[name = tensor("op_52273_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52273_cast_fp16 = slice_by_index(begin = var_52273_begin_0, end = var_52273_end_0, end_mask = var_52273_end_mask_0, x = v_231_cast_fp16)[name = tensor("op_52273_cast_fp16")]; + tensor var_52277_begin_0 = const()[name = tensor("op_52277_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_52277_end_0 = const()[name = tensor("op_52277_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_52277_end_mask_0 = const()[name = tensor("op_52277_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52277_cast_fp16 = slice_by_index(begin = var_52277_begin_0, end = var_52277_end_0, end_mask = var_52277_end_mask_0, x = v_231_cast_fp16)[name = tensor("op_52277_cast_fp16")]; + tensor var_52281_begin_0 = const()[name = tensor("op_52281_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_52281_end_0 = const()[name = tensor("op_52281_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_52281_end_mask_0 = const()[name = tensor("op_52281_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52281_cast_fp16 = slice_by_index(begin = var_52281_begin_0, end = var_52281_end_0, end_mask = var_52281_end_mask_0, x = v_231_cast_fp16)[name = tensor("op_52281_cast_fp16")]; + tensor var_52285_begin_0 = const()[name = tensor("op_52285_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_52285_end_0 = const()[name = tensor("op_52285_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_52285_end_mask_0 = const()[name = tensor("op_52285_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52285_cast_fp16 = slice_by_index(begin = var_52285_begin_0, end = var_52285_end_0, end_mask = var_52285_end_mask_0, x = v_231_cast_fp16)[name = tensor("op_52285_cast_fp16")]; + tensor var_52289_begin_0 = const()[name = tensor("op_52289_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_52289_end_0 = const()[name = tensor("op_52289_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_52289_end_mask_0 = const()[name = tensor("op_52289_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52289_cast_fp16 = slice_by_index(begin = var_52289_begin_0, end = var_52289_end_0, end_mask = var_52289_end_mask_0, x = v_231_cast_fp16)[name = tensor("op_52289_cast_fp16")]; + tensor var_52293_begin_0 = const()[name = tensor("op_52293_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_52293_end_0 = const()[name = tensor("op_52293_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_52293_end_mask_0 = const()[name = tensor("op_52293_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52293_cast_fp16 = slice_by_index(begin = var_52293_begin_0, end = var_52293_end_0, end_mask = var_52293_end_mask_0, x = v_231_cast_fp16)[name = tensor("op_52293_cast_fp16")]; + tensor var_52297_begin_0 = const()[name = tensor("op_52297_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_52297_end_0 = const()[name = tensor("op_52297_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_52297_end_mask_0 = const()[name = tensor("op_52297_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52297_cast_fp16 = slice_by_index(begin = var_52297_begin_0, end = var_52297_end_0, end_mask = var_52297_end_mask_0, x = v_231_cast_fp16)[name = tensor("op_52297_cast_fp16")]; + tensor var_52301_begin_0 = const()[name = tensor("op_52301_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_52301_end_0 = const()[name = tensor("op_52301_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_52301_end_mask_0 = const()[name = tensor("op_52301_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52301_cast_fp16 = slice_by_index(begin = var_52301_begin_0, end = var_52301_end_0, end_mask = var_52301_end_mask_0, x = v_231_cast_fp16)[name = tensor("op_52301_cast_fp16")]; + tensor var_52305_begin_0 = const()[name = tensor("op_52305_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_52305_end_0 = const()[name = tensor("op_52305_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_52305_end_mask_0 = const()[name = tensor("op_52305_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52305_cast_fp16 = slice_by_index(begin = var_52305_begin_0, end = var_52305_end_0, end_mask = var_52305_end_mask_0, x = v_231_cast_fp16)[name = tensor("op_52305_cast_fp16")]; + tensor var_52309_begin_0 = const()[name = tensor("op_52309_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_52309_end_0 = const()[name = tensor("op_52309_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_52309_end_mask_0 = const()[name = tensor("op_52309_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52309_cast_fp16 = slice_by_index(begin = var_52309_begin_0, end = var_52309_end_0, end_mask = var_52309_end_mask_0, x = v_231_cast_fp16)[name = tensor("op_52309_cast_fp16")]; + tensor var_52313_begin_0 = const()[name = tensor("op_52313_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_52313_end_0 = const()[name = tensor("op_52313_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_52313_end_mask_0 = const()[name = tensor("op_52313_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52313_cast_fp16 = slice_by_index(begin = var_52313_begin_0, end = var_52313_end_0, end_mask = var_52313_end_mask_0, x = v_231_cast_fp16)[name = tensor("op_52313_cast_fp16")]; + tensor var_52317_begin_0 = const()[name = tensor("op_52317_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_52317_end_0 = const()[name = tensor("op_52317_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_52317_end_mask_0 = const()[name = tensor("op_52317_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52317_cast_fp16 = slice_by_index(begin = var_52317_begin_0, end = var_52317_end_0, end_mask = var_52317_end_mask_0, x = v_231_cast_fp16)[name = tensor("op_52317_cast_fp16")]; + tensor var_52321_begin_0 = const()[name = tensor("op_52321_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_52321_end_0 = const()[name = tensor("op_52321_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_52321_end_mask_0 = const()[name = tensor("op_52321_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52321_cast_fp16 = slice_by_index(begin = var_52321_begin_0, end = var_52321_end_0, end_mask = var_52321_end_mask_0, x = v_231_cast_fp16)[name = tensor("op_52321_cast_fp16")]; + tensor var_52325_begin_0 = const()[name = tensor("op_52325_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_52325_end_0 = const()[name = tensor("op_52325_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_52325_end_mask_0 = const()[name = tensor("op_52325_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52325_cast_fp16 = slice_by_index(begin = var_52325_begin_0, end = var_52325_end_0, end_mask = var_52325_end_mask_0, x = v_231_cast_fp16)[name = tensor("op_52325_cast_fp16")]; + tensor var_52329_begin_0 = const()[name = tensor("op_52329_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_52329_end_0 = const()[name = tensor("op_52329_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_52329_end_mask_0 = const()[name = tensor("op_52329_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52329_cast_fp16 = slice_by_index(begin = var_52329_begin_0, end = var_52329_end_0, end_mask = var_52329_end_mask_0, x = v_231_cast_fp16)[name = tensor("op_52329_cast_fp16")]; + tensor var_52333_equation_0 = const()[name = tensor("op_52333_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52333_cast_fp16 = einsum(equation = var_52333_equation_0, values = (var_52175_cast_fp16, var_52092_cast_fp16))[name = tensor("op_52333_cast_fp16")]; + tensor var_52334_to_fp16 = const()[name = tensor("op_52334_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4441_cast_fp16 = mul(x = var_52333_cast_fp16, y = var_52334_to_fp16)[name = tensor("aw_4441_cast_fp16")]; + tensor var_52337_equation_0 = const()[name = tensor("op_52337_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52337_cast_fp16 = einsum(equation = var_52337_equation_0, values = (var_52179_cast_fp16, var_52096_cast_fp16))[name = tensor("op_52337_cast_fp16")]; + tensor var_52338_to_fp16 = const()[name = tensor("op_52338_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4443_cast_fp16 = mul(x = var_52337_cast_fp16, y = var_52338_to_fp16)[name = tensor("aw_4443_cast_fp16")]; + tensor var_52341_equation_0 = const()[name = tensor("op_52341_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52341_cast_fp16 = einsum(equation = var_52341_equation_0, values = (var_52183_cast_fp16, var_52100_cast_fp16))[name = tensor("op_52341_cast_fp16")]; + tensor var_52342_to_fp16 = const()[name = tensor("op_52342_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4445_cast_fp16 = mul(x = var_52341_cast_fp16, y = var_52342_to_fp16)[name = tensor("aw_4445_cast_fp16")]; + tensor var_52345_equation_0 = const()[name = tensor("op_52345_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52345_cast_fp16 = einsum(equation = var_52345_equation_0, values = (var_52187_cast_fp16, var_52104_cast_fp16))[name = tensor("op_52345_cast_fp16")]; + tensor var_52346_to_fp16 = const()[name = tensor("op_52346_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4447_cast_fp16 = mul(x = var_52345_cast_fp16, y = var_52346_to_fp16)[name = tensor("aw_4447_cast_fp16")]; + tensor var_52349_equation_0 = const()[name = tensor("op_52349_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52349_cast_fp16 = einsum(equation = var_52349_equation_0, values = (var_52191_cast_fp16, var_52108_cast_fp16))[name = tensor("op_52349_cast_fp16")]; + tensor var_52350_to_fp16 = const()[name = tensor("op_52350_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4449_cast_fp16 = mul(x = var_52349_cast_fp16, y = var_52350_to_fp16)[name = tensor("aw_4449_cast_fp16")]; + tensor var_52353_equation_0 = const()[name = tensor("op_52353_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52353_cast_fp16 = einsum(equation = var_52353_equation_0, values = (var_52195_cast_fp16, var_52112_cast_fp16))[name = tensor("op_52353_cast_fp16")]; + tensor var_52354_to_fp16 = const()[name = tensor("op_52354_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4451_cast_fp16 = mul(x = var_52353_cast_fp16, y = var_52354_to_fp16)[name = tensor("aw_4451_cast_fp16")]; + tensor var_52357_equation_0 = const()[name = tensor("op_52357_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52357_cast_fp16 = einsum(equation = var_52357_equation_0, values = (var_52199_cast_fp16, var_52116_cast_fp16))[name = tensor("op_52357_cast_fp16")]; + tensor var_52358_to_fp16 = const()[name = tensor("op_52358_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4453_cast_fp16 = mul(x = var_52357_cast_fp16, y = var_52358_to_fp16)[name = tensor("aw_4453_cast_fp16")]; + tensor var_52361_equation_0 = const()[name = tensor("op_52361_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52361_cast_fp16 = einsum(equation = var_52361_equation_0, values = (var_52203_cast_fp16, var_52120_cast_fp16))[name = tensor("op_52361_cast_fp16")]; + tensor var_52362_to_fp16 = const()[name = tensor("op_52362_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4455_cast_fp16 = mul(x = var_52361_cast_fp16, y = var_52362_to_fp16)[name = tensor("aw_4455_cast_fp16")]; + tensor var_52365_equation_0 = const()[name = tensor("op_52365_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52365_cast_fp16 = einsum(equation = var_52365_equation_0, values = (var_52207_cast_fp16, var_52124_cast_fp16))[name = tensor("op_52365_cast_fp16")]; + tensor var_52366_to_fp16 = const()[name = tensor("op_52366_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4457_cast_fp16 = mul(x = var_52365_cast_fp16, y = var_52366_to_fp16)[name = tensor("aw_4457_cast_fp16")]; + tensor var_52369_equation_0 = const()[name = tensor("op_52369_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52369_cast_fp16 = einsum(equation = var_52369_equation_0, values = (var_52211_cast_fp16, var_52128_cast_fp16))[name = tensor("op_52369_cast_fp16")]; + tensor var_52370_to_fp16 = const()[name = tensor("op_52370_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4459_cast_fp16 = mul(x = var_52369_cast_fp16, y = var_52370_to_fp16)[name = tensor("aw_4459_cast_fp16")]; + tensor var_52373_equation_0 = const()[name = tensor("op_52373_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52373_cast_fp16 = einsum(equation = var_52373_equation_0, values = (var_52215_cast_fp16, var_52132_cast_fp16))[name = tensor("op_52373_cast_fp16")]; + tensor var_52374_to_fp16 = const()[name = tensor("op_52374_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4461_cast_fp16 = mul(x = var_52373_cast_fp16, y = var_52374_to_fp16)[name = tensor("aw_4461_cast_fp16")]; + tensor var_52377_equation_0 = const()[name = tensor("op_52377_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52377_cast_fp16 = einsum(equation = var_52377_equation_0, values = (var_52219_cast_fp16, var_52136_cast_fp16))[name = tensor("op_52377_cast_fp16")]; + tensor var_52378_to_fp16 = const()[name = tensor("op_52378_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4463_cast_fp16 = mul(x = var_52377_cast_fp16, y = var_52378_to_fp16)[name = tensor("aw_4463_cast_fp16")]; + tensor var_52381_equation_0 = const()[name = tensor("op_52381_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52381_cast_fp16 = einsum(equation = var_52381_equation_0, values = (var_52223_cast_fp16, var_52140_cast_fp16))[name = tensor("op_52381_cast_fp16")]; + tensor var_52382_to_fp16 = const()[name = tensor("op_52382_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4465_cast_fp16 = mul(x = var_52381_cast_fp16, y = var_52382_to_fp16)[name = tensor("aw_4465_cast_fp16")]; + tensor var_52385_equation_0 = const()[name = tensor("op_52385_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52385_cast_fp16 = einsum(equation = var_52385_equation_0, values = (var_52227_cast_fp16, var_52144_cast_fp16))[name = tensor("op_52385_cast_fp16")]; + tensor var_52386_to_fp16 = const()[name = tensor("op_52386_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4467_cast_fp16 = mul(x = var_52385_cast_fp16, y = var_52386_to_fp16)[name = tensor("aw_4467_cast_fp16")]; + tensor var_52389_equation_0 = const()[name = tensor("op_52389_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52389_cast_fp16 = einsum(equation = var_52389_equation_0, values = (var_52231_cast_fp16, var_52148_cast_fp16))[name = tensor("op_52389_cast_fp16")]; + tensor var_52390_to_fp16 = const()[name = tensor("op_52390_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4469_cast_fp16 = mul(x = var_52389_cast_fp16, y = var_52390_to_fp16)[name = tensor("aw_4469_cast_fp16")]; + tensor var_52393_equation_0 = const()[name = tensor("op_52393_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52393_cast_fp16 = einsum(equation = var_52393_equation_0, values = (var_52235_cast_fp16, var_52152_cast_fp16))[name = tensor("op_52393_cast_fp16")]; + tensor var_52394_to_fp16 = const()[name = tensor("op_52394_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4471_cast_fp16 = mul(x = var_52393_cast_fp16, y = var_52394_to_fp16)[name = tensor("aw_4471_cast_fp16")]; + tensor var_52397_equation_0 = const()[name = tensor("op_52397_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52397_cast_fp16 = einsum(equation = var_52397_equation_0, values = (var_52239_cast_fp16, var_52156_cast_fp16))[name = tensor("op_52397_cast_fp16")]; + tensor var_52398_to_fp16 = const()[name = tensor("op_52398_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4473_cast_fp16 = mul(x = var_52397_cast_fp16, y = var_52398_to_fp16)[name = tensor("aw_4473_cast_fp16")]; + tensor var_52401_equation_0 = const()[name = tensor("op_52401_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52401_cast_fp16 = einsum(equation = var_52401_equation_0, values = (var_52243_cast_fp16, var_52160_cast_fp16))[name = tensor("op_52401_cast_fp16")]; + tensor var_52402_to_fp16 = const()[name = tensor("op_52402_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4475_cast_fp16 = mul(x = var_52401_cast_fp16, y = var_52402_to_fp16)[name = tensor("aw_4475_cast_fp16")]; + tensor var_52405_equation_0 = const()[name = tensor("op_52405_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52405_cast_fp16 = einsum(equation = var_52405_equation_0, values = (var_52247_cast_fp16, var_52164_cast_fp16))[name = tensor("op_52405_cast_fp16")]; + tensor var_52406_to_fp16 = const()[name = tensor("op_52406_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4477_cast_fp16 = mul(x = var_52405_cast_fp16, y = var_52406_to_fp16)[name = tensor("aw_4477_cast_fp16")]; + tensor var_52409_equation_0 = const()[name = tensor("op_52409_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52409_cast_fp16 = einsum(equation = var_52409_equation_0, values = (var_52251_cast_fp16, var_52168_cast_fp16))[name = tensor("op_52409_cast_fp16")]; + tensor var_52410_to_fp16 = const()[name = tensor("op_52410_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4479_cast_fp16 = mul(x = var_52409_cast_fp16, y = var_52410_to_fp16)[name = tensor("aw_4479_cast_fp16")]; + tensor var_52412_cast_fp16 = softmax(axis = var_30355, x = aw_4441_cast_fp16)[name = tensor("op_52412_cast_fp16")]; + tensor var_52413_cast_fp16 = softmax(axis = var_30355, x = aw_4443_cast_fp16)[name = tensor("op_52413_cast_fp16")]; + tensor var_52414_cast_fp16 = softmax(axis = var_30355, x = aw_4445_cast_fp16)[name = tensor("op_52414_cast_fp16")]; + tensor var_52415_cast_fp16 = softmax(axis = var_30355, x = aw_4447_cast_fp16)[name = tensor("op_52415_cast_fp16")]; + tensor var_52416_cast_fp16 = softmax(axis = var_30355, x = aw_4449_cast_fp16)[name = tensor("op_52416_cast_fp16")]; + tensor var_52417_cast_fp16 = softmax(axis = var_30355, x = aw_4451_cast_fp16)[name = tensor("op_52417_cast_fp16")]; + tensor var_52418_cast_fp16 = softmax(axis = var_30355, x = aw_4453_cast_fp16)[name = tensor("op_52418_cast_fp16")]; + tensor var_52419_cast_fp16 = softmax(axis = var_30355, x = aw_4455_cast_fp16)[name = tensor("op_52419_cast_fp16")]; + tensor var_52420_cast_fp16 = softmax(axis = var_30355, x = aw_4457_cast_fp16)[name = tensor("op_52420_cast_fp16")]; + tensor var_52421_cast_fp16 = softmax(axis = var_30355, x = aw_4459_cast_fp16)[name = tensor("op_52421_cast_fp16")]; + tensor var_52422_cast_fp16 = softmax(axis = var_30355, x = aw_4461_cast_fp16)[name = tensor("op_52422_cast_fp16")]; + tensor var_52423_cast_fp16 = softmax(axis = var_30355, x = aw_4463_cast_fp16)[name = tensor("op_52423_cast_fp16")]; + tensor var_52424_cast_fp16 = softmax(axis = var_30355, x = aw_4465_cast_fp16)[name = tensor("op_52424_cast_fp16")]; + tensor var_52425_cast_fp16 = softmax(axis = var_30355, x = aw_4467_cast_fp16)[name = tensor("op_52425_cast_fp16")]; + tensor var_52426_cast_fp16 = softmax(axis = var_30355, x = aw_4469_cast_fp16)[name = tensor("op_52426_cast_fp16")]; + tensor var_52427_cast_fp16 = softmax(axis = var_30355, x = aw_4471_cast_fp16)[name = tensor("op_52427_cast_fp16")]; + tensor var_52428_cast_fp16 = softmax(axis = var_30355, x = aw_4473_cast_fp16)[name = tensor("op_52428_cast_fp16")]; + tensor var_52429_cast_fp16 = softmax(axis = var_30355, x = aw_4475_cast_fp16)[name = tensor("op_52429_cast_fp16")]; + tensor var_52430_cast_fp16 = softmax(axis = var_30355, x = aw_4477_cast_fp16)[name = tensor("op_52430_cast_fp16")]; + tensor var_52431_cast_fp16 = softmax(axis = var_30355, x = aw_4479_cast_fp16)[name = tensor("op_52431_cast_fp16")]; + tensor var_52433_equation_0 = const()[name = tensor("op_52433_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52433_cast_fp16 = einsum(equation = var_52433_equation_0, values = (var_52253_cast_fp16, var_52412_cast_fp16))[name = tensor("op_52433_cast_fp16")]; + tensor var_52435_equation_0 = const()[name = tensor("op_52435_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52435_cast_fp16 = einsum(equation = var_52435_equation_0, values = (var_52257_cast_fp16, var_52413_cast_fp16))[name = tensor("op_52435_cast_fp16")]; + tensor var_52437_equation_0 = const()[name = tensor("op_52437_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52437_cast_fp16 = einsum(equation = var_52437_equation_0, values = (var_52261_cast_fp16, var_52414_cast_fp16))[name = tensor("op_52437_cast_fp16")]; + tensor var_52439_equation_0 = const()[name = tensor("op_52439_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52439_cast_fp16 = einsum(equation = var_52439_equation_0, values = (var_52265_cast_fp16, var_52415_cast_fp16))[name = tensor("op_52439_cast_fp16")]; + tensor var_52441_equation_0 = const()[name = tensor("op_52441_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52441_cast_fp16 = einsum(equation = var_52441_equation_0, values = (var_52269_cast_fp16, var_52416_cast_fp16))[name = tensor("op_52441_cast_fp16")]; + tensor var_52443_equation_0 = const()[name = tensor("op_52443_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52443_cast_fp16 = einsum(equation = var_52443_equation_0, values = (var_52273_cast_fp16, var_52417_cast_fp16))[name = tensor("op_52443_cast_fp16")]; + tensor var_52445_equation_0 = const()[name = tensor("op_52445_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52445_cast_fp16 = einsum(equation = var_52445_equation_0, values = (var_52277_cast_fp16, var_52418_cast_fp16))[name = tensor("op_52445_cast_fp16")]; + tensor var_52447_equation_0 = const()[name = tensor("op_52447_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52447_cast_fp16 = einsum(equation = var_52447_equation_0, values = (var_52281_cast_fp16, var_52419_cast_fp16))[name = tensor("op_52447_cast_fp16")]; + tensor var_52449_equation_0 = const()[name = tensor("op_52449_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52449_cast_fp16 = einsum(equation = var_52449_equation_0, values = (var_52285_cast_fp16, var_52420_cast_fp16))[name = tensor("op_52449_cast_fp16")]; + tensor var_52451_equation_0 = const()[name = tensor("op_52451_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52451_cast_fp16 = einsum(equation = var_52451_equation_0, values = (var_52289_cast_fp16, var_52421_cast_fp16))[name = tensor("op_52451_cast_fp16")]; + tensor var_52453_equation_0 = const()[name = tensor("op_52453_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52453_cast_fp16 = einsum(equation = var_52453_equation_0, values = (var_52293_cast_fp16, var_52422_cast_fp16))[name = tensor("op_52453_cast_fp16")]; + tensor var_52455_equation_0 = const()[name = tensor("op_52455_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52455_cast_fp16 = einsum(equation = var_52455_equation_0, values = (var_52297_cast_fp16, var_52423_cast_fp16))[name = tensor("op_52455_cast_fp16")]; + tensor var_52457_equation_0 = const()[name = tensor("op_52457_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52457_cast_fp16 = einsum(equation = var_52457_equation_0, values = (var_52301_cast_fp16, var_52424_cast_fp16))[name = tensor("op_52457_cast_fp16")]; + tensor var_52459_equation_0 = const()[name = tensor("op_52459_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52459_cast_fp16 = einsum(equation = var_52459_equation_0, values = (var_52305_cast_fp16, var_52425_cast_fp16))[name = tensor("op_52459_cast_fp16")]; + tensor var_52461_equation_0 = const()[name = tensor("op_52461_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52461_cast_fp16 = einsum(equation = var_52461_equation_0, values = (var_52309_cast_fp16, var_52426_cast_fp16))[name = tensor("op_52461_cast_fp16")]; + tensor var_52463_equation_0 = const()[name = tensor("op_52463_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52463_cast_fp16 = einsum(equation = var_52463_equation_0, values = (var_52313_cast_fp16, var_52427_cast_fp16))[name = tensor("op_52463_cast_fp16")]; + tensor var_52465_equation_0 = const()[name = tensor("op_52465_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52465_cast_fp16 = einsum(equation = var_52465_equation_0, values = (var_52317_cast_fp16, var_52428_cast_fp16))[name = tensor("op_52465_cast_fp16")]; + tensor var_52467_equation_0 = const()[name = tensor("op_52467_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52467_cast_fp16 = einsum(equation = var_52467_equation_0, values = (var_52321_cast_fp16, var_52429_cast_fp16))[name = tensor("op_52467_cast_fp16")]; + tensor var_52469_equation_0 = const()[name = tensor("op_52469_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52469_cast_fp16 = einsum(equation = var_52469_equation_0, values = (var_52325_cast_fp16, var_52430_cast_fp16))[name = tensor("op_52469_cast_fp16")]; + tensor var_52471_equation_0 = const()[name = tensor("op_52471_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52471_cast_fp16 = einsum(equation = var_52471_equation_0, values = (var_52329_cast_fp16, var_52431_cast_fp16))[name = tensor("op_52471_cast_fp16")]; + tensor input_661_interleave_0 = const()[name = tensor("input_661_interleave_0"), val = tensor(false)]; + tensor input_661_cast_fp16 = concat(axis = var_30355, interleave = input_661_interleave_0, values = (var_52433_cast_fp16, var_52435_cast_fp16, var_52437_cast_fp16, var_52439_cast_fp16, var_52441_cast_fp16, var_52443_cast_fp16, var_52445_cast_fp16, var_52447_cast_fp16, var_52449_cast_fp16, var_52451_cast_fp16, var_52453_cast_fp16, var_52455_cast_fp16, var_52457_cast_fp16, var_52459_cast_fp16, var_52461_cast_fp16, var_52463_cast_fp16, var_52465_cast_fp16, var_52467_cast_fp16, var_52469_cast_fp16, var_52471_cast_fp16))[name = tensor("input_661_cast_fp16")]; + tensor var_52477 = const()[name = tensor("op_52477"), val = tensor([1, 1])]; + tensor var_52479 = const()[name = tensor("op_52479"), val = tensor([1, 1])]; + tensor var_52481_pad_type_0 = const()[name = tensor("op_52481_pad_type_0"), val = tensor("custom")]; + tensor var_52481_pad_0 = const()[name = tensor("op_52481_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_3_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_3_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4407513216)))]; + tensor up_blocks_0_attentions_2_transformer_blocks_3_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_3_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4410790080)))]; + tensor var_52481_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_3_attn2_to_out_0_bias_to_fp16, dilations = var_52479, groups = var_30355, pad = var_52481_pad_0, pad_type = var_52481_pad_type_0, strides = var_52477, weight = up_blocks_0_attentions_2_transformer_blocks_3_attn2_to_out_0_weight_to_fp16, x = input_661_cast_fp16)[name = tensor("op_52481_cast_fp16")]; + tensor inputs_347_cast_fp16 = add(x = var_52481_cast_fp16, y = inputs_345_cast_fp16)[name = tensor("inputs_347_cast_fp16")]; + tensor input_663_axes_0 = const()[name = tensor("input_663_axes_0"), val = tensor([1])]; + tensor input_663_gamma_0_to_fp16 = const()[name = tensor("input_663_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4410792704)))]; + tensor input_663_beta_0_to_fp16 = const()[name = tensor("input_663_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4410795328)))]; + tensor var_52491_to_fp16 = const()[name = tensor("op_52491_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_663_cast_fp16 = layer_norm(axes = input_663_axes_0, beta = input_663_beta_0_to_fp16, epsilon = var_52491_to_fp16, gamma = input_663_gamma_0_to_fp16, x = inputs_347_cast_fp16)[name = tensor("input_663_cast_fp16")]; + tensor var_52507 = const()[name = tensor("op_52507"), val = tensor([1, 1])]; + tensor var_52509 = const()[name = tensor("op_52509"), val = tensor([1, 1])]; + tensor var_52511_pad_type_0 = const()[name = tensor("op_52511_pad_type_0"), val = tensor("custom")]; + tensor var_52511_pad_0 = const()[name = tensor("op_52511_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_3_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_3_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4410797952)))]; + tensor up_blocks_0_attentions_2_transformer_blocks_3_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_3_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4437012416)))]; + tensor var_52511_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_3_ff_net_0_proj_bias_to_fp16, dilations = var_52509, groups = var_30355, pad = var_52511_pad_0, pad_type = var_52511_pad_type_0, strides = var_52507, weight = up_blocks_0_attentions_2_transformer_blocks_3_ff_net_0_proj_weight_to_fp16, x = input_663_cast_fp16)[name = tensor("op_52511_cast_fp16")]; + tensor var_52512_split_sizes_0 = const()[name = tensor("op_52512_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_52512_axis_0 = const()[name = tensor("op_52512_axis_0"), val = tensor(1)]; + tensor var_52512_cast_fp16_0, tensor var_52512_cast_fp16_1 = split(axis = var_52512_axis_0, split_sizes = var_52512_split_sizes_0, x = var_52511_cast_fp16)[name = tensor("op_52512_cast_fp16")]; + tensor var_52514_mode_0 = const()[name = tensor("op_52514_mode_0"), val = tensor("EXACT")]; + tensor var_52514_cast_fp16 = gelu(mode = var_52514_mode_0, x = var_52512_cast_fp16_1)[name = tensor("op_52514_cast_fp16")]; + tensor input_665_cast_fp16 = mul(x = var_52512_cast_fp16_0, y = var_52514_cast_fp16)[name = tensor("input_665_cast_fp16")]; + tensor var_52518 = const()[name = tensor("op_52518"), val = tensor([1, 1])]; + tensor var_52520 = const()[name = tensor("op_52520"), val = tensor([1, 1])]; + tensor var_52522_pad_type_0 = const()[name = tensor("op_52522_pad_type_0"), val = tensor("custom")]; + tensor var_52522_pad_0 = const()[name = tensor("op_52522_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_3_ff_net_2_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_3_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4437032960)))]; + tensor up_blocks_0_attentions_2_transformer_blocks_3_ff_net_2_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_3_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4450140224)))]; + tensor var_52522_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_3_ff_net_2_bias_to_fp16, dilations = var_52520, groups = var_30355, pad = var_52522_pad_0, pad_type = var_52522_pad_type_0, strides = var_52518, weight = up_blocks_0_attentions_2_transformer_blocks_3_ff_net_2_weight_to_fp16, x = input_665_cast_fp16)[name = tensor("op_52522_cast_fp16")]; + tensor inputs_349_cast_fp16 = add(x = var_52522_cast_fp16, y = inputs_347_cast_fp16)[name = tensor("inputs_349_cast_fp16")]; + tensor hidden_states_459_axes_0 = const()[name = tensor("hidden_states_459_axes_0"), val = tensor([1])]; + tensor hidden_states_459_gamma_0_to_fp16 = const()[name = tensor("hidden_states_459_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4450142848)))]; + tensor hidden_states_459_beta_0_to_fp16 = const()[name = tensor("hidden_states_459_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4450145472)))]; + tensor var_52538_to_fp16 = const()[name = tensor("op_52538_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_459_cast_fp16 = layer_norm(axes = hidden_states_459_axes_0, beta = hidden_states_459_beta_0_to_fp16, epsilon = var_52538_to_fp16, gamma = hidden_states_459_gamma_0_to_fp16, x = inputs_349_cast_fp16)[name = tensor("hidden_states_459_cast_fp16")]; + tensor var_52553 = const()[name = tensor("op_52553"), val = tensor([1, 1])]; + tensor var_52555 = const()[name = tensor("op_52555"), val = tensor([1, 1])]; + tensor q_233_pad_type_0 = const()[name = tensor("q_233_pad_type_0"), val = tensor("custom")]; + tensor q_233_pad_0 = const()[name = tensor("q_233_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_4_attn1_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_4_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4450148096)))]; + tensor q_233_cast_fp16 = conv(dilations = var_52555, groups = var_30355, pad = q_233_pad_0, pad_type = q_233_pad_type_0, strides = var_52553, weight = up_blocks_0_attentions_2_transformer_blocks_4_attn1_to_q_weight_to_fp16, x = hidden_states_459_cast_fp16)[name = tensor("q_233_cast_fp16")]; + tensor var_52559 = const()[name = tensor("op_52559"), val = tensor([1, 1])]; + tensor var_52561 = const()[name = tensor("op_52561"), val = tensor([1, 1])]; + tensor k_465_pad_type_0 = const()[name = tensor("k_465_pad_type_0"), val = tensor("custom")]; + tensor k_465_pad_0 = const()[name = tensor("k_465_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_4_attn1_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_4_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4453424960)))]; + tensor k_465_cast_fp16 = conv(dilations = var_52561, groups = var_30355, pad = k_465_pad_0, pad_type = k_465_pad_type_0, strides = var_52559, weight = up_blocks_0_attentions_2_transformer_blocks_4_attn1_to_k_weight_to_fp16, x = hidden_states_459_cast_fp16)[name = tensor("k_465_cast_fp16")]; + tensor var_52565 = const()[name = tensor("op_52565"), val = tensor([1, 1])]; + tensor var_52567 = const()[name = tensor("op_52567"), val = tensor([1, 1])]; + tensor v_233_pad_type_0 = const()[name = tensor("v_233_pad_type_0"), val = tensor("custom")]; + tensor v_233_pad_0 = const()[name = tensor("v_233_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_4_attn1_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_4_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4456701824)))]; + tensor v_233_cast_fp16 = conv(dilations = var_52567, groups = var_30355, pad = v_233_pad_0, pad_type = v_233_pad_type_0, strides = var_52565, weight = up_blocks_0_attentions_2_transformer_blocks_4_attn1_to_v_weight_to_fp16, x = hidden_states_459_cast_fp16)[name = tensor("v_233_cast_fp16")]; + tensor var_52571_begin_0 = const()[name = tensor("op_52571_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_52571_end_0 = const()[name = tensor("op_52571_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_52571_end_mask_0 = const()[name = tensor("op_52571_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52571_cast_fp16 = slice_by_index(begin = var_52571_begin_0, end = var_52571_end_0, end_mask = var_52571_end_mask_0, x = q_233_cast_fp16)[name = tensor("op_52571_cast_fp16")]; + tensor var_52575_begin_0 = const()[name = tensor("op_52575_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_52575_end_0 = const()[name = tensor("op_52575_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_52575_end_mask_0 = const()[name = tensor("op_52575_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52575_cast_fp16 = slice_by_index(begin = var_52575_begin_0, end = var_52575_end_0, end_mask = var_52575_end_mask_0, x = q_233_cast_fp16)[name = tensor("op_52575_cast_fp16")]; + tensor var_52579_begin_0 = const()[name = tensor("op_52579_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_52579_end_0 = const()[name = tensor("op_52579_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_52579_end_mask_0 = const()[name = tensor("op_52579_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52579_cast_fp16 = slice_by_index(begin = var_52579_begin_0, end = var_52579_end_0, end_mask = var_52579_end_mask_0, x = q_233_cast_fp16)[name = tensor("op_52579_cast_fp16")]; + tensor var_52583_begin_0 = const()[name = tensor("op_52583_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_52583_end_0 = const()[name = tensor("op_52583_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_52583_end_mask_0 = const()[name = tensor("op_52583_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52583_cast_fp16 = slice_by_index(begin = var_52583_begin_0, end = var_52583_end_0, end_mask = var_52583_end_mask_0, x = q_233_cast_fp16)[name = tensor("op_52583_cast_fp16")]; + tensor var_52587_begin_0 = const()[name = tensor("op_52587_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_52587_end_0 = const()[name = tensor("op_52587_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_52587_end_mask_0 = const()[name = tensor("op_52587_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52587_cast_fp16 = slice_by_index(begin = var_52587_begin_0, end = var_52587_end_0, end_mask = var_52587_end_mask_0, x = q_233_cast_fp16)[name = tensor("op_52587_cast_fp16")]; + tensor var_52591_begin_0 = const()[name = tensor("op_52591_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_52591_end_0 = const()[name = tensor("op_52591_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_52591_end_mask_0 = const()[name = tensor("op_52591_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52591_cast_fp16 = slice_by_index(begin = var_52591_begin_0, end = var_52591_end_0, end_mask = var_52591_end_mask_0, x = q_233_cast_fp16)[name = tensor("op_52591_cast_fp16")]; + tensor var_52595_begin_0 = const()[name = tensor("op_52595_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_52595_end_0 = const()[name = tensor("op_52595_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_52595_end_mask_0 = const()[name = tensor("op_52595_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52595_cast_fp16 = slice_by_index(begin = var_52595_begin_0, end = var_52595_end_0, end_mask = var_52595_end_mask_0, x = q_233_cast_fp16)[name = tensor("op_52595_cast_fp16")]; + tensor var_52599_begin_0 = const()[name = tensor("op_52599_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_52599_end_0 = const()[name = tensor("op_52599_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_52599_end_mask_0 = const()[name = tensor("op_52599_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52599_cast_fp16 = slice_by_index(begin = var_52599_begin_0, end = var_52599_end_0, end_mask = var_52599_end_mask_0, x = q_233_cast_fp16)[name = tensor("op_52599_cast_fp16")]; + tensor var_52603_begin_0 = const()[name = tensor("op_52603_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_52603_end_0 = const()[name = tensor("op_52603_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_52603_end_mask_0 = const()[name = tensor("op_52603_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52603_cast_fp16 = slice_by_index(begin = var_52603_begin_0, end = var_52603_end_0, end_mask = var_52603_end_mask_0, x = q_233_cast_fp16)[name = tensor("op_52603_cast_fp16")]; + tensor var_52607_begin_0 = const()[name = tensor("op_52607_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_52607_end_0 = const()[name = tensor("op_52607_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_52607_end_mask_0 = const()[name = tensor("op_52607_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52607_cast_fp16 = slice_by_index(begin = var_52607_begin_0, end = var_52607_end_0, end_mask = var_52607_end_mask_0, x = q_233_cast_fp16)[name = tensor("op_52607_cast_fp16")]; + tensor var_52611_begin_0 = const()[name = tensor("op_52611_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_52611_end_0 = const()[name = tensor("op_52611_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_52611_end_mask_0 = const()[name = tensor("op_52611_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52611_cast_fp16 = slice_by_index(begin = var_52611_begin_0, end = var_52611_end_0, end_mask = var_52611_end_mask_0, x = q_233_cast_fp16)[name = tensor("op_52611_cast_fp16")]; + tensor var_52615_begin_0 = const()[name = tensor("op_52615_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_52615_end_0 = const()[name = tensor("op_52615_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_52615_end_mask_0 = const()[name = tensor("op_52615_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52615_cast_fp16 = slice_by_index(begin = var_52615_begin_0, end = var_52615_end_0, end_mask = var_52615_end_mask_0, x = q_233_cast_fp16)[name = tensor("op_52615_cast_fp16")]; + tensor var_52619_begin_0 = const()[name = tensor("op_52619_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_52619_end_0 = const()[name = tensor("op_52619_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_52619_end_mask_0 = const()[name = tensor("op_52619_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52619_cast_fp16 = slice_by_index(begin = var_52619_begin_0, end = var_52619_end_0, end_mask = var_52619_end_mask_0, x = q_233_cast_fp16)[name = tensor("op_52619_cast_fp16")]; + tensor var_52623_begin_0 = const()[name = tensor("op_52623_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_52623_end_0 = const()[name = tensor("op_52623_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_52623_end_mask_0 = const()[name = tensor("op_52623_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52623_cast_fp16 = slice_by_index(begin = var_52623_begin_0, end = var_52623_end_0, end_mask = var_52623_end_mask_0, x = q_233_cast_fp16)[name = tensor("op_52623_cast_fp16")]; + tensor var_52627_begin_0 = const()[name = tensor("op_52627_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_52627_end_0 = const()[name = tensor("op_52627_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_52627_end_mask_0 = const()[name = tensor("op_52627_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52627_cast_fp16 = slice_by_index(begin = var_52627_begin_0, end = var_52627_end_0, end_mask = var_52627_end_mask_0, x = q_233_cast_fp16)[name = tensor("op_52627_cast_fp16")]; + tensor var_52631_begin_0 = const()[name = tensor("op_52631_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_52631_end_0 = const()[name = tensor("op_52631_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_52631_end_mask_0 = const()[name = tensor("op_52631_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52631_cast_fp16 = slice_by_index(begin = var_52631_begin_0, end = var_52631_end_0, end_mask = var_52631_end_mask_0, x = q_233_cast_fp16)[name = tensor("op_52631_cast_fp16")]; + tensor var_52635_begin_0 = const()[name = tensor("op_52635_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_52635_end_0 = const()[name = tensor("op_52635_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_52635_end_mask_0 = const()[name = tensor("op_52635_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52635_cast_fp16 = slice_by_index(begin = var_52635_begin_0, end = var_52635_end_0, end_mask = var_52635_end_mask_0, x = q_233_cast_fp16)[name = tensor("op_52635_cast_fp16")]; + tensor var_52639_begin_0 = const()[name = tensor("op_52639_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_52639_end_0 = const()[name = tensor("op_52639_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_52639_end_mask_0 = const()[name = tensor("op_52639_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52639_cast_fp16 = slice_by_index(begin = var_52639_begin_0, end = var_52639_end_0, end_mask = var_52639_end_mask_0, x = q_233_cast_fp16)[name = tensor("op_52639_cast_fp16")]; + tensor var_52643_begin_0 = const()[name = tensor("op_52643_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_52643_end_0 = const()[name = tensor("op_52643_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_52643_end_mask_0 = const()[name = tensor("op_52643_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52643_cast_fp16 = slice_by_index(begin = var_52643_begin_0, end = var_52643_end_0, end_mask = var_52643_end_mask_0, x = q_233_cast_fp16)[name = tensor("op_52643_cast_fp16")]; + tensor var_52647_begin_0 = const()[name = tensor("op_52647_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_52647_end_0 = const()[name = tensor("op_52647_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_52647_end_mask_0 = const()[name = tensor("op_52647_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52647_cast_fp16 = slice_by_index(begin = var_52647_begin_0, end = var_52647_end_0, end_mask = var_52647_end_mask_0, x = q_233_cast_fp16)[name = tensor("op_52647_cast_fp16")]; + tensor k_467_perm_0 = const()[name = tensor("k_467_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_52654_begin_0 = const()[name = tensor("op_52654_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_52654_end_0 = const()[name = tensor("op_52654_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_52654_end_mask_0 = const()[name = tensor("op_52654_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_467_cast_fp16 = transpose(perm = k_467_perm_0, x = k_465_cast_fp16)[name = tensor("transpose_23")]; + tensor var_52654_cast_fp16 = slice_by_index(begin = var_52654_begin_0, end = var_52654_end_0, end_mask = var_52654_end_mask_0, x = k_467_cast_fp16)[name = tensor("op_52654_cast_fp16")]; + tensor var_52658_begin_0 = const()[name = tensor("op_52658_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_52658_end_0 = const()[name = tensor("op_52658_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_52658_end_mask_0 = const()[name = tensor("op_52658_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_52658_cast_fp16 = slice_by_index(begin = var_52658_begin_0, end = var_52658_end_0, end_mask = var_52658_end_mask_0, x = k_467_cast_fp16)[name = tensor("op_52658_cast_fp16")]; + tensor var_52662_begin_0 = const()[name = tensor("op_52662_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_52662_end_0 = const()[name = tensor("op_52662_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_52662_end_mask_0 = const()[name = tensor("op_52662_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_52662_cast_fp16 = slice_by_index(begin = var_52662_begin_0, end = var_52662_end_0, end_mask = var_52662_end_mask_0, x = k_467_cast_fp16)[name = tensor("op_52662_cast_fp16")]; + tensor var_52666_begin_0 = const()[name = tensor("op_52666_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_52666_end_0 = const()[name = tensor("op_52666_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_52666_end_mask_0 = const()[name = tensor("op_52666_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_52666_cast_fp16 = slice_by_index(begin = var_52666_begin_0, end = var_52666_end_0, end_mask = var_52666_end_mask_0, x = k_467_cast_fp16)[name = tensor("op_52666_cast_fp16")]; + tensor var_52670_begin_0 = const()[name = tensor("op_52670_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_52670_end_0 = const()[name = tensor("op_52670_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_52670_end_mask_0 = const()[name = tensor("op_52670_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_52670_cast_fp16 = slice_by_index(begin = var_52670_begin_0, end = var_52670_end_0, end_mask = var_52670_end_mask_0, x = k_467_cast_fp16)[name = tensor("op_52670_cast_fp16")]; + tensor var_52674_begin_0 = const()[name = tensor("op_52674_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_52674_end_0 = const()[name = tensor("op_52674_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_52674_end_mask_0 = const()[name = tensor("op_52674_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_52674_cast_fp16 = slice_by_index(begin = var_52674_begin_0, end = var_52674_end_0, end_mask = var_52674_end_mask_0, x = k_467_cast_fp16)[name = tensor("op_52674_cast_fp16")]; + tensor var_52678_begin_0 = const()[name = tensor("op_52678_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_52678_end_0 = const()[name = tensor("op_52678_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_52678_end_mask_0 = const()[name = tensor("op_52678_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_52678_cast_fp16 = slice_by_index(begin = var_52678_begin_0, end = var_52678_end_0, end_mask = var_52678_end_mask_0, x = k_467_cast_fp16)[name = tensor("op_52678_cast_fp16")]; + tensor var_52682_begin_0 = const()[name = tensor("op_52682_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_52682_end_0 = const()[name = tensor("op_52682_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_52682_end_mask_0 = const()[name = tensor("op_52682_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_52682_cast_fp16 = slice_by_index(begin = var_52682_begin_0, end = var_52682_end_0, end_mask = var_52682_end_mask_0, x = k_467_cast_fp16)[name = tensor("op_52682_cast_fp16")]; + tensor var_52686_begin_0 = const()[name = tensor("op_52686_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_52686_end_0 = const()[name = tensor("op_52686_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_52686_end_mask_0 = const()[name = tensor("op_52686_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_52686_cast_fp16 = slice_by_index(begin = var_52686_begin_0, end = var_52686_end_0, end_mask = var_52686_end_mask_0, x = k_467_cast_fp16)[name = tensor("op_52686_cast_fp16")]; + tensor var_52690_begin_0 = const()[name = tensor("op_52690_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_52690_end_0 = const()[name = tensor("op_52690_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_52690_end_mask_0 = const()[name = tensor("op_52690_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_52690_cast_fp16 = slice_by_index(begin = var_52690_begin_0, end = var_52690_end_0, end_mask = var_52690_end_mask_0, x = k_467_cast_fp16)[name = tensor("op_52690_cast_fp16")]; + tensor var_52694_begin_0 = const()[name = tensor("op_52694_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_52694_end_0 = const()[name = tensor("op_52694_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_52694_end_mask_0 = const()[name = tensor("op_52694_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_52694_cast_fp16 = slice_by_index(begin = var_52694_begin_0, end = var_52694_end_0, end_mask = var_52694_end_mask_0, x = k_467_cast_fp16)[name = tensor("op_52694_cast_fp16")]; + tensor var_52698_begin_0 = const()[name = tensor("op_52698_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_52698_end_0 = const()[name = tensor("op_52698_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_52698_end_mask_0 = const()[name = tensor("op_52698_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_52698_cast_fp16 = slice_by_index(begin = var_52698_begin_0, end = var_52698_end_0, end_mask = var_52698_end_mask_0, x = k_467_cast_fp16)[name = tensor("op_52698_cast_fp16")]; + tensor var_52702_begin_0 = const()[name = tensor("op_52702_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_52702_end_0 = const()[name = tensor("op_52702_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_52702_end_mask_0 = const()[name = tensor("op_52702_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_52702_cast_fp16 = slice_by_index(begin = var_52702_begin_0, end = var_52702_end_0, end_mask = var_52702_end_mask_0, x = k_467_cast_fp16)[name = tensor("op_52702_cast_fp16")]; + tensor var_52706_begin_0 = const()[name = tensor("op_52706_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_52706_end_0 = const()[name = tensor("op_52706_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_52706_end_mask_0 = const()[name = tensor("op_52706_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_52706_cast_fp16 = slice_by_index(begin = var_52706_begin_0, end = var_52706_end_0, end_mask = var_52706_end_mask_0, x = k_467_cast_fp16)[name = tensor("op_52706_cast_fp16")]; + tensor var_52710_begin_0 = const()[name = tensor("op_52710_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_52710_end_0 = const()[name = tensor("op_52710_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_52710_end_mask_0 = const()[name = tensor("op_52710_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_52710_cast_fp16 = slice_by_index(begin = var_52710_begin_0, end = var_52710_end_0, end_mask = var_52710_end_mask_0, x = k_467_cast_fp16)[name = tensor("op_52710_cast_fp16")]; + tensor var_52714_begin_0 = const()[name = tensor("op_52714_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_52714_end_0 = const()[name = tensor("op_52714_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_52714_end_mask_0 = const()[name = tensor("op_52714_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_52714_cast_fp16 = slice_by_index(begin = var_52714_begin_0, end = var_52714_end_0, end_mask = var_52714_end_mask_0, x = k_467_cast_fp16)[name = tensor("op_52714_cast_fp16")]; + tensor var_52718_begin_0 = const()[name = tensor("op_52718_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_52718_end_0 = const()[name = tensor("op_52718_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_52718_end_mask_0 = const()[name = tensor("op_52718_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_52718_cast_fp16 = slice_by_index(begin = var_52718_begin_0, end = var_52718_end_0, end_mask = var_52718_end_mask_0, x = k_467_cast_fp16)[name = tensor("op_52718_cast_fp16")]; + tensor var_52722_begin_0 = const()[name = tensor("op_52722_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_52722_end_0 = const()[name = tensor("op_52722_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_52722_end_mask_0 = const()[name = tensor("op_52722_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_52722_cast_fp16 = slice_by_index(begin = var_52722_begin_0, end = var_52722_end_0, end_mask = var_52722_end_mask_0, x = k_467_cast_fp16)[name = tensor("op_52722_cast_fp16")]; + tensor var_52726_begin_0 = const()[name = tensor("op_52726_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_52726_end_0 = const()[name = tensor("op_52726_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_52726_end_mask_0 = const()[name = tensor("op_52726_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_52726_cast_fp16 = slice_by_index(begin = var_52726_begin_0, end = var_52726_end_0, end_mask = var_52726_end_mask_0, x = k_467_cast_fp16)[name = tensor("op_52726_cast_fp16")]; + tensor var_52730_begin_0 = const()[name = tensor("op_52730_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_52730_end_0 = const()[name = tensor("op_52730_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_52730_end_mask_0 = const()[name = tensor("op_52730_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_52730_cast_fp16 = slice_by_index(begin = var_52730_begin_0, end = var_52730_end_0, end_mask = var_52730_end_mask_0, x = k_467_cast_fp16)[name = tensor("op_52730_cast_fp16")]; + tensor var_52732_begin_0 = const()[name = tensor("op_52732_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_52732_end_0 = const()[name = tensor("op_52732_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_52732_end_mask_0 = const()[name = tensor("op_52732_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52732_cast_fp16 = slice_by_index(begin = var_52732_begin_0, end = var_52732_end_0, end_mask = var_52732_end_mask_0, x = v_233_cast_fp16)[name = tensor("op_52732_cast_fp16")]; + tensor var_52736_begin_0 = const()[name = tensor("op_52736_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_52736_end_0 = const()[name = tensor("op_52736_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_52736_end_mask_0 = const()[name = tensor("op_52736_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52736_cast_fp16 = slice_by_index(begin = var_52736_begin_0, end = var_52736_end_0, end_mask = var_52736_end_mask_0, x = v_233_cast_fp16)[name = tensor("op_52736_cast_fp16")]; + tensor var_52740_begin_0 = const()[name = tensor("op_52740_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_52740_end_0 = const()[name = tensor("op_52740_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_52740_end_mask_0 = const()[name = tensor("op_52740_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52740_cast_fp16 = slice_by_index(begin = var_52740_begin_0, end = var_52740_end_0, end_mask = var_52740_end_mask_0, x = v_233_cast_fp16)[name = tensor("op_52740_cast_fp16")]; + tensor var_52744_begin_0 = const()[name = tensor("op_52744_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_52744_end_0 = const()[name = tensor("op_52744_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_52744_end_mask_0 = const()[name = tensor("op_52744_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52744_cast_fp16 = slice_by_index(begin = var_52744_begin_0, end = var_52744_end_0, end_mask = var_52744_end_mask_0, x = v_233_cast_fp16)[name = tensor("op_52744_cast_fp16")]; + tensor var_52748_begin_0 = const()[name = tensor("op_52748_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_52748_end_0 = const()[name = tensor("op_52748_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_52748_end_mask_0 = const()[name = tensor("op_52748_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52748_cast_fp16 = slice_by_index(begin = var_52748_begin_0, end = var_52748_end_0, end_mask = var_52748_end_mask_0, x = v_233_cast_fp16)[name = tensor("op_52748_cast_fp16")]; + tensor var_52752_begin_0 = const()[name = tensor("op_52752_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_52752_end_0 = const()[name = tensor("op_52752_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_52752_end_mask_0 = const()[name = tensor("op_52752_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52752_cast_fp16 = slice_by_index(begin = var_52752_begin_0, end = var_52752_end_0, end_mask = var_52752_end_mask_0, x = v_233_cast_fp16)[name = tensor("op_52752_cast_fp16")]; + tensor var_52756_begin_0 = const()[name = tensor("op_52756_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_52756_end_0 = const()[name = tensor("op_52756_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_52756_end_mask_0 = const()[name = tensor("op_52756_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52756_cast_fp16 = slice_by_index(begin = var_52756_begin_0, end = var_52756_end_0, end_mask = var_52756_end_mask_0, x = v_233_cast_fp16)[name = tensor("op_52756_cast_fp16")]; + tensor var_52760_begin_0 = const()[name = tensor("op_52760_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_52760_end_0 = const()[name = tensor("op_52760_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_52760_end_mask_0 = const()[name = tensor("op_52760_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52760_cast_fp16 = slice_by_index(begin = var_52760_begin_0, end = var_52760_end_0, end_mask = var_52760_end_mask_0, x = v_233_cast_fp16)[name = tensor("op_52760_cast_fp16")]; + tensor var_52764_begin_0 = const()[name = tensor("op_52764_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_52764_end_0 = const()[name = tensor("op_52764_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_52764_end_mask_0 = const()[name = tensor("op_52764_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52764_cast_fp16 = slice_by_index(begin = var_52764_begin_0, end = var_52764_end_0, end_mask = var_52764_end_mask_0, x = v_233_cast_fp16)[name = tensor("op_52764_cast_fp16")]; + tensor var_52768_begin_0 = const()[name = tensor("op_52768_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_52768_end_0 = const()[name = tensor("op_52768_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_52768_end_mask_0 = const()[name = tensor("op_52768_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52768_cast_fp16 = slice_by_index(begin = var_52768_begin_0, end = var_52768_end_0, end_mask = var_52768_end_mask_0, x = v_233_cast_fp16)[name = tensor("op_52768_cast_fp16")]; + tensor var_52772_begin_0 = const()[name = tensor("op_52772_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_52772_end_0 = const()[name = tensor("op_52772_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_52772_end_mask_0 = const()[name = tensor("op_52772_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52772_cast_fp16 = slice_by_index(begin = var_52772_begin_0, end = var_52772_end_0, end_mask = var_52772_end_mask_0, x = v_233_cast_fp16)[name = tensor("op_52772_cast_fp16")]; + tensor var_52776_begin_0 = const()[name = tensor("op_52776_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_52776_end_0 = const()[name = tensor("op_52776_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_52776_end_mask_0 = const()[name = tensor("op_52776_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52776_cast_fp16 = slice_by_index(begin = var_52776_begin_0, end = var_52776_end_0, end_mask = var_52776_end_mask_0, x = v_233_cast_fp16)[name = tensor("op_52776_cast_fp16")]; + tensor var_52780_begin_0 = const()[name = tensor("op_52780_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_52780_end_0 = const()[name = tensor("op_52780_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_52780_end_mask_0 = const()[name = tensor("op_52780_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52780_cast_fp16 = slice_by_index(begin = var_52780_begin_0, end = var_52780_end_0, end_mask = var_52780_end_mask_0, x = v_233_cast_fp16)[name = tensor("op_52780_cast_fp16")]; + tensor var_52784_begin_0 = const()[name = tensor("op_52784_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_52784_end_0 = const()[name = tensor("op_52784_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_52784_end_mask_0 = const()[name = tensor("op_52784_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52784_cast_fp16 = slice_by_index(begin = var_52784_begin_0, end = var_52784_end_0, end_mask = var_52784_end_mask_0, x = v_233_cast_fp16)[name = tensor("op_52784_cast_fp16")]; + tensor var_52788_begin_0 = const()[name = tensor("op_52788_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_52788_end_0 = const()[name = tensor("op_52788_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_52788_end_mask_0 = const()[name = tensor("op_52788_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52788_cast_fp16 = slice_by_index(begin = var_52788_begin_0, end = var_52788_end_0, end_mask = var_52788_end_mask_0, x = v_233_cast_fp16)[name = tensor("op_52788_cast_fp16")]; + tensor var_52792_begin_0 = const()[name = tensor("op_52792_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_52792_end_0 = const()[name = tensor("op_52792_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_52792_end_mask_0 = const()[name = tensor("op_52792_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52792_cast_fp16 = slice_by_index(begin = var_52792_begin_0, end = var_52792_end_0, end_mask = var_52792_end_mask_0, x = v_233_cast_fp16)[name = tensor("op_52792_cast_fp16")]; + tensor var_52796_begin_0 = const()[name = tensor("op_52796_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_52796_end_0 = const()[name = tensor("op_52796_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_52796_end_mask_0 = const()[name = tensor("op_52796_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52796_cast_fp16 = slice_by_index(begin = var_52796_begin_0, end = var_52796_end_0, end_mask = var_52796_end_mask_0, x = v_233_cast_fp16)[name = tensor("op_52796_cast_fp16")]; + tensor var_52800_begin_0 = const()[name = tensor("op_52800_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_52800_end_0 = const()[name = tensor("op_52800_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_52800_end_mask_0 = const()[name = tensor("op_52800_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52800_cast_fp16 = slice_by_index(begin = var_52800_begin_0, end = var_52800_end_0, end_mask = var_52800_end_mask_0, x = v_233_cast_fp16)[name = tensor("op_52800_cast_fp16")]; + tensor var_52804_begin_0 = const()[name = tensor("op_52804_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_52804_end_0 = const()[name = tensor("op_52804_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_52804_end_mask_0 = const()[name = tensor("op_52804_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52804_cast_fp16 = slice_by_index(begin = var_52804_begin_0, end = var_52804_end_0, end_mask = var_52804_end_mask_0, x = v_233_cast_fp16)[name = tensor("op_52804_cast_fp16")]; + tensor var_52808_begin_0 = const()[name = tensor("op_52808_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_52808_end_0 = const()[name = tensor("op_52808_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_52808_end_mask_0 = const()[name = tensor("op_52808_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_52808_cast_fp16 = slice_by_index(begin = var_52808_begin_0, end = var_52808_end_0, end_mask = var_52808_end_mask_0, x = v_233_cast_fp16)[name = tensor("op_52808_cast_fp16")]; + tensor var_52812_equation_0 = const()[name = tensor("op_52812_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52812_cast_fp16 = einsum(equation = var_52812_equation_0, values = (var_52654_cast_fp16, var_52571_cast_fp16))[name = tensor("op_52812_cast_fp16")]; + tensor var_52813_to_fp16 = const()[name = tensor("op_52813_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4481_cast_fp16 = mul(x = var_52812_cast_fp16, y = var_52813_to_fp16)[name = tensor("aw_4481_cast_fp16")]; + tensor var_52816_equation_0 = const()[name = tensor("op_52816_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52816_cast_fp16 = einsum(equation = var_52816_equation_0, values = (var_52658_cast_fp16, var_52575_cast_fp16))[name = tensor("op_52816_cast_fp16")]; + tensor var_52817_to_fp16 = const()[name = tensor("op_52817_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4483_cast_fp16 = mul(x = var_52816_cast_fp16, y = var_52817_to_fp16)[name = tensor("aw_4483_cast_fp16")]; + tensor var_52820_equation_0 = const()[name = tensor("op_52820_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52820_cast_fp16 = einsum(equation = var_52820_equation_0, values = (var_52662_cast_fp16, var_52579_cast_fp16))[name = tensor("op_52820_cast_fp16")]; + tensor var_52821_to_fp16 = const()[name = tensor("op_52821_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4485_cast_fp16 = mul(x = var_52820_cast_fp16, y = var_52821_to_fp16)[name = tensor("aw_4485_cast_fp16")]; + tensor var_52824_equation_0 = const()[name = tensor("op_52824_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52824_cast_fp16 = einsum(equation = var_52824_equation_0, values = (var_52666_cast_fp16, var_52583_cast_fp16))[name = tensor("op_52824_cast_fp16")]; + tensor var_52825_to_fp16 = const()[name = tensor("op_52825_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4487_cast_fp16 = mul(x = var_52824_cast_fp16, y = var_52825_to_fp16)[name = tensor("aw_4487_cast_fp16")]; + tensor var_52828_equation_0 = const()[name = tensor("op_52828_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52828_cast_fp16 = einsum(equation = var_52828_equation_0, values = (var_52670_cast_fp16, var_52587_cast_fp16))[name = tensor("op_52828_cast_fp16")]; + tensor var_52829_to_fp16 = const()[name = tensor("op_52829_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4489_cast_fp16 = mul(x = var_52828_cast_fp16, y = var_52829_to_fp16)[name = tensor("aw_4489_cast_fp16")]; + tensor var_52832_equation_0 = const()[name = tensor("op_52832_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52832_cast_fp16 = einsum(equation = var_52832_equation_0, values = (var_52674_cast_fp16, var_52591_cast_fp16))[name = tensor("op_52832_cast_fp16")]; + tensor var_52833_to_fp16 = const()[name = tensor("op_52833_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4491_cast_fp16 = mul(x = var_52832_cast_fp16, y = var_52833_to_fp16)[name = tensor("aw_4491_cast_fp16")]; + tensor var_52836_equation_0 = const()[name = tensor("op_52836_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52836_cast_fp16 = einsum(equation = var_52836_equation_0, values = (var_52678_cast_fp16, var_52595_cast_fp16))[name = tensor("op_52836_cast_fp16")]; + tensor var_52837_to_fp16 = const()[name = tensor("op_52837_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4493_cast_fp16 = mul(x = var_52836_cast_fp16, y = var_52837_to_fp16)[name = tensor("aw_4493_cast_fp16")]; + tensor var_52840_equation_0 = const()[name = tensor("op_52840_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52840_cast_fp16 = einsum(equation = var_52840_equation_0, values = (var_52682_cast_fp16, var_52599_cast_fp16))[name = tensor("op_52840_cast_fp16")]; + tensor var_52841_to_fp16 = const()[name = tensor("op_52841_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4495_cast_fp16 = mul(x = var_52840_cast_fp16, y = var_52841_to_fp16)[name = tensor("aw_4495_cast_fp16")]; + tensor var_52844_equation_0 = const()[name = tensor("op_52844_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52844_cast_fp16 = einsum(equation = var_52844_equation_0, values = (var_52686_cast_fp16, var_52603_cast_fp16))[name = tensor("op_52844_cast_fp16")]; + tensor var_52845_to_fp16 = const()[name = tensor("op_52845_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4497_cast_fp16 = mul(x = var_52844_cast_fp16, y = var_52845_to_fp16)[name = tensor("aw_4497_cast_fp16")]; + tensor var_52848_equation_0 = const()[name = tensor("op_52848_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52848_cast_fp16 = einsum(equation = var_52848_equation_0, values = (var_52690_cast_fp16, var_52607_cast_fp16))[name = tensor("op_52848_cast_fp16")]; + tensor var_52849_to_fp16 = const()[name = tensor("op_52849_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4499_cast_fp16 = mul(x = var_52848_cast_fp16, y = var_52849_to_fp16)[name = tensor("aw_4499_cast_fp16")]; + tensor var_52852_equation_0 = const()[name = tensor("op_52852_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52852_cast_fp16 = einsum(equation = var_52852_equation_0, values = (var_52694_cast_fp16, var_52611_cast_fp16))[name = tensor("op_52852_cast_fp16")]; + tensor var_52853_to_fp16 = const()[name = tensor("op_52853_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4501_cast_fp16 = mul(x = var_52852_cast_fp16, y = var_52853_to_fp16)[name = tensor("aw_4501_cast_fp16")]; + tensor var_52856_equation_0 = const()[name = tensor("op_52856_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52856_cast_fp16 = einsum(equation = var_52856_equation_0, values = (var_52698_cast_fp16, var_52615_cast_fp16))[name = tensor("op_52856_cast_fp16")]; + tensor var_52857_to_fp16 = const()[name = tensor("op_52857_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4503_cast_fp16 = mul(x = var_52856_cast_fp16, y = var_52857_to_fp16)[name = tensor("aw_4503_cast_fp16")]; + tensor var_52860_equation_0 = const()[name = tensor("op_52860_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52860_cast_fp16 = einsum(equation = var_52860_equation_0, values = (var_52702_cast_fp16, var_52619_cast_fp16))[name = tensor("op_52860_cast_fp16")]; + tensor var_52861_to_fp16 = const()[name = tensor("op_52861_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4505_cast_fp16 = mul(x = var_52860_cast_fp16, y = var_52861_to_fp16)[name = tensor("aw_4505_cast_fp16")]; + tensor var_52864_equation_0 = const()[name = tensor("op_52864_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52864_cast_fp16 = einsum(equation = var_52864_equation_0, values = (var_52706_cast_fp16, var_52623_cast_fp16))[name = tensor("op_52864_cast_fp16")]; + tensor var_52865_to_fp16 = const()[name = tensor("op_52865_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4507_cast_fp16 = mul(x = var_52864_cast_fp16, y = var_52865_to_fp16)[name = tensor("aw_4507_cast_fp16")]; + tensor var_52868_equation_0 = const()[name = tensor("op_52868_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52868_cast_fp16 = einsum(equation = var_52868_equation_0, values = (var_52710_cast_fp16, var_52627_cast_fp16))[name = tensor("op_52868_cast_fp16")]; + tensor var_52869_to_fp16 = const()[name = tensor("op_52869_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4509_cast_fp16 = mul(x = var_52868_cast_fp16, y = var_52869_to_fp16)[name = tensor("aw_4509_cast_fp16")]; + tensor var_52872_equation_0 = const()[name = tensor("op_52872_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52872_cast_fp16 = einsum(equation = var_52872_equation_0, values = (var_52714_cast_fp16, var_52631_cast_fp16))[name = tensor("op_52872_cast_fp16")]; + tensor var_52873_to_fp16 = const()[name = tensor("op_52873_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4511_cast_fp16 = mul(x = var_52872_cast_fp16, y = var_52873_to_fp16)[name = tensor("aw_4511_cast_fp16")]; + tensor var_52876_equation_0 = const()[name = tensor("op_52876_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52876_cast_fp16 = einsum(equation = var_52876_equation_0, values = (var_52718_cast_fp16, var_52635_cast_fp16))[name = tensor("op_52876_cast_fp16")]; + tensor var_52877_to_fp16 = const()[name = tensor("op_52877_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4513_cast_fp16 = mul(x = var_52876_cast_fp16, y = var_52877_to_fp16)[name = tensor("aw_4513_cast_fp16")]; + tensor var_52880_equation_0 = const()[name = tensor("op_52880_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52880_cast_fp16 = einsum(equation = var_52880_equation_0, values = (var_52722_cast_fp16, var_52639_cast_fp16))[name = tensor("op_52880_cast_fp16")]; + tensor var_52881_to_fp16 = const()[name = tensor("op_52881_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4515_cast_fp16 = mul(x = var_52880_cast_fp16, y = var_52881_to_fp16)[name = tensor("aw_4515_cast_fp16")]; + tensor var_52884_equation_0 = const()[name = tensor("op_52884_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52884_cast_fp16 = einsum(equation = var_52884_equation_0, values = (var_52726_cast_fp16, var_52643_cast_fp16))[name = tensor("op_52884_cast_fp16")]; + tensor var_52885_to_fp16 = const()[name = tensor("op_52885_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4517_cast_fp16 = mul(x = var_52884_cast_fp16, y = var_52885_to_fp16)[name = tensor("aw_4517_cast_fp16")]; + tensor var_52888_equation_0 = const()[name = tensor("op_52888_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52888_cast_fp16 = einsum(equation = var_52888_equation_0, values = (var_52730_cast_fp16, var_52647_cast_fp16))[name = tensor("op_52888_cast_fp16")]; + tensor var_52889_to_fp16 = const()[name = tensor("op_52889_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4519_cast_fp16 = mul(x = var_52888_cast_fp16, y = var_52889_to_fp16)[name = tensor("aw_4519_cast_fp16")]; + tensor var_52891_cast_fp16 = softmax(axis = var_30355, x = aw_4481_cast_fp16)[name = tensor("op_52891_cast_fp16")]; + tensor var_52892_cast_fp16 = softmax(axis = var_30355, x = aw_4483_cast_fp16)[name = tensor("op_52892_cast_fp16")]; + tensor var_52893_cast_fp16 = softmax(axis = var_30355, x = aw_4485_cast_fp16)[name = tensor("op_52893_cast_fp16")]; + tensor var_52894_cast_fp16 = softmax(axis = var_30355, x = aw_4487_cast_fp16)[name = tensor("op_52894_cast_fp16")]; + tensor var_52895_cast_fp16 = softmax(axis = var_30355, x = aw_4489_cast_fp16)[name = tensor("op_52895_cast_fp16")]; + tensor var_52896_cast_fp16 = softmax(axis = var_30355, x = aw_4491_cast_fp16)[name = tensor("op_52896_cast_fp16")]; + tensor var_52897_cast_fp16 = softmax(axis = var_30355, x = aw_4493_cast_fp16)[name = tensor("op_52897_cast_fp16")]; + tensor var_52898_cast_fp16 = softmax(axis = var_30355, x = aw_4495_cast_fp16)[name = tensor("op_52898_cast_fp16")]; + tensor var_52899_cast_fp16 = softmax(axis = var_30355, x = aw_4497_cast_fp16)[name = tensor("op_52899_cast_fp16")]; + tensor var_52900_cast_fp16 = softmax(axis = var_30355, x = aw_4499_cast_fp16)[name = tensor("op_52900_cast_fp16")]; + tensor var_52901_cast_fp16 = softmax(axis = var_30355, x = aw_4501_cast_fp16)[name = tensor("op_52901_cast_fp16")]; + tensor var_52902_cast_fp16 = softmax(axis = var_30355, x = aw_4503_cast_fp16)[name = tensor("op_52902_cast_fp16")]; + tensor var_52903_cast_fp16 = softmax(axis = var_30355, x = aw_4505_cast_fp16)[name = tensor("op_52903_cast_fp16")]; + tensor var_52904_cast_fp16 = softmax(axis = var_30355, x = aw_4507_cast_fp16)[name = tensor("op_52904_cast_fp16")]; + tensor var_52905_cast_fp16 = softmax(axis = var_30355, x = aw_4509_cast_fp16)[name = tensor("op_52905_cast_fp16")]; + tensor var_52906_cast_fp16 = softmax(axis = var_30355, x = aw_4511_cast_fp16)[name = tensor("op_52906_cast_fp16")]; + tensor var_52907_cast_fp16 = softmax(axis = var_30355, x = aw_4513_cast_fp16)[name = tensor("op_52907_cast_fp16")]; + tensor var_52908_cast_fp16 = softmax(axis = var_30355, x = aw_4515_cast_fp16)[name = tensor("op_52908_cast_fp16")]; + tensor var_52909_cast_fp16 = softmax(axis = var_30355, x = aw_4517_cast_fp16)[name = tensor("op_52909_cast_fp16")]; + tensor var_52910_cast_fp16 = softmax(axis = var_30355, x = aw_4519_cast_fp16)[name = tensor("op_52910_cast_fp16")]; + tensor var_52912_equation_0 = const()[name = tensor("op_52912_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52912_cast_fp16 = einsum(equation = var_52912_equation_0, values = (var_52732_cast_fp16, var_52891_cast_fp16))[name = tensor("op_52912_cast_fp16")]; + tensor var_52914_equation_0 = const()[name = tensor("op_52914_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52914_cast_fp16 = einsum(equation = var_52914_equation_0, values = (var_52736_cast_fp16, var_52892_cast_fp16))[name = tensor("op_52914_cast_fp16")]; + tensor var_52916_equation_0 = const()[name = tensor("op_52916_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52916_cast_fp16 = einsum(equation = var_52916_equation_0, values = (var_52740_cast_fp16, var_52893_cast_fp16))[name = tensor("op_52916_cast_fp16")]; + tensor var_52918_equation_0 = const()[name = tensor("op_52918_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52918_cast_fp16 = einsum(equation = var_52918_equation_0, values = (var_52744_cast_fp16, var_52894_cast_fp16))[name = tensor("op_52918_cast_fp16")]; + tensor var_52920_equation_0 = const()[name = tensor("op_52920_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52920_cast_fp16 = einsum(equation = var_52920_equation_0, values = (var_52748_cast_fp16, var_52895_cast_fp16))[name = tensor("op_52920_cast_fp16")]; + tensor var_52922_equation_0 = const()[name = tensor("op_52922_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52922_cast_fp16 = einsum(equation = var_52922_equation_0, values = (var_52752_cast_fp16, var_52896_cast_fp16))[name = tensor("op_52922_cast_fp16")]; + tensor var_52924_equation_0 = const()[name = tensor("op_52924_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52924_cast_fp16 = einsum(equation = var_52924_equation_0, values = (var_52756_cast_fp16, var_52897_cast_fp16))[name = tensor("op_52924_cast_fp16")]; + tensor var_52926_equation_0 = const()[name = tensor("op_52926_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52926_cast_fp16 = einsum(equation = var_52926_equation_0, values = (var_52760_cast_fp16, var_52898_cast_fp16))[name = tensor("op_52926_cast_fp16")]; + tensor var_52928_equation_0 = const()[name = tensor("op_52928_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52928_cast_fp16 = einsum(equation = var_52928_equation_0, values = (var_52764_cast_fp16, var_52899_cast_fp16))[name = tensor("op_52928_cast_fp16")]; + tensor var_52930_equation_0 = const()[name = tensor("op_52930_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52930_cast_fp16 = einsum(equation = var_52930_equation_0, values = (var_52768_cast_fp16, var_52900_cast_fp16))[name = tensor("op_52930_cast_fp16")]; + tensor var_52932_equation_0 = const()[name = tensor("op_52932_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52932_cast_fp16 = einsum(equation = var_52932_equation_0, values = (var_52772_cast_fp16, var_52901_cast_fp16))[name = tensor("op_52932_cast_fp16")]; + tensor var_52934_equation_0 = const()[name = tensor("op_52934_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52934_cast_fp16 = einsum(equation = var_52934_equation_0, values = (var_52776_cast_fp16, var_52902_cast_fp16))[name = tensor("op_52934_cast_fp16")]; + tensor var_52936_equation_0 = const()[name = tensor("op_52936_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52936_cast_fp16 = einsum(equation = var_52936_equation_0, values = (var_52780_cast_fp16, var_52903_cast_fp16))[name = tensor("op_52936_cast_fp16")]; + tensor var_52938_equation_0 = const()[name = tensor("op_52938_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52938_cast_fp16 = einsum(equation = var_52938_equation_0, values = (var_52784_cast_fp16, var_52904_cast_fp16))[name = tensor("op_52938_cast_fp16")]; + tensor var_52940_equation_0 = const()[name = tensor("op_52940_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52940_cast_fp16 = einsum(equation = var_52940_equation_0, values = (var_52788_cast_fp16, var_52905_cast_fp16))[name = tensor("op_52940_cast_fp16")]; + tensor var_52942_equation_0 = const()[name = tensor("op_52942_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52942_cast_fp16 = einsum(equation = var_52942_equation_0, values = (var_52792_cast_fp16, var_52906_cast_fp16))[name = tensor("op_52942_cast_fp16")]; + tensor var_52944_equation_0 = const()[name = tensor("op_52944_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52944_cast_fp16 = einsum(equation = var_52944_equation_0, values = (var_52796_cast_fp16, var_52907_cast_fp16))[name = tensor("op_52944_cast_fp16")]; + tensor var_52946_equation_0 = const()[name = tensor("op_52946_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52946_cast_fp16 = einsum(equation = var_52946_equation_0, values = (var_52800_cast_fp16, var_52908_cast_fp16))[name = tensor("op_52946_cast_fp16")]; + tensor var_52948_equation_0 = const()[name = tensor("op_52948_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52948_cast_fp16 = einsum(equation = var_52948_equation_0, values = (var_52804_cast_fp16, var_52909_cast_fp16))[name = tensor("op_52948_cast_fp16")]; + tensor var_52950_equation_0 = const()[name = tensor("op_52950_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52950_cast_fp16 = einsum(equation = var_52950_equation_0, values = (var_52808_cast_fp16, var_52910_cast_fp16))[name = tensor("op_52950_cast_fp16")]; + tensor input_667_interleave_0 = const()[name = tensor("input_667_interleave_0"), val = tensor(false)]; + tensor input_667_cast_fp16 = concat(axis = var_30355, interleave = input_667_interleave_0, values = (var_52912_cast_fp16, var_52914_cast_fp16, var_52916_cast_fp16, var_52918_cast_fp16, var_52920_cast_fp16, var_52922_cast_fp16, var_52924_cast_fp16, var_52926_cast_fp16, var_52928_cast_fp16, var_52930_cast_fp16, var_52932_cast_fp16, var_52934_cast_fp16, var_52936_cast_fp16, var_52938_cast_fp16, var_52940_cast_fp16, var_52942_cast_fp16, var_52944_cast_fp16, var_52946_cast_fp16, var_52948_cast_fp16, var_52950_cast_fp16))[name = tensor("input_667_cast_fp16")]; + tensor var_52956 = const()[name = tensor("op_52956"), val = tensor([1, 1])]; + tensor var_52958 = const()[name = tensor("op_52958"), val = tensor([1, 1])]; + tensor var_52960_pad_type_0 = const()[name = tensor("op_52960_pad_type_0"), val = tensor("custom")]; + tensor var_52960_pad_0 = const()[name = tensor("op_52960_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_4_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_4_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4459978688)))]; + tensor up_blocks_0_attentions_2_transformer_blocks_4_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_4_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4463255552)))]; + tensor var_52960_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_4_attn1_to_out_0_bias_to_fp16, dilations = var_52958, groups = var_30355, pad = var_52960_pad_0, pad_type = var_52960_pad_type_0, strides = var_52956, weight = up_blocks_0_attentions_2_transformer_blocks_4_attn1_to_out_0_weight_to_fp16, x = input_667_cast_fp16)[name = tensor("op_52960_cast_fp16")]; + tensor inputs_351_cast_fp16 = add(x = var_52960_cast_fp16, y = inputs_349_cast_fp16)[name = tensor("inputs_351_cast_fp16")]; + tensor hidden_states_461_axes_0 = const()[name = tensor("hidden_states_461_axes_0"), val = tensor([1])]; + tensor hidden_states_461_gamma_0_to_fp16 = const()[name = tensor("hidden_states_461_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4463258176)))]; + tensor hidden_states_461_beta_0_to_fp16 = const()[name = tensor("hidden_states_461_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4463260800)))]; + tensor var_52970_to_fp16 = const()[name = tensor("op_52970_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_461_cast_fp16 = layer_norm(axes = hidden_states_461_axes_0, beta = hidden_states_461_beta_0_to_fp16, epsilon = var_52970_to_fp16, gamma = hidden_states_461_gamma_0_to_fp16, x = inputs_351_cast_fp16)[name = tensor("hidden_states_461_cast_fp16")]; + tensor var_52985 = const()[name = tensor("op_52985"), val = tensor([1, 1])]; + tensor var_52987 = const()[name = tensor("op_52987"), val = tensor([1, 1])]; + tensor q_235_pad_type_0 = const()[name = tensor("q_235_pad_type_0"), val = tensor("custom")]; + tensor q_235_pad_0 = const()[name = tensor("q_235_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_4_attn2_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_4_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4463263424)))]; + tensor q_235_cast_fp16 = conv(dilations = var_52987, groups = var_30355, pad = q_235_pad_0, pad_type = q_235_pad_type_0, strides = var_52985, weight = up_blocks_0_attentions_2_transformer_blocks_4_attn2_to_q_weight_to_fp16, x = hidden_states_461_cast_fp16)[name = tensor("q_235_cast_fp16")]; + tensor var_52991 = const()[name = tensor("op_52991"), val = tensor([1, 1])]; + tensor var_52993 = const()[name = tensor("op_52993"), val = tensor([1, 1])]; + tensor k_469_pad_type_0 = const()[name = tensor("k_469_pad_type_0"), val = tensor("custom")]; + tensor k_469_pad_0 = const()[name = tensor("k_469_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_4_attn2_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_4_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4466540288)))]; + tensor k_469_cast_fp16 = conv(dilations = var_52993, groups = var_30355, pad = k_469_pad_0, pad_type = k_469_pad_type_0, strides = var_52991, weight = up_blocks_0_attentions_2_transformer_blocks_4_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_469_cast_fp16")]; + tensor var_52997 = const()[name = tensor("op_52997"), val = tensor([1, 1])]; + tensor var_52999 = const()[name = tensor("op_52999"), val = tensor([1, 1])]; + tensor v_235_pad_type_0 = const()[name = tensor("v_235_pad_type_0"), val = tensor("custom")]; + tensor v_235_pad_0 = const()[name = tensor("v_235_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_4_attn2_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_4_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4471783232)))]; + tensor v_235_cast_fp16 = conv(dilations = var_52999, groups = var_30355, pad = v_235_pad_0, pad_type = v_235_pad_type_0, strides = var_52997, weight = up_blocks_0_attentions_2_transformer_blocks_4_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_235_cast_fp16")]; + tensor var_53003_begin_0 = const()[name = tensor("op_53003_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_53003_end_0 = const()[name = tensor("op_53003_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_53003_end_mask_0 = const()[name = tensor("op_53003_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53003_cast_fp16 = slice_by_index(begin = var_53003_begin_0, end = var_53003_end_0, end_mask = var_53003_end_mask_0, x = q_235_cast_fp16)[name = tensor("op_53003_cast_fp16")]; + tensor var_53007_begin_0 = const()[name = tensor("op_53007_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_53007_end_0 = const()[name = tensor("op_53007_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_53007_end_mask_0 = const()[name = tensor("op_53007_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53007_cast_fp16 = slice_by_index(begin = var_53007_begin_0, end = var_53007_end_0, end_mask = var_53007_end_mask_0, x = q_235_cast_fp16)[name = tensor("op_53007_cast_fp16")]; + tensor var_53011_begin_0 = const()[name = tensor("op_53011_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_53011_end_0 = const()[name = tensor("op_53011_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_53011_end_mask_0 = const()[name = tensor("op_53011_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53011_cast_fp16 = slice_by_index(begin = var_53011_begin_0, end = var_53011_end_0, end_mask = var_53011_end_mask_0, x = q_235_cast_fp16)[name = tensor("op_53011_cast_fp16")]; + tensor var_53015_begin_0 = const()[name = tensor("op_53015_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_53015_end_0 = const()[name = tensor("op_53015_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_53015_end_mask_0 = const()[name = tensor("op_53015_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53015_cast_fp16 = slice_by_index(begin = var_53015_begin_0, end = var_53015_end_0, end_mask = var_53015_end_mask_0, x = q_235_cast_fp16)[name = tensor("op_53015_cast_fp16")]; + tensor var_53019_begin_0 = const()[name = tensor("op_53019_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_53019_end_0 = const()[name = tensor("op_53019_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_53019_end_mask_0 = const()[name = tensor("op_53019_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53019_cast_fp16 = slice_by_index(begin = var_53019_begin_0, end = var_53019_end_0, end_mask = var_53019_end_mask_0, x = q_235_cast_fp16)[name = tensor("op_53019_cast_fp16")]; + tensor var_53023_begin_0 = const()[name = tensor("op_53023_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_53023_end_0 = const()[name = tensor("op_53023_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_53023_end_mask_0 = const()[name = tensor("op_53023_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53023_cast_fp16 = slice_by_index(begin = var_53023_begin_0, end = var_53023_end_0, end_mask = var_53023_end_mask_0, x = q_235_cast_fp16)[name = tensor("op_53023_cast_fp16")]; + tensor var_53027_begin_0 = const()[name = tensor("op_53027_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_53027_end_0 = const()[name = tensor("op_53027_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_53027_end_mask_0 = const()[name = tensor("op_53027_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53027_cast_fp16 = slice_by_index(begin = var_53027_begin_0, end = var_53027_end_0, end_mask = var_53027_end_mask_0, x = q_235_cast_fp16)[name = tensor("op_53027_cast_fp16")]; + tensor var_53031_begin_0 = const()[name = tensor("op_53031_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_53031_end_0 = const()[name = tensor("op_53031_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_53031_end_mask_0 = const()[name = tensor("op_53031_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53031_cast_fp16 = slice_by_index(begin = var_53031_begin_0, end = var_53031_end_0, end_mask = var_53031_end_mask_0, x = q_235_cast_fp16)[name = tensor("op_53031_cast_fp16")]; + tensor var_53035_begin_0 = const()[name = tensor("op_53035_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_53035_end_0 = const()[name = tensor("op_53035_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_53035_end_mask_0 = const()[name = tensor("op_53035_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53035_cast_fp16 = slice_by_index(begin = var_53035_begin_0, end = var_53035_end_0, end_mask = var_53035_end_mask_0, x = q_235_cast_fp16)[name = tensor("op_53035_cast_fp16")]; + tensor var_53039_begin_0 = const()[name = tensor("op_53039_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_53039_end_0 = const()[name = tensor("op_53039_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_53039_end_mask_0 = const()[name = tensor("op_53039_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53039_cast_fp16 = slice_by_index(begin = var_53039_begin_0, end = var_53039_end_0, end_mask = var_53039_end_mask_0, x = q_235_cast_fp16)[name = tensor("op_53039_cast_fp16")]; + tensor var_53043_begin_0 = const()[name = tensor("op_53043_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_53043_end_0 = const()[name = tensor("op_53043_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_53043_end_mask_0 = const()[name = tensor("op_53043_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53043_cast_fp16 = slice_by_index(begin = var_53043_begin_0, end = var_53043_end_0, end_mask = var_53043_end_mask_0, x = q_235_cast_fp16)[name = tensor("op_53043_cast_fp16")]; + tensor var_53047_begin_0 = const()[name = tensor("op_53047_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_53047_end_0 = const()[name = tensor("op_53047_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_53047_end_mask_0 = const()[name = tensor("op_53047_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53047_cast_fp16 = slice_by_index(begin = var_53047_begin_0, end = var_53047_end_0, end_mask = var_53047_end_mask_0, x = q_235_cast_fp16)[name = tensor("op_53047_cast_fp16")]; + tensor var_53051_begin_0 = const()[name = tensor("op_53051_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_53051_end_0 = const()[name = tensor("op_53051_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_53051_end_mask_0 = const()[name = tensor("op_53051_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53051_cast_fp16 = slice_by_index(begin = var_53051_begin_0, end = var_53051_end_0, end_mask = var_53051_end_mask_0, x = q_235_cast_fp16)[name = tensor("op_53051_cast_fp16")]; + tensor var_53055_begin_0 = const()[name = tensor("op_53055_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_53055_end_0 = const()[name = tensor("op_53055_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_53055_end_mask_0 = const()[name = tensor("op_53055_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53055_cast_fp16 = slice_by_index(begin = var_53055_begin_0, end = var_53055_end_0, end_mask = var_53055_end_mask_0, x = q_235_cast_fp16)[name = tensor("op_53055_cast_fp16")]; + tensor var_53059_begin_0 = const()[name = tensor("op_53059_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_53059_end_0 = const()[name = tensor("op_53059_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_53059_end_mask_0 = const()[name = tensor("op_53059_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53059_cast_fp16 = slice_by_index(begin = var_53059_begin_0, end = var_53059_end_0, end_mask = var_53059_end_mask_0, x = q_235_cast_fp16)[name = tensor("op_53059_cast_fp16")]; + tensor var_53063_begin_0 = const()[name = tensor("op_53063_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_53063_end_0 = const()[name = tensor("op_53063_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_53063_end_mask_0 = const()[name = tensor("op_53063_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53063_cast_fp16 = slice_by_index(begin = var_53063_begin_0, end = var_53063_end_0, end_mask = var_53063_end_mask_0, x = q_235_cast_fp16)[name = tensor("op_53063_cast_fp16")]; + tensor var_53067_begin_0 = const()[name = tensor("op_53067_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_53067_end_0 = const()[name = tensor("op_53067_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_53067_end_mask_0 = const()[name = tensor("op_53067_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53067_cast_fp16 = slice_by_index(begin = var_53067_begin_0, end = var_53067_end_0, end_mask = var_53067_end_mask_0, x = q_235_cast_fp16)[name = tensor("op_53067_cast_fp16")]; + tensor var_53071_begin_0 = const()[name = tensor("op_53071_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_53071_end_0 = const()[name = tensor("op_53071_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_53071_end_mask_0 = const()[name = tensor("op_53071_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53071_cast_fp16 = slice_by_index(begin = var_53071_begin_0, end = var_53071_end_0, end_mask = var_53071_end_mask_0, x = q_235_cast_fp16)[name = tensor("op_53071_cast_fp16")]; + tensor var_53075_begin_0 = const()[name = tensor("op_53075_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_53075_end_0 = const()[name = tensor("op_53075_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_53075_end_mask_0 = const()[name = tensor("op_53075_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53075_cast_fp16 = slice_by_index(begin = var_53075_begin_0, end = var_53075_end_0, end_mask = var_53075_end_mask_0, x = q_235_cast_fp16)[name = tensor("op_53075_cast_fp16")]; + tensor var_53079_begin_0 = const()[name = tensor("op_53079_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_53079_end_0 = const()[name = tensor("op_53079_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_53079_end_mask_0 = const()[name = tensor("op_53079_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53079_cast_fp16 = slice_by_index(begin = var_53079_begin_0, end = var_53079_end_0, end_mask = var_53079_end_mask_0, x = q_235_cast_fp16)[name = tensor("op_53079_cast_fp16")]; + tensor k_471_perm_0 = const()[name = tensor("k_471_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_53086_begin_0 = const()[name = tensor("op_53086_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_53086_end_0 = const()[name = tensor("op_53086_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_53086_end_mask_0 = const()[name = tensor("op_53086_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_471_cast_fp16 = transpose(perm = k_471_perm_0, x = k_469_cast_fp16)[name = tensor("transpose_22")]; + tensor var_53086_cast_fp16 = slice_by_index(begin = var_53086_begin_0, end = var_53086_end_0, end_mask = var_53086_end_mask_0, x = k_471_cast_fp16)[name = tensor("op_53086_cast_fp16")]; + tensor var_53090_begin_0 = const()[name = tensor("op_53090_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_53090_end_0 = const()[name = tensor("op_53090_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_53090_end_mask_0 = const()[name = tensor("op_53090_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_53090_cast_fp16 = slice_by_index(begin = var_53090_begin_0, end = var_53090_end_0, end_mask = var_53090_end_mask_0, x = k_471_cast_fp16)[name = tensor("op_53090_cast_fp16")]; + tensor var_53094_begin_0 = const()[name = tensor("op_53094_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_53094_end_0 = const()[name = tensor("op_53094_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_53094_end_mask_0 = const()[name = tensor("op_53094_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_53094_cast_fp16 = slice_by_index(begin = var_53094_begin_0, end = var_53094_end_0, end_mask = var_53094_end_mask_0, x = k_471_cast_fp16)[name = tensor("op_53094_cast_fp16")]; + tensor var_53098_begin_0 = const()[name = tensor("op_53098_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_53098_end_0 = const()[name = tensor("op_53098_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_53098_end_mask_0 = const()[name = tensor("op_53098_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_53098_cast_fp16 = slice_by_index(begin = var_53098_begin_0, end = var_53098_end_0, end_mask = var_53098_end_mask_0, x = k_471_cast_fp16)[name = tensor("op_53098_cast_fp16")]; + tensor var_53102_begin_0 = const()[name = tensor("op_53102_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_53102_end_0 = const()[name = tensor("op_53102_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_53102_end_mask_0 = const()[name = tensor("op_53102_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_53102_cast_fp16 = slice_by_index(begin = var_53102_begin_0, end = var_53102_end_0, end_mask = var_53102_end_mask_0, x = k_471_cast_fp16)[name = tensor("op_53102_cast_fp16")]; + tensor var_53106_begin_0 = const()[name = tensor("op_53106_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_53106_end_0 = const()[name = tensor("op_53106_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_53106_end_mask_0 = const()[name = tensor("op_53106_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_53106_cast_fp16 = slice_by_index(begin = var_53106_begin_0, end = var_53106_end_0, end_mask = var_53106_end_mask_0, x = k_471_cast_fp16)[name = tensor("op_53106_cast_fp16")]; + tensor var_53110_begin_0 = const()[name = tensor("op_53110_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_53110_end_0 = const()[name = tensor("op_53110_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_53110_end_mask_0 = const()[name = tensor("op_53110_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_53110_cast_fp16 = slice_by_index(begin = var_53110_begin_0, end = var_53110_end_0, end_mask = var_53110_end_mask_0, x = k_471_cast_fp16)[name = tensor("op_53110_cast_fp16")]; + tensor var_53114_begin_0 = const()[name = tensor("op_53114_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_53114_end_0 = const()[name = tensor("op_53114_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_53114_end_mask_0 = const()[name = tensor("op_53114_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_53114_cast_fp16 = slice_by_index(begin = var_53114_begin_0, end = var_53114_end_0, end_mask = var_53114_end_mask_0, x = k_471_cast_fp16)[name = tensor("op_53114_cast_fp16")]; + tensor var_53118_begin_0 = const()[name = tensor("op_53118_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_53118_end_0 = const()[name = tensor("op_53118_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_53118_end_mask_0 = const()[name = tensor("op_53118_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_53118_cast_fp16 = slice_by_index(begin = var_53118_begin_0, end = var_53118_end_0, end_mask = var_53118_end_mask_0, x = k_471_cast_fp16)[name = tensor("op_53118_cast_fp16")]; + tensor var_53122_begin_0 = const()[name = tensor("op_53122_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_53122_end_0 = const()[name = tensor("op_53122_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_53122_end_mask_0 = const()[name = tensor("op_53122_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_53122_cast_fp16 = slice_by_index(begin = var_53122_begin_0, end = var_53122_end_0, end_mask = var_53122_end_mask_0, x = k_471_cast_fp16)[name = tensor("op_53122_cast_fp16")]; + tensor var_53126_begin_0 = const()[name = tensor("op_53126_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_53126_end_0 = const()[name = tensor("op_53126_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_53126_end_mask_0 = const()[name = tensor("op_53126_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_53126_cast_fp16 = slice_by_index(begin = var_53126_begin_0, end = var_53126_end_0, end_mask = var_53126_end_mask_0, x = k_471_cast_fp16)[name = tensor("op_53126_cast_fp16")]; + tensor var_53130_begin_0 = const()[name = tensor("op_53130_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_53130_end_0 = const()[name = tensor("op_53130_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_53130_end_mask_0 = const()[name = tensor("op_53130_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_53130_cast_fp16 = slice_by_index(begin = var_53130_begin_0, end = var_53130_end_0, end_mask = var_53130_end_mask_0, x = k_471_cast_fp16)[name = tensor("op_53130_cast_fp16")]; + tensor var_53134_begin_0 = const()[name = tensor("op_53134_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_53134_end_0 = const()[name = tensor("op_53134_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_53134_end_mask_0 = const()[name = tensor("op_53134_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_53134_cast_fp16 = slice_by_index(begin = var_53134_begin_0, end = var_53134_end_0, end_mask = var_53134_end_mask_0, x = k_471_cast_fp16)[name = tensor("op_53134_cast_fp16")]; + tensor var_53138_begin_0 = const()[name = tensor("op_53138_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_53138_end_0 = const()[name = tensor("op_53138_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_53138_end_mask_0 = const()[name = tensor("op_53138_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_53138_cast_fp16 = slice_by_index(begin = var_53138_begin_0, end = var_53138_end_0, end_mask = var_53138_end_mask_0, x = k_471_cast_fp16)[name = tensor("op_53138_cast_fp16")]; + tensor var_53142_begin_0 = const()[name = tensor("op_53142_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_53142_end_0 = const()[name = tensor("op_53142_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_53142_end_mask_0 = const()[name = tensor("op_53142_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_53142_cast_fp16 = slice_by_index(begin = var_53142_begin_0, end = var_53142_end_0, end_mask = var_53142_end_mask_0, x = k_471_cast_fp16)[name = tensor("op_53142_cast_fp16")]; + tensor var_53146_begin_0 = const()[name = tensor("op_53146_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_53146_end_0 = const()[name = tensor("op_53146_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_53146_end_mask_0 = const()[name = tensor("op_53146_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_53146_cast_fp16 = slice_by_index(begin = var_53146_begin_0, end = var_53146_end_0, end_mask = var_53146_end_mask_0, x = k_471_cast_fp16)[name = tensor("op_53146_cast_fp16")]; + tensor var_53150_begin_0 = const()[name = tensor("op_53150_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_53150_end_0 = const()[name = tensor("op_53150_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_53150_end_mask_0 = const()[name = tensor("op_53150_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_53150_cast_fp16 = slice_by_index(begin = var_53150_begin_0, end = var_53150_end_0, end_mask = var_53150_end_mask_0, x = k_471_cast_fp16)[name = tensor("op_53150_cast_fp16")]; + tensor var_53154_begin_0 = const()[name = tensor("op_53154_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_53154_end_0 = const()[name = tensor("op_53154_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_53154_end_mask_0 = const()[name = tensor("op_53154_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_53154_cast_fp16 = slice_by_index(begin = var_53154_begin_0, end = var_53154_end_0, end_mask = var_53154_end_mask_0, x = k_471_cast_fp16)[name = tensor("op_53154_cast_fp16")]; + tensor var_53158_begin_0 = const()[name = tensor("op_53158_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_53158_end_0 = const()[name = tensor("op_53158_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_53158_end_mask_0 = const()[name = tensor("op_53158_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_53158_cast_fp16 = slice_by_index(begin = var_53158_begin_0, end = var_53158_end_0, end_mask = var_53158_end_mask_0, x = k_471_cast_fp16)[name = tensor("op_53158_cast_fp16")]; + tensor var_53162_begin_0 = const()[name = tensor("op_53162_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_53162_end_0 = const()[name = tensor("op_53162_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_53162_end_mask_0 = const()[name = tensor("op_53162_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_53162_cast_fp16 = slice_by_index(begin = var_53162_begin_0, end = var_53162_end_0, end_mask = var_53162_end_mask_0, x = k_471_cast_fp16)[name = tensor("op_53162_cast_fp16")]; + tensor var_53164_begin_0 = const()[name = tensor("op_53164_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_53164_end_0 = const()[name = tensor("op_53164_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_53164_end_mask_0 = const()[name = tensor("op_53164_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53164_cast_fp16 = slice_by_index(begin = var_53164_begin_0, end = var_53164_end_0, end_mask = var_53164_end_mask_0, x = v_235_cast_fp16)[name = tensor("op_53164_cast_fp16")]; + tensor var_53168_begin_0 = const()[name = tensor("op_53168_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_53168_end_0 = const()[name = tensor("op_53168_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_53168_end_mask_0 = const()[name = tensor("op_53168_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53168_cast_fp16 = slice_by_index(begin = var_53168_begin_0, end = var_53168_end_0, end_mask = var_53168_end_mask_0, x = v_235_cast_fp16)[name = tensor("op_53168_cast_fp16")]; + tensor var_53172_begin_0 = const()[name = tensor("op_53172_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_53172_end_0 = const()[name = tensor("op_53172_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_53172_end_mask_0 = const()[name = tensor("op_53172_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53172_cast_fp16 = slice_by_index(begin = var_53172_begin_0, end = var_53172_end_0, end_mask = var_53172_end_mask_0, x = v_235_cast_fp16)[name = tensor("op_53172_cast_fp16")]; + tensor var_53176_begin_0 = const()[name = tensor("op_53176_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_53176_end_0 = const()[name = tensor("op_53176_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_53176_end_mask_0 = const()[name = tensor("op_53176_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53176_cast_fp16 = slice_by_index(begin = var_53176_begin_0, end = var_53176_end_0, end_mask = var_53176_end_mask_0, x = v_235_cast_fp16)[name = tensor("op_53176_cast_fp16")]; + tensor var_53180_begin_0 = const()[name = tensor("op_53180_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_53180_end_0 = const()[name = tensor("op_53180_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_53180_end_mask_0 = const()[name = tensor("op_53180_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53180_cast_fp16 = slice_by_index(begin = var_53180_begin_0, end = var_53180_end_0, end_mask = var_53180_end_mask_0, x = v_235_cast_fp16)[name = tensor("op_53180_cast_fp16")]; + tensor var_53184_begin_0 = const()[name = tensor("op_53184_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_53184_end_0 = const()[name = tensor("op_53184_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_53184_end_mask_0 = const()[name = tensor("op_53184_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53184_cast_fp16 = slice_by_index(begin = var_53184_begin_0, end = var_53184_end_0, end_mask = var_53184_end_mask_0, x = v_235_cast_fp16)[name = tensor("op_53184_cast_fp16")]; + tensor var_53188_begin_0 = const()[name = tensor("op_53188_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_53188_end_0 = const()[name = tensor("op_53188_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_53188_end_mask_0 = const()[name = tensor("op_53188_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53188_cast_fp16 = slice_by_index(begin = var_53188_begin_0, end = var_53188_end_0, end_mask = var_53188_end_mask_0, x = v_235_cast_fp16)[name = tensor("op_53188_cast_fp16")]; + tensor var_53192_begin_0 = const()[name = tensor("op_53192_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_53192_end_0 = const()[name = tensor("op_53192_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_53192_end_mask_0 = const()[name = tensor("op_53192_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53192_cast_fp16 = slice_by_index(begin = var_53192_begin_0, end = var_53192_end_0, end_mask = var_53192_end_mask_0, x = v_235_cast_fp16)[name = tensor("op_53192_cast_fp16")]; + tensor var_53196_begin_0 = const()[name = tensor("op_53196_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_53196_end_0 = const()[name = tensor("op_53196_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_53196_end_mask_0 = const()[name = tensor("op_53196_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53196_cast_fp16 = slice_by_index(begin = var_53196_begin_0, end = var_53196_end_0, end_mask = var_53196_end_mask_0, x = v_235_cast_fp16)[name = tensor("op_53196_cast_fp16")]; + tensor var_53200_begin_0 = const()[name = tensor("op_53200_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_53200_end_0 = const()[name = tensor("op_53200_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_53200_end_mask_0 = const()[name = tensor("op_53200_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53200_cast_fp16 = slice_by_index(begin = var_53200_begin_0, end = var_53200_end_0, end_mask = var_53200_end_mask_0, x = v_235_cast_fp16)[name = tensor("op_53200_cast_fp16")]; + tensor var_53204_begin_0 = const()[name = tensor("op_53204_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_53204_end_0 = const()[name = tensor("op_53204_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_53204_end_mask_0 = const()[name = tensor("op_53204_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53204_cast_fp16 = slice_by_index(begin = var_53204_begin_0, end = var_53204_end_0, end_mask = var_53204_end_mask_0, x = v_235_cast_fp16)[name = tensor("op_53204_cast_fp16")]; + tensor var_53208_begin_0 = const()[name = tensor("op_53208_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_53208_end_0 = const()[name = tensor("op_53208_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_53208_end_mask_0 = const()[name = tensor("op_53208_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53208_cast_fp16 = slice_by_index(begin = var_53208_begin_0, end = var_53208_end_0, end_mask = var_53208_end_mask_0, x = v_235_cast_fp16)[name = tensor("op_53208_cast_fp16")]; + tensor var_53212_begin_0 = const()[name = tensor("op_53212_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_53212_end_0 = const()[name = tensor("op_53212_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_53212_end_mask_0 = const()[name = tensor("op_53212_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53212_cast_fp16 = slice_by_index(begin = var_53212_begin_0, end = var_53212_end_0, end_mask = var_53212_end_mask_0, x = v_235_cast_fp16)[name = tensor("op_53212_cast_fp16")]; + tensor var_53216_begin_0 = const()[name = tensor("op_53216_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_53216_end_0 = const()[name = tensor("op_53216_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_53216_end_mask_0 = const()[name = tensor("op_53216_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53216_cast_fp16 = slice_by_index(begin = var_53216_begin_0, end = var_53216_end_0, end_mask = var_53216_end_mask_0, x = v_235_cast_fp16)[name = tensor("op_53216_cast_fp16")]; + tensor var_53220_begin_0 = const()[name = tensor("op_53220_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_53220_end_0 = const()[name = tensor("op_53220_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_53220_end_mask_0 = const()[name = tensor("op_53220_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53220_cast_fp16 = slice_by_index(begin = var_53220_begin_0, end = var_53220_end_0, end_mask = var_53220_end_mask_0, x = v_235_cast_fp16)[name = tensor("op_53220_cast_fp16")]; + tensor var_53224_begin_0 = const()[name = tensor("op_53224_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_53224_end_0 = const()[name = tensor("op_53224_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_53224_end_mask_0 = const()[name = tensor("op_53224_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53224_cast_fp16 = slice_by_index(begin = var_53224_begin_0, end = var_53224_end_0, end_mask = var_53224_end_mask_0, x = v_235_cast_fp16)[name = tensor("op_53224_cast_fp16")]; + tensor var_53228_begin_0 = const()[name = tensor("op_53228_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_53228_end_0 = const()[name = tensor("op_53228_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_53228_end_mask_0 = const()[name = tensor("op_53228_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53228_cast_fp16 = slice_by_index(begin = var_53228_begin_0, end = var_53228_end_0, end_mask = var_53228_end_mask_0, x = v_235_cast_fp16)[name = tensor("op_53228_cast_fp16")]; + tensor var_53232_begin_0 = const()[name = tensor("op_53232_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_53232_end_0 = const()[name = tensor("op_53232_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_53232_end_mask_0 = const()[name = tensor("op_53232_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53232_cast_fp16 = slice_by_index(begin = var_53232_begin_0, end = var_53232_end_0, end_mask = var_53232_end_mask_0, x = v_235_cast_fp16)[name = tensor("op_53232_cast_fp16")]; + tensor var_53236_begin_0 = const()[name = tensor("op_53236_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_53236_end_0 = const()[name = tensor("op_53236_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_53236_end_mask_0 = const()[name = tensor("op_53236_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53236_cast_fp16 = slice_by_index(begin = var_53236_begin_0, end = var_53236_end_0, end_mask = var_53236_end_mask_0, x = v_235_cast_fp16)[name = tensor("op_53236_cast_fp16")]; + tensor var_53240_begin_0 = const()[name = tensor("op_53240_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_53240_end_0 = const()[name = tensor("op_53240_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_53240_end_mask_0 = const()[name = tensor("op_53240_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53240_cast_fp16 = slice_by_index(begin = var_53240_begin_0, end = var_53240_end_0, end_mask = var_53240_end_mask_0, x = v_235_cast_fp16)[name = tensor("op_53240_cast_fp16")]; + tensor var_53244_equation_0 = const()[name = tensor("op_53244_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_53244_cast_fp16 = einsum(equation = var_53244_equation_0, values = (var_53086_cast_fp16, var_53003_cast_fp16))[name = tensor("op_53244_cast_fp16")]; + tensor var_53245_to_fp16 = const()[name = tensor("op_53245_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4521_cast_fp16 = mul(x = var_53244_cast_fp16, y = var_53245_to_fp16)[name = tensor("aw_4521_cast_fp16")]; + tensor var_53248_equation_0 = const()[name = tensor("op_53248_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_53248_cast_fp16 = einsum(equation = var_53248_equation_0, values = (var_53090_cast_fp16, var_53007_cast_fp16))[name = tensor("op_53248_cast_fp16")]; + tensor var_53249_to_fp16 = const()[name = tensor("op_53249_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4523_cast_fp16 = mul(x = var_53248_cast_fp16, y = var_53249_to_fp16)[name = tensor("aw_4523_cast_fp16")]; + tensor var_53252_equation_0 = const()[name = tensor("op_53252_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_53252_cast_fp16 = einsum(equation = var_53252_equation_0, values = (var_53094_cast_fp16, var_53011_cast_fp16))[name = tensor("op_53252_cast_fp16")]; + tensor var_53253_to_fp16 = const()[name = tensor("op_53253_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4525_cast_fp16 = mul(x = var_53252_cast_fp16, y = var_53253_to_fp16)[name = tensor("aw_4525_cast_fp16")]; + tensor var_53256_equation_0 = const()[name = tensor("op_53256_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_53256_cast_fp16 = einsum(equation = var_53256_equation_0, values = (var_53098_cast_fp16, var_53015_cast_fp16))[name = tensor("op_53256_cast_fp16")]; + tensor var_53257_to_fp16 = const()[name = tensor("op_53257_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4527_cast_fp16 = mul(x = var_53256_cast_fp16, y = var_53257_to_fp16)[name = tensor("aw_4527_cast_fp16")]; + tensor var_53260_equation_0 = const()[name = tensor("op_53260_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_53260_cast_fp16 = einsum(equation = var_53260_equation_0, values = (var_53102_cast_fp16, var_53019_cast_fp16))[name = tensor("op_53260_cast_fp16")]; + tensor var_53261_to_fp16 = const()[name = tensor("op_53261_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4529_cast_fp16 = mul(x = var_53260_cast_fp16, y = var_53261_to_fp16)[name = tensor("aw_4529_cast_fp16")]; + tensor var_53264_equation_0 = const()[name = tensor("op_53264_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_53264_cast_fp16 = einsum(equation = var_53264_equation_0, values = (var_53106_cast_fp16, var_53023_cast_fp16))[name = tensor("op_53264_cast_fp16")]; + tensor var_53265_to_fp16 = const()[name = tensor("op_53265_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4531_cast_fp16 = mul(x = var_53264_cast_fp16, y = var_53265_to_fp16)[name = tensor("aw_4531_cast_fp16")]; + tensor var_53268_equation_0 = const()[name = tensor("op_53268_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_53268_cast_fp16 = einsum(equation = var_53268_equation_0, values = (var_53110_cast_fp16, var_53027_cast_fp16))[name = tensor("op_53268_cast_fp16")]; + tensor var_53269_to_fp16 = const()[name = tensor("op_53269_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4533_cast_fp16 = mul(x = var_53268_cast_fp16, y = var_53269_to_fp16)[name = tensor("aw_4533_cast_fp16")]; + tensor var_53272_equation_0 = const()[name = tensor("op_53272_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_53272_cast_fp16 = einsum(equation = var_53272_equation_0, values = (var_53114_cast_fp16, var_53031_cast_fp16))[name = tensor("op_53272_cast_fp16")]; + tensor var_53273_to_fp16 = const()[name = tensor("op_53273_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4535_cast_fp16 = mul(x = var_53272_cast_fp16, y = var_53273_to_fp16)[name = tensor("aw_4535_cast_fp16")]; + tensor var_53276_equation_0 = const()[name = tensor("op_53276_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_53276_cast_fp16 = einsum(equation = var_53276_equation_0, values = (var_53118_cast_fp16, var_53035_cast_fp16))[name = tensor("op_53276_cast_fp16")]; + tensor var_53277_to_fp16 = const()[name = tensor("op_53277_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4537_cast_fp16 = mul(x = var_53276_cast_fp16, y = var_53277_to_fp16)[name = tensor("aw_4537_cast_fp16")]; + tensor var_53280_equation_0 = const()[name = tensor("op_53280_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_53280_cast_fp16 = einsum(equation = var_53280_equation_0, values = (var_53122_cast_fp16, var_53039_cast_fp16))[name = tensor("op_53280_cast_fp16")]; + tensor var_53281_to_fp16 = const()[name = tensor("op_53281_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4539_cast_fp16 = mul(x = var_53280_cast_fp16, y = var_53281_to_fp16)[name = tensor("aw_4539_cast_fp16")]; + tensor var_53284_equation_0 = const()[name = tensor("op_53284_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_53284_cast_fp16 = einsum(equation = var_53284_equation_0, values = (var_53126_cast_fp16, var_53043_cast_fp16))[name = tensor("op_53284_cast_fp16")]; + tensor var_53285_to_fp16 = const()[name = tensor("op_53285_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4541_cast_fp16 = mul(x = var_53284_cast_fp16, y = var_53285_to_fp16)[name = tensor("aw_4541_cast_fp16")]; + tensor var_53288_equation_0 = const()[name = tensor("op_53288_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_53288_cast_fp16 = einsum(equation = var_53288_equation_0, values = (var_53130_cast_fp16, var_53047_cast_fp16))[name = tensor("op_53288_cast_fp16")]; + tensor var_53289_to_fp16 = const()[name = tensor("op_53289_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4543_cast_fp16 = mul(x = var_53288_cast_fp16, y = var_53289_to_fp16)[name = tensor("aw_4543_cast_fp16")]; + tensor var_53292_equation_0 = const()[name = tensor("op_53292_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_53292_cast_fp16 = einsum(equation = var_53292_equation_0, values = (var_53134_cast_fp16, var_53051_cast_fp16))[name = tensor("op_53292_cast_fp16")]; + tensor var_53293_to_fp16 = const()[name = tensor("op_53293_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4545_cast_fp16 = mul(x = var_53292_cast_fp16, y = var_53293_to_fp16)[name = tensor("aw_4545_cast_fp16")]; + tensor var_53296_equation_0 = const()[name = tensor("op_53296_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_53296_cast_fp16 = einsum(equation = var_53296_equation_0, values = (var_53138_cast_fp16, var_53055_cast_fp16))[name = tensor("op_53296_cast_fp16")]; + tensor var_53297_to_fp16 = const()[name = tensor("op_53297_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4547_cast_fp16 = mul(x = var_53296_cast_fp16, y = var_53297_to_fp16)[name = tensor("aw_4547_cast_fp16")]; + tensor var_53300_equation_0 = const()[name = tensor("op_53300_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_53300_cast_fp16 = einsum(equation = var_53300_equation_0, values = (var_53142_cast_fp16, var_53059_cast_fp16))[name = tensor("op_53300_cast_fp16")]; + tensor var_53301_to_fp16 = const()[name = tensor("op_53301_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4549_cast_fp16 = mul(x = var_53300_cast_fp16, y = var_53301_to_fp16)[name = tensor("aw_4549_cast_fp16")]; + tensor var_53304_equation_0 = const()[name = tensor("op_53304_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_53304_cast_fp16 = einsum(equation = var_53304_equation_0, values = (var_53146_cast_fp16, var_53063_cast_fp16))[name = tensor("op_53304_cast_fp16")]; + tensor var_53305_to_fp16 = const()[name = tensor("op_53305_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4551_cast_fp16 = mul(x = var_53304_cast_fp16, y = var_53305_to_fp16)[name = tensor("aw_4551_cast_fp16")]; + tensor var_53308_equation_0 = const()[name = tensor("op_53308_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_53308_cast_fp16 = einsum(equation = var_53308_equation_0, values = (var_53150_cast_fp16, var_53067_cast_fp16))[name = tensor("op_53308_cast_fp16")]; + tensor var_53309_to_fp16 = const()[name = tensor("op_53309_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4553_cast_fp16 = mul(x = var_53308_cast_fp16, y = var_53309_to_fp16)[name = tensor("aw_4553_cast_fp16")]; + tensor var_53312_equation_0 = const()[name = tensor("op_53312_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_53312_cast_fp16 = einsum(equation = var_53312_equation_0, values = (var_53154_cast_fp16, var_53071_cast_fp16))[name = tensor("op_53312_cast_fp16")]; + tensor var_53313_to_fp16 = const()[name = tensor("op_53313_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4555_cast_fp16 = mul(x = var_53312_cast_fp16, y = var_53313_to_fp16)[name = tensor("aw_4555_cast_fp16")]; + tensor var_53316_equation_0 = const()[name = tensor("op_53316_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_53316_cast_fp16 = einsum(equation = var_53316_equation_0, values = (var_53158_cast_fp16, var_53075_cast_fp16))[name = tensor("op_53316_cast_fp16")]; + tensor var_53317_to_fp16 = const()[name = tensor("op_53317_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4557_cast_fp16 = mul(x = var_53316_cast_fp16, y = var_53317_to_fp16)[name = tensor("aw_4557_cast_fp16")]; + tensor var_53320_equation_0 = const()[name = tensor("op_53320_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_53320_cast_fp16 = einsum(equation = var_53320_equation_0, values = (var_53162_cast_fp16, var_53079_cast_fp16))[name = tensor("op_53320_cast_fp16")]; + tensor var_53321_to_fp16 = const()[name = tensor("op_53321_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4559_cast_fp16 = mul(x = var_53320_cast_fp16, y = var_53321_to_fp16)[name = tensor("aw_4559_cast_fp16")]; + tensor var_53323_cast_fp16 = softmax(axis = var_30355, x = aw_4521_cast_fp16)[name = tensor("op_53323_cast_fp16")]; + tensor var_53324_cast_fp16 = softmax(axis = var_30355, x = aw_4523_cast_fp16)[name = tensor("op_53324_cast_fp16")]; + tensor var_53325_cast_fp16 = softmax(axis = var_30355, x = aw_4525_cast_fp16)[name = tensor("op_53325_cast_fp16")]; + tensor var_53326_cast_fp16 = softmax(axis = var_30355, x = aw_4527_cast_fp16)[name = tensor("op_53326_cast_fp16")]; + tensor var_53327_cast_fp16 = softmax(axis = var_30355, x = aw_4529_cast_fp16)[name = tensor("op_53327_cast_fp16")]; + tensor var_53328_cast_fp16 = softmax(axis = var_30355, x = aw_4531_cast_fp16)[name = tensor("op_53328_cast_fp16")]; + tensor var_53329_cast_fp16 = softmax(axis = var_30355, x = aw_4533_cast_fp16)[name = tensor("op_53329_cast_fp16")]; + tensor var_53330_cast_fp16 = softmax(axis = var_30355, x = aw_4535_cast_fp16)[name = tensor("op_53330_cast_fp16")]; + tensor var_53331_cast_fp16 = softmax(axis = var_30355, x = aw_4537_cast_fp16)[name = tensor("op_53331_cast_fp16")]; + tensor var_53332_cast_fp16 = softmax(axis = var_30355, x = aw_4539_cast_fp16)[name = tensor("op_53332_cast_fp16")]; + tensor var_53333_cast_fp16 = softmax(axis = var_30355, x = aw_4541_cast_fp16)[name = tensor("op_53333_cast_fp16")]; + tensor var_53334_cast_fp16 = softmax(axis = var_30355, x = aw_4543_cast_fp16)[name = tensor("op_53334_cast_fp16")]; + tensor var_53335_cast_fp16 = softmax(axis = var_30355, x = aw_4545_cast_fp16)[name = tensor("op_53335_cast_fp16")]; + tensor var_53336_cast_fp16 = softmax(axis = var_30355, x = aw_4547_cast_fp16)[name = tensor("op_53336_cast_fp16")]; + tensor var_53337_cast_fp16 = softmax(axis = var_30355, x = aw_4549_cast_fp16)[name = tensor("op_53337_cast_fp16")]; + tensor var_53338_cast_fp16 = softmax(axis = var_30355, x = aw_4551_cast_fp16)[name = tensor("op_53338_cast_fp16")]; + tensor var_53339_cast_fp16 = softmax(axis = var_30355, x = aw_4553_cast_fp16)[name = tensor("op_53339_cast_fp16")]; + tensor var_53340_cast_fp16 = softmax(axis = var_30355, x = aw_4555_cast_fp16)[name = tensor("op_53340_cast_fp16")]; + tensor var_53341_cast_fp16 = softmax(axis = var_30355, x = aw_4557_cast_fp16)[name = tensor("op_53341_cast_fp16")]; + tensor var_53342_cast_fp16 = softmax(axis = var_30355, x = aw_4559_cast_fp16)[name = tensor("op_53342_cast_fp16")]; + tensor var_53344_equation_0 = const()[name = tensor("op_53344_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_53344_cast_fp16 = einsum(equation = var_53344_equation_0, values = (var_53164_cast_fp16, var_53323_cast_fp16))[name = tensor("op_53344_cast_fp16")]; + tensor var_53346_equation_0 = const()[name = tensor("op_53346_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_53346_cast_fp16 = einsum(equation = var_53346_equation_0, values = (var_53168_cast_fp16, var_53324_cast_fp16))[name = tensor("op_53346_cast_fp16")]; + tensor var_53348_equation_0 = const()[name = tensor("op_53348_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_53348_cast_fp16 = einsum(equation = var_53348_equation_0, values = (var_53172_cast_fp16, var_53325_cast_fp16))[name = tensor("op_53348_cast_fp16")]; + tensor var_53350_equation_0 = const()[name = tensor("op_53350_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_53350_cast_fp16 = einsum(equation = var_53350_equation_0, values = (var_53176_cast_fp16, var_53326_cast_fp16))[name = tensor("op_53350_cast_fp16")]; + tensor var_53352_equation_0 = const()[name = tensor("op_53352_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_53352_cast_fp16 = einsum(equation = var_53352_equation_0, values = (var_53180_cast_fp16, var_53327_cast_fp16))[name = tensor("op_53352_cast_fp16")]; + tensor var_53354_equation_0 = const()[name = tensor("op_53354_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_53354_cast_fp16 = einsum(equation = var_53354_equation_0, values = (var_53184_cast_fp16, var_53328_cast_fp16))[name = tensor("op_53354_cast_fp16")]; + tensor var_53356_equation_0 = const()[name = tensor("op_53356_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_53356_cast_fp16 = einsum(equation = var_53356_equation_0, values = (var_53188_cast_fp16, var_53329_cast_fp16))[name = tensor("op_53356_cast_fp16")]; + tensor var_53358_equation_0 = const()[name = tensor("op_53358_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_53358_cast_fp16 = einsum(equation = var_53358_equation_0, values = (var_53192_cast_fp16, var_53330_cast_fp16))[name = tensor("op_53358_cast_fp16")]; + tensor var_53360_equation_0 = const()[name = tensor("op_53360_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_53360_cast_fp16 = einsum(equation = var_53360_equation_0, values = (var_53196_cast_fp16, var_53331_cast_fp16))[name = tensor("op_53360_cast_fp16")]; + tensor var_53362_equation_0 = const()[name = tensor("op_53362_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_53362_cast_fp16 = einsum(equation = var_53362_equation_0, values = (var_53200_cast_fp16, var_53332_cast_fp16))[name = tensor("op_53362_cast_fp16")]; + tensor var_53364_equation_0 = const()[name = tensor("op_53364_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_53364_cast_fp16 = einsum(equation = var_53364_equation_0, values = (var_53204_cast_fp16, var_53333_cast_fp16))[name = tensor("op_53364_cast_fp16")]; + tensor var_53366_equation_0 = const()[name = tensor("op_53366_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_53366_cast_fp16 = einsum(equation = var_53366_equation_0, values = (var_53208_cast_fp16, var_53334_cast_fp16))[name = tensor("op_53366_cast_fp16")]; + tensor var_53368_equation_0 = const()[name = tensor("op_53368_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_53368_cast_fp16 = einsum(equation = var_53368_equation_0, values = (var_53212_cast_fp16, var_53335_cast_fp16))[name = tensor("op_53368_cast_fp16")]; + tensor var_53370_equation_0 = const()[name = tensor("op_53370_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_53370_cast_fp16 = einsum(equation = var_53370_equation_0, values = (var_53216_cast_fp16, var_53336_cast_fp16))[name = tensor("op_53370_cast_fp16")]; + tensor var_53372_equation_0 = const()[name = tensor("op_53372_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_53372_cast_fp16 = einsum(equation = var_53372_equation_0, values = (var_53220_cast_fp16, var_53337_cast_fp16))[name = tensor("op_53372_cast_fp16")]; + tensor var_53374_equation_0 = const()[name = tensor("op_53374_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_53374_cast_fp16 = einsum(equation = var_53374_equation_0, values = (var_53224_cast_fp16, var_53338_cast_fp16))[name = tensor("op_53374_cast_fp16")]; + tensor var_53376_equation_0 = const()[name = tensor("op_53376_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_53376_cast_fp16 = einsum(equation = var_53376_equation_0, values = (var_53228_cast_fp16, var_53339_cast_fp16))[name = tensor("op_53376_cast_fp16")]; + tensor var_53378_equation_0 = const()[name = tensor("op_53378_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_53378_cast_fp16 = einsum(equation = var_53378_equation_0, values = (var_53232_cast_fp16, var_53340_cast_fp16))[name = tensor("op_53378_cast_fp16")]; + tensor var_53380_equation_0 = const()[name = tensor("op_53380_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_53380_cast_fp16 = einsum(equation = var_53380_equation_0, values = (var_53236_cast_fp16, var_53341_cast_fp16))[name = tensor("op_53380_cast_fp16")]; + tensor var_53382_equation_0 = const()[name = tensor("op_53382_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_53382_cast_fp16 = einsum(equation = var_53382_equation_0, values = (var_53240_cast_fp16, var_53342_cast_fp16))[name = tensor("op_53382_cast_fp16")]; + tensor input_669_interleave_0 = const()[name = tensor("input_669_interleave_0"), val = tensor(false)]; + tensor input_669_cast_fp16 = concat(axis = var_30355, interleave = input_669_interleave_0, values = (var_53344_cast_fp16, var_53346_cast_fp16, var_53348_cast_fp16, var_53350_cast_fp16, var_53352_cast_fp16, var_53354_cast_fp16, var_53356_cast_fp16, var_53358_cast_fp16, var_53360_cast_fp16, var_53362_cast_fp16, var_53364_cast_fp16, var_53366_cast_fp16, var_53368_cast_fp16, var_53370_cast_fp16, var_53372_cast_fp16, var_53374_cast_fp16, var_53376_cast_fp16, var_53378_cast_fp16, var_53380_cast_fp16, var_53382_cast_fp16))[name = tensor("input_669_cast_fp16")]; + tensor var_53388 = const()[name = tensor("op_53388"), val = tensor([1, 1])]; + tensor var_53390 = const()[name = tensor("op_53390"), val = tensor([1, 1])]; + tensor var_53392_pad_type_0 = const()[name = tensor("op_53392_pad_type_0"), val = tensor("custom")]; + tensor var_53392_pad_0 = const()[name = tensor("op_53392_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_4_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_4_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4477026176)))]; + tensor up_blocks_0_attentions_2_transformer_blocks_4_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_4_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4480303040)))]; + tensor var_53392_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_4_attn2_to_out_0_bias_to_fp16, dilations = var_53390, groups = var_30355, pad = var_53392_pad_0, pad_type = var_53392_pad_type_0, strides = var_53388, weight = up_blocks_0_attentions_2_transformer_blocks_4_attn2_to_out_0_weight_to_fp16, x = input_669_cast_fp16)[name = tensor("op_53392_cast_fp16")]; + tensor inputs_353_cast_fp16 = add(x = var_53392_cast_fp16, y = inputs_351_cast_fp16)[name = tensor("inputs_353_cast_fp16")]; + tensor input_671_axes_0 = const()[name = tensor("input_671_axes_0"), val = tensor([1])]; + tensor input_671_gamma_0_to_fp16 = const()[name = tensor("input_671_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4480305664)))]; + tensor input_671_beta_0_to_fp16 = const()[name = tensor("input_671_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4480308288)))]; + tensor var_53402_to_fp16 = const()[name = tensor("op_53402_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_671_cast_fp16 = layer_norm(axes = input_671_axes_0, beta = input_671_beta_0_to_fp16, epsilon = var_53402_to_fp16, gamma = input_671_gamma_0_to_fp16, x = inputs_353_cast_fp16)[name = tensor("input_671_cast_fp16")]; + tensor var_53418 = const()[name = tensor("op_53418"), val = tensor([1, 1])]; + tensor var_53420 = const()[name = tensor("op_53420"), val = tensor([1, 1])]; + tensor var_53422_pad_type_0 = const()[name = tensor("op_53422_pad_type_0"), val = tensor("custom")]; + tensor var_53422_pad_0 = const()[name = tensor("op_53422_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_4_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_4_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4480310912)))]; + tensor up_blocks_0_attentions_2_transformer_blocks_4_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_4_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4506525376)))]; + tensor var_53422_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_4_ff_net_0_proj_bias_to_fp16, dilations = var_53420, groups = var_30355, pad = var_53422_pad_0, pad_type = var_53422_pad_type_0, strides = var_53418, weight = up_blocks_0_attentions_2_transformer_blocks_4_ff_net_0_proj_weight_to_fp16, x = input_671_cast_fp16)[name = tensor("op_53422_cast_fp16")]; + tensor var_53423_split_sizes_0 = const()[name = tensor("op_53423_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_53423_axis_0 = const()[name = tensor("op_53423_axis_0"), val = tensor(1)]; + tensor var_53423_cast_fp16_0, tensor var_53423_cast_fp16_1 = split(axis = var_53423_axis_0, split_sizes = var_53423_split_sizes_0, x = var_53422_cast_fp16)[name = tensor("op_53423_cast_fp16")]; + tensor var_53425_mode_0 = const()[name = tensor("op_53425_mode_0"), val = tensor("EXACT")]; + tensor var_53425_cast_fp16 = gelu(mode = var_53425_mode_0, x = var_53423_cast_fp16_1)[name = tensor("op_53425_cast_fp16")]; + tensor input_673_cast_fp16 = mul(x = var_53423_cast_fp16_0, y = var_53425_cast_fp16)[name = tensor("input_673_cast_fp16")]; + tensor var_53429 = const()[name = tensor("op_53429"), val = tensor([1, 1])]; + tensor var_53431 = const()[name = tensor("op_53431"), val = tensor([1, 1])]; + tensor var_53433_pad_type_0 = const()[name = tensor("op_53433_pad_type_0"), val = tensor("custom")]; + tensor var_53433_pad_0 = const()[name = tensor("op_53433_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_4_ff_net_2_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_4_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4506545920)))]; + tensor up_blocks_0_attentions_2_transformer_blocks_4_ff_net_2_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_4_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4519653184)))]; + tensor var_53433_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_4_ff_net_2_bias_to_fp16, dilations = var_53431, groups = var_30355, pad = var_53433_pad_0, pad_type = var_53433_pad_type_0, strides = var_53429, weight = up_blocks_0_attentions_2_transformer_blocks_4_ff_net_2_weight_to_fp16, x = input_673_cast_fp16)[name = tensor("op_53433_cast_fp16")]; + tensor inputs_355_cast_fp16 = add(x = var_53433_cast_fp16, y = inputs_353_cast_fp16)[name = tensor("inputs_355_cast_fp16")]; + tensor hidden_states_465_axes_0 = const()[name = tensor("hidden_states_465_axes_0"), val = tensor([1])]; + tensor hidden_states_465_gamma_0_to_fp16 = const()[name = tensor("hidden_states_465_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4519655808)))]; + tensor hidden_states_465_beta_0_to_fp16 = const()[name = tensor("hidden_states_465_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4519658432)))]; + tensor var_53449_to_fp16 = const()[name = tensor("op_53449_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_465_cast_fp16 = layer_norm(axes = hidden_states_465_axes_0, beta = hidden_states_465_beta_0_to_fp16, epsilon = var_53449_to_fp16, gamma = hidden_states_465_gamma_0_to_fp16, x = inputs_355_cast_fp16)[name = tensor("hidden_states_465_cast_fp16")]; + tensor var_53464 = const()[name = tensor("op_53464"), val = tensor([1, 1])]; + tensor var_53466 = const()[name = tensor("op_53466"), val = tensor([1, 1])]; + tensor q_237_pad_type_0 = const()[name = tensor("q_237_pad_type_0"), val = tensor("custom")]; + tensor q_237_pad_0 = const()[name = tensor("q_237_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_5_attn1_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_5_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4519661056)))]; + tensor q_237_cast_fp16 = conv(dilations = var_53466, groups = var_30355, pad = q_237_pad_0, pad_type = q_237_pad_type_0, strides = var_53464, weight = up_blocks_0_attentions_2_transformer_blocks_5_attn1_to_q_weight_to_fp16, x = hidden_states_465_cast_fp16)[name = tensor("q_237_cast_fp16")]; + tensor var_53470 = const()[name = tensor("op_53470"), val = tensor([1, 1])]; + tensor var_53472 = const()[name = tensor("op_53472"), val = tensor([1, 1])]; + tensor k_473_pad_type_0 = const()[name = tensor("k_473_pad_type_0"), val = tensor("custom")]; + tensor k_473_pad_0 = const()[name = tensor("k_473_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_5_attn1_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_5_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4522937920)))]; + tensor k_473_cast_fp16 = conv(dilations = var_53472, groups = var_30355, pad = k_473_pad_0, pad_type = k_473_pad_type_0, strides = var_53470, weight = up_blocks_0_attentions_2_transformer_blocks_5_attn1_to_k_weight_to_fp16, x = hidden_states_465_cast_fp16)[name = tensor("k_473_cast_fp16")]; + tensor var_53476 = const()[name = tensor("op_53476"), val = tensor([1, 1])]; + tensor var_53478 = const()[name = tensor("op_53478"), val = tensor([1, 1])]; + tensor v_237_pad_type_0 = const()[name = tensor("v_237_pad_type_0"), val = tensor("custom")]; + tensor v_237_pad_0 = const()[name = tensor("v_237_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_5_attn1_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_5_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4526214784)))]; + tensor v_237_cast_fp16 = conv(dilations = var_53478, groups = var_30355, pad = v_237_pad_0, pad_type = v_237_pad_type_0, strides = var_53476, weight = up_blocks_0_attentions_2_transformer_blocks_5_attn1_to_v_weight_to_fp16, x = hidden_states_465_cast_fp16)[name = tensor("v_237_cast_fp16")]; + tensor var_53482_begin_0 = const()[name = tensor("op_53482_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_53482_end_0 = const()[name = tensor("op_53482_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_53482_end_mask_0 = const()[name = tensor("op_53482_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53482_cast_fp16 = slice_by_index(begin = var_53482_begin_0, end = var_53482_end_0, end_mask = var_53482_end_mask_0, x = q_237_cast_fp16)[name = tensor("op_53482_cast_fp16")]; + tensor var_53486_begin_0 = const()[name = tensor("op_53486_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_53486_end_0 = const()[name = tensor("op_53486_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_53486_end_mask_0 = const()[name = tensor("op_53486_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53486_cast_fp16 = slice_by_index(begin = var_53486_begin_0, end = var_53486_end_0, end_mask = var_53486_end_mask_0, x = q_237_cast_fp16)[name = tensor("op_53486_cast_fp16")]; + tensor var_53490_begin_0 = const()[name = tensor("op_53490_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_53490_end_0 = const()[name = tensor("op_53490_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_53490_end_mask_0 = const()[name = tensor("op_53490_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53490_cast_fp16 = slice_by_index(begin = var_53490_begin_0, end = var_53490_end_0, end_mask = var_53490_end_mask_0, x = q_237_cast_fp16)[name = tensor("op_53490_cast_fp16")]; + tensor var_53494_begin_0 = const()[name = tensor("op_53494_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_53494_end_0 = const()[name = tensor("op_53494_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_53494_end_mask_0 = const()[name = tensor("op_53494_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53494_cast_fp16 = slice_by_index(begin = var_53494_begin_0, end = var_53494_end_0, end_mask = var_53494_end_mask_0, x = q_237_cast_fp16)[name = tensor("op_53494_cast_fp16")]; + tensor var_53498_begin_0 = const()[name = tensor("op_53498_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_53498_end_0 = const()[name = tensor("op_53498_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_53498_end_mask_0 = const()[name = tensor("op_53498_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53498_cast_fp16 = slice_by_index(begin = var_53498_begin_0, end = var_53498_end_0, end_mask = var_53498_end_mask_0, x = q_237_cast_fp16)[name = tensor("op_53498_cast_fp16")]; + tensor var_53502_begin_0 = const()[name = tensor("op_53502_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_53502_end_0 = const()[name = tensor("op_53502_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_53502_end_mask_0 = const()[name = tensor("op_53502_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53502_cast_fp16 = slice_by_index(begin = var_53502_begin_0, end = var_53502_end_0, end_mask = var_53502_end_mask_0, x = q_237_cast_fp16)[name = tensor("op_53502_cast_fp16")]; + tensor var_53506_begin_0 = const()[name = tensor("op_53506_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_53506_end_0 = const()[name = tensor("op_53506_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_53506_end_mask_0 = const()[name = tensor("op_53506_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53506_cast_fp16 = slice_by_index(begin = var_53506_begin_0, end = var_53506_end_0, end_mask = var_53506_end_mask_0, x = q_237_cast_fp16)[name = tensor("op_53506_cast_fp16")]; + tensor var_53510_begin_0 = const()[name = tensor("op_53510_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_53510_end_0 = const()[name = tensor("op_53510_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_53510_end_mask_0 = const()[name = tensor("op_53510_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53510_cast_fp16 = slice_by_index(begin = var_53510_begin_0, end = var_53510_end_0, end_mask = var_53510_end_mask_0, x = q_237_cast_fp16)[name = tensor("op_53510_cast_fp16")]; + tensor var_53514_begin_0 = const()[name = tensor("op_53514_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_53514_end_0 = const()[name = tensor("op_53514_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_53514_end_mask_0 = const()[name = tensor("op_53514_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53514_cast_fp16 = slice_by_index(begin = var_53514_begin_0, end = var_53514_end_0, end_mask = var_53514_end_mask_0, x = q_237_cast_fp16)[name = tensor("op_53514_cast_fp16")]; + tensor var_53518_begin_0 = const()[name = tensor("op_53518_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_53518_end_0 = const()[name = tensor("op_53518_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_53518_end_mask_0 = const()[name = tensor("op_53518_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53518_cast_fp16 = slice_by_index(begin = var_53518_begin_0, end = var_53518_end_0, end_mask = var_53518_end_mask_0, x = q_237_cast_fp16)[name = tensor("op_53518_cast_fp16")]; + tensor var_53522_begin_0 = const()[name = tensor("op_53522_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_53522_end_0 = const()[name = tensor("op_53522_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_53522_end_mask_0 = const()[name = tensor("op_53522_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53522_cast_fp16 = slice_by_index(begin = var_53522_begin_0, end = var_53522_end_0, end_mask = var_53522_end_mask_0, x = q_237_cast_fp16)[name = tensor("op_53522_cast_fp16")]; + tensor var_53526_begin_0 = const()[name = tensor("op_53526_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_53526_end_0 = const()[name = tensor("op_53526_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_53526_end_mask_0 = const()[name = tensor("op_53526_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53526_cast_fp16 = slice_by_index(begin = var_53526_begin_0, end = var_53526_end_0, end_mask = var_53526_end_mask_0, x = q_237_cast_fp16)[name = tensor("op_53526_cast_fp16")]; + tensor var_53530_begin_0 = const()[name = tensor("op_53530_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_53530_end_0 = const()[name = tensor("op_53530_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_53530_end_mask_0 = const()[name = tensor("op_53530_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53530_cast_fp16 = slice_by_index(begin = var_53530_begin_0, end = var_53530_end_0, end_mask = var_53530_end_mask_0, x = q_237_cast_fp16)[name = tensor("op_53530_cast_fp16")]; + tensor var_53534_begin_0 = const()[name = tensor("op_53534_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_53534_end_0 = const()[name = tensor("op_53534_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_53534_end_mask_0 = const()[name = tensor("op_53534_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53534_cast_fp16 = slice_by_index(begin = var_53534_begin_0, end = var_53534_end_0, end_mask = var_53534_end_mask_0, x = q_237_cast_fp16)[name = tensor("op_53534_cast_fp16")]; + tensor var_53538_begin_0 = const()[name = tensor("op_53538_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_53538_end_0 = const()[name = tensor("op_53538_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_53538_end_mask_0 = const()[name = tensor("op_53538_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53538_cast_fp16 = slice_by_index(begin = var_53538_begin_0, end = var_53538_end_0, end_mask = var_53538_end_mask_0, x = q_237_cast_fp16)[name = tensor("op_53538_cast_fp16")]; + tensor var_53542_begin_0 = const()[name = tensor("op_53542_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_53542_end_0 = const()[name = tensor("op_53542_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_53542_end_mask_0 = const()[name = tensor("op_53542_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53542_cast_fp16 = slice_by_index(begin = var_53542_begin_0, end = var_53542_end_0, end_mask = var_53542_end_mask_0, x = q_237_cast_fp16)[name = tensor("op_53542_cast_fp16")]; + tensor var_53546_begin_0 = const()[name = tensor("op_53546_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_53546_end_0 = const()[name = tensor("op_53546_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_53546_end_mask_0 = const()[name = tensor("op_53546_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53546_cast_fp16 = slice_by_index(begin = var_53546_begin_0, end = var_53546_end_0, end_mask = var_53546_end_mask_0, x = q_237_cast_fp16)[name = tensor("op_53546_cast_fp16")]; + tensor var_53550_begin_0 = const()[name = tensor("op_53550_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_53550_end_0 = const()[name = tensor("op_53550_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_53550_end_mask_0 = const()[name = tensor("op_53550_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53550_cast_fp16 = slice_by_index(begin = var_53550_begin_0, end = var_53550_end_0, end_mask = var_53550_end_mask_0, x = q_237_cast_fp16)[name = tensor("op_53550_cast_fp16")]; + tensor var_53554_begin_0 = const()[name = tensor("op_53554_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_53554_end_0 = const()[name = tensor("op_53554_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_53554_end_mask_0 = const()[name = tensor("op_53554_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53554_cast_fp16 = slice_by_index(begin = var_53554_begin_0, end = var_53554_end_0, end_mask = var_53554_end_mask_0, x = q_237_cast_fp16)[name = tensor("op_53554_cast_fp16")]; + tensor var_53558_begin_0 = const()[name = tensor("op_53558_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_53558_end_0 = const()[name = tensor("op_53558_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_53558_end_mask_0 = const()[name = tensor("op_53558_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53558_cast_fp16 = slice_by_index(begin = var_53558_begin_0, end = var_53558_end_0, end_mask = var_53558_end_mask_0, x = q_237_cast_fp16)[name = tensor("op_53558_cast_fp16")]; + tensor k_475_perm_0 = const()[name = tensor("k_475_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_53565_begin_0 = const()[name = tensor("op_53565_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_53565_end_0 = const()[name = tensor("op_53565_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_53565_end_mask_0 = const()[name = tensor("op_53565_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_475_cast_fp16 = transpose(perm = k_475_perm_0, x = k_473_cast_fp16)[name = tensor("transpose_21")]; + tensor var_53565_cast_fp16 = slice_by_index(begin = var_53565_begin_0, end = var_53565_end_0, end_mask = var_53565_end_mask_0, x = k_475_cast_fp16)[name = tensor("op_53565_cast_fp16")]; + tensor var_53569_begin_0 = const()[name = tensor("op_53569_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_53569_end_0 = const()[name = tensor("op_53569_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_53569_end_mask_0 = const()[name = tensor("op_53569_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_53569_cast_fp16 = slice_by_index(begin = var_53569_begin_0, end = var_53569_end_0, end_mask = var_53569_end_mask_0, x = k_475_cast_fp16)[name = tensor("op_53569_cast_fp16")]; + tensor var_53573_begin_0 = const()[name = tensor("op_53573_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_53573_end_0 = const()[name = tensor("op_53573_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_53573_end_mask_0 = const()[name = tensor("op_53573_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_53573_cast_fp16 = slice_by_index(begin = var_53573_begin_0, end = var_53573_end_0, end_mask = var_53573_end_mask_0, x = k_475_cast_fp16)[name = tensor("op_53573_cast_fp16")]; + tensor var_53577_begin_0 = const()[name = tensor("op_53577_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_53577_end_0 = const()[name = tensor("op_53577_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_53577_end_mask_0 = const()[name = tensor("op_53577_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_53577_cast_fp16 = slice_by_index(begin = var_53577_begin_0, end = var_53577_end_0, end_mask = var_53577_end_mask_0, x = k_475_cast_fp16)[name = tensor("op_53577_cast_fp16")]; + tensor var_53581_begin_0 = const()[name = tensor("op_53581_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_53581_end_0 = const()[name = tensor("op_53581_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_53581_end_mask_0 = const()[name = tensor("op_53581_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_53581_cast_fp16 = slice_by_index(begin = var_53581_begin_0, end = var_53581_end_0, end_mask = var_53581_end_mask_0, x = k_475_cast_fp16)[name = tensor("op_53581_cast_fp16")]; + tensor var_53585_begin_0 = const()[name = tensor("op_53585_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_53585_end_0 = const()[name = tensor("op_53585_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_53585_end_mask_0 = const()[name = tensor("op_53585_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_53585_cast_fp16 = slice_by_index(begin = var_53585_begin_0, end = var_53585_end_0, end_mask = var_53585_end_mask_0, x = k_475_cast_fp16)[name = tensor("op_53585_cast_fp16")]; + tensor var_53589_begin_0 = const()[name = tensor("op_53589_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_53589_end_0 = const()[name = tensor("op_53589_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_53589_end_mask_0 = const()[name = tensor("op_53589_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_53589_cast_fp16 = slice_by_index(begin = var_53589_begin_0, end = var_53589_end_0, end_mask = var_53589_end_mask_0, x = k_475_cast_fp16)[name = tensor("op_53589_cast_fp16")]; + tensor var_53593_begin_0 = const()[name = tensor("op_53593_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_53593_end_0 = const()[name = tensor("op_53593_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_53593_end_mask_0 = const()[name = tensor("op_53593_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_53593_cast_fp16 = slice_by_index(begin = var_53593_begin_0, end = var_53593_end_0, end_mask = var_53593_end_mask_0, x = k_475_cast_fp16)[name = tensor("op_53593_cast_fp16")]; + tensor var_53597_begin_0 = const()[name = tensor("op_53597_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_53597_end_0 = const()[name = tensor("op_53597_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_53597_end_mask_0 = const()[name = tensor("op_53597_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_53597_cast_fp16 = slice_by_index(begin = var_53597_begin_0, end = var_53597_end_0, end_mask = var_53597_end_mask_0, x = k_475_cast_fp16)[name = tensor("op_53597_cast_fp16")]; + tensor var_53601_begin_0 = const()[name = tensor("op_53601_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_53601_end_0 = const()[name = tensor("op_53601_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_53601_end_mask_0 = const()[name = tensor("op_53601_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_53601_cast_fp16 = slice_by_index(begin = var_53601_begin_0, end = var_53601_end_0, end_mask = var_53601_end_mask_0, x = k_475_cast_fp16)[name = tensor("op_53601_cast_fp16")]; + tensor var_53605_begin_0 = const()[name = tensor("op_53605_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_53605_end_0 = const()[name = tensor("op_53605_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_53605_end_mask_0 = const()[name = tensor("op_53605_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_53605_cast_fp16 = slice_by_index(begin = var_53605_begin_0, end = var_53605_end_0, end_mask = var_53605_end_mask_0, x = k_475_cast_fp16)[name = tensor("op_53605_cast_fp16")]; + tensor var_53609_begin_0 = const()[name = tensor("op_53609_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_53609_end_0 = const()[name = tensor("op_53609_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_53609_end_mask_0 = const()[name = tensor("op_53609_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_53609_cast_fp16 = slice_by_index(begin = var_53609_begin_0, end = var_53609_end_0, end_mask = var_53609_end_mask_0, x = k_475_cast_fp16)[name = tensor("op_53609_cast_fp16")]; + tensor var_53613_begin_0 = const()[name = tensor("op_53613_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_53613_end_0 = const()[name = tensor("op_53613_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_53613_end_mask_0 = const()[name = tensor("op_53613_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_53613_cast_fp16 = slice_by_index(begin = var_53613_begin_0, end = var_53613_end_0, end_mask = var_53613_end_mask_0, x = k_475_cast_fp16)[name = tensor("op_53613_cast_fp16")]; + tensor var_53617_begin_0 = const()[name = tensor("op_53617_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_53617_end_0 = const()[name = tensor("op_53617_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_53617_end_mask_0 = const()[name = tensor("op_53617_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_53617_cast_fp16 = slice_by_index(begin = var_53617_begin_0, end = var_53617_end_0, end_mask = var_53617_end_mask_0, x = k_475_cast_fp16)[name = tensor("op_53617_cast_fp16")]; + tensor var_53621_begin_0 = const()[name = tensor("op_53621_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_53621_end_0 = const()[name = tensor("op_53621_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_53621_end_mask_0 = const()[name = tensor("op_53621_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_53621_cast_fp16 = slice_by_index(begin = var_53621_begin_0, end = var_53621_end_0, end_mask = var_53621_end_mask_0, x = k_475_cast_fp16)[name = tensor("op_53621_cast_fp16")]; + tensor var_53625_begin_0 = const()[name = tensor("op_53625_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_53625_end_0 = const()[name = tensor("op_53625_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_53625_end_mask_0 = const()[name = tensor("op_53625_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_53625_cast_fp16 = slice_by_index(begin = var_53625_begin_0, end = var_53625_end_0, end_mask = var_53625_end_mask_0, x = k_475_cast_fp16)[name = tensor("op_53625_cast_fp16")]; + tensor var_53629_begin_0 = const()[name = tensor("op_53629_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_53629_end_0 = const()[name = tensor("op_53629_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_53629_end_mask_0 = const()[name = tensor("op_53629_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_53629_cast_fp16 = slice_by_index(begin = var_53629_begin_0, end = var_53629_end_0, end_mask = var_53629_end_mask_0, x = k_475_cast_fp16)[name = tensor("op_53629_cast_fp16")]; + tensor var_53633_begin_0 = const()[name = tensor("op_53633_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_53633_end_0 = const()[name = tensor("op_53633_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_53633_end_mask_0 = const()[name = tensor("op_53633_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_53633_cast_fp16 = slice_by_index(begin = var_53633_begin_0, end = var_53633_end_0, end_mask = var_53633_end_mask_0, x = k_475_cast_fp16)[name = tensor("op_53633_cast_fp16")]; + tensor var_53637_begin_0 = const()[name = tensor("op_53637_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_53637_end_0 = const()[name = tensor("op_53637_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_53637_end_mask_0 = const()[name = tensor("op_53637_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_53637_cast_fp16 = slice_by_index(begin = var_53637_begin_0, end = var_53637_end_0, end_mask = var_53637_end_mask_0, x = k_475_cast_fp16)[name = tensor("op_53637_cast_fp16")]; + tensor var_53641_begin_0 = const()[name = tensor("op_53641_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_53641_end_0 = const()[name = tensor("op_53641_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_53641_end_mask_0 = const()[name = tensor("op_53641_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_53641_cast_fp16 = slice_by_index(begin = var_53641_begin_0, end = var_53641_end_0, end_mask = var_53641_end_mask_0, x = k_475_cast_fp16)[name = tensor("op_53641_cast_fp16")]; + tensor var_53643_begin_0 = const()[name = tensor("op_53643_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_53643_end_0 = const()[name = tensor("op_53643_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_53643_end_mask_0 = const()[name = tensor("op_53643_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53643_cast_fp16 = slice_by_index(begin = var_53643_begin_0, end = var_53643_end_0, end_mask = var_53643_end_mask_0, x = v_237_cast_fp16)[name = tensor("op_53643_cast_fp16")]; + tensor var_53647_begin_0 = const()[name = tensor("op_53647_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_53647_end_0 = const()[name = tensor("op_53647_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_53647_end_mask_0 = const()[name = tensor("op_53647_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53647_cast_fp16 = slice_by_index(begin = var_53647_begin_0, end = var_53647_end_0, end_mask = var_53647_end_mask_0, x = v_237_cast_fp16)[name = tensor("op_53647_cast_fp16")]; + tensor var_53651_begin_0 = const()[name = tensor("op_53651_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_53651_end_0 = const()[name = tensor("op_53651_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_53651_end_mask_0 = const()[name = tensor("op_53651_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53651_cast_fp16 = slice_by_index(begin = var_53651_begin_0, end = var_53651_end_0, end_mask = var_53651_end_mask_0, x = v_237_cast_fp16)[name = tensor("op_53651_cast_fp16")]; + tensor var_53655_begin_0 = const()[name = tensor("op_53655_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_53655_end_0 = const()[name = tensor("op_53655_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_53655_end_mask_0 = const()[name = tensor("op_53655_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53655_cast_fp16 = slice_by_index(begin = var_53655_begin_0, end = var_53655_end_0, end_mask = var_53655_end_mask_0, x = v_237_cast_fp16)[name = tensor("op_53655_cast_fp16")]; + tensor var_53659_begin_0 = const()[name = tensor("op_53659_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_53659_end_0 = const()[name = tensor("op_53659_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_53659_end_mask_0 = const()[name = tensor("op_53659_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53659_cast_fp16 = slice_by_index(begin = var_53659_begin_0, end = var_53659_end_0, end_mask = var_53659_end_mask_0, x = v_237_cast_fp16)[name = tensor("op_53659_cast_fp16")]; + tensor var_53663_begin_0 = const()[name = tensor("op_53663_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_53663_end_0 = const()[name = tensor("op_53663_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_53663_end_mask_0 = const()[name = tensor("op_53663_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53663_cast_fp16 = slice_by_index(begin = var_53663_begin_0, end = var_53663_end_0, end_mask = var_53663_end_mask_0, x = v_237_cast_fp16)[name = tensor("op_53663_cast_fp16")]; + tensor var_53667_begin_0 = const()[name = tensor("op_53667_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_53667_end_0 = const()[name = tensor("op_53667_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_53667_end_mask_0 = const()[name = tensor("op_53667_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53667_cast_fp16 = slice_by_index(begin = var_53667_begin_0, end = var_53667_end_0, end_mask = var_53667_end_mask_0, x = v_237_cast_fp16)[name = tensor("op_53667_cast_fp16")]; + tensor var_53671_begin_0 = const()[name = tensor("op_53671_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_53671_end_0 = const()[name = tensor("op_53671_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_53671_end_mask_0 = const()[name = tensor("op_53671_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53671_cast_fp16 = slice_by_index(begin = var_53671_begin_0, end = var_53671_end_0, end_mask = var_53671_end_mask_0, x = v_237_cast_fp16)[name = tensor("op_53671_cast_fp16")]; + tensor var_53675_begin_0 = const()[name = tensor("op_53675_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_53675_end_0 = const()[name = tensor("op_53675_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_53675_end_mask_0 = const()[name = tensor("op_53675_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53675_cast_fp16 = slice_by_index(begin = var_53675_begin_0, end = var_53675_end_0, end_mask = var_53675_end_mask_0, x = v_237_cast_fp16)[name = tensor("op_53675_cast_fp16")]; + tensor var_53679_begin_0 = const()[name = tensor("op_53679_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_53679_end_0 = const()[name = tensor("op_53679_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_53679_end_mask_0 = const()[name = tensor("op_53679_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53679_cast_fp16 = slice_by_index(begin = var_53679_begin_0, end = var_53679_end_0, end_mask = var_53679_end_mask_0, x = v_237_cast_fp16)[name = tensor("op_53679_cast_fp16")]; + tensor var_53683_begin_0 = const()[name = tensor("op_53683_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_53683_end_0 = const()[name = tensor("op_53683_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_53683_end_mask_0 = const()[name = tensor("op_53683_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53683_cast_fp16 = slice_by_index(begin = var_53683_begin_0, end = var_53683_end_0, end_mask = var_53683_end_mask_0, x = v_237_cast_fp16)[name = tensor("op_53683_cast_fp16")]; + tensor var_53687_begin_0 = const()[name = tensor("op_53687_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_53687_end_0 = const()[name = tensor("op_53687_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_53687_end_mask_0 = const()[name = tensor("op_53687_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53687_cast_fp16 = slice_by_index(begin = var_53687_begin_0, end = var_53687_end_0, end_mask = var_53687_end_mask_0, x = v_237_cast_fp16)[name = tensor("op_53687_cast_fp16")]; + tensor var_53691_begin_0 = const()[name = tensor("op_53691_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_53691_end_0 = const()[name = tensor("op_53691_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_53691_end_mask_0 = const()[name = tensor("op_53691_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53691_cast_fp16 = slice_by_index(begin = var_53691_begin_0, end = var_53691_end_0, end_mask = var_53691_end_mask_0, x = v_237_cast_fp16)[name = tensor("op_53691_cast_fp16")]; + tensor var_53695_begin_0 = const()[name = tensor("op_53695_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_53695_end_0 = const()[name = tensor("op_53695_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_53695_end_mask_0 = const()[name = tensor("op_53695_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53695_cast_fp16 = slice_by_index(begin = var_53695_begin_0, end = var_53695_end_0, end_mask = var_53695_end_mask_0, x = v_237_cast_fp16)[name = tensor("op_53695_cast_fp16")]; + tensor var_53699_begin_0 = const()[name = tensor("op_53699_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_53699_end_0 = const()[name = tensor("op_53699_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_53699_end_mask_0 = const()[name = tensor("op_53699_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53699_cast_fp16 = slice_by_index(begin = var_53699_begin_0, end = var_53699_end_0, end_mask = var_53699_end_mask_0, x = v_237_cast_fp16)[name = tensor("op_53699_cast_fp16")]; + tensor var_53703_begin_0 = const()[name = tensor("op_53703_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_53703_end_0 = const()[name = tensor("op_53703_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_53703_end_mask_0 = const()[name = tensor("op_53703_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53703_cast_fp16 = slice_by_index(begin = var_53703_begin_0, end = var_53703_end_0, end_mask = var_53703_end_mask_0, x = v_237_cast_fp16)[name = tensor("op_53703_cast_fp16")]; + tensor var_53707_begin_0 = const()[name = tensor("op_53707_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_53707_end_0 = const()[name = tensor("op_53707_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_53707_end_mask_0 = const()[name = tensor("op_53707_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53707_cast_fp16 = slice_by_index(begin = var_53707_begin_0, end = var_53707_end_0, end_mask = var_53707_end_mask_0, x = v_237_cast_fp16)[name = tensor("op_53707_cast_fp16")]; + tensor var_53711_begin_0 = const()[name = tensor("op_53711_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_53711_end_0 = const()[name = tensor("op_53711_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_53711_end_mask_0 = const()[name = tensor("op_53711_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53711_cast_fp16 = slice_by_index(begin = var_53711_begin_0, end = var_53711_end_0, end_mask = var_53711_end_mask_0, x = v_237_cast_fp16)[name = tensor("op_53711_cast_fp16")]; + tensor var_53715_begin_0 = const()[name = tensor("op_53715_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_53715_end_0 = const()[name = tensor("op_53715_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_53715_end_mask_0 = const()[name = tensor("op_53715_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53715_cast_fp16 = slice_by_index(begin = var_53715_begin_0, end = var_53715_end_0, end_mask = var_53715_end_mask_0, x = v_237_cast_fp16)[name = tensor("op_53715_cast_fp16")]; + tensor var_53719_begin_0 = const()[name = tensor("op_53719_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_53719_end_0 = const()[name = tensor("op_53719_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_53719_end_mask_0 = const()[name = tensor("op_53719_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53719_cast_fp16 = slice_by_index(begin = var_53719_begin_0, end = var_53719_end_0, end_mask = var_53719_end_mask_0, x = v_237_cast_fp16)[name = tensor("op_53719_cast_fp16")]; + tensor var_53723_equation_0 = const()[name = tensor("op_53723_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_53723_cast_fp16 = einsum(equation = var_53723_equation_0, values = (var_53565_cast_fp16, var_53482_cast_fp16))[name = tensor("op_53723_cast_fp16")]; + tensor var_53724_to_fp16 = const()[name = tensor("op_53724_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4561_cast_fp16 = mul(x = var_53723_cast_fp16, y = var_53724_to_fp16)[name = tensor("aw_4561_cast_fp16")]; + tensor var_53727_equation_0 = const()[name = tensor("op_53727_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_53727_cast_fp16 = einsum(equation = var_53727_equation_0, values = (var_53569_cast_fp16, var_53486_cast_fp16))[name = tensor("op_53727_cast_fp16")]; + tensor var_53728_to_fp16 = const()[name = tensor("op_53728_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4563_cast_fp16 = mul(x = var_53727_cast_fp16, y = var_53728_to_fp16)[name = tensor("aw_4563_cast_fp16")]; + tensor var_53731_equation_0 = const()[name = tensor("op_53731_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_53731_cast_fp16 = einsum(equation = var_53731_equation_0, values = (var_53573_cast_fp16, var_53490_cast_fp16))[name = tensor("op_53731_cast_fp16")]; + tensor var_53732_to_fp16 = const()[name = tensor("op_53732_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4565_cast_fp16 = mul(x = var_53731_cast_fp16, y = var_53732_to_fp16)[name = tensor("aw_4565_cast_fp16")]; + tensor var_53735_equation_0 = const()[name = tensor("op_53735_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_53735_cast_fp16 = einsum(equation = var_53735_equation_0, values = (var_53577_cast_fp16, var_53494_cast_fp16))[name = tensor("op_53735_cast_fp16")]; + tensor var_53736_to_fp16 = const()[name = tensor("op_53736_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4567_cast_fp16 = mul(x = var_53735_cast_fp16, y = var_53736_to_fp16)[name = tensor("aw_4567_cast_fp16")]; + tensor var_53739_equation_0 = const()[name = tensor("op_53739_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_53739_cast_fp16 = einsum(equation = var_53739_equation_0, values = (var_53581_cast_fp16, var_53498_cast_fp16))[name = tensor("op_53739_cast_fp16")]; + tensor var_53740_to_fp16 = const()[name = tensor("op_53740_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4569_cast_fp16 = mul(x = var_53739_cast_fp16, y = var_53740_to_fp16)[name = tensor("aw_4569_cast_fp16")]; + tensor var_53743_equation_0 = const()[name = tensor("op_53743_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_53743_cast_fp16 = einsum(equation = var_53743_equation_0, values = (var_53585_cast_fp16, var_53502_cast_fp16))[name = tensor("op_53743_cast_fp16")]; + tensor var_53744_to_fp16 = const()[name = tensor("op_53744_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4571_cast_fp16 = mul(x = var_53743_cast_fp16, y = var_53744_to_fp16)[name = tensor("aw_4571_cast_fp16")]; + tensor var_53747_equation_0 = const()[name = tensor("op_53747_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_53747_cast_fp16 = einsum(equation = var_53747_equation_0, values = (var_53589_cast_fp16, var_53506_cast_fp16))[name = tensor("op_53747_cast_fp16")]; + tensor var_53748_to_fp16 = const()[name = tensor("op_53748_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4573_cast_fp16 = mul(x = var_53747_cast_fp16, y = var_53748_to_fp16)[name = tensor("aw_4573_cast_fp16")]; + tensor var_53751_equation_0 = const()[name = tensor("op_53751_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_53751_cast_fp16 = einsum(equation = var_53751_equation_0, values = (var_53593_cast_fp16, var_53510_cast_fp16))[name = tensor("op_53751_cast_fp16")]; + tensor var_53752_to_fp16 = const()[name = tensor("op_53752_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4575_cast_fp16 = mul(x = var_53751_cast_fp16, y = var_53752_to_fp16)[name = tensor("aw_4575_cast_fp16")]; + tensor var_53755_equation_0 = const()[name = tensor("op_53755_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_53755_cast_fp16 = einsum(equation = var_53755_equation_0, values = (var_53597_cast_fp16, var_53514_cast_fp16))[name = tensor("op_53755_cast_fp16")]; + tensor var_53756_to_fp16 = const()[name = tensor("op_53756_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4577_cast_fp16 = mul(x = var_53755_cast_fp16, y = var_53756_to_fp16)[name = tensor("aw_4577_cast_fp16")]; + tensor var_53759_equation_0 = const()[name = tensor("op_53759_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_53759_cast_fp16 = einsum(equation = var_53759_equation_0, values = (var_53601_cast_fp16, var_53518_cast_fp16))[name = tensor("op_53759_cast_fp16")]; + tensor var_53760_to_fp16 = const()[name = tensor("op_53760_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4579_cast_fp16 = mul(x = var_53759_cast_fp16, y = var_53760_to_fp16)[name = tensor("aw_4579_cast_fp16")]; + tensor var_53763_equation_0 = const()[name = tensor("op_53763_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_53763_cast_fp16 = einsum(equation = var_53763_equation_0, values = (var_53605_cast_fp16, var_53522_cast_fp16))[name = tensor("op_53763_cast_fp16")]; + tensor var_53764_to_fp16 = const()[name = tensor("op_53764_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4581_cast_fp16 = mul(x = var_53763_cast_fp16, y = var_53764_to_fp16)[name = tensor("aw_4581_cast_fp16")]; + tensor var_53767_equation_0 = const()[name = tensor("op_53767_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_53767_cast_fp16 = einsum(equation = var_53767_equation_0, values = (var_53609_cast_fp16, var_53526_cast_fp16))[name = tensor("op_53767_cast_fp16")]; + tensor var_53768_to_fp16 = const()[name = tensor("op_53768_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4583_cast_fp16 = mul(x = var_53767_cast_fp16, y = var_53768_to_fp16)[name = tensor("aw_4583_cast_fp16")]; + tensor var_53771_equation_0 = const()[name = tensor("op_53771_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_53771_cast_fp16 = einsum(equation = var_53771_equation_0, values = (var_53613_cast_fp16, var_53530_cast_fp16))[name = tensor("op_53771_cast_fp16")]; + tensor var_53772_to_fp16 = const()[name = tensor("op_53772_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4585_cast_fp16 = mul(x = var_53771_cast_fp16, y = var_53772_to_fp16)[name = tensor("aw_4585_cast_fp16")]; + tensor var_53775_equation_0 = const()[name = tensor("op_53775_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_53775_cast_fp16 = einsum(equation = var_53775_equation_0, values = (var_53617_cast_fp16, var_53534_cast_fp16))[name = tensor("op_53775_cast_fp16")]; + tensor var_53776_to_fp16 = const()[name = tensor("op_53776_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4587_cast_fp16 = mul(x = var_53775_cast_fp16, y = var_53776_to_fp16)[name = tensor("aw_4587_cast_fp16")]; + tensor var_53779_equation_0 = const()[name = tensor("op_53779_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_53779_cast_fp16 = einsum(equation = var_53779_equation_0, values = (var_53621_cast_fp16, var_53538_cast_fp16))[name = tensor("op_53779_cast_fp16")]; + tensor var_53780_to_fp16 = const()[name = tensor("op_53780_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4589_cast_fp16 = mul(x = var_53779_cast_fp16, y = var_53780_to_fp16)[name = tensor("aw_4589_cast_fp16")]; + tensor var_53783_equation_0 = const()[name = tensor("op_53783_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_53783_cast_fp16 = einsum(equation = var_53783_equation_0, values = (var_53625_cast_fp16, var_53542_cast_fp16))[name = tensor("op_53783_cast_fp16")]; + tensor var_53784_to_fp16 = const()[name = tensor("op_53784_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4591_cast_fp16 = mul(x = var_53783_cast_fp16, y = var_53784_to_fp16)[name = tensor("aw_4591_cast_fp16")]; + tensor var_53787_equation_0 = const()[name = tensor("op_53787_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_53787_cast_fp16 = einsum(equation = var_53787_equation_0, values = (var_53629_cast_fp16, var_53546_cast_fp16))[name = tensor("op_53787_cast_fp16")]; + tensor var_53788_to_fp16 = const()[name = tensor("op_53788_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4593_cast_fp16 = mul(x = var_53787_cast_fp16, y = var_53788_to_fp16)[name = tensor("aw_4593_cast_fp16")]; + tensor var_53791_equation_0 = const()[name = tensor("op_53791_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_53791_cast_fp16 = einsum(equation = var_53791_equation_0, values = (var_53633_cast_fp16, var_53550_cast_fp16))[name = tensor("op_53791_cast_fp16")]; + tensor var_53792_to_fp16 = const()[name = tensor("op_53792_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4595_cast_fp16 = mul(x = var_53791_cast_fp16, y = var_53792_to_fp16)[name = tensor("aw_4595_cast_fp16")]; + tensor var_53795_equation_0 = const()[name = tensor("op_53795_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_53795_cast_fp16 = einsum(equation = var_53795_equation_0, values = (var_53637_cast_fp16, var_53554_cast_fp16))[name = tensor("op_53795_cast_fp16")]; + tensor var_53796_to_fp16 = const()[name = tensor("op_53796_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4597_cast_fp16 = mul(x = var_53795_cast_fp16, y = var_53796_to_fp16)[name = tensor("aw_4597_cast_fp16")]; + tensor var_53799_equation_0 = const()[name = tensor("op_53799_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_53799_cast_fp16 = einsum(equation = var_53799_equation_0, values = (var_53641_cast_fp16, var_53558_cast_fp16))[name = tensor("op_53799_cast_fp16")]; + tensor var_53800_to_fp16 = const()[name = tensor("op_53800_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4599_cast_fp16 = mul(x = var_53799_cast_fp16, y = var_53800_to_fp16)[name = tensor("aw_4599_cast_fp16")]; + tensor var_53802_cast_fp16 = softmax(axis = var_30355, x = aw_4561_cast_fp16)[name = tensor("op_53802_cast_fp16")]; + tensor var_53803_cast_fp16 = softmax(axis = var_30355, x = aw_4563_cast_fp16)[name = tensor("op_53803_cast_fp16")]; + tensor var_53804_cast_fp16 = softmax(axis = var_30355, x = aw_4565_cast_fp16)[name = tensor("op_53804_cast_fp16")]; + tensor var_53805_cast_fp16 = softmax(axis = var_30355, x = aw_4567_cast_fp16)[name = tensor("op_53805_cast_fp16")]; + tensor var_53806_cast_fp16 = softmax(axis = var_30355, x = aw_4569_cast_fp16)[name = tensor("op_53806_cast_fp16")]; + tensor var_53807_cast_fp16 = softmax(axis = var_30355, x = aw_4571_cast_fp16)[name = tensor("op_53807_cast_fp16")]; + tensor var_53808_cast_fp16 = softmax(axis = var_30355, x = aw_4573_cast_fp16)[name = tensor("op_53808_cast_fp16")]; + tensor var_53809_cast_fp16 = softmax(axis = var_30355, x = aw_4575_cast_fp16)[name = tensor("op_53809_cast_fp16")]; + tensor var_53810_cast_fp16 = softmax(axis = var_30355, x = aw_4577_cast_fp16)[name = tensor("op_53810_cast_fp16")]; + tensor var_53811_cast_fp16 = softmax(axis = var_30355, x = aw_4579_cast_fp16)[name = tensor("op_53811_cast_fp16")]; + tensor var_53812_cast_fp16 = softmax(axis = var_30355, x = aw_4581_cast_fp16)[name = tensor("op_53812_cast_fp16")]; + tensor var_53813_cast_fp16 = softmax(axis = var_30355, x = aw_4583_cast_fp16)[name = tensor("op_53813_cast_fp16")]; + tensor var_53814_cast_fp16 = softmax(axis = var_30355, x = aw_4585_cast_fp16)[name = tensor("op_53814_cast_fp16")]; + tensor var_53815_cast_fp16 = softmax(axis = var_30355, x = aw_4587_cast_fp16)[name = tensor("op_53815_cast_fp16")]; + tensor var_53816_cast_fp16 = softmax(axis = var_30355, x = aw_4589_cast_fp16)[name = tensor("op_53816_cast_fp16")]; + tensor var_53817_cast_fp16 = softmax(axis = var_30355, x = aw_4591_cast_fp16)[name = tensor("op_53817_cast_fp16")]; + tensor var_53818_cast_fp16 = softmax(axis = var_30355, x = aw_4593_cast_fp16)[name = tensor("op_53818_cast_fp16")]; + tensor var_53819_cast_fp16 = softmax(axis = var_30355, x = aw_4595_cast_fp16)[name = tensor("op_53819_cast_fp16")]; + tensor var_53820_cast_fp16 = softmax(axis = var_30355, x = aw_4597_cast_fp16)[name = tensor("op_53820_cast_fp16")]; + tensor var_53821_cast_fp16 = softmax(axis = var_30355, x = aw_4599_cast_fp16)[name = tensor("op_53821_cast_fp16")]; + tensor var_53823_equation_0 = const()[name = tensor("op_53823_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_53823_cast_fp16 = einsum(equation = var_53823_equation_0, values = (var_53643_cast_fp16, var_53802_cast_fp16))[name = tensor("op_53823_cast_fp16")]; + tensor var_53825_equation_0 = const()[name = tensor("op_53825_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_53825_cast_fp16 = einsum(equation = var_53825_equation_0, values = (var_53647_cast_fp16, var_53803_cast_fp16))[name = tensor("op_53825_cast_fp16")]; + tensor var_53827_equation_0 = const()[name = tensor("op_53827_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_53827_cast_fp16 = einsum(equation = var_53827_equation_0, values = (var_53651_cast_fp16, var_53804_cast_fp16))[name = tensor("op_53827_cast_fp16")]; + tensor var_53829_equation_0 = const()[name = tensor("op_53829_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_53829_cast_fp16 = einsum(equation = var_53829_equation_0, values = (var_53655_cast_fp16, var_53805_cast_fp16))[name = tensor("op_53829_cast_fp16")]; + tensor var_53831_equation_0 = const()[name = tensor("op_53831_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_53831_cast_fp16 = einsum(equation = var_53831_equation_0, values = (var_53659_cast_fp16, var_53806_cast_fp16))[name = tensor("op_53831_cast_fp16")]; + tensor var_53833_equation_0 = const()[name = tensor("op_53833_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_53833_cast_fp16 = einsum(equation = var_53833_equation_0, values = (var_53663_cast_fp16, var_53807_cast_fp16))[name = tensor("op_53833_cast_fp16")]; + tensor var_53835_equation_0 = const()[name = tensor("op_53835_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_53835_cast_fp16 = einsum(equation = var_53835_equation_0, values = (var_53667_cast_fp16, var_53808_cast_fp16))[name = tensor("op_53835_cast_fp16")]; + tensor var_53837_equation_0 = const()[name = tensor("op_53837_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_53837_cast_fp16 = einsum(equation = var_53837_equation_0, values = (var_53671_cast_fp16, var_53809_cast_fp16))[name = tensor("op_53837_cast_fp16")]; + tensor var_53839_equation_0 = const()[name = tensor("op_53839_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_53839_cast_fp16 = einsum(equation = var_53839_equation_0, values = (var_53675_cast_fp16, var_53810_cast_fp16))[name = tensor("op_53839_cast_fp16")]; + tensor var_53841_equation_0 = const()[name = tensor("op_53841_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_53841_cast_fp16 = einsum(equation = var_53841_equation_0, values = (var_53679_cast_fp16, var_53811_cast_fp16))[name = tensor("op_53841_cast_fp16")]; + tensor var_53843_equation_0 = const()[name = tensor("op_53843_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_53843_cast_fp16 = einsum(equation = var_53843_equation_0, values = (var_53683_cast_fp16, var_53812_cast_fp16))[name = tensor("op_53843_cast_fp16")]; + tensor var_53845_equation_0 = const()[name = tensor("op_53845_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_53845_cast_fp16 = einsum(equation = var_53845_equation_0, values = (var_53687_cast_fp16, var_53813_cast_fp16))[name = tensor("op_53845_cast_fp16")]; + tensor var_53847_equation_0 = const()[name = tensor("op_53847_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_53847_cast_fp16 = einsum(equation = var_53847_equation_0, values = (var_53691_cast_fp16, var_53814_cast_fp16))[name = tensor("op_53847_cast_fp16")]; + tensor var_53849_equation_0 = const()[name = tensor("op_53849_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_53849_cast_fp16 = einsum(equation = var_53849_equation_0, values = (var_53695_cast_fp16, var_53815_cast_fp16))[name = tensor("op_53849_cast_fp16")]; + tensor var_53851_equation_0 = const()[name = tensor("op_53851_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_53851_cast_fp16 = einsum(equation = var_53851_equation_0, values = (var_53699_cast_fp16, var_53816_cast_fp16))[name = tensor("op_53851_cast_fp16")]; + tensor var_53853_equation_0 = const()[name = tensor("op_53853_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_53853_cast_fp16 = einsum(equation = var_53853_equation_0, values = (var_53703_cast_fp16, var_53817_cast_fp16))[name = tensor("op_53853_cast_fp16")]; + tensor var_53855_equation_0 = const()[name = tensor("op_53855_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_53855_cast_fp16 = einsum(equation = var_53855_equation_0, values = (var_53707_cast_fp16, var_53818_cast_fp16))[name = tensor("op_53855_cast_fp16")]; + tensor var_53857_equation_0 = const()[name = tensor("op_53857_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_53857_cast_fp16 = einsum(equation = var_53857_equation_0, values = (var_53711_cast_fp16, var_53819_cast_fp16))[name = tensor("op_53857_cast_fp16")]; + tensor var_53859_equation_0 = const()[name = tensor("op_53859_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_53859_cast_fp16 = einsum(equation = var_53859_equation_0, values = (var_53715_cast_fp16, var_53820_cast_fp16))[name = tensor("op_53859_cast_fp16")]; + tensor var_53861_equation_0 = const()[name = tensor("op_53861_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_53861_cast_fp16 = einsum(equation = var_53861_equation_0, values = (var_53719_cast_fp16, var_53821_cast_fp16))[name = tensor("op_53861_cast_fp16")]; + tensor input_675_interleave_0 = const()[name = tensor("input_675_interleave_0"), val = tensor(false)]; + tensor input_675_cast_fp16 = concat(axis = var_30355, interleave = input_675_interleave_0, values = (var_53823_cast_fp16, var_53825_cast_fp16, var_53827_cast_fp16, var_53829_cast_fp16, var_53831_cast_fp16, var_53833_cast_fp16, var_53835_cast_fp16, var_53837_cast_fp16, var_53839_cast_fp16, var_53841_cast_fp16, var_53843_cast_fp16, var_53845_cast_fp16, var_53847_cast_fp16, var_53849_cast_fp16, var_53851_cast_fp16, var_53853_cast_fp16, var_53855_cast_fp16, var_53857_cast_fp16, var_53859_cast_fp16, var_53861_cast_fp16))[name = tensor("input_675_cast_fp16")]; + tensor var_53867 = const()[name = tensor("op_53867"), val = tensor([1, 1])]; + tensor var_53869 = const()[name = tensor("op_53869"), val = tensor([1, 1])]; + tensor var_53871_pad_type_0 = const()[name = tensor("op_53871_pad_type_0"), val = tensor("custom")]; + tensor var_53871_pad_0 = const()[name = tensor("op_53871_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_5_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_5_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4529491648)))]; + tensor up_blocks_0_attentions_2_transformer_blocks_5_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_5_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4532768512)))]; + tensor var_53871_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_5_attn1_to_out_0_bias_to_fp16, dilations = var_53869, groups = var_30355, pad = var_53871_pad_0, pad_type = var_53871_pad_type_0, strides = var_53867, weight = up_blocks_0_attentions_2_transformer_blocks_5_attn1_to_out_0_weight_to_fp16, x = input_675_cast_fp16)[name = tensor("op_53871_cast_fp16")]; + tensor inputs_357_cast_fp16 = add(x = var_53871_cast_fp16, y = inputs_355_cast_fp16)[name = tensor("inputs_357_cast_fp16")]; + tensor hidden_states_467_axes_0 = const()[name = tensor("hidden_states_467_axes_0"), val = tensor([1])]; + tensor hidden_states_467_gamma_0_to_fp16 = const()[name = tensor("hidden_states_467_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4532771136)))]; + tensor hidden_states_467_beta_0_to_fp16 = const()[name = tensor("hidden_states_467_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4532773760)))]; + tensor var_53881_to_fp16 = const()[name = tensor("op_53881_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_467_cast_fp16 = layer_norm(axes = hidden_states_467_axes_0, beta = hidden_states_467_beta_0_to_fp16, epsilon = var_53881_to_fp16, gamma = hidden_states_467_gamma_0_to_fp16, x = inputs_357_cast_fp16)[name = tensor("hidden_states_467_cast_fp16")]; + tensor var_53896 = const()[name = tensor("op_53896"), val = tensor([1, 1])]; + tensor var_53898 = const()[name = tensor("op_53898"), val = tensor([1, 1])]; + tensor q_239_pad_type_0 = const()[name = tensor("q_239_pad_type_0"), val = tensor("custom")]; + tensor q_239_pad_0 = const()[name = tensor("q_239_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_5_attn2_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_5_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4532776384)))]; + tensor q_239_cast_fp16 = conv(dilations = var_53898, groups = var_30355, pad = q_239_pad_0, pad_type = q_239_pad_type_0, strides = var_53896, weight = up_blocks_0_attentions_2_transformer_blocks_5_attn2_to_q_weight_to_fp16, x = hidden_states_467_cast_fp16)[name = tensor("q_239_cast_fp16")]; + tensor var_53902 = const()[name = tensor("op_53902"), val = tensor([1, 1])]; + tensor var_53904 = const()[name = tensor("op_53904"), val = tensor([1, 1])]; + tensor k_477_pad_type_0 = const()[name = tensor("k_477_pad_type_0"), val = tensor("custom")]; + tensor k_477_pad_0 = const()[name = tensor("k_477_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_5_attn2_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_5_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4536053248)))]; + tensor k_477_cast_fp16 = conv(dilations = var_53904, groups = var_30355, pad = k_477_pad_0, pad_type = k_477_pad_type_0, strides = var_53902, weight = up_blocks_0_attentions_2_transformer_blocks_5_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_477_cast_fp16")]; + tensor var_53908 = const()[name = tensor("op_53908"), val = tensor([1, 1])]; + tensor var_53910 = const()[name = tensor("op_53910"), val = tensor([1, 1])]; + tensor v_239_pad_type_0 = const()[name = tensor("v_239_pad_type_0"), val = tensor("custom")]; + tensor v_239_pad_0 = const()[name = tensor("v_239_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_5_attn2_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_5_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4541296192)))]; + tensor v_239_cast_fp16 = conv(dilations = var_53910, groups = var_30355, pad = v_239_pad_0, pad_type = v_239_pad_type_0, strides = var_53908, weight = up_blocks_0_attentions_2_transformer_blocks_5_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_239_cast_fp16")]; + tensor var_53914_begin_0 = const()[name = tensor("op_53914_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_53914_end_0 = const()[name = tensor("op_53914_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_53914_end_mask_0 = const()[name = tensor("op_53914_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53914_cast_fp16 = slice_by_index(begin = var_53914_begin_0, end = var_53914_end_0, end_mask = var_53914_end_mask_0, x = q_239_cast_fp16)[name = tensor("op_53914_cast_fp16")]; + tensor var_53918_begin_0 = const()[name = tensor("op_53918_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_53918_end_0 = const()[name = tensor("op_53918_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_53918_end_mask_0 = const()[name = tensor("op_53918_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53918_cast_fp16 = slice_by_index(begin = var_53918_begin_0, end = var_53918_end_0, end_mask = var_53918_end_mask_0, x = q_239_cast_fp16)[name = tensor("op_53918_cast_fp16")]; + tensor var_53922_begin_0 = const()[name = tensor("op_53922_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_53922_end_0 = const()[name = tensor("op_53922_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_53922_end_mask_0 = const()[name = tensor("op_53922_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53922_cast_fp16 = slice_by_index(begin = var_53922_begin_0, end = var_53922_end_0, end_mask = var_53922_end_mask_0, x = q_239_cast_fp16)[name = tensor("op_53922_cast_fp16")]; + tensor var_53926_begin_0 = const()[name = tensor("op_53926_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_53926_end_0 = const()[name = tensor("op_53926_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_53926_end_mask_0 = const()[name = tensor("op_53926_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53926_cast_fp16 = slice_by_index(begin = var_53926_begin_0, end = var_53926_end_0, end_mask = var_53926_end_mask_0, x = q_239_cast_fp16)[name = tensor("op_53926_cast_fp16")]; + tensor var_53930_begin_0 = const()[name = tensor("op_53930_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_53930_end_0 = const()[name = tensor("op_53930_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_53930_end_mask_0 = const()[name = tensor("op_53930_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53930_cast_fp16 = slice_by_index(begin = var_53930_begin_0, end = var_53930_end_0, end_mask = var_53930_end_mask_0, x = q_239_cast_fp16)[name = tensor("op_53930_cast_fp16")]; + tensor var_53934_begin_0 = const()[name = tensor("op_53934_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_53934_end_0 = const()[name = tensor("op_53934_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_53934_end_mask_0 = const()[name = tensor("op_53934_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53934_cast_fp16 = slice_by_index(begin = var_53934_begin_0, end = var_53934_end_0, end_mask = var_53934_end_mask_0, x = q_239_cast_fp16)[name = tensor("op_53934_cast_fp16")]; + tensor var_53938_begin_0 = const()[name = tensor("op_53938_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_53938_end_0 = const()[name = tensor("op_53938_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_53938_end_mask_0 = const()[name = tensor("op_53938_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53938_cast_fp16 = slice_by_index(begin = var_53938_begin_0, end = var_53938_end_0, end_mask = var_53938_end_mask_0, x = q_239_cast_fp16)[name = tensor("op_53938_cast_fp16")]; + tensor var_53942_begin_0 = const()[name = tensor("op_53942_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_53942_end_0 = const()[name = tensor("op_53942_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_53942_end_mask_0 = const()[name = tensor("op_53942_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53942_cast_fp16 = slice_by_index(begin = var_53942_begin_0, end = var_53942_end_0, end_mask = var_53942_end_mask_0, x = q_239_cast_fp16)[name = tensor("op_53942_cast_fp16")]; + tensor var_53946_begin_0 = const()[name = tensor("op_53946_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_53946_end_0 = const()[name = tensor("op_53946_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_53946_end_mask_0 = const()[name = tensor("op_53946_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53946_cast_fp16 = slice_by_index(begin = var_53946_begin_0, end = var_53946_end_0, end_mask = var_53946_end_mask_0, x = q_239_cast_fp16)[name = tensor("op_53946_cast_fp16")]; + tensor var_53950_begin_0 = const()[name = tensor("op_53950_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_53950_end_0 = const()[name = tensor("op_53950_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_53950_end_mask_0 = const()[name = tensor("op_53950_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53950_cast_fp16 = slice_by_index(begin = var_53950_begin_0, end = var_53950_end_0, end_mask = var_53950_end_mask_0, x = q_239_cast_fp16)[name = tensor("op_53950_cast_fp16")]; + tensor var_53954_begin_0 = const()[name = tensor("op_53954_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_53954_end_0 = const()[name = tensor("op_53954_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_53954_end_mask_0 = const()[name = tensor("op_53954_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53954_cast_fp16 = slice_by_index(begin = var_53954_begin_0, end = var_53954_end_0, end_mask = var_53954_end_mask_0, x = q_239_cast_fp16)[name = tensor("op_53954_cast_fp16")]; + tensor var_53958_begin_0 = const()[name = tensor("op_53958_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_53958_end_0 = const()[name = tensor("op_53958_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_53958_end_mask_0 = const()[name = tensor("op_53958_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53958_cast_fp16 = slice_by_index(begin = var_53958_begin_0, end = var_53958_end_0, end_mask = var_53958_end_mask_0, x = q_239_cast_fp16)[name = tensor("op_53958_cast_fp16")]; + tensor var_53962_begin_0 = const()[name = tensor("op_53962_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_53962_end_0 = const()[name = tensor("op_53962_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_53962_end_mask_0 = const()[name = tensor("op_53962_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53962_cast_fp16 = slice_by_index(begin = var_53962_begin_0, end = var_53962_end_0, end_mask = var_53962_end_mask_0, x = q_239_cast_fp16)[name = tensor("op_53962_cast_fp16")]; + tensor var_53966_begin_0 = const()[name = tensor("op_53966_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_53966_end_0 = const()[name = tensor("op_53966_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_53966_end_mask_0 = const()[name = tensor("op_53966_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53966_cast_fp16 = slice_by_index(begin = var_53966_begin_0, end = var_53966_end_0, end_mask = var_53966_end_mask_0, x = q_239_cast_fp16)[name = tensor("op_53966_cast_fp16")]; + tensor var_53970_begin_0 = const()[name = tensor("op_53970_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_53970_end_0 = const()[name = tensor("op_53970_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_53970_end_mask_0 = const()[name = tensor("op_53970_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53970_cast_fp16 = slice_by_index(begin = var_53970_begin_0, end = var_53970_end_0, end_mask = var_53970_end_mask_0, x = q_239_cast_fp16)[name = tensor("op_53970_cast_fp16")]; + tensor var_53974_begin_0 = const()[name = tensor("op_53974_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_53974_end_0 = const()[name = tensor("op_53974_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_53974_end_mask_0 = const()[name = tensor("op_53974_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53974_cast_fp16 = slice_by_index(begin = var_53974_begin_0, end = var_53974_end_0, end_mask = var_53974_end_mask_0, x = q_239_cast_fp16)[name = tensor("op_53974_cast_fp16")]; + tensor var_53978_begin_0 = const()[name = tensor("op_53978_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_53978_end_0 = const()[name = tensor("op_53978_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_53978_end_mask_0 = const()[name = tensor("op_53978_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53978_cast_fp16 = slice_by_index(begin = var_53978_begin_0, end = var_53978_end_0, end_mask = var_53978_end_mask_0, x = q_239_cast_fp16)[name = tensor("op_53978_cast_fp16")]; + tensor var_53982_begin_0 = const()[name = tensor("op_53982_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_53982_end_0 = const()[name = tensor("op_53982_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_53982_end_mask_0 = const()[name = tensor("op_53982_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53982_cast_fp16 = slice_by_index(begin = var_53982_begin_0, end = var_53982_end_0, end_mask = var_53982_end_mask_0, x = q_239_cast_fp16)[name = tensor("op_53982_cast_fp16")]; + tensor var_53986_begin_0 = const()[name = tensor("op_53986_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_53986_end_0 = const()[name = tensor("op_53986_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_53986_end_mask_0 = const()[name = tensor("op_53986_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53986_cast_fp16 = slice_by_index(begin = var_53986_begin_0, end = var_53986_end_0, end_mask = var_53986_end_mask_0, x = q_239_cast_fp16)[name = tensor("op_53986_cast_fp16")]; + tensor var_53990_begin_0 = const()[name = tensor("op_53990_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_53990_end_0 = const()[name = tensor("op_53990_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_53990_end_mask_0 = const()[name = tensor("op_53990_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_53990_cast_fp16 = slice_by_index(begin = var_53990_begin_0, end = var_53990_end_0, end_mask = var_53990_end_mask_0, x = q_239_cast_fp16)[name = tensor("op_53990_cast_fp16")]; + tensor k_479_perm_0 = const()[name = tensor("k_479_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_53997_begin_0 = const()[name = tensor("op_53997_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_53997_end_0 = const()[name = tensor("op_53997_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_53997_end_mask_0 = const()[name = tensor("op_53997_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_479_cast_fp16 = transpose(perm = k_479_perm_0, x = k_477_cast_fp16)[name = tensor("transpose_20")]; + tensor var_53997_cast_fp16 = slice_by_index(begin = var_53997_begin_0, end = var_53997_end_0, end_mask = var_53997_end_mask_0, x = k_479_cast_fp16)[name = tensor("op_53997_cast_fp16")]; + tensor var_54001_begin_0 = const()[name = tensor("op_54001_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_54001_end_0 = const()[name = tensor("op_54001_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_54001_end_mask_0 = const()[name = tensor("op_54001_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_54001_cast_fp16 = slice_by_index(begin = var_54001_begin_0, end = var_54001_end_0, end_mask = var_54001_end_mask_0, x = k_479_cast_fp16)[name = tensor("op_54001_cast_fp16")]; + tensor var_54005_begin_0 = const()[name = tensor("op_54005_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_54005_end_0 = const()[name = tensor("op_54005_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_54005_end_mask_0 = const()[name = tensor("op_54005_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_54005_cast_fp16 = slice_by_index(begin = var_54005_begin_0, end = var_54005_end_0, end_mask = var_54005_end_mask_0, x = k_479_cast_fp16)[name = tensor("op_54005_cast_fp16")]; + tensor var_54009_begin_0 = const()[name = tensor("op_54009_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_54009_end_0 = const()[name = tensor("op_54009_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_54009_end_mask_0 = const()[name = tensor("op_54009_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_54009_cast_fp16 = slice_by_index(begin = var_54009_begin_0, end = var_54009_end_0, end_mask = var_54009_end_mask_0, x = k_479_cast_fp16)[name = tensor("op_54009_cast_fp16")]; + tensor var_54013_begin_0 = const()[name = tensor("op_54013_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_54013_end_0 = const()[name = tensor("op_54013_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_54013_end_mask_0 = const()[name = tensor("op_54013_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_54013_cast_fp16 = slice_by_index(begin = var_54013_begin_0, end = var_54013_end_0, end_mask = var_54013_end_mask_0, x = k_479_cast_fp16)[name = tensor("op_54013_cast_fp16")]; + tensor var_54017_begin_0 = const()[name = tensor("op_54017_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_54017_end_0 = const()[name = tensor("op_54017_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_54017_end_mask_0 = const()[name = tensor("op_54017_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_54017_cast_fp16 = slice_by_index(begin = var_54017_begin_0, end = var_54017_end_0, end_mask = var_54017_end_mask_0, x = k_479_cast_fp16)[name = tensor("op_54017_cast_fp16")]; + tensor var_54021_begin_0 = const()[name = tensor("op_54021_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_54021_end_0 = const()[name = tensor("op_54021_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_54021_end_mask_0 = const()[name = tensor("op_54021_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_54021_cast_fp16 = slice_by_index(begin = var_54021_begin_0, end = var_54021_end_0, end_mask = var_54021_end_mask_0, x = k_479_cast_fp16)[name = tensor("op_54021_cast_fp16")]; + tensor var_54025_begin_0 = const()[name = tensor("op_54025_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_54025_end_0 = const()[name = tensor("op_54025_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_54025_end_mask_0 = const()[name = tensor("op_54025_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_54025_cast_fp16 = slice_by_index(begin = var_54025_begin_0, end = var_54025_end_0, end_mask = var_54025_end_mask_0, x = k_479_cast_fp16)[name = tensor("op_54025_cast_fp16")]; + tensor var_54029_begin_0 = const()[name = tensor("op_54029_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_54029_end_0 = const()[name = tensor("op_54029_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_54029_end_mask_0 = const()[name = tensor("op_54029_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_54029_cast_fp16 = slice_by_index(begin = var_54029_begin_0, end = var_54029_end_0, end_mask = var_54029_end_mask_0, x = k_479_cast_fp16)[name = tensor("op_54029_cast_fp16")]; + tensor var_54033_begin_0 = const()[name = tensor("op_54033_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_54033_end_0 = const()[name = tensor("op_54033_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_54033_end_mask_0 = const()[name = tensor("op_54033_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_54033_cast_fp16 = slice_by_index(begin = var_54033_begin_0, end = var_54033_end_0, end_mask = var_54033_end_mask_0, x = k_479_cast_fp16)[name = tensor("op_54033_cast_fp16")]; + tensor var_54037_begin_0 = const()[name = tensor("op_54037_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_54037_end_0 = const()[name = tensor("op_54037_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_54037_end_mask_0 = const()[name = tensor("op_54037_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_54037_cast_fp16 = slice_by_index(begin = var_54037_begin_0, end = var_54037_end_0, end_mask = var_54037_end_mask_0, x = k_479_cast_fp16)[name = tensor("op_54037_cast_fp16")]; + tensor var_54041_begin_0 = const()[name = tensor("op_54041_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_54041_end_0 = const()[name = tensor("op_54041_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_54041_end_mask_0 = const()[name = tensor("op_54041_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_54041_cast_fp16 = slice_by_index(begin = var_54041_begin_0, end = var_54041_end_0, end_mask = var_54041_end_mask_0, x = k_479_cast_fp16)[name = tensor("op_54041_cast_fp16")]; + tensor var_54045_begin_0 = const()[name = tensor("op_54045_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_54045_end_0 = const()[name = tensor("op_54045_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_54045_end_mask_0 = const()[name = tensor("op_54045_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_54045_cast_fp16 = slice_by_index(begin = var_54045_begin_0, end = var_54045_end_0, end_mask = var_54045_end_mask_0, x = k_479_cast_fp16)[name = tensor("op_54045_cast_fp16")]; + tensor var_54049_begin_0 = const()[name = tensor("op_54049_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_54049_end_0 = const()[name = tensor("op_54049_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_54049_end_mask_0 = const()[name = tensor("op_54049_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_54049_cast_fp16 = slice_by_index(begin = var_54049_begin_0, end = var_54049_end_0, end_mask = var_54049_end_mask_0, x = k_479_cast_fp16)[name = tensor("op_54049_cast_fp16")]; + tensor var_54053_begin_0 = const()[name = tensor("op_54053_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_54053_end_0 = const()[name = tensor("op_54053_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_54053_end_mask_0 = const()[name = tensor("op_54053_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_54053_cast_fp16 = slice_by_index(begin = var_54053_begin_0, end = var_54053_end_0, end_mask = var_54053_end_mask_0, x = k_479_cast_fp16)[name = tensor("op_54053_cast_fp16")]; + tensor var_54057_begin_0 = const()[name = tensor("op_54057_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_54057_end_0 = const()[name = tensor("op_54057_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_54057_end_mask_0 = const()[name = tensor("op_54057_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_54057_cast_fp16 = slice_by_index(begin = var_54057_begin_0, end = var_54057_end_0, end_mask = var_54057_end_mask_0, x = k_479_cast_fp16)[name = tensor("op_54057_cast_fp16")]; + tensor var_54061_begin_0 = const()[name = tensor("op_54061_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_54061_end_0 = const()[name = tensor("op_54061_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_54061_end_mask_0 = const()[name = tensor("op_54061_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_54061_cast_fp16 = slice_by_index(begin = var_54061_begin_0, end = var_54061_end_0, end_mask = var_54061_end_mask_0, x = k_479_cast_fp16)[name = tensor("op_54061_cast_fp16")]; + tensor var_54065_begin_0 = const()[name = tensor("op_54065_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_54065_end_0 = const()[name = tensor("op_54065_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_54065_end_mask_0 = const()[name = tensor("op_54065_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_54065_cast_fp16 = slice_by_index(begin = var_54065_begin_0, end = var_54065_end_0, end_mask = var_54065_end_mask_0, x = k_479_cast_fp16)[name = tensor("op_54065_cast_fp16")]; + tensor var_54069_begin_0 = const()[name = tensor("op_54069_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_54069_end_0 = const()[name = tensor("op_54069_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_54069_end_mask_0 = const()[name = tensor("op_54069_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_54069_cast_fp16 = slice_by_index(begin = var_54069_begin_0, end = var_54069_end_0, end_mask = var_54069_end_mask_0, x = k_479_cast_fp16)[name = tensor("op_54069_cast_fp16")]; + tensor var_54073_begin_0 = const()[name = tensor("op_54073_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_54073_end_0 = const()[name = tensor("op_54073_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_54073_end_mask_0 = const()[name = tensor("op_54073_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_54073_cast_fp16 = slice_by_index(begin = var_54073_begin_0, end = var_54073_end_0, end_mask = var_54073_end_mask_0, x = k_479_cast_fp16)[name = tensor("op_54073_cast_fp16")]; + tensor var_54075_begin_0 = const()[name = tensor("op_54075_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_54075_end_0 = const()[name = tensor("op_54075_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_54075_end_mask_0 = const()[name = tensor("op_54075_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54075_cast_fp16 = slice_by_index(begin = var_54075_begin_0, end = var_54075_end_0, end_mask = var_54075_end_mask_0, x = v_239_cast_fp16)[name = tensor("op_54075_cast_fp16")]; + tensor var_54079_begin_0 = const()[name = tensor("op_54079_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_54079_end_0 = const()[name = tensor("op_54079_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_54079_end_mask_0 = const()[name = tensor("op_54079_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54079_cast_fp16 = slice_by_index(begin = var_54079_begin_0, end = var_54079_end_0, end_mask = var_54079_end_mask_0, x = v_239_cast_fp16)[name = tensor("op_54079_cast_fp16")]; + tensor var_54083_begin_0 = const()[name = tensor("op_54083_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_54083_end_0 = const()[name = tensor("op_54083_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_54083_end_mask_0 = const()[name = tensor("op_54083_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54083_cast_fp16 = slice_by_index(begin = var_54083_begin_0, end = var_54083_end_0, end_mask = var_54083_end_mask_0, x = v_239_cast_fp16)[name = tensor("op_54083_cast_fp16")]; + tensor var_54087_begin_0 = const()[name = tensor("op_54087_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_54087_end_0 = const()[name = tensor("op_54087_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_54087_end_mask_0 = const()[name = tensor("op_54087_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54087_cast_fp16 = slice_by_index(begin = var_54087_begin_0, end = var_54087_end_0, end_mask = var_54087_end_mask_0, x = v_239_cast_fp16)[name = tensor("op_54087_cast_fp16")]; + tensor var_54091_begin_0 = const()[name = tensor("op_54091_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_54091_end_0 = const()[name = tensor("op_54091_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_54091_end_mask_0 = const()[name = tensor("op_54091_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54091_cast_fp16 = slice_by_index(begin = var_54091_begin_0, end = var_54091_end_0, end_mask = var_54091_end_mask_0, x = v_239_cast_fp16)[name = tensor("op_54091_cast_fp16")]; + tensor var_54095_begin_0 = const()[name = tensor("op_54095_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_54095_end_0 = const()[name = tensor("op_54095_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_54095_end_mask_0 = const()[name = tensor("op_54095_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54095_cast_fp16 = slice_by_index(begin = var_54095_begin_0, end = var_54095_end_0, end_mask = var_54095_end_mask_0, x = v_239_cast_fp16)[name = tensor("op_54095_cast_fp16")]; + tensor var_54099_begin_0 = const()[name = tensor("op_54099_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_54099_end_0 = const()[name = tensor("op_54099_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_54099_end_mask_0 = const()[name = tensor("op_54099_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54099_cast_fp16 = slice_by_index(begin = var_54099_begin_0, end = var_54099_end_0, end_mask = var_54099_end_mask_0, x = v_239_cast_fp16)[name = tensor("op_54099_cast_fp16")]; + tensor var_54103_begin_0 = const()[name = tensor("op_54103_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_54103_end_0 = const()[name = tensor("op_54103_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_54103_end_mask_0 = const()[name = tensor("op_54103_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54103_cast_fp16 = slice_by_index(begin = var_54103_begin_0, end = var_54103_end_0, end_mask = var_54103_end_mask_0, x = v_239_cast_fp16)[name = tensor("op_54103_cast_fp16")]; + tensor var_54107_begin_0 = const()[name = tensor("op_54107_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_54107_end_0 = const()[name = tensor("op_54107_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_54107_end_mask_0 = const()[name = tensor("op_54107_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54107_cast_fp16 = slice_by_index(begin = var_54107_begin_0, end = var_54107_end_0, end_mask = var_54107_end_mask_0, x = v_239_cast_fp16)[name = tensor("op_54107_cast_fp16")]; + tensor var_54111_begin_0 = const()[name = tensor("op_54111_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_54111_end_0 = const()[name = tensor("op_54111_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_54111_end_mask_0 = const()[name = tensor("op_54111_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54111_cast_fp16 = slice_by_index(begin = var_54111_begin_0, end = var_54111_end_0, end_mask = var_54111_end_mask_0, x = v_239_cast_fp16)[name = tensor("op_54111_cast_fp16")]; + tensor var_54115_begin_0 = const()[name = tensor("op_54115_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_54115_end_0 = const()[name = tensor("op_54115_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_54115_end_mask_0 = const()[name = tensor("op_54115_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54115_cast_fp16 = slice_by_index(begin = var_54115_begin_0, end = var_54115_end_0, end_mask = var_54115_end_mask_0, x = v_239_cast_fp16)[name = tensor("op_54115_cast_fp16")]; + tensor var_54119_begin_0 = const()[name = tensor("op_54119_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_54119_end_0 = const()[name = tensor("op_54119_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_54119_end_mask_0 = const()[name = tensor("op_54119_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54119_cast_fp16 = slice_by_index(begin = var_54119_begin_0, end = var_54119_end_0, end_mask = var_54119_end_mask_0, x = v_239_cast_fp16)[name = tensor("op_54119_cast_fp16")]; + tensor var_54123_begin_0 = const()[name = tensor("op_54123_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_54123_end_0 = const()[name = tensor("op_54123_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_54123_end_mask_0 = const()[name = tensor("op_54123_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54123_cast_fp16 = slice_by_index(begin = var_54123_begin_0, end = var_54123_end_0, end_mask = var_54123_end_mask_0, x = v_239_cast_fp16)[name = tensor("op_54123_cast_fp16")]; + tensor var_54127_begin_0 = const()[name = tensor("op_54127_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_54127_end_0 = const()[name = tensor("op_54127_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_54127_end_mask_0 = const()[name = tensor("op_54127_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54127_cast_fp16 = slice_by_index(begin = var_54127_begin_0, end = var_54127_end_0, end_mask = var_54127_end_mask_0, x = v_239_cast_fp16)[name = tensor("op_54127_cast_fp16")]; + tensor var_54131_begin_0 = const()[name = tensor("op_54131_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_54131_end_0 = const()[name = tensor("op_54131_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_54131_end_mask_0 = const()[name = tensor("op_54131_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54131_cast_fp16 = slice_by_index(begin = var_54131_begin_0, end = var_54131_end_0, end_mask = var_54131_end_mask_0, x = v_239_cast_fp16)[name = tensor("op_54131_cast_fp16")]; + tensor var_54135_begin_0 = const()[name = tensor("op_54135_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_54135_end_0 = const()[name = tensor("op_54135_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_54135_end_mask_0 = const()[name = tensor("op_54135_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54135_cast_fp16 = slice_by_index(begin = var_54135_begin_0, end = var_54135_end_0, end_mask = var_54135_end_mask_0, x = v_239_cast_fp16)[name = tensor("op_54135_cast_fp16")]; + tensor var_54139_begin_0 = const()[name = tensor("op_54139_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_54139_end_0 = const()[name = tensor("op_54139_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_54139_end_mask_0 = const()[name = tensor("op_54139_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54139_cast_fp16 = slice_by_index(begin = var_54139_begin_0, end = var_54139_end_0, end_mask = var_54139_end_mask_0, x = v_239_cast_fp16)[name = tensor("op_54139_cast_fp16")]; + tensor var_54143_begin_0 = const()[name = tensor("op_54143_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_54143_end_0 = const()[name = tensor("op_54143_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_54143_end_mask_0 = const()[name = tensor("op_54143_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54143_cast_fp16 = slice_by_index(begin = var_54143_begin_0, end = var_54143_end_0, end_mask = var_54143_end_mask_0, x = v_239_cast_fp16)[name = tensor("op_54143_cast_fp16")]; + tensor var_54147_begin_0 = const()[name = tensor("op_54147_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_54147_end_0 = const()[name = tensor("op_54147_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_54147_end_mask_0 = const()[name = tensor("op_54147_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54147_cast_fp16 = slice_by_index(begin = var_54147_begin_0, end = var_54147_end_0, end_mask = var_54147_end_mask_0, x = v_239_cast_fp16)[name = tensor("op_54147_cast_fp16")]; + tensor var_54151_begin_0 = const()[name = tensor("op_54151_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_54151_end_0 = const()[name = tensor("op_54151_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_54151_end_mask_0 = const()[name = tensor("op_54151_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54151_cast_fp16 = slice_by_index(begin = var_54151_begin_0, end = var_54151_end_0, end_mask = var_54151_end_mask_0, x = v_239_cast_fp16)[name = tensor("op_54151_cast_fp16")]; + tensor var_54155_equation_0 = const()[name = tensor("op_54155_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_54155_cast_fp16 = einsum(equation = var_54155_equation_0, values = (var_53997_cast_fp16, var_53914_cast_fp16))[name = tensor("op_54155_cast_fp16")]; + tensor var_54156_to_fp16 = const()[name = tensor("op_54156_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4601_cast_fp16 = mul(x = var_54155_cast_fp16, y = var_54156_to_fp16)[name = tensor("aw_4601_cast_fp16")]; + tensor var_54159_equation_0 = const()[name = tensor("op_54159_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_54159_cast_fp16 = einsum(equation = var_54159_equation_0, values = (var_54001_cast_fp16, var_53918_cast_fp16))[name = tensor("op_54159_cast_fp16")]; + tensor var_54160_to_fp16 = const()[name = tensor("op_54160_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4603_cast_fp16 = mul(x = var_54159_cast_fp16, y = var_54160_to_fp16)[name = tensor("aw_4603_cast_fp16")]; + tensor var_54163_equation_0 = const()[name = tensor("op_54163_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_54163_cast_fp16 = einsum(equation = var_54163_equation_0, values = (var_54005_cast_fp16, var_53922_cast_fp16))[name = tensor("op_54163_cast_fp16")]; + tensor var_54164_to_fp16 = const()[name = tensor("op_54164_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4605_cast_fp16 = mul(x = var_54163_cast_fp16, y = var_54164_to_fp16)[name = tensor("aw_4605_cast_fp16")]; + tensor var_54167_equation_0 = const()[name = tensor("op_54167_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_54167_cast_fp16 = einsum(equation = var_54167_equation_0, values = (var_54009_cast_fp16, var_53926_cast_fp16))[name = tensor("op_54167_cast_fp16")]; + tensor var_54168_to_fp16 = const()[name = tensor("op_54168_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4607_cast_fp16 = mul(x = var_54167_cast_fp16, y = var_54168_to_fp16)[name = tensor("aw_4607_cast_fp16")]; + tensor var_54171_equation_0 = const()[name = tensor("op_54171_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_54171_cast_fp16 = einsum(equation = var_54171_equation_0, values = (var_54013_cast_fp16, var_53930_cast_fp16))[name = tensor("op_54171_cast_fp16")]; + tensor var_54172_to_fp16 = const()[name = tensor("op_54172_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4609_cast_fp16 = mul(x = var_54171_cast_fp16, y = var_54172_to_fp16)[name = tensor("aw_4609_cast_fp16")]; + tensor var_54175_equation_0 = const()[name = tensor("op_54175_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_54175_cast_fp16 = einsum(equation = var_54175_equation_0, values = (var_54017_cast_fp16, var_53934_cast_fp16))[name = tensor("op_54175_cast_fp16")]; + tensor var_54176_to_fp16 = const()[name = tensor("op_54176_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4611_cast_fp16 = mul(x = var_54175_cast_fp16, y = var_54176_to_fp16)[name = tensor("aw_4611_cast_fp16")]; + tensor var_54179_equation_0 = const()[name = tensor("op_54179_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_54179_cast_fp16 = einsum(equation = var_54179_equation_0, values = (var_54021_cast_fp16, var_53938_cast_fp16))[name = tensor("op_54179_cast_fp16")]; + tensor var_54180_to_fp16 = const()[name = tensor("op_54180_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4613_cast_fp16 = mul(x = var_54179_cast_fp16, y = var_54180_to_fp16)[name = tensor("aw_4613_cast_fp16")]; + tensor var_54183_equation_0 = const()[name = tensor("op_54183_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_54183_cast_fp16 = einsum(equation = var_54183_equation_0, values = (var_54025_cast_fp16, var_53942_cast_fp16))[name = tensor("op_54183_cast_fp16")]; + tensor var_54184_to_fp16 = const()[name = tensor("op_54184_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4615_cast_fp16 = mul(x = var_54183_cast_fp16, y = var_54184_to_fp16)[name = tensor("aw_4615_cast_fp16")]; + tensor var_54187_equation_0 = const()[name = tensor("op_54187_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_54187_cast_fp16 = einsum(equation = var_54187_equation_0, values = (var_54029_cast_fp16, var_53946_cast_fp16))[name = tensor("op_54187_cast_fp16")]; + tensor var_54188_to_fp16 = const()[name = tensor("op_54188_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4617_cast_fp16 = mul(x = var_54187_cast_fp16, y = var_54188_to_fp16)[name = tensor("aw_4617_cast_fp16")]; + tensor var_54191_equation_0 = const()[name = tensor("op_54191_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_54191_cast_fp16 = einsum(equation = var_54191_equation_0, values = (var_54033_cast_fp16, var_53950_cast_fp16))[name = tensor("op_54191_cast_fp16")]; + tensor var_54192_to_fp16 = const()[name = tensor("op_54192_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4619_cast_fp16 = mul(x = var_54191_cast_fp16, y = var_54192_to_fp16)[name = tensor("aw_4619_cast_fp16")]; + tensor var_54195_equation_0 = const()[name = tensor("op_54195_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_54195_cast_fp16 = einsum(equation = var_54195_equation_0, values = (var_54037_cast_fp16, var_53954_cast_fp16))[name = tensor("op_54195_cast_fp16")]; + tensor var_54196_to_fp16 = const()[name = tensor("op_54196_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4621_cast_fp16 = mul(x = var_54195_cast_fp16, y = var_54196_to_fp16)[name = tensor("aw_4621_cast_fp16")]; + tensor var_54199_equation_0 = const()[name = tensor("op_54199_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_54199_cast_fp16 = einsum(equation = var_54199_equation_0, values = (var_54041_cast_fp16, var_53958_cast_fp16))[name = tensor("op_54199_cast_fp16")]; + tensor var_54200_to_fp16 = const()[name = tensor("op_54200_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4623_cast_fp16 = mul(x = var_54199_cast_fp16, y = var_54200_to_fp16)[name = tensor("aw_4623_cast_fp16")]; + tensor var_54203_equation_0 = const()[name = tensor("op_54203_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_54203_cast_fp16 = einsum(equation = var_54203_equation_0, values = (var_54045_cast_fp16, var_53962_cast_fp16))[name = tensor("op_54203_cast_fp16")]; + tensor var_54204_to_fp16 = const()[name = tensor("op_54204_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4625_cast_fp16 = mul(x = var_54203_cast_fp16, y = var_54204_to_fp16)[name = tensor("aw_4625_cast_fp16")]; + tensor var_54207_equation_0 = const()[name = tensor("op_54207_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_54207_cast_fp16 = einsum(equation = var_54207_equation_0, values = (var_54049_cast_fp16, var_53966_cast_fp16))[name = tensor("op_54207_cast_fp16")]; + tensor var_54208_to_fp16 = const()[name = tensor("op_54208_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4627_cast_fp16 = mul(x = var_54207_cast_fp16, y = var_54208_to_fp16)[name = tensor("aw_4627_cast_fp16")]; + tensor var_54211_equation_0 = const()[name = tensor("op_54211_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_54211_cast_fp16 = einsum(equation = var_54211_equation_0, values = (var_54053_cast_fp16, var_53970_cast_fp16))[name = tensor("op_54211_cast_fp16")]; + tensor var_54212_to_fp16 = const()[name = tensor("op_54212_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4629_cast_fp16 = mul(x = var_54211_cast_fp16, y = var_54212_to_fp16)[name = tensor("aw_4629_cast_fp16")]; + tensor var_54215_equation_0 = const()[name = tensor("op_54215_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_54215_cast_fp16 = einsum(equation = var_54215_equation_0, values = (var_54057_cast_fp16, var_53974_cast_fp16))[name = tensor("op_54215_cast_fp16")]; + tensor var_54216_to_fp16 = const()[name = tensor("op_54216_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4631_cast_fp16 = mul(x = var_54215_cast_fp16, y = var_54216_to_fp16)[name = tensor("aw_4631_cast_fp16")]; + tensor var_54219_equation_0 = const()[name = tensor("op_54219_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_54219_cast_fp16 = einsum(equation = var_54219_equation_0, values = (var_54061_cast_fp16, var_53978_cast_fp16))[name = tensor("op_54219_cast_fp16")]; + tensor var_54220_to_fp16 = const()[name = tensor("op_54220_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4633_cast_fp16 = mul(x = var_54219_cast_fp16, y = var_54220_to_fp16)[name = tensor("aw_4633_cast_fp16")]; + tensor var_54223_equation_0 = const()[name = tensor("op_54223_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_54223_cast_fp16 = einsum(equation = var_54223_equation_0, values = (var_54065_cast_fp16, var_53982_cast_fp16))[name = tensor("op_54223_cast_fp16")]; + tensor var_54224_to_fp16 = const()[name = tensor("op_54224_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4635_cast_fp16 = mul(x = var_54223_cast_fp16, y = var_54224_to_fp16)[name = tensor("aw_4635_cast_fp16")]; + tensor var_54227_equation_0 = const()[name = tensor("op_54227_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_54227_cast_fp16 = einsum(equation = var_54227_equation_0, values = (var_54069_cast_fp16, var_53986_cast_fp16))[name = tensor("op_54227_cast_fp16")]; + tensor var_54228_to_fp16 = const()[name = tensor("op_54228_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4637_cast_fp16 = mul(x = var_54227_cast_fp16, y = var_54228_to_fp16)[name = tensor("aw_4637_cast_fp16")]; + tensor var_54231_equation_0 = const()[name = tensor("op_54231_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_54231_cast_fp16 = einsum(equation = var_54231_equation_0, values = (var_54073_cast_fp16, var_53990_cast_fp16))[name = tensor("op_54231_cast_fp16")]; + tensor var_54232_to_fp16 = const()[name = tensor("op_54232_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4639_cast_fp16 = mul(x = var_54231_cast_fp16, y = var_54232_to_fp16)[name = tensor("aw_4639_cast_fp16")]; + tensor var_54234_cast_fp16 = softmax(axis = var_30355, x = aw_4601_cast_fp16)[name = tensor("op_54234_cast_fp16")]; + tensor var_54235_cast_fp16 = softmax(axis = var_30355, x = aw_4603_cast_fp16)[name = tensor("op_54235_cast_fp16")]; + tensor var_54236_cast_fp16 = softmax(axis = var_30355, x = aw_4605_cast_fp16)[name = tensor("op_54236_cast_fp16")]; + tensor var_54237_cast_fp16 = softmax(axis = var_30355, x = aw_4607_cast_fp16)[name = tensor("op_54237_cast_fp16")]; + tensor var_54238_cast_fp16 = softmax(axis = var_30355, x = aw_4609_cast_fp16)[name = tensor("op_54238_cast_fp16")]; + tensor var_54239_cast_fp16 = softmax(axis = var_30355, x = aw_4611_cast_fp16)[name = tensor("op_54239_cast_fp16")]; + tensor var_54240_cast_fp16 = softmax(axis = var_30355, x = aw_4613_cast_fp16)[name = tensor("op_54240_cast_fp16")]; + tensor var_54241_cast_fp16 = softmax(axis = var_30355, x = aw_4615_cast_fp16)[name = tensor("op_54241_cast_fp16")]; + tensor var_54242_cast_fp16 = softmax(axis = var_30355, x = aw_4617_cast_fp16)[name = tensor("op_54242_cast_fp16")]; + tensor var_54243_cast_fp16 = softmax(axis = var_30355, x = aw_4619_cast_fp16)[name = tensor("op_54243_cast_fp16")]; + tensor var_54244_cast_fp16 = softmax(axis = var_30355, x = aw_4621_cast_fp16)[name = tensor("op_54244_cast_fp16")]; + tensor var_54245_cast_fp16 = softmax(axis = var_30355, x = aw_4623_cast_fp16)[name = tensor("op_54245_cast_fp16")]; + tensor var_54246_cast_fp16 = softmax(axis = var_30355, x = aw_4625_cast_fp16)[name = tensor("op_54246_cast_fp16")]; + tensor var_54247_cast_fp16 = softmax(axis = var_30355, x = aw_4627_cast_fp16)[name = tensor("op_54247_cast_fp16")]; + tensor var_54248_cast_fp16 = softmax(axis = var_30355, x = aw_4629_cast_fp16)[name = tensor("op_54248_cast_fp16")]; + tensor var_54249_cast_fp16 = softmax(axis = var_30355, x = aw_4631_cast_fp16)[name = tensor("op_54249_cast_fp16")]; + tensor var_54250_cast_fp16 = softmax(axis = var_30355, x = aw_4633_cast_fp16)[name = tensor("op_54250_cast_fp16")]; + tensor var_54251_cast_fp16 = softmax(axis = var_30355, x = aw_4635_cast_fp16)[name = tensor("op_54251_cast_fp16")]; + tensor var_54252_cast_fp16 = softmax(axis = var_30355, x = aw_4637_cast_fp16)[name = tensor("op_54252_cast_fp16")]; + tensor var_54253_cast_fp16 = softmax(axis = var_30355, x = aw_4639_cast_fp16)[name = tensor("op_54253_cast_fp16")]; + tensor var_54255_equation_0 = const()[name = tensor("op_54255_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_54255_cast_fp16 = einsum(equation = var_54255_equation_0, values = (var_54075_cast_fp16, var_54234_cast_fp16))[name = tensor("op_54255_cast_fp16")]; + tensor var_54257_equation_0 = const()[name = tensor("op_54257_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_54257_cast_fp16 = einsum(equation = var_54257_equation_0, values = (var_54079_cast_fp16, var_54235_cast_fp16))[name = tensor("op_54257_cast_fp16")]; + tensor var_54259_equation_0 = const()[name = tensor("op_54259_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_54259_cast_fp16 = einsum(equation = var_54259_equation_0, values = (var_54083_cast_fp16, var_54236_cast_fp16))[name = tensor("op_54259_cast_fp16")]; + tensor var_54261_equation_0 = const()[name = tensor("op_54261_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_54261_cast_fp16 = einsum(equation = var_54261_equation_0, values = (var_54087_cast_fp16, var_54237_cast_fp16))[name = tensor("op_54261_cast_fp16")]; + tensor var_54263_equation_0 = const()[name = tensor("op_54263_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_54263_cast_fp16 = einsum(equation = var_54263_equation_0, values = (var_54091_cast_fp16, var_54238_cast_fp16))[name = tensor("op_54263_cast_fp16")]; + tensor var_54265_equation_0 = const()[name = tensor("op_54265_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_54265_cast_fp16 = einsum(equation = var_54265_equation_0, values = (var_54095_cast_fp16, var_54239_cast_fp16))[name = tensor("op_54265_cast_fp16")]; + tensor var_54267_equation_0 = const()[name = tensor("op_54267_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_54267_cast_fp16 = einsum(equation = var_54267_equation_0, values = (var_54099_cast_fp16, var_54240_cast_fp16))[name = tensor("op_54267_cast_fp16")]; + tensor var_54269_equation_0 = const()[name = tensor("op_54269_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_54269_cast_fp16 = einsum(equation = var_54269_equation_0, values = (var_54103_cast_fp16, var_54241_cast_fp16))[name = tensor("op_54269_cast_fp16")]; + tensor var_54271_equation_0 = const()[name = tensor("op_54271_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_54271_cast_fp16 = einsum(equation = var_54271_equation_0, values = (var_54107_cast_fp16, var_54242_cast_fp16))[name = tensor("op_54271_cast_fp16")]; + tensor var_54273_equation_0 = const()[name = tensor("op_54273_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_54273_cast_fp16 = einsum(equation = var_54273_equation_0, values = (var_54111_cast_fp16, var_54243_cast_fp16))[name = tensor("op_54273_cast_fp16")]; + tensor var_54275_equation_0 = const()[name = tensor("op_54275_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_54275_cast_fp16 = einsum(equation = var_54275_equation_0, values = (var_54115_cast_fp16, var_54244_cast_fp16))[name = tensor("op_54275_cast_fp16")]; + tensor var_54277_equation_0 = const()[name = tensor("op_54277_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_54277_cast_fp16 = einsum(equation = var_54277_equation_0, values = (var_54119_cast_fp16, var_54245_cast_fp16))[name = tensor("op_54277_cast_fp16")]; + tensor var_54279_equation_0 = const()[name = tensor("op_54279_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_54279_cast_fp16 = einsum(equation = var_54279_equation_0, values = (var_54123_cast_fp16, var_54246_cast_fp16))[name = tensor("op_54279_cast_fp16")]; + tensor var_54281_equation_0 = const()[name = tensor("op_54281_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_54281_cast_fp16 = einsum(equation = var_54281_equation_0, values = (var_54127_cast_fp16, var_54247_cast_fp16))[name = tensor("op_54281_cast_fp16")]; + tensor var_54283_equation_0 = const()[name = tensor("op_54283_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_54283_cast_fp16 = einsum(equation = var_54283_equation_0, values = (var_54131_cast_fp16, var_54248_cast_fp16))[name = tensor("op_54283_cast_fp16")]; + tensor var_54285_equation_0 = const()[name = tensor("op_54285_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_54285_cast_fp16 = einsum(equation = var_54285_equation_0, values = (var_54135_cast_fp16, var_54249_cast_fp16))[name = tensor("op_54285_cast_fp16")]; + tensor var_54287_equation_0 = const()[name = tensor("op_54287_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_54287_cast_fp16 = einsum(equation = var_54287_equation_0, values = (var_54139_cast_fp16, var_54250_cast_fp16))[name = tensor("op_54287_cast_fp16")]; + tensor var_54289_equation_0 = const()[name = tensor("op_54289_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_54289_cast_fp16 = einsum(equation = var_54289_equation_0, values = (var_54143_cast_fp16, var_54251_cast_fp16))[name = tensor("op_54289_cast_fp16")]; + tensor var_54291_equation_0 = const()[name = tensor("op_54291_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_54291_cast_fp16 = einsum(equation = var_54291_equation_0, values = (var_54147_cast_fp16, var_54252_cast_fp16))[name = tensor("op_54291_cast_fp16")]; + tensor var_54293_equation_0 = const()[name = tensor("op_54293_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_54293_cast_fp16 = einsum(equation = var_54293_equation_0, values = (var_54151_cast_fp16, var_54253_cast_fp16))[name = tensor("op_54293_cast_fp16")]; + tensor input_677_interleave_0 = const()[name = tensor("input_677_interleave_0"), val = tensor(false)]; + tensor input_677_cast_fp16 = concat(axis = var_30355, interleave = input_677_interleave_0, values = (var_54255_cast_fp16, var_54257_cast_fp16, var_54259_cast_fp16, var_54261_cast_fp16, var_54263_cast_fp16, var_54265_cast_fp16, var_54267_cast_fp16, var_54269_cast_fp16, var_54271_cast_fp16, var_54273_cast_fp16, var_54275_cast_fp16, var_54277_cast_fp16, var_54279_cast_fp16, var_54281_cast_fp16, var_54283_cast_fp16, var_54285_cast_fp16, var_54287_cast_fp16, var_54289_cast_fp16, var_54291_cast_fp16, var_54293_cast_fp16))[name = tensor("input_677_cast_fp16")]; + tensor var_54299 = const()[name = tensor("op_54299"), val = tensor([1, 1])]; + tensor var_54301 = const()[name = tensor("op_54301"), val = tensor([1, 1])]; + tensor var_54303_pad_type_0 = const()[name = tensor("op_54303_pad_type_0"), val = tensor("custom")]; + tensor var_54303_pad_0 = const()[name = tensor("op_54303_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_5_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_5_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4546539136)))]; + tensor up_blocks_0_attentions_2_transformer_blocks_5_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_5_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4549816000)))]; + tensor var_54303_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_5_attn2_to_out_0_bias_to_fp16, dilations = var_54301, groups = var_30355, pad = var_54303_pad_0, pad_type = var_54303_pad_type_0, strides = var_54299, weight = up_blocks_0_attentions_2_transformer_blocks_5_attn2_to_out_0_weight_to_fp16, x = input_677_cast_fp16)[name = tensor("op_54303_cast_fp16")]; + tensor inputs_359_cast_fp16 = add(x = var_54303_cast_fp16, y = inputs_357_cast_fp16)[name = tensor("inputs_359_cast_fp16")]; + tensor input_679_axes_0 = const()[name = tensor("input_679_axes_0"), val = tensor([1])]; + tensor input_679_gamma_0_to_fp16 = const()[name = tensor("input_679_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4549818624)))]; + tensor input_679_beta_0_to_fp16 = const()[name = tensor("input_679_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4549821248)))]; + tensor var_54313_to_fp16 = const()[name = tensor("op_54313_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_679_cast_fp16 = layer_norm(axes = input_679_axes_0, beta = input_679_beta_0_to_fp16, epsilon = var_54313_to_fp16, gamma = input_679_gamma_0_to_fp16, x = inputs_359_cast_fp16)[name = tensor("input_679_cast_fp16")]; + tensor var_54329 = const()[name = tensor("op_54329"), val = tensor([1, 1])]; + tensor var_54331 = const()[name = tensor("op_54331"), val = tensor([1, 1])]; + tensor var_54333_pad_type_0 = const()[name = tensor("op_54333_pad_type_0"), val = tensor("custom")]; + tensor var_54333_pad_0 = const()[name = tensor("op_54333_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_5_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_5_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4549823872)))]; + tensor up_blocks_0_attentions_2_transformer_blocks_5_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_5_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4576038336)))]; + tensor var_54333_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_5_ff_net_0_proj_bias_to_fp16, dilations = var_54331, groups = var_30355, pad = var_54333_pad_0, pad_type = var_54333_pad_type_0, strides = var_54329, weight = up_blocks_0_attentions_2_transformer_blocks_5_ff_net_0_proj_weight_to_fp16, x = input_679_cast_fp16)[name = tensor("op_54333_cast_fp16")]; + tensor var_54334_split_sizes_0 = const()[name = tensor("op_54334_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_54334_axis_0 = const()[name = tensor("op_54334_axis_0"), val = tensor(1)]; + tensor var_54334_cast_fp16_0, tensor var_54334_cast_fp16_1 = split(axis = var_54334_axis_0, split_sizes = var_54334_split_sizes_0, x = var_54333_cast_fp16)[name = tensor("op_54334_cast_fp16")]; + tensor var_54336_mode_0 = const()[name = tensor("op_54336_mode_0"), val = tensor("EXACT")]; + tensor var_54336_cast_fp16 = gelu(mode = var_54336_mode_0, x = var_54334_cast_fp16_1)[name = tensor("op_54336_cast_fp16")]; + tensor input_681_cast_fp16 = mul(x = var_54334_cast_fp16_0, y = var_54336_cast_fp16)[name = tensor("input_681_cast_fp16")]; + tensor var_54340 = const()[name = tensor("op_54340"), val = tensor([1, 1])]; + tensor var_54342 = const()[name = tensor("op_54342"), val = tensor([1, 1])]; + tensor var_54344_pad_type_0 = const()[name = tensor("op_54344_pad_type_0"), val = tensor("custom")]; + tensor var_54344_pad_0 = const()[name = tensor("op_54344_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_5_ff_net_2_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_5_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4576058880)))]; + tensor up_blocks_0_attentions_2_transformer_blocks_5_ff_net_2_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_5_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4589166144)))]; + tensor var_54344_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_5_ff_net_2_bias_to_fp16, dilations = var_54342, groups = var_30355, pad = var_54344_pad_0, pad_type = var_54344_pad_type_0, strides = var_54340, weight = up_blocks_0_attentions_2_transformer_blocks_5_ff_net_2_weight_to_fp16, x = input_681_cast_fp16)[name = tensor("op_54344_cast_fp16")]; + tensor inputs_361_cast_fp16 = add(x = var_54344_cast_fp16, y = inputs_359_cast_fp16)[name = tensor("inputs_361_cast_fp16")]; + tensor hidden_states_471_axes_0 = const()[name = tensor("hidden_states_471_axes_0"), val = tensor([1])]; + tensor hidden_states_471_gamma_0_to_fp16 = const()[name = tensor("hidden_states_471_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4589168768)))]; + tensor hidden_states_471_beta_0_to_fp16 = const()[name = tensor("hidden_states_471_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4589171392)))]; + tensor var_54360_to_fp16 = const()[name = tensor("op_54360_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_471_cast_fp16 = layer_norm(axes = hidden_states_471_axes_0, beta = hidden_states_471_beta_0_to_fp16, epsilon = var_54360_to_fp16, gamma = hidden_states_471_gamma_0_to_fp16, x = inputs_361_cast_fp16)[name = tensor("hidden_states_471_cast_fp16")]; + tensor var_54375 = const()[name = tensor("op_54375"), val = tensor([1, 1])]; + tensor var_54377 = const()[name = tensor("op_54377"), val = tensor([1, 1])]; + tensor q_241_pad_type_0 = const()[name = tensor("q_241_pad_type_0"), val = tensor("custom")]; + tensor q_241_pad_0 = const()[name = tensor("q_241_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_6_attn1_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_6_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4589174016)))]; + tensor q_241_cast_fp16 = conv(dilations = var_54377, groups = var_30355, pad = q_241_pad_0, pad_type = q_241_pad_type_0, strides = var_54375, weight = up_blocks_0_attentions_2_transformer_blocks_6_attn1_to_q_weight_to_fp16, x = hidden_states_471_cast_fp16)[name = tensor("q_241_cast_fp16")]; + tensor var_54381 = const()[name = tensor("op_54381"), val = tensor([1, 1])]; + tensor var_54383 = const()[name = tensor("op_54383"), val = tensor([1, 1])]; + tensor k_481_pad_type_0 = const()[name = tensor("k_481_pad_type_0"), val = tensor("custom")]; + tensor k_481_pad_0 = const()[name = tensor("k_481_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_6_attn1_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_6_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4592450880)))]; + tensor k_481_cast_fp16 = conv(dilations = var_54383, groups = var_30355, pad = k_481_pad_0, pad_type = k_481_pad_type_0, strides = var_54381, weight = up_blocks_0_attentions_2_transformer_blocks_6_attn1_to_k_weight_to_fp16, x = hidden_states_471_cast_fp16)[name = tensor("k_481_cast_fp16")]; + tensor var_54387 = const()[name = tensor("op_54387"), val = tensor([1, 1])]; + tensor var_54389 = const()[name = tensor("op_54389"), val = tensor([1, 1])]; + tensor v_241_pad_type_0 = const()[name = tensor("v_241_pad_type_0"), val = tensor("custom")]; + tensor v_241_pad_0 = const()[name = tensor("v_241_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_6_attn1_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_6_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4595727744)))]; + tensor v_241_cast_fp16 = conv(dilations = var_54389, groups = var_30355, pad = v_241_pad_0, pad_type = v_241_pad_type_0, strides = var_54387, weight = up_blocks_0_attentions_2_transformer_blocks_6_attn1_to_v_weight_to_fp16, x = hidden_states_471_cast_fp16)[name = tensor("v_241_cast_fp16")]; + tensor var_54393_begin_0 = const()[name = tensor("op_54393_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_54393_end_0 = const()[name = tensor("op_54393_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_54393_end_mask_0 = const()[name = tensor("op_54393_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54393_cast_fp16 = slice_by_index(begin = var_54393_begin_0, end = var_54393_end_0, end_mask = var_54393_end_mask_0, x = q_241_cast_fp16)[name = tensor("op_54393_cast_fp16")]; + tensor var_54397_begin_0 = const()[name = tensor("op_54397_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_54397_end_0 = const()[name = tensor("op_54397_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_54397_end_mask_0 = const()[name = tensor("op_54397_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54397_cast_fp16 = slice_by_index(begin = var_54397_begin_0, end = var_54397_end_0, end_mask = var_54397_end_mask_0, x = q_241_cast_fp16)[name = tensor("op_54397_cast_fp16")]; + tensor var_54401_begin_0 = const()[name = tensor("op_54401_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_54401_end_0 = const()[name = tensor("op_54401_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_54401_end_mask_0 = const()[name = tensor("op_54401_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54401_cast_fp16 = slice_by_index(begin = var_54401_begin_0, end = var_54401_end_0, end_mask = var_54401_end_mask_0, x = q_241_cast_fp16)[name = tensor("op_54401_cast_fp16")]; + tensor var_54405_begin_0 = const()[name = tensor("op_54405_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_54405_end_0 = const()[name = tensor("op_54405_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_54405_end_mask_0 = const()[name = tensor("op_54405_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54405_cast_fp16 = slice_by_index(begin = var_54405_begin_0, end = var_54405_end_0, end_mask = var_54405_end_mask_0, x = q_241_cast_fp16)[name = tensor("op_54405_cast_fp16")]; + tensor var_54409_begin_0 = const()[name = tensor("op_54409_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_54409_end_0 = const()[name = tensor("op_54409_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_54409_end_mask_0 = const()[name = tensor("op_54409_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54409_cast_fp16 = slice_by_index(begin = var_54409_begin_0, end = var_54409_end_0, end_mask = var_54409_end_mask_0, x = q_241_cast_fp16)[name = tensor("op_54409_cast_fp16")]; + tensor var_54413_begin_0 = const()[name = tensor("op_54413_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_54413_end_0 = const()[name = tensor("op_54413_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_54413_end_mask_0 = const()[name = tensor("op_54413_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54413_cast_fp16 = slice_by_index(begin = var_54413_begin_0, end = var_54413_end_0, end_mask = var_54413_end_mask_0, x = q_241_cast_fp16)[name = tensor("op_54413_cast_fp16")]; + tensor var_54417_begin_0 = const()[name = tensor("op_54417_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_54417_end_0 = const()[name = tensor("op_54417_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_54417_end_mask_0 = const()[name = tensor("op_54417_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54417_cast_fp16 = slice_by_index(begin = var_54417_begin_0, end = var_54417_end_0, end_mask = var_54417_end_mask_0, x = q_241_cast_fp16)[name = tensor("op_54417_cast_fp16")]; + tensor var_54421_begin_0 = const()[name = tensor("op_54421_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_54421_end_0 = const()[name = tensor("op_54421_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_54421_end_mask_0 = const()[name = tensor("op_54421_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54421_cast_fp16 = slice_by_index(begin = var_54421_begin_0, end = var_54421_end_0, end_mask = var_54421_end_mask_0, x = q_241_cast_fp16)[name = tensor("op_54421_cast_fp16")]; + tensor var_54425_begin_0 = const()[name = tensor("op_54425_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_54425_end_0 = const()[name = tensor("op_54425_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_54425_end_mask_0 = const()[name = tensor("op_54425_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54425_cast_fp16 = slice_by_index(begin = var_54425_begin_0, end = var_54425_end_0, end_mask = var_54425_end_mask_0, x = q_241_cast_fp16)[name = tensor("op_54425_cast_fp16")]; + tensor var_54429_begin_0 = const()[name = tensor("op_54429_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_54429_end_0 = const()[name = tensor("op_54429_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_54429_end_mask_0 = const()[name = tensor("op_54429_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54429_cast_fp16 = slice_by_index(begin = var_54429_begin_0, end = var_54429_end_0, end_mask = var_54429_end_mask_0, x = q_241_cast_fp16)[name = tensor("op_54429_cast_fp16")]; + tensor var_54433_begin_0 = const()[name = tensor("op_54433_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_54433_end_0 = const()[name = tensor("op_54433_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_54433_end_mask_0 = const()[name = tensor("op_54433_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54433_cast_fp16 = slice_by_index(begin = var_54433_begin_0, end = var_54433_end_0, end_mask = var_54433_end_mask_0, x = q_241_cast_fp16)[name = tensor("op_54433_cast_fp16")]; + tensor var_54437_begin_0 = const()[name = tensor("op_54437_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_54437_end_0 = const()[name = tensor("op_54437_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_54437_end_mask_0 = const()[name = tensor("op_54437_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54437_cast_fp16 = slice_by_index(begin = var_54437_begin_0, end = var_54437_end_0, end_mask = var_54437_end_mask_0, x = q_241_cast_fp16)[name = tensor("op_54437_cast_fp16")]; + tensor var_54441_begin_0 = const()[name = tensor("op_54441_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_54441_end_0 = const()[name = tensor("op_54441_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_54441_end_mask_0 = const()[name = tensor("op_54441_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54441_cast_fp16 = slice_by_index(begin = var_54441_begin_0, end = var_54441_end_0, end_mask = var_54441_end_mask_0, x = q_241_cast_fp16)[name = tensor("op_54441_cast_fp16")]; + tensor var_54445_begin_0 = const()[name = tensor("op_54445_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_54445_end_0 = const()[name = tensor("op_54445_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_54445_end_mask_0 = const()[name = tensor("op_54445_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54445_cast_fp16 = slice_by_index(begin = var_54445_begin_0, end = var_54445_end_0, end_mask = var_54445_end_mask_0, x = q_241_cast_fp16)[name = tensor("op_54445_cast_fp16")]; + tensor var_54449_begin_0 = const()[name = tensor("op_54449_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_54449_end_0 = const()[name = tensor("op_54449_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_54449_end_mask_0 = const()[name = tensor("op_54449_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54449_cast_fp16 = slice_by_index(begin = var_54449_begin_0, end = var_54449_end_0, end_mask = var_54449_end_mask_0, x = q_241_cast_fp16)[name = tensor("op_54449_cast_fp16")]; + tensor var_54453_begin_0 = const()[name = tensor("op_54453_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_54453_end_0 = const()[name = tensor("op_54453_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_54453_end_mask_0 = const()[name = tensor("op_54453_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54453_cast_fp16 = slice_by_index(begin = var_54453_begin_0, end = var_54453_end_0, end_mask = var_54453_end_mask_0, x = q_241_cast_fp16)[name = tensor("op_54453_cast_fp16")]; + tensor var_54457_begin_0 = const()[name = tensor("op_54457_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_54457_end_0 = const()[name = tensor("op_54457_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_54457_end_mask_0 = const()[name = tensor("op_54457_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54457_cast_fp16 = slice_by_index(begin = var_54457_begin_0, end = var_54457_end_0, end_mask = var_54457_end_mask_0, x = q_241_cast_fp16)[name = tensor("op_54457_cast_fp16")]; + tensor var_54461_begin_0 = const()[name = tensor("op_54461_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_54461_end_0 = const()[name = tensor("op_54461_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_54461_end_mask_0 = const()[name = tensor("op_54461_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54461_cast_fp16 = slice_by_index(begin = var_54461_begin_0, end = var_54461_end_0, end_mask = var_54461_end_mask_0, x = q_241_cast_fp16)[name = tensor("op_54461_cast_fp16")]; + tensor var_54465_begin_0 = const()[name = tensor("op_54465_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_54465_end_0 = const()[name = tensor("op_54465_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_54465_end_mask_0 = const()[name = tensor("op_54465_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54465_cast_fp16 = slice_by_index(begin = var_54465_begin_0, end = var_54465_end_0, end_mask = var_54465_end_mask_0, x = q_241_cast_fp16)[name = tensor("op_54465_cast_fp16")]; + tensor var_54469_begin_0 = const()[name = tensor("op_54469_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_54469_end_0 = const()[name = tensor("op_54469_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_54469_end_mask_0 = const()[name = tensor("op_54469_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54469_cast_fp16 = slice_by_index(begin = var_54469_begin_0, end = var_54469_end_0, end_mask = var_54469_end_mask_0, x = q_241_cast_fp16)[name = tensor("op_54469_cast_fp16")]; + tensor k_483_perm_0 = const()[name = tensor("k_483_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_54476_begin_0 = const()[name = tensor("op_54476_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_54476_end_0 = const()[name = tensor("op_54476_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_54476_end_mask_0 = const()[name = tensor("op_54476_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_483_cast_fp16 = transpose(perm = k_483_perm_0, x = k_481_cast_fp16)[name = tensor("transpose_19")]; + tensor var_54476_cast_fp16 = slice_by_index(begin = var_54476_begin_0, end = var_54476_end_0, end_mask = var_54476_end_mask_0, x = k_483_cast_fp16)[name = tensor("op_54476_cast_fp16")]; + tensor var_54480_begin_0 = const()[name = tensor("op_54480_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_54480_end_0 = const()[name = tensor("op_54480_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_54480_end_mask_0 = const()[name = tensor("op_54480_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_54480_cast_fp16 = slice_by_index(begin = var_54480_begin_0, end = var_54480_end_0, end_mask = var_54480_end_mask_0, x = k_483_cast_fp16)[name = tensor("op_54480_cast_fp16")]; + tensor var_54484_begin_0 = const()[name = tensor("op_54484_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_54484_end_0 = const()[name = tensor("op_54484_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_54484_end_mask_0 = const()[name = tensor("op_54484_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_54484_cast_fp16 = slice_by_index(begin = var_54484_begin_0, end = var_54484_end_0, end_mask = var_54484_end_mask_0, x = k_483_cast_fp16)[name = tensor("op_54484_cast_fp16")]; + tensor var_54488_begin_0 = const()[name = tensor("op_54488_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_54488_end_0 = const()[name = tensor("op_54488_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_54488_end_mask_0 = const()[name = tensor("op_54488_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_54488_cast_fp16 = slice_by_index(begin = var_54488_begin_0, end = var_54488_end_0, end_mask = var_54488_end_mask_0, x = k_483_cast_fp16)[name = tensor("op_54488_cast_fp16")]; + tensor var_54492_begin_0 = const()[name = tensor("op_54492_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_54492_end_0 = const()[name = tensor("op_54492_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_54492_end_mask_0 = const()[name = tensor("op_54492_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_54492_cast_fp16 = slice_by_index(begin = var_54492_begin_0, end = var_54492_end_0, end_mask = var_54492_end_mask_0, x = k_483_cast_fp16)[name = tensor("op_54492_cast_fp16")]; + tensor var_54496_begin_0 = const()[name = tensor("op_54496_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_54496_end_0 = const()[name = tensor("op_54496_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_54496_end_mask_0 = const()[name = tensor("op_54496_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_54496_cast_fp16 = slice_by_index(begin = var_54496_begin_0, end = var_54496_end_0, end_mask = var_54496_end_mask_0, x = k_483_cast_fp16)[name = tensor("op_54496_cast_fp16")]; + tensor var_54500_begin_0 = const()[name = tensor("op_54500_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_54500_end_0 = const()[name = tensor("op_54500_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_54500_end_mask_0 = const()[name = tensor("op_54500_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_54500_cast_fp16 = slice_by_index(begin = var_54500_begin_0, end = var_54500_end_0, end_mask = var_54500_end_mask_0, x = k_483_cast_fp16)[name = tensor("op_54500_cast_fp16")]; + tensor var_54504_begin_0 = const()[name = tensor("op_54504_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_54504_end_0 = const()[name = tensor("op_54504_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_54504_end_mask_0 = const()[name = tensor("op_54504_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_54504_cast_fp16 = slice_by_index(begin = var_54504_begin_0, end = var_54504_end_0, end_mask = var_54504_end_mask_0, x = k_483_cast_fp16)[name = tensor("op_54504_cast_fp16")]; + tensor var_54508_begin_0 = const()[name = tensor("op_54508_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_54508_end_0 = const()[name = tensor("op_54508_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_54508_end_mask_0 = const()[name = tensor("op_54508_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_54508_cast_fp16 = slice_by_index(begin = var_54508_begin_0, end = var_54508_end_0, end_mask = var_54508_end_mask_0, x = k_483_cast_fp16)[name = tensor("op_54508_cast_fp16")]; + tensor var_54512_begin_0 = const()[name = tensor("op_54512_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_54512_end_0 = const()[name = tensor("op_54512_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_54512_end_mask_0 = const()[name = tensor("op_54512_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_54512_cast_fp16 = slice_by_index(begin = var_54512_begin_0, end = var_54512_end_0, end_mask = var_54512_end_mask_0, x = k_483_cast_fp16)[name = tensor("op_54512_cast_fp16")]; + tensor var_54516_begin_0 = const()[name = tensor("op_54516_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_54516_end_0 = const()[name = tensor("op_54516_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_54516_end_mask_0 = const()[name = tensor("op_54516_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_54516_cast_fp16 = slice_by_index(begin = var_54516_begin_0, end = var_54516_end_0, end_mask = var_54516_end_mask_0, x = k_483_cast_fp16)[name = tensor("op_54516_cast_fp16")]; + tensor var_54520_begin_0 = const()[name = tensor("op_54520_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_54520_end_0 = const()[name = tensor("op_54520_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_54520_end_mask_0 = const()[name = tensor("op_54520_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_54520_cast_fp16 = slice_by_index(begin = var_54520_begin_0, end = var_54520_end_0, end_mask = var_54520_end_mask_0, x = k_483_cast_fp16)[name = tensor("op_54520_cast_fp16")]; + tensor var_54524_begin_0 = const()[name = tensor("op_54524_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_54524_end_0 = const()[name = tensor("op_54524_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_54524_end_mask_0 = const()[name = tensor("op_54524_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_54524_cast_fp16 = slice_by_index(begin = var_54524_begin_0, end = var_54524_end_0, end_mask = var_54524_end_mask_0, x = k_483_cast_fp16)[name = tensor("op_54524_cast_fp16")]; + tensor var_54528_begin_0 = const()[name = tensor("op_54528_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_54528_end_0 = const()[name = tensor("op_54528_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_54528_end_mask_0 = const()[name = tensor("op_54528_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_54528_cast_fp16 = slice_by_index(begin = var_54528_begin_0, end = var_54528_end_0, end_mask = var_54528_end_mask_0, x = k_483_cast_fp16)[name = tensor("op_54528_cast_fp16")]; + tensor var_54532_begin_0 = const()[name = tensor("op_54532_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_54532_end_0 = const()[name = tensor("op_54532_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_54532_end_mask_0 = const()[name = tensor("op_54532_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_54532_cast_fp16 = slice_by_index(begin = var_54532_begin_0, end = var_54532_end_0, end_mask = var_54532_end_mask_0, x = k_483_cast_fp16)[name = tensor("op_54532_cast_fp16")]; + tensor var_54536_begin_0 = const()[name = tensor("op_54536_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_54536_end_0 = const()[name = tensor("op_54536_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_54536_end_mask_0 = const()[name = tensor("op_54536_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_54536_cast_fp16 = slice_by_index(begin = var_54536_begin_0, end = var_54536_end_0, end_mask = var_54536_end_mask_0, x = k_483_cast_fp16)[name = tensor("op_54536_cast_fp16")]; + tensor var_54540_begin_0 = const()[name = tensor("op_54540_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_54540_end_0 = const()[name = tensor("op_54540_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_54540_end_mask_0 = const()[name = tensor("op_54540_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_54540_cast_fp16 = slice_by_index(begin = var_54540_begin_0, end = var_54540_end_0, end_mask = var_54540_end_mask_0, x = k_483_cast_fp16)[name = tensor("op_54540_cast_fp16")]; + tensor var_54544_begin_0 = const()[name = tensor("op_54544_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_54544_end_0 = const()[name = tensor("op_54544_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_54544_end_mask_0 = const()[name = tensor("op_54544_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_54544_cast_fp16 = slice_by_index(begin = var_54544_begin_0, end = var_54544_end_0, end_mask = var_54544_end_mask_0, x = k_483_cast_fp16)[name = tensor("op_54544_cast_fp16")]; + tensor var_54548_begin_0 = const()[name = tensor("op_54548_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_54548_end_0 = const()[name = tensor("op_54548_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_54548_end_mask_0 = const()[name = tensor("op_54548_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_54548_cast_fp16 = slice_by_index(begin = var_54548_begin_0, end = var_54548_end_0, end_mask = var_54548_end_mask_0, x = k_483_cast_fp16)[name = tensor("op_54548_cast_fp16")]; + tensor var_54552_begin_0 = const()[name = tensor("op_54552_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_54552_end_0 = const()[name = tensor("op_54552_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_54552_end_mask_0 = const()[name = tensor("op_54552_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_54552_cast_fp16 = slice_by_index(begin = var_54552_begin_0, end = var_54552_end_0, end_mask = var_54552_end_mask_0, x = k_483_cast_fp16)[name = tensor("op_54552_cast_fp16")]; + tensor var_54554_begin_0 = const()[name = tensor("op_54554_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_54554_end_0 = const()[name = tensor("op_54554_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_54554_end_mask_0 = const()[name = tensor("op_54554_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54554_cast_fp16 = slice_by_index(begin = var_54554_begin_0, end = var_54554_end_0, end_mask = var_54554_end_mask_0, x = v_241_cast_fp16)[name = tensor("op_54554_cast_fp16")]; + tensor var_54558_begin_0 = const()[name = tensor("op_54558_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_54558_end_0 = const()[name = tensor("op_54558_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_54558_end_mask_0 = const()[name = tensor("op_54558_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54558_cast_fp16 = slice_by_index(begin = var_54558_begin_0, end = var_54558_end_0, end_mask = var_54558_end_mask_0, x = v_241_cast_fp16)[name = tensor("op_54558_cast_fp16")]; + tensor var_54562_begin_0 = const()[name = tensor("op_54562_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_54562_end_0 = const()[name = tensor("op_54562_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_54562_end_mask_0 = const()[name = tensor("op_54562_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54562_cast_fp16 = slice_by_index(begin = var_54562_begin_0, end = var_54562_end_0, end_mask = var_54562_end_mask_0, x = v_241_cast_fp16)[name = tensor("op_54562_cast_fp16")]; + tensor var_54566_begin_0 = const()[name = tensor("op_54566_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_54566_end_0 = const()[name = tensor("op_54566_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_54566_end_mask_0 = const()[name = tensor("op_54566_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54566_cast_fp16 = slice_by_index(begin = var_54566_begin_0, end = var_54566_end_0, end_mask = var_54566_end_mask_0, x = v_241_cast_fp16)[name = tensor("op_54566_cast_fp16")]; + tensor var_54570_begin_0 = const()[name = tensor("op_54570_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_54570_end_0 = const()[name = tensor("op_54570_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_54570_end_mask_0 = const()[name = tensor("op_54570_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54570_cast_fp16 = slice_by_index(begin = var_54570_begin_0, end = var_54570_end_0, end_mask = var_54570_end_mask_0, x = v_241_cast_fp16)[name = tensor("op_54570_cast_fp16")]; + tensor var_54574_begin_0 = const()[name = tensor("op_54574_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_54574_end_0 = const()[name = tensor("op_54574_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_54574_end_mask_0 = const()[name = tensor("op_54574_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54574_cast_fp16 = slice_by_index(begin = var_54574_begin_0, end = var_54574_end_0, end_mask = var_54574_end_mask_0, x = v_241_cast_fp16)[name = tensor("op_54574_cast_fp16")]; + tensor var_54578_begin_0 = const()[name = tensor("op_54578_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_54578_end_0 = const()[name = tensor("op_54578_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_54578_end_mask_0 = const()[name = tensor("op_54578_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54578_cast_fp16 = slice_by_index(begin = var_54578_begin_0, end = var_54578_end_0, end_mask = var_54578_end_mask_0, x = v_241_cast_fp16)[name = tensor("op_54578_cast_fp16")]; + tensor var_54582_begin_0 = const()[name = tensor("op_54582_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_54582_end_0 = const()[name = tensor("op_54582_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_54582_end_mask_0 = const()[name = tensor("op_54582_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54582_cast_fp16 = slice_by_index(begin = var_54582_begin_0, end = var_54582_end_0, end_mask = var_54582_end_mask_0, x = v_241_cast_fp16)[name = tensor("op_54582_cast_fp16")]; + tensor var_54586_begin_0 = const()[name = tensor("op_54586_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_54586_end_0 = const()[name = tensor("op_54586_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_54586_end_mask_0 = const()[name = tensor("op_54586_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54586_cast_fp16 = slice_by_index(begin = var_54586_begin_0, end = var_54586_end_0, end_mask = var_54586_end_mask_0, x = v_241_cast_fp16)[name = tensor("op_54586_cast_fp16")]; + tensor var_54590_begin_0 = const()[name = tensor("op_54590_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_54590_end_0 = const()[name = tensor("op_54590_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_54590_end_mask_0 = const()[name = tensor("op_54590_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54590_cast_fp16 = slice_by_index(begin = var_54590_begin_0, end = var_54590_end_0, end_mask = var_54590_end_mask_0, x = v_241_cast_fp16)[name = tensor("op_54590_cast_fp16")]; + tensor var_54594_begin_0 = const()[name = tensor("op_54594_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_54594_end_0 = const()[name = tensor("op_54594_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_54594_end_mask_0 = const()[name = tensor("op_54594_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54594_cast_fp16 = slice_by_index(begin = var_54594_begin_0, end = var_54594_end_0, end_mask = var_54594_end_mask_0, x = v_241_cast_fp16)[name = tensor("op_54594_cast_fp16")]; + tensor var_54598_begin_0 = const()[name = tensor("op_54598_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_54598_end_0 = const()[name = tensor("op_54598_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_54598_end_mask_0 = const()[name = tensor("op_54598_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54598_cast_fp16 = slice_by_index(begin = var_54598_begin_0, end = var_54598_end_0, end_mask = var_54598_end_mask_0, x = v_241_cast_fp16)[name = tensor("op_54598_cast_fp16")]; + tensor var_54602_begin_0 = const()[name = tensor("op_54602_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_54602_end_0 = const()[name = tensor("op_54602_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_54602_end_mask_0 = const()[name = tensor("op_54602_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54602_cast_fp16 = slice_by_index(begin = var_54602_begin_0, end = var_54602_end_0, end_mask = var_54602_end_mask_0, x = v_241_cast_fp16)[name = tensor("op_54602_cast_fp16")]; + tensor var_54606_begin_0 = const()[name = tensor("op_54606_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_54606_end_0 = const()[name = tensor("op_54606_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_54606_end_mask_0 = const()[name = tensor("op_54606_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54606_cast_fp16 = slice_by_index(begin = var_54606_begin_0, end = var_54606_end_0, end_mask = var_54606_end_mask_0, x = v_241_cast_fp16)[name = tensor("op_54606_cast_fp16")]; + tensor var_54610_begin_0 = const()[name = tensor("op_54610_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_54610_end_0 = const()[name = tensor("op_54610_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_54610_end_mask_0 = const()[name = tensor("op_54610_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54610_cast_fp16 = slice_by_index(begin = var_54610_begin_0, end = var_54610_end_0, end_mask = var_54610_end_mask_0, x = v_241_cast_fp16)[name = tensor("op_54610_cast_fp16")]; + tensor var_54614_begin_0 = const()[name = tensor("op_54614_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_54614_end_0 = const()[name = tensor("op_54614_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_54614_end_mask_0 = const()[name = tensor("op_54614_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54614_cast_fp16 = slice_by_index(begin = var_54614_begin_0, end = var_54614_end_0, end_mask = var_54614_end_mask_0, x = v_241_cast_fp16)[name = tensor("op_54614_cast_fp16")]; + tensor var_54618_begin_0 = const()[name = tensor("op_54618_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_54618_end_0 = const()[name = tensor("op_54618_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_54618_end_mask_0 = const()[name = tensor("op_54618_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54618_cast_fp16 = slice_by_index(begin = var_54618_begin_0, end = var_54618_end_0, end_mask = var_54618_end_mask_0, x = v_241_cast_fp16)[name = tensor("op_54618_cast_fp16")]; + tensor var_54622_begin_0 = const()[name = tensor("op_54622_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_54622_end_0 = const()[name = tensor("op_54622_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_54622_end_mask_0 = const()[name = tensor("op_54622_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54622_cast_fp16 = slice_by_index(begin = var_54622_begin_0, end = var_54622_end_0, end_mask = var_54622_end_mask_0, x = v_241_cast_fp16)[name = tensor("op_54622_cast_fp16")]; + tensor var_54626_begin_0 = const()[name = tensor("op_54626_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_54626_end_0 = const()[name = tensor("op_54626_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_54626_end_mask_0 = const()[name = tensor("op_54626_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54626_cast_fp16 = slice_by_index(begin = var_54626_begin_0, end = var_54626_end_0, end_mask = var_54626_end_mask_0, x = v_241_cast_fp16)[name = tensor("op_54626_cast_fp16")]; + tensor var_54630_begin_0 = const()[name = tensor("op_54630_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_54630_end_0 = const()[name = tensor("op_54630_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_54630_end_mask_0 = const()[name = tensor("op_54630_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54630_cast_fp16 = slice_by_index(begin = var_54630_begin_0, end = var_54630_end_0, end_mask = var_54630_end_mask_0, x = v_241_cast_fp16)[name = tensor("op_54630_cast_fp16")]; + tensor var_54634_equation_0 = const()[name = tensor("op_54634_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_54634_cast_fp16 = einsum(equation = var_54634_equation_0, values = (var_54476_cast_fp16, var_54393_cast_fp16))[name = tensor("op_54634_cast_fp16")]; + tensor var_54635_to_fp16 = const()[name = tensor("op_54635_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4641_cast_fp16 = mul(x = var_54634_cast_fp16, y = var_54635_to_fp16)[name = tensor("aw_4641_cast_fp16")]; + tensor var_54638_equation_0 = const()[name = tensor("op_54638_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_54638_cast_fp16 = einsum(equation = var_54638_equation_0, values = (var_54480_cast_fp16, var_54397_cast_fp16))[name = tensor("op_54638_cast_fp16")]; + tensor var_54639_to_fp16 = const()[name = tensor("op_54639_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4643_cast_fp16 = mul(x = var_54638_cast_fp16, y = var_54639_to_fp16)[name = tensor("aw_4643_cast_fp16")]; + tensor var_54642_equation_0 = const()[name = tensor("op_54642_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_54642_cast_fp16 = einsum(equation = var_54642_equation_0, values = (var_54484_cast_fp16, var_54401_cast_fp16))[name = tensor("op_54642_cast_fp16")]; + tensor var_54643_to_fp16 = const()[name = tensor("op_54643_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4645_cast_fp16 = mul(x = var_54642_cast_fp16, y = var_54643_to_fp16)[name = tensor("aw_4645_cast_fp16")]; + tensor var_54646_equation_0 = const()[name = tensor("op_54646_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_54646_cast_fp16 = einsum(equation = var_54646_equation_0, values = (var_54488_cast_fp16, var_54405_cast_fp16))[name = tensor("op_54646_cast_fp16")]; + tensor var_54647_to_fp16 = const()[name = tensor("op_54647_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4647_cast_fp16 = mul(x = var_54646_cast_fp16, y = var_54647_to_fp16)[name = tensor("aw_4647_cast_fp16")]; + tensor var_54650_equation_0 = const()[name = tensor("op_54650_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_54650_cast_fp16 = einsum(equation = var_54650_equation_0, values = (var_54492_cast_fp16, var_54409_cast_fp16))[name = tensor("op_54650_cast_fp16")]; + tensor var_54651_to_fp16 = const()[name = tensor("op_54651_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4649_cast_fp16 = mul(x = var_54650_cast_fp16, y = var_54651_to_fp16)[name = tensor("aw_4649_cast_fp16")]; + tensor var_54654_equation_0 = const()[name = tensor("op_54654_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_54654_cast_fp16 = einsum(equation = var_54654_equation_0, values = (var_54496_cast_fp16, var_54413_cast_fp16))[name = tensor("op_54654_cast_fp16")]; + tensor var_54655_to_fp16 = const()[name = tensor("op_54655_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4651_cast_fp16 = mul(x = var_54654_cast_fp16, y = var_54655_to_fp16)[name = tensor("aw_4651_cast_fp16")]; + tensor var_54658_equation_0 = const()[name = tensor("op_54658_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_54658_cast_fp16 = einsum(equation = var_54658_equation_0, values = (var_54500_cast_fp16, var_54417_cast_fp16))[name = tensor("op_54658_cast_fp16")]; + tensor var_54659_to_fp16 = const()[name = tensor("op_54659_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4653_cast_fp16 = mul(x = var_54658_cast_fp16, y = var_54659_to_fp16)[name = tensor("aw_4653_cast_fp16")]; + tensor var_54662_equation_0 = const()[name = tensor("op_54662_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_54662_cast_fp16 = einsum(equation = var_54662_equation_0, values = (var_54504_cast_fp16, var_54421_cast_fp16))[name = tensor("op_54662_cast_fp16")]; + tensor var_54663_to_fp16 = const()[name = tensor("op_54663_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4655_cast_fp16 = mul(x = var_54662_cast_fp16, y = var_54663_to_fp16)[name = tensor("aw_4655_cast_fp16")]; + tensor var_54666_equation_0 = const()[name = tensor("op_54666_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_54666_cast_fp16 = einsum(equation = var_54666_equation_0, values = (var_54508_cast_fp16, var_54425_cast_fp16))[name = tensor("op_54666_cast_fp16")]; + tensor var_54667_to_fp16 = const()[name = tensor("op_54667_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4657_cast_fp16 = mul(x = var_54666_cast_fp16, y = var_54667_to_fp16)[name = tensor("aw_4657_cast_fp16")]; + tensor var_54670_equation_0 = const()[name = tensor("op_54670_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_54670_cast_fp16 = einsum(equation = var_54670_equation_0, values = (var_54512_cast_fp16, var_54429_cast_fp16))[name = tensor("op_54670_cast_fp16")]; + tensor var_54671_to_fp16 = const()[name = tensor("op_54671_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4659_cast_fp16 = mul(x = var_54670_cast_fp16, y = var_54671_to_fp16)[name = tensor("aw_4659_cast_fp16")]; + tensor var_54674_equation_0 = const()[name = tensor("op_54674_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_54674_cast_fp16 = einsum(equation = var_54674_equation_0, values = (var_54516_cast_fp16, var_54433_cast_fp16))[name = tensor("op_54674_cast_fp16")]; + tensor var_54675_to_fp16 = const()[name = tensor("op_54675_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4661_cast_fp16 = mul(x = var_54674_cast_fp16, y = var_54675_to_fp16)[name = tensor("aw_4661_cast_fp16")]; + tensor var_54678_equation_0 = const()[name = tensor("op_54678_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_54678_cast_fp16 = einsum(equation = var_54678_equation_0, values = (var_54520_cast_fp16, var_54437_cast_fp16))[name = tensor("op_54678_cast_fp16")]; + tensor var_54679_to_fp16 = const()[name = tensor("op_54679_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4663_cast_fp16 = mul(x = var_54678_cast_fp16, y = var_54679_to_fp16)[name = tensor("aw_4663_cast_fp16")]; + tensor var_54682_equation_0 = const()[name = tensor("op_54682_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_54682_cast_fp16 = einsum(equation = var_54682_equation_0, values = (var_54524_cast_fp16, var_54441_cast_fp16))[name = tensor("op_54682_cast_fp16")]; + tensor var_54683_to_fp16 = const()[name = tensor("op_54683_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4665_cast_fp16 = mul(x = var_54682_cast_fp16, y = var_54683_to_fp16)[name = tensor("aw_4665_cast_fp16")]; + tensor var_54686_equation_0 = const()[name = tensor("op_54686_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_54686_cast_fp16 = einsum(equation = var_54686_equation_0, values = (var_54528_cast_fp16, var_54445_cast_fp16))[name = tensor("op_54686_cast_fp16")]; + tensor var_54687_to_fp16 = const()[name = tensor("op_54687_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4667_cast_fp16 = mul(x = var_54686_cast_fp16, y = var_54687_to_fp16)[name = tensor("aw_4667_cast_fp16")]; + tensor var_54690_equation_0 = const()[name = tensor("op_54690_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_54690_cast_fp16 = einsum(equation = var_54690_equation_0, values = (var_54532_cast_fp16, var_54449_cast_fp16))[name = tensor("op_54690_cast_fp16")]; + tensor var_54691_to_fp16 = const()[name = tensor("op_54691_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4669_cast_fp16 = mul(x = var_54690_cast_fp16, y = var_54691_to_fp16)[name = tensor("aw_4669_cast_fp16")]; + tensor var_54694_equation_0 = const()[name = tensor("op_54694_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_54694_cast_fp16 = einsum(equation = var_54694_equation_0, values = (var_54536_cast_fp16, var_54453_cast_fp16))[name = tensor("op_54694_cast_fp16")]; + tensor var_54695_to_fp16 = const()[name = tensor("op_54695_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4671_cast_fp16 = mul(x = var_54694_cast_fp16, y = var_54695_to_fp16)[name = tensor("aw_4671_cast_fp16")]; + tensor var_54698_equation_0 = const()[name = tensor("op_54698_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_54698_cast_fp16 = einsum(equation = var_54698_equation_0, values = (var_54540_cast_fp16, var_54457_cast_fp16))[name = tensor("op_54698_cast_fp16")]; + tensor var_54699_to_fp16 = const()[name = tensor("op_54699_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4673_cast_fp16 = mul(x = var_54698_cast_fp16, y = var_54699_to_fp16)[name = tensor("aw_4673_cast_fp16")]; + tensor var_54702_equation_0 = const()[name = tensor("op_54702_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_54702_cast_fp16 = einsum(equation = var_54702_equation_0, values = (var_54544_cast_fp16, var_54461_cast_fp16))[name = tensor("op_54702_cast_fp16")]; + tensor var_54703_to_fp16 = const()[name = tensor("op_54703_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4675_cast_fp16 = mul(x = var_54702_cast_fp16, y = var_54703_to_fp16)[name = tensor("aw_4675_cast_fp16")]; + tensor var_54706_equation_0 = const()[name = tensor("op_54706_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_54706_cast_fp16 = einsum(equation = var_54706_equation_0, values = (var_54548_cast_fp16, var_54465_cast_fp16))[name = tensor("op_54706_cast_fp16")]; + tensor var_54707_to_fp16 = const()[name = tensor("op_54707_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4677_cast_fp16 = mul(x = var_54706_cast_fp16, y = var_54707_to_fp16)[name = tensor("aw_4677_cast_fp16")]; + tensor var_54710_equation_0 = const()[name = tensor("op_54710_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_54710_cast_fp16 = einsum(equation = var_54710_equation_0, values = (var_54552_cast_fp16, var_54469_cast_fp16))[name = tensor("op_54710_cast_fp16")]; + tensor var_54711_to_fp16 = const()[name = tensor("op_54711_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4679_cast_fp16 = mul(x = var_54710_cast_fp16, y = var_54711_to_fp16)[name = tensor("aw_4679_cast_fp16")]; + tensor var_54713_cast_fp16 = softmax(axis = var_30355, x = aw_4641_cast_fp16)[name = tensor("op_54713_cast_fp16")]; + tensor var_54714_cast_fp16 = softmax(axis = var_30355, x = aw_4643_cast_fp16)[name = tensor("op_54714_cast_fp16")]; + tensor var_54715_cast_fp16 = softmax(axis = var_30355, x = aw_4645_cast_fp16)[name = tensor("op_54715_cast_fp16")]; + tensor var_54716_cast_fp16 = softmax(axis = var_30355, x = aw_4647_cast_fp16)[name = tensor("op_54716_cast_fp16")]; + tensor var_54717_cast_fp16 = softmax(axis = var_30355, x = aw_4649_cast_fp16)[name = tensor("op_54717_cast_fp16")]; + tensor var_54718_cast_fp16 = softmax(axis = var_30355, x = aw_4651_cast_fp16)[name = tensor("op_54718_cast_fp16")]; + tensor var_54719_cast_fp16 = softmax(axis = var_30355, x = aw_4653_cast_fp16)[name = tensor("op_54719_cast_fp16")]; + tensor var_54720_cast_fp16 = softmax(axis = var_30355, x = aw_4655_cast_fp16)[name = tensor("op_54720_cast_fp16")]; + tensor var_54721_cast_fp16 = softmax(axis = var_30355, x = aw_4657_cast_fp16)[name = tensor("op_54721_cast_fp16")]; + tensor var_54722_cast_fp16 = softmax(axis = var_30355, x = aw_4659_cast_fp16)[name = tensor("op_54722_cast_fp16")]; + tensor var_54723_cast_fp16 = softmax(axis = var_30355, x = aw_4661_cast_fp16)[name = tensor("op_54723_cast_fp16")]; + tensor var_54724_cast_fp16 = softmax(axis = var_30355, x = aw_4663_cast_fp16)[name = tensor("op_54724_cast_fp16")]; + tensor var_54725_cast_fp16 = softmax(axis = var_30355, x = aw_4665_cast_fp16)[name = tensor("op_54725_cast_fp16")]; + tensor var_54726_cast_fp16 = softmax(axis = var_30355, x = aw_4667_cast_fp16)[name = tensor("op_54726_cast_fp16")]; + tensor var_54727_cast_fp16 = softmax(axis = var_30355, x = aw_4669_cast_fp16)[name = tensor("op_54727_cast_fp16")]; + tensor var_54728_cast_fp16 = softmax(axis = var_30355, x = aw_4671_cast_fp16)[name = tensor("op_54728_cast_fp16")]; + tensor var_54729_cast_fp16 = softmax(axis = var_30355, x = aw_4673_cast_fp16)[name = tensor("op_54729_cast_fp16")]; + tensor var_54730_cast_fp16 = softmax(axis = var_30355, x = aw_4675_cast_fp16)[name = tensor("op_54730_cast_fp16")]; + tensor var_54731_cast_fp16 = softmax(axis = var_30355, x = aw_4677_cast_fp16)[name = tensor("op_54731_cast_fp16")]; + tensor var_54732_cast_fp16 = softmax(axis = var_30355, x = aw_4679_cast_fp16)[name = tensor("op_54732_cast_fp16")]; + tensor var_54734_equation_0 = const()[name = tensor("op_54734_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_54734_cast_fp16 = einsum(equation = var_54734_equation_0, values = (var_54554_cast_fp16, var_54713_cast_fp16))[name = tensor("op_54734_cast_fp16")]; + tensor var_54736_equation_0 = const()[name = tensor("op_54736_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_54736_cast_fp16 = einsum(equation = var_54736_equation_0, values = (var_54558_cast_fp16, var_54714_cast_fp16))[name = tensor("op_54736_cast_fp16")]; + tensor var_54738_equation_0 = const()[name = tensor("op_54738_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_54738_cast_fp16 = einsum(equation = var_54738_equation_0, values = (var_54562_cast_fp16, var_54715_cast_fp16))[name = tensor("op_54738_cast_fp16")]; + tensor var_54740_equation_0 = const()[name = tensor("op_54740_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_54740_cast_fp16 = einsum(equation = var_54740_equation_0, values = (var_54566_cast_fp16, var_54716_cast_fp16))[name = tensor("op_54740_cast_fp16")]; + tensor var_54742_equation_0 = const()[name = tensor("op_54742_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_54742_cast_fp16 = einsum(equation = var_54742_equation_0, values = (var_54570_cast_fp16, var_54717_cast_fp16))[name = tensor("op_54742_cast_fp16")]; + tensor var_54744_equation_0 = const()[name = tensor("op_54744_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_54744_cast_fp16 = einsum(equation = var_54744_equation_0, values = (var_54574_cast_fp16, var_54718_cast_fp16))[name = tensor("op_54744_cast_fp16")]; + tensor var_54746_equation_0 = const()[name = tensor("op_54746_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_54746_cast_fp16 = einsum(equation = var_54746_equation_0, values = (var_54578_cast_fp16, var_54719_cast_fp16))[name = tensor("op_54746_cast_fp16")]; + tensor var_54748_equation_0 = const()[name = tensor("op_54748_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_54748_cast_fp16 = einsum(equation = var_54748_equation_0, values = (var_54582_cast_fp16, var_54720_cast_fp16))[name = tensor("op_54748_cast_fp16")]; + tensor var_54750_equation_0 = const()[name = tensor("op_54750_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_54750_cast_fp16 = einsum(equation = var_54750_equation_0, values = (var_54586_cast_fp16, var_54721_cast_fp16))[name = tensor("op_54750_cast_fp16")]; + tensor var_54752_equation_0 = const()[name = tensor("op_54752_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_54752_cast_fp16 = einsum(equation = var_54752_equation_0, values = (var_54590_cast_fp16, var_54722_cast_fp16))[name = tensor("op_54752_cast_fp16")]; + tensor var_54754_equation_0 = const()[name = tensor("op_54754_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_54754_cast_fp16 = einsum(equation = var_54754_equation_0, values = (var_54594_cast_fp16, var_54723_cast_fp16))[name = tensor("op_54754_cast_fp16")]; + tensor var_54756_equation_0 = const()[name = tensor("op_54756_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_54756_cast_fp16 = einsum(equation = var_54756_equation_0, values = (var_54598_cast_fp16, var_54724_cast_fp16))[name = tensor("op_54756_cast_fp16")]; + tensor var_54758_equation_0 = const()[name = tensor("op_54758_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_54758_cast_fp16 = einsum(equation = var_54758_equation_0, values = (var_54602_cast_fp16, var_54725_cast_fp16))[name = tensor("op_54758_cast_fp16")]; + tensor var_54760_equation_0 = const()[name = tensor("op_54760_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_54760_cast_fp16 = einsum(equation = var_54760_equation_0, values = (var_54606_cast_fp16, var_54726_cast_fp16))[name = tensor("op_54760_cast_fp16")]; + tensor var_54762_equation_0 = const()[name = tensor("op_54762_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_54762_cast_fp16 = einsum(equation = var_54762_equation_0, values = (var_54610_cast_fp16, var_54727_cast_fp16))[name = tensor("op_54762_cast_fp16")]; + tensor var_54764_equation_0 = const()[name = tensor("op_54764_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_54764_cast_fp16 = einsum(equation = var_54764_equation_0, values = (var_54614_cast_fp16, var_54728_cast_fp16))[name = tensor("op_54764_cast_fp16")]; + tensor var_54766_equation_0 = const()[name = tensor("op_54766_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_54766_cast_fp16 = einsum(equation = var_54766_equation_0, values = (var_54618_cast_fp16, var_54729_cast_fp16))[name = tensor("op_54766_cast_fp16")]; + tensor var_54768_equation_0 = const()[name = tensor("op_54768_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_54768_cast_fp16 = einsum(equation = var_54768_equation_0, values = (var_54622_cast_fp16, var_54730_cast_fp16))[name = tensor("op_54768_cast_fp16")]; + tensor var_54770_equation_0 = const()[name = tensor("op_54770_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_54770_cast_fp16 = einsum(equation = var_54770_equation_0, values = (var_54626_cast_fp16, var_54731_cast_fp16))[name = tensor("op_54770_cast_fp16")]; + tensor var_54772_equation_0 = const()[name = tensor("op_54772_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_54772_cast_fp16 = einsum(equation = var_54772_equation_0, values = (var_54630_cast_fp16, var_54732_cast_fp16))[name = tensor("op_54772_cast_fp16")]; + tensor input_683_interleave_0 = const()[name = tensor("input_683_interleave_0"), val = tensor(false)]; + tensor input_683_cast_fp16 = concat(axis = var_30355, interleave = input_683_interleave_0, values = (var_54734_cast_fp16, var_54736_cast_fp16, var_54738_cast_fp16, var_54740_cast_fp16, var_54742_cast_fp16, var_54744_cast_fp16, var_54746_cast_fp16, var_54748_cast_fp16, var_54750_cast_fp16, var_54752_cast_fp16, var_54754_cast_fp16, var_54756_cast_fp16, var_54758_cast_fp16, var_54760_cast_fp16, var_54762_cast_fp16, var_54764_cast_fp16, var_54766_cast_fp16, var_54768_cast_fp16, var_54770_cast_fp16, var_54772_cast_fp16))[name = tensor("input_683_cast_fp16")]; + tensor var_54778 = const()[name = tensor("op_54778"), val = tensor([1, 1])]; + tensor var_54780 = const()[name = tensor("op_54780"), val = tensor([1, 1])]; + tensor var_54782_pad_type_0 = const()[name = tensor("op_54782_pad_type_0"), val = tensor("custom")]; + tensor var_54782_pad_0 = const()[name = tensor("op_54782_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_6_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_6_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4599004608)))]; + tensor up_blocks_0_attentions_2_transformer_blocks_6_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_6_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4602281472)))]; + tensor var_54782_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_6_attn1_to_out_0_bias_to_fp16, dilations = var_54780, groups = var_30355, pad = var_54782_pad_0, pad_type = var_54782_pad_type_0, strides = var_54778, weight = up_blocks_0_attentions_2_transformer_blocks_6_attn1_to_out_0_weight_to_fp16, x = input_683_cast_fp16)[name = tensor("op_54782_cast_fp16")]; + tensor inputs_363_cast_fp16 = add(x = var_54782_cast_fp16, y = inputs_361_cast_fp16)[name = tensor("inputs_363_cast_fp16")]; + tensor hidden_states_473_axes_0 = const()[name = tensor("hidden_states_473_axes_0"), val = tensor([1])]; + tensor hidden_states_473_gamma_0_to_fp16 = const()[name = tensor("hidden_states_473_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4602284096)))]; + tensor hidden_states_473_beta_0_to_fp16 = const()[name = tensor("hidden_states_473_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4602286720)))]; + tensor var_54792_to_fp16 = const()[name = tensor("op_54792_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_473_cast_fp16 = layer_norm(axes = hidden_states_473_axes_0, beta = hidden_states_473_beta_0_to_fp16, epsilon = var_54792_to_fp16, gamma = hidden_states_473_gamma_0_to_fp16, x = inputs_363_cast_fp16)[name = tensor("hidden_states_473_cast_fp16")]; + tensor var_54807 = const()[name = tensor("op_54807"), val = tensor([1, 1])]; + tensor var_54809 = const()[name = tensor("op_54809"), val = tensor([1, 1])]; + tensor q_243_pad_type_0 = const()[name = tensor("q_243_pad_type_0"), val = tensor("custom")]; + tensor q_243_pad_0 = const()[name = tensor("q_243_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_6_attn2_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_6_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4602289344)))]; + tensor q_243_cast_fp16 = conv(dilations = var_54809, groups = var_30355, pad = q_243_pad_0, pad_type = q_243_pad_type_0, strides = var_54807, weight = up_blocks_0_attentions_2_transformer_blocks_6_attn2_to_q_weight_to_fp16, x = hidden_states_473_cast_fp16)[name = tensor("q_243_cast_fp16")]; + tensor var_54813 = const()[name = tensor("op_54813"), val = tensor([1, 1])]; + tensor var_54815 = const()[name = tensor("op_54815"), val = tensor([1, 1])]; + tensor k_485_pad_type_0 = const()[name = tensor("k_485_pad_type_0"), val = tensor("custom")]; + tensor k_485_pad_0 = const()[name = tensor("k_485_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_6_attn2_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_6_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4605566208)))]; + tensor k_485_cast_fp16 = conv(dilations = var_54815, groups = var_30355, pad = k_485_pad_0, pad_type = k_485_pad_type_0, strides = var_54813, weight = up_blocks_0_attentions_2_transformer_blocks_6_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_485_cast_fp16")]; + tensor var_54819 = const()[name = tensor("op_54819"), val = tensor([1, 1])]; + tensor var_54821 = const()[name = tensor("op_54821"), val = tensor([1, 1])]; + tensor v_243_pad_type_0 = const()[name = tensor("v_243_pad_type_0"), val = tensor("custom")]; + tensor v_243_pad_0 = const()[name = tensor("v_243_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_6_attn2_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_6_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4610809152)))]; + tensor v_243_cast_fp16 = conv(dilations = var_54821, groups = var_30355, pad = v_243_pad_0, pad_type = v_243_pad_type_0, strides = var_54819, weight = up_blocks_0_attentions_2_transformer_blocks_6_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_243_cast_fp16")]; + tensor var_54825_begin_0 = const()[name = tensor("op_54825_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_54825_end_0 = const()[name = tensor("op_54825_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_54825_end_mask_0 = const()[name = tensor("op_54825_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54825_cast_fp16 = slice_by_index(begin = var_54825_begin_0, end = var_54825_end_0, end_mask = var_54825_end_mask_0, x = q_243_cast_fp16)[name = tensor("op_54825_cast_fp16")]; + tensor var_54829_begin_0 = const()[name = tensor("op_54829_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_54829_end_0 = const()[name = tensor("op_54829_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_54829_end_mask_0 = const()[name = tensor("op_54829_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54829_cast_fp16 = slice_by_index(begin = var_54829_begin_0, end = var_54829_end_0, end_mask = var_54829_end_mask_0, x = q_243_cast_fp16)[name = tensor("op_54829_cast_fp16")]; + tensor var_54833_begin_0 = const()[name = tensor("op_54833_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_54833_end_0 = const()[name = tensor("op_54833_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_54833_end_mask_0 = const()[name = tensor("op_54833_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54833_cast_fp16 = slice_by_index(begin = var_54833_begin_0, end = var_54833_end_0, end_mask = var_54833_end_mask_0, x = q_243_cast_fp16)[name = tensor("op_54833_cast_fp16")]; + tensor var_54837_begin_0 = const()[name = tensor("op_54837_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_54837_end_0 = const()[name = tensor("op_54837_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_54837_end_mask_0 = const()[name = tensor("op_54837_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54837_cast_fp16 = slice_by_index(begin = var_54837_begin_0, end = var_54837_end_0, end_mask = var_54837_end_mask_0, x = q_243_cast_fp16)[name = tensor("op_54837_cast_fp16")]; + tensor var_54841_begin_0 = const()[name = tensor("op_54841_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_54841_end_0 = const()[name = tensor("op_54841_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_54841_end_mask_0 = const()[name = tensor("op_54841_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54841_cast_fp16 = slice_by_index(begin = var_54841_begin_0, end = var_54841_end_0, end_mask = var_54841_end_mask_0, x = q_243_cast_fp16)[name = tensor("op_54841_cast_fp16")]; + tensor var_54845_begin_0 = const()[name = tensor("op_54845_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_54845_end_0 = const()[name = tensor("op_54845_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_54845_end_mask_0 = const()[name = tensor("op_54845_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54845_cast_fp16 = slice_by_index(begin = var_54845_begin_0, end = var_54845_end_0, end_mask = var_54845_end_mask_0, x = q_243_cast_fp16)[name = tensor("op_54845_cast_fp16")]; + tensor var_54849_begin_0 = const()[name = tensor("op_54849_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_54849_end_0 = const()[name = tensor("op_54849_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_54849_end_mask_0 = const()[name = tensor("op_54849_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54849_cast_fp16 = slice_by_index(begin = var_54849_begin_0, end = var_54849_end_0, end_mask = var_54849_end_mask_0, x = q_243_cast_fp16)[name = tensor("op_54849_cast_fp16")]; + tensor var_54853_begin_0 = const()[name = tensor("op_54853_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_54853_end_0 = const()[name = tensor("op_54853_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_54853_end_mask_0 = const()[name = tensor("op_54853_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54853_cast_fp16 = slice_by_index(begin = var_54853_begin_0, end = var_54853_end_0, end_mask = var_54853_end_mask_0, x = q_243_cast_fp16)[name = tensor("op_54853_cast_fp16")]; + tensor var_54857_begin_0 = const()[name = tensor("op_54857_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_54857_end_0 = const()[name = tensor("op_54857_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_54857_end_mask_0 = const()[name = tensor("op_54857_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54857_cast_fp16 = slice_by_index(begin = var_54857_begin_0, end = var_54857_end_0, end_mask = var_54857_end_mask_0, x = q_243_cast_fp16)[name = tensor("op_54857_cast_fp16")]; + tensor var_54861_begin_0 = const()[name = tensor("op_54861_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_54861_end_0 = const()[name = tensor("op_54861_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_54861_end_mask_0 = const()[name = tensor("op_54861_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54861_cast_fp16 = slice_by_index(begin = var_54861_begin_0, end = var_54861_end_0, end_mask = var_54861_end_mask_0, x = q_243_cast_fp16)[name = tensor("op_54861_cast_fp16")]; + tensor var_54865_begin_0 = const()[name = tensor("op_54865_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_54865_end_0 = const()[name = tensor("op_54865_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_54865_end_mask_0 = const()[name = tensor("op_54865_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54865_cast_fp16 = slice_by_index(begin = var_54865_begin_0, end = var_54865_end_0, end_mask = var_54865_end_mask_0, x = q_243_cast_fp16)[name = tensor("op_54865_cast_fp16")]; + tensor var_54869_begin_0 = const()[name = tensor("op_54869_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_54869_end_0 = const()[name = tensor("op_54869_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_54869_end_mask_0 = const()[name = tensor("op_54869_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54869_cast_fp16 = slice_by_index(begin = var_54869_begin_0, end = var_54869_end_0, end_mask = var_54869_end_mask_0, x = q_243_cast_fp16)[name = tensor("op_54869_cast_fp16")]; + tensor var_54873_begin_0 = const()[name = tensor("op_54873_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_54873_end_0 = const()[name = tensor("op_54873_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_54873_end_mask_0 = const()[name = tensor("op_54873_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54873_cast_fp16 = slice_by_index(begin = var_54873_begin_0, end = var_54873_end_0, end_mask = var_54873_end_mask_0, x = q_243_cast_fp16)[name = tensor("op_54873_cast_fp16")]; + tensor var_54877_begin_0 = const()[name = tensor("op_54877_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_54877_end_0 = const()[name = tensor("op_54877_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_54877_end_mask_0 = const()[name = tensor("op_54877_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54877_cast_fp16 = slice_by_index(begin = var_54877_begin_0, end = var_54877_end_0, end_mask = var_54877_end_mask_0, x = q_243_cast_fp16)[name = tensor("op_54877_cast_fp16")]; + tensor var_54881_begin_0 = const()[name = tensor("op_54881_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_54881_end_0 = const()[name = tensor("op_54881_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_54881_end_mask_0 = const()[name = tensor("op_54881_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54881_cast_fp16 = slice_by_index(begin = var_54881_begin_0, end = var_54881_end_0, end_mask = var_54881_end_mask_0, x = q_243_cast_fp16)[name = tensor("op_54881_cast_fp16")]; + tensor var_54885_begin_0 = const()[name = tensor("op_54885_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_54885_end_0 = const()[name = tensor("op_54885_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_54885_end_mask_0 = const()[name = tensor("op_54885_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54885_cast_fp16 = slice_by_index(begin = var_54885_begin_0, end = var_54885_end_0, end_mask = var_54885_end_mask_0, x = q_243_cast_fp16)[name = tensor("op_54885_cast_fp16")]; + tensor var_54889_begin_0 = const()[name = tensor("op_54889_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_54889_end_0 = const()[name = tensor("op_54889_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_54889_end_mask_0 = const()[name = tensor("op_54889_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54889_cast_fp16 = slice_by_index(begin = var_54889_begin_0, end = var_54889_end_0, end_mask = var_54889_end_mask_0, x = q_243_cast_fp16)[name = tensor("op_54889_cast_fp16")]; + tensor var_54893_begin_0 = const()[name = tensor("op_54893_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_54893_end_0 = const()[name = tensor("op_54893_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_54893_end_mask_0 = const()[name = tensor("op_54893_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54893_cast_fp16 = slice_by_index(begin = var_54893_begin_0, end = var_54893_end_0, end_mask = var_54893_end_mask_0, x = q_243_cast_fp16)[name = tensor("op_54893_cast_fp16")]; + tensor var_54897_begin_0 = const()[name = tensor("op_54897_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_54897_end_0 = const()[name = tensor("op_54897_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_54897_end_mask_0 = const()[name = tensor("op_54897_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54897_cast_fp16 = slice_by_index(begin = var_54897_begin_0, end = var_54897_end_0, end_mask = var_54897_end_mask_0, x = q_243_cast_fp16)[name = tensor("op_54897_cast_fp16")]; + tensor var_54901_begin_0 = const()[name = tensor("op_54901_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_54901_end_0 = const()[name = tensor("op_54901_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_54901_end_mask_0 = const()[name = tensor("op_54901_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54901_cast_fp16 = slice_by_index(begin = var_54901_begin_0, end = var_54901_end_0, end_mask = var_54901_end_mask_0, x = q_243_cast_fp16)[name = tensor("op_54901_cast_fp16")]; + tensor k_487_perm_0 = const()[name = tensor("k_487_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_54908_begin_0 = const()[name = tensor("op_54908_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_54908_end_0 = const()[name = tensor("op_54908_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_54908_end_mask_0 = const()[name = tensor("op_54908_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_487_cast_fp16 = transpose(perm = k_487_perm_0, x = k_485_cast_fp16)[name = tensor("transpose_18")]; + tensor var_54908_cast_fp16 = slice_by_index(begin = var_54908_begin_0, end = var_54908_end_0, end_mask = var_54908_end_mask_0, x = k_487_cast_fp16)[name = tensor("op_54908_cast_fp16")]; + tensor var_54912_begin_0 = const()[name = tensor("op_54912_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_54912_end_0 = const()[name = tensor("op_54912_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_54912_end_mask_0 = const()[name = tensor("op_54912_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_54912_cast_fp16 = slice_by_index(begin = var_54912_begin_0, end = var_54912_end_0, end_mask = var_54912_end_mask_0, x = k_487_cast_fp16)[name = tensor("op_54912_cast_fp16")]; + tensor var_54916_begin_0 = const()[name = tensor("op_54916_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_54916_end_0 = const()[name = tensor("op_54916_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_54916_end_mask_0 = const()[name = tensor("op_54916_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_54916_cast_fp16 = slice_by_index(begin = var_54916_begin_0, end = var_54916_end_0, end_mask = var_54916_end_mask_0, x = k_487_cast_fp16)[name = tensor("op_54916_cast_fp16")]; + tensor var_54920_begin_0 = const()[name = tensor("op_54920_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_54920_end_0 = const()[name = tensor("op_54920_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_54920_end_mask_0 = const()[name = tensor("op_54920_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_54920_cast_fp16 = slice_by_index(begin = var_54920_begin_0, end = var_54920_end_0, end_mask = var_54920_end_mask_0, x = k_487_cast_fp16)[name = tensor("op_54920_cast_fp16")]; + tensor var_54924_begin_0 = const()[name = tensor("op_54924_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_54924_end_0 = const()[name = tensor("op_54924_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_54924_end_mask_0 = const()[name = tensor("op_54924_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_54924_cast_fp16 = slice_by_index(begin = var_54924_begin_0, end = var_54924_end_0, end_mask = var_54924_end_mask_0, x = k_487_cast_fp16)[name = tensor("op_54924_cast_fp16")]; + tensor var_54928_begin_0 = const()[name = tensor("op_54928_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_54928_end_0 = const()[name = tensor("op_54928_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_54928_end_mask_0 = const()[name = tensor("op_54928_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_54928_cast_fp16 = slice_by_index(begin = var_54928_begin_0, end = var_54928_end_0, end_mask = var_54928_end_mask_0, x = k_487_cast_fp16)[name = tensor("op_54928_cast_fp16")]; + tensor var_54932_begin_0 = const()[name = tensor("op_54932_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_54932_end_0 = const()[name = tensor("op_54932_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_54932_end_mask_0 = const()[name = tensor("op_54932_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_54932_cast_fp16 = slice_by_index(begin = var_54932_begin_0, end = var_54932_end_0, end_mask = var_54932_end_mask_0, x = k_487_cast_fp16)[name = tensor("op_54932_cast_fp16")]; + tensor var_54936_begin_0 = const()[name = tensor("op_54936_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_54936_end_0 = const()[name = tensor("op_54936_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_54936_end_mask_0 = const()[name = tensor("op_54936_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_54936_cast_fp16 = slice_by_index(begin = var_54936_begin_0, end = var_54936_end_0, end_mask = var_54936_end_mask_0, x = k_487_cast_fp16)[name = tensor("op_54936_cast_fp16")]; + tensor var_54940_begin_0 = const()[name = tensor("op_54940_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_54940_end_0 = const()[name = tensor("op_54940_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_54940_end_mask_0 = const()[name = tensor("op_54940_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_54940_cast_fp16 = slice_by_index(begin = var_54940_begin_0, end = var_54940_end_0, end_mask = var_54940_end_mask_0, x = k_487_cast_fp16)[name = tensor("op_54940_cast_fp16")]; + tensor var_54944_begin_0 = const()[name = tensor("op_54944_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_54944_end_0 = const()[name = tensor("op_54944_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_54944_end_mask_0 = const()[name = tensor("op_54944_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_54944_cast_fp16 = slice_by_index(begin = var_54944_begin_0, end = var_54944_end_0, end_mask = var_54944_end_mask_0, x = k_487_cast_fp16)[name = tensor("op_54944_cast_fp16")]; + tensor var_54948_begin_0 = const()[name = tensor("op_54948_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_54948_end_0 = const()[name = tensor("op_54948_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_54948_end_mask_0 = const()[name = tensor("op_54948_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_54948_cast_fp16 = slice_by_index(begin = var_54948_begin_0, end = var_54948_end_0, end_mask = var_54948_end_mask_0, x = k_487_cast_fp16)[name = tensor("op_54948_cast_fp16")]; + tensor var_54952_begin_0 = const()[name = tensor("op_54952_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_54952_end_0 = const()[name = tensor("op_54952_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_54952_end_mask_0 = const()[name = tensor("op_54952_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_54952_cast_fp16 = slice_by_index(begin = var_54952_begin_0, end = var_54952_end_0, end_mask = var_54952_end_mask_0, x = k_487_cast_fp16)[name = tensor("op_54952_cast_fp16")]; + tensor var_54956_begin_0 = const()[name = tensor("op_54956_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_54956_end_0 = const()[name = tensor("op_54956_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_54956_end_mask_0 = const()[name = tensor("op_54956_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_54956_cast_fp16 = slice_by_index(begin = var_54956_begin_0, end = var_54956_end_0, end_mask = var_54956_end_mask_0, x = k_487_cast_fp16)[name = tensor("op_54956_cast_fp16")]; + tensor var_54960_begin_0 = const()[name = tensor("op_54960_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_54960_end_0 = const()[name = tensor("op_54960_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_54960_end_mask_0 = const()[name = tensor("op_54960_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_54960_cast_fp16 = slice_by_index(begin = var_54960_begin_0, end = var_54960_end_0, end_mask = var_54960_end_mask_0, x = k_487_cast_fp16)[name = tensor("op_54960_cast_fp16")]; + tensor var_54964_begin_0 = const()[name = tensor("op_54964_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_54964_end_0 = const()[name = tensor("op_54964_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_54964_end_mask_0 = const()[name = tensor("op_54964_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_54964_cast_fp16 = slice_by_index(begin = var_54964_begin_0, end = var_54964_end_0, end_mask = var_54964_end_mask_0, x = k_487_cast_fp16)[name = tensor("op_54964_cast_fp16")]; + tensor var_54968_begin_0 = const()[name = tensor("op_54968_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_54968_end_0 = const()[name = tensor("op_54968_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_54968_end_mask_0 = const()[name = tensor("op_54968_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_54968_cast_fp16 = slice_by_index(begin = var_54968_begin_0, end = var_54968_end_0, end_mask = var_54968_end_mask_0, x = k_487_cast_fp16)[name = tensor("op_54968_cast_fp16")]; + tensor var_54972_begin_0 = const()[name = tensor("op_54972_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_54972_end_0 = const()[name = tensor("op_54972_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_54972_end_mask_0 = const()[name = tensor("op_54972_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_54972_cast_fp16 = slice_by_index(begin = var_54972_begin_0, end = var_54972_end_0, end_mask = var_54972_end_mask_0, x = k_487_cast_fp16)[name = tensor("op_54972_cast_fp16")]; + tensor var_54976_begin_0 = const()[name = tensor("op_54976_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_54976_end_0 = const()[name = tensor("op_54976_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_54976_end_mask_0 = const()[name = tensor("op_54976_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_54976_cast_fp16 = slice_by_index(begin = var_54976_begin_0, end = var_54976_end_0, end_mask = var_54976_end_mask_0, x = k_487_cast_fp16)[name = tensor("op_54976_cast_fp16")]; + tensor var_54980_begin_0 = const()[name = tensor("op_54980_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_54980_end_0 = const()[name = tensor("op_54980_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_54980_end_mask_0 = const()[name = tensor("op_54980_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_54980_cast_fp16 = slice_by_index(begin = var_54980_begin_0, end = var_54980_end_0, end_mask = var_54980_end_mask_0, x = k_487_cast_fp16)[name = tensor("op_54980_cast_fp16")]; + tensor var_54984_begin_0 = const()[name = tensor("op_54984_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_54984_end_0 = const()[name = tensor("op_54984_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_54984_end_mask_0 = const()[name = tensor("op_54984_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_54984_cast_fp16 = slice_by_index(begin = var_54984_begin_0, end = var_54984_end_0, end_mask = var_54984_end_mask_0, x = k_487_cast_fp16)[name = tensor("op_54984_cast_fp16")]; + tensor var_54986_begin_0 = const()[name = tensor("op_54986_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_54986_end_0 = const()[name = tensor("op_54986_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_54986_end_mask_0 = const()[name = tensor("op_54986_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54986_cast_fp16 = slice_by_index(begin = var_54986_begin_0, end = var_54986_end_0, end_mask = var_54986_end_mask_0, x = v_243_cast_fp16)[name = tensor("op_54986_cast_fp16")]; + tensor var_54990_begin_0 = const()[name = tensor("op_54990_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_54990_end_0 = const()[name = tensor("op_54990_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_54990_end_mask_0 = const()[name = tensor("op_54990_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54990_cast_fp16 = slice_by_index(begin = var_54990_begin_0, end = var_54990_end_0, end_mask = var_54990_end_mask_0, x = v_243_cast_fp16)[name = tensor("op_54990_cast_fp16")]; + tensor var_54994_begin_0 = const()[name = tensor("op_54994_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_54994_end_0 = const()[name = tensor("op_54994_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_54994_end_mask_0 = const()[name = tensor("op_54994_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54994_cast_fp16 = slice_by_index(begin = var_54994_begin_0, end = var_54994_end_0, end_mask = var_54994_end_mask_0, x = v_243_cast_fp16)[name = tensor("op_54994_cast_fp16")]; + tensor var_54998_begin_0 = const()[name = tensor("op_54998_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_54998_end_0 = const()[name = tensor("op_54998_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_54998_end_mask_0 = const()[name = tensor("op_54998_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_54998_cast_fp16 = slice_by_index(begin = var_54998_begin_0, end = var_54998_end_0, end_mask = var_54998_end_mask_0, x = v_243_cast_fp16)[name = tensor("op_54998_cast_fp16")]; + tensor var_55002_begin_0 = const()[name = tensor("op_55002_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_55002_end_0 = const()[name = tensor("op_55002_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_55002_end_mask_0 = const()[name = tensor("op_55002_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55002_cast_fp16 = slice_by_index(begin = var_55002_begin_0, end = var_55002_end_0, end_mask = var_55002_end_mask_0, x = v_243_cast_fp16)[name = tensor("op_55002_cast_fp16")]; + tensor var_55006_begin_0 = const()[name = tensor("op_55006_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_55006_end_0 = const()[name = tensor("op_55006_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_55006_end_mask_0 = const()[name = tensor("op_55006_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55006_cast_fp16 = slice_by_index(begin = var_55006_begin_0, end = var_55006_end_0, end_mask = var_55006_end_mask_0, x = v_243_cast_fp16)[name = tensor("op_55006_cast_fp16")]; + tensor var_55010_begin_0 = const()[name = tensor("op_55010_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_55010_end_0 = const()[name = tensor("op_55010_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_55010_end_mask_0 = const()[name = tensor("op_55010_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55010_cast_fp16 = slice_by_index(begin = var_55010_begin_0, end = var_55010_end_0, end_mask = var_55010_end_mask_0, x = v_243_cast_fp16)[name = tensor("op_55010_cast_fp16")]; + tensor var_55014_begin_0 = const()[name = tensor("op_55014_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_55014_end_0 = const()[name = tensor("op_55014_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_55014_end_mask_0 = const()[name = tensor("op_55014_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55014_cast_fp16 = slice_by_index(begin = var_55014_begin_0, end = var_55014_end_0, end_mask = var_55014_end_mask_0, x = v_243_cast_fp16)[name = tensor("op_55014_cast_fp16")]; + tensor var_55018_begin_0 = const()[name = tensor("op_55018_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_55018_end_0 = const()[name = tensor("op_55018_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_55018_end_mask_0 = const()[name = tensor("op_55018_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55018_cast_fp16 = slice_by_index(begin = var_55018_begin_0, end = var_55018_end_0, end_mask = var_55018_end_mask_0, x = v_243_cast_fp16)[name = tensor("op_55018_cast_fp16")]; + tensor var_55022_begin_0 = const()[name = tensor("op_55022_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_55022_end_0 = const()[name = tensor("op_55022_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_55022_end_mask_0 = const()[name = tensor("op_55022_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55022_cast_fp16 = slice_by_index(begin = var_55022_begin_0, end = var_55022_end_0, end_mask = var_55022_end_mask_0, x = v_243_cast_fp16)[name = tensor("op_55022_cast_fp16")]; + tensor var_55026_begin_0 = const()[name = tensor("op_55026_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_55026_end_0 = const()[name = tensor("op_55026_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_55026_end_mask_0 = const()[name = tensor("op_55026_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55026_cast_fp16 = slice_by_index(begin = var_55026_begin_0, end = var_55026_end_0, end_mask = var_55026_end_mask_0, x = v_243_cast_fp16)[name = tensor("op_55026_cast_fp16")]; + tensor var_55030_begin_0 = const()[name = tensor("op_55030_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_55030_end_0 = const()[name = tensor("op_55030_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_55030_end_mask_0 = const()[name = tensor("op_55030_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55030_cast_fp16 = slice_by_index(begin = var_55030_begin_0, end = var_55030_end_0, end_mask = var_55030_end_mask_0, x = v_243_cast_fp16)[name = tensor("op_55030_cast_fp16")]; + tensor var_55034_begin_0 = const()[name = tensor("op_55034_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_55034_end_0 = const()[name = tensor("op_55034_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_55034_end_mask_0 = const()[name = tensor("op_55034_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55034_cast_fp16 = slice_by_index(begin = var_55034_begin_0, end = var_55034_end_0, end_mask = var_55034_end_mask_0, x = v_243_cast_fp16)[name = tensor("op_55034_cast_fp16")]; + tensor var_55038_begin_0 = const()[name = tensor("op_55038_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_55038_end_0 = const()[name = tensor("op_55038_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_55038_end_mask_0 = const()[name = tensor("op_55038_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55038_cast_fp16 = slice_by_index(begin = var_55038_begin_0, end = var_55038_end_0, end_mask = var_55038_end_mask_0, x = v_243_cast_fp16)[name = tensor("op_55038_cast_fp16")]; + tensor var_55042_begin_0 = const()[name = tensor("op_55042_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_55042_end_0 = const()[name = tensor("op_55042_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_55042_end_mask_0 = const()[name = tensor("op_55042_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55042_cast_fp16 = slice_by_index(begin = var_55042_begin_0, end = var_55042_end_0, end_mask = var_55042_end_mask_0, x = v_243_cast_fp16)[name = tensor("op_55042_cast_fp16")]; + tensor var_55046_begin_0 = const()[name = tensor("op_55046_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_55046_end_0 = const()[name = tensor("op_55046_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_55046_end_mask_0 = const()[name = tensor("op_55046_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55046_cast_fp16 = slice_by_index(begin = var_55046_begin_0, end = var_55046_end_0, end_mask = var_55046_end_mask_0, x = v_243_cast_fp16)[name = tensor("op_55046_cast_fp16")]; + tensor var_55050_begin_0 = const()[name = tensor("op_55050_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_55050_end_0 = const()[name = tensor("op_55050_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_55050_end_mask_0 = const()[name = tensor("op_55050_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55050_cast_fp16 = slice_by_index(begin = var_55050_begin_0, end = var_55050_end_0, end_mask = var_55050_end_mask_0, x = v_243_cast_fp16)[name = tensor("op_55050_cast_fp16")]; + tensor var_55054_begin_0 = const()[name = tensor("op_55054_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_55054_end_0 = const()[name = tensor("op_55054_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_55054_end_mask_0 = const()[name = tensor("op_55054_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55054_cast_fp16 = slice_by_index(begin = var_55054_begin_0, end = var_55054_end_0, end_mask = var_55054_end_mask_0, x = v_243_cast_fp16)[name = tensor("op_55054_cast_fp16")]; + tensor var_55058_begin_0 = const()[name = tensor("op_55058_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_55058_end_0 = const()[name = tensor("op_55058_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_55058_end_mask_0 = const()[name = tensor("op_55058_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55058_cast_fp16 = slice_by_index(begin = var_55058_begin_0, end = var_55058_end_0, end_mask = var_55058_end_mask_0, x = v_243_cast_fp16)[name = tensor("op_55058_cast_fp16")]; + tensor var_55062_begin_0 = const()[name = tensor("op_55062_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_55062_end_0 = const()[name = tensor("op_55062_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_55062_end_mask_0 = const()[name = tensor("op_55062_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55062_cast_fp16 = slice_by_index(begin = var_55062_begin_0, end = var_55062_end_0, end_mask = var_55062_end_mask_0, x = v_243_cast_fp16)[name = tensor("op_55062_cast_fp16")]; + tensor var_55066_equation_0 = const()[name = tensor("op_55066_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_55066_cast_fp16 = einsum(equation = var_55066_equation_0, values = (var_54908_cast_fp16, var_54825_cast_fp16))[name = tensor("op_55066_cast_fp16")]; + tensor var_55067_to_fp16 = const()[name = tensor("op_55067_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4681_cast_fp16 = mul(x = var_55066_cast_fp16, y = var_55067_to_fp16)[name = tensor("aw_4681_cast_fp16")]; + tensor var_55070_equation_0 = const()[name = tensor("op_55070_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_55070_cast_fp16 = einsum(equation = var_55070_equation_0, values = (var_54912_cast_fp16, var_54829_cast_fp16))[name = tensor("op_55070_cast_fp16")]; + tensor var_55071_to_fp16 = const()[name = tensor("op_55071_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4683_cast_fp16 = mul(x = var_55070_cast_fp16, y = var_55071_to_fp16)[name = tensor("aw_4683_cast_fp16")]; + tensor var_55074_equation_0 = const()[name = tensor("op_55074_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_55074_cast_fp16 = einsum(equation = var_55074_equation_0, values = (var_54916_cast_fp16, var_54833_cast_fp16))[name = tensor("op_55074_cast_fp16")]; + tensor var_55075_to_fp16 = const()[name = tensor("op_55075_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4685_cast_fp16 = mul(x = var_55074_cast_fp16, y = var_55075_to_fp16)[name = tensor("aw_4685_cast_fp16")]; + tensor var_55078_equation_0 = const()[name = tensor("op_55078_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_55078_cast_fp16 = einsum(equation = var_55078_equation_0, values = (var_54920_cast_fp16, var_54837_cast_fp16))[name = tensor("op_55078_cast_fp16")]; + tensor var_55079_to_fp16 = const()[name = tensor("op_55079_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4687_cast_fp16 = mul(x = var_55078_cast_fp16, y = var_55079_to_fp16)[name = tensor("aw_4687_cast_fp16")]; + tensor var_55082_equation_0 = const()[name = tensor("op_55082_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_55082_cast_fp16 = einsum(equation = var_55082_equation_0, values = (var_54924_cast_fp16, var_54841_cast_fp16))[name = tensor("op_55082_cast_fp16")]; + tensor var_55083_to_fp16 = const()[name = tensor("op_55083_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4689_cast_fp16 = mul(x = var_55082_cast_fp16, y = var_55083_to_fp16)[name = tensor("aw_4689_cast_fp16")]; + tensor var_55086_equation_0 = const()[name = tensor("op_55086_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_55086_cast_fp16 = einsum(equation = var_55086_equation_0, values = (var_54928_cast_fp16, var_54845_cast_fp16))[name = tensor("op_55086_cast_fp16")]; + tensor var_55087_to_fp16 = const()[name = tensor("op_55087_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4691_cast_fp16 = mul(x = var_55086_cast_fp16, y = var_55087_to_fp16)[name = tensor("aw_4691_cast_fp16")]; + tensor var_55090_equation_0 = const()[name = tensor("op_55090_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_55090_cast_fp16 = einsum(equation = var_55090_equation_0, values = (var_54932_cast_fp16, var_54849_cast_fp16))[name = tensor("op_55090_cast_fp16")]; + tensor var_55091_to_fp16 = const()[name = tensor("op_55091_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4693_cast_fp16 = mul(x = var_55090_cast_fp16, y = var_55091_to_fp16)[name = tensor("aw_4693_cast_fp16")]; + tensor var_55094_equation_0 = const()[name = tensor("op_55094_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_55094_cast_fp16 = einsum(equation = var_55094_equation_0, values = (var_54936_cast_fp16, var_54853_cast_fp16))[name = tensor("op_55094_cast_fp16")]; + tensor var_55095_to_fp16 = const()[name = tensor("op_55095_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4695_cast_fp16 = mul(x = var_55094_cast_fp16, y = var_55095_to_fp16)[name = tensor("aw_4695_cast_fp16")]; + tensor var_55098_equation_0 = const()[name = tensor("op_55098_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_55098_cast_fp16 = einsum(equation = var_55098_equation_0, values = (var_54940_cast_fp16, var_54857_cast_fp16))[name = tensor("op_55098_cast_fp16")]; + tensor var_55099_to_fp16 = const()[name = tensor("op_55099_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4697_cast_fp16 = mul(x = var_55098_cast_fp16, y = var_55099_to_fp16)[name = tensor("aw_4697_cast_fp16")]; + tensor var_55102_equation_0 = const()[name = tensor("op_55102_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_55102_cast_fp16 = einsum(equation = var_55102_equation_0, values = (var_54944_cast_fp16, var_54861_cast_fp16))[name = tensor("op_55102_cast_fp16")]; + tensor var_55103_to_fp16 = const()[name = tensor("op_55103_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4699_cast_fp16 = mul(x = var_55102_cast_fp16, y = var_55103_to_fp16)[name = tensor("aw_4699_cast_fp16")]; + tensor var_55106_equation_0 = const()[name = tensor("op_55106_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_55106_cast_fp16 = einsum(equation = var_55106_equation_0, values = (var_54948_cast_fp16, var_54865_cast_fp16))[name = tensor("op_55106_cast_fp16")]; + tensor var_55107_to_fp16 = const()[name = tensor("op_55107_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4701_cast_fp16 = mul(x = var_55106_cast_fp16, y = var_55107_to_fp16)[name = tensor("aw_4701_cast_fp16")]; + tensor var_55110_equation_0 = const()[name = tensor("op_55110_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_55110_cast_fp16 = einsum(equation = var_55110_equation_0, values = (var_54952_cast_fp16, var_54869_cast_fp16))[name = tensor("op_55110_cast_fp16")]; + tensor var_55111_to_fp16 = const()[name = tensor("op_55111_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4703_cast_fp16 = mul(x = var_55110_cast_fp16, y = var_55111_to_fp16)[name = tensor("aw_4703_cast_fp16")]; + tensor var_55114_equation_0 = const()[name = tensor("op_55114_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_55114_cast_fp16 = einsum(equation = var_55114_equation_0, values = (var_54956_cast_fp16, var_54873_cast_fp16))[name = tensor("op_55114_cast_fp16")]; + tensor var_55115_to_fp16 = const()[name = tensor("op_55115_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4705_cast_fp16 = mul(x = var_55114_cast_fp16, y = var_55115_to_fp16)[name = tensor("aw_4705_cast_fp16")]; + tensor var_55118_equation_0 = const()[name = tensor("op_55118_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_55118_cast_fp16 = einsum(equation = var_55118_equation_0, values = (var_54960_cast_fp16, var_54877_cast_fp16))[name = tensor("op_55118_cast_fp16")]; + tensor var_55119_to_fp16 = const()[name = tensor("op_55119_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4707_cast_fp16 = mul(x = var_55118_cast_fp16, y = var_55119_to_fp16)[name = tensor("aw_4707_cast_fp16")]; + tensor var_55122_equation_0 = const()[name = tensor("op_55122_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_55122_cast_fp16 = einsum(equation = var_55122_equation_0, values = (var_54964_cast_fp16, var_54881_cast_fp16))[name = tensor("op_55122_cast_fp16")]; + tensor var_55123_to_fp16 = const()[name = tensor("op_55123_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4709_cast_fp16 = mul(x = var_55122_cast_fp16, y = var_55123_to_fp16)[name = tensor("aw_4709_cast_fp16")]; + tensor var_55126_equation_0 = const()[name = tensor("op_55126_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_55126_cast_fp16 = einsum(equation = var_55126_equation_0, values = (var_54968_cast_fp16, var_54885_cast_fp16))[name = tensor("op_55126_cast_fp16")]; + tensor var_55127_to_fp16 = const()[name = tensor("op_55127_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4711_cast_fp16 = mul(x = var_55126_cast_fp16, y = var_55127_to_fp16)[name = tensor("aw_4711_cast_fp16")]; + tensor var_55130_equation_0 = const()[name = tensor("op_55130_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_55130_cast_fp16 = einsum(equation = var_55130_equation_0, values = (var_54972_cast_fp16, var_54889_cast_fp16))[name = tensor("op_55130_cast_fp16")]; + tensor var_55131_to_fp16 = const()[name = tensor("op_55131_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4713_cast_fp16 = mul(x = var_55130_cast_fp16, y = var_55131_to_fp16)[name = tensor("aw_4713_cast_fp16")]; + tensor var_55134_equation_0 = const()[name = tensor("op_55134_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_55134_cast_fp16 = einsum(equation = var_55134_equation_0, values = (var_54976_cast_fp16, var_54893_cast_fp16))[name = tensor("op_55134_cast_fp16")]; + tensor var_55135_to_fp16 = const()[name = tensor("op_55135_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4715_cast_fp16 = mul(x = var_55134_cast_fp16, y = var_55135_to_fp16)[name = tensor("aw_4715_cast_fp16")]; + tensor var_55138_equation_0 = const()[name = tensor("op_55138_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_55138_cast_fp16 = einsum(equation = var_55138_equation_0, values = (var_54980_cast_fp16, var_54897_cast_fp16))[name = tensor("op_55138_cast_fp16")]; + tensor var_55139_to_fp16 = const()[name = tensor("op_55139_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4717_cast_fp16 = mul(x = var_55138_cast_fp16, y = var_55139_to_fp16)[name = tensor("aw_4717_cast_fp16")]; + tensor var_55142_equation_0 = const()[name = tensor("op_55142_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_55142_cast_fp16 = einsum(equation = var_55142_equation_0, values = (var_54984_cast_fp16, var_54901_cast_fp16))[name = tensor("op_55142_cast_fp16")]; + tensor var_55143_to_fp16 = const()[name = tensor("op_55143_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4719_cast_fp16 = mul(x = var_55142_cast_fp16, y = var_55143_to_fp16)[name = tensor("aw_4719_cast_fp16")]; + tensor var_55145_cast_fp16 = softmax(axis = var_30355, x = aw_4681_cast_fp16)[name = tensor("op_55145_cast_fp16")]; + tensor var_55146_cast_fp16 = softmax(axis = var_30355, x = aw_4683_cast_fp16)[name = tensor("op_55146_cast_fp16")]; + tensor var_55147_cast_fp16 = softmax(axis = var_30355, x = aw_4685_cast_fp16)[name = tensor("op_55147_cast_fp16")]; + tensor var_55148_cast_fp16 = softmax(axis = var_30355, x = aw_4687_cast_fp16)[name = tensor("op_55148_cast_fp16")]; + tensor var_55149_cast_fp16 = softmax(axis = var_30355, x = aw_4689_cast_fp16)[name = tensor("op_55149_cast_fp16")]; + tensor var_55150_cast_fp16 = softmax(axis = var_30355, x = aw_4691_cast_fp16)[name = tensor("op_55150_cast_fp16")]; + tensor var_55151_cast_fp16 = softmax(axis = var_30355, x = aw_4693_cast_fp16)[name = tensor("op_55151_cast_fp16")]; + tensor var_55152_cast_fp16 = softmax(axis = var_30355, x = aw_4695_cast_fp16)[name = tensor("op_55152_cast_fp16")]; + tensor var_55153_cast_fp16 = softmax(axis = var_30355, x = aw_4697_cast_fp16)[name = tensor("op_55153_cast_fp16")]; + tensor var_55154_cast_fp16 = softmax(axis = var_30355, x = aw_4699_cast_fp16)[name = tensor("op_55154_cast_fp16")]; + tensor var_55155_cast_fp16 = softmax(axis = var_30355, x = aw_4701_cast_fp16)[name = tensor("op_55155_cast_fp16")]; + tensor var_55156_cast_fp16 = softmax(axis = var_30355, x = aw_4703_cast_fp16)[name = tensor("op_55156_cast_fp16")]; + tensor var_55157_cast_fp16 = softmax(axis = var_30355, x = aw_4705_cast_fp16)[name = tensor("op_55157_cast_fp16")]; + tensor var_55158_cast_fp16 = softmax(axis = var_30355, x = aw_4707_cast_fp16)[name = tensor("op_55158_cast_fp16")]; + tensor var_55159_cast_fp16 = softmax(axis = var_30355, x = aw_4709_cast_fp16)[name = tensor("op_55159_cast_fp16")]; + tensor var_55160_cast_fp16 = softmax(axis = var_30355, x = aw_4711_cast_fp16)[name = tensor("op_55160_cast_fp16")]; + tensor var_55161_cast_fp16 = softmax(axis = var_30355, x = aw_4713_cast_fp16)[name = tensor("op_55161_cast_fp16")]; + tensor var_55162_cast_fp16 = softmax(axis = var_30355, x = aw_4715_cast_fp16)[name = tensor("op_55162_cast_fp16")]; + tensor var_55163_cast_fp16 = softmax(axis = var_30355, x = aw_4717_cast_fp16)[name = tensor("op_55163_cast_fp16")]; + tensor var_55164_cast_fp16 = softmax(axis = var_30355, x = aw_4719_cast_fp16)[name = tensor("op_55164_cast_fp16")]; + tensor var_55166_equation_0 = const()[name = tensor("op_55166_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_55166_cast_fp16 = einsum(equation = var_55166_equation_0, values = (var_54986_cast_fp16, var_55145_cast_fp16))[name = tensor("op_55166_cast_fp16")]; + tensor var_55168_equation_0 = const()[name = tensor("op_55168_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_55168_cast_fp16 = einsum(equation = var_55168_equation_0, values = (var_54990_cast_fp16, var_55146_cast_fp16))[name = tensor("op_55168_cast_fp16")]; + tensor var_55170_equation_0 = const()[name = tensor("op_55170_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_55170_cast_fp16 = einsum(equation = var_55170_equation_0, values = (var_54994_cast_fp16, var_55147_cast_fp16))[name = tensor("op_55170_cast_fp16")]; + tensor var_55172_equation_0 = const()[name = tensor("op_55172_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_55172_cast_fp16 = einsum(equation = var_55172_equation_0, values = (var_54998_cast_fp16, var_55148_cast_fp16))[name = tensor("op_55172_cast_fp16")]; + tensor var_55174_equation_0 = const()[name = tensor("op_55174_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_55174_cast_fp16 = einsum(equation = var_55174_equation_0, values = (var_55002_cast_fp16, var_55149_cast_fp16))[name = tensor("op_55174_cast_fp16")]; + tensor var_55176_equation_0 = const()[name = tensor("op_55176_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_55176_cast_fp16 = einsum(equation = var_55176_equation_0, values = (var_55006_cast_fp16, var_55150_cast_fp16))[name = tensor("op_55176_cast_fp16")]; + tensor var_55178_equation_0 = const()[name = tensor("op_55178_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_55178_cast_fp16 = einsum(equation = var_55178_equation_0, values = (var_55010_cast_fp16, var_55151_cast_fp16))[name = tensor("op_55178_cast_fp16")]; + tensor var_55180_equation_0 = const()[name = tensor("op_55180_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_55180_cast_fp16 = einsum(equation = var_55180_equation_0, values = (var_55014_cast_fp16, var_55152_cast_fp16))[name = tensor("op_55180_cast_fp16")]; + tensor var_55182_equation_0 = const()[name = tensor("op_55182_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_55182_cast_fp16 = einsum(equation = var_55182_equation_0, values = (var_55018_cast_fp16, var_55153_cast_fp16))[name = tensor("op_55182_cast_fp16")]; + tensor var_55184_equation_0 = const()[name = tensor("op_55184_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_55184_cast_fp16 = einsum(equation = var_55184_equation_0, values = (var_55022_cast_fp16, var_55154_cast_fp16))[name = tensor("op_55184_cast_fp16")]; + tensor var_55186_equation_0 = const()[name = tensor("op_55186_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_55186_cast_fp16 = einsum(equation = var_55186_equation_0, values = (var_55026_cast_fp16, var_55155_cast_fp16))[name = tensor("op_55186_cast_fp16")]; + tensor var_55188_equation_0 = const()[name = tensor("op_55188_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_55188_cast_fp16 = einsum(equation = var_55188_equation_0, values = (var_55030_cast_fp16, var_55156_cast_fp16))[name = tensor("op_55188_cast_fp16")]; + tensor var_55190_equation_0 = const()[name = tensor("op_55190_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_55190_cast_fp16 = einsum(equation = var_55190_equation_0, values = (var_55034_cast_fp16, var_55157_cast_fp16))[name = tensor("op_55190_cast_fp16")]; + tensor var_55192_equation_0 = const()[name = tensor("op_55192_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_55192_cast_fp16 = einsum(equation = var_55192_equation_0, values = (var_55038_cast_fp16, var_55158_cast_fp16))[name = tensor("op_55192_cast_fp16")]; + tensor var_55194_equation_0 = const()[name = tensor("op_55194_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_55194_cast_fp16 = einsum(equation = var_55194_equation_0, values = (var_55042_cast_fp16, var_55159_cast_fp16))[name = tensor("op_55194_cast_fp16")]; + tensor var_55196_equation_0 = const()[name = tensor("op_55196_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_55196_cast_fp16 = einsum(equation = var_55196_equation_0, values = (var_55046_cast_fp16, var_55160_cast_fp16))[name = tensor("op_55196_cast_fp16")]; + tensor var_55198_equation_0 = const()[name = tensor("op_55198_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_55198_cast_fp16 = einsum(equation = var_55198_equation_0, values = (var_55050_cast_fp16, var_55161_cast_fp16))[name = tensor("op_55198_cast_fp16")]; + tensor var_55200_equation_0 = const()[name = tensor("op_55200_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_55200_cast_fp16 = einsum(equation = var_55200_equation_0, values = (var_55054_cast_fp16, var_55162_cast_fp16))[name = tensor("op_55200_cast_fp16")]; + tensor var_55202_equation_0 = const()[name = tensor("op_55202_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_55202_cast_fp16 = einsum(equation = var_55202_equation_0, values = (var_55058_cast_fp16, var_55163_cast_fp16))[name = tensor("op_55202_cast_fp16")]; + tensor var_55204_equation_0 = const()[name = tensor("op_55204_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_55204_cast_fp16 = einsum(equation = var_55204_equation_0, values = (var_55062_cast_fp16, var_55164_cast_fp16))[name = tensor("op_55204_cast_fp16")]; + tensor input_685_interleave_0 = const()[name = tensor("input_685_interleave_0"), val = tensor(false)]; + tensor input_685_cast_fp16 = concat(axis = var_30355, interleave = input_685_interleave_0, values = (var_55166_cast_fp16, var_55168_cast_fp16, var_55170_cast_fp16, var_55172_cast_fp16, var_55174_cast_fp16, var_55176_cast_fp16, var_55178_cast_fp16, var_55180_cast_fp16, var_55182_cast_fp16, var_55184_cast_fp16, var_55186_cast_fp16, var_55188_cast_fp16, var_55190_cast_fp16, var_55192_cast_fp16, var_55194_cast_fp16, var_55196_cast_fp16, var_55198_cast_fp16, var_55200_cast_fp16, var_55202_cast_fp16, var_55204_cast_fp16))[name = tensor("input_685_cast_fp16")]; + tensor var_55210 = const()[name = tensor("op_55210"), val = tensor([1, 1])]; + tensor var_55212 = const()[name = tensor("op_55212"), val = tensor([1, 1])]; + tensor var_55214_pad_type_0 = const()[name = tensor("op_55214_pad_type_0"), val = tensor("custom")]; + tensor var_55214_pad_0 = const()[name = tensor("op_55214_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_6_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_6_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4616052096)))]; + tensor up_blocks_0_attentions_2_transformer_blocks_6_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_6_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4619328960)))]; + tensor var_55214_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_6_attn2_to_out_0_bias_to_fp16, dilations = var_55212, groups = var_30355, pad = var_55214_pad_0, pad_type = var_55214_pad_type_0, strides = var_55210, weight = up_blocks_0_attentions_2_transformer_blocks_6_attn2_to_out_0_weight_to_fp16, x = input_685_cast_fp16)[name = tensor("op_55214_cast_fp16")]; + tensor inputs_365_cast_fp16 = add(x = var_55214_cast_fp16, y = inputs_363_cast_fp16)[name = tensor("inputs_365_cast_fp16")]; + tensor input_687_axes_0 = const()[name = tensor("input_687_axes_0"), val = tensor([1])]; + tensor input_687_gamma_0_to_fp16 = const()[name = tensor("input_687_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4619331584)))]; + tensor input_687_beta_0_to_fp16 = const()[name = tensor("input_687_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4619334208)))]; + tensor var_55224_to_fp16 = const()[name = tensor("op_55224_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_687_cast_fp16 = layer_norm(axes = input_687_axes_0, beta = input_687_beta_0_to_fp16, epsilon = var_55224_to_fp16, gamma = input_687_gamma_0_to_fp16, x = inputs_365_cast_fp16)[name = tensor("input_687_cast_fp16")]; + tensor var_55240 = const()[name = tensor("op_55240"), val = tensor([1, 1])]; + tensor var_55242 = const()[name = tensor("op_55242"), val = tensor([1, 1])]; + tensor var_55244_pad_type_0 = const()[name = tensor("op_55244_pad_type_0"), val = tensor("custom")]; + tensor var_55244_pad_0 = const()[name = tensor("op_55244_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_6_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_6_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4619336832)))]; + tensor up_blocks_0_attentions_2_transformer_blocks_6_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_6_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4645551296)))]; + tensor var_55244_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_6_ff_net_0_proj_bias_to_fp16, dilations = var_55242, groups = var_30355, pad = var_55244_pad_0, pad_type = var_55244_pad_type_0, strides = var_55240, weight = up_blocks_0_attentions_2_transformer_blocks_6_ff_net_0_proj_weight_to_fp16, x = input_687_cast_fp16)[name = tensor("op_55244_cast_fp16")]; + tensor var_55245_split_sizes_0 = const()[name = tensor("op_55245_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_55245_axis_0 = const()[name = tensor("op_55245_axis_0"), val = tensor(1)]; + tensor var_55245_cast_fp16_0, tensor var_55245_cast_fp16_1 = split(axis = var_55245_axis_0, split_sizes = var_55245_split_sizes_0, x = var_55244_cast_fp16)[name = tensor("op_55245_cast_fp16")]; + tensor var_55247_mode_0 = const()[name = tensor("op_55247_mode_0"), val = tensor("EXACT")]; + tensor var_55247_cast_fp16 = gelu(mode = var_55247_mode_0, x = var_55245_cast_fp16_1)[name = tensor("op_55247_cast_fp16")]; + tensor input_689_cast_fp16 = mul(x = var_55245_cast_fp16_0, y = var_55247_cast_fp16)[name = tensor("input_689_cast_fp16")]; + tensor var_55251 = const()[name = tensor("op_55251"), val = tensor([1, 1])]; + tensor var_55253 = const()[name = tensor("op_55253"), val = tensor([1, 1])]; + tensor var_55255_pad_type_0 = const()[name = tensor("op_55255_pad_type_0"), val = tensor("custom")]; + tensor var_55255_pad_0 = const()[name = tensor("op_55255_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_6_ff_net_2_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_6_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4645571840)))]; + tensor up_blocks_0_attentions_2_transformer_blocks_6_ff_net_2_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_6_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4658679104)))]; + tensor var_55255_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_6_ff_net_2_bias_to_fp16, dilations = var_55253, groups = var_30355, pad = var_55255_pad_0, pad_type = var_55255_pad_type_0, strides = var_55251, weight = up_blocks_0_attentions_2_transformer_blocks_6_ff_net_2_weight_to_fp16, x = input_689_cast_fp16)[name = tensor("op_55255_cast_fp16")]; + tensor inputs_367_cast_fp16 = add(x = var_55255_cast_fp16, y = inputs_365_cast_fp16)[name = tensor("inputs_367_cast_fp16")]; + tensor hidden_states_477_axes_0 = const()[name = tensor("hidden_states_477_axes_0"), val = tensor([1])]; + tensor hidden_states_477_gamma_0_to_fp16 = const()[name = tensor("hidden_states_477_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4658681728)))]; + tensor hidden_states_477_beta_0_to_fp16 = const()[name = tensor("hidden_states_477_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4658684352)))]; + tensor var_55271_to_fp16 = const()[name = tensor("op_55271_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_477_cast_fp16 = layer_norm(axes = hidden_states_477_axes_0, beta = hidden_states_477_beta_0_to_fp16, epsilon = var_55271_to_fp16, gamma = hidden_states_477_gamma_0_to_fp16, x = inputs_367_cast_fp16)[name = tensor("hidden_states_477_cast_fp16")]; + tensor var_55286 = const()[name = tensor("op_55286"), val = tensor([1, 1])]; + tensor var_55288 = const()[name = tensor("op_55288"), val = tensor([1, 1])]; + tensor q_245_pad_type_0 = const()[name = tensor("q_245_pad_type_0"), val = tensor("custom")]; + tensor q_245_pad_0 = const()[name = tensor("q_245_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_7_attn1_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_7_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4658686976)))]; + tensor q_245_cast_fp16 = conv(dilations = var_55288, groups = var_30355, pad = q_245_pad_0, pad_type = q_245_pad_type_0, strides = var_55286, weight = up_blocks_0_attentions_2_transformer_blocks_7_attn1_to_q_weight_to_fp16, x = hidden_states_477_cast_fp16)[name = tensor("q_245_cast_fp16")]; + tensor var_55292 = const()[name = tensor("op_55292"), val = tensor([1, 1])]; + tensor var_55294 = const()[name = tensor("op_55294"), val = tensor([1, 1])]; + tensor k_489_pad_type_0 = const()[name = tensor("k_489_pad_type_0"), val = tensor("custom")]; + tensor k_489_pad_0 = const()[name = tensor("k_489_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_7_attn1_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_7_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4661963840)))]; + tensor k_489_cast_fp16 = conv(dilations = var_55294, groups = var_30355, pad = k_489_pad_0, pad_type = k_489_pad_type_0, strides = var_55292, weight = up_blocks_0_attentions_2_transformer_blocks_7_attn1_to_k_weight_to_fp16, x = hidden_states_477_cast_fp16)[name = tensor("k_489_cast_fp16")]; + tensor var_55298 = const()[name = tensor("op_55298"), val = tensor([1, 1])]; + tensor var_55300 = const()[name = tensor("op_55300"), val = tensor([1, 1])]; + tensor v_245_pad_type_0 = const()[name = tensor("v_245_pad_type_0"), val = tensor("custom")]; + tensor v_245_pad_0 = const()[name = tensor("v_245_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_7_attn1_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_7_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4665240704)))]; + tensor v_245_cast_fp16 = conv(dilations = var_55300, groups = var_30355, pad = v_245_pad_0, pad_type = v_245_pad_type_0, strides = var_55298, weight = up_blocks_0_attentions_2_transformer_blocks_7_attn1_to_v_weight_to_fp16, x = hidden_states_477_cast_fp16)[name = tensor("v_245_cast_fp16")]; + tensor var_55304_begin_0 = const()[name = tensor("op_55304_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_55304_end_0 = const()[name = tensor("op_55304_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_55304_end_mask_0 = const()[name = tensor("op_55304_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55304_cast_fp16 = slice_by_index(begin = var_55304_begin_0, end = var_55304_end_0, end_mask = var_55304_end_mask_0, x = q_245_cast_fp16)[name = tensor("op_55304_cast_fp16")]; + tensor var_55308_begin_0 = const()[name = tensor("op_55308_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_55308_end_0 = const()[name = tensor("op_55308_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_55308_end_mask_0 = const()[name = tensor("op_55308_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55308_cast_fp16 = slice_by_index(begin = var_55308_begin_0, end = var_55308_end_0, end_mask = var_55308_end_mask_0, x = q_245_cast_fp16)[name = tensor("op_55308_cast_fp16")]; + tensor var_55312_begin_0 = const()[name = tensor("op_55312_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_55312_end_0 = const()[name = tensor("op_55312_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_55312_end_mask_0 = const()[name = tensor("op_55312_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55312_cast_fp16 = slice_by_index(begin = var_55312_begin_0, end = var_55312_end_0, end_mask = var_55312_end_mask_0, x = q_245_cast_fp16)[name = tensor("op_55312_cast_fp16")]; + tensor var_55316_begin_0 = const()[name = tensor("op_55316_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_55316_end_0 = const()[name = tensor("op_55316_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_55316_end_mask_0 = const()[name = tensor("op_55316_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55316_cast_fp16 = slice_by_index(begin = var_55316_begin_0, end = var_55316_end_0, end_mask = var_55316_end_mask_0, x = q_245_cast_fp16)[name = tensor("op_55316_cast_fp16")]; + tensor var_55320_begin_0 = const()[name = tensor("op_55320_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_55320_end_0 = const()[name = tensor("op_55320_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_55320_end_mask_0 = const()[name = tensor("op_55320_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55320_cast_fp16 = slice_by_index(begin = var_55320_begin_0, end = var_55320_end_0, end_mask = var_55320_end_mask_0, x = q_245_cast_fp16)[name = tensor("op_55320_cast_fp16")]; + tensor var_55324_begin_0 = const()[name = tensor("op_55324_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_55324_end_0 = const()[name = tensor("op_55324_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_55324_end_mask_0 = const()[name = tensor("op_55324_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55324_cast_fp16 = slice_by_index(begin = var_55324_begin_0, end = var_55324_end_0, end_mask = var_55324_end_mask_0, x = q_245_cast_fp16)[name = tensor("op_55324_cast_fp16")]; + tensor var_55328_begin_0 = const()[name = tensor("op_55328_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_55328_end_0 = const()[name = tensor("op_55328_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_55328_end_mask_0 = const()[name = tensor("op_55328_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55328_cast_fp16 = slice_by_index(begin = var_55328_begin_0, end = var_55328_end_0, end_mask = var_55328_end_mask_0, x = q_245_cast_fp16)[name = tensor("op_55328_cast_fp16")]; + tensor var_55332_begin_0 = const()[name = tensor("op_55332_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_55332_end_0 = const()[name = tensor("op_55332_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_55332_end_mask_0 = const()[name = tensor("op_55332_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55332_cast_fp16 = slice_by_index(begin = var_55332_begin_0, end = var_55332_end_0, end_mask = var_55332_end_mask_0, x = q_245_cast_fp16)[name = tensor("op_55332_cast_fp16")]; + tensor var_55336_begin_0 = const()[name = tensor("op_55336_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_55336_end_0 = const()[name = tensor("op_55336_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_55336_end_mask_0 = const()[name = tensor("op_55336_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55336_cast_fp16 = slice_by_index(begin = var_55336_begin_0, end = var_55336_end_0, end_mask = var_55336_end_mask_0, x = q_245_cast_fp16)[name = tensor("op_55336_cast_fp16")]; + tensor var_55340_begin_0 = const()[name = tensor("op_55340_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_55340_end_0 = const()[name = tensor("op_55340_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_55340_end_mask_0 = const()[name = tensor("op_55340_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55340_cast_fp16 = slice_by_index(begin = var_55340_begin_0, end = var_55340_end_0, end_mask = var_55340_end_mask_0, x = q_245_cast_fp16)[name = tensor("op_55340_cast_fp16")]; + tensor var_55344_begin_0 = const()[name = tensor("op_55344_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_55344_end_0 = const()[name = tensor("op_55344_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_55344_end_mask_0 = const()[name = tensor("op_55344_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55344_cast_fp16 = slice_by_index(begin = var_55344_begin_0, end = var_55344_end_0, end_mask = var_55344_end_mask_0, x = q_245_cast_fp16)[name = tensor("op_55344_cast_fp16")]; + tensor var_55348_begin_0 = const()[name = tensor("op_55348_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_55348_end_0 = const()[name = tensor("op_55348_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_55348_end_mask_0 = const()[name = tensor("op_55348_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55348_cast_fp16 = slice_by_index(begin = var_55348_begin_0, end = var_55348_end_0, end_mask = var_55348_end_mask_0, x = q_245_cast_fp16)[name = tensor("op_55348_cast_fp16")]; + tensor var_55352_begin_0 = const()[name = tensor("op_55352_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_55352_end_0 = const()[name = tensor("op_55352_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_55352_end_mask_0 = const()[name = tensor("op_55352_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55352_cast_fp16 = slice_by_index(begin = var_55352_begin_0, end = var_55352_end_0, end_mask = var_55352_end_mask_0, x = q_245_cast_fp16)[name = tensor("op_55352_cast_fp16")]; + tensor var_55356_begin_0 = const()[name = tensor("op_55356_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_55356_end_0 = const()[name = tensor("op_55356_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_55356_end_mask_0 = const()[name = tensor("op_55356_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55356_cast_fp16 = slice_by_index(begin = var_55356_begin_0, end = var_55356_end_0, end_mask = var_55356_end_mask_0, x = q_245_cast_fp16)[name = tensor("op_55356_cast_fp16")]; + tensor var_55360_begin_0 = const()[name = tensor("op_55360_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_55360_end_0 = const()[name = tensor("op_55360_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_55360_end_mask_0 = const()[name = tensor("op_55360_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55360_cast_fp16 = slice_by_index(begin = var_55360_begin_0, end = var_55360_end_0, end_mask = var_55360_end_mask_0, x = q_245_cast_fp16)[name = tensor("op_55360_cast_fp16")]; + tensor var_55364_begin_0 = const()[name = tensor("op_55364_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_55364_end_0 = const()[name = tensor("op_55364_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_55364_end_mask_0 = const()[name = tensor("op_55364_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55364_cast_fp16 = slice_by_index(begin = var_55364_begin_0, end = var_55364_end_0, end_mask = var_55364_end_mask_0, x = q_245_cast_fp16)[name = tensor("op_55364_cast_fp16")]; + tensor var_55368_begin_0 = const()[name = tensor("op_55368_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_55368_end_0 = const()[name = tensor("op_55368_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_55368_end_mask_0 = const()[name = tensor("op_55368_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55368_cast_fp16 = slice_by_index(begin = var_55368_begin_0, end = var_55368_end_0, end_mask = var_55368_end_mask_0, x = q_245_cast_fp16)[name = tensor("op_55368_cast_fp16")]; + tensor var_55372_begin_0 = const()[name = tensor("op_55372_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_55372_end_0 = const()[name = tensor("op_55372_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_55372_end_mask_0 = const()[name = tensor("op_55372_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55372_cast_fp16 = slice_by_index(begin = var_55372_begin_0, end = var_55372_end_0, end_mask = var_55372_end_mask_0, x = q_245_cast_fp16)[name = tensor("op_55372_cast_fp16")]; + tensor var_55376_begin_0 = const()[name = tensor("op_55376_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_55376_end_0 = const()[name = tensor("op_55376_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_55376_end_mask_0 = const()[name = tensor("op_55376_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55376_cast_fp16 = slice_by_index(begin = var_55376_begin_0, end = var_55376_end_0, end_mask = var_55376_end_mask_0, x = q_245_cast_fp16)[name = tensor("op_55376_cast_fp16")]; + tensor var_55380_begin_0 = const()[name = tensor("op_55380_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_55380_end_0 = const()[name = tensor("op_55380_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_55380_end_mask_0 = const()[name = tensor("op_55380_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55380_cast_fp16 = slice_by_index(begin = var_55380_begin_0, end = var_55380_end_0, end_mask = var_55380_end_mask_0, x = q_245_cast_fp16)[name = tensor("op_55380_cast_fp16")]; + tensor k_491_perm_0 = const()[name = tensor("k_491_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_55387_begin_0 = const()[name = tensor("op_55387_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_55387_end_0 = const()[name = tensor("op_55387_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_55387_end_mask_0 = const()[name = tensor("op_55387_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_491_cast_fp16 = transpose(perm = k_491_perm_0, x = k_489_cast_fp16)[name = tensor("transpose_17")]; + tensor var_55387_cast_fp16 = slice_by_index(begin = var_55387_begin_0, end = var_55387_end_0, end_mask = var_55387_end_mask_0, x = k_491_cast_fp16)[name = tensor("op_55387_cast_fp16")]; + tensor var_55391_begin_0 = const()[name = tensor("op_55391_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_55391_end_0 = const()[name = tensor("op_55391_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_55391_end_mask_0 = const()[name = tensor("op_55391_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_55391_cast_fp16 = slice_by_index(begin = var_55391_begin_0, end = var_55391_end_0, end_mask = var_55391_end_mask_0, x = k_491_cast_fp16)[name = tensor("op_55391_cast_fp16")]; + tensor var_55395_begin_0 = const()[name = tensor("op_55395_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_55395_end_0 = const()[name = tensor("op_55395_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_55395_end_mask_0 = const()[name = tensor("op_55395_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_55395_cast_fp16 = slice_by_index(begin = var_55395_begin_0, end = var_55395_end_0, end_mask = var_55395_end_mask_0, x = k_491_cast_fp16)[name = tensor("op_55395_cast_fp16")]; + tensor var_55399_begin_0 = const()[name = tensor("op_55399_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_55399_end_0 = const()[name = tensor("op_55399_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_55399_end_mask_0 = const()[name = tensor("op_55399_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_55399_cast_fp16 = slice_by_index(begin = var_55399_begin_0, end = var_55399_end_0, end_mask = var_55399_end_mask_0, x = k_491_cast_fp16)[name = tensor("op_55399_cast_fp16")]; + tensor var_55403_begin_0 = const()[name = tensor("op_55403_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_55403_end_0 = const()[name = tensor("op_55403_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_55403_end_mask_0 = const()[name = tensor("op_55403_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_55403_cast_fp16 = slice_by_index(begin = var_55403_begin_0, end = var_55403_end_0, end_mask = var_55403_end_mask_0, x = k_491_cast_fp16)[name = tensor("op_55403_cast_fp16")]; + tensor var_55407_begin_0 = const()[name = tensor("op_55407_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_55407_end_0 = const()[name = tensor("op_55407_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_55407_end_mask_0 = const()[name = tensor("op_55407_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_55407_cast_fp16 = slice_by_index(begin = var_55407_begin_0, end = var_55407_end_0, end_mask = var_55407_end_mask_0, x = k_491_cast_fp16)[name = tensor("op_55407_cast_fp16")]; + tensor var_55411_begin_0 = const()[name = tensor("op_55411_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_55411_end_0 = const()[name = tensor("op_55411_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_55411_end_mask_0 = const()[name = tensor("op_55411_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_55411_cast_fp16 = slice_by_index(begin = var_55411_begin_0, end = var_55411_end_0, end_mask = var_55411_end_mask_0, x = k_491_cast_fp16)[name = tensor("op_55411_cast_fp16")]; + tensor var_55415_begin_0 = const()[name = tensor("op_55415_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_55415_end_0 = const()[name = tensor("op_55415_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_55415_end_mask_0 = const()[name = tensor("op_55415_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_55415_cast_fp16 = slice_by_index(begin = var_55415_begin_0, end = var_55415_end_0, end_mask = var_55415_end_mask_0, x = k_491_cast_fp16)[name = tensor("op_55415_cast_fp16")]; + tensor var_55419_begin_0 = const()[name = tensor("op_55419_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_55419_end_0 = const()[name = tensor("op_55419_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_55419_end_mask_0 = const()[name = tensor("op_55419_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_55419_cast_fp16 = slice_by_index(begin = var_55419_begin_0, end = var_55419_end_0, end_mask = var_55419_end_mask_0, x = k_491_cast_fp16)[name = tensor("op_55419_cast_fp16")]; + tensor var_55423_begin_0 = const()[name = tensor("op_55423_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_55423_end_0 = const()[name = tensor("op_55423_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_55423_end_mask_0 = const()[name = tensor("op_55423_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_55423_cast_fp16 = slice_by_index(begin = var_55423_begin_0, end = var_55423_end_0, end_mask = var_55423_end_mask_0, x = k_491_cast_fp16)[name = tensor("op_55423_cast_fp16")]; + tensor var_55427_begin_0 = const()[name = tensor("op_55427_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_55427_end_0 = const()[name = tensor("op_55427_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_55427_end_mask_0 = const()[name = tensor("op_55427_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_55427_cast_fp16 = slice_by_index(begin = var_55427_begin_0, end = var_55427_end_0, end_mask = var_55427_end_mask_0, x = k_491_cast_fp16)[name = tensor("op_55427_cast_fp16")]; + tensor var_55431_begin_0 = const()[name = tensor("op_55431_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_55431_end_0 = const()[name = tensor("op_55431_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_55431_end_mask_0 = const()[name = tensor("op_55431_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_55431_cast_fp16 = slice_by_index(begin = var_55431_begin_0, end = var_55431_end_0, end_mask = var_55431_end_mask_0, x = k_491_cast_fp16)[name = tensor("op_55431_cast_fp16")]; + tensor var_55435_begin_0 = const()[name = tensor("op_55435_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_55435_end_0 = const()[name = tensor("op_55435_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_55435_end_mask_0 = const()[name = tensor("op_55435_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_55435_cast_fp16 = slice_by_index(begin = var_55435_begin_0, end = var_55435_end_0, end_mask = var_55435_end_mask_0, x = k_491_cast_fp16)[name = tensor("op_55435_cast_fp16")]; + tensor var_55439_begin_0 = const()[name = tensor("op_55439_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_55439_end_0 = const()[name = tensor("op_55439_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_55439_end_mask_0 = const()[name = tensor("op_55439_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_55439_cast_fp16 = slice_by_index(begin = var_55439_begin_0, end = var_55439_end_0, end_mask = var_55439_end_mask_0, x = k_491_cast_fp16)[name = tensor("op_55439_cast_fp16")]; + tensor var_55443_begin_0 = const()[name = tensor("op_55443_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_55443_end_0 = const()[name = tensor("op_55443_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_55443_end_mask_0 = const()[name = tensor("op_55443_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_55443_cast_fp16 = slice_by_index(begin = var_55443_begin_0, end = var_55443_end_0, end_mask = var_55443_end_mask_0, x = k_491_cast_fp16)[name = tensor("op_55443_cast_fp16")]; + tensor var_55447_begin_0 = const()[name = tensor("op_55447_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_55447_end_0 = const()[name = tensor("op_55447_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_55447_end_mask_0 = const()[name = tensor("op_55447_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_55447_cast_fp16 = slice_by_index(begin = var_55447_begin_0, end = var_55447_end_0, end_mask = var_55447_end_mask_0, x = k_491_cast_fp16)[name = tensor("op_55447_cast_fp16")]; + tensor var_55451_begin_0 = const()[name = tensor("op_55451_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_55451_end_0 = const()[name = tensor("op_55451_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_55451_end_mask_0 = const()[name = tensor("op_55451_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_55451_cast_fp16 = slice_by_index(begin = var_55451_begin_0, end = var_55451_end_0, end_mask = var_55451_end_mask_0, x = k_491_cast_fp16)[name = tensor("op_55451_cast_fp16")]; + tensor var_55455_begin_0 = const()[name = tensor("op_55455_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_55455_end_0 = const()[name = tensor("op_55455_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_55455_end_mask_0 = const()[name = tensor("op_55455_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_55455_cast_fp16 = slice_by_index(begin = var_55455_begin_0, end = var_55455_end_0, end_mask = var_55455_end_mask_0, x = k_491_cast_fp16)[name = tensor("op_55455_cast_fp16")]; + tensor var_55459_begin_0 = const()[name = tensor("op_55459_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_55459_end_0 = const()[name = tensor("op_55459_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_55459_end_mask_0 = const()[name = tensor("op_55459_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_55459_cast_fp16 = slice_by_index(begin = var_55459_begin_0, end = var_55459_end_0, end_mask = var_55459_end_mask_0, x = k_491_cast_fp16)[name = tensor("op_55459_cast_fp16")]; + tensor var_55463_begin_0 = const()[name = tensor("op_55463_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_55463_end_0 = const()[name = tensor("op_55463_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_55463_end_mask_0 = const()[name = tensor("op_55463_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_55463_cast_fp16 = slice_by_index(begin = var_55463_begin_0, end = var_55463_end_0, end_mask = var_55463_end_mask_0, x = k_491_cast_fp16)[name = tensor("op_55463_cast_fp16")]; + tensor var_55465_begin_0 = const()[name = tensor("op_55465_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_55465_end_0 = const()[name = tensor("op_55465_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_55465_end_mask_0 = const()[name = tensor("op_55465_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55465_cast_fp16 = slice_by_index(begin = var_55465_begin_0, end = var_55465_end_0, end_mask = var_55465_end_mask_0, x = v_245_cast_fp16)[name = tensor("op_55465_cast_fp16")]; + tensor var_55469_begin_0 = const()[name = tensor("op_55469_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_55469_end_0 = const()[name = tensor("op_55469_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_55469_end_mask_0 = const()[name = tensor("op_55469_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55469_cast_fp16 = slice_by_index(begin = var_55469_begin_0, end = var_55469_end_0, end_mask = var_55469_end_mask_0, x = v_245_cast_fp16)[name = tensor("op_55469_cast_fp16")]; + tensor var_55473_begin_0 = const()[name = tensor("op_55473_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_55473_end_0 = const()[name = tensor("op_55473_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_55473_end_mask_0 = const()[name = tensor("op_55473_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55473_cast_fp16 = slice_by_index(begin = var_55473_begin_0, end = var_55473_end_0, end_mask = var_55473_end_mask_0, x = v_245_cast_fp16)[name = tensor("op_55473_cast_fp16")]; + tensor var_55477_begin_0 = const()[name = tensor("op_55477_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_55477_end_0 = const()[name = tensor("op_55477_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_55477_end_mask_0 = const()[name = tensor("op_55477_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55477_cast_fp16 = slice_by_index(begin = var_55477_begin_0, end = var_55477_end_0, end_mask = var_55477_end_mask_0, x = v_245_cast_fp16)[name = tensor("op_55477_cast_fp16")]; + tensor var_55481_begin_0 = const()[name = tensor("op_55481_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_55481_end_0 = const()[name = tensor("op_55481_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_55481_end_mask_0 = const()[name = tensor("op_55481_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55481_cast_fp16 = slice_by_index(begin = var_55481_begin_0, end = var_55481_end_0, end_mask = var_55481_end_mask_0, x = v_245_cast_fp16)[name = tensor("op_55481_cast_fp16")]; + tensor var_55485_begin_0 = const()[name = tensor("op_55485_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_55485_end_0 = const()[name = tensor("op_55485_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_55485_end_mask_0 = const()[name = tensor("op_55485_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55485_cast_fp16 = slice_by_index(begin = var_55485_begin_0, end = var_55485_end_0, end_mask = var_55485_end_mask_0, x = v_245_cast_fp16)[name = tensor("op_55485_cast_fp16")]; + tensor var_55489_begin_0 = const()[name = tensor("op_55489_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_55489_end_0 = const()[name = tensor("op_55489_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_55489_end_mask_0 = const()[name = tensor("op_55489_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55489_cast_fp16 = slice_by_index(begin = var_55489_begin_0, end = var_55489_end_0, end_mask = var_55489_end_mask_0, x = v_245_cast_fp16)[name = tensor("op_55489_cast_fp16")]; + tensor var_55493_begin_0 = const()[name = tensor("op_55493_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_55493_end_0 = const()[name = tensor("op_55493_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_55493_end_mask_0 = const()[name = tensor("op_55493_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55493_cast_fp16 = slice_by_index(begin = var_55493_begin_0, end = var_55493_end_0, end_mask = var_55493_end_mask_0, x = v_245_cast_fp16)[name = tensor("op_55493_cast_fp16")]; + tensor var_55497_begin_0 = const()[name = tensor("op_55497_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_55497_end_0 = const()[name = tensor("op_55497_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_55497_end_mask_0 = const()[name = tensor("op_55497_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55497_cast_fp16 = slice_by_index(begin = var_55497_begin_0, end = var_55497_end_0, end_mask = var_55497_end_mask_0, x = v_245_cast_fp16)[name = tensor("op_55497_cast_fp16")]; + tensor var_55501_begin_0 = const()[name = tensor("op_55501_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_55501_end_0 = const()[name = tensor("op_55501_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_55501_end_mask_0 = const()[name = tensor("op_55501_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55501_cast_fp16 = slice_by_index(begin = var_55501_begin_0, end = var_55501_end_0, end_mask = var_55501_end_mask_0, x = v_245_cast_fp16)[name = tensor("op_55501_cast_fp16")]; + tensor var_55505_begin_0 = const()[name = tensor("op_55505_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_55505_end_0 = const()[name = tensor("op_55505_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_55505_end_mask_0 = const()[name = tensor("op_55505_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55505_cast_fp16 = slice_by_index(begin = var_55505_begin_0, end = var_55505_end_0, end_mask = var_55505_end_mask_0, x = v_245_cast_fp16)[name = tensor("op_55505_cast_fp16")]; + tensor var_55509_begin_0 = const()[name = tensor("op_55509_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_55509_end_0 = const()[name = tensor("op_55509_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_55509_end_mask_0 = const()[name = tensor("op_55509_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55509_cast_fp16 = slice_by_index(begin = var_55509_begin_0, end = var_55509_end_0, end_mask = var_55509_end_mask_0, x = v_245_cast_fp16)[name = tensor("op_55509_cast_fp16")]; + tensor var_55513_begin_0 = const()[name = tensor("op_55513_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_55513_end_0 = const()[name = tensor("op_55513_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_55513_end_mask_0 = const()[name = tensor("op_55513_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55513_cast_fp16 = slice_by_index(begin = var_55513_begin_0, end = var_55513_end_0, end_mask = var_55513_end_mask_0, x = v_245_cast_fp16)[name = tensor("op_55513_cast_fp16")]; + tensor var_55517_begin_0 = const()[name = tensor("op_55517_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_55517_end_0 = const()[name = tensor("op_55517_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_55517_end_mask_0 = const()[name = tensor("op_55517_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55517_cast_fp16 = slice_by_index(begin = var_55517_begin_0, end = var_55517_end_0, end_mask = var_55517_end_mask_0, x = v_245_cast_fp16)[name = tensor("op_55517_cast_fp16")]; + tensor var_55521_begin_0 = const()[name = tensor("op_55521_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_55521_end_0 = const()[name = tensor("op_55521_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_55521_end_mask_0 = const()[name = tensor("op_55521_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55521_cast_fp16 = slice_by_index(begin = var_55521_begin_0, end = var_55521_end_0, end_mask = var_55521_end_mask_0, x = v_245_cast_fp16)[name = tensor("op_55521_cast_fp16")]; + tensor var_55525_begin_0 = const()[name = tensor("op_55525_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_55525_end_0 = const()[name = tensor("op_55525_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_55525_end_mask_0 = const()[name = tensor("op_55525_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55525_cast_fp16 = slice_by_index(begin = var_55525_begin_0, end = var_55525_end_0, end_mask = var_55525_end_mask_0, x = v_245_cast_fp16)[name = tensor("op_55525_cast_fp16")]; + tensor var_55529_begin_0 = const()[name = tensor("op_55529_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_55529_end_0 = const()[name = tensor("op_55529_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_55529_end_mask_0 = const()[name = tensor("op_55529_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55529_cast_fp16 = slice_by_index(begin = var_55529_begin_0, end = var_55529_end_0, end_mask = var_55529_end_mask_0, x = v_245_cast_fp16)[name = tensor("op_55529_cast_fp16")]; + tensor var_55533_begin_0 = const()[name = tensor("op_55533_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_55533_end_0 = const()[name = tensor("op_55533_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_55533_end_mask_0 = const()[name = tensor("op_55533_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55533_cast_fp16 = slice_by_index(begin = var_55533_begin_0, end = var_55533_end_0, end_mask = var_55533_end_mask_0, x = v_245_cast_fp16)[name = tensor("op_55533_cast_fp16")]; + tensor var_55537_begin_0 = const()[name = tensor("op_55537_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_55537_end_0 = const()[name = tensor("op_55537_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_55537_end_mask_0 = const()[name = tensor("op_55537_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55537_cast_fp16 = slice_by_index(begin = var_55537_begin_0, end = var_55537_end_0, end_mask = var_55537_end_mask_0, x = v_245_cast_fp16)[name = tensor("op_55537_cast_fp16")]; + tensor var_55541_begin_0 = const()[name = tensor("op_55541_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_55541_end_0 = const()[name = tensor("op_55541_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_55541_end_mask_0 = const()[name = tensor("op_55541_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55541_cast_fp16 = slice_by_index(begin = var_55541_begin_0, end = var_55541_end_0, end_mask = var_55541_end_mask_0, x = v_245_cast_fp16)[name = tensor("op_55541_cast_fp16")]; + tensor var_55545_equation_0 = const()[name = tensor("op_55545_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_55545_cast_fp16 = einsum(equation = var_55545_equation_0, values = (var_55387_cast_fp16, var_55304_cast_fp16))[name = tensor("op_55545_cast_fp16")]; + tensor var_55546_to_fp16 = const()[name = tensor("op_55546_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4721_cast_fp16 = mul(x = var_55545_cast_fp16, y = var_55546_to_fp16)[name = tensor("aw_4721_cast_fp16")]; + tensor var_55549_equation_0 = const()[name = tensor("op_55549_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_55549_cast_fp16 = einsum(equation = var_55549_equation_0, values = (var_55391_cast_fp16, var_55308_cast_fp16))[name = tensor("op_55549_cast_fp16")]; + tensor var_55550_to_fp16 = const()[name = tensor("op_55550_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4723_cast_fp16 = mul(x = var_55549_cast_fp16, y = var_55550_to_fp16)[name = tensor("aw_4723_cast_fp16")]; + tensor var_55553_equation_0 = const()[name = tensor("op_55553_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_55553_cast_fp16 = einsum(equation = var_55553_equation_0, values = (var_55395_cast_fp16, var_55312_cast_fp16))[name = tensor("op_55553_cast_fp16")]; + tensor var_55554_to_fp16 = const()[name = tensor("op_55554_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4725_cast_fp16 = mul(x = var_55553_cast_fp16, y = var_55554_to_fp16)[name = tensor("aw_4725_cast_fp16")]; + tensor var_55557_equation_0 = const()[name = tensor("op_55557_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_55557_cast_fp16 = einsum(equation = var_55557_equation_0, values = (var_55399_cast_fp16, var_55316_cast_fp16))[name = tensor("op_55557_cast_fp16")]; + tensor var_55558_to_fp16 = const()[name = tensor("op_55558_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4727_cast_fp16 = mul(x = var_55557_cast_fp16, y = var_55558_to_fp16)[name = tensor("aw_4727_cast_fp16")]; + tensor var_55561_equation_0 = const()[name = tensor("op_55561_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_55561_cast_fp16 = einsum(equation = var_55561_equation_0, values = (var_55403_cast_fp16, var_55320_cast_fp16))[name = tensor("op_55561_cast_fp16")]; + tensor var_55562_to_fp16 = const()[name = tensor("op_55562_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4729_cast_fp16 = mul(x = var_55561_cast_fp16, y = var_55562_to_fp16)[name = tensor("aw_4729_cast_fp16")]; + tensor var_55565_equation_0 = const()[name = tensor("op_55565_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_55565_cast_fp16 = einsum(equation = var_55565_equation_0, values = (var_55407_cast_fp16, var_55324_cast_fp16))[name = tensor("op_55565_cast_fp16")]; + tensor var_55566_to_fp16 = const()[name = tensor("op_55566_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4731_cast_fp16 = mul(x = var_55565_cast_fp16, y = var_55566_to_fp16)[name = tensor("aw_4731_cast_fp16")]; + tensor var_55569_equation_0 = const()[name = tensor("op_55569_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_55569_cast_fp16 = einsum(equation = var_55569_equation_0, values = (var_55411_cast_fp16, var_55328_cast_fp16))[name = tensor("op_55569_cast_fp16")]; + tensor var_55570_to_fp16 = const()[name = tensor("op_55570_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4733_cast_fp16 = mul(x = var_55569_cast_fp16, y = var_55570_to_fp16)[name = tensor("aw_4733_cast_fp16")]; + tensor var_55573_equation_0 = const()[name = tensor("op_55573_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_55573_cast_fp16 = einsum(equation = var_55573_equation_0, values = (var_55415_cast_fp16, var_55332_cast_fp16))[name = tensor("op_55573_cast_fp16")]; + tensor var_55574_to_fp16 = const()[name = tensor("op_55574_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4735_cast_fp16 = mul(x = var_55573_cast_fp16, y = var_55574_to_fp16)[name = tensor("aw_4735_cast_fp16")]; + tensor var_55577_equation_0 = const()[name = tensor("op_55577_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_55577_cast_fp16 = einsum(equation = var_55577_equation_0, values = (var_55419_cast_fp16, var_55336_cast_fp16))[name = tensor("op_55577_cast_fp16")]; + tensor var_55578_to_fp16 = const()[name = tensor("op_55578_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4737_cast_fp16 = mul(x = var_55577_cast_fp16, y = var_55578_to_fp16)[name = tensor("aw_4737_cast_fp16")]; + tensor var_55581_equation_0 = const()[name = tensor("op_55581_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_55581_cast_fp16 = einsum(equation = var_55581_equation_0, values = (var_55423_cast_fp16, var_55340_cast_fp16))[name = tensor("op_55581_cast_fp16")]; + tensor var_55582_to_fp16 = const()[name = tensor("op_55582_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4739_cast_fp16 = mul(x = var_55581_cast_fp16, y = var_55582_to_fp16)[name = tensor("aw_4739_cast_fp16")]; + tensor var_55585_equation_0 = const()[name = tensor("op_55585_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_55585_cast_fp16 = einsum(equation = var_55585_equation_0, values = (var_55427_cast_fp16, var_55344_cast_fp16))[name = tensor("op_55585_cast_fp16")]; + tensor var_55586_to_fp16 = const()[name = tensor("op_55586_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4741_cast_fp16 = mul(x = var_55585_cast_fp16, y = var_55586_to_fp16)[name = tensor("aw_4741_cast_fp16")]; + tensor var_55589_equation_0 = const()[name = tensor("op_55589_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_55589_cast_fp16 = einsum(equation = var_55589_equation_0, values = (var_55431_cast_fp16, var_55348_cast_fp16))[name = tensor("op_55589_cast_fp16")]; + tensor var_55590_to_fp16 = const()[name = tensor("op_55590_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4743_cast_fp16 = mul(x = var_55589_cast_fp16, y = var_55590_to_fp16)[name = tensor("aw_4743_cast_fp16")]; + tensor var_55593_equation_0 = const()[name = tensor("op_55593_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_55593_cast_fp16 = einsum(equation = var_55593_equation_0, values = (var_55435_cast_fp16, var_55352_cast_fp16))[name = tensor("op_55593_cast_fp16")]; + tensor var_55594_to_fp16 = const()[name = tensor("op_55594_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4745_cast_fp16 = mul(x = var_55593_cast_fp16, y = var_55594_to_fp16)[name = tensor("aw_4745_cast_fp16")]; + tensor var_55597_equation_0 = const()[name = tensor("op_55597_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_55597_cast_fp16 = einsum(equation = var_55597_equation_0, values = (var_55439_cast_fp16, var_55356_cast_fp16))[name = tensor("op_55597_cast_fp16")]; + tensor var_55598_to_fp16 = const()[name = tensor("op_55598_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4747_cast_fp16 = mul(x = var_55597_cast_fp16, y = var_55598_to_fp16)[name = tensor("aw_4747_cast_fp16")]; + tensor var_55601_equation_0 = const()[name = tensor("op_55601_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_55601_cast_fp16 = einsum(equation = var_55601_equation_0, values = (var_55443_cast_fp16, var_55360_cast_fp16))[name = tensor("op_55601_cast_fp16")]; + tensor var_55602_to_fp16 = const()[name = tensor("op_55602_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4749_cast_fp16 = mul(x = var_55601_cast_fp16, y = var_55602_to_fp16)[name = tensor("aw_4749_cast_fp16")]; + tensor var_55605_equation_0 = const()[name = tensor("op_55605_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_55605_cast_fp16 = einsum(equation = var_55605_equation_0, values = (var_55447_cast_fp16, var_55364_cast_fp16))[name = tensor("op_55605_cast_fp16")]; + tensor var_55606_to_fp16 = const()[name = tensor("op_55606_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4751_cast_fp16 = mul(x = var_55605_cast_fp16, y = var_55606_to_fp16)[name = tensor("aw_4751_cast_fp16")]; + tensor var_55609_equation_0 = const()[name = tensor("op_55609_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_55609_cast_fp16 = einsum(equation = var_55609_equation_0, values = (var_55451_cast_fp16, var_55368_cast_fp16))[name = tensor("op_55609_cast_fp16")]; + tensor var_55610_to_fp16 = const()[name = tensor("op_55610_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4753_cast_fp16 = mul(x = var_55609_cast_fp16, y = var_55610_to_fp16)[name = tensor("aw_4753_cast_fp16")]; + tensor var_55613_equation_0 = const()[name = tensor("op_55613_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_55613_cast_fp16 = einsum(equation = var_55613_equation_0, values = (var_55455_cast_fp16, var_55372_cast_fp16))[name = tensor("op_55613_cast_fp16")]; + tensor var_55614_to_fp16 = const()[name = tensor("op_55614_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4755_cast_fp16 = mul(x = var_55613_cast_fp16, y = var_55614_to_fp16)[name = tensor("aw_4755_cast_fp16")]; + tensor var_55617_equation_0 = const()[name = tensor("op_55617_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_55617_cast_fp16 = einsum(equation = var_55617_equation_0, values = (var_55459_cast_fp16, var_55376_cast_fp16))[name = tensor("op_55617_cast_fp16")]; + tensor var_55618_to_fp16 = const()[name = tensor("op_55618_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4757_cast_fp16 = mul(x = var_55617_cast_fp16, y = var_55618_to_fp16)[name = tensor("aw_4757_cast_fp16")]; + tensor var_55621_equation_0 = const()[name = tensor("op_55621_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_55621_cast_fp16 = einsum(equation = var_55621_equation_0, values = (var_55463_cast_fp16, var_55380_cast_fp16))[name = tensor("op_55621_cast_fp16")]; + tensor var_55622_to_fp16 = const()[name = tensor("op_55622_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4759_cast_fp16 = mul(x = var_55621_cast_fp16, y = var_55622_to_fp16)[name = tensor("aw_4759_cast_fp16")]; + tensor var_55624_cast_fp16 = softmax(axis = var_30355, x = aw_4721_cast_fp16)[name = tensor("op_55624_cast_fp16")]; + tensor var_55625_cast_fp16 = softmax(axis = var_30355, x = aw_4723_cast_fp16)[name = tensor("op_55625_cast_fp16")]; + tensor var_55626_cast_fp16 = softmax(axis = var_30355, x = aw_4725_cast_fp16)[name = tensor("op_55626_cast_fp16")]; + tensor var_55627_cast_fp16 = softmax(axis = var_30355, x = aw_4727_cast_fp16)[name = tensor("op_55627_cast_fp16")]; + tensor var_55628_cast_fp16 = softmax(axis = var_30355, x = aw_4729_cast_fp16)[name = tensor("op_55628_cast_fp16")]; + tensor var_55629_cast_fp16 = softmax(axis = var_30355, x = aw_4731_cast_fp16)[name = tensor("op_55629_cast_fp16")]; + tensor var_55630_cast_fp16 = softmax(axis = var_30355, x = aw_4733_cast_fp16)[name = tensor("op_55630_cast_fp16")]; + tensor var_55631_cast_fp16 = softmax(axis = var_30355, x = aw_4735_cast_fp16)[name = tensor("op_55631_cast_fp16")]; + tensor var_55632_cast_fp16 = softmax(axis = var_30355, x = aw_4737_cast_fp16)[name = tensor("op_55632_cast_fp16")]; + tensor var_55633_cast_fp16 = softmax(axis = var_30355, x = aw_4739_cast_fp16)[name = tensor("op_55633_cast_fp16")]; + tensor var_55634_cast_fp16 = softmax(axis = var_30355, x = aw_4741_cast_fp16)[name = tensor("op_55634_cast_fp16")]; + tensor var_55635_cast_fp16 = softmax(axis = var_30355, x = aw_4743_cast_fp16)[name = tensor("op_55635_cast_fp16")]; + tensor var_55636_cast_fp16 = softmax(axis = var_30355, x = aw_4745_cast_fp16)[name = tensor("op_55636_cast_fp16")]; + tensor var_55637_cast_fp16 = softmax(axis = var_30355, x = aw_4747_cast_fp16)[name = tensor("op_55637_cast_fp16")]; + tensor var_55638_cast_fp16 = softmax(axis = var_30355, x = aw_4749_cast_fp16)[name = tensor("op_55638_cast_fp16")]; + tensor var_55639_cast_fp16 = softmax(axis = var_30355, x = aw_4751_cast_fp16)[name = tensor("op_55639_cast_fp16")]; + tensor var_55640_cast_fp16 = softmax(axis = var_30355, x = aw_4753_cast_fp16)[name = tensor("op_55640_cast_fp16")]; + tensor var_55641_cast_fp16 = softmax(axis = var_30355, x = aw_4755_cast_fp16)[name = tensor("op_55641_cast_fp16")]; + tensor var_55642_cast_fp16 = softmax(axis = var_30355, x = aw_4757_cast_fp16)[name = tensor("op_55642_cast_fp16")]; + tensor var_55643_cast_fp16 = softmax(axis = var_30355, x = aw_4759_cast_fp16)[name = tensor("op_55643_cast_fp16")]; + tensor var_55645_equation_0 = const()[name = tensor("op_55645_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_55645_cast_fp16 = einsum(equation = var_55645_equation_0, values = (var_55465_cast_fp16, var_55624_cast_fp16))[name = tensor("op_55645_cast_fp16")]; + tensor var_55647_equation_0 = const()[name = tensor("op_55647_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_55647_cast_fp16 = einsum(equation = var_55647_equation_0, values = (var_55469_cast_fp16, var_55625_cast_fp16))[name = tensor("op_55647_cast_fp16")]; + tensor var_55649_equation_0 = const()[name = tensor("op_55649_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_55649_cast_fp16 = einsum(equation = var_55649_equation_0, values = (var_55473_cast_fp16, var_55626_cast_fp16))[name = tensor("op_55649_cast_fp16")]; + tensor var_55651_equation_0 = const()[name = tensor("op_55651_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_55651_cast_fp16 = einsum(equation = var_55651_equation_0, values = (var_55477_cast_fp16, var_55627_cast_fp16))[name = tensor("op_55651_cast_fp16")]; + tensor var_55653_equation_0 = const()[name = tensor("op_55653_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_55653_cast_fp16 = einsum(equation = var_55653_equation_0, values = (var_55481_cast_fp16, var_55628_cast_fp16))[name = tensor("op_55653_cast_fp16")]; + tensor var_55655_equation_0 = const()[name = tensor("op_55655_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_55655_cast_fp16 = einsum(equation = var_55655_equation_0, values = (var_55485_cast_fp16, var_55629_cast_fp16))[name = tensor("op_55655_cast_fp16")]; + tensor var_55657_equation_0 = const()[name = tensor("op_55657_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_55657_cast_fp16 = einsum(equation = var_55657_equation_0, values = (var_55489_cast_fp16, var_55630_cast_fp16))[name = tensor("op_55657_cast_fp16")]; + tensor var_55659_equation_0 = const()[name = tensor("op_55659_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_55659_cast_fp16 = einsum(equation = var_55659_equation_0, values = (var_55493_cast_fp16, var_55631_cast_fp16))[name = tensor("op_55659_cast_fp16")]; + tensor var_55661_equation_0 = const()[name = tensor("op_55661_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_55661_cast_fp16 = einsum(equation = var_55661_equation_0, values = (var_55497_cast_fp16, var_55632_cast_fp16))[name = tensor("op_55661_cast_fp16")]; + tensor var_55663_equation_0 = const()[name = tensor("op_55663_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_55663_cast_fp16 = einsum(equation = var_55663_equation_0, values = (var_55501_cast_fp16, var_55633_cast_fp16))[name = tensor("op_55663_cast_fp16")]; + tensor var_55665_equation_0 = const()[name = tensor("op_55665_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_55665_cast_fp16 = einsum(equation = var_55665_equation_0, values = (var_55505_cast_fp16, var_55634_cast_fp16))[name = tensor("op_55665_cast_fp16")]; + tensor var_55667_equation_0 = const()[name = tensor("op_55667_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_55667_cast_fp16 = einsum(equation = var_55667_equation_0, values = (var_55509_cast_fp16, var_55635_cast_fp16))[name = tensor("op_55667_cast_fp16")]; + tensor var_55669_equation_0 = const()[name = tensor("op_55669_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_55669_cast_fp16 = einsum(equation = var_55669_equation_0, values = (var_55513_cast_fp16, var_55636_cast_fp16))[name = tensor("op_55669_cast_fp16")]; + tensor var_55671_equation_0 = const()[name = tensor("op_55671_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_55671_cast_fp16 = einsum(equation = var_55671_equation_0, values = (var_55517_cast_fp16, var_55637_cast_fp16))[name = tensor("op_55671_cast_fp16")]; + tensor var_55673_equation_0 = const()[name = tensor("op_55673_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_55673_cast_fp16 = einsum(equation = var_55673_equation_0, values = (var_55521_cast_fp16, var_55638_cast_fp16))[name = tensor("op_55673_cast_fp16")]; + tensor var_55675_equation_0 = const()[name = tensor("op_55675_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_55675_cast_fp16 = einsum(equation = var_55675_equation_0, values = (var_55525_cast_fp16, var_55639_cast_fp16))[name = tensor("op_55675_cast_fp16")]; + tensor var_55677_equation_0 = const()[name = tensor("op_55677_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_55677_cast_fp16 = einsum(equation = var_55677_equation_0, values = (var_55529_cast_fp16, var_55640_cast_fp16))[name = tensor("op_55677_cast_fp16")]; + tensor var_55679_equation_0 = const()[name = tensor("op_55679_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_55679_cast_fp16 = einsum(equation = var_55679_equation_0, values = (var_55533_cast_fp16, var_55641_cast_fp16))[name = tensor("op_55679_cast_fp16")]; + tensor var_55681_equation_0 = const()[name = tensor("op_55681_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_55681_cast_fp16 = einsum(equation = var_55681_equation_0, values = (var_55537_cast_fp16, var_55642_cast_fp16))[name = tensor("op_55681_cast_fp16")]; + tensor var_55683_equation_0 = const()[name = tensor("op_55683_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_55683_cast_fp16 = einsum(equation = var_55683_equation_0, values = (var_55541_cast_fp16, var_55643_cast_fp16))[name = tensor("op_55683_cast_fp16")]; + tensor input_691_interleave_0 = const()[name = tensor("input_691_interleave_0"), val = tensor(false)]; + tensor input_691_cast_fp16 = concat(axis = var_30355, interleave = input_691_interleave_0, values = (var_55645_cast_fp16, var_55647_cast_fp16, var_55649_cast_fp16, var_55651_cast_fp16, var_55653_cast_fp16, var_55655_cast_fp16, var_55657_cast_fp16, var_55659_cast_fp16, var_55661_cast_fp16, var_55663_cast_fp16, var_55665_cast_fp16, var_55667_cast_fp16, var_55669_cast_fp16, var_55671_cast_fp16, var_55673_cast_fp16, var_55675_cast_fp16, var_55677_cast_fp16, var_55679_cast_fp16, var_55681_cast_fp16, var_55683_cast_fp16))[name = tensor("input_691_cast_fp16")]; + tensor var_55689 = const()[name = tensor("op_55689"), val = tensor([1, 1])]; + tensor var_55691 = const()[name = tensor("op_55691"), val = tensor([1, 1])]; + tensor var_55693_pad_type_0 = const()[name = tensor("op_55693_pad_type_0"), val = tensor("custom")]; + tensor var_55693_pad_0 = const()[name = tensor("op_55693_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_7_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_7_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4668517568)))]; + tensor up_blocks_0_attentions_2_transformer_blocks_7_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_7_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4671794432)))]; + tensor var_55693_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_7_attn1_to_out_0_bias_to_fp16, dilations = var_55691, groups = var_30355, pad = var_55693_pad_0, pad_type = var_55693_pad_type_0, strides = var_55689, weight = up_blocks_0_attentions_2_transformer_blocks_7_attn1_to_out_0_weight_to_fp16, x = input_691_cast_fp16)[name = tensor("op_55693_cast_fp16")]; + tensor inputs_369_cast_fp16 = add(x = var_55693_cast_fp16, y = inputs_367_cast_fp16)[name = tensor("inputs_369_cast_fp16")]; + tensor hidden_states_479_axes_0 = const()[name = tensor("hidden_states_479_axes_0"), val = tensor([1])]; + tensor hidden_states_479_gamma_0_to_fp16 = const()[name = tensor("hidden_states_479_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4671797056)))]; + tensor hidden_states_479_beta_0_to_fp16 = const()[name = tensor("hidden_states_479_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4671799680)))]; + tensor var_55703_to_fp16 = const()[name = tensor("op_55703_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_479_cast_fp16 = layer_norm(axes = hidden_states_479_axes_0, beta = hidden_states_479_beta_0_to_fp16, epsilon = var_55703_to_fp16, gamma = hidden_states_479_gamma_0_to_fp16, x = inputs_369_cast_fp16)[name = tensor("hidden_states_479_cast_fp16")]; + tensor var_55718 = const()[name = tensor("op_55718"), val = tensor([1, 1])]; + tensor var_55720 = const()[name = tensor("op_55720"), val = tensor([1, 1])]; + tensor q_247_pad_type_0 = const()[name = tensor("q_247_pad_type_0"), val = tensor("custom")]; + tensor q_247_pad_0 = const()[name = tensor("q_247_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_7_attn2_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_7_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4671802304)))]; + tensor q_247_cast_fp16 = conv(dilations = var_55720, groups = var_30355, pad = q_247_pad_0, pad_type = q_247_pad_type_0, strides = var_55718, weight = up_blocks_0_attentions_2_transformer_blocks_7_attn2_to_q_weight_to_fp16, x = hidden_states_479_cast_fp16)[name = tensor("q_247_cast_fp16")]; + tensor var_55724 = const()[name = tensor("op_55724"), val = tensor([1, 1])]; + tensor var_55726 = const()[name = tensor("op_55726"), val = tensor([1, 1])]; + tensor k_493_pad_type_0 = const()[name = tensor("k_493_pad_type_0"), val = tensor("custom")]; + tensor k_493_pad_0 = const()[name = tensor("k_493_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_7_attn2_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_7_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4675079168)))]; + tensor k_493_cast_fp16 = conv(dilations = var_55726, groups = var_30355, pad = k_493_pad_0, pad_type = k_493_pad_type_0, strides = var_55724, weight = up_blocks_0_attentions_2_transformer_blocks_7_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_493_cast_fp16")]; + tensor var_55730 = const()[name = tensor("op_55730"), val = tensor([1, 1])]; + tensor var_55732 = const()[name = tensor("op_55732"), val = tensor([1, 1])]; + tensor v_247_pad_type_0 = const()[name = tensor("v_247_pad_type_0"), val = tensor("custom")]; + tensor v_247_pad_0 = const()[name = tensor("v_247_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_7_attn2_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_7_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4680322112)))]; + tensor v_247_cast_fp16 = conv(dilations = var_55732, groups = var_30355, pad = v_247_pad_0, pad_type = v_247_pad_type_0, strides = var_55730, weight = up_blocks_0_attentions_2_transformer_blocks_7_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_247_cast_fp16")]; + tensor var_55736_begin_0 = const()[name = tensor("op_55736_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_55736_end_0 = const()[name = tensor("op_55736_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_55736_end_mask_0 = const()[name = tensor("op_55736_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55736_cast_fp16 = slice_by_index(begin = var_55736_begin_0, end = var_55736_end_0, end_mask = var_55736_end_mask_0, x = q_247_cast_fp16)[name = tensor("op_55736_cast_fp16")]; + tensor var_55740_begin_0 = const()[name = tensor("op_55740_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_55740_end_0 = const()[name = tensor("op_55740_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_55740_end_mask_0 = const()[name = tensor("op_55740_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55740_cast_fp16 = slice_by_index(begin = var_55740_begin_0, end = var_55740_end_0, end_mask = var_55740_end_mask_0, x = q_247_cast_fp16)[name = tensor("op_55740_cast_fp16")]; + tensor var_55744_begin_0 = const()[name = tensor("op_55744_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_55744_end_0 = const()[name = tensor("op_55744_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_55744_end_mask_0 = const()[name = tensor("op_55744_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55744_cast_fp16 = slice_by_index(begin = var_55744_begin_0, end = var_55744_end_0, end_mask = var_55744_end_mask_0, x = q_247_cast_fp16)[name = tensor("op_55744_cast_fp16")]; + tensor var_55748_begin_0 = const()[name = tensor("op_55748_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_55748_end_0 = const()[name = tensor("op_55748_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_55748_end_mask_0 = const()[name = tensor("op_55748_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55748_cast_fp16 = slice_by_index(begin = var_55748_begin_0, end = var_55748_end_0, end_mask = var_55748_end_mask_0, x = q_247_cast_fp16)[name = tensor("op_55748_cast_fp16")]; + tensor var_55752_begin_0 = const()[name = tensor("op_55752_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_55752_end_0 = const()[name = tensor("op_55752_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_55752_end_mask_0 = const()[name = tensor("op_55752_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55752_cast_fp16 = slice_by_index(begin = var_55752_begin_0, end = var_55752_end_0, end_mask = var_55752_end_mask_0, x = q_247_cast_fp16)[name = tensor("op_55752_cast_fp16")]; + tensor var_55756_begin_0 = const()[name = tensor("op_55756_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_55756_end_0 = const()[name = tensor("op_55756_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_55756_end_mask_0 = const()[name = tensor("op_55756_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55756_cast_fp16 = slice_by_index(begin = var_55756_begin_0, end = var_55756_end_0, end_mask = var_55756_end_mask_0, x = q_247_cast_fp16)[name = tensor("op_55756_cast_fp16")]; + tensor var_55760_begin_0 = const()[name = tensor("op_55760_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_55760_end_0 = const()[name = tensor("op_55760_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_55760_end_mask_0 = const()[name = tensor("op_55760_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55760_cast_fp16 = slice_by_index(begin = var_55760_begin_0, end = var_55760_end_0, end_mask = var_55760_end_mask_0, x = q_247_cast_fp16)[name = tensor("op_55760_cast_fp16")]; + tensor var_55764_begin_0 = const()[name = tensor("op_55764_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_55764_end_0 = const()[name = tensor("op_55764_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_55764_end_mask_0 = const()[name = tensor("op_55764_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55764_cast_fp16 = slice_by_index(begin = var_55764_begin_0, end = var_55764_end_0, end_mask = var_55764_end_mask_0, x = q_247_cast_fp16)[name = tensor("op_55764_cast_fp16")]; + tensor var_55768_begin_0 = const()[name = tensor("op_55768_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_55768_end_0 = const()[name = tensor("op_55768_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_55768_end_mask_0 = const()[name = tensor("op_55768_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55768_cast_fp16 = slice_by_index(begin = var_55768_begin_0, end = var_55768_end_0, end_mask = var_55768_end_mask_0, x = q_247_cast_fp16)[name = tensor("op_55768_cast_fp16")]; + tensor var_55772_begin_0 = const()[name = tensor("op_55772_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_55772_end_0 = const()[name = tensor("op_55772_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_55772_end_mask_0 = const()[name = tensor("op_55772_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55772_cast_fp16 = slice_by_index(begin = var_55772_begin_0, end = var_55772_end_0, end_mask = var_55772_end_mask_0, x = q_247_cast_fp16)[name = tensor("op_55772_cast_fp16")]; + tensor var_55776_begin_0 = const()[name = tensor("op_55776_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_55776_end_0 = const()[name = tensor("op_55776_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_55776_end_mask_0 = const()[name = tensor("op_55776_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55776_cast_fp16 = slice_by_index(begin = var_55776_begin_0, end = var_55776_end_0, end_mask = var_55776_end_mask_0, x = q_247_cast_fp16)[name = tensor("op_55776_cast_fp16")]; + tensor var_55780_begin_0 = const()[name = tensor("op_55780_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_55780_end_0 = const()[name = tensor("op_55780_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_55780_end_mask_0 = const()[name = tensor("op_55780_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55780_cast_fp16 = slice_by_index(begin = var_55780_begin_0, end = var_55780_end_0, end_mask = var_55780_end_mask_0, x = q_247_cast_fp16)[name = tensor("op_55780_cast_fp16")]; + tensor var_55784_begin_0 = const()[name = tensor("op_55784_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_55784_end_0 = const()[name = tensor("op_55784_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_55784_end_mask_0 = const()[name = tensor("op_55784_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55784_cast_fp16 = slice_by_index(begin = var_55784_begin_0, end = var_55784_end_0, end_mask = var_55784_end_mask_0, x = q_247_cast_fp16)[name = tensor("op_55784_cast_fp16")]; + tensor var_55788_begin_0 = const()[name = tensor("op_55788_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_55788_end_0 = const()[name = tensor("op_55788_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_55788_end_mask_0 = const()[name = tensor("op_55788_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55788_cast_fp16 = slice_by_index(begin = var_55788_begin_0, end = var_55788_end_0, end_mask = var_55788_end_mask_0, x = q_247_cast_fp16)[name = tensor("op_55788_cast_fp16")]; + tensor var_55792_begin_0 = const()[name = tensor("op_55792_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_55792_end_0 = const()[name = tensor("op_55792_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_55792_end_mask_0 = const()[name = tensor("op_55792_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55792_cast_fp16 = slice_by_index(begin = var_55792_begin_0, end = var_55792_end_0, end_mask = var_55792_end_mask_0, x = q_247_cast_fp16)[name = tensor("op_55792_cast_fp16")]; + tensor var_55796_begin_0 = const()[name = tensor("op_55796_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_55796_end_0 = const()[name = tensor("op_55796_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_55796_end_mask_0 = const()[name = tensor("op_55796_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55796_cast_fp16 = slice_by_index(begin = var_55796_begin_0, end = var_55796_end_0, end_mask = var_55796_end_mask_0, x = q_247_cast_fp16)[name = tensor("op_55796_cast_fp16")]; + tensor var_55800_begin_0 = const()[name = tensor("op_55800_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_55800_end_0 = const()[name = tensor("op_55800_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_55800_end_mask_0 = const()[name = tensor("op_55800_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55800_cast_fp16 = slice_by_index(begin = var_55800_begin_0, end = var_55800_end_0, end_mask = var_55800_end_mask_0, x = q_247_cast_fp16)[name = tensor("op_55800_cast_fp16")]; + tensor var_55804_begin_0 = const()[name = tensor("op_55804_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_55804_end_0 = const()[name = tensor("op_55804_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_55804_end_mask_0 = const()[name = tensor("op_55804_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55804_cast_fp16 = slice_by_index(begin = var_55804_begin_0, end = var_55804_end_0, end_mask = var_55804_end_mask_0, x = q_247_cast_fp16)[name = tensor("op_55804_cast_fp16")]; + tensor var_55808_begin_0 = const()[name = tensor("op_55808_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_55808_end_0 = const()[name = tensor("op_55808_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_55808_end_mask_0 = const()[name = tensor("op_55808_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55808_cast_fp16 = slice_by_index(begin = var_55808_begin_0, end = var_55808_end_0, end_mask = var_55808_end_mask_0, x = q_247_cast_fp16)[name = tensor("op_55808_cast_fp16")]; + tensor var_55812_begin_0 = const()[name = tensor("op_55812_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_55812_end_0 = const()[name = tensor("op_55812_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_55812_end_mask_0 = const()[name = tensor("op_55812_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55812_cast_fp16 = slice_by_index(begin = var_55812_begin_0, end = var_55812_end_0, end_mask = var_55812_end_mask_0, x = q_247_cast_fp16)[name = tensor("op_55812_cast_fp16")]; + tensor k_495_perm_0 = const()[name = tensor("k_495_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_55819_begin_0 = const()[name = tensor("op_55819_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_55819_end_0 = const()[name = tensor("op_55819_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_55819_end_mask_0 = const()[name = tensor("op_55819_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_495_cast_fp16 = transpose(perm = k_495_perm_0, x = k_493_cast_fp16)[name = tensor("transpose_16")]; + tensor var_55819_cast_fp16 = slice_by_index(begin = var_55819_begin_0, end = var_55819_end_0, end_mask = var_55819_end_mask_0, x = k_495_cast_fp16)[name = tensor("op_55819_cast_fp16")]; + tensor var_55823_begin_0 = const()[name = tensor("op_55823_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_55823_end_0 = const()[name = tensor("op_55823_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_55823_end_mask_0 = const()[name = tensor("op_55823_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_55823_cast_fp16 = slice_by_index(begin = var_55823_begin_0, end = var_55823_end_0, end_mask = var_55823_end_mask_0, x = k_495_cast_fp16)[name = tensor("op_55823_cast_fp16")]; + tensor var_55827_begin_0 = const()[name = tensor("op_55827_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_55827_end_0 = const()[name = tensor("op_55827_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_55827_end_mask_0 = const()[name = tensor("op_55827_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_55827_cast_fp16 = slice_by_index(begin = var_55827_begin_0, end = var_55827_end_0, end_mask = var_55827_end_mask_0, x = k_495_cast_fp16)[name = tensor("op_55827_cast_fp16")]; + tensor var_55831_begin_0 = const()[name = tensor("op_55831_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_55831_end_0 = const()[name = tensor("op_55831_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_55831_end_mask_0 = const()[name = tensor("op_55831_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_55831_cast_fp16 = slice_by_index(begin = var_55831_begin_0, end = var_55831_end_0, end_mask = var_55831_end_mask_0, x = k_495_cast_fp16)[name = tensor("op_55831_cast_fp16")]; + tensor var_55835_begin_0 = const()[name = tensor("op_55835_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_55835_end_0 = const()[name = tensor("op_55835_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_55835_end_mask_0 = const()[name = tensor("op_55835_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_55835_cast_fp16 = slice_by_index(begin = var_55835_begin_0, end = var_55835_end_0, end_mask = var_55835_end_mask_0, x = k_495_cast_fp16)[name = tensor("op_55835_cast_fp16")]; + tensor var_55839_begin_0 = const()[name = tensor("op_55839_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_55839_end_0 = const()[name = tensor("op_55839_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_55839_end_mask_0 = const()[name = tensor("op_55839_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_55839_cast_fp16 = slice_by_index(begin = var_55839_begin_0, end = var_55839_end_0, end_mask = var_55839_end_mask_0, x = k_495_cast_fp16)[name = tensor("op_55839_cast_fp16")]; + tensor var_55843_begin_0 = const()[name = tensor("op_55843_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_55843_end_0 = const()[name = tensor("op_55843_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_55843_end_mask_0 = const()[name = tensor("op_55843_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_55843_cast_fp16 = slice_by_index(begin = var_55843_begin_0, end = var_55843_end_0, end_mask = var_55843_end_mask_0, x = k_495_cast_fp16)[name = tensor("op_55843_cast_fp16")]; + tensor var_55847_begin_0 = const()[name = tensor("op_55847_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_55847_end_0 = const()[name = tensor("op_55847_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_55847_end_mask_0 = const()[name = tensor("op_55847_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_55847_cast_fp16 = slice_by_index(begin = var_55847_begin_0, end = var_55847_end_0, end_mask = var_55847_end_mask_0, x = k_495_cast_fp16)[name = tensor("op_55847_cast_fp16")]; + tensor var_55851_begin_0 = const()[name = tensor("op_55851_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_55851_end_0 = const()[name = tensor("op_55851_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_55851_end_mask_0 = const()[name = tensor("op_55851_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_55851_cast_fp16 = slice_by_index(begin = var_55851_begin_0, end = var_55851_end_0, end_mask = var_55851_end_mask_0, x = k_495_cast_fp16)[name = tensor("op_55851_cast_fp16")]; + tensor var_55855_begin_0 = const()[name = tensor("op_55855_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_55855_end_0 = const()[name = tensor("op_55855_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_55855_end_mask_0 = const()[name = tensor("op_55855_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_55855_cast_fp16 = slice_by_index(begin = var_55855_begin_0, end = var_55855_end_0, end_mask = var_55855_end_mask_0, x = k_495_cast_fp16)[name = tensor("op_55855_cast_fp16")]; + tensor var_55859_begin_0 = const()[name = tensor("op_55859_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_55859_end_0 = const()[name = tensor("op_55859_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_55859_end_mask_0 = const()[name = tensor("op_55859_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_55859_cast_fp16 = slice_by_index(begin = var_55859_begin_0, end = var_55859_end_0, end_mask = var_55859_end_mask_0, x = k_495_cast_fp16)[name = tensor("op_55859_cast_fp16")]; + tensor var_55863_begin_0 = const()[name = tensor("op_55863_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_55863_end_0 = const()[name = tensor("op_55863_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_55863_end_mask_0 = const()[name = tensor("op_55863_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_55863_cast_fp16 = slice_by_index(begin = var_55863_begin_0, end = var_55863_end_0, end_mask = var_55863_end_mask_0, x = k_495_cast_fp16)[name = tensor("op_55863_cast_fp16")]; + tensor var_55867_begin_0 = const()[name = tensor("op_55867_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_55867_end_0 = const()[name = tensor("op_55867_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_55867_end_mask_0 = const()[name = tensor("op_55867_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_55867_cast_fp16 = slice_by_index(begin = var_55867_begin_0, end = var_55867_end_0, end_mask = var_55867_end_mask_0, x = k_495_cast_fp16)[name = tensor("op_55867_cast_fp16")]; + tensor var_55871_begin_0 = const()[name = tensor("op_55871_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_55871_end_0 = const()[name = tensor("op_55871_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_55871_end_mask_0 = const()[name = tensor("op_55871_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_55871_cast_fp16 = slice_by_index(begin = var_55871_begin_0, end = var_55871_end_0, end_mask = var_55871_end_mask_0, x = k_495_cast_fp16)[name = tensor("op_55871_cast_fp16")]; + tensor var_55875_begin_0 = const()[name = tensor("op_55875_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_55875_end_0 = const()[name = tensor("op_55875_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_55875_end_mask_0 = const()[name = tensor("op_55875_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_55875_cast_fp16 = slice_by_index(begin = var_55875_begin_0, end = var_55875_end_0, end_mask = var_55875_end_mask_0, x = k_495_cast_fp16)[name = tensor("op_55875_cast_fp16")]; + tensor var_55879_begin_0 = const()[name = tensor("op_55879_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_55879_end_0 = const()[name = tensor("op_55879_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_55879_end_mask_0 = const()[name = tensor("op_55879_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_55879_cast_fp16 = slice_by_index(begin = var_55879_begin_0, end = var_55879_end_0, end_mask = var_55879_end_mask_0, x = k_495_cast_fp16)[name = tensor("op_55879_cast_fp16")]; + tensor var_55883_begin_0 = const()[name = tensor("op_55883_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_55883_end_0 = const()[name = tensor("op_55883_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_55883_end_mask_0 = const()[name = tensor("op_55883_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_55883_cast_fp16 = slice_by_index(begin = var_55883_begin_0, end = var_55883_end_0, end_mask = var_55883_end_mask_0, x = k_495_cast_fp16)[name = tensor("op_55883_cast_fp16")]; + tensor var_55887_begin_0 = const()[name = tensor("op_55887_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_55887_end_0 = const()[name = tensor("op_55887_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_55887_end_mask_0 = const()[name = tensor("op_55887_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_55887_cast_fp16 = slice_by_index(begin = var_55887_begin_0, end = var_55887_end_0, end_mask = var_55887_end_mask_0, x = k_495_cast_fp16)[name = tensor("op_55887_cast_fp16")]; + tensor var_55891_begin_0 = const()[name = tensor("op_55891_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_55891_end_0 = const()[name = tensor("op_55891_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_55891_end_mask_0 = const()[name = tensor("op_55891_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_55891_cast_fp16 = slice_by_index(begin = var_55891_begin_0, end = var_55891_end_0, end_mask = var_55891_end_mask_0, x = k_495_cast_fp16)[name = tensor("op_55891_cast_fp16")]; + tensor var_55895_begin_0 = const()[name = tensor("op_55895_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_55895_end_0 = const()[name = tensor("op_55895_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_55895_end_mask_0 = const()[name = tensor("op_55895_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_55895_cast_fp16 = slice_by_index(begin = var_55895_begin_0, end = var_55895_end_0, end_mask = var_55895_end_mask_0, x = k_495_cast_fp16)[name = tensor("op_55895_cast_fp16")]; + tensor var_55897_begin_0 = const()[name = tensor("op_55897_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_55897_end_0 = const()[name = tensor("op_55897_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_55897_end_mask_0 = const()[name = tensor("op_55897_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55897_cast_fp16 = slice_by_index(begin = var_55897_begin_0, end = var_55897_end_0, end_mask = var_55897_end_mask_0, x = v_247_cast_fp16)[name = tensor("op_55897_cast_fp16")]; + tensor var_55901_begin_0 = const()[name = tensor("op_55901_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_55901_end_0 = const()[name = tensor("op_55901_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_55901_end_mask_0 = const()[name = tensor("op_55901_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55901_cast_fp16 = slice_by_index(begin = var_55901_begin_0, end = var_55901_end_0, end_mask = var_55901_end_mask_0, x = v_247_cast_fp16)[name = tensor("op_55901_cast_fp16")]; + tensor var_55905_begin_0 = const()[name = tensor("op_55905_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_55905_end_0 = const()[name = tensor("op_55905_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_55905_end_mask_0 = const()[name = tensor("op_55905_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55905_cast_fp16 = slice_by_index(begin = var_55905_begin_0, end = var_55905_end_0, end_mask = var_55905_end_mask_0, x = v_247_cast_fp16)[name = tensor("op_55905_cast_fp16")]; + tensor var_55909_begin_0 = const()[name = tensor("op_55909_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_55909_end_0 = const()[name = tensor("op_55909_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_55909_end_mask_0 = const()[name = tensor("op_55909_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55909_cast_fp16 = slice_by_index(begin = var_55909_begin_0, end = var_55909_end_0, end_mask = var_55909_end_mask_0, x = v_247_cast_fp16)[name = tensor("op_55909_cast_fp16")]; + tensor var_55913_begin_0 = const()[name = tensor("op_55913_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_55913_end_0 = const()[name = tensor("op_55913_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_55913_end_mask_0 = const()[name = tensor("op_55913_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55913_cast_fp16 = slice_by_index(begin = var_55913_begin_0, end = var_55913_end_0, end_mask = var_55913_end_mask_0, x = v_247_cast_fp16)[name = tensor("op_55913_cast_fp16")]; + tensor var_55917_begin_0 = const()[name = tensor("op_55917_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_55917_end_0 = const()[name = tensor("op_55917_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_55917_end_mask_0 = const()[name = tensor("op_55917_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55917_cast_fp16 = slice_by_index(begin = var_55917_begin_0, end = var_55917_end_0, end_mask = var_55917_end_mask_0, x = v_247_cast_fp16)[name = tensor("op_55917_cast_fp16")]; + tensor var_55921_begin_0 = const()[name = tensor("op_55921_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_55921_end_0 = const()[name = tensor("op_55921_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_55921_end_mask_0 = const()[name = tensor("op_55921_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55921_cast_fp16 = slice_by_index(begin = var_55921_begin_0, end = var_55921_end_0, end_mask = var_55921_end_mask_0, x = v_247_cast_fp16)[name = tensor("op_55921_cast_fp16")]; + tensor var_55925_begin_0 = const()[name = tensor("op_55925_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_55925_end_0 = const()[name = tensor("op_55925_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_55925_end_mask_0 = const()[name = tensor("op_55925_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55925_cast_fp16 = slice_by_index(begin = var_55925_begin_0, end = var_55925_end_0, end_mask = var_55925_end_mask_0, x = v_247_cast_fp16)[name = tensor("op_55925_cast_fp16")]; + tensor var_55929_begin_0 = const()[name = tensor("op_55929_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_55929_end_0 = const()[name = tensor("op_55929_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_55929_end_mask_0 = const()[name = tensor("op_55929_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55929_cast_fp16 = slice_by_index(begin = var_55929_begin_0, end = var_55929_end_0, end_mask = var_55929_end_mask_0, x = v_247_cast_fp16)[name = tensor("op_55929_cast_fp16")]; + tensor var_55933_begin_0 = const()[name = tensor("op_55933_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_55933_end_0 = const()[name = tensor("op_55933_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_55933_end_mask_0 = const()[name = tensor("op_55933_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55933_cast_fp16 = slice_by_index(begin = var_55933_begin_0, end = var_55933_end_0, end_mask = var_55933_end_mask_0, x = v_247_cast_fp16)[name = tensor("op_55933_cast_fp16")]; + tensor var_55937_begin_0 = const()[name = tensor("op_55937_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_55937_end_0 = const()[name = tensor("op_55937_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_55937_end_mask_0 = const()[name = tensor("op_55937_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55937_cast_fp16 = slice_by_index(begin = var_55937_begin_0, end = var_55937_end_0, end_mask = var_55937_end_mask_0, x = v_247_cast_fp16)[name = tensor("op_55937_cast_fp16")]; + tensor var_55941_begin_0 = const()[name = tensor("op_55941_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_55941_end_0 = const()[name = tensor("op_55941_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_55941_end_mask_0 = const()[name = tensor("op_55941_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55941_cast_fp16 = slice_by_index(begin = var_55941_begin_0, end = var_55941_end_0, end_mask = var_55941_end_mask_0, x = v_247_cast_fp16)[name = tensor("op_55941_cast_fp16")]; + tensor var_55945_begin_0 = const()[name = tensor("op_55945_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_55945_end_0 = const()[name = tensor("op_55945_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_55945_end_mask_0 = const()[name = tensor("op_55945_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55945_cast_fp16 = slice_by_index(begin = var_55945_begin_0, end = var_55945_end_0, end_mask = var_55945_end_mask_0, x = v_247_cast_fp16)[name = tensor("op_55945_cast_fp16")]; + tensor var_55949_begin_0 = const()[name = tensor("op_55949_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_55949_end_0 = const()[name = tensor("op_55949_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_55949_end_mask_0 = const()[name = tensor("op_55949_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55949_cast_fp16 = slice_by_index(begin = var_55949_begin_0, end = var_55949_end_0, end_mask = var_55949_end_mask_0, x = v_247_cast_fp16)[name = tensor("op_55949_cast_fp16")]; + tensor var_55953_begin_0 = const()[name = tensor("op_55953_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_55953_end_0 = const()[name = tensor("op_55953_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_55953_end_mask_0 = const()[name = tensor("op_55953_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55953_cast_fp16 = slice_by_index(begin = var_55953_begin_0, end = var_55953_end_0, end_mask = var_55953_end_mask_0, x = v_247_cast_fp16)[name = tensor("op_55953_cast_fp16")]; + tensor var_55957_begin_0 = const()[name = tensor("op_55957_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_55957_end_0 = const()[name = tensor("op_55957_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_55957_end_mask_0 = const()[name = tensor("op_55957_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55957_cast_fp16 = slice_by_index(begin = var_55957_begin_0, end = var_55957_end_0, end_mask = var_55957_end_mask_0, x = v_247_cast_fp16)[name = tensor("op_55957_cast_fp16")]; + tensor var_55961_begin_0 = const()[name = tensor("op_55961_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_55961_end_0 = const()[name = tensor("op_55961_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_55961_end_mask_0 = const()[name = tensor("op_55961_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55961_cast_fp16 = slice_by_index(begin = var_55961_begin_0, end = var_55961_end_0, end_mask = var_55961_end_mask_0, x = v_247_cast_fp16)[name = tensor("op_55961_cast_fp16")]; + tensor var_55965_begin_0 = const()[name = tensor("op_55965_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_55965_end_0 = const()[name = tensor("op_55965_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_55965_end_mask_0 = const()[name = tensor("op_55965_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55965_cast_fp16 = slice_by_index(begin = var_55965_begin_0, end = var_55965_end_0, end_mask = var_55965_end_mask_0, x = v_247_cast_fp16)[name = tensor("op_55965_cast_fp16")]; + tensor var_55969_begin_0 = const()[name = tensor("op_55969_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_55969_end_0 = const()[name = tensor("op_55969_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_55969_end_mask_0 = const()[name = tensor("op_55969_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55969_cast_fp16 = slice_by_index(begin = var_55969_begin_0, end = var_55969_end_0, end_mask = var_55969_end_mask_0, x = v_247_cast_fp16)[name = tensor("op_55969_cast_fp16")]; + tensor var_55973_begin_0 = const()[name = tensor("op_55973_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_55973_end_0 = const()[name = tensor("op_55973_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_55973_end_mask_0 = const()[name = tensor("op_55973_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_55973_cast_fp16 = slice_by_index(begin = var_55973_begin_0, end = var_55973_end_0, end_mask = var_55973_end_mask_0, x = v_247_cast_fp16)[name = tensor("op_55973_cast_fp16")]; + tensor var_55977_equation_0 = const()[name = tensor("op_55977_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_55977_cast_fp16 = einsum(equation = var_55977_equation_0, values = (var_55819_cast_fp16, var_55736_cast_fp16))[name = tensor("op_55977_cast_fp16")]; + tensor var_55978_to_fp16 = const()[name = tensor("op_55978_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4761_cast_fp16 = mul(x = var_55977_cast_fp16, y = var_55978_to_fp16)[name = tensor("aw_4761_cast_fp16")]; + tensor var_55981_equation_0 = const()[name = tensor("op_55981_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_55981_cast_fp16 = einsum(equation = var_55981_equation_0, values = (var_55823_cast_fp16, var_55740_cast_fp16))[name = tensor("op_55981_cast_fp16")]; + tensor var_55982_to_fp16 = const()[name = tensor("op_55982_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4763_cast_fp16 = mul(x = var_55981_cast_fp16, y = var_55982_to_fp16)[name = tensor("aw_4763_cast_fp16")]; + tensor var_55985_equation_0 = const()[name = tensor("op_55985_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_55985_cast_fp16 = einsum(equation = var_55985_equation_0, values = (var_55827_cast_fp16, var_55744_cast_fp16))[name = tensor("op_55985_cast_fp16")]; + tensor var_55986_to_fp16 = const()[name = tensor("op_55986_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4765_cast_fp16 = mul(x = var_55985_cast_fp16, y = var_55986_to_fp16)[name = tensor("aw_4765_cast_fp16")]; + tensor var_55989_equation_0 = const()[name = tensor("op_55989_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_55989_cast_fp16 = einsum(equation = var_55989_equation_0, values = (var_55831_cast_fp16, var_55748_cast_fp16))[name = tensor("op_55989_cast_fp16")]; + tensor var_55990_to_fp16 = const()[name = tensor("op_55990_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4767_cast_fp16 = mul(x = var_55989_cast_fp16, y = var_55990_to_fp16)[name = tensor("aw_4767_cast_fp16")]; + tensor var_55993_equation_0 = const()[name = tensor("op_55993_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_55993_cast_fp16 = einsum(equation = var_55993_equation_0, values = (var_55835_cast_fp16, var_55752_cast_fp16))[name = tensor("op_55993_cast_fp16")]; + tensor var_55994_to_fp16 = const()[name = tensor("op_55994_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4769_cast_fp16 = mul(x = var_55993_cast_fp16, y = var_55994_to_fp16)[name = tensor("aw_4769_cast_fp16")]; + tensor var_55997_equation_0 = const()[name = tensor("op_55997_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_55997_cast_fp16 = einsum(equation = var_55997_equation_0, values = (var_55839_cast_fp16, var_55756_cast_fp16))[name = tensor("op_55997_cast_fp16")]; + tensor var_55998_to_fp16 = const()[name = tensor("op_55998_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4771_cast_fp16 = mul(x = var_55997_cast_fp16, y = var_55998_to_fp16)[name = tensor("aw_4771_cast_fp16")]; + tensor var_56001_equation_0 = const()[name = tensor("op_56001_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_56001_cast_fp16 = einsum(equation = var_56001_equation_0, values = (var_55843_cast_fp16, var_55760_cast_fp16))[name = tensor("op_56001_cast_fp16")]; + tensor var_56002_to_fp16 = const()[name = tensor("op_56002_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4773_cast_fp16 = mul(x = var_56001_cast_fp16, y = var_56002_to_fp16)[name = tensor("aw_4773_cast_fp16")]; + tensor var_56005_equation_0 = const()[name = tensor("op_56005_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_56005_cast_fp16 = einsum(equation = var_56005_equation_0, values = (var_55847_cast_fp16, var_55764_cast_fp16))[name = tensor("op_56005_cast_fp16")]; + tensor var_56006_to_fp16 = const()[name = tensor("op_56006_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4775_cast_fp16 = mul(x = var_56005_cast_fp16, y = var_56006_to_fp16)[name = tensor("aw_4775_cast_fp16")]; + tensor var_56009_equation_0 = const()[name = tensor("op_56009_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_56009_cast_fp16 = einsum(equation = var_56009_equation_0, values = (var_55851_cast_fp16, var_55768_cast_fp16))[name = tensor("op_56009_cast_fp16")]; + tensor var_56010_to_fp16 = const()[name = tensor("op_56010_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4777_cast_fp16 = mul(x = var_56009_cast_fp16, y = var_56010_to_fp16)[name = tensor("aw_4777_cast_fp16")]; + tensor var_56013_equation_0 = const()[name = tensor("op_56013_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_56013_cast_fp16 = einsum(equation = var_56013_equation_0, values = (var_55855_cast_fp16, var_55772_cast_fp16))[name = tensor("op_56013_cast_fp16")]; + tensor var_56014_to_fp16 = const()[name = tensor("op_56014_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4779_cast_fp16 = mul(x = var_56013_cast_fp16, y = var_56014_to_fp16)[name = tensor("aw_4779_cast_fp16")]; + tensor var_56017_equation_0 = const()[name = tensor("op_56017_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_56017_cast_fp16 = einsum(equation = var_56017_equation_0, values = (var_55859_cast_fp16, var_55776_cast_fp16))[name = tensor("op_56017_cast_fp16")]; + tensor var_56018_to_fp16 = const()[name = tensor("op_56018_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4781_cast_fp16 = mul(x = var_56017_cast_fp16, y = var_56018_to_fp16)[name = tensor("aw_4781_cast_fp16")]; + tensor var_56021_equation_0 = const()[name = tensor("op_56021_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_56021_cast_fp16 = einsum(equation = var_56021_equation_0, values = (var_55863_cast_fp16, var_55780_cast_fp16))[name = tensor("op_56021_cast_fp16")]; + tensor var_56022_to_fp16 = const()[name = tensor("op_56022_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4783_cast_fp16 = mul(x = var_56021_cast_fp16, y = var_56022_to_fp16)[name = tensor("aw_4783_cast_fp16")]; + tensor var_56025_equation_0 = const()[name = tensor("op_56025_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_56025_cast_fp16 = einsum(equation = var_56025_equation_0, values = (var_55867_cast_fp16, var_55784_cast_fp16))[name = tensor("op_56025_cast_fp16")]; + tensor var_56026_to_fp16 = const()[name = tensor("op_56026_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4785_cast_fp16 = mul(x = var_56025_cast_fp16, y = var_56026_to_fp16)[name = tensor("aw_4785_cast_fp16")]; + tensor var_56029_equation_0 = const()[name = tensor("op_56029_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_56029_cast_fp16 = einsum(equation = var_56029_equation_0, values = (var_55871_cast_fp16, var_55788_cast_fp16))[name = tensor("op_56029_cast_fp16")]; + tensor var_56030_to_fp16 = const()[name = tensor("op_56030_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4787_cast_fp16 = mul(x = var_56029_cast_fp16, y = var_56030_to_fp16)[name = tensor("aw_4787_cast_fp16")]; + tensor var_56033_equation_0 = const()[name = tensor("op_56033_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_56033_cast_fp16 = einsum(equation = var_56033_equation_0, values = (var_55875_cast_fp16, var_55792_cast_fp16))[name = tensor("op_56033_cast_fp16")]; + tensor var_56034_to_fp16 = const()[name = tensor("op_56034_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4789_cast_fp16 = mul(x = var_56033_cast_fp16, y = var_56034_to_fp16)[name = tensor("aw_4789_cast_fp16")]; + tensor var_56037_equation_0 = const()[name = tensor("op_56037_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_56037_cast_fp16 = einsum(equation = var_56037_equation_0, values = (var_55879_cast_fp16, var_55796_cast_fp16))[name = tensor("op_56037_cast_fp16")]; + tensor var_56038_to_fp16 = const()[name = tensor("op_56038_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4791_cast_fp16 = mul(x = var_56037_cast_fp16, y = var_56038_to_fp16)[name = tensor("aw_4791_cast_fp16")]; + tensor var_56041_equation_0 = const()[name = tensor("op_56041_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_56041_cast_fp16 = einsum(equation = var_56041_equation_0, values = (var_55883_cast_fp16, var_55800_cast_fp16))[name = tensor("op_56041_cast_fp16")]; + tensor var_56042_to_fp16 = const()[name = tensor("op_56042_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4793_cast_fp16 = mul(x = var_56041_cast_fp16, y = var_56042_to_fp16)[name = tensor("aw_4793_cast_fp16")]; + tensor var_56045_equation_0 = const()[name = tensor("op_56045_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_56045_cast_fp16 = einsum(equation = var_56045_equation_0, values = (var_55887_cast_fp16, var_55804_cast_fp16))[name = tensor("op_56045_cast_fp16")]; + tensor var_56046_to_fp16 = const()[name = tensor("op_56046_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4795_cast_fp16 = mul(x = var_56045_cast_fp16, y = var_56046_to_fp16)[name = tensor("aw_4795_cast_fp16")]; + tensor var_56049_equation_0 = const()[name = tensor("op_56049_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_56049_cast_fp16 = einsum(equation = var_56049_equation_0, values = (var_55891_cast_fp16, var_55808_cast_fp16))[name = tensor("op_56049_cast_fp16")]; + tensor var_56050_to_fp16 = const()[name = tensor("op_56050_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4797_cast_fp16 = mul(x = var_56049_cast_fp16, y = var_56050_to_fp16)[name = tensor("aw_4797_cast_fp16")]; + tensor var_56053_equation_0 = const()[name = tensor("op_56053_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_56053_cast_fp16 = einsum(equation = var_56053_equation_0, values = (var_55895_cast_fp16, var_55812_cast_fp16))[name = tensor("op_56053_cast_fp16")]; + tensor var_56054_to_fp16 = const()[name = tensor("op_56054_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4799_cast_fp16 = mul(x = var_56053_cast_fp16, y = var_56054_to_fp16)[name = tensor("aw_4799_cast_fp16")]; + tensor var_56056_cast_fp16 = softmax(axis = var_30355, x = aw_4761_cast_fp16)[name = tensor("op_56056_cast_fp16")]; + tensor var_56057_cast_fp16 = softmax(axis = var_30355, x = aw_4763_cast_fp16)[name = tensor("op_56057_cast_fp16")]; + tensor var_56058_cast_fp16 = softmax(axis = var_30355, x = aw_4765_cast_fp16)[name = tensor("op_56058_cast_fp16")]; + tensor var_56059_cast_fp16 = softmax(axis = var_30355, x = aw_4767_cast_fp16)[name = tensor("op_56059_cast_fp16")]; + tensor var_56060_cast_fp16 = softmax(axis = var_30355, x = aw_4769_cast_fp16)[name = tensor("op_56060_cast_fp16")]; + tensor var_56061_cast_fp16 = softmax(axis = var_30355, x = aw_4771_cast_fp16)[name = tensor("op_56061_cast_fp16")]; + tensor var_56062_cast_fp16 = softmax(axis = var_30355, x = aw_4773_cast_fp16)[name = tensor("op_56062_cast_fp16")]; + tensor var_56063_cast_fp16 = softmax(axis = var_30355, x = aw_4775_cast_fp16)[name = tensor("op_56063_cast_fp16")]; + tensor var_56064_cast_fp16 = softmax(axis = var_30355, x = aw_4777_cast_fp16)[name = tensor("op_56064_cast_fp16")]; + tensor var_56065_cast_fp16 = softmax(axis = var_30355, x = aw_4779_cast_fp16)[name = tensor("op_56065_cast_fp16")]; + tensor var_56066_cast_fp16 = softmax(axis = var_30355, x = aw_4781_cast_fp16)[name = tensor("op_56066_cast_fp16")]; + tensor var_56067_cast_fp16 = softmax(axis = var_30355, x = aw_4783_cast_fp16)[name = tensor("op_56067_cast_fp16")]; + tensor var_56068_cast_fp16 = softmax(axis = var_30355, x = aw_4785_cast_fp16)[name = tensor("op_56068_cast_fp16")]; + tensor var_56069_cast_fp16 = softmax(axis = var_30355, x = aw_4787_cast_fp16)[name = tensor("op_56069_cast_fp16")]; + tensor var_56070_cast_fp16 = softmax(axis = var_30355, x = aw_4789_cast_fp16)[name = tensor("op_56070_cast_fp16")]; + tensor var_56071_cast_fp16 = softmax(axis = var_30355, x = aw_4791_cast_fp16)[name = tensor("op_56071_cast_fp16")]; + tensor var_56072_cast_fp16 = softmax(axis = var_30355, x = aw_4793_cast_fp16)[name = tensor("op_56072_cast_fp16")]; + tensor var_56073_cast_fp16 = softmax(axis = var_30355, x = aw_4795_cast_fp16)[name = tensor("op_56073_cast_fp16")]; + tensor var_56074_cast_fp16 = softmax(axis = var_30355, x = aw_4797_cast_fp16)[name = tensor("op_56074_cast_fp16")]; + tensor var_56075_cast_fp16 = softmax(axis = var_30355, x = aw_4799_cast_fp16)[name = tensor("op_56075_cast_fp16")]; + tensor var_56077_equation_0 = const()[name = tensor("op_56077_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_56077_cast_fp16 = einsum(equation = var_56077_equation_0, values = (var_55897_cast_fp16, var_56056_cast_fp16))[name = tensor("op_56077_cast_fp16")]; + tensor var_56079_equation_0 = const()[name = tensor("op_56079_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_56079_cast_fp16 = einsum(equation = var_56079_equation_0, values = (var_55901_cast_fp16, var_56057_cast_fp16))[name = tensor("op_56079_cast_fp16")]; + tensor var_56081_equation_0 = const()[name = tensor("op_56081_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_56081_cast_fp16 = einsum(equation = var_56081_equation_0, values = (var_55905_cast_fp16, var_56058_cast_fp16))[name = tensor("op_56081_cast_fp16")]; + tensor var_56083_equation_0 = const()[name = tensor("op_56083_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_56083_cast_fp16 = einsum(equation = var_56083_equation_0, values = (var_55909_cast_fp16, var_56059_cast_fp16))[name = tensor("op_56083_cast_fp16")]; + tensor var_56085_equation_0 = const()[name = tensor("op_56085_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_56085_cast_fp16 = einsum(equation = var_56085_equation_0, values = (var_55913_cast_fp16, var_56060_cast_fp16))[name = tensor("op_56085_cast_fp16")]; + tensor var_56087_equation_0 = const()[name = tensor("op_56087_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_56087_cast_fp16 = einsum(equation = var_56087_equation_0, values = (var_55917_cast_fp16, var_56061_cast_fp16))[name = tensor("op_56087_cast_fp16")]; + tensor var_56089_equation_0 = const()[name = tensor("op_56089_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_56089_cast_fp16 = einsum(equation = var_56089_equation_0, values = (var_55921_cast_fp16, var_56062_cast_fp16))[name = tensor("op_56089_cast_fp16")]; + tensor var_56091_equation_0 = const()[name = tensor("op_56091_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_56091_cast_fp16 = einsum(equation = var_56091_equation_0, values = (var_55925_cast_fp16, var_56063_cast_fp16))[name = tensor("op_56091_cast_fp16")]; + tensor var_56093_equation_0 = const()[name = tensor("op_56093_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_56093_cast_fp16 = einsum(equation = var_56093_equation_0, values = (var_55929_cast_fp16, var_56064_cast_fp16))[name = tensor("op_56093_cast_fp16")]; + tensor var_56095_equation_0 = const()[name = tensor("op_56095_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_56095_cast_fp16 = einsum(equation = var_56095_equation_0, values = (var_55933_cast_fp16, var_56065_cast_fp16))[name = tensor("op_56095_cast_fp16")]; + tensor var_56097_equation_0 = const()[name = tensor("op_56097_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_56097_cast_fp16 = einsum(equation = var_56097_equation_0, values = (var_55937_cast_fp16, var_56066_cast_fp16))[name = tensor("op_56097_cast_fp16")]; + tensor var_56099_equation_0 = const()[name = tensor("op_56099_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_56099_cast_fp16 = einsum(equation = var_56099_equation_0, values = (var_55941_cast_fp16, var_56067_cast_fp16))[name = tensor("op_56099_cast_fp16")]; + tensor var_56101_equation_0 = const()[name = tensor("op_56101_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_56101_cast_fp16 = einsum(equation = var_56101_equation_0, values = (var_55945_cast_fp16, var_56068_cast_fp16))[name = tensor("op_56101_cast_fp16")]; + tensor var_56103_equation_0 = const()[name = tensor("op_56103_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_56103_cast_fp16 = einsum(equation = var_56103_equation_0, values = (var_55949_cast_fp16, var_56069_cast_fp16))[name = tensor("op_56103_cast_fp16")]; + tensor var_56105_equation_0 = const()[name = tensor("op_56105_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_56105_cast_fp16 = einsum(equation = var_56105_equation_0, values = (var_55953_cast_fp16, var_56070_cast_fp16))[name = tensor("op_56105_cast_fp16")]; + tensor var_56107_equation_0 = const()[name = tensor("op_56107_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_56107_cast_fp16 = einsum(equation = var_56107_equation_0, values = (var_55957_cast_fp16, var_56071_cast_fp16))[name = tensor("op_56107_cast_fp16")]; + tensor var_56109_equation_0 = const()[name = tensor("op_56109_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_56109_cast_fp16 = einsum(equation = var_56109_equation_0, values = (var_55961_cast_fp16, var_56072_cast_fp16))[name = tensor("op_56109_cast_fp16")]; + tensor var_56111_equation_0 = const()[name = tensor("op_56111_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_56111_cast_fp16 = einsum(equation = var_56111_equation_0, values = (var_55965_cast_fp16, var_56073_cast_fp16))[name = tensor("op_56111_cast_fp16")]; + tensor var_56113_equation_0 = const()[name = tensor("op_56113_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_56113_cast_fp16 = einsum(equation = var_56113_equation_0, values = (var_55969_cast_fp16, var_56074_cast_fp16))[name = tensor("op_56113_cast_fp16")]; + tensor var_56115_equation_0 = const()[name = tensor("op_56115_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_56115_cast_fp16 = einsum(equation = var_56115_equation_0, values = (var_55973_cast_fp16, var_56075_cast_fp16))[name = tensor("op_56115_cast_fp16")]; + tensor input_693_interleave_0 = const()[name = tensor("input_693_interleave_0"), val = tensor(false)]; + tensor input_693_cast_fp16 = concat(axis = var_30355, interleave = input_693_interleave_0, values = (var_56077_cast_fp16, var_56079_cast_fp16, var_56081_cast_fp16, var_56083_cast_fp16, var_56085_cast_fp16, var_56087_cast_fp16, var_56089_cast_fp16, var_56091_cast_fp16, var_56093_cast_fp16, var_56095_cast_fp16, var_56097_cast_fp16, var_56099_cast_fp16, var_56101_cast_fp16, var_56103_cast_fp16, var_56105_cast_fp16, var_56107_cast_fp16, var_56109_cast_fp16, var_56111_cast_fp16, var_56113_cast_fp16, var_56115_cast_fp16))[name = tensor("input_693_cast_fp16")]; + tensor var_56121 = const()[name = tensor("op_56121"), val = tensor([1, 1])]; + tensor var_56123 = const()[name = tensor("op_56123"), val = tensor([1, 1])]; + tensor var_56125_pad_type_0 = const()[name = tensor("op_56125_pad_type_0"), val = tensor("custom")]; + tensor var_56125_pad_0 = const()[name = tensor("op_56125_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_7_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_7_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4685565056)))]; + tensor up_blocks_0_attentions_2_transformer_blocks_7_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_7_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4688841920)))]; + tensor var_56125_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_7_attn2_to_out_0_bias_to_fp16, dilations = var_56123, groups = var_30355, pad = var_56125_pad_0, pad_type = var_56125_pad_type_0, strides = var_56121, weight = up_blocks_0_attentions_2_transformer_blocks_7_attn2_to_out_0_weight_to_fp16, x = input_693_cast_fp16)[name = tensor("op_56125_cast_fp16")]; + tensor inputs_371_cast_fp16 = add(x = var_56125_cast_fp16, y = inputs_369_cast_fp16)[name = tensor("inputs_371_cast_fp16")]; + tensor input_695_axes_0 = const()[name = tensor("input_695_axes_0"), val = tensor([1])]; + tensor input_695_gamma_0_to_fp16 = const()[name = tensor("input_695_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4688844544)))]; + tensor input_695_beta_0_to_fp16 = const()[name = tensor("input_695_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4688847168)))]; + tensor var_56135_to_fp16 = const()[name = tensor("op_56135_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_695_cast_fp16 = layer_norm(axes = input_695_axes_0, beta = input_695_beta_0_to_fp16, epsilon = var_56135_to_fp16, gamma = input_695_gamma_0_to_fp16, x = inputs_371_cast_fp16)[name = tensor("input_695_cast_fp16")]; + tensor var_56151 = const()[name = tensor("op_56151"), val = tensor([1, 1])]; + tensor var_56153 = const()[name = tensor("op_56153"), val = tensor([1, 1])]; + tensor var_56155_pad_type_0 = const()[name = tensor("op_56155_pad_type_0"), val = tensor("custom")]; + tensor var_56155_pad_0 = const()[name = tensor("op_56155_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_7_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_7_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4688849792)))]; + tensor up_blocks_0_attentions_2_transformer_blocks_7_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_7_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4715064256)))]; + tensor var_56155_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_7_ff_net_0_proj_bias_to_fp16, dilations = var_56153, groups = var_30355, pad = var_56155_pad_0, pad_type = var_56155_pad_type_0, strides = var_56151, weight = up_blocks_0_attentions_2_transformer_blocks_7_ff_net_0_proj_weight_to_fp16, x = input_695_cast_fp16)[name = tensor("op_56155_cast_fp16")]; + tensor var_56156_split_sizes_0 = const()[name = tensor("op_56156_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_56156_axis_0 = const()[name = tensor("op_56156_axis_0"), val = tensor(1)]; + tensor var_56156_cast_fp16_0, tensor var_56156_cast_fp16_1 = split(axis = var_56156_axis_0, split_sizes = var_56156_split_sizes_0, x = var_56155_cast_fp16)[name = tensor("op_56156_cast_fp16")]; + tensor var_56158_mode_0 = const()[name = tensor("op_56158_mode_0"), val = tensor("EXACT")]; + tensor var_56158_cast_fp16 = gelu(mode = var_56158_mode_0, x = var_56156_cast_fp16_1)[name = tensor("op_56158_cast_fp16")]; + tensor input_697_cast_fp16 = mul(x = var_56156_cast_fp16_0, y = var_56158_cast_fp16)[name = tensor("input_697_cast_fp16")]; + tensor var_56162 = const()[name = tensor("op_56162"), val = tensor([1, 1])]; + tensor var_56164 = const()[name = tensor("op_56164"), val = tensor([1, 1])]; + tensor var_56166_pad_type_0 = const()[name = tensor("op_56166_pad_type_0"), val = tensor("custom")]; + tensor var_56166_pad_0 = const()[name = tensor("op_56166_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_7_ff_net_2_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_7_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4715084800)))]; + tensor up_blocks_0_attentions_2_transformer_blocks_7_ff_net_2_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_7_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4728192064)))]; + tensor var_56166_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_7_ff_net_2_bias_to_fp16, dilations = var_56164, groups = var_30355, pad = var_56166_pad_0, pad_type = var_56166_pad_type_0, strides = var_56162, weight = up_blocks_0_attentions_2_transformer_blocks_7_ff_net_2_weight_to_fp16, x = input_697_cast_fp16)[name = tensor("op_56166_cast_fp16")]; + tensor inputs_373_cast_fp16 = add(x = var_56166_cast_fp16, y = inputs_371_cast_fp16)[name = tensor("inputs_373_cast_fp16")]; + tensor hidden_states_483_axes_0 = const()[name = tensor("hidden_states_483_axes_0"), val = tensor([1])]; + tensor hidden_states_483_gamma_0_to_fp16 = const()[name = tensor("hidden_states_483_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4728194688)))]; + tensor hidden_states_483_beta_0_to_fp16 = const()[name = tensor("hidden_states_483_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4728197312)))]; + tensor var_56182_to_fp16 = const()[name = tensor("op_56182_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_483_cast_fp16 = layer_norm(axes = hidden_states_483_axes_0, beta = hidden_states_483_beta_0_to_fp16, epsilon = var_56182_to_fp16, gamma = hidden_states_483_gamma_0_to_fp16, x = inputs_373_cast_fp16)[name = tensor("hidden_states_483_cast_fp16")]; + tensor var_56197 = const()[name = tensor("op_56197"), val = tensor([1, 1])]; + tensor var_56199 = const()[name = tensor("op_56199"), val = tensor([1, 1])]; + tensor q_249_pad_type_0 = const()[name = tensor("q_249_pad_type_0"), val = tensor("custom")]; + tensor q_249_pad_0 = const()[name = tensor("q_249_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_8_attn1_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_8_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4728199936)))]; + tensor q_249_cast_fp16 = conv(dilations = var_56199, groups = var_30355, pad = q_249_pad_0, pad_type = q_249_pad_type_0, strides = var_56197, weight = up_blocks_0_attentions_2_transformer_blocks_8_attn1_to_q_weight_to_fp16, x = hidden_states_483_cast_fp16)[name = tensor("q_249_cast_fp16")]; + tensor var_56203 = const()[name = tensor("op_56203"), val = tensor([1, 1])]; + tensor var_56205 = const()[name = tensor("op_56205"), val = tensor([1, 1])]; + tensor k_497_pad_type_0 = const()[name = tensor("k_497_pad_type_0"), val = tensor("custom")]; + tensor k_497_pad_0 = const()[name = tensor("k_497_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_8_attn1_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_8_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4731476800)))]; + tensor k_497_cast_fp16 = conv(dilations = var_56205, groups = var_30355, pad = k_497_pad_0, pad_type = k_497_pad_type_0, strides = var_56203, weight = up_blocks_0_attentions_2_transformer_blocks_8_attn1_to_k_weight_to_fp16, x = hidden_states_483_cast_fp16)[name = tensor("k_497_cast_fp16")]; + tensor var_56209 = const()[name = tensor("op_56209"), val = tensor([1, 1])]; + tensor var_56211 = const()[name = tensor("op_56211"), val = tensor([1, 1])]; + tensor v_249_pad_type_0 = const()[name = tensor("v_249_pad_type_0"), val = tensor("custom")]; + tensor v_249_pad_0 = const()[name = tensor("v_249_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_8_attn1_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_8_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4734753664)))]; + tensor v_249_cast_fp16 = conv(dilations = var_56211, groups = var_30355, pad = v_249_pad_0, pad_type = v_249_pad_type_0, strides = var_56209, weight = up_blocks_0_attentions_2_transformer_blocks_8_attn1_to_v_weight_to_fp16, x = hidden_states_483_cast_fp16)[name = tensor("v_249_cast_fp16")]; + tensor var_56215_begin_0 = const()[name = tensor("op_56215_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_56215_end_0 = const()[name = tensor("op_56215_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_56215_end_mask_0 = const()[name = tensor("op_56215_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56215_cast_fp16 = slice_by_index(begin = var_56215_begin_0, end = var_56215_end_0, end_mask = var_56215_end_mask_0, x = q_249_cast_fp16)[name = tensor("op_56215_cast_fp16")]; + tensor var_56219_begin_0 = const()[name = tensor("op_56219_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_56219_end_0 = const()[name = tensor("op_56219_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_56219_end_mask_0 = const()[name = tensor("op_56219_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56219_cast_fp16 = slice_by_index(begin = var_56219_begin_0, end = var_56219_end_0, end_mask = var_56219_end_mask_0, x = q_249_cast_fp16)[name = tensor("op_56219_cast_fp16")]; + tensor var_56223_begin_0 = const()[name = tensor("op_56223_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_56223_end_0 = const()[name = tensor("op_56223_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_56223_end_mask_0 = const()[name = tensor("op_56223_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56223_cast_fp16 = slice_by_index(begin = var_56223_begin_0, end = var_56223_end_0, end_mask = var_56223_end_mask_0, x = q_249_cast_fp16)[name = tensor("op_56223_cast_fp16")]; + tensor var_56227_begin_0 = const()[name = tensor("op_56227_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_56227_end_0 = const()[name = tensor("op_56227_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_56227_end_mask_0 = const()[name = tensor("op_56227_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56227_cast_fp16 = slice_by_index(begin = var_56227_begin_0, end = var_56227_end_0, end_mask = var_56227_end_mask_0, x = q_249_cast_fp16)[name = tensor("op_56227_cast_fp16")]; + tensor var_56231_begin_0 = const()[name = tensor("op_56231_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_56231_end_0 = const()[name = tensor("op_56231_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_56231_end_mask_0 = const()[name = tensor("op_56231_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56231_cast_fp16 = slice_by_index(begin = var_56231_begin_0, end = var_56231_end_0, end_mask = var_56231_end_mask_0, x = q_249_cast_fp16)[name = tensor("op_56231_cast_fp16")]; + tensor var_56235_begin_0 = const()[name = tensor("op_56235_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_56235_end_0 = const()[name = tensor("op_56235_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_56235_end_mask_0 = const()[name = tensor("op_56235_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56235_cast_fp16 = slice_by_index(begin = var_56235_begin_0, end = var_56235_end_0, end_mask = var_56235_end_mask_0, x = q_249_cast_fp16)[name = tensor("op_56235_cast_fp16")]; + tensor var_56239_begin_0 = const()[name = tensor("op_56239_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_56239_end_0 = const()[name = tensor("op_56239_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_56239_end_mask_0 = const()[name = tensor("op_56239_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56239_cast_fp16 = slice_by_index(begin = var_56239_begin_0, end = var_56239_end_0, end_mask = var_56239_end_mask_0, x = q_249_cast_fp16)[name = tensor("op_56239_cast_fp16")]; + tensor var_56243_begin_0 = const()[name = tensor("op_56243_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_56243_end_0 = const()[name = tensor("op_56243_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_56243_end_mask_0 = const()[name = tensor("op_56243_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56243_cast_fp16 = slice_by_index(begin = var_56243_begin_0, end = var_56243_end_0, end_mask = var_56243_end_mask_0, x = q_249_cast_fp16)[name = tensor("op_56243_cast_fp16")]; + tensor var_56247_begin_0 = const()[name = tensor("op_56247_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_56247_end_0 = const()[name = tensor("op_56247_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_56247_end_mask_0 = const()[name = tensor("op_56247_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56247_cast_fp16 = slice_by_index(begin = var_56247_begin_0, end = var_56247_end_0, end_mask = var_56247_end_mask_0, x = q_249_cast_fp16)[name = tensor("op_56247_cast_fp16")]; + tensor var_56251_begin_0 = const()[name = tensor("op_56251_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_56251_end_0 = const()[name = tensor("op_56251_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_56251_end_mask_0 = const()[name = tensor("op_56251_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56251_cast_fp16 = slice_by_index(begin = var_56251_begin_0, end = var_56251_end_0, end_mask = var_56251_end_mask_0, x = q_249_cast_fp16)[name = tensor("op_56251_cast_fp16")]; + tensor var_56255_begin_0 = const()[name = tensor("op_56255_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_56255_end_0 = const()[name = tensor("op_56255_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_56255_end_mask_0 = const()[name = tensor("op_56255_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56255_cast_fp16 = slice_by_index(begin = var_56255_begin_0, end = var_56255_end_0, end_mask = var_56255_end_mask_0, x = q_249_cast_fp16)[name = tensor("op_56255_cast_fp16")]; + tensor var_56259_begin_0 = const()[name = tensor("op_56259_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_56259_end_0 = const()[name = tensor("op_56259_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_56259_end_mask_0 = const()[name = tensor("op_56259_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56259_cast_fp16 = slice_by_index(begin = var_56259_begin_0, end = var_56259_end_0, end_mask = var_56259_end_mask_0, x = q_249_cast_fp16)[name = tensor("op_56259_cast_fp16")]; + tensor var_56263_begin_0 = const()[name = tensor("op_56263_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_56263_end_0 = const()[name = tensor("op_56263_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_56263_end_mask_0 = const()[name = tensor("op_56263_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56263_cast_fp16 = slice_by_index(begin = var_56263_begin_0, end = var_56263_end_0, end_mask = var_56263_end_mask_0, x = q_249_cast_fp16)[name = tensor("op_56263_cast_fp16")]; + tensor var_56267_begin_0 = const()[name = tensor("op_56267_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_56267_end_0 = const()[name = tensor("op_56267_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_56267_end_mask_0 = const()[name = tensor("op_56267_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56267_cast_fp16 = slice_by_index(begin = var_56267_begin_0, end = var_56267_end_0, end_mask = var_56267_end_mask_0, x = q_249_cast_fp16)[name = tensor("op_56267_cast_fp16")]; + tensor var_56271_begin_0 = const()[name = tensor("op_56271_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_56271_end_0 = const()[name = tensor("op_56271_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_56271_end_mask_0 = const()[name = tensor("op_56271_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56271_cast_fp16 = slice_by_index(begin = var_56271_begin_0, end = var_56271_end_0, end_mask = var_56271_end_mask_0, x = q_249_cast_fp16)[name = tensor("op_56271_cast_fp16")]; + tensor var_56275_begin_0 = const()[name = tensor("op_56275_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_56275_end_0 = const()[name = tensor("op_56275_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_56275_end_mask_0 = const()[name = tensor("op_56275_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56275_cast_fp16 = slice_by_index(begin = var_56275_begin_0, end = var_56275_end_0, end_mask = var_56275_end_mask_0, x = q_249_cast_fp16)[name = tensor("op_56275_cast_fp16")]; + tensor var_56279_begin_0 = const()[name = tensor("op_56279_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_56279_end_0 = const()[name = tensor("op_56279_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_56279_end_mask_0 = const()[name = tensor("op_56279_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56279_cast_fp16 = slice_by_index(begin = var_56279_begin_0, end = var_56279_end_0, end_mask = var_56279_end_mask_0, x = q_249_cast_fp16)[name = tensor("op_56279_cast_fp16")]; + tensor var_56283_begin_0 = const()[name = tensor("op_56283_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_56283_end_0 = const()[name = tensor("op_56283_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_56283_end_mask_0 = const()[name = tensor("op_56283_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56283_cast_fp16 = slice_by_index(begin = var_56283_begin_0, end = var_56283_end_0, end_mask = var_56283_end_mask_0, x = q_249_cast_fp16)[name = tensor("op_56283_cast_fp16")]; + tensor var_56287_begin_0 = const()[name = tensor("op_56287_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_56287_end_0 = const()[name = tensor("op_56287_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_56287_end_mask_0 = const()[name = tensor("op_56287_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56287_cast_fp16 = slice_by_index(begin = var_56287_begin_0, end = var_56287_end_0, end_mask = var_56287_end_mask_0, x = q_249_cast_fp16)[name = tensor("op_56287_cast_fp16")]; + tensor var_56291_begin_0 = const()[name = tensor("op_56291_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_56291_end_0 = const()[name = tensor("op_56291_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_56291_end_mask_0 = const()[name = tensor("op_56291_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56291_cast_fp16 = slice_by_index(begin = var_56291_begin_0, end = var_56291_end_0, end_mask = var_56291_end_mask_0, x = q_249_cast_fp16)[name = tensor("op_56291_cast_fp16")]; + tensor k_499_perm_0 = const()[name = tensor("k_499_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_56298_begin_0 = const()[name = tensor("op_56298_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_56298_end_0 = const()[name = tensor("op_56298_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_56298_end_mask_0 = const()[name = tensor("op_56298_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_499_cast_fp16 = transpose(perm = k_499_perm_0, x = k_497_cast_fp16)[name = tensor("transpose_15")]; + tensor var_56298_cast_fp16 = slice_by_index(begin = var_56298_begin_0, end = var_56298_end_0, end_mask = var_56298_end_mask_0, x = k_499_cast_fp16)[name = tensor("op_56298_cast_fp16")]; + tensor var_56302_begin_0 = const()[name = tensor("op_56302_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_56302_end_0 = const()[name = tensor("op_56302_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_56302_end_mask_0 = const()[name = tensor("op_56302_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_56302_cast_fp16 = slice_by_index(begin = var_56302_begin_0, end = var_56302_end_0, end_mask = var_56302_end_mask_0, x = k_499_cast_fp16)[name = tensor("op_56302_cast_fp16")]; + tensor var_56306_begin_0 = const()[name = tensor("op_56306_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_56306_end_0 = const()[name = tensor("op_56306_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_56306_end_mask_0 = const()[name = tensor("op_56306_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_56306_cast_fp16 = slice_by_index(begin = var_56306_begin_0, end = var_56306_end_0, end_mask = var_56306_end_mask_0, x = k_499_cast_fp16)[name = tensor("op_56306_cast_fp16")]; + tensor var_56310_begin_0 = const()[name = tensor("op_56310_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_56310_end_0 = const()[name = tensor("op_56310_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_56310_end_mask_0 = const()[name = tensor("op_56310_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_56310_cast_fp16 = slice_by_index(begin = var_56310_begin_0, end = var_56310_end_0, end_mask = var_56310_end_mask_0, x = k_499_cast_fp16)[name = tensor("op_56310_cast_fp16")]; + tensor var_56314_begin_0 = const()[name = tensor("op_56314_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_56314_end_0 = const()[name = tensor("op_56314_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_56314_end_mask_0 = const()[name = tensor("op_56314_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_56314_cast_fp16 = slice_by_index(begin = var_56314_begin_0, end = var_56314_end_0, end_mask = var_56314_end_mask_0, x = k_499_cast_fp16)[name = tensor("op_56314_cast_fp16")]; + tensor var_56318_begin_0 = const()[name = tensor("op_56318_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_56318_end_0 = const()[name = tensor("op_56318_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_56318_end_mask_0 = const()[name = tensor("op_56318_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_56318_cast_fp16 = slice_by_index(begin = var_56318_begin_0, end = var_56318_end_0, end_mask = var_56318_end_mask_0, x = k_499_cast_fp16)[name = tensor("op_56318_cast_fp16")]; + tensor var_56322_begin_0 = const()[name = tensor("op_56322_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_56322_end_0 = const()[name = tensor("op_56322_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_56322_end_mask_0 = const()[name = tensor("op_56322_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_56322_cast_fp16 = slice_by_index(begin = var_56322_begin_0, end = var_56322_end_0, end_mask = var_56322_end_mask_0, x = k_499_cast_fp16)[name = tensor("op_56322_cast_fp16")]; + tensor var_56326_begin_0 = const()[name = tensor("op_56326_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_56326_end_0 = const()[name = tensor("op_56326_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_56326_end_mask_0 = const()[name = tensor("op_56326_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_56326_cast_fp16 = slice_by_index(begin = var_56326_begin_0, end = var_56326_end_0, end_mask = var_56326_end_mask_0, x = k_499_cast_fp16)[name = tensor("op_56326_cast_fp16")]; + tensor var_56330_begin_0 = const()[name = tensor("op_56330_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_56330_end_0 = const()[name = tensor("op_56330_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_56330_end_mask_0 = const()[name = tensor("op_56330_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_56330_cast_fp16 = slice_by_index(begin = var_56330_begin_0, end = var_56330_end_0, end_mask = var_56330_end_mask_0, x = k_499_cast_fp16)[name = tensor("op_56330_cast_fp16")]; + tensor var_56334_begin_0 = const()[name = tensor("op_56334_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_56334_end_0 = const()[name = tensor("op_56334_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_56334_end_mask_0 = const()[name = tensor("op_56334_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_56334_cast_fp16 = slice_by_index(begin = var_56334_begin_0, end = var_56334_end_0, end_mask = var_56334_end_mask_0, x = k_499_cast_fp16)[name = tensor("op_56334_cast_fp16")]; + tensor var_56338_begin_0 = const()[name = tensor("op_56338_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_56338_end_0 = const()[name = tensor("op_56338_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_56338_end_mask_0 = const()[name = tensor("op_56338_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_56338_cast_fp16 = slice_by_index(begin = var_56338_begin_0, end = var_56338_end_0, end_mask = var_56338_end_mask_0, x = k_499_cast_fp16)[name = tensor("op_56338_cast_fp16")]; + tensor var_56342_begin_0 = const()[name = tensor("op_56342_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_56342_end_0 = const()[name = tensor("op_56342_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_56342_end_mask_0 = const()[name = tensor("op_56342_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_56342_cast_fp16 = slice_by_index(begin = var_56342_begin_0, end = var_56342_end_0, end_mask = var_56342_end_mask_0, x = k_499_cast_fp16)[name = tensor("op_56342_cast_fp16")]; + tensor var_56346_begin_0 = const()[name = tensor("op_56346_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_56346_end_0 = const()[name = tensor("op_56346_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_56346_end_mask_0 = const()[name = tensor("op_56346_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_56346_cast_fp16 = slice_by_index(begin = var_56346_begin_0, end = var_56346_end_0, end_mask = var_56346_end_mask_0, x = k_499_cast_fp16)[name = tensor("op_56346_cast_fp16")]; + tensor var_56350_begin_0 = const()[name = tensor("op_56350_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_56350_end_0 = const()[name = tensor("op_56350_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_56350_end_mask_0 = const()[name = tensor("op_56350_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_56350_cast_fp16 = slice_by_index(begin = var_56350_begin_0, end = var_56350_end_0, end_mask = var_56350_end_mask_0, x = k_499_cast_fp16)[name = tensor("op_56350_cast_fp16")]; + tensor var_56354_begin_0 = const()[name = tensor("op_56354_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_56354_end_0 = const()[name = tensor("op_56354_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_56354_end_mask_0 = const()[name = tensor("op_56354_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_56354_cast_fp16 = slice_by_index(begin = var_56354_begin_0, end = var_56354_end_0, end_mask = var_56354_end_mask_0, x = k_499_cast_fp16)[name = tensor("op_56354_cast_fp16")]; + tensor var_56358_begin_0 = const()[name = tensor("op_56358_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_56358_end_0 = const()[name = tensor("op_56358_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_56358_end_mask_0 = const()[name = tensor("op_56358_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_56358_cast_fp16 = slice_by_index(begin = var_56358_begin_0, end = var_56358_end_0, end_mask = var_56358_end_mask_0, x = k_499_cast_fp16)[name = tensor("op_56358_cast_fp16")]; + tensor var_56362_begin_0 = const()[name = tensor("op_56362_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_56362_end_0 = const()[name = tensor("op_56362_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_56362_end_mask_0 = const()[name = tensor("op_56362_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_56362_cast_fp16 = slice_by_index(begin = var_56362_begin_0, end = var_56362_end_0, end_mask = var_56362_end_mask_0, x = k_499_cast_fp16)[name = tensor("op_56362_cast_fp16")]; + tensor var_56366_begin_0 = const()[name = tensor("op_56366_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_56366_end_0 = const()[name = tensor("op_56366_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_56366_end_mask_0 = const()[name = tensor("op_56366_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_56366_cast_fp16 = slice_by_index(begin = var_56366_begin_0, end = var_56366_end_0, end_mask = var_56366_end_mask_0, x = k_499_cast_fp16)[name = tensor("op_56366_cast_fp16")]; + tensor var_56370_begin_0 = const()[name = tensor("op_56370_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_56370_end_0 = const()[name = tensor("op_56370_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_56370_end_mask_0 = const()[name = tensor("op_56370_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_56370_cast_fp16 = slice_by_index(begin = var_56370_begin_0, end = var_56370_end_0, end_mask = var_56370_end_mask_0, x = k_499_cast_fp16)[name = tensor("op_56370_cast_fp16")]; + tensor var_56374_begin_0 = const()[name = tensor("op_56374_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_56374_end_0 = const()[name = tensor("op_56374_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_56374_end_mask_0 = const()[name = tensor("op_56374_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_56374_cast_fp16 = slice_by_index(begin = var_56374_begin_0, end = var_56374_end_0, end_mask = var_56374_end_mask_0, x = k_499_cast_fp16)[name = tensor("op_56374_cast_fp16")]; + tensor var_56376_begin_0 = const()[name = tensor("op_56376_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_56376_end_0 = const()[name = tensor("op_56376_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_56376_end_mask_0 = const()[name = tensor("op_56376_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56376_cast_fp16 = slice_by_index(begin = var_56376_begin_0, end = var_56376_end_0, end_mask = var_56376_end_mask_0, x = v_249_cast_fp16)[name = tensor("op_56376_cast_fp16")]; + tensor var_56380_begin_0 = const()[name = tensor("op_56380_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_56380_end_0 = const()[name = tensor("op_56380_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_56380_end_mask_0 = const()[name = tensor("op_56380_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56380_cast_fp16 = slice_by_index(begin = var_56380_begin_0, end = var_56380_end_0, end_mask = var_56380_end_mask_0, x = v_249_cast_fp16)[name = tensor("op_56380_cast_fp16")]; + tensor var_56384_begin_0 = const()[name = tensor("op_56384_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_56384_end_0 = const()[name = tensor("op_56384_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_56384_end_mask_0 = const()[name = tensor("op_56384_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56384_cast_fp16 = slice_by_index(begin = var_56384_begin_0, end = var_56384_end_0, end_mask = var_56384_end_mask_0, x = v_249_cast_fp16)[name = tensor("op_56384_cast_fp16")]; + tensor var_56388_begin_0 = const()[name = tensor("op_56388_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_56388_end_0 = const()[name = tensor("op_56388_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_56388_end_mask_0 = const()[name = tensor("op_56388_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56388_cast_fp16 = slice_by_index(begin = var_56388_begin_0, end = var_56388_end_0, end_mask = var_56388_end_mask_0, x = v_249_cast_fp16)[name = tensor("op_56388_cast_fp16")]; + tensor var_56392_begin_0 = const()[name = tensor("op_56392_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_56392_end_0 = const()[name = tensor("op_56392_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_56392_end_mask_0 = const()[name = tensor("op_56392_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56392_cast_fp16 = slice_by_index(begin = var_56392_begin_0, end = var_56392_end_0, end_mask = var_56392_end_mask_0, x = v_249_cast_fp16)[name = tensor("op_56392_cast_fp16")]; + tensor var_56396_begin_0 = const()[name = tensor("op_56396_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_56396_end_0 = const()[name = tensor("op_56396_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_56396_end_mask_0 = const()[name = tensor("op_56396_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56396_cast_fp16 = slice_by_index(begin = var_56396_begin_0, end = var_56396_end_0, end_mask = var_56396_end_mask_0, x = v_249_cast_fp16)[name = tensor("op_56396_cast_fp16")]; + tensor var_56400_begin_0 = const()[name = tensor("op_56400_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_56400_end_0 = const()[name = tensor("op_56400_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_56400_end_mask_0 = const()[name = tensor("op_56400_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56400_cast_fp16 = slice_by_index(begin = var_56400_begin_0, end = var_56400_end_0, end_mask = var_56400_end_mask_0, x = v_249_cast_fp16)[name = tensor("op_56400_cast_fp16")]; + tensor var_56404_begin_0 = const()[name = tensor("op_56404_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_56404_end_0 = const()[name = tensor("op_56404_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_56404_end_mask_0 = const()[name = tensor("op_56404_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56404_cast_fp16 = slice_by_index(begin = var_56404_begin_0, end = var_56404_end_0, end_mask = var_56404_end_mask_0, x = v_249_cast_fp16)[name = tensor("op_56404_cast_fp16")]; + tensor var_56408_begin_0 = const()[name = tensor("op_56408_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_56408_end_0 = const()[name = tensor("op_56408_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_56408_end_mask_0 = const()[name = tensor("op_56408_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56408_cast_fp16 = slice_by_index(begin = var_56408_begin_0, end = var_56408_end_0, end_mask = var_56408_end_mask_0, x = v_249_cast_fp16)[name = tensor("op_56408_cast_fp16")]; + tensor var_56412_begin_0 = const()[name = tensor("op_56412_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_56412_end_0 = const()[name = tensor("op_56412_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_56412_end_mask_0 = const()[name = tensor("op_56412_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56412_cast_fp16 = slice_by_index(begin = var_56412_begin_0, end = var_56412_end_0, end_mask = var_56412_end_mask_0, x = v_249_cast_fp16)[name = tensor("op_56412_cast_fp16")]; + tensor var_56416_begin_0 = const()[name = tensor("op_56416_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_56416_end_0 = const()[name = tensor("op_56416_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_56416_end_mask_0 = const()[name = tensor("op_56416_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56416_cast_fp16 = slice_by_index(begin = var_56416_begin_0, end = var_56416_end_0, end_mask = var_56416_end_mask_0, x = v_249_cast_fp16)[name = tensor("op_56416_cast_fp16")]; + tensor var_56420_begin_0 = const()[name = tensor("op_56420_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_56420_end_0 = const()[name = tensor("op_56420_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_56420_end_mask_0 = const()[name = tensor("op_56420_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56420_cast_fp16 = slice_by_index(begin = var_56420_begin_0, end = var_56420_end_0, end_mask = var_56420_end_mask_0, x = v_249_cast_fp16)[name = tensor("op_56420_cast_fp16")]; + tensor var_56424_begin_0 = const()[name = tensor("op_56424_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_56424_end_0 = const()[name = tensor("op_56424_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_56424_end_mask_0 = const()[name = tensor("op_56424_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56424_cast_fp16 = slice_by_index(begin = var_56424_begin_0, end = var_56424_end_0, end_mask = var_56424_end_mask_0, x = v_249_cast_fp16)[name = tensor("op_56424_cast_fp16")]; + tensor var_56428_begin_0 = const()[name = tensor("op_56428_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_56428_end_0 = const()[name = tensor("op_56428_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_56428_end_mask_0 = const()[name = tensor("op_56428_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56428_cast_fp16 = slice_by_index(begin = var_56428_begin_0, end = var_56428_end_0, end_mask = var_56428_end_mask_0, x = v_249_cast_fp16)[name = tensor("op_56428_cast_fp16")]; + tensor var_56432_begin_0 = const()[name = tensor("op_56432_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_56432_end_0 = const()[name = tensor("op_56432_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_56432_end_mask_0 = const()[name = tensor("op_56432_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56432_cast_fp16 = slice_by_index(begin = var_56432_begin_0, end = var_56432_end_0, end_mask = var_56432_end_mask_0, x = v_249_cast_fp16)[name = tensor("op_56432_cast_fp16")]; + tensor var_56436_begin_0 = const()[name = tensor("op_56436_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_56436_end_0 = const()[name = tensor("op_56436_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_56436_end_mask_0 = const()[name = tensor("op_56436_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56436_cast_fp16 = slice_by_index(begin = var_56436_begin_0, end = var_56436_end_0, end_mask = var_56436_end_mask_0, x = v_249_cast_fp16)[name = tensor("op_56436_cast_fp16")]; + tensor var_56440_begin_0 = const()[name = tensor("op_56440_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_56440_end_0 = const()[name = tensor("op_56440_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_56440_end_mask_0 = const()[name = tensor("op_56440_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56440_cast_fp16 = slice_by_index(begin = var_56440_begin_0, end = var_56440_end_0, end_mask = var_56440_end_mask_0, x = v_249_cast_fp16)[name = tensor("op_56440_cast_fp16")]; + tensor var_56444_begin_0 = const()[name = tensor("op_56444_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_56444_end_0 = const()[name = tensor("op_56444_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_56444_end_mask_0 = const()[name = tensor("op_56444_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56444_cast_fp16 = slice_by_index(begin = var_56444_begin_0, end = var_56444_end_0, end_mask = var_56444_end_mask_0, x = v_249_cast_fp16)[name = tensor("op_56444_cast_fp16")]; + tensor var_56448_begin_0 = const()[name = tensor("op_56448_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_56448_end_0 = const()[name = tensor("op_56448_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_56448_end_mask_0 = const()[name = tensor("op_56448_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56448_cast_fp16 = slice_by_index(begin = var_56448_begin_0, end = var_56448_end_0, end_mask = var_56448_end_mask_0, x = v_249_cast_fp16)[name = tensor("op_56448_cast_fp16")]; + tensor var_56452_begin_0 = const()[name = tensor("op_56452_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_56452_end_0 = const()[name = tensor("op_56452_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_56452_end_mask_0 = const()[name = tensor("op_56452_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56452_cast_fp16 = slice_by_index(begin = var_56452_begin_0, end = var_56452_end_0, end_mask = var_56452_end_mask_0, x = v_249_cast_fp16)[name = tensor("op_56452_cast_fp16")]; + tensor var_56456_equation_0 = const()[name = tensor("op_56456_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_56456_cast_fp16 = einsum(equation = var_56456_equation_0, values = (var_56298_cast_fp16, var_56215_cast_fp16))[name = tensor("op_56456_cast_fp16")]; + tensor var_56457_to_fp16 = const()[name = tensor("op_56457_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4801_cast_fp16 = mul(x = var_56456_cast_fp16, y = var_56457_to_fp16)[name = tensor("aw_4801_cast_fp16")]; + tensor var_56460_equation_0 = const()[name = tensor("op_56460_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_56460_cast_fp16 = einsum(equation = var_56460_equation_0, values = (var_56302_cast_fp16, var_56219_cast_fp16))[name = tensor("op_56460_cast_fp16")]; + tensor var_56461_to_fp16 = const()[name = tensor("op_56461_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4803_cast_fp16 = mul(x = var_56460_cast_fp16, y = var_56461_to_fp16)[name = tensor("aw_4803_cast_fp16")]; + tensor var_56464_equation_0 = const()[name = tensor("op_56464_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_56464_cast_fp16 = einsum(equation = var_56464_equation_0, values = (var_56306_cast_fp16, var_56223_cast_fp16))[name = tensor("op_56464_cast_fp16")]; + tensor var_56465_to_fp16 = const()[name = tensor("op_56465_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4805_cast_fp16 = mul(x = var_56464_cast_fp16, y = var_56465_to_fp16)[name = tensor("aw_4805_cast_fp16")]; + tensor var_56468_equation_0 = const()[name = tensor("op_56468_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_56468_cast_fp16 = einsum(equation = var_56468_equation_0, values = (var_56310_cast_fp16, var_56227_cast_fp16))[name = tensor("op_56468_cast_fp16")]; + tensor var_56469_to_fp16 = const()[name = tensor("op_56469_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4807_cast_fp16 = mul(x = var_56468_cast_fp16, y = var_56469_to_fp16)[name = tensor("aw_4807_cast_fp16")]; + tensor var_56472_equation_0 = const()[name = tensor("op_56472_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_56472_cast_fp16 = einsum(equation = var_56472_equation_0, values = (var_56314_cast_fp16, var_56231_cast_fp16))[name = tensor("op_56472_cast_fp16")]; + tensor var_56473_to_fp16 = const()[name = tensor("op_56473_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4809_cast_fp16 = mul(x = var_56472_cast_fp16, y = var_56473_to_fp16)[name = tensor("aw_4809_cast_fp16")]; + tensor var_56476_equation_0 = const()[name = tensor("op_56476_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_56476_cast_fp16 = einsum(equation = var_56476_equation_0, values = (var_56318_cast_fp16, var_56235_cast_fp16))[name = tensor("op_56476_cast_fp16")]; + tensor var_56477_to_fp16 = const()[name = tensor("op_56477_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4811_cast_fp16 = mul(x = var_56476_cast_fp16, y = var_56477_to_fp16)[name = tensor("aw_4811_cast_fp16")]; + tensor var_56480_equation_0 = const()[name = tensor("op_56480_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_56480_cast_fp16 = einsum(equation = var_56480_equation_0, values = (var_56322_cast_fp16, var_56239_cast_fp16))[name = tensor("op_56480_cast_fp16")]; + tensor var_56481_to_fp16 = const()[name = tensor("op_56481_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4813_cast_fp16 = mul(x = var_56480_cast_fp16, y = var_56481_to_fp16)[name = tensor("aw_4813_cast_fp16")]; + tensor var_56484_equation_0 = const()[name = tensor("op_56484_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_56484_cast_fp16 = einsum(equation = var_56484_equation_0, values = (var_56326_cast_fp16, var_56243_cast_fp16))[name = tensor("op_56484_cast_fp16")]; + tensor var_56485_to_fp16 = const()[name = tensor("op_56485_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4815_cast_fp16 = mul(x = var_56484_cast_fp16, y = var_56485_to_fp16)[name = tensor("aw_4815_cast_fp16")]; + tensor var_56488_equation_0 = const()[name = tensor("op_56488_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_56488_cast_fp16 = einsum(equation = var_56488_equation_0, values = (var_56330_cast_fp16, var_56247_cast_fp16))[name = tensor("op_56488_cast_fp16")]; + tensor var_56489_to_fp16 = const()[name = tensor("op_56489_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4817_cast_fp16 = mul(x = var_56488_cast_fp16, y = var_56489_to_fp16)[name = tensor("aw_4817_cast_fp16")]; + tensor var_56492_equation_0 = const()[name = tensor("op_56492_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_56492_cast_fp16 = einsum(equation = var_56492_equation_0, values = (var_56334_cast_fp16, var_56251_cast_fp16))[name = tensor("op_56492_cast_fp16")]; + tensor var_56493_to_fp16 = const()[name = tensor("op_56493_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4819_cast_fp16 = mul(x = var_56492_cast_fp16, y = var_56493_to_fp16)[name = tensor("aw_4819_cast_fp16")]; + tensor var_56496_equation_0 = const()[name = tensor("op_56496_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_56496_cast_fp16 = einsum(equation = var_56496_equation_0, values = (var_56338_cast_fp16, var_56255_cast_fp16))[name = tensor("op_56496_cast_fp16")]; + tensor var_56497_to_fp16 = const()[name = tensor("op_56497_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4821_cast_fp16 = mul(x = var_56496_cast_fp16, y = var_56497_to_fp16)[name = tensor("aw_4821_cast_fp16")]; + tensor var_56500_equation_0 = const()[name = tensor("op_56500_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_56500_cast_fp16 = einsum(equation = var_56500_equation_0, values = (var_56342_cast_fp16, var_56259_cast_fp16))[name = tensor("op_56500_cast_fp16")]; + tensor var_56501_to_fp16 = const()[name = tensor("op_56501_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4823_cast_fp16 = mul(x = var_56500_cast_fp16, y = var_56501_to_fp16)[name = tensor("aw_4823_cast_fp16")]; + tensor var_56504_equation_0 = const()[name = tensor("op_56504_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_56504_cast_fp16 = einsum(equation = var_56504_equation_0, values = (var_56346_cast_fp16, var_56263_cast_fp16))[name = tensor("op_56504_cast_fp16")]; + tensor var_56505_to_fp16 = const()[name = tensor("op_56505_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4825_cast_fp16 = mul(x = var_56504_cast_fp16, y = var_56505_to_fp16)[name = tensor("aw_4825_cast_fp16")]; + tensor var_56508_equation_0 = const()[name = tensor("op_56508_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_56508_cast_fp16 = einsum(equation = var_56508_equation_0, values = (var_56350_cast_fp16, var_56267_cast_fp16))[name = tensor("op_56508_cast_fp16")]; + tensor var_56509_to_fp16 = const()[name = tensor("op_56509_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4827_cast_fp16 = mul(x = var_56508_cast_fp16, y = var_56509_to_fp16)[name = tensor("aw_4827_cast_fp16")]; + tensor var_56512_equation_0 = const()[name = tensor("op_56512_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_56512_cast_fp16 = einsum(equation = var_56512_equation_0, values = (var_56354_cast_fp16, var_56271_cast_fp16))[name = tensor("op_56512_cast_fp16")]; + tensor var_56513_to_fp16 = const()[name = tensor("op_56513_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4829_cast_fp16 = mul(x = var_56512_cast_fp16, y = var_56513_to_fp16)[name = tensor("aw_4829_cast_fp16")]; + tensor var_56516_equation_0 = const()[name = tensor("op_56516_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_56516_cast_fp16 = einsum(equation = var_56516_equation_0, values = (var_56358_cast_fp16, var_56275_cast_fp16))[name = tensor("op_56516_cast_fp16")]; + tensor var_56517_to_fp16 = const()[name = tensor("op_56517_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4831_cast_fp16 = mul(x = var_56516_cast_fp16, y = var_56517_to_fp16)[name = tensor("aw_4831_cast_fp16")]; + tensor var_56520_equation_0 = const()[name = tensor("op_56520_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_56520_cast_fp16 = einsum(equation = var_56520_equation_0, values = (var_56362_cast_fp16, var_56279_cast_fp16))[name = tensor("op_56520_cast_fp16")]; + tensor var_56521_to_fp16 = const()[name = tensor("op_56521_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4833_cast_fp16 = mul(x = var_56520_cast_fp16, y = var_56521_to_fp16)[name = tensor("aw_4833_cast_fp16")]; + tensor var_56524_equation_0 = const()[name = tensor("op_56524_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_56524_cast_fp16 = einsum(equation = var_56524_equation_0, values = (var_56366_cast_fp16, var_56283_cast_fp16))[name = tensor("op_56524_cast_fp16")]; + tensor var_56525_to_fp16 = const()[name = tensor("op_56525_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4835_cast_fp16 = mul(x = var_56524_cast_fp16, y = var_56525_to_fp16)[name = tensor("aw_4835_cast_fp16")]; + tensor var_56528_equation_0 = const()[name = tensor("op_56528_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_56528_cast_fp16 = einsum(equation = var_56528_equation_0, values = (var_56370_cast_fp16, var_56287_cast_fp16))[name = tensor("op_56528_cast_fp16")]; + tensor var_56529_to_fp16 = const()[name = tensor("op_56529_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4837_cast_fp16 = mul(x = var_56528_cast_fp16, y = var_56529_to_fp16)[name = tensor("aw_4837_cast_fp16")]; + tensor var_56532_equation_0 = const()[name = tensor("op_56532_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_56532_cast_fp16 = einsum(equation = var_56532_equation_0, values = (var_56374_cast_fp16, var_56291_cast_fp16))[name = tensor("op_56532_cast_fp16")]; + tensor var_56533_to_fp16 = const()[name = tensor("op_56533_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4839_cast_fp16 = mul(x = var_56532_cast_fp16, y = var_56533_to_fp16)[name = tensor("aw_4839_cast_fp16")]; + tensor var_56535_cast_fp16 = softmax(axis = var_30355, x = aw_4801_cast_fp16)[name = tensor("op_56535_cast_fp16")]; + tensor var_56536_cast_fp16 = softmax(axis = var_30355, x = aw_4803_cast_fp16)[name = tensor("op_56536_cast_fp16")]; + tensor var_56537_cast_fp16 = softmax(axis = var_30355, x = aw_4805_cast_fp16)[name = tensor("op_56537_cast_fp16")]; + tensor var_56538_cast_fp16 = softmax(axis = var_30355, x = aw_4807_cast_fp16)[name = tensor("op_56538_cast_fp16")]; + tensor var_56539_cast_fp16 = softmax(axis = var_30355, x = aw_4809_cast_fp16)[name = tensor("op_56539_cast_fp16")]; + tensor var_56540_cast_fp16 = softmax(axis = var_30355, x = aw_4811_cast_fp16)[name = tensor("op_56540_cast_fp16")]; + tensor var_56541_cast_fp16 = softmax(axis = var_30355, x = aw_4813_cast_fp16)[name = tensor("op_56541_cast_fp16")]; + tensor var_56542_cast_fp16 = softmax(axis = var_30355, x = aw_4815_cast_fp16)[name = tensor("op_56542_cast_fp16")]; + tensor var_56543_cast_fp16 = softmax(axis = var_30355, x = aw_4817_cast_fp16)[name = tensor("op_56543_cast_fp16")]; + tensor var_56544_cast_fp16 = softmax(axis = var_30355, x = aw_4819_cast_fp16)[name = tensor("op_56544_cast_fp16")]; + tensor var_56545_cast_fp16 = softmax(axis = var_30355, x = aw_4821_cast_fp16)[name = tensor("op_56545_cast_fp16")]; + tensor var_56546_cast_fp16 = softmax(axis = var_30355, x = aw_4823_cast_fp16)[name = tensor("op_56546_cast_fp16")]; + tensor var_56547_cast_fp16 = softmax(axis = var_30355, x = aw_4825_cast_fp16)[name = tensor("op_56547_cast_fp16")]; + tensor var_56548_cast_fp16 = softmax(axis = var_30355, x = aw_4827_cast_fp16)[name = tensor("op_56548_cast_fp16")]; + tensor var_56549_cast_fp16 = softmax(axis = var_30355, x = aw_4829_cast_fp16)[name = tensor("op_56549_cast_fp16")]; + tensor var_56550_cast_fp16 = softmax(axis = var_30355, x = aw_4831_cast_fp16)[name = tensor("op_56550_cast_fp16")]; + tensor var_56551_cast_fp16 = softmax(axis = var_30355, x = aw_4833_cast_fp16)[name = tensor("op_56551_cast_fp16")]; + tensor var_56552_cast_fp16 = softmax(axis = var_30355, x = aw_4835_cast_fp16)[name = tensor("op_56552_cast_fp16")]; + tensor var_56553_cast_fp16 = softmax(axis = var_30355, x = aw_4837_cast_fp16)[name = tensor("op_56553_cast_fp16")]; + tensor var_56554_cast_fp16 = softmax(axis = var_30355, x = aw_4839_cast_fp16)[name = tensor("op_56554_cast_fp16")]; + tensor var_56556_equation_0 = const()[name = tensor("op_56556_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_56556_cast_fp16 = einsum(equation = var_56556_equation_0, values = (var_56376_cast_fp16, var_56535_cast_fp16))[name = tensor("op_56556_cast_fp16")]; + tensor var_56558_equation_0 = const()[name = tensor("op_56558_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_56558_cast_fp16 = einsum(equation = var_56558_equation_0, values = (var_56380_cast_fp16, var_56536_cast_fp16))[name = tensor("op_56558_cast_fp16")]; + tensor var_56560_equation_0 = const()[name = tensor("op_56560_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_56560_cast_fp16 = einsum(equation = var_56560_equation_0, values = (var_56384_cast_fp16, var_56537_cast_fp16))[name = tensor("op_56560_cast_fp16")]; + tensor var_56562_equation_0 = const()[name = tensor("op_56562_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_56562_cast_fp16 = einsum(equation = var_56562_equation_0, values = (var_56388_cast_fp16, var_56538_cast_fp16))[name = tensor("op_56562_cast_fp16")]; + tensor var_56564_equation_0 = const()[name = tensor("op_56564_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_56564_cast_fp16 = einsum(equation = var_56564_equation_0, values = (var_56392_cast_fp16, var_56539_cast_fp16))[name = tensor("op_56564_cast_fp16")]; + tensor var_56566_equation_0 = const()[name = tensor("op_56566_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_56566_cast_fp16 = einsum(equation = var_56566_equation_0, values = (var_56396_cast_fp16, var_56540_cast_fp16))[name = tensor("op_56566_cast_fp16")]; + tensor var_56568_equation_0 = const()[name = tensor("op_56568_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_56568_cast_fp16 = einsum(equation = var_56568_equation_0, values = (var_56400_cast_fp16, var_56541_cast_fp16))[name = tensor("op_56568_cast_fp16")]; + tensor var_56570_equation_0 = const()[name = tensor("op_56570_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_56570_cast_fp16 = einsum(equation = var_56570_equation_0, values = (var_56404_cast_fp16, var_56542_cast_fp16))[name = tensor("op_56570_cast_fp16")]; + tensor var_56572_equation_0 = const()[name = tensor("op_56572_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_56572_cast_fp16 = einsum(equation = var_56572_equation_0, values = (var_56408_cast_fp16, var_56543_cast_fp16))[name = tensor("op_56572_cast_fp16")]; + tensor var_56574_equation_0 = const()[name = tensor("op_56574_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_56574_cast_fp16 = einsum(equation = var_56574_equation_0, values = (var_56412_cast_fp16, var_56544_cast_fp16))[name = tensor("op_56574_cast_fp16")]; + tensor var_56576_equation_0 = const()[name = tensor("op_56576_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_56576_cast_fp16 = einsum(equation = var_56576_equation_0, values = (var_56416_cast_fp16, var_56545_cast_fp16))[name = tensor("op_56576_cast_fp16")]; + tensor var_56578_equation_0 = const()[name = tensor("op_56578_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_56578_cast_fp16 = einsum(equation = var_56578_equation_0, values = (var_56420_cast_fp16, var_56546_cast_fp16))[name = tensor("op_56578_cast_fp16")]; + tensor var_56580_equation_0 = const()[name = tensor("op_56580_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_56580_cast_fp16 = einsum(equation = var_56580_equation_0, values = (var_56424_cast_fp16, var_56547_cast_fp16))[name = tensor("op_56580_cast_fp16")]; + tensor var_56582_equation_0 = const()[name = tensor("op_56582_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_56582_cast_fp16 = einsum(equation = var_56582_equation_0, values = (var_56428_cast_fp16, var_56548_cast_fp16))[name = tensor("op_56582_cast_fp16")]; + tensor var_56584_equation_0 = const()[name = tensor("op_56584_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_56584_cast_fp16 = einsum(equation = var_56584_equation_0, values = (var_56432_cast_fp16, var_56549_cast_fp16))[name = tensor("op_56584_cast_fp16")]; + tensor var_56586_equation_0 = const()[name = tensor("op_56586_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_56586_cast_fp16 = einsum(equation = var_56586_equation_0, values = (var_56436_cast_fp16, var_56550_cast_fp16))[name = tensor("op_56586_cast_fp16")]; + tensor var_56588_equation_0 = const()[name = tensor("op_56588_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_56588_cast_fp16 = einsum(equation = var_56588_equation_0, values = (var_56440_cast_fp16, var_56551_cast_fp16))[name = tensor("op_56588_cast_fp16")]; + tensor var_56590_equation_0 = const()[name = tensor("op_56590_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_56590_cast_fp16 = einsum(equation = var_56590_equation_0, values = (var_56444_cast_fp16, var_56552_cast_fp16))[name = tensor("op_56590_cast_fp16")]; + tensor var_56592_equation_0 = const()[name = tensor("op_56592_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_56592_cast_fp16 = einsum(equation = var_56592_equation_0, values = (var_56448_cast_fp16, var_56553_cast_fp16))[name = tensor("op_56592_cast_fp16")]; + tensor var_56594_equation_0 = const()[name = tensor("op_56594_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_56594_cast_fp16 = einsum(equation = var_56594_equation_0, values = (var_56452_cast_fp16, var_56554_cast_fp16))[name = tensor("op_56594_cast_fp16")]; + tensor input_699_interleave_0 = const()[name = tensor("input_699_interleave_0"), val = tensor(false)]; + tensor input_699_cast_fp16 = concat(axis = var_30355, interleave = input_699_interleave_0, values = (var_56556_cast_fp16, var_56558_cast_fp16, var_56560_cast_fp16, var_56562_cast_fp16, var_56564_cast_fp16, var_56566_cast_fp16, var_56568_cast_fp16, var_56570_cast_fp16, var_56572_cast_fp16, var_56574_cast_fp16, var_56576_cast_fp16, var_56578_cast_fp16, var_56580_cast_fp16, var_56582_cast_fp16, var_56584_cast_fp16, var_56586_cast_fp16, var_56588_cast_fp16, var_56590_cast_fp16, var_56592_cast_fp16, var_56594_cast_fp16))[name = tensor("input_699_cast_fp16")]; + tensor var_56600 = const()[name = tensor("op_56600"), val = tensor([1, 1])]; + tensor var_56602 = const()[name = tensor("op_56602"), val = tensor([1, 1])]; + tensor var_56604_pad_type_0 = const()[name = tensor("op_56604_pad_type_0"), val = tensor("custom")]; + tensor var_56604_pad_0 = const()[name = tensor("op_56604_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_8_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_8_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4738030528)))]; + tensor up_blocks_0_attentions_2_transformer_blocks_8_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_8_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4741307392)))]; + tensor var_56604_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_8_attn1_to_out_0_bias_to_fp16, dilations = var_56602, groups = var_30355, pad = var_56604_pad_0, pad_type = var_56604_pad_type_0, strides = var_56600, weight = up_blocks_0_attentions_2_transformer_blocks_8_attn1_to_out_0_weight_to_fp16, x = input_699_cast_fp16)[name = tensor("op_56604_cast_fp16")]; + tensor inputs_375_cast_fp16 = add(x = var_56604_cast_fp16, y = inputs_373_cast_fp16)[name = tensor("inputs_375_cast_fp16")]; + tensor hidden_states_485_axes_0 = const()[name = tensor("hidden_states_485_axes_0"), val = tensor([1])]; + tensor hidden_states_485_gamma_0_to_fp16 = const()[name = tensor("hidden_states_485_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4741310016)))]; + tensor hidden_states_485_beta_0_to_fp16 = const()[name = tensor("hidden_states_485_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4741312640)))]; + tensor var_56614_to_fp16 = const()[name = tensor("op_56614_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_485_cast_fp16 = layer_norm(axes = hidden_states_485_axes_0, beta = hidden_states_485_beta_0_to_fp16, epsilon = var_56614_to_fp16, gamma = hidden_states_485_gamma_0_to_fp16, x = inputs_375_cast_fp16)[name = tensor("hidden_states_485_cast_fp16")]; + tensor var_56629 = const()[name = tensor("op_56629"), val = tensor([1, 1])]; + tensor var_56631 = const()[name = tensor("op_56631"), val = tensor([1, 1])]; + tensor q_251_pad_type_0 = const()[name = tensor("q_251_pad_type_0"), val = tensor("custom")]; + tensor q_251_pad_0 = const()[name = tensor("q_251_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_8_attn2_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_8_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4741315264)))]; + tensor q_251_cast_fp16 = conv(dilations = var_56631, groups = var_30355, pad = q_251_pad_0, pad_type = q_251_pad_type_0, strides = var_56629, weight = up_blocks_0_attentions_2_transformer_blocks_8_attn2_to_q_weight_to_fp16, x = hidden_states_485_cast_fp16)[name = tensor("q_251_cast_fp16")]; + tensor var_56635 = const()[name = tensor("op_56635"), val = tensor([1, 1])]; + tensor var_56637 = const()[name = tensor("op_56637"), val = tensor([1, 1])]; + tensor k_501_pad_type_0 = const()[name = tensor("k_501_pad_type_0"), val = tensor("custom")]; + tensor k_501_pad_0 = const()[name = tensor("k_501_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_8_attn2_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_8_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4744592128)))]; + tensor k_501_cast_fp16 = conv(dilations = var_56637, groups = var_30355, pad = k_501_pad_0, pad_type = k_501_pad_type_0, strides = var_56635, weight = up_blocks_0_attentions_2_transformer_blocks_8_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_501_cast_fp16")]; + tensor var_56641 = const()[name = tensor("op_56641"), val = tensor([1, 1])]; + tensor var_56643 = const()[name = tensor("op_56643"), val = tensor([1, 1])]; + tensor v_251_pad_type_0 = const()[name = tensor("v_251_pad_type_0"), val = tensor("custom")]; + tensor v_251_pad_0 = const()[name = tensor("v_251_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_8_attn2_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_8_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4749835072)))]; + tensor v_251_cast_fp16 = conv(dilations = var_56643, groups = var_30355, pad = v_251_pad_0, pad_type = v_251_pad_type_0, strides = var_56641, weight = up_blocks_0_attentions_2_transformer_blocks_8_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_251_cast_fp16")]; + tensor var_56647_begin_0 = const()[name = tensor("op_56647_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_56647_end_0 = const()[name = tensor("op_56647_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_56647_end_mask_0 = const()[name = tensor("op_56647_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56647_cast_fp16 = slice_by_index(begin = var_56647_begin_0, end = var_56647_end_0, end_mask = var_56647_end_mask_0, x = q_251_cast_fp16)[name = tensor("op_56647_cast_fp16")]; + tensor var_56651_begin_0 = const()[name = tensor("op_56651_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_56651_end_0 = const()[name = tensor("op_56651_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_56651_end_mask_0 = const()[name = tensor("op_56651_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56651_cast_fp16 = slice_by_index(begin = var_56651_begin_0, end = var_56651_end_0, end_mask = var_56651_end_mask_0, x = q_251_cast_fp16)[name = tensor("op_56651_cast_fp16")]; + tensor var_56655_begin_0 = const()[name = tensor("op_56655_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_56655_end_0 = const()[name = tensor("op_56655_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_56655_end_mask_0 = const()[name = tensor("op_56655_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56655_cast_fp16 = slice_by_index(begin = var_56655_begin_0, end = var_56655_end_0, end_mask = var_56655_end_mask_0, x = q_251_cast_fp16)[name = tensor("op_56655_cast_fp16")]; + tensor var_56659_begin_0 = const()[name = tensor("op_56659_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_56659_end_0 = const()[name = tensor("op_56659_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_56659_end_mask_0 = const()[name = tensor("op_56659_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56659_cast_fp16 = slice_by_index(begin = var_56659_begin_0, end = var_56659_end_0, end_mask = var_56659_end_mask_0, x = q_251_cast_fp16)[name = tensor("op_56659_cast_fp16")]; + tensor var_56663_begin_0 = const()[name = tensor("op_56663_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_56663_end_0 = const()[name = tensor("op_56663_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_56663_end_mask_0 = const()[name = tensor("op_56663_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56663_cast_fp16 = slice_by_index(begin = var_56663_begin_0, end = var_56663_end_0, end_mask = var_56663_end_mask_0, x = q_251_cast_fp16)[name = tensor("op_56663_cast_fp16")]; + tensor var_56667_begin_0 = const()[name = tensor("op_56667_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_56667_end_0 = const()[name = tensor("op_56667_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_56667_end_mask_0 = const()[name = tensor("op_56667_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56667_cast_fp16 = slice_by_index(begin = var_56667_begin_0, end = var_56667_end_0, end_mask = var_56667_end_mask_0, x = q_251_cast_fp16)[name = tensor("op_56667_cast_fp16")]; + tensor var_56671_begin_0 = const()[name = tensor("op_56671_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_56671_end_0 = const()[name = tensor("op_56671_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_56671_end_mask_0 = const()[name = tensor("op_56671_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56671_cast_fp16 = slice_by_index(begin = var_56671_begin_0, end = var_56671_end_0, end_mask = var_56671_end_mask_0, x = q_251_cast_fp16)[name = tensor("op_56671_cast_fp16")]; + tensor var_56675_begin_0 = const()[name = tensor("op_56675_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_56675_end_0 = const()[name = tensor("op_56675_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_56675_end_mask_0 = const()[name = tensor("op_56675_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56675_cast_fp16 = slice_by_index(begin = var_56675_begin_0, end = var_56675_end_0, end_mask = var_56675_end_mask_0, x = q_251_cast_fp16)[name = tensor("op_56675_cast_fp16")]; + tensor var_56679_begin_0 = const()[name = tensor("op_56679_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_56679_end_0 = const()[name = tensor("op_56679_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_56679_end_mask_0 = const()[name = tensor("op_56679_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56679_cast_fp16 = slice_by_index(begin = var_56679_begin_0, end = var_56679_end_0, end_mask = var_56679_end_mask_0, x = q_251_cast_fp16)[name = tensor("op_56679_cast_fp16")]; + tensor var_56683_begin_0 = const()[name = tensor("op_56683_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_56683_end_0 = const()[name = tensor("op_56683_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_56683_end_mask_0 = const()[name = tensor("op_56683_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56683_cast_fp16 = slice_by_index(begin = var_56683_begin_0, end = var_56683_end_0, end_mask = var_56683_end_mask_0, x = q_251_cast_fp16)[name = tensor("op_56683_cast_fp16")]; + tensor var_56687_begin_0 = const()[name = tensor("op_56687_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_56687_end_0 = const()[name = tensor("op_56687_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_56687_end_mask_0 = const()[name = tensor("op_56687_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56687_cast_fp16 = slice_by_index(begin = var_56687_begin_0, end = var_56687_end_0, end_mask = var_56687_end_mask_0, x = q_251_cast_fp16)[name = tensor("op_56687_cast_fp16")]; + tensor var_56691_begin_0 = const()[name = tensor("op_56691_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_56691_end_0 = const()[name = tensor("op_56691_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_56691_end_mask_0 = const()[name = tensor("op_56691_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56691_cast_fp16 = slice_by_index(begin = var_56691_begin_0, end = var_56691_end_0, end_mask = var_56691_end_mask_0, x = q_251_cast_fp16)[name = tensor("op_56691_cast_fp16")]; + tensor var_56695_begin_0 = const()[name = tensor("op_56695_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_56695_end_0 = const()[name = tensor("op_56695_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_56695_end_mask_0 = const()[name = tensor("op_56695_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56695_cast_fp16 = slice_by_index(begin = var_56695_begin_0, end = var_56695_end_0, end_mask = var_56695_end_mask_0, x = q_251_cast_fp16)[name = tensor("op_56695_cast_fp16")]; + tensor var_56699_begin_0 = const()[name = tensor("op_56699_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_56699_end_0 = const()[name = tensor("op_56699_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_56699_end_mask_0 = const()[name = tensor("op_56699_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56699_cast_fp16 = slice_by_index(begin = var_56699_begin_0, end = var_56699_end_0, end_mask = var_56699_end_mask_0, x = q_251_cast_fp16)[name = tensor("op_56699_cast_fp16")]; + tensor var_56703_begin_0 = const()[name = tensor("op_56703_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_56703_end_0 = const()[name = tensor("op_56703_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_56703_end_mask_0 = const()[name = tensor("op_56703_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56703_cast_fp16 = slice_by_index(begin = var_56703_begin_0, end = var_56703_end_0, end_mask = var_56703_end_mask_0, x = q_251_cast_fp16)[name = tensor("op_56703_cast_fp16")]; + tensor var_56707_begin_0 = const()[name = tensor("op_56707_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_56707_end_0 = const()[name = tensor("op_56707_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_56707_end_mask_0 = const()[name = tensor("op_56707_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56707_cast_fp16 = slice_by_index(begin = var_56707_begin_0, end = var_56707_end_0, end_mask = var_56707_end_mask_0, x = q_251_cast_fp16)[name = tensor("op_56707_cast_fp16")]; + tensor var_56711_begin_0 = const()[name = tensor("op_56711_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_56711_end_0 = const()[name = tensor("op_56711_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_56711_end_mask_0 = const()[name = tensor("op_56711_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56711_cast_fp16 = slice_by_index(begin = var_56711_begin_0, end = var_56711_end_0, end_mask = var_56711_end_mask_0, x = q_251_cast_fp16)[name = tensor("op_56711_cast_fp16")]; + tensor var_56715_begin_0 = const()[name = tensor("op_56715_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_56715_end_0 = const()[name = tensor("op_56715_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_56715_end_mask_0 = const()[name = tensor("op_56715_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56715_cast_fp16 = slice_by_index(begin = var_56715_begin_0, end = var_56715_end_0, end_mask = var_56715_end_mask_0, x = q_251_cast_fp16)[name = tensor("op_56715_cast_fp16")]; + tensor var_56719_begin_0 = const()[name = tensor("op_56719_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_56719_end_0 = const()[name = tensor("op_56719_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_56719_end_mask_0 = const()[name = tensor("op_56719_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56719_cast_fp16 = slice_by_index(begin = var_56719_begin_0, end = var_56719_end_0, end_mask = var_56719_end_mask_0, x = q_251_cast_fp16)[name = tensor("op_56719_cast_fp16")]; + tensor var_56723_begin_0 = const()[name = tensor("op_56723_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_56723_end_0 = const()[name = tensor("op_56723_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_56723_end_mask_0 = const()[name = tensor("op_56723_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56723_cast_fp16 = slice_by_index(begin = var_56723_begin_0, end = var_56723_end_0, end_mask = var_56723_end_mask_0, x = q_251_cast_fp16)[name = tensor("op_56723_cast_fp16")]; + tensor k_503_perm_0 = const()[name = tensor("k_503_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_56730_begin_0 = const()[name = tensor("op_56730_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_56730_end_0 = const()[name = tensor("op_56730_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_56730_end_mask_0 = const()[name = tensor("op_56730_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_503_cast_fp16 = transpose(perm = k_503_perm_0, x = k_501_cast_fp16)[name = tensor("transpose_14")]; + tensor var_56730_cast_fp16 = slice_by_index(begin = var_56730_begin_0, end = var_56730_end_0, end_mask = var_56730_end_mask_0, x = k_503_cast_fp16)[name = tensor("op_56730_cast_fp16")]; + tensor var_56734_begin_0 = const()[name = tensor("op_56734_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_56734_end_0 = const()[name = tensor("op_56734_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_56734_end_mask_0 = const()[name = tensor("op_56734_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_56734_cast_fp16 = slice_by_index(begin = var_56734_begin_0, end = var_56734_end_0, end_mask = var_56734_end_mask_0, x = k_503_cast_fp16)[name = tensor("op_56734_cast_fp16")]; + tensor var_56738_begin_0 = const()[name = tensor("op_56738_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_56738_end_0 = const()[name = tensor("op_56738_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_56738_end_mask_0 = const()[name = tensor("op_56738_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_56738_cast_fp16 = slice_by_index(begin = var_56738_begin_0, end = var_56738_end_0, end_mask = var_56738_end_mask_0, x = k_503_cast_fp16)[name = tensor("op_56738_cast_fp16")]; + tensor var_56742_begin_0 = const()[name = tensor("op_56742_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_56742_end_0 = const()[name = tensor("op_56742_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_56742_end_mask_0 = const()[name = tensor("op_56742_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_56742_cast_fp16 = slice_by_index(begin = var_56742_begin_0, end = var_56742_end_0, end_mask = var_56742_end_mask_0, x = k_503_cast_fp16)[name = tensor("op_56742_cast_fp16")]; + tensor var_56746_begin_0 = const()[name = tensor("op_56746_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_56746_end_0 = const()[name = tensor("op_56746_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_56746_end_mask_0 = const()[name = tensor("op_56746_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_56746_cast_fp16 = slice_by_index(begin = var_56746_begin_0, end = var_56746_end_0, end_mask = var_56746_end_mask_0, x = k_503_cast_fp16)[name = tensor("op_56746_cast_fp16")]; + tensor var_56750_begin_0 = const()[name = tensor("op_56750_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_56750_end_0 = const()[name = tensor("op_56750_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_56750_end_mask_0 = const()[name = tensor("op_56750_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_56750_cast_fp16 = slice_by_index(begin = var_56750_begin_0, end = var_56750_end_0, end_mask = var_56750_end_mask_0, x = k_503_cast_fp16)[name = tensor("op_56750_cast_fp16")]; + tensor var_56754_begin_0 = const()[name = tensor("op_56754_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_56754_end_0 = const()[name = tensor("op_56754_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_56754_end_mask_0 = const()[name = tensor("op_56754_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_56754_cast_fp16 = slice_by_index(begin = var_56754_begin_0, end = var_56754_end_0, end_mask = var_56754_end_mask_0, x = k_503_cast_fp16)[name = tensor("op_56754_cast_fp16")]; + tensor var_56758_begin_0 = const()[name = tensor("op_56758_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_56758_end_0 = const()[name = tensor("op_56758_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_56758_end_mask_0 = const()[name = tensor("op_56758_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_56758_cast_fp16 = slice_by_index(begin = var_56758_begin_0, end = var_56758_end_0, end_mask = var_56758_end_mask_0, x = k_503_cast_fp16)[name = tensor("op_56758_cast_fp16")]; + tensor var_56762_begin_0 = const()[name = tensor("op_56762_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_56762_end_0 = const()[name = tensor("op_56762_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_56762_end_mask_0 = const()[name = tensor("op_56762_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_56762_cast_fp16 = slice_by_index(begin = var_56762_begin_0, end = var_56762_end_0, end_mask = var_56762_end_mask_0, x = k_503_cast_fp16)[name = tensor("op_56762_cast_fp16")]; + tensor var_56766_begin_0 = const()[name = tensor("op_56766_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_56766_end_0 = const()[name = tensor("op_56766_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_56766_end_mask_0 = const()[name = tensor("op_56766_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_56766_cast_fp16 = slice_by_index(begin = var_56766_begin_0, end = var_56766_end_0, end_mask = var_56766_end_mask_0, x = k_503_cast_fp16)[name = tensor("op_56766_cast_fp16")]; + tensor var_56770_begin_0 = const()[name = tensor("op_56770_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_56770_end_0 = const()[name = tensor("op_56770_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_56770_end_mask_0 = const()[name = tensor("op_56770_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_56770_cast_fp16 = slice_by_index(begin = var_56770_begin_0, end = var_56770_end_0, end_mask = var_56770_end_mask_0, x = k_503_cast_fp16)[name = tensor("op_56770_cast_fp16")]; + tensor var_56774_begin_0 = const()[name = tensor("op_56774_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_56774_end_0 = const()[name = tensor("op_56774_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_56774_end_mask_0 = const()[name = tensor("op_56774_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_56774_cast_fp16 = slice_by_index(begin = var_56774_begin_0, end = var_56774_end_0, end_mask = var_56774_end_mask_0, x = k_503_cast_fp16)[name = tensor("op_56774_cast_fp16")]; + tensor var_56778_begin_0 = const()[name = tensor("op_56778_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_56778_end_0 = const()[name = tensor("op_56778_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_56778_end_mask_0 = const()[name = tensor("op_56778_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_56778_cast_fp16 = slice_by_index(begin = var_56778_begin_0, end = var_56778_end_0, end_mask = var_56778_end_mask_0, x = k_503_cast_fp16)[name = tensor("op_56778_cast_fp16")]; + tensor var_56782_begin_0 = const()[name = tensor("op_56782_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_56782_end_0 = const()[name = tensor("op_56782_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_56782_end_mask_0 = const()[name = tensor("op_56782_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_56782_cast_fp16 = slice_by_index(begin = var_56782_begin_0, end = var_56782_end_0, end_mask = var_56782_end_mask_0, x = k_503_cast_fp16)[name = tensor("op_56782_cast_fp16")]; + tensor var_56786_begin_0 = const()[name = tensor("op_56786_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_56786_end_0 = const()[name = tensor("op_56786_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_56786_end_mask_0 = const()[name = tensor("op_56786_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_56786_cast_fp16 = slice_by_index(begin = var_56786_begin_0, end = var_56786_end_0, end_mask = var_56786_end_mask_0, x = k_503_cast_fp16)[name = tensor("op_56786_cast_fp16")]; + tensor var_56790_begin_0 = const()[name = tensor("op_56790_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_56790_end_0 = const()[name = tensor("op_56790_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_56790_end_mask_0 = const()[name = tensor("op_56790_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_56790_cast_fp16 = slice_by_index(begin = var_56790_begin_0, end = var_56790_end_0, end_mask = var_56790_end_mask_0, x = k_503_cast_fp16)[name = tensor("op_56790_cast_fp16")]; + tensor var_56794_begin_0 = const()[name = tensor("op_56794_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_56794_end_0 = const()[name = tensor("op_56794_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_56794_end_mask_0 = const()[name = tensor("op_56794_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_56794_cast_fp16 = slice_by_index(begin = var_56794_begin_0, end = var_56794_end_0, end_mask = var_56794_end_mask_0, x = k_503_cast_fp16)[name = tensor("op_56794_cast_fp16")]; + tensor var_56798_begin_0 = const()[name = tensor("op_56798_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_56798_end_0 = const()[name = tensor("op_56798_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_56798_end_mask_0 = const()[name = tensor("op_56798_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_56798_cast_fp16 = slice_by_index(begin = var_56798_begin_0, end = var_56798_end_0, end_mask = var_56798_end_mask_0, x = k_503_cast_fp16)[name = tensor("op_56798_cast_fp16")]; + tensor var_56802_begin_0 = const()[name = tensor("op_56802_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_56802_end_0 = const()[name = tensor("op_56802_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_56802_end_mask_0 = const()[name = tensor("op_56802_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_56802_cast_fp16 = slice_by_index(begin = var_56802_begin_0, end = var_56802_end_0, end_mask = var_56802_end_mask_0, x = k_503_cast_fp16)[name = tensor("op_56802_cast_fp16")]; + tensor var_56806_begin_0 = const()[name = tensor("op_56806_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_56806_end_0 = const()[name = tensor("op_56806_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_56806_end_mask_0 = const()[name = tensor("op_56806_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_56806_cast_fp16 = slice_by_index(begin = var_56806_begin_0, end = var_56806_end_0, end_mask = var_56806_end_mask_0, x = k_503_cast_fp16)[name = tensor("op_56806_cast_fp16")]; + tensor var_56808_begin_0 = const()[name = tensor("op_56808_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_56808_end_0 = const()[name = tensor("op_56808_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_56808_end_mask_0 = const()[name = tensor("op_56808_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56808_cast_fp16 = slice_by_index(begin = var_56808_begin_0, end = var_56808_end_0, end_mask = var_56808_end_mask_0, x = v_251_cast_fp16)[name = tensor("op_56808_cast_fp16")]; + tensor var_56812_begin_0 = const()[name = tensor("op_56812_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_56812_end_0 = const()[name = tensor("op_56812_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_56812_end_mask_0 = const()[name = tensor("op_56812_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56812_cast_fp16 = slice_by_index(begin = var_56812_begin_0, end = var_56812_end_0, end_mask = var_56812_end_mask_0, x = v_251_cast_fp16)[name = tensor("op_56812_cast_fp16")]; + tensor var_56816_begin_0 = const()[name = tensor("op_56816_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_56816_end_0 = const()[name = tensor("op_56816_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_56816_end_mask_0 = const()[name = tensor("op_56816_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56816_cast_fp16 = slice_by_index(begin = var_56816_begin_0, end = var_56816_end_0, end_mask = var_56816_end_mask_0, x = v_251_cast_fp16)[name = tensor("op_56816_cast_fp16")]; + tensor var_56820_begin_0 = const()[name = tensor("op_56820_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_56820_end_0 = const()[name = tensor("op_56820_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_56820_end_mask_0 = const()[name = tensor("op_56820_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56820_cast_fp16 = slice_by_index(begin = var_56820_begin_0, end = var_56820_end_0, end_mask = var_56820_end_mask_0, x = v_251_cast_fp16)[name = tensor("op_56820_cast_fp16")]; + tensor var_56824_begin_0 = const()[name = tensor("op_56824_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_56824_end_0 = const()[name = tensor("op_56824_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_56824_end_mask_0 = const()[name = tensor("op_56824_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56824_cast_fp16 = slice_by_index(begin = var_56824_begin_0, end = var_56824_end_0, end_mask = var_56824_end_mask_0, x = v_251_cast_fp16)[name = tensor("op_56824_cast_fp16")]; + tensor var_56828_begin_0 = const()[name = tensor("op_56828_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_56828_end_0 = const()[name = tensor("op_56828_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_56828_end_mask_0 = const()[name = tensor("op_56828_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56828_cast_fp16 = slice_by_index(begin = var_56828_begin_0, end = var_56828_end_0, end_mask = var_56828_end_mask_0, x = v_251_cast_fp16)[name = tensor("op_56828_cast_fp16")]; + tensor var_56832_begin_0 = const()[name = tensor("op_56832_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_56832_end_0 = const()[name = tensor("op_56832_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_56832_end_mask_0 = const()[name = tensor("op_56832_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56832_cast_fp16 = slice_by_index(begin = var_56832_begin_0, end = var_56832_end_0, end_mask = var_56832_end_mask_0, x = v_251_cast_fp16)[name = tensor("op_56832_cast_fp16")]; + tensor var_56836_begin_0 = const()[name = tensor("op_56836_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_56836_end_0 = const()[name = tensor("op_56836_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_56836_end_mask_0 = const()[name = tensor("op_56836_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56836_cast_fp16 = slice_by_index(begin = var_56836_begin_0, end = var_56836_end_0, end_mask = var_56836_end_mask_0, x = v_251_cast_fp16)[name = tensor("op_56836_cast_fp16")]; + tensor var_56840_begin_0 = const()[name = tensor("op_56840_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_56840_end_0 = const()[name = tensor("op_56840_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_56840_end_mask_0 = const()[name = tensor("op_56840_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56840_cast_fp16 = slice_by_index(begin = var_56840_begin_0, end = var_56840_end_0, end_mask = var_56840_end_mask_0, x = v_251_cast_fp16)[name = tensor("op_56840_cast_fp16")]; + tensor var_56844_begin_0 = const()[name = tensor("op_56844_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_56844_end_0 = const()[name = tensor("op_56844_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_56844_end_mask_0 = const()[name = tensor("op_56844_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56844_cast_fp16 = slice_by_index(begin = var_56844_begin_0, end = var_56844_end_0, end_mask = var_56844_end_mask_0, x = v_251_cast_fp16)[name = tensor("op_56844_cast_fp16")]; + tensor var_56848_begin_0 = const()[name = tensor("op_56848_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_56848_end_0 = const()[name = tensor("op_56848_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_56848_end_mask_0 = const()[name = tensor("op_56848_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56848_cast_fp16 = slice_by_index(begin = var_56848_begin_0, end = var_56848_end_0, end_mask = var_56848_end_mask_0, x = v_251_cast_fp16)[name = tensor("op_56848_cast_fp16")]; + tensor var_56852_begin_0 = const()[name = tensor("op_56852_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_56852_end_0 = const()[name = tensor("op_56852_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_56852_end_mask_0 = const()[name = tensor("op_56852_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56852_cast_fp16 = slice_by_index(begin = var_56852_begin_0, end = var_56852_end_0, end_mask = var_56852_end_mask_0, x = v_251_cast_fp16)[name = tensor("op_56852_cast_fp16")]; + tensor var_56856_begin_0 = const()[name = tensor("op_56856_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_56856_end_0 = const()[name = tensor("op_56856_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_56856_end_mask_0 = const()[name = tensor("op_56856_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56856_cast_fp16 = slice_by_index(begin = var_56856_begin_0, end = var_56856_end_0, end_mask = var_56856_end_mask_0, x = v_251_cast_fp16)[name = tensor("op_56856_cast_fp16")]; + tensor var_56860_begin_0 = const()[name = tensor("op_56860_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_56860_end_0 = const()[name = tensor("op_56860_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_56860_end_mask_0 = const()[name = tensor("op_56860_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56860_cast_fp16 = slice_by_index(begin = var_56860_begin_0, end = var_56860_end_0, end_mask = var_56860_end_mask_0, x = v_251_cast_fp16)[name = tensor("op_56860_cast_fp16")]; + tensor var_56864_begin_0 = const()[name = tensor("op_56864_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_56864_end_0 = const()[name = tensor("op_56864_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_56864_end_mask_0 = const()[name = tensor("op_56864_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56864_cast_fp16 = slice_by_index(begin = var_56864_begin_0, end = var_56864_end_0, end_mask = var_56864_end_mask_0, x = v_251_cast_fp16)[name = tensor("op_56864_cast_fp16")]; + tensor var_56868_begin_0 = const()[name = tensor("op_56868_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_56868_end_0 = const()[name = tensor("op_56868_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_56868_end_mask_0 = const()[name = tensor("op_56868_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56868_cast_fp16 = slice_by_index(begin = var_56868_begin_0, end = var_56868_end_0, end_mask = var_56868_end_mask_0, x = v_251_cast_fp16)[name = tensor("op_56868_cast_fp16")]; + tensor var_56872_begin_0 = const()[name = tensor("op_56872_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_56872_end_0 = const()[name = tensor("op_56872_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_56872_end_mask_0 = const()[name = tensor("op_56872_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56872_cast_fp16 = slice_by_index(begin = var_56872_begin_0, end = var_56872_end_0, end_mask = var_56872_end_mask_0, x = v_251_cast_fp16)[name = tensor("op_56872_cast_fp16")]; + tensor var_56876_begin_0 = const()[name = tensor("op_56876_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_56876_end_0 = const()[name = tensor("op_56876_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_56876_end_mask_0 = const()[name = tensor("op_56876_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56876_cast_fp16 = slice_by_index(begin = var_56876_begin_0, end = var_56876_end_0, end_mask = var_56876_end_mask_0, x = v_251_cast_fp16)[name = tensor("op_56876_cast_fp16")]; + tensor var_56880_begin_0 = const()[name = tensor("op_56880_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_56880_end_0 = const()[name = tensor("op_56880_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_56880_end_mask_0 = const()[name = tensor("op_56880_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56880_cast_fp16 = slice_by_index(begin = var_56880_begin_0, end = var_56880_end_0, end_mask = var_56880_end_mask_0, x = v_251_cast_fp16)[name = tensor("op_56880_cast_fp16")]; + tensor var_56884_begin_0 = const()[name = tensor("op_56884_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_56884_end_0 = const()[name = tensor("op_56884_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_56884_end_mask_0 = const()[name = tensor("op_56884_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_56884_cast_fp16 = slice_by_index(begin = var_56884_begin_0, end = var_56884_end_0, end_mask = var_56884_end_mask_0, x = v_251_cast_fp16)[name = tensor("op_56884_cast_fp16")]; + tensor var_56888_equation_0 = const()[name = tensor("op_56888_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_56888_cast_fp16 = einsum(equation = var_56888_equation_0, values = (var_56730_cast_fp16, var_56647_cast_fp16))[name = tensor("op_56888_cast_fp16")]; + tensor var_56889_to_fp16 = const()[name = tensor("op_56889_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4841_cast_fp16 = mul(x = var_56888_cast_fp16, y = var_56889_to_fp16)[name = tensor("aw_4841_cast_fp16")]; + tensor var_56892_equation_0 = const()[name = tensor("op_56892_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_56892_cast_fp16 = einsum(equation = var_56892_equation_0, values = (var_56734_cast_fp16, var_56651_cast_fp16))[name = tensor("op_56892_cast_fp16")]; + tensor var_56893_to_fp16 = const()[name = tensor("op_56893_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4843_cast_fp16 = mul(x = var_56892_cast_fp16, y = var_56893_to_fp16)[name = tensor("aw_4843_cast_fp16")]; + tensor var_56896_equation_0 = const()[name = tensor("op_56896_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_56896_cast_fp16 = einsum(equation = var_56896_equation_0, values = (var_56738_cast_fp16, var_56655_cast_fp16))[name = tensor("op_56896_cast_fp16")]; + tensor var_56897_to_fp16 = const()[name = tensor("op_56897_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4845_cast_fp16 = mul(x = var_56896_cast_fp16, y = var_56897_to_fp16)[name = tensor("aw_4845_cast_fp16")]; + tensor var_56900_equation_0 = const()[name = tensor("op_56900_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_56900_cast_fp16 = einsum(equation = var_56900_equation_0, values = (var_56742_cast_fp16, var_56659_cast_fp16))[name = tensor("op_56900_cast_fp16")]; + tensor var_56901_to_fp16 = const()[name = tensor("op_56901_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4847_cast_fp16 = mul(x = var_56900_cast_fp16, y = var_56901_to_fp16)[name = tensor("aw_4847_cast_fp16")]; + tensor var_56904_equation_0 = const()[name = tensor("op_56904_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_56904_cast_fp16 = einsum(equation = var_56904_equation_0, values = (var_56746_cast_fp16, var_56663_cast_fp16))[name = tensor("op_56904_cast_fp16")]; + tensor var_56905_to_fp16 = const()[name = tensor("op_56905_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4849_cast_fp16 = mul(x = var_56904_cast_fp16, y = var_56905_to_fp16)[name = tensor("aw_4849_cast_fp16")]; + tensor var_56908_equation_0 = const()[name = tensor("op_56908_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_56908_cast_fp16 = einsum(equation = var_56908_equation_0, values = (var_56750_cast_fp16, var_56667_cast_fp16))[name = tensor("op_56908_cast_fp16")]; + tensor var_56909_to_fp16 = const()[name = tensor("op_56909_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4851_cast_fp16 = mul(x = var_56908_cast_fp16, y = var_56909_to_fp16)[name = tensor("aw_4851_cast_fp16")]; + tensor var_56912_equation_0 = const()[name = tensor("op_56912_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_56912_cast_fp16 = einsum(equation = var_56912_equation_0, values = (var_56754_cast_fp16, var_56671_cast_fp16))[name = tensor("op_56912_cast_fp16")]; + tensor var_56913_to_fp16 = const()[name = tensor("op_56913_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4853_cast_fp16 = mul(x = var_56912_cast_fp16, y = var_56913_to_fp16)[name = tensor("aw_4853_cast_fp16")]; + tensor var_56916_equation_0 = const()[name = tensor("op_56916_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_56916_cast_fp16 = einsum(equation = var_56916_equation_0, values = (var_56758_cast_fp16, var_56675_cast_fp16))[name = tensor("op_56916_cast_fp16")]; + tensor var_56917_to_fp16 = const()[name = tensor("op_56917_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4855_cast_fp16 = mul(x = var_56916_cast_fp16, y = var_56917_to_fp16)[name = tensor("aw_4855_cast_fp16")]; + tensor var_56920_equation_0 = const()[name = tensor("op_56920_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_56920_cast_fp16 = einsum(equation = var_56920_equation_0, values = (var_56762_cast_fp16, var_56679_cast_fp16))[name = tensor("op_56920_cast_fp16")]; + tensor var_56921_to_fp16 = const()[name = tensor("op_56921_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4857_cast_fp16 = mul(x = var_56920_cast_fp16, y = var_56921_to_fp16)[name = tensor("aw_4857_cast_fp16")]; + tensor var_56924_equation_0 = const()[name = tensor("op_56924_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_56924_cast_fp16 = einsum(equation = var_56924_equation_0, values = (var_56766_cast_fp16, var_56683_cast_fp16))[name = tensor("op_56924_cast_fp16")]; + tensor var_56925_to_fp16 = const()[name = tensor("op_56925_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4859_cast_fp16 = mul(x = var_56924_cast_fp16, y = var_56925_to_fp16)[name = tensor("aw_4859_cast_fp16")]; + tensor var_56928_equation_0 = const()[name = tensor("op_56928_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_56928_cast_fp16 = einsum(equation = var_56928_equation_0, values = (var_56770_cast_fp16, var_56687_cast_fp16))[name = tensor("op_56928_cast_fp16")]; + tensor var_56929_to_fp16 = const()[name = tensor("op_56929_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4861_cast_fp16 = mul(x = var_56928_cast_fp16, y = var_56929_to_fp16)[name = tensor("aw_4861_cast_fp16")]; + tensor var_56932_equation_0 = const()[name = tensor("op_56932_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_56932_cast_fp16 = einsum(equation = var_56932_equation_0, values = (var_56774_cast_fp16, var_56691_cast_fp16))[name = tensor("op_56932_cast_fp16")]; + tensor var_56933_to_fp16 = const()[name = tensor("op_56933_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4863_cast_fp16 = mul(x = var_56932_cast_fp16, y = var_56933_to_fp16)[name = tensor("aw_4863_cast_fp16")]; + tensor var_56936_equation_0 = const()[name = tensor("op_56936_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_56936_cast_fp16 = einsum(equation = var_56936_equation_0, values = (var_56778_cast_fp16, var_56695_cast_fp16))[name = tensor("op_56936_cast_fp16")]; + tensor var_56937_to_fp16 = const()[name = tensor("op_56937_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4865_cast_fp16 = mul(x = var_56936_cast_fp16, y = var_56937_to_fp16)[name = tensor("aw_4865_cast_fp16")]; + tensor var_56940_equation_0 = const()[name = tensor("op_56940_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_56940_cast_fp16 = einsum(equation = var_56940_equation_0, values = (var_56782_cast_fp16, var_56699_cast_fp16))[name = tensor("op_56940_cast_fp16")]; + tensor var_56941_to_fp16 = const()[name = tensor("op_56941_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4867_cast_fp16 = mul(x = var_56940_cast_fp16, y = var_56941_to_fp16)[name = tensor("aw_4867_cast_fp16")]; + tensor var_56944_equation_0 = const()[name = tensor("op_56944_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_56944_cast_fp16 = einsum(equation = var_56944_equation_0, values = (var_56786_cast_fp16, var_56703_cast_fp16))[name = tensor("op_56944_cast_fp16")]; + tensor var_56945_to_fp16 = const()[name = tensor("op_56945_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4869_cast_fp16 = mul(x = var_56944_cast_fp16, y = var_56945_to_fp16)[name = tensor("aw_4869_cast_fp16")]; + tensor var_56948_equation_0 = const()[name = tensor("op_56948_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_56948_cast_fp16 = einsum(equation = var_56948_equation_0, values = (var_56790_cast_fp16, var_56707_cast_fp16))[name = tensor("op_56948_cast_fp16")]; + tensor var_56949_to_fp16 = const()[name = tensor("op_56949_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4871_cast_fp16 = mul(x = var_56948_cast_fp16, y = var_56949_to_fp16)[name = tensor("aw_4871_cast_fp16")]; + tensor var_56952_equation_0 = const()[name = tensor("op_56952_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_56952_cast_fp16 = einsum(equation = var_56952_equation_0, values = (var_56794_cast_fp16, var_56711_cast_fp16))[name = tensor("op_56952_cast_fp16")]; + tensor var_56953_to_fp16 = const()[name = tensor("op_56953_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4873_cast_fp16 = mul(x = var_56952_cast_fp16, y = var_56953_to_fp16)[name = tensor("aw_4873_cast_fp16")]; + tensor var_56956_equation_0 = const()[name = tensor("op_56956_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_56956_cast_fp16 = einsum(equation = var_56956_equation_0, values = (var_56798_cast_fp16, var_56715_cast_fp16))[name = tensor("op_56956_cast_fp16")]; + tensor var_56957_to_fp16 = const()[name = tensor("op_56957_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4875_cast_fp16 = mul(x = var_56956_cast_fp16, y = var_56957_to_fp16)[name = tensor("aw_4875_cast_fp16")]; + tensor var_56960_equation_0 = const()[name = tensor("op_56960_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_56960_cast_fp16 = einsum(equation = var_56960_equation_0, values = (var_56802_cast_fp16, var_56719_cast_fp16))[name = tensor("op_56960_cast_fp16")]; + tensor var_56961_to_fp16 = const()[name = tensor("op_56961_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4877_cast_fp16 = mul(x = var_56960_cast_fp16, y = var_56961_to_fp16)[name = tensor("aw_4877_cast_fp16")]; + tensor var_56964_equation_0 = const()[name = tensor("op_56964_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_56964_cast_fp16 = einsum(equation = var_56964_equation_0, values = (var_56806_cast_fp16, var_56723_cast_fp16))[name = tensor("op_56964_cast_fp16")]; + tensor var_56965_to_fp16 = const()[name = tensor("op_56965_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4879_cast_fp16 = mul(x = var_56964_cast_fp16, y = var_56965_to_fp16)[name = tensor("aw_4879_cast_fp16")]; + tensor var_56967_cast_fp16 = softmax(axis = var_30355, x = aw_4841_cast_fp16)[name = tensor("op_56967_cast_fp16")]; + tensor var_56968_cast_fp16 = softmax(axis = var_30355, x = aw_4843_cast_fp16)[name = tensor("op_56968_cast_fp16")]; + tensor var_56969_cast_fp16 = softmax(axis = var_30355, x = aw_4845_cast_fp16)[name = tensor("op_56969_cast_fp16")]; + tensor var_56970_cast_fp16 = softmax(axis = var_30355, x = aw_4847_cast_fp16)[name = tensor("op_56970_cast_fp16")]; + tensor var_56971_cast_fp16 = softmax(axis = var_30355, x = aw_4849_cast_fp16)[name = tensor("op_56971_cast_fp16")]; + tensor var_56972_cast_fp16 = softmax(axis = var_30355, x = aw_4851_cast_fp16)[name = tensor("op_56972_cast_fp16")]; + tensor var_56973_cast_fp16 = softmax(axis = var_30355, x = aw_4853_cast_fp16)[name = tensor("op_56973_cast_fp16")]; + tensor var_56974_cast_fp16 = softmax(axis = var_30355, x = aw_4855_cast_fp16)[name = tensor("op_56974_cast_fp16")]; + tensor var_56975_cast_fp16 = softmax(axis = var_30355, x = aw_4857_cast_fp16)[name = tensor("op_56975_cast_fp16")]; + tensor var_56976_cast_fp16 = softmax(axis = var_30355, x = aw_4859_cast_fp16)[name = tensor("op_56976_cast_fp16")]; + tensor var_56977_cast_fp16 = softmax(axis = var_30355, x = aw_4861_cast_fp16)[name = tensor("op_56977_cast_fp16")]; + tensor var_56978_cast_fp16 = softmax(axis = var_30355, x = aw_4863_cast_fp16)[name = tensor("op_56978_cast_fp16")]; + tensor var_56979_cast_fp16 = softmax(axis = var_30355, x = aw_4865_cast_fp16)[name = tensor("op_56979_cast_fp16")]; + tensor var_56980_cast_fp16 = softmax(axis = var_30355, x = aw_4867_cast_fp16)[name = tensor("op_56980_cast_fp16")]; + tensor var_56981_cast_fp16 = softmax(axis = var_30355, x = aw_4869_cast_fp16)[name = tensor("op_56981_cast_fp16")]; + tensor var_56982_cast_fp16 = softmax(axis = var_30355, x = aw_4871_cast_fp16)[name = tensor("op_56982_cast_fp16")]; + tensor var_56983_cast_fp16 = softmax(axis = var_30355, x = aw_4873_cast_fp16)[name = tensor("op_56983_cast_fp16")]; + tensor var_56984_cast_fp16 = softmax(axis = var_30355, x = aw_4875_cast_fp16)[name = tensor("op_56984_cast_fp16")]; + tensor var_56985_cast_fp16 = softmax(axis = var_30355, x = aw_4877_cast_fp16)[name = tensor("op_56985_cast_fp16")]; + tensor var_56986_cast_fp16 = softmax(axis = var_30355, x = aw_4879_cast_fp16)[name = tensor("op_56986_cast_fp16")]; + tensor var_56988_equation_0 = const()[name = tensor("op_56988_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_56988_cast_fp16 = einsum(equation = var_56988_equation_0, values = (var_56808_cast_fp16, var_56967_cast_fp16))[name = tensor("op_56988_cast_fp16")]; + tensor var_56990_equation_0 = const()[name = tensor("op_56990_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_56990_cast_fp16 = einsum(equation = var_56990_equation_0, values = (var_56812_cast_fp16, var_56968_cast_fp16))[name = tensor("op_56990_cast_fp16")]; + tensor var_56992_equation_0 = const()[name = tensor("op_56992_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_56992_cast_fp16 = einsum(equation = var_56992_equation_0, values = (var_56816_cast_fp16, var_56969_cast_fp16))[name = tensor("op_56992_cast_fp16")]; + tensor var_56994_equation_0 = const()[name = tensor("op_56994_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_56994_cast_fp16 = einsum(equation = var_56994_equation_0, values = (var_56820_cast_fp16, var_56970_cast_fp16))[name = tensor("op_56994_cast_fp16")]; + tensor var_56996_equation_0 = const()[name = tensor("op_56996_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_56996_cast_fp16 = einsum(equation = var_56996_equation_0, values = (var_56824_cast_fp16, var_56971_cast_fp16))[name = tensor("op_56996_cast_fp16")]; + tensor var_56998_equation_0 = const()[name = tensor("op_56998_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_56998_cast_fp16 = einsum(equation = var_56998_equation_0, values = (var_56828_cast_fp16, var_56972_cast_fp16))[name = tensor("op_56998_cast_fp16")]; + tensor var_57000_equation_0 = const()[name = tensor("op_57000_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_57000_cast_fp16 = einsum(equation = var_57000_equation_0, values = (var_56832_cast_fp16, var_56973_cast_fp16))[name = tensor("op_57000_cast_fp16")]; + tensor var_57002_equation_0 = const()[name = tensor("op_57002_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_57002_cast_fp16 = einsum(equation = var_57002_equation_0, values = (var_56836_cast_fp16, var_56974_cast_fp16))[name = tensor("op_57002_cast_fp16")]; + tensor var_57004_equation_0 = const()[name = tensor("op_57004_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_57004_cast_fp16 = einsum(equation = var_57004_equation_0, values = (var_56840_cast_fp16, var_56975_cast_fp16))[name = tensor("op_57004_cast_fp16")]; + tensor var_57006_equation_0 = const()[name = tensor("op_57006_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_57006_cast_fp16 = einsum(equation = var_57006_equation_0, values = (var_56844_cast_fp16, var_56976_cast_fp16))[name = tensor("op_57006_cast_fp16")]; + tensor var_57008_equation_0 = const()[name = tensor("op_57008_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_57008_cast_fp16 = einsum(equation = var_57008_equation_0, values = (var_56848_cast_fp16, var_56977_cast_fp16))[name = tensor("op_57008_cast_fp16")]; + tensor var_57010_equation_0 = const()[name = tensor("op_57010_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_57010_cast_fp16 = einsum(equation = var_57010_equation_0, values = (var_56852_cast_fp16, var_56978_cast_fp16))[name = tensor("op_57010_cast_fp16")]; + tensor var_57012_equation_0 = const()[name = tensor("op_57012_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_57012_cast_fp16 = einsum(equation = var_57012_equation_0, values = (var_56856_cast_fp16, var_56979_cast_fp16))[name = tensor("op_57012_cast_fp16")]; + tensor var_57014_equation_0 = const()[name = tensor("op_57014_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_57014_cast_fp16 = einsum(equation = var_57014_equation_0, values = (var_56860_cast_fp16, var_56980_cast_fp16))[name = tensor("op_57014_cast_fp16")]; + tensor var_57016_equation_0 = const()[name = tensor("op_57016_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_57016_cast_fp16 = einsum(equation = var_57016_equation_0, values = (var_56864_cast_fp16, var_56981_cast_fp16))[name = tensor("op_57016_cast_fp16")]; + tensor var_57018_equation_0 = const()[name = tensor("op_57018_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_57018_cast_fp16 = einsum(equation = var_57018_equation_0, values = (var_56868_cast_fp16, var_56982_cast_fp16))[name = tensor("op_57018_cast_fp16")]; + tensor var_57020_equation_0 = const()[name = tensor("op_57020_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_57020_cast_fp16 = einsum(equation = var_57020_equation_0, values = (var_56872_cast_fp16, var_56983_cast_fp16))[name = tensor("op_57020_cast_fp16")]; + tensor var_57022_equation_0 = const()[name = tensor("op_57022_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_57022_cast_fp16 = einsum(equation = var_57022_equation_0, values = (var_56876_cast_fp16, var_56984_cast_fp16))[name = tensor("op_57022_cast_fp16")]; + tensor var_57024_equation_0 = const()[name = tensor("op_57024_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_57024_cast_fp16 = einsum(equation = var_57024_equation_0, values = (var_56880_cast_fp16, var_56985_cast_fp16))[name = tensor("op_57024_cast_fp16")]; + tensor var_57026_equation_0 = const()[name = tensor("op_57026_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_57026_cast_fp16 = einsum(equation = var_57026_equation_0, values = (var_56884_cast_fp16, var_56986_cast_fp16))[name = tensor("op_57026_cast_fp16")]; + tensor input_701_interleave_0 = const()[name = tensor("input_701_interleave_0"), val = tensor(false)]; + tensor input_701_cast_fp16 = concat(axis = var_30355, interleave = input_701_interleave_0, values = (var_56988_cast_fp16, var_56990_cast_fp16, var_56992_cast_fp16, var_56994_cast_fp16, var_56996_cast_fp16, var_56998_cast_fp16, var_57000_cast_fp16, var_57002_cast_fp16, var_57004_cast_fp16, var_57006_cast_fp16, var_57008_cast_fp16, var_57010_cast_fp16, var_57012_cast_fp16, var_57014_cast_fp16, var_57016_cast_fp16, var_57018_cast_fp16, var_57020_cast_fp16, var_57022_cast_fp16, var_57024_cast_fp16, var_57026_cast_fp16))[name = tensor("input_701_cast_fp16")]; + tensor var_57032 = const()[name = tensor("op_57032"), val = tensor([1, 1])]; + tensor var_57034 = const()[name = tensor("op_57034"), val = tensor([1, 1])]; + tensor var_57036_pad_type_0 = const()[name = tensor("op_57036_pad_type_0"), val = tensor("custom")]; + tensor var_57036_pad_0 = const()[name = tensor("op_57036_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_8_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_8_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4755078016)))]; + tensor up_blocks_0_attentions_2_transformer_blocks_8_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_8_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4758354880)))]; + tensor var_57036_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_8_attn2_to_out_0_bias_to_fp16, dilations = var_57034, groups = var_30355, pad = var_57036_pad_0, pad_type = var_57036_pad_type_0, strides = var_57032, weight = up_blocks_0_attentions_2_transformer_blocks_8_attn2_to_out_0_weight_to_fp16, x = input_701_cast_fp16)[name = tensor("op_57036_cast_fp16")]; + tensor inputs_377_cast_fp16 = add(x = var_57036_cast_fp16, y = inputs_375_cast_fp16)[name = tensor("inputs_377_cast_fp16")]; + tensor input_703_axes_0 = const()[name = tensor("input_703_axes_0"), val = tensor([1])]; + tensor input_703_gamma_0_to_fp16 = const()[name = tensor("input_703_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4758357504)))]; + tensor input_703_beta_0_to_fp16 = const()[name = tensor("input_703_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4758360128)))]; + tensor var_57046_to_fp16 = const()[name = tensor("op_57046_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_703_cast_fp16 = layer_norm(axes = input_703_axes_0, beta = input_703_beta_0_to_fp16, epsilon = var_57046_to_fp16, gamma = input_703_gamma_0_to_fp16, x = inputs_377_cast_fp16)[name = tensor("input_703_cast_fp16")]; + tensor var_57062 = const()[name = tensor("op_57062"), val = tensor([1, 1])]; + tensor var_57064 = const()[name = tensor("op_57064"), val = tensor([1, 1])]; + tensor var_57066_pad_type_0 = const()[name = tensor("op_57066_pad_type_0"), val = tensor("custom")]; + tensor var_57066_pad_0 = const()[name = tensor("op_57066_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_8_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_8_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4758362752)))]; + tensor up_blocks_0_attentions_2_transformer_blocks_8_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_8_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4784577216)))]; + tensor var_57066_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_8_ff_net_0_proj_bias_to_fp16, dilations = var_57064, groups = var_30355, pad = var_57066_pad_0, pad_type = var_57066_pad_type_0, strides = var_57062, weight = up_blocks_0_attentions_2_transformer_blocks_8_ff_net_0_proj_weight_to_fp16, x = input_703_cast_fp16)[name = tensor("op_57066_cast_fp16")]; + tensor var_57067_split_sizes_0 = const()[name = tensor("op_57067_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_57067_axis_0 = const()[name = tensor("op_57067_axis_0"), val = tensor(1)]; + tensor var_57067_cast_fp16_0, tensor var_57067_cast_fp16_1 = split(axis = var_57067_axis_0, split_sizes = var_57067_split_sizes_0, x = var_57066_cast_fp16)[name = tensor("op_57067_cast_fp16")]; + tensor var_57069_mode_0 = const()[name = tensor("op_57069_mode_0"), val = tensor("EXACT")]; + tensor var_57069_cast_fp16 = gelu(mode = var_57069_mode_0, x = var_57067_cast_fp16_1)[name = tensor("op_57069_cast_fp16")]; + tensor input_705_cast_fp16 = mul(x = var_57067_cast_fp16_0, y = var_57069_cast_fp16)[name = tensor("input_705_cast_fp16")]; + tensor var_57073 = const()[name = tensor("op_57073"), val = tensor([1, 1])]; + tensor var_57075 = const()[name = tensor("op_57075"), val = tensor([1, 1])]; + tensor var_57077_pad_type_0 = const()[name = tensor("op_57077_pad_type_0"), val = tensor("custom")]; + tensor var_57077_pad_0 = const()[name = tensor("op_57077_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_8_ff_net_2_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_8_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4784597760)))]; + tensor up_blocks_0_attentions_2_transformer_blocks_8_ff_net_2_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_8_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4797705024)))]; + tensor var_57077_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_8_ff_net_2_bias_to_fp16, dilations = var_57075, groups = var_30355, pad = var_57077_pad_0, pad_type = var_57077_pad_type_0, strides = var_57073, weight = up_blocks_0_attentions_2_transformer_blocks_8_ff_net_2_weight_to_fp16, x = input_705_cast_fp16)[name = tensor("op_57077_cast_fp16")]; + tensor inputs_379_cast_fp16 = add(x = var_57077_cast_fp16, y = inputs_377_cast_fp16)[name = tensor("inputs_379_cast_fp16")]; + tensor hidden_states_489_axes_0 = const()[name = tensor("hidden_states_489_axes_0"), val = tensor([1])]; + tensor hidden_states_489_gamma_0_to_fp16 = const()[name = tensor("hidden_states_489_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4797707648)))]; + tensor hidden_states_489_beta_0_to_fp16 = const()[name = tensor("hidden_states_489_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4797710272)))]; + tensor var_57093_to_fp16 = const()[name = tensor("op_57093_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_489_cast_fp16 = layer_norm(axes = hidden_states_489_axes_0, beta = hidden_states_489_beta_0_to_fp16, epsilon = var_57093_to_fp16, gamma = hidden_states_489_gamma_0_to_fp16, x = inputs_379_cast_fp16)[name = tensor("hidden_states_489_cast_fp16")]; + tensor var_57108 = const()[name = tensor("op_57108"), val = tensor([1, 1])]; + tensor var_57110 = const()[name = tensor("op_57110"), val = tensor([1, 1])]; + tensor q_253_pad_type_0 = const()[name = tensor("q_253_pad_type_0"), val = tensor("custom")]; + tensor q_253_pad_0 = const()[name = tensor("q_253_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_9_attn1_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_9_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4797712896)))]; + tensor q_253_cast_fp16 = conv(dilations = var_57110, groups = var_30355, pad = q_253_pad_0, pad_type = q_253_pad_type_0, strides = var_57108, weight = up_blocks_0_attentions_2_transformer_blocks_9_attn1_to_q_weight_to_fp16, x = hidden_states_489_cast_fp16)[name = tensor("q_253_cast_fp16")]; + tensor var_57114 = const()[name = tensor("op_57114"), val = tensor([1, 1])]; + tensor var_57116 = const()[name = tensor("op_57116"), val = tensor([1, 1])]; + tensor k_505_pad_type_0 = const()[name = tensor("k_505_pad_type_0"), val = tensor("custom")]; + tensor k_505_pad_0 = const()[name = tensor("k_505_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_9_attn1_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_9_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4800989760)))]; + tensor k_505_cast_fp16 = conv(dilations = var_57116, groups = var_30355, pad = k_505_pad_0, pad_type = k_505_pad_type_0, strides = var_57114, weight = up_blocks_0_attentions_2_transformer_blocks_9_attn1_to_k_weight_to_fp16, x = hidden_states_489_cast_fp16)[name = tensor("k_505_cast_fp16")]; + tensor var_57120 = const()[name = tensor("op_57120"), val = tensor([1, 1])]; + tensor var_57122 = const()[name = tensor("op_57122"), val = tensor([1, 1])]; + tensor v_253_pad_type_0 = const()[name = tensor("v_253_pad_type_0"), val = tensor("custom")]; + tensor v_253_pad_0 = const()[name = tensor("v_253_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_9_attn1_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_9_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4804266624)))]; + tensor v_253_cast_fp16 = conv(dilations = var_57122, groups = var_30355, pad = v_253_pad_0, pad_type = v_253_pad_type_0, strides = var_57120, weight = up_blocks_0_attentions_2_transformer_blocks_9_attn1_to_v_weight_to_fp16, x = hidden_states_489_cast_fp16)[name = tensor("v_253_cast_fp16")]; + tensor var_57126_begin_0 = const()[name = tensor("op_57126_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_57126_end_0 = const()[name = tensor("op_57126_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_57126_end_mask_0 = const()[name = tensor("op_57126_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57126_cast_fp16 = slice_by_index(begin = var_57126_begin_0, end = var_57126_end_0, end_mask = var_57126_end_mask_0, x = q_253_cast_fp16)[name = tensor("op_57126_cast_fp16")]; + tensor var_57130_begin_0 = const()[name = tensor("op_57130_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_57130_end_0 = const()[name = tensor("op_57130_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_57130_end_mask_0 = const()[name = tensor("op_57130_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57130_cast_fp16 = slice_by_index(begin = var_57130_begin_0, end = var_57130_end_0, end_mask = var_57130_end_mask_0, x = q_253_cast_fp16)[name = tensor("op_57130_cast_fp16")]; + tensor var_57134_begin_0 = const()[name = tensor("op_57134_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_57134_end_0 = const()[name = tensor("op_57134_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_57134_end_mask_0 = const()[name = tensor("op_57134_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57134_cast_fp16 = slice_by_index(begin = var_57134_begin_0, end = var_57134_end_0, end_mask = var_57134_end_mask_0, x = q_253_cast_fp16)[name = tensor("op_57134_cast_fp16")]; + tensor var_57138_begin_0 = const()[name = tensor("op_57138_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_57138_end_0 = const()[name = tensor("op_57138_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_57138_end_mask_0 = const()[name = tensor("op_57138_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57138_cast_fp16 = slice_by_index(begin = var_57138_begin_0, end = var_57138_end_0, end_mask = var_57138_end_mask_0, x = q_253_cast_fp16)[name = tensor("op_57138_cast_fp16")]; + tensor var_57142_begin_0 = const()[name = tensor("op_57142_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_57142_end_0 = const()[name = tensor("op_57142_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_57142_end_mask_0 = const()[name = tensor("op_57142_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57142_cast_fp16 = slice_by_index(begin = var_57142_begin_0, end = var_57142_end_0, end_mask = var_57142_end_mask_0, x = q_253_cast_fp16)[name = tensor("op_57142_cast_fp16")]; + tensor var_57146_begin_0 = const()[name = tensor("op_57146_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_57146_end_0 = const()[name = tensor("op_57146_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_57146_end_mask_0 = const()[name = tensor("op_57146_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57146_cast_fp16 = slice_by_index(begin = var_57146_begin_0, end = var_57146_end_0, end_mask = var_57146_end_mask_0, x = q_253_cast_fp16)[name = tensor("op_57146_cast_fp16")]; + tensor var_57150_begin_0 = const()[name = tensor("op_57150_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_57150_end_0 = const()[name = tensor("op_57150_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_57150_end_mask_0 = const()[name = tensor("op_57150_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57150_cast_fp16 = slice_by_index(begin = var_57150_begin_0, end = var_57150_end_0, end_mask = var_57150_end_mask_0, x = q_253_cast_fp16)[name = tensor("op_57150_cast_fp16")]; + tensor var_57154_begin_0 = const()[name = tensor("op_57154_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_57154_end_0 = const()[name = tensor("op_57154_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_57154_end_mask_0 = const()[name = tensor("op_57154_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57154_cast_fp16 = slice_by_index(begin = var_57154_begin_0, end = var_57154_end_0, end_mask = var_57154_end_mask_0, x = q_253_cast_fp16)[name = tensor("op_57154_cast_fp16")]; + tensor var_57158_begin_0 = const()[name = tensor("op_57158_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_57158_end_0 = const()[name = tensor("op_57158_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_57158_end_mask_0 = const()[name = tensor("op_57158_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57158_cast_fp16 = slice_by_index(begin = var_57158_begin_0, end = var_57158_end_0, end_mask = var_57158_end_mask_0, x = q_253_cast_fp16)[name = tensor("op_57158_cast_fp16")]; + tensor var_57162_begin_0 = const()[name = tensor("op_57162_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_57162_end_0 = const()[name = tensor("op_57162_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_57162_end_mask_0 = const()[name = tensor("op_57162_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57162_cast_fp16 = slice_by_index(begin = var_57162_begin_0, end = var_57162_end_0, end_mask = var_57162_end_mask_0, x = q_253_cast_fp16)[name = tensor("op_57162_cast_fp16")]; + tensor var_57166_begin_0 = const()[name = tensor("op_57166_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_57166_end_0 = const()[name = tensor("op_57166_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_57166_end_mask_0 = const()[name = tensor("op_57166_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57166_cast_fp16 = slice_by_index(begin = var_57166_begin_0, end = var_57166_end_0, end_mask = var_57166_end_mask_0, x = q_253_cast_fp16)[name = tensor("op_57166_cast_fp16")]; + tensor var_57170_begin_0 = const()[name = tensor("op_57170_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_57170_end_0 = const()[name = tensor("op_57170_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_57170_end_mask_0 = const()[name = tensor("op_57170_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57170_cast_fp16 = slice_by_index(begin = var_57170_begin_0, end = var_57170_end_0, end_mask = var_57170_end_mask_0, x = q_253_cast_fp16)[name = tensor("op_57170_cast_fp16")]; + tensor var_57174_begin_0 = const()[name = tensor("op_57174_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_57174_end_0 = const()[name = tensor("op_57174_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_57174_end_mask_0 = const()[name = tensor("op_57174_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57174_cast_fp16 = slice_by_index(begin = var_57174_begin_0, end = var_57174_end_0, end_mask = var_57174_end_mask_0, x = q_253_cast_fp16)[name = tensor("op_57174_cast_fp16")]; + tensor var_57178_begin_0 = const()[name = tensor("op_57178_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_57178_end_0 = const()[name = tensor("op_57178_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_57178_end_mask_0 = const()[name = tensor("op_57178_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57178_cast_fp16 = slice_by_index(begin = var_57178_begin_0, end = var_57178_end_0, end_mask = var_57178_end_mask_0, x = q_253_cast_fp16)[name = tensor("op_57178_cast_fp16")]; + tensor var_57182_begin_0 = const()[name = tensor("op_57182_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_57182_end_0 = const()[name = tensor("op_57182_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_57182_end_mask_0 = const()[name = tensor("op_57182_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57182_cast_fp16 = slice_by_index(begin = var_57182_begin_0, end = var_57182_end_0, end_mask = var_57182_end_mask_0, x = q_253_cast_fp16)[name = tensor("op_57182_cast_fp16")]; + tensor var_57186_begin_0 = const()[name = tensor("op_57186_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_57186_end_0 = const()[name = tensor("op_57186_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_57186_end_mask_0 = const()[name = tensor("op_57186_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57186_cast_fp16 = slice_by_index(begin = var_57186_begin_0, end = var_57186_end_0, end_mask = var_57186_end_mask_0, x = q_253_cast_fp16)[name = tensor("op_57186_cast_fp16")]; + tensor var_57190_begin_0 = const()[name = tensor("op_57190_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_57190_end_0 = const()[name = tensor("op_57190_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_57190_end_mask_0 = const()[name = tensor("op_57190_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57190_cast_fp16 = slice_by_index(begin = var_57190_begin_0, end = var_57190_end_0, end_mask = var_57190_end_mask_0, x = q_253_cast_fp16)[name = tensor("op_57190_cast_fp16")]; + tensor var_57194_begin_0 = const()[name = tensor("op_57194_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_57194_end_0 = const()[name = tensor("op_57194_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_57194_end_mask_0 = const()[name = tensor("op_57194_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57194_cast_fp16 = slice_by_index(begin = var_57194_begin_0, end = var_57194_end_0, end_mask = var_57194_end_mask_0, x = q_253_cast_fp16)[name = tensor("op_57194_cast_fp16")]; + tensor var_57198_begin_0 = const()[name = tensor("op_57198_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_57198_end_0 = const()[name = tensor("op_57198_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_57198_end_mask_0 = const()[name = tensor("op_57198_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57198_cast_fp16 = slice_by_index(begin = var_57198_begin_0, end = var_57198_end_0, end_mask = var_57198_end_mask_0, x = q_253_cast_fp16)[name = tensor("op_57198_cast_fp16")]; + tensor var_57202_begin_0 = const()[name = tensor("op_57202_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_57202_end_0 = const()[name = tensor("op_57202_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_57202_end_mask_0 = const()[name = tensor("op_57202_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57202_cast_fp16 = slice_by_index(begin = var_57202_begin_0, end = var_57202_end_0, end_mask = var_57202_end_mask_0, x = q_253_cast_fp16)[name = tensor("op_57202_cast_fp16")]; + tensor k_507_perm_0 = const()[name = tensor("k_507_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_57209_begin_0 = const()[name = tensor("op_57209_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_57209_end_0 = const()[name = tensor("op_57209_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_57209_end_mask_0 = const()[name = tensor("op_57209_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_507_cast_fp16 = transpose(perm = k_507_perm_0, x = k_505_cast_fp16)[name = tensor("transpose_13")]; + tensor var_57209_cast_fp16 = slice_by_index(begin = var_57209_begin_0, end = var_57209_end_0, end_mask = var_57209_end_mask_0, x = k_507_cast_fp16)[name = tensor("op_57209_cast_fp16")]; + tensor var_57213_begin_0 = const()[name = tensor("op_57213_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_57213_end_0 = const()[name = tensor("op_57213_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_57213_end_mask_0 = const()[name = tensor("op_57213_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_57213_cast_fp16 = slice_by_index(begin = var_57213_begin_0, end = var_57213_end_0, end_mask = var_57213_end_mask_0, x = k_507_cast_fp16)[name = tensor("op_57213_cast_fp16")]; + tensor var_57217_begin_0 = const()[name = tensor("op_57217_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_57217_end_0 = const()[name = tensor("op_57217_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_57217_end_mask_0 = const()[name = tensor("op_57217_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_57217_cast_fp16 = slice_by_index(begin = var_57217_begin_0, end = var_57217_end_0, end_mask = var_57217_end_mask_0, x = k_507_cast_fp16)[name = tensor("op_57217_cast_fp16")]; + tensor var_57221_begin_0 = const()[name = tensor("op_57221_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_57221_end_0 = const()[name = tensor("op_57221_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_57221_end_mask_0 = const()[name = tensor("op_57221_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_57221_cast_fp16 = slice_by_index(begin = var_57221_begin_0, end = var_57221_end_0, end_mask = var_57221_end_mask_0, x = k_507_cast_fp16)[name = tensor("op_57221_cast_fp16")]; + tensor var_57225_begin_0 = const()[name = tensor("op_57225_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_57225_end_0 = const()[name = tensor("op_57225_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_57225_end_mask_0 = const()[name = tensor("op_57225_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_57225_cast_fp16 = slice_by_index(begin = var_57225_begin_0, end = var_57225_end_0, end_mask = var_57225_end_mask_0, x = k_507_cast_fp16)[name = tensor("op_57225_cast_fp16")]; + tensor var_57229_begin_0 = const()[name = tensor("op_57229_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_57229_end_0 = const()[name = tensor("op_57229_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_57229_end_mask_0 = const()[name = tensor("op_57229_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_57229_cast_fp16 = slice_by_index(begin = var_57229_begin_0, end = var_57229_end_0, end_mask = var_57229_end_mask_0, x = k_507_cast_fp16)[name = tensor("op_57229_cast_fp16")]; + tensor var_57233_begin_0 = const()[name = tensor("op_57233_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_57233_end_0 = const()[name = tensor("op_57233_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_57233_end_mask_0 = const()[name = tensor("op_57233_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_57233_cast_fp16 = slice_by_index(begin = var_57233_begin_0, end = var_57233_end_0, end_mask = var_57233_end_mask_0, x = k_507_cast_fp16)[name = tensor("op_57233_cast_fp16")]; + tensor var_57237_begin_0 = const()[name = tensor("op_57237_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_57237_end_0 = const()[name = tensor("op_57237_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_57237_end_mask_0 = const()[name = tensor("op_57237_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_57237_cast_fp16 = slice_by_index(begin = var_57237_begin_0, end = var_57237_end_0, end_mask = var_57237_end_mask_0, x = k_507_cast_fp16)[name = tensor("op_57237_cast_fp16")]; + tensor var_57241_begin_0 = const()[name = tensor("op_57241_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_57241_end_0 = const()[name = tensor("op_57241_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_57241_end_mask_0 = const()[name = tensor("op_57241_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_57241_cast_fp16 = slice_by_index(begin = var_57241_begin_0, end = var_57241_end_0, end_mask = var_57241_end_mask_0, x = k_507_cast_fp16)[name = tensor("op_57241_cast_fp16")]; + tensor var_57245_begin_0 = const()[name = tensor("op_57245_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_57245_end_0 = const()[name = tensor("op_57245_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_57245_end_mask_0 = const()[name = tensor("op_57245_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_57245_cast_fp16 = slice_by_index(begin = var_57245_begin_0, end = var_57245_end_0, end_mask = var_57245_end_mask_0, x = k_507_cast_fp16)[name = tensor("op_57245_cast_fp16")]; + tensor var_57249_begin_0 = const()[name = tensor("op_57249_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_57249_end_0 = const()[name = tensor("op_57249_end_0"), val = tensor([2, 1024, 1, 704])]; + tensor var_57249_end_mask_0 = const()[name = tensor("op_57249_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_57249_cast_fp16 = slice_by_index(begin = var_57249_begin_0, end = var_57249_end_0, end_mask = var_57249_end_mask_0, x = k_507_cast_fp16)[name = tensor("op_57249_cast_fp16")]; + tensor var_57253_begin_0 = const()[name = tensor("op_57253_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_57253_end_0 = const()[name = tensor("op_57253_end_0"), val = tensor([2, 1024, 1, 768])]; + tensor var_57253_end_mask_0 = const()[name = tensor("op_57253_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_57253_cast_fp16 = slice_by_index(begin = var_57253_begin_0, end = var_57253_end_0, end_mask = var_57253_end_mask_0, x = k_507_cast_fp16)[name = tensor("op_57253_cast_fp16")]; + tensor var_57257_begin_0 = const()[name = tensor("op_57257_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_57257_end_0 = const()[name = tensor("op_57257_end_0"), val = tensor([2, 1024, 1, 832])]; + tensor var_57257_end_mask_0 = const()[name = tensor("op_57257_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_57257_cast_fp16 = slice_by_index(begin = var_57257_begin_0, end = var_57257_end_0, end_mask = var_57257_end_mask_0, x = k_507_cast_fp16)[name = tensor("op_57257_cast_fp16")]; + tensor var_57261_begin_0 = const()[name = tensor("op_57261_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_57261_end_0 = const()[name = tensor("op_57261_end_0"), val = tensor([2, 1024, 1, 896])]; + tensor var_57261_end_mask_0 = const()[name = tensor("op_57261_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_57261_cast_fp16 = slice_by_index(begin = var_57261_begin_0, end = var_57261_end_0, end_mask = var_57261_end_mask_0, x = k_507_cast_fp16)[name = tensor("op_57261_cast_fp16")]; + tensor var_57265_begin_0 = const()[name = tensor("op_57265_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_57265_end_0 = const()[name = tensor("op_57265_end_0"), val = tensor([2, 1024, 1, 960])]; + tensor var_57265_end_mask_0 = const()[name = tensor("op_57265_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_57265_cast_fp16 = slice_by_index(begin = var_57265_begin_0, end = var_57265_end_0, end_mask = var_57265_end_mask_0, x = k_507_cast_fp16)[name = tensor("op_57265_cast_fp16")]; + tensor var_57269_begin_0 = const()[name = tensor("op_57269_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_57269_end_0 = const()[name = tensor("op_57269_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_57269_end_mask_0 = const()[name = tensor("op_57269_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_57269_cast_fp16 = slice_by_index(begin = var_57269_begin_0, end = var_57269_end_0, end_mask = var_57269_end_mask_0, x = k_507_cast_fp16)[name = tensor("op_57269_cast_fp16")]; + tensor var_57273_begin_0 = const()[name = tensor("op_57273_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_57273_end_0 = const()[name = tensor("op_57273_end_0"), val = tensor([2, 1024, 1, 1088])]; + tensor var_57273_end_mask_0 = const()[name = tensor("op_57273_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_57273_cast_fp16 = slice_by_index(begin = var_57273_begin_0, end = var_57273_end_0, end_mask = var_57273_end_mask_0, x = k_507_cast_fp16)[name = tensor("op_57273_cast_fp16")]; + tensor var_57277_begin_0 = const()[name = tensor("op_57277_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_57277_end_0 = const()[name = tensor("op_57277_end_0"), val = tensor([2, 1024, 1, 1152])]; + tensor var_57277_end_mask_0 = const()[name = tensor("op_57277_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_57277_cast_fp16 = slice_by_index(begin = var_57277_begin_0, end = var_57277_end_0, end_mask = var_57277_end_mask_0, x = k_507_cast_fp16)[name = tensor("op_57277_cast_fp16")]; + tensor var_57281_begin_0 = const()[name = tensor("op_57281_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_57281_end_0 = const()[name = tensor("op_57281_end_0"), val = tensor([2, 1024, 1, 1216])]; + tensor var_57281_end_mask_0 = const()[name = tensor("op_57281_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_57281_cast_fp16 = slice_by_index(begin = var_57281_begin_0, end = var_57281_end_0, end_mask = var_57281_end_mask_0, x = k_507_cast_fp16)[name = tensor("op_57281_cast_fp16")]; + tensor var_57285_begin_0 = const()[name = tensor("op_57285_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_57285_end_0 = const()[name = tensor("op_57285_end_0"), val = tensor([2, 1024, 1, 1280])]; + tensor var_57285_end_mask_0 = const()[name = tensor("op_57285_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_57285_cast_fp16 = slice_by_index(begin = var_57285_begin_0, end = var_57285_end_0, end_mask = var_57285_end_mask_0, x = k_507_cast_fp16)[name = tensor("op_57285_cast_fp16")]; + tensor var_57287_begin_0 = const()[name = tensor("op_57287_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_57287_end_0 = const()[name = tensor("op_57287_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_57287_end_mask_0 = const()[name = tensor("op_57287_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57287_cast_fp16 = slice_by_index(begin = var_57287_begin_0, end = var_57287_end_0, end_mask = var_57287_end_mask_0, x = v_253_cast_fp16)[name = tensor("op_57287_cast_fp16")]; + tensor var_57291_begin_0 = const()[name = tensor("op_57291_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_57291_end_0 = const()[name = tensor("op_57291_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_57291_end_mask_0 = const()[name = tensor("op_57291_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57291_cast_fp16 = slice_by_index(begin = var_57291_begin_0, end = var_57291_end_0, end_mask = var_57291_end_mask_0, x = v_253_cast_fp16)[name = tensor("op_57291_cast_fp16")]; + tensor var_57295_begin_0 = const()[name = tensor("op_57295_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_57295_end_0 = const()[name = tensor("op_57295_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_57295_end_mask_0 = const()[name = tensor("op_57295_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57295_cast_fp16 = slice_by_index(begin = var_57295_begin_0, end = var_57295_end_0, end_mask = var_57295_end_mask_0, x = v_253_cast_fp16)[name = tensor("op_57295_cast_fp16")]; + tensor var_57299_begin_0 = const()[name = tensor("op_57299_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_57299_end_0 = const()[name = tensor("op_57299_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_57299_end_mask_0 = const()[name = tensor("op_57299_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57299_cast_fp16 = slice_by_index(begin = var_57299_begin_0, end = var_57299_end_0, end_mask = var_57299_end_mask_0, x = v_253_cast_fp16)[name = tensor("op_57299_cast_fp16")]; + tensor var_57303_begin_0 = const()[name = tensor("op_57303_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_57303_end_0 = const()[name = tensor("op_57303_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_57303_end_mask_0 = const()[name = tensor("op_57303_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57303_cast_fp16 = slice_by_index(begin = var_57303_begin_0, end = var_57303_end_0, end_mask = var_57303_end_mask_0, x = v_253_cast_fp16)[name = tensor("op_57303_cast_fp16")]; + tensor var_57307_begin_0 = const()[name = tensor("op_57307_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_57307_end_0 = const()[name = tensor("op_57307_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_57307_end_mask_0 = const()[name = tensor("op_57307_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57307_cast_fp16 = slice_by_index(begin = var_57307_begin_0, end = var_57307_end_0, end_mask = var_57307_end_mask_0, x = v_253_cast_fp16)[name = tensor("op_57307_cast_fp16")]; + tensor var_57311_begin_0 = const()[name = tensor("op_57311_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_57311_end_0 = const()[name = tensor("op_57311_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_57311_end_mask_0 = const()[name = tensor("op_57311_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57311_cast_fp16 = slice_by_index(begin = var_57311_begin_0, end = var_57311_end_0, end_mask = var_57311_end_mask_0, x = v_253_cast_fp16)[name = tensor("op_57311_cast_fp16")]; + tensor var_57315_begin_0 = const()[name = tensor("op_57315_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_57315_end_0 = const()[name = tensor("op_57315_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_57315_end_mask_0 = const()[name = tensor("op_57315_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57315_cast_fp16 = slice_by_index(begin = var_57315_begin_0, end = var_57315_end_0, end_mask = var_57315_end_mask_0, x = v_253_cast_fp16)[name = tensor("op_57315_cast_fp16")]; + tensor var_57319_begin_0 = const()[name = tensor("op_57319_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_57319_end_0 = const()[name = tensor("op_57319_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_57319_end_mask_0 = const()[name = tensor("op_57319_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57319_cast_fp16 = slice_by_index(begin = var_57319_begin_0, end = var_57319_end_0, end_mask = var_57319_end_mask_0, x = v_253_cast_fp16)[name = tensor("op_57319_cast_fp16")]; + tensor var_57323_begin_0 = const()[name = tensor("op_57323_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_57323_end_0 = const()[name = tensor("op_57323_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_57323_end_mask_0 = const()[name = tensor("op_57323_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57323_cast_fp16 = slice_by_index(begin = var_57323_begin_0, end = var_57323_end_0, end_mask = var_57323_end_mask_0, x = v_253_cast_fp16)[name = tensor("op_57323_cast_fp16")]; + tensor var_57327_begin_0 = const()[name = tensor("op_57327_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_57327_end_0 = const()[name = tensor("op_57327_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_57327_end_mask_0 = const()[name = tensor("op_57327_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57327_cast_fp16 = slice_by_index(begin = var_57327_begin_0, end = var_57327_end_0, end_mask = var_57327_end_mask_0, x = v_253_cast_fp16)[name = tensor("op_57327_cast_fp16")]; + tensor var_57331_begin_0 = const()[name = tensor("op_57331_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_57331_end_0 = const()[name = tensor("op_57331_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_57331_end_mask_0 = const()[name = tensor("op_57331_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57331_cast_fp16 = slice_by_index(begin = var_57331_begin_0, end = var_57331_end_0, end_mask = var_57331_end_mask_0, x = v_253_cast_fp16)[name = tensor("op_57331_cast_fp16")]; + tensor var_57335_begin_0 = const()[name = tensor("op_57335_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_57335_end_0 = const()[name = tensor("op_57335_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_57335_end_mask_0 = const()[name = tensor("op_57335_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57335_cast_fp16 = slice_by_index(begin = var_57335_begin_0, end = var_57335_end_0, end_mask = var_57335_end_mask_0, x = v_253_cast_fp16)[name = tensor("op_57335_cast_fp16")]; + tensor var_57339_begin_0 = const()[name = tensor("op_57339_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_57339_end_0 = const()[name = tensor("op_57339_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_57339_end_mask_0 = const()[name = tensor("op_57339_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57339_cast_fp16 = slice_by_index(begin = var_57339_begin_0, end = var_57339_end_0, end_mask = var_57339_end_mask_0, x = v_253_cast_fp16)[name = tensor("op_57339_cast_fp16")]; + tensor var_57343_begin_0 = const()[name = tensor("op_57343_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_57343_end_0 = const()[name = tensor("op_57343_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_57343_end_mask_0 = const()[name = tensor("op_57343_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57343_cast_fp16 = slice_by_index(begin = var_57343_begin_0, end = var_57343_end_0, end_mask = var_57343_end_mask_0, x = v_253_cast_fp16)[name = tensor("op_57343_cast_fp16")]; + tensor var_57347_begin_0 = const()[name = tensor("op_57347_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_57347_end_0 = const()[name = tensor("op_57347_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_57347_end_mask_0 = const()[name = tensor("op_57347_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57347_cast_fp16 = slice_by_index(begin = var_57347_begin_0, end = var_57347_end_0, end_mask = var_57347_end_mask_0, x = v_253_cast_fp16)[name = tensor("op_57347_cast_fp16")]; + tensor var_57351_begin_0 = const()[name = tensor("op_57351_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_57351_end_0 = const()[name = tensor("op_57351_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_57351_end_mask_0 = const()[name = tensor("op_57351_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57351_cast_fp16 = slice_by_index(begin = var_57351_begin_0, end = var_57351_end_0, end_mask = var_57351_end_mask_0, x = v_253_cast_fp16)[name = tensor("op_57351_cast_fp16")]; + tensor var_57355_begin_0 = const()[name = tensor("op_57355_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_57355_end_0 = const()[name = tensor("op_57355_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_57355_end_mask_0 = const()[name = tensor("op_57355_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57355_cast_fp16 = slice_by_index(begin = var_57355_begin_0, end = var_57355_end_0, end_mask = var_57355_end_mask_0, x = v_253_cast_fp16)[name = tensor("op_57355_cast_fp16")]; + tensor var_57359_begin_0 = const()[name = tensor("op_57359_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_57359_end_0 = const()[name = tensor("op_57359_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_57359_end_mask_0 = const()[name = tensor("op_57359_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57359_cast_fp16 = slice_by_index(begin = var_57359_begin_0, end = var_57359_end_0, end_mask = var_57359_end_mask_0, x = v_253_cast_fp16)[name = tensor("op_57359_cast_fp16")]; + tensor var_57363_begin_0 = const()[name = tensor("op_57363_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_57363_end_0 = const()[name = tensor("op_57363_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_57363_end_mask_0 = const()[name = tensor("op_57363_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57363_cast_fp16 = slice_by_index(begin = var_57363_begin_0, end = var_57363_end_0, end_mask = var_57363_end_mask_0, x = v_253_cast_fp16)[name = tensor("op_57363_cast_fp16")]; + tensor var_57367_equation_0 = const()[name = tensor("op_57367_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_57367_cast_fp16 = einsum(equation = var_57367_equation_0, values = (var_57209_cast_fp16, var_57126_cast_fp16))[name = tensor("op_57367_cast_fp16")]; + tensor var_57368_to_fp16 = const()[name = tensor("op_57368_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4881_cast_fp16 = mul(x = var_57367_cast_fp16, y = var_57368_to_fp16)[name = tensor("aw_4881_cast_fp16")]; + tensor var_57371_equation_0 = const()[name = tensor("op_57371_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_57371_cast_fp16 = einsum(equation = var_57371_equation_0, values = (var_57213_cast_fp16, var_57130_cast_fp16))[name = tensor("op_57371_cast_fp16")]; + tensor var_57372_to_fp16 = const()[name = tensor("op_57372_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4883_cast_fp16 = mul(x = var_57371_cast_fp16, y = var_57372_to_fp16)[name = tensor("aw_4883_cast_fp16")]; + tensor var_57375_equation_0 = const()[name = tensor("op_57375_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_57375_cast_fp16 = einsum(equation = var_57375_equation_0, values = (var_57217_cast_fp16, var_57134_cast_fp16))[name = tensor("op_57375_cast_fp16")]; + tensor var_57376_to_fp16 = const()[name = tensor("op_57376_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4885_cast_fp16 = mul(x = var_57375_cast_fp16, y = var_57376_to_fp16)[name = tensor("aw_4885_cast_fp16")]; + tensor var_57379_equation_0 = const()[name = tensor("op_57379_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_57379_cast_fp16 = einsum(equation = var_57379_equation_0, values = (var_57221_cast_fp16, var_57138_cast_fp16))[name = tensor("op_57379_cast_fp16")]; + tensor var_57380_to_fp16 = const()[name = tensor("op_57380_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4887_cast_fp16 = mul(x = var_57379_cast_fp16, y = var_57380_to_fp16)[name = tensor("aw_4887_cast_fp16")]; + tensor var_57383_equation_0 = const()[name = tensor("op_57383_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_57383_cast_fp16 = einsum(equation = var_57383_equation_0, values = (var_57225_cast_fp16, var_57142_cast_fp16))[name = tensor("op_57383_cast_fp16")]; + tensor var_57384_to_fp16 = const()[name = tensor("op_57384_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4889_cast_fp16 = mul(x = var_57383_cast_fp16, y = var_57384_to_fp16)[name = tensor("aw_4889_cast_fp16")]; + tensor var_57387_equation_0 = const()[name = tensor("op_57387_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_57387_cast_fp16 = einsum(equation = var_57387_equation_0, values = (var_57229_cast_fp16, var_57146_cast_fp16))[name = tensor("op_57387_cast_fp16")]; + tensor var_57388_to_fp16 = const()[name = tensor("op_57388_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4891_cast_fp16 = mul(x = var_57387_cast_fp16, y = var_57388_to_fp16)[name = tensor("aw_4891_cast_fp16")]; + tensor var_57391_equation_0 = const()[name = tensor("op_57391_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_57391_cast_fp16 = einsum(equation = var_57391_equation_0, values = (var_57233_cast_fp16, var_57150_cast_fp16))[name = tensor("op_57391_cast_fp16")]; + tensor var_57392_to_fp16 = const()[name = tensor("op_57392_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4893_cast_fp16 = mul(x = var_57391_cast_fp16, y = var_57392_to_fp16)[name = tensor("aw_4893_cast_fp16")]; + tensor var_57395_equation_0 = const()[name = tensor("op_57395_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_57395_cast_fp16 = einsum(equation = var_57395_equation_0, values = (var_57237_cast_fp16, var_57154_cast_fp16))[name = tensor("op_57395_cast_fp16")]; + tensor var_57396_to_fp16 = const()[name = tensor("op_57396_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4895_cast_fp16 = mul(x = var_57395_cast_fp16, y = var_57396_to_fp16)[name = tensor("aw_4895_cast_fp16")]; + tensor var_57399_equation_0 = const()[name = tensor("op_57399_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_57399_cast_fp16 = einsum(equation = var_57399_equation_0, values = (var_57241_cast_fp16, var_57158_cast_fp16))[name = tensor("op_57399_cast_fp16")]; + tensor var_57400_to_fp16 = const()[name = tensor("op_57400_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4897_cast_fp16 = mul(x = var_57399_cast_fp16, y = var_57400_to_fp16)[name = tensor("aw_4897_cast_fp16")]; + tensor var_57403_equation_0 = const()[name = tensor("op_57403_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_57403_cast_fp16 = einsum(equation = var_57403_equation_0, values = (var_57245_cast_fp16, var_57162_cast_fp16))[name = tensor("op_57403_cast_fp16")]; + tensor var_57404_to_fp16 = const()[name = tensor("op_57404_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4899_cast_fp16 = mul(x = var_57403_cast_fp16, y = var_57404_to_fp16)[name = tensor("aw_4899_cast_fp16")]; + tensor var_57407_equation_0 = const()[name = tensor("op_57407_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_57407_cast_fp16 = einsum(equation = var_57407_equation_0, values = (var_57249_cast_fp16, var_57166_cast_fp16))[name = tensor("op_57407_cast_fp16")]; + tensor var_57408_to_fp16 = const()[name = tensor("op_57408_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4901_cast_fp16 = mul(x = var_57407_cast_fp16, y = var_57408_to_fp16)[name = tensor("aw_4901_cast_fp16")]; + tensor var_57411_equation_0 = const()[name = tensor("op_57411_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_57411_cast_fp16 = einsum(equation = var_57411_equation_0, values = (var_57253_cast_fp16, var_57170_cast_fp16))[name = tensor("op_57411_cast_fp16")]; + tensor var_57412_to_fp16 = const()[name = tensor("op_57412_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4903_cast_fp16 = mul(x = var_57411_cast_fp16, y = var_57412_to_fp16)[name = tensor("aw_4903_cast_fp16")]; + tensor var_57415_equation_0 = const()[name = tensor("op_57415_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_57415_cast_fp16 = einsum(equation = var_57415_equation_0, values = (var_57257_cast_fp16, var_57174_cast_fp16))[name = tensor("op_57415_cast_fp16")]; + tensor var_57416_to_fp16 = const()[name = tensor("op_57416_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4905_cast_fp16 = mul(x = var_57415_cast_fp16, y = var_57416_to_fp16)[name = tensor("aw_4905_cast_fp16")]; + tensor var_57419_equation_0 = const()[name = tensor("op_57419_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_57419_cast_fp16 = einsum(equation = var_57419_equation_0, values = (var_57261_cast_fp16, var_57178_cast_fp16))[name = tensor("op_57419_cast_fp16")]; + tensor var_57420_to_fp16 = const()[name = tensor("op_57420_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4907_cast_fp16 = mul(x = var_57419_cast_fp16, y = var_57420_to_fp16)[name = tensor("aw_4907_cast_fp16")]; + tensor var_57423_equation_0 = const()[name = tensor("op_57423_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_57423_cast_fp16 = einsum(equation = var_57423_equation_0, values = (var_57265_cast_fp16, var_57182_cast_fp16))[name = tensor("op_57423_cast_fp16")]; + tensor var_57424_to_fp16 = const()[name = tensor("op_57424_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4909_cast_fp16 = mul(x = var_57423_cast_fp16, y = var_57424_to_fp16)[name = tensor("aw_4909_cast_fp16")]; + tensor var_57427_equation_0 = const()[name = tensor("op_57427_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_57427_cast_fp16 = einsum(equation = var_57427_equation_0, values = (var_57269_cast_fp16, var_57186_cast_fp16))[name = tensor("op_57427_cast_fp16")]; + tensor var_57428_to_fp16 = const()[name = tensor("op_57428_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4911_cast_fp16 = mul(x = var_57427_cast_fp16, y = var_57428_to_fp16)[name = tensor("aw_4911_cast_fp16")]; + tensor var_57431_equation_0 = const()[name = tensor("op_57431_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_57431_cast_fp16 = einsum(equation = var_57431_equation_0, values = (var_57273_cast_fp16, var_57190_cast_fp16))[name = tensor("op_57431_cast_fp16")]; + tensor var_57432_to_fp16 = const()[name = tensor("op_57432_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4913_cast_fp16 = mul(x = var_57431_cast_fp16, y = var_57432_to_fp16)[name = tensor("aw_4913_cast_fp16")]; + tensor var_57435_equation_0 = const()[name = tensor("op_57435_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_57435_cast_fp16 = einsum(equation = var_57435_equation_0, values = (var_57277_cast_fp16, var_57194_cast_fp16))[name = tensor("op_57435_cast_fp16")]; + tensor var_57436_to_fp16 = const()[name = tensor("op_57436_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4915_cast_fp16 = mul(x = var_57435_cast_fp16, y = var_57436_to_fp16)[name = tensor("aw_4915_cast_fp16")]; + tensor var_57439_equation_0 = const()[name = tensor("op_57439_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_57439_cast_fp16 = einsum(equation = var_57439_equation_0, values = (var_57281_cast_fp16, var_57198_cast_fp16))[name = tensor("op_57439_cast_fp16")]; + tensor var_57440_to_fp16 = const()[name = tensor("op_57440_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4917_cast_fp16 = mul(x = var_57439_cast_fp16, y = var_57440_to_fp16)[name = tensor("aw_4917_cast_fp16")]; + tensor var_57443_equation_0 = const()[name = tensor("op_57443_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_57443_cast_fp16 = einsum(equation = var_57443_equation_0, values = (var_57285_cast_fp16, var_57202_cast_fp16))[name = tensor("op_57443_cast_fp16")]; + tensor var_57444_to_fp16 = const()[name = tensor("op_57444_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4919_cast_fp16 = mul(x = var_57443_cast_fp16, y = var_57444_to_fp16)[name = tensor("aw_4919_cast_fp16")]; + tensor var_57446_cast_fp16 = softmax(axis = var_30355, x = aw_4881_cast_fp16)[name = tensor("op_57446_cast_fp16")]; + tensor var_57447_cast_fp16 = softmax(axis = var_30355, x = aw_4883_cast_fp16)[name = tensor("op_57447_cast_fp16")]; + tensor var_57448_cast_fp16 = softmax(axis = var_30355, x = aw_4885_cast_fp16)[name = tensor("op_57448_cast_fp16")]; + tensor var_57449_cast_fp16 = softmax(axis = var_30355, x = aw_4887_cast_fp16)[name = tensor("op_57449_cast_fp16")]; + tensor var_57450_cast_fp16 = softmax(axis = var_30355, x = aw_4889_cast_fp16)[name = tensor("op_57450_cast_fp16")]; + tensor var_57451_cast_fp16 = softmax(axis = var_30355, x = aw_4891_cast_fp16)[name = tensor("op_57451_cast_fp16")]; + tensor var_57452_cast_fp16 = softmax(axis = var_30355, x = aw_4893_cast_fp16)[name = tensor("op_57452_cast_fp16")]; + tensor var_57453_cast_fp16 = softmax(axis = var_30355, x = aw_4895_cast_fp16)[name = tensor("op_57453_cast_fp16")]; + tensor var_57454_cast_fp16 = softmax(axis = var_30355, x = aw_4897_cast_fp16)[name = tensor("op_57454_cast_fp16")]; + tensor var_57455_cast_fp16 = softmax(axis = var_30355, x = aw_4899_cast_fp16)[name = tensor("op_57455_cast_fp16")]; + tensor var_57456_cast_fp16 = softmax(axis = var_30355, x = aw_4901_cast_fp16)[name = tensor("op_57456_cast_fp16")]; + tensor var_57457_cast_fp16 = softmax(axis = var_30355, x = aw_4903_cast_fp16)[name = tensor("op_57457_cast_fp16")]; + tensor var_57458_cast_fp16 = softmax(axis = var_30355, x = aw_4905_cast_fp16)[name = tensor("op_57458_cast_fp16")]; + tensor var_57459_cast_fp16 = softmax(axis = var_30355, x = aw_4907_cast_fp16)[name = tensor("op_57459_cast_fp16")]; + tensor var_57460_cast_fp16 = softmax(axis = var_30355, x = aw_4909_cast_fp16)[name = tensor("op_57460_cast_fp16")]; + tensor var_57461_cast_fp16 = softmax(axis = var_30355, x = aw_4911_cast_fp16)[name = tensor("op_57461_cast_fp16")]; + tensor var_57462_cast_fp16 = softmax(axis = var_30355, x = aw_4913_cast_fp16)[name = tensor("op_57462_cast_fp16")]; + tensor var_57463_cast_fp16 = softmax(axis = var_30355, x = aw_4915_cast_fp16)[name = tensor("op_57463_cast_fp16")]; + tensor var_57464_cast_fp16 = softmax(axis = var_30355, x = aw_4917_cast_fp16)[name = tensor("op_57464_cast_fp16")]; + tensor var_57465_cast_fp16 = softmax(axis = var_30355, x = aw_4919_cast_fp16)[name = tensor("op_57465_cast_fp16")]; + tensor var_57467_equation_0 = const()[name = tensor("op_57467_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_57467_cast_fp16 = einsum(equation = var_57467_equation_0, values = (var_57287_cast_fp16, var_57446_cast_fp16))[name = tensor("op_57467_cast_fp16")]; + tensor var_57469_equation_0 = const()[name = tensor("op_57469_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_57469_cast_fp16 = einsum(equation = var_57469_equation_0, values = (var_57291_cast_fp16, var_57447_cast_fp16))[name = tensor("op_57469_cast_fp16")]; + tensor var_57471_equation_0 = const()[name = tensor("op_57471_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_57471_cast_fp16 = einsum(equation = var_57471_equation_0, values = (var_57295_cast_fp16, var_57448_cast_fp16))[name = tensor("op_57471_cast_fp16")]; + tensor var_57473_equation_0 = const()[name = tensor("op_57473_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_57473_cast_fp16 = einsum(equation = var_57473_equation_0, values = (var_57299_cast_fp16, var_57449_cast_fp16))[name = tensor("op_57473_cast_fp16")]; + tensor var_57475_equation_0 = const()[name = tensor("op_57475_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_57475_cast_fp16 = einsum(equation = var_57475_equation_0, values = (var_57303_cast_fp16, var_57450_cast_fp16))[name = tensor("op_57475_cast_fp16")]; + tensor var_57477_equation_0 = const()[name = tensor("op_57477_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_57477_cast_fp16 = einsum(equation = var_57477_equation_0, values = (var_57307_cast_fp16, var_57451_cast_fp16))[name = tensor("op_57477_cast_fp16")]; + tensor var_57479_equation_0 = const()[name = tensor("op_57479_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_57479_cast_fp16 = einsum(equation = var_57479_equation_0, values = (var_57311_cast_fp16, var_57452_cast_fp16))[name = tensor("op_57479_cast_fp16")]; + tensor var_57481_equation_0 = const()[name = tensor("op_57481_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_57481_cast_fp16 = einsum(equation = var_57481_equation_0, values = (var_57315_cast_fp16, var_57453_cast_fp16))[name = tensor("op_57481_cast_fp16")]; + tensor var_57483_equation_0 = const()[name = tensor("op_57483_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_57483_cast_fp16 = einsum(equation = var_57483_equation_0, values = (var_57319_cast_fp16, var_57454_cast_fp16))[name = tensor("op_57483_cast_fp16")]; + tensor var_57485_equation_0 = const()[name = tensor("op_57485_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_57485_cast_fp16 = einsum(equation = var_57485_equation_0, values = (var_57323_cast_fp16, var_57455_cast_fp16))[name = tensor("op_57485_cast_fp16")]; + tensor var_57487_equation_0 = const()[name = tensor("op_57487_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_57487_cast_fp16 = einsum(equation = var_57487_equation_0, values = (var_57327_cast_fp16, var_57456_cast_fp16))[name = tensor("op_57487_cast_fp16")]; + tensor var_57489_equation_0 = const()[name = tensor("op_57489_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_57489_cast_fp16 = einsum(equation = var_57489_equation_0, values = (var_57331_cast_fp16, var_57457_cast_fp16))[name = tensor("op_57489_cast_fp16")]; + tensor var_57491_equation_0 = const()[name = tensor("op_57491_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_57491_cast_fp16 = einsum(equation = var_57491_equation_0, values = (var_57335_cast_fp16, var_57458_cast_fp16))[name = tensor("op_57491_cast_fp16")]; + tensor var_57493_equation_0 = const()[name = tensor("op_57493_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_57493_cast_fp16 = einsum(equation = var_57493_equation_0, values = (var_57339_cast_fp16, var_57459_cast_fp16))[name = tensor("op_57493_cast_fp16")]; + tensor var_57495_equation_0 = const()[name = tensor("op_57495_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_57495_cast_fp16 = einsum(equation = var_57495_equation_0, values = (var_57343_cast_fp16, var_57460_cast_fp16))[name = tensor("op_57495_cast_fp16")]; + tensor var_57497_equation_0 = const()[name = tensor("op_57497_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_57497_cast_fp16 = einsum(equation = var_57497_equation_0, values = (var_57347_cast_fp16, var_57461_cast_fp16))[name = tensor("op_57497_cast_fp16")]; + tensor var_57499_equation_0 = const()[name = tensor("op_57499_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_57499_cast_fp16 = einsum(equation = var_57499_equation_0, values = (var_57351_cast_fp16, var_57462_cast_fp16))[name = tensor("op_57499_cast_fp16")]; + tensor var_57501_equation_0 = const()[name = tensor("op_57501_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_57501_cast_fp16 = einsum(equation = var_57501_equation_0, values = (var_57355_cast_fp16, var_57463_cast_fp16))[name = tensor("op_57501_cast_fp16")]; + tensor var_57503_equation_0 = const()[name = tensor("op_57503_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_57503_cast_fp16 = einsum(equation = var_57503_equation_0, values = (var_57359_cast_fp16, var_57464_cast_fp16))[name = tensor("op_57503_cast_fp16")]; + tensor var_57505_equation_0 = const()[name = tensor("op_57505_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_57505_cast_fp16 = einsum(equation = var_57505_equation_0, values = (var_57363_cast_fp16, var_57465_cast_fp16))[name = tensor("op_57505_cast_fp16")]; + tensor input_707_interleave_0 = const()[name = tensor("input_707_interleave_0"), val = tensor(false)]; + tensor input_707_cast_fp16 = concat(axis = var_30355, interleave = input_707_interleave_0, values = (var_57467_cast_fp16, var_57469_cast_fp16, var_57471_cast_fp16, var_57473_cast_fp16, var_57475_cast_fp16, var_57477_cast_fp16, var_57479_cast_fp16, var_57481_cast_fp16, var_57483_cast_fp16, var_57485_cast_fp16, var_57487_cast_fp16, var_57489_cast_fp16, var_57491_cast_fp16, var_57493_cast_fp16, var_57495_cast_fp16, var_57497_cast_fp16, var_57499_cast_fp16, var_57501_cast_fp16, var_57503_cast_fp16, var_57505_cast_fp16))[name = tensor("input_707_cast_fp16")]; + tensor var_57511 = const()[name = tensor("op_57511"), val = tensor([1, 1])]; + tensor var_57513 = const()[name = tensor("op_57513"), val = tensor([1, 1])]; + tensor var_57515_pad_type_0 = const()[name = tensor("op_57515_pad_type_0"), val = tensor("custom")]; + tensor var_57515_pad_0 = const()[name = tensor("op_57515_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_9_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_9_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4807543488)))]; + tensor up_blocks_0_attentions_2_transformer_blocks_9_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_9_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4810820352)))]; + tensor var_57515_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_9_attn1_to_out_0_bias_to_fp16, dilations = var_57513, groups = var_30355, pad = var_57515_pad_0, pad_type = var_57515_pad_type_0, strides = var_57511, weight = up_blocks_0_attentions_2_transformer_blocks_9_attn1_to_out_0_weight_to_fp16, x = input_707_cast_fp16)[name = tensor("op_57515_cast_fp16")]; + tensor inputs_381_cast_fp16 = add(x = var_57515_cast_fp16, y = inputs_379_cast_fp16)[name = tensor("inputs_381_cast_fp16")]; + tensor hidden_states_491_axes_0 = const()[name = tensor("hidden_states_491_axes_0"), val = tensor([1])]; + tensor hidden_states_491_gamma_0_to_fp16 = const()[name = tensor("hidden_states_491_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4810822976)))]; + tensor hidden_states_491_beta_0_to_fp16 = const()[name = tensor("hidden_states_491_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4810825600)))]; + tensor var_57525_to_fp16 = const()[name = tensor("op_57525_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_491_cast_fp16 = layer_norm(axes = hidden_states_491_axes_0, beta = hidden_states_491_beta_0_to_fp16, epsilon = var_57525_to_fp16, gamma = hidden_states_491_gamma_0_to_fp16, x = inputs_381_cast_fp16)[name = tensor("hidden_states_491_cast_fp16")]; + tensor var_57540 = const()[name = tensor("op_57540"), val = tensor([1, 1])]; + tensor var_57542 = const()[name = tensor("op_57542"), val = tensor([1, 1])]; + tensor q_255_pad_type_0 = const()[name = tensor("q_255_pad_type_0"), val = tensor("custom")]; + tensor q_255_pad_0 = const()[name = tensor("q_255_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_9_attn2_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_9_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4810828224)))]; + tensor q_255_cast_fp16 = conv(dilations = var_57542, groups = var_30355, pad = q_255_pad_0, pad_type = q_255_pad_type_0, strides = var_57540, weight = up_blocks_0_attentions_2_transformer_blocks_9_attn2_to_q_weight_to_fp16, x = hidden_states_491_cast_fp16)[name = tensor("q_255_cast_fp16")]; + tensor var_57546 = const()[name = tensor("op_57546"), val = tensor([1, 1])]; + tensor var_57548 = const()[name = tensor("op_57548"), val = tensor([1, 1])]; + tensor k_509_pad_type_0 = const()[name = tensor("k_509_pad_type_0"), val = tensor("custom")]; + tensor k_509_pad_0 = const()[name = tensor("k_509_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_9_attn2_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_9_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4814105088)))]; + tensor k_509_cast_fp16 = conv(dilations = var_57548, groups = var_30355, pad = k_509_pad_0, pad_type = k_509_pad_type_0, strides = var_57546, weight = up_blocks_0_attentions_2_transformer_blocks_9_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_509_cast_fp16")]; + tensor var_57552 = const()[name = tensor("op_57552"), val = tensor([1, 1])]; + tensor var_57554 = const()[name = tensor("op_57554"), val = tensor([1, 1])]; + tensor v_255_pad_type_0 = const()[name = tensor("v_255_pad_type_0"), val = tensor("custom")]; + tensor v_255_pad_0 = const()[name = tensor("v_255_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_9_attn2_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_9_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4819348032)))]; + tensor v_255_cast_fp16 = conv(dilations = var_57554, groups = var_30355, pad = v_255_pad_0, pad_type = v_255_pad_type_0, strides = var_57552, weight = up_blocks_0_attentions_2_transformer_blocks_9_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_255_cast_fp16")]; + tensor var_57558_begin_0 = const()[name = tensor("op_57558_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_57558_end_0 = const()[name = tensor("op_57558_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_57558_end_mask_0 = const()[name = tensor("op_57558_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57558_cast_fp16 = slice_by_index(begin = var_57558_begin_0, end = var_57558_end_0, end_mask = var_57558_end_mask_0, x = q_255_cast_fp16)[name = tensor("op_57558_cast_fp16")]; + tensor var_57562_begin_0 = const()[name = tensor("op_57562_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_57562_end_0 = const()[name = tensor("op_57562_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_57562_end_mask_0 = const()[name = tensor("op_57562_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57562_cast_fp16 = slice_by_index(begin = var_57562_begin_0, end = var_57562_end_0, end_mask = var_57562_end_mask_0, x = q_255_cast_fp16)[name = tensor("op_57562_cast_fp16")]; + tensor var_57566_begin_0 = const()[name = tensor("op_57566_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_57566_end_0 = const()[name = tensor("op_57566_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_57566_end_mask_0 = const()[name = tensor("op_57566_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57566_cast_fp16 = slice_by_index(begin = var_57566_begin_0, end = var_57566_end_0, end_mask = var_57566_end_mask_0, x = q_255_cast_fp16)[name = tensor("op_57566_cast_fp16")]; + tensor var_57570_begin_0 = const()[name = tensor("op_57570_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_57570_end_0 = const()[name = tensor("op_57570_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_57570_end_mask_0 = const()[name = tensor("op_57570_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57570_cast_fp16 = slice_by_index(begin = var_57570_begin_0, end = var_57570_end_0, end_mask = var_57570_end_mask_0, x = q_255_cast_fp16)[name = tensor("op_57570_cast_fp16")]; + tensor var_57574_begin_0 = const()[name = tensor("op_57574_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_57574_end_0 = const()[name = tensor("op_57574_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_57574_end_mask_0 = const()[name = tensor("op_57574_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57574_cast_fp16 = slice_by_index(begin = var_57574_begin_0, end = var_57574_end_0, end_mask = var_57574_end_mask_0, x = q_255_cast_fp16)[name = tensor("op_57574_cast_fp16")]; + tensor var_57578_begin_0 = const()[name = tensor("op_57578_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_57578_end_0 = const()[name = tensor("op_57578_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_57578_end_mask_0 = const()[name = tensor("op_57578_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57578_cast_fp16 = slice_by_index(begin = var_57578_begin_0, end = var_57578_end_0, end_mask = var_57578_end_mask_0, x = q_255_cast_fp16)[name = tensor("op_57578_cast_fp16")]; + tensor var_57582_begin_0 = const()[name = tensor("op_57582_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_57582_end_0 = const()[name = tensor("op_57582_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_57582_end_mask_0 = const()[name = tensor("op_57582_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57582_cast_fp16 = slice_by_index(begin = var_57582_begin_0, end = var_57582_end_0, end_mask = var_57582_end_mask_0, x = q_255_cast_fp16)[name = tensor("op_57582_cast_fp16")]; + tensor var_57586_begin_0 = const()[name = tensor("op_57586_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_57586_end_0 = const()[name = tensor("op_57586_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_57586_end_mask_0 = const()[name = tensor("op_57586_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57586_cast_fp16 = slice_by_index(begin = var_57586_begin_0, end = var_57586_end_0, end_mask = var_57586_end_mask_0, x = q_255_cast_fp16)[name = tensor("op_57586_cast_fp16")]; + tensor var_57590_begin_0 = const()[name = tensor("op_57590_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_57590_end_0 = const()[name = tensor("op_57590_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_57590_end_mask_0 = const()[name = tensor("op_57590_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57590_cast_fp16 = slice_by_index(begin = var_57590_begin_0, end = var_57590_end_0, end_mask = var_57590_end_mask_0, x = q_255_cast_fp16)[name = tensor("op_57590_cast_fp16")]; + tensor var_57594_begin_0 = const()[name = tensor("op_57594_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_57594_end_0 = const()[name = tensor("op_57594_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_57594_end_mask_0 = const()[name = tensor("op_57594_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57594_cast_fp16 = slice_by_index(begin = var_57594_begin_0, end = var_57594_end_0, end_mask = var_57594_end_mask_0, x = q_255_cast_fp16)[name = tensor("op_57594_cast_fp16")]; + tensor var_57598_begin_0 = const()[name = tensor("op_57598_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_57598_end_0 = const()[name = tensor("op_57598_end_0"), val = tensor([2, 704, 1, 1024])]; + tensor var_57598_end_mask_0 = const()[name = tensor("op_57598_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57598_cast_fp16 = slice_by_index(begin = var_57598_begin_0, end = var_57598_end_0, end_mask = var_57598_end_mask_0, x = q_255_cast_fp16)[name = tensor("op_57598_cast_fp16")]; + tensor var_57602_begin_0 = const()[name = tensor("op_57602_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_57602_end_0 = const()[name = tensor("op_57602_end_0"), val = tensor([2, 768, 1, 1024])]; + tensor var_57602_end_mask_0 = const()[name = tensor("op_57602_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57602_cast_fp16 = slice_by_index(begin = var_57602_begin_0, end = var_57602_end_0, end_mask = var_57602_end_mask_0, x = q_255_cast_fp16)[name = tensor("op_57602_cast_fp16")]; + tensor var_57606_begin_0 = const()[name = tensor("op_57606_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_57606_end_0 = const()[name = tensor("op_57606_end_0"), val = tensor([2, 832, 1, 1024])]; + tensor var_57606_end_mask_0 = const()[name = tensor("op_57606_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57606_cast_fp16 = slice_by_index(begin = var_57606_begin_0, end = var_57606_end_0, end_mask = var_57606_end_mask_0, x = q_255_cast_fp16)[name = tensor("op_57606_cast_fp16")]; + tensor var_57610_begin_0 = const()[name = tensor("op_57610_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_57610_end_0 = const()[name = tensor("op_57610_end_0"), val = tensor([2, 896, 1, 1024])]; + tensor var_57610_end_mask_0 = const()[name = tensor("op_57610_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57610_cast_fp16 = slice_by_index(begin = var_57610_begin_0, end = var_57610_end_0, end_mask = var_57610_end_mask_0, x = q_255_cast_fp16)[name = tensor("op_57610_cast_fp16")]; + tensor var_57614_begin_0 = const()[name = tensor("op_57614_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_57614_end_0 = const()[name = tensor("op_57614_end_0"), val = tensor([2, 960, 1, 1024])]; + tensor var_57614_end_mask_0 = const()[name = tensor("op_57614_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57614_cast_fp16 = slice_by_index(begin = var_57614_begin_0, end = var_57614_end_0, end_mask = var_57614_end_mask_0, x = q_255_cast_fp16)[name = tensor("op_57614_cast_fp16")]; + tensor var_57618_begin_0 = const()[name = tensor("op_57618_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_57618_end_0 = const()[name = tensor("op_57618_end_0"), val = tensor([2, 1024, 1, 1024])]; + tensor var_57618_end_mask_0 = const()[name = tensor("op_57618_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57618_cast_fp16 = slice_by_index(begin = var_57618_begin_0, end = var_57618_end_0, end_mask = var_57618_end_mask_0, x = q_255_cast_fp16)[name = tensor("op_57618_cast_fp16")]; + tensor var_57622_begin_0 = const()[name = tensor("op_57622_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_57622_end_0 = const()[name = tensor("op_57622_end_0"), val = tensor([2, 1088, 1, 1024])]; + tensor var_57622_end_mask_0 = const()[name = tensor("op_57622_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57622_cast_fp16 = slice_by_index(begin = var_57622_begin_0, end = var_57622_end_0, end_mask = var_57622_end_mask_0, x = q_255_cast_fp16)[name = tensor("op_57622_cast_fp16")]; + tensor var_57626_begin_0 = const()[name = tensor("op_57626_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_57626_end_0 = const()[name = tensor("op_57626_end_0"), val = tensor([2, 1152, 1, 1024])]; + tensor var_57626_end_mask_0 = const()[name = tensor("op_57626_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57626_cast_fp16 = slice_by_index(begin = var_57626_begin_0, end = var_57626_end_0, end_mask = var_57626_end_mask_0, x = q_255_cast_fp16)[name = tensor("op_57626_cast_fp16")]; + tensor var_57630_begin_0 = const()[name = tensor("op_57630_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_57630_end_0 = const()[name = tensor("op_57630_end_0"), val = tensor([2, 1216, 1, 1024])]; + tensor var_57630_end_mask_0 = const()[name = tensor("op_57630_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57630_cast_fp16 = slice_by_index(begin = var_57630_begin_0, end = var_57630_end_0, end_mask = var_57630_end_mask_0, x = q_255_cast_fp16)[name = tensor("op_57630_cast_fp16")]; + tensor var_57634_begin_0 = const()[name = tensor("op_57634_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_57634_end_0 = const()[name = tensor("op_57634_end_0"), val = tensor([2, 1280, 1, 1024])]; + tensor var_57634_end_mask_0 = const()[name = tensor("op_57634_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57634_cast_fp16 = slice_by_index(begin = var_57634_begin_0, end = var_57634_end_0, end_mask = var_57634_end_mask_0, x = q_255_cast_fp16)[name = tensor("op_57634_cast_fp16")]; + tensor k_511_perm_0 = const()[name = tensor("k_511_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_57641_begin_0 = const()[name = tensor("op_57641_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_57641_end_0 = const()[name = tensor("op_57641_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_57641_end_mask_0 = const()[name = tensor("op_57641_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_511_cast_fp16 = transpose(perm = k_511_perm_0, x = k_509_cast_fp16)[name = tensor("transpose_12")]; + tensor var_57641_cast_fp16 = slice_by_index(begin = var_57641_begin_0, end = var_57641_end_0, end_mask = var_57641_end_mask_0, x = k_511_cast_fp16)[name = tensor("op_57641_cast_fp16")]; + tensor var_57645_begin_0 = const()[name = tensor("op_57645_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_57645_end_0 = const()[name = tensor("op_57645_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_57645_end_mask_0 = const()[name = tensor("op_57645_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_57645_cast_fp16 = slice_by_index(begin = var_57645_begin_0, end = var_57645_end_0, end_mask = var_57645_end_mask_0, x = k_511_cast_fp16)[name = tensor("op_57645_cast_fp16")]; + tensor var_57649_begin_0 = const()[name = tensor("op_57649_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_57649_end_0 = const()[name = tensor("op_57649_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_57649_end_mask_0 = const()[name = tensor("op_57649_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_57649_cast_fp16 = slice_by_index(begin = var_57649_begin_0, end = var_57649_end_0, end_mask = var_57649_end_mask_0, x = k_511_cast_fp16)[name = tensor("op_57649_cast_fp16")]; + tensor var_57653_begin_0 = const()[name = tensor("op_57653_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_57653_end_0 = const()[name = tensor("op_57653_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_57653_end_mask_0 = const()[name = tensor("op_57653_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_57653_cast_fp16 = slice_by_index(begin = var_57653_begin_0, end = var_57653_end_0, end_mask = var_57653_end_mask_0, x = k_511_cast_fp16)[name = tensor("op_57653_cast_fp16")]; + tensor var_57657_begin_0 = const()[name = tensor("op_57657_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_57657_end_0 = const()[name = tensor("op_57657_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_57657_end_mask_0 = const()[name = tensor("op_57657_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_57657_cast_fp16 = slice_by_index(begin = var_57657_begin_0, end = var_57657_end_0, end_mask = var_57657_end_mask_0, x = k_511_cast_fp16)[name = tensor("op_57657_cast_fp16")]; + tensor var_57661_begin_0 = const()[name = tensor("op_57661_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_57661_end_0 = const()[name = tensor("op_57661_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_57661_end_mask_0 = const()[name = tensor("op_57661_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_57661_cast_fp16 = slice_by_index(begin = var_57661_begin_0, end = var_57661_end_0, end_mask = var_57661_end_mask_0, x = k_511_cast_fp16)[name = tensor("op_57661_cast_fp16")]; + tensor var_57665_begin_0 = const()[name = tensor("op_57665_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_57665_end_0 = const()[name = tensor("op_57665_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_57665_end_mask_0 = const()[name = tensor("op_57665_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_57665_cast_fp16 = slice_by_index(begin = var_57665_begin_0, end = var_57665_end_0, end_mask = var_57665_end_mask_0, x = k_511_cast_fp16)[name = tensor("op_57665_cast_fp16")]; + tensor var_57669_begin_0 = const()[name = tensor("op_57669_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_57669_end_0 = const()[name = tensor("op_57669_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_57669_end_mask_0 = const()[name = tensor("op_57669_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_57669_cast_fp16 = slice_by_index(begin = var_57669_begin_0, end = var_57669_end_0, end_mask = var_57669_end_mask_0, x = k_511_cast_fp16)[name = tensor("op_57669_cast_fp16")]; + tensor var_57673_begin_0 = const()[name = tensor("op_57673_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_57673_end_0 = const()[name = tensor("op_57673_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_57673_end_mask_0 = const()[name = tensor("op_57673_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_57673_cast_fp16 = slice_by_index(begin = var_57673_begin_0, end = var_57673_end_0, end_mask = var_57673_end_mask_0, x = k_511_cast_fp16)[name = tensor("op_57673_cast_fp16")]; + tensor var_57677_begin_0 = const()[name = tensor("op_57677_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_57677_end_0 = const()[name = tensor("op_57677_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_57677_end_mask_0 = const()[name = tensor("op_57677_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_57677_cast_fp16 = slice_by_index(begin = var_57677_begin_0, end = var_57677_end_0, end_mask = var_57677_end_mask_0, x = k_511_cast_fp16)[name = tensor("op_57677_cast_fp16")]; + tensor var_57681_begin_0 = const()[name = tensor("op_57681_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_57681_end_0 = const()[name = tensor("op_57681_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_57681_end_mask_0 = const()[name = tensor("op_57681_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_57681_cast_fp16 = slice_by_index(begin = var_57681_begin_0, end = var_57681_end_0, end_mask = var_57681_end_mask_0, x = k_511_cast_fp16)[name = tensor("op_57681_cast_fp16")]; + tensor var_57685_begin_0 = const()[name = tensor("op_57685_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_57685_end_0 = const()[name = tensor("op_57685_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_57685_end_mask_0 = const()[name = tensor("op_57685_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_57685_cast_fp16 = slice_by_index(begin = var_57685_begin_0, end = var_57685_end_0, end_mask = var_57685_end_mask_0, x = k_511_cast_fp16)[name = tensor("op_57685_cast_fp16")]; + tensor var_57689_begin_0 = const()[name = tensor("op_57689_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_57689_end_0 = const()[name = tensor("op_57689_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_57689_end_mask_0 = const()[name = tensor("op_57689_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_57689_cast_fp16 = slice_by_index(begin = var_57689_begin_0, end = var_57689_end_0, end_mask = var_57689_end_mask_0, x = k_511_cast_fp16)[name = tensor("op_57689_cast_fp16")]; + tensor var_57693_begin_0 = const()[name = tensor("op_57693_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_57693_end_0 = const()[name = tensor("op_57693_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_57693_end_mask_0 = const()[name = tensor("op_57693_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_57693_cast_fp16 = slice_by_index(begin = var_57693_begin_0, end = var_57693_end_0, end_mask = var_57693_end_mask_0, x = k_511_cast_fp16)[name = tensor("op_57693_cast_fp16")]; + tensor var_57697_begin_0 = const()[name = tensor("op_57697_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_57697_end_0 = const()[name = tensor("op_57697_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_57697_end_mask_0 = const()[name = tensor("op_57697_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_57697_cast_fp16 = slice_by_index(begin = var_57697_begin_0, end = var_57697_end_0, end_mask = var_57697_end_mask_0, x = k_511_cast_fp16)[name = tensor("op_57697_cast_fp16")]; + tensor var_57701_begin_0 = const()[name = tensor("op_57701_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_57701_end_0 = const()[name = tensor("op_57701_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_57701_end_mask_0 = const()[name = tensor("op_57701_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_57701_cast_fp16 = slice_by_index(begin = var_57701_begin_0, end = var_57701_end_0, end_mask = var_57701_end_mask_0, x = k_511_cast_fp16)[name = tensor("op_57701_cast_fp16")]; + tensor var_57705_begin_0 = const()[name = tensor("op_57705_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_57705_end_0 = const()[name = tensor("op_57705_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_57705_end_mask_0 = const()[name = tensor("op_57705_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_57705_cast_fp16 = slice_by_index(begin = var_57705_begin_0, end = var_57705_end_0, end_mask = var_57705_end_mask_0, x = k_511_cast_fp16)[name = tensor("op_57705_cast_fp16")]; + tensor var_57709_begin_0 = const()[name = tensor("op_57709_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_57709_end_0 = const()[name = tensor("op_57709_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_57709_end_mask_0 = const()[name = tensor("op_57709_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_57709_cast_fp16 = slice_by_index(begin = var_57709_begin_0, end = var_57709_end_0, end_mask = var_57709_end_mask_0, x = k_511_cast_fp16)[name = tensor("op_57709_cast_fp16")]; + tensor var_57713_begin_0 = const()[name = tensor("op_57713_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_57713_end_0 = const()[name = tensor("op_57713_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_57713_end_mask_0 = const()[name = tensor("op_57713_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_57713_cast_fp16 = slice_by_index(begin = var_57713_begin_0, end = var_57713_end_0, end_mask = var_57713_end_mask_0, x = k_511_cast_fp16)[name = tensor("op_57713_cast_fp16")]; + tensor var_57717_begin_0 = const()[name = tensor("op_57717_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_57717_end_0 = const()[name = tensor("op_57717_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_57717_end_mask_0 = const()[name = tensor("op_57717_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_57717_cast_fp16 = slice_by_index(begin = var_57717_begin_0, end = var_57717_end_0, end_mask = var_57717_end_mask_0, x = k_511_cast_fp16)[name = tensor("op_57717_cast_fp16")]; + tensor var_57719_begin_0 = const()[name = tensor("op_57719_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_57719_end_0 = const()[name = tensor("op_57719_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_57719_end_mask_0 = const()[name = tensor("op_57719_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57719_cast_fp16 = slice_by_index(begin = var_57719_begin_0, end = var_57719_end_0, end_mask = var_57719_end_mask_0, x = v_255_cast_fp16)[name = tensor("op_57719_cast_fp16")]; + tensor var_57723_begin_0 = const()[name = tensor("op_57723_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_57723_end_0 = const()[name = tensor("op_57723_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_57723_end_mask_0 = const()[name = tensor("op_57723_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57723_cast_fp16 = slice_by_index(begin = var_57723_begin_0, end = var_57723_end_0, end_mask = var_57723_end_mask_0, x = v_255_cast_fp16)[name = tensor("op_57723_cast_fp16")]; + tensor var_57727_begin_0 = const()[name = tensor("op_57727_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_57727_end_0 = const()[name = tensor("op_57727_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_57727_end_mask_0 = const()[name = tensor("op_57727_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57727_cast_fp16 = slice_by_index(begin = var_57727_begin_0, end = var_57727_end_0, end_mask = var_57727_end_mask_0, x = v_255_cast_fp16)[name = tensor("op_57727_cast_fp16")]; + tensor var_57731_begin_0 = const()[name = tensor("op_57731_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_57731_end_0 = const()[name = tensor("op_57731_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_57731_end_mask_0 = const()[name = tensor("op_57731_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57731_cast_fp16 = slice_by_index(begin = var_57731_begin_0, end = var_57731_end_0, end_mask = var_57731_end_mask_0, x = v_255_cast_fp16)[name = tensor("op_57731_cast_fp16")]; + tensor var_57735_begin_0 = const()[name = tensor("op_57735_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_57735_end_0 = const()[name = tensor("op_57735_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_57735_end_mask_0 = const()[name = tensor("op_57735_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57735_cast_fp16 = slice_by_index(begin = var_57735_begin_0, end = var_57735_end_0, end_mask = var_57735_end_mask_0, x = v_255_cast_fp16)[name = tensor("op_57735_cast_fp16")]; + tensor var_57739_begin_0 = const()[name = tensor("op_57739_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_57739_end_0 = const()[name = tensor("op_57739_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_57739_end_mask_0 = const()[name = tensor("op_57739_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57739_cast_fp16 = slice_by_index(begin = var_57739_begin_0, end = var_57739_end_0, end_mask = var_57739_end_mask_0, x = v_255_cast_fp16)[name = tensor("op_57739_cast_fp16")]; + tensor var_57743_begin_0 = const()[name = tensor("op_57743_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_57743_end_0 = const()[name = tensor("op_57743_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_57743_end_mask_0 = const()[name = tensor("op_57743_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57743_cast_fp16 = slice_by_index(begin = var_57743_begin_0, end = var_57743_end_0, end_mask = var_57743_end_mask_0, x = v_255_cast_fp16)[name = tensor("op_57743_cast_fp16")]; + tensor var_57747_begin_0 = const()[name = tensor("op_57747_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_57747_end_0 = const()[name = tensor("op_57747_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_57747_end_mask_0 = const()[name = tensor("op_57747_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57747_cast_fp16 = slice_by_index(begin = var_57747_begin_0, end = var_57747_end_0, end_mask = var_57747_end_mask_0, x = v_255_cast_fp16)[name = tensor("op_57747_cast_fp16")]; + tensor var_57751_begin_0 = const()[name = tensor("op_57751_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_57751_end_0 = const()[name = tensor("op_57751_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_57751_end_mask_0 = const()[name = tensor("op_57751_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57751_cast_fp16 = slice_by_index(begin = var_57751_begin_0, end = var_57751_end_0, end_mask = var_57751_end_mask_0, x = v_255_cast_fp16)[name = tensor("op_57751_cast_fp16")]; + tensor var_57755_begin_0 = const()[name = tensor("op_57755_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_57755_end_0 = const()[name = tensor("op_57755_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_57755_end_mask_0 = const()[name = tensor("op_57755_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57755_cast_fp16 = slice_by_index(begin = var_57755_begin_0, end = var_57755_end_0, end_mask = var_57755_end_mask_0, x = v_255_cast_fp16)[name = tensor("op_57755_cast_fp16")]; + tensor var_57759_begin_0 = const()[name = tensor("op_57759_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_57759_end_0 = const()[name = tensor("op_57759_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_57759_end_mask_0 = const()[name = tensor("op_57759_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57759_cast_fp16 = slice_by_index(begin = var_57759_begin_0, end = var_57759_end_0, end_mask = var_57759_end_mask_0, x = v_255_cast_fp16)[name = tensor("op_57759_cast_fp16")]; + tensor var_57763_begin_0 = const()[name = tensor("op_57763_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_57763_end_0 = const()[name = tensor("op_57763_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_57763_end_mask_0 = const()[name = tensor("op_57763_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57763_cast_fp16 = slice_by_index(begin = var_57763_begin_0, end = var_57763_end_0, end_mask = var_57763_end_mask_0, x = v_255_cast_fp16)[name = tensor("op_57763_cast_fp16")]; + tensor var_57767_begin_0 = const()[name = tensor("op_57767_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_57767_end_0 = const()[name = tensor("op_57767_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_57767_end_mask_0 = const()[name = tensor("op_57767_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57767_cast_fp16 = slice_by_index(begin = var_57767_begin_0, end = var_57767_end_0, end_mask = var_57767_end_mask_0, x = v_255_cast_fp16)[name = tensor("op_57767_cast_fp16")]; + tensor var_57771_begin_0 = const()[name = tensor("op_57771_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_57771_end_0 = const()[name = tensor("op_57771_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_57771_end_mask_0 = const()[name = tensor("op_57771_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57771_cast_fp16 = slice_by_index(begin = var_57771_begin_0, end = var_57771_end_0, end_mask = var_57771_end_mask_0, x = v_255_cast_fp16)[name = tensor("op_57771_cast_fp16")]; + tensor var_57775_begin_0 = const()[name = tensor("op_57775_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_57775_end_0 = const()[name = tensor("op_57775_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_57775_end_mask_0 = const()[name = tensor("op_57775_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57775_cast_fp16 = slice_by_index(begin = var_57775_begin_0, end = var_57775_end_0, end_mask = var_57775_end_mask_0, x = v_255_cast_fp16)[name = tensor("op_57775_cast_fp16")]; + tensor var_57779_begin_0 = const()[name = tensor("op_57779_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_57779_end_0 = const()[name = tensor("op_57779_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_57779_end_mask_0 = const()[name = tensor("op_57779_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57779_cast_fp16 = slice_by_index(begin = var_57779_begin_0, end = var_57779_end_0, end_mask = var_57779_end_mask_0, x = v_255_cast_fp16)[name = tensor("op_57779_cast_fp16")]; + tensor var_57783_begin_0 = const()[name = tensor("op_57783_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_57783_end_0 = const()[name = tensor("op_57783_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_57783_end_mask_0 = const()[name = tensor("op_57783_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57783_cast_fp16 = slice_by_index(begin = var_57783_begin_0, end = var_57783_end_0, end_mask = var_57783_end_mask_0, x = v_255_cast_fp16)[name = tensor("op_57783_cast_fp16")]; + tensor var_57787_begin_0 = const()[name = tensor("op_57787_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_57787_end_0 = const()[name = tensor("op_57787_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_57787_end_mask_0 = const()[name = tensor("op_57787_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57787_cast_fp16 = slice_by_index(begin = var_57787_begin_0, end = var_57787_end_0, end_mask = var_57787_end_mask_0, x = v_255_cast_fp16)[name = tensor("op_57787_cast_fp16")]; + tensor var_57791_begin_0 = const()[name = tensor("op_57791_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_57791_end_0 = const()[name = tensor("op_57791_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_57791_end_mask_0 = const()[name = tensor("op_57791_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57791_cast_fp16 = slice_by_index(begin = var_57791_begin_0, end = var_57791_end_0, end_mask = var_57791_end_mask_0, x = v_255_cast_fp16)[name = tensor("op_57791_cast_fp16")]; + tensor var_57795_begin_0 = const()[name = tensor("op_57795_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_57795_end_0 = const()[name = tensor("op_57795_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_57795_end_mask_0 = const()[name = tensor("op_57795_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_57795_cast_fp16 = slice_by_index(begin = var_57795_begin_0, end = var_57795_end_0, end_mask = var_57795_end_mask_0, x = v_255_cast_fp16)[name = tensor("op_57795_cast_fp16")]; + tensor var_57799_equation_0 = const()[name = tensor("op_57799_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_57799_cast_fp16 = einsum(equation = var_57799_equation_0, values = (var_57641_cast_fp16, var_57558_cast_fp16))[name = tensor("op_57799_cast_fp16")]; + tensor var_57800_to_fp16 = const()[name = tensor("op_57800_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4921_cast_fp16 = mul(x = var_57799_cast_fp16, y = var_57800_to_fp16)[name = tensor("aw_4921_cast_fp16")]; + tensor var_57803_equation_0 = const()[name = tensor("op_57803_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_57803_cast_fp16 = einsum(equation = var_57803_equation_0, values = (var_57645_cast_fp16, var_57562_cast_fp16))[name = tensor("op_57803_cast_fp16")]; + tensor var_57804_to_fp16 = const()[name = tensor("op_57804_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4923_cast_fp16 = mul(x = var_57803_cast_fp16, y = var_57804_to_fp16)[name = tensor("aw_4923_cast_fp16")]; + tensor var_57807_equation_0 = const()[name = tensor("op_57807_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_57807_cast_fp16 = einsum(equation = var_57807_equation_0, values = (var_57649_cast_fp16, var_57566_cast_fp16))[name = tensor("op_57807_cast_fp16")]; + tensor var_57808_to_fp16 = const()[name = tensor("op_57808_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4925_cast_fp16 = mul(x = var_57807_cast_fp16, y = var_57808_to_fp16)[name = tensor("aw_4925_cast_fp16")]; + tensor var_57811_equation_0 = const()[name = tensor("op_57811_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_57811_cast_fp16 = einsum(equation = var_57811_equation_0, values = (var_57653_cast_fp16, var_57570_cast_fp16))[name = tensor("op_57811_cast_fp16")]; + tensor var_57812_to_fp16 = const()[name = tensor("op_57812_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4927_cast_fp16 = mul(x = var_57811_cast_fp16, y = var_57812_to_fp16)[name = tensor("aw_4927_cast_fp16")]; + tensor var_57815_equation_0 = const()[name = tensor("op_57815_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_57815_cast_fp16 = einsum(equation = var_57815_equation_0, values = (var_57657_cast_fp16, var_57574_cast_fp16))[name = tensor("op_57815_cast_fp16")]; + tensor var_57816_to_fp16 = const()[name = tensor("op_57816_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4929_cast_fp16 = mul(x = var_57815_cast_fp16, y = var_57816_to_fp16)[name = tensor("aw_4929_cast_fp16")]; + tensor var_57819_equation_0 = const()[name = tensor("op_57819_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_57819_cast_fp16 = einsum(equation = var_57819_equation_0, values = (var_57661_cast_fp16, var_57578_cast_fp16))[name = tensor("op_57819_cast_fp16")]; + tensor var_57820_to_fp16 = const()[name = tensor("op_57820_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4931_cast_fp16 = mul(x = var_57819_cast_fp16, y = var_57820_to_fp16)[name = tensor("aw_4931_cast_fp16")]; + tensor var_57823_equation_0 = const()[name = tensor("op_57823_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_57823_cast_fp16 = einsum(equation = var_57823_equation_0, values = (var_57665_cast_fp16, var_57582_cast_fp16))[name = tensor("op_57823_cast_fp16")]; + tensor var_57824_to_fp16 = const()[name = tensor("op_57824_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4933_cast_fp16 = mul(x = var_57823_cast_fp16, y = var_57824_to_fp16)[name = tensor("aw_4933_cast_fp16")]; + tensor var_57827_equation_0 = const()[name = tensor("op_57827_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_57827_cast_fp16 = einsum(equation = var_57827_equation_0, values = (var_57669_cast_fp16, var_57586_cast_fp16))[name = tensor("op_57827_cast_fp16")]; + tensor var_57828_to_fp16 = const()[name = tensor("op_57828_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4935_cast_fp16 = mul(x = var_57827_cast_fp16, y = var_57828_to_fp16)[name = tensor("aw_4935_cast_fp16")]; + tensor var_57831_equation_0 = const()[name = tensor("op_57831_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_57831_cast_fp16 = einsum(equation = var_57831_equation_0, values = (var_57673_cast_fp16, var_57590_cast_fp16))[name = tensor("op_57831_cast_fp16")]; + tensor var_57832_to_fp16 = const()[name = tensor("op_57832_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4937_cast_fp16 = mul(x = var_57831_cast_fp16, y = var_57832_to_fp16)[name = tensor("aw_4937_cast_fp16")]; + tensor var_57835_equation_0 = const()[name = tensor("op_57835_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_57835_cast_fp16 = einsum(equation = var_57835_equation_0, values = (var_57677_cast_fp16, var_57594_cast_fp16))[name = tensor("op_57835_cast_fp16")]; + tensor var_57836_to_fp16 = const()[name = tensor("op_57836_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4939_cast_fp16 = mul(x = var_57835_cast_fp16, y = var_57836_to_fp16)[name = tensor("aw_4939_cast_fp16")]; + tensor var_57839_equation_0 = const()[name = tensor("op_57839_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_57839_cast_fp16 = einsum(equation = var_57839_equation_0, values = (var_57681_cast_fp16, var_57598_cast_fp16))[name = tensor("op_57839_cast_fp16")]; + tensor var_57840_to_fp16 = const()[name = tensor("op_57840_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4941_cast_fp16 = mul(x = var_57839_cast_fp16, y = var_57840_to_fp16)[name = tensor("aw_4941_cast_fp16")]; + tensor var_57843_equation_0 = const()[name = tensor("op_57843_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_57843_cast_fp16 = einsum(equation = var_57843_equation_0, values = (var_57685_cast_fp16, var_57602_cast_fp16))[name = tensor("op_57843_cast_fp16")]; + tensor var_57844_to_fp16 = const()[name = tensor("op_57844_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4943_cast_fp16 = mul(x = var_57843_cast_fp16, y = var_57844_to_fp16)[name = tensor("aw_4943_cast_fp16")]; + tensor var_57847_equation_0 = const()[name = tensor("op_57847_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_57847_cast_fp16 = einsum(equation = var_57847_equation_0, values = (var_57689_cast_fp16, var_57606_cast_fp16))[name = tensor("op_57847_cast_fp16")]; + tensor var_57848_to_fp16 = const()[name = tensor("op_57848_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4945_cast_fp16 = mul(x = var_57847_cast_fp16, y = var_57848_to_fp16)[name = tensor("aw_4945_cast_fp16")]; + tensor var_57851_equation_0 = const()[name = tensor("op_57851_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_57851_cast_fp16 = einsum(equation = var_57851_equation_0, values = (var_57693_cast_fp16, var_57610_cast_fp16))[name = tensor("op_57851_cast_fp16")]; + tensor var_57852_to_fp16 = const()[name = tensor("op_57852_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4947_cast_fp16 = mul(x = var_57851_cast_fp16, y = var_57852_to_fp16)[name = tensor("aw_4947_cast_fp16")]; + tensor var_57855_equation_0 = const()[name = tensor("op_57855_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_57855_cast_fp16 = einsum(equation = var_57855_equation_0, values = (var_57697_cast_fp16, var_57614_cast_fp16))[name = tensor("op_57855_cast_fp16")]; + tensor var_57856_to_fp16 = const()[name = tensor("op_57856_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4949_cast_fp16 = mul(x = var_57855_cast_fp16, y = var_57856_to_fp16)[name = tensor("aw_4949_cast_fp16")]; + tensor var_57859_equation_0 = const()[name = tensor("op_57859_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_57859_cast_fp16 = einsum(equation = var_57859_equation_0, values = (var_57701_cast_fp16, var_57618_cast_fp16))[name = tensor("op_57859_cast_fp16")]; + tensor var_57860_to_fp16 = const()[name = tensor("op_57860_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4951_cast_fp16 = mul(x = var_57859_cast_fp16, y = var_57860_to_fp16)[name = tensor("aw_4951_cast_fp16")]; + tensor var_57863_equation_0 = const()[name = tensor("op_57863_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_57863_cast_fp16 = einsum(equation = var_57863_equation_0, values = (var_57705_cast_fp16, var_57622_cast_fp16))[name = tensor("op_57863_cast_fp16")]; + tensor var_57864_to_fp16 = const()[name = tensor("op_57864_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4953_cast_fp16 = mul(x = var_57863_cast_fp16, y = var_57864_to_fp16)[name = tensor("aw_4953_cast_fp16")]; + tensor var_57867_equation_0 = const()[name = tensor("op_57867_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_57867_cast_fp16 = einsum(equation = var_57867_equation_0, values = (var_57709_cast_fp16, var_57626_cast_fp16))[name = tensor("op_57867_cast_fp16")]; + tensor var_57868_to_fp16 = const()[name = tensor("op_57868_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4955_cast_fp16 = mul(x = var_57867_cast_fp16, y = var_57868_to_fp16)[name = tensor("aw_4955_cast_fp16")]; + tensor var_57871_equation_0 = const()[name = tensor("op_57871_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_57871_cast_fp16 = einsum(equation = var_57871_equation_0, values = (var_57713_cast_fp16, var_57630_cast_fp16))[name = tensor("op_57871_cast_fp16")]; + tensor var_57872_to_fp16 = const()[name = tensor("op_57872_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4957_cast_fp16 = mul(x = var_57871_cast_fp16, y = var_57872_to_fp16)[name = tensor("aw_4957_cast_fp16")]; + tensor var_57875_equation_0 = const()[name = tensor("op_57875_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_57875_cast_fp16 = einsum(equation = var_57875_equation_0, values = (var_57717_cast_fp16, var_57634_cast_fp16))[name = tensor("op_57875_cast_fp16")]; + tensor var_57876_to_fp16 = const()[name = tensor("op_57876_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4959_cast_fp16 = mul(x = var_57875_cast_fp16, y = var_57876_to_fp16)[name = tensor("aw_4959_cast_fp16")]; + tensor var_57878_cast_fp16 = softmax(axis = var_30355, x = aw_4921_cast_fp16)[name = tensor("op_57878_cast_fp16")]; + tensor var_57879_cast_fp16 = softmax(axis = var_30355, x = aw_4923_cast_fp16)[name = tensor("op_57879_cast_fp16")]; + tensor var_57880_cast_fp16 = softmax(axis = var_30355, x = aw_4925_cast_fp16)[name = tensor("op_57880_cast_fp16")]; + tensor var_57881_cast_fp16 = softmax(axis = var_30355, x = aw_4927_cast_fp16)[name = tensor("op_57881_cast_fp16")]; + tensor var_57882_cast_fp16 = softmax(axis = var_30355, x = aw_4929_cast_fp16)[name = tensor("op_57882_cast_fp16")]; + tensor var_57883_cast_fp16 = softmax(axis = var_30355, x = aw_4931_cast_fp16)[name = tensor("op_57883_cast_fp16")]; + tensor var_57884_cast_fp16 = softmax(axis = var_30355, x = aw_4933_cast_fp16)[name = tensor("op_57884_cast_fp16")]; + tensor var_57885_cast_fp16 = softmax(axis = var_30355, x = aw_4935_cast_fp16)[name = tensor("op_57885_cast_fp16")]; + tensor var_57886_cast_fp16 = softmax(axis = var_30355, x = aw_4937_cast_fp16)[name = tensor("op_57886_cast_fp16")]; + tensor var_57887_cast_fp16 = softmax(axis = var_30355, x = aw_4939_cast_fp16)[name = tensor("op_57887_cast_fp16")]; + tensor var_57888_cast_fp16 = softmax(axis = var_30355, x = aw_4941_cast_fp16)[name = tensor("op_57888_cast_fp16")]; + tensor var_57889_cast_fp16 = softmax(axis = var_30355, x = aw_4943_cast_fp16)[name = tensor("op_57889_cast_fp16")]; + tensor var_57890_cast_fp16 = softmax(axis = var_30355, x = aw_4945_cast_fp16)[name = tensor("op_57890_cast_fp16")]; + tensor var_57891_cast_fp16 = softmax(axis = var_30355, x = aw_4947_cast_fp16)[name = tensor("op_57891_cast_fp16")]; + tensor var_57892_cast_fp16 = softmax(axis = var_30355, x = aw_4949_cast_fp16)[name = tensor("op_57892_cast_fp16")]; + tensor var_57893_cast_fp16 = softmax(axis = var_30355, x = aw_4951_cast_fp16)[name = tensor("op_57893_cast_fp16")]; + tensor var_57894_cast_fp16 = softmax(axis = var_30355, x = aw_4953_cast_fp16)[name = tensor("op_57894_cast_fp16")]; + tensor var_57895_cast_fp16 = softmax(axis = var_30355, x = aw_4955_cast_fp16)[name = tensor("op_57895_cast_fp16")]; + tensor var_57896_cast_fp16 = softmax(axis = var_30355, x = aw_4957_cast_fp16)[name = tensor("op_57896_cast_fp16")]; + tensor var_57897_cast_fp16 = softmax(axis = var_30355, x = aw_4959_cast_fp16)[name = tensor("op_57897_cast_fp16")]; + tensor var_57899_equation_0 = const()[name = tensor("op_57899_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_57899_cast_fp16 = einsum(equation = var_57899_equation_0, values = (var_57719_cast_fp16, var_57878_cast_fp16))[name = tensor("op_57899_cast_fp16")]; + tensor var_57901_equation_0 = const()[name = tensor("op_57901_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_57901_cast_fp16 = einsum(equation = var_57901_equation_0, values = (var_57723_cast_fp16, var_57879_cast_fp16))[name = tensor("op_57901_cast_fp16")]; + tensor var_57903_equation_0 = const()[name = tensor("op_57903_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_57903_cast_fp16 = einsum(equation = var_57903_equation_0, values = (var_57727_cast_fp16, var_57880_cast_fp16))[name = tensor("op_57903_cast_fp16")]; + tensor var_57905_equation_0 = const()[name = tensor("op_57905_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_57905_cast_fp16 = einsum(equation = var_57905_equation_0, values = (var_57731_cast_fp16, var_57881_cast_fp16))[name = tensor("op_57905_cast_fp16")]; + tensor var_57907_equation_0 = const()[name = tensor("op_57907_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_57907_cast_fp16 = einsum(equation = var_57907_equation_0, values = (var_57735_cast_fp16, var_57882_cast_fp16))[name = tensor("op_57907_cast_fp16")]; + tensor var_57909_equation_0 = const()[name = tensor("op_57909_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_57909_cast_fp16 = einsum(equation = var_57909_equation_0, values = (var_57739_cast_fp16, var_57883_cast_fp16))[name = tensor("op_57909_cast_fp16")]; + tensor var_57911_equation_0 = const()[name = tensor("op_57911_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_57911_cast_fp16 = einsum(equation = var_57911_equation_0, values = (var_57743_cast_fp16, var_57884_cast_fp16))[name = tensor("op_57911_cast_fp16")]; + tensor var_57913_equation_0 = const()[name = tensor("op_57913_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_57913_cast_fp16 = einsum(equation = var_57913_equation_0, values = (var_57747_cast_fp16, var_57885_cast_fp16))[name = tensor("op_57913_cast_fp16")]; + tensor var_57915_equation_0 = const()[name = tensor("op_57915_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_57915_cast_fp16 = einsum(equation = var_57915_equation_0, values = (var_57751_cast_fp16, var_57886_cast_fp16))[name = tensor("op_57915_cast_fp16")]; + tensor var_57917_equation_0 = const()[name = tensor("op_57917_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_57917_cast_fp16 = einsum(equation = var_57917_equation_0, values = (var_57755_cast_fp16, var_57887_cast_fp16))[name = tensor("op_57917_cast_fp16")]; + tensor var_57919_equation_0 = const()[name = tensor("op_57919_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_57919_cast_fp16 = einsum(equation = var_57919_equation_0, values = (var_57759_cast_fp16, var_57888_cast_fp16))[name = tensor("op_57919_cast_fp16")]; + tensor var_57921_equation_0 = const()[name = tensor("op_57921_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_57921_cast_fp16 = einsum(equation = var_57921_equation_0, values = (var_57763_cast_fp16, var_57889_cast_fp16))[name = tensor("op_57921_cast_fp16")]; + tensor var_57923_equation_0 = const()[name = tensor("op_57923_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_57923_cast_fp16 = einsum(equation = var_57923_equation_0, values = (var_57767_cast_fp16, var_57890_cast_fp16))[name = tensor("op_57923_cast_fp16")]; + tensor var_57925_equation_0 = const()[name = tensor("op_57925_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_57925_cast_fp16 = einsum(equation = var_57925_equation_0, values = (var_57771_cast_fp16, var_57891_cast_fp16))[name = tensor("op_57925_cast_fp16")]; + tensor var_57927_equation_0 = const()[name = tensor("op_57927_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_57927_cast_fp16 = einsum(equation = var_57927_equation_0, values = (var_57775_cast_fp16, var_57892_cast_fp16))[name = tensor("op_57927_cast_fp16")]; + tensor var_57929_equation_0 = const()[name = tensor("op_57929_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_57929_cast_fp16 = einsum(equation = var_57929_equation_0, values = (var_57779_cast_fp16, var_57893_cast_fp16))[name = tensor("op_57929_cast_fp16")]; + tensor var_57931_equation_0 = const()[name = tensor("op_57931_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_57931_cast_fp16 = einsum(equation = var_57931_equation_0, values = (var_57783_cast_fp16, var_57894_cast_fp16))[name = tensor("op_57931_cast_fp16")]; + tensor var_57933_equation_0 = const()[name = tensor("op_57933_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_57933_cast_fp16 = einsum(equation = var_57933_equation_0, values = (var_57787_cast_fp16, var_57895_cast_fp16))[name = tensor("op_57933_cast_fp16")]; + tensor var_57935_equation_0 = const()[name = tensor("op_57935_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_57935_cast_fp16 = einsum(equation = var_57935_equation_0, values = (var_57791_cast_fp16, var_57896_cast_fp16))[name = tensor("op_57935_cast_fp16")]; + tensor var_57937_equation_0 = const()[name = tensor("op_57937_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_57937_cast_fp16 = einsum(equation = var_57937_equation_0, values = (var_57795_cast_fp16, var_57897_cast_fp16))[name = tensor("op_57937_cast_fp16")]; + tensor input_709_interleave_0 = const()[name = tensor("input_709_interleave_0"), val = tensor(false)]; + tensor input_709_cast_fp16 = concat(axis = var_30355, interleave = input_709_interleave_0, values = (var_57899_cast_fp16, var_57901_cast_fp16, var_57903_cast_fp16, var_57905_cast_fp16, var_57907_cast_fp16, var_57909_cast_fp16, var_57911_cast_fp16, var_57913_cast_fp16, var_57915_cast_fp16, var_57917_cast_fp16, var_57919_cast_fp16, var_57921_cast_fp16, var_57923_cast_fp16, var_57925_cast_fp16, var_57927_cast_fp16, var_57929_cast_fp16, var_57931_cast_fp16, var_57933_cast_fp16, var_57935_cast_fp16, var_57937_cast_fp16))[name = tensor("input_709_cast_fp16")]; + tensor var_57943 = const()[name = tensor("op_57943"), val = tensor([1, 1])]; + tensor var_57945 = const()[name = tensor("op_57945"), val = tensor([1, 1])]; + tensor var_57947_pad_type_0 = const()[name = tensor("op_57947_pad_type_0"), val = tensor("custom")]; + tensor var_57947_pad_0 = const()[name = tensor("op_57947_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_9_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_9_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4824590976)))]; + tensor up_blocks_0_attentions_2_transformer_blocks_9_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_9_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4827867840)))]; + tensor var_57947_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_9_attn2_to_out_0_bias_to_fp16, dilations = var_57945, groups = var_30355, pad = var_57947_pad_0, pad_type = var_57947_pad_type_0, strides = var_57943, weight = up_blocks_0_attentions_2_transformer_blocks_9_attn2_to_out_0_weight_to_fp16, x = input_709_cast_fp16)[name = tensor("op_57947_cast_fp16")]; + tensor inputs_383_cast_fp16 = add(x = var_57947_cast_fp16, y = inputs_381_cast_fp16)[name = tensor("inputs_383_cast_fp16")]; + tensor input_711_axes_0 = const()[name = tensor("input_711_axes_0"), val = tensor([1])]; + tensor input_711_gamma_0_to_fp16 = const()[name = tensor("input_711_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4827870464)))]; + tensor input_711_beta_0_to_fp16 = const()[name = tensor("input_711_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4827873088)))]; + tensor var_57957_to_fp16 = const()[name = tensor("op_57957_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_711_cast_fp16 = layer_norm(axes = input_711_axes_0, beta = input_711_beta_0_to_fp16, epsilon = var_57957_to_fp16, gamma = input_711_gamma_0_to_fp16, x = inputs_383_cast_fp16)[name = tensor("input_711_cast_fp16")]; + tensor var_57973 = const()[name = tensor("op_57973"), val = tensor([1, 1])]; + tensor var_57975 = const()[name = tensor("op_57975"), val = tensor([1, 1])]; + tensor var_57977_pad_type_0 = const()[name = tensor("op_57977_pad_type_0"), val = tensor("custom")]; + tensor var_57977_pad_0 = const()[name = tensor("op_57977_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_9_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_9_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4827875712)))]; + tensor up_blocks_0_attentions_2_transformer_blocks_9_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_9_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4854090176)))]; + tensor var_57977_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_9_ff_net_0_proj_bias_to_fp16, dilations = var_57975, groups = var_30355, pad = var_57977_pad_0, pad_type = var_57977_pad_type_0, strides = var_57973, weight = up_blocks_0_attentions_2_transformer_blocks_9_ff_net_0_proj_weight_to_fp16, x = input_711_cast_fp16)[name = tensor("op_57977_cast_fp16")]; + tensor var_57978_split_sizes_0 = const()[name = tensor("op_57978_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_57978_axis_0 = const()[name = tensor("op_57978_axis_0"), val = tensor(1)]; + tensor var_57978_cast_fp16_0, tensor var_57978_cast_fp16_1 = split(axis = var_57978_axis_0, split_sizes = var_57978_split_sizes_0, x = var_57977_cast_fp16)[name = tensor("op_57978_cast_fp16")]; + tensor var_57980_mode_0 = const()[name = tensor("op_57980_mode_0"), val = tensor("EXACT")]; + tensor var_57980_cast_fp16 = gelu(mode = var_57980_mode_0, x = var_57978_cast_fp16_1)[name = tensor("op_57980_cast_fp16")]; + tensor input_713_cast_fp16 = mul(x = var_57978_cast_fp16_0, y = var_57980_cast_fp16)[name = tensor("input_713_cast_fp16")]; + tensor var_57984 = const()[name = tensor("op_57984"), val = tensor([1, 1])]; + tensor var_57986 = const()[name = tensor("op_57986"), val = tensor([1, 1])]; + tensor var_57988_pad_type_0 = const()[name = tensor("op_57988_pad_type_0"), val = tensor("custom")]; + tensor var_57988_pad_0 = const()[name = tensor("op_57988_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_transformer_blocks_9_ff_net_2_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_9_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4854110720)))]; + tensor up_blocks_0_attentions_2_transformer_blocks_9_ff_net_2_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_transformer_blocks_9_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4867217984)))]; + tensor var_57988_cast_fp16 = conv(bias = up_blocks_0_attentions_2_transformer_blocks_9_ff_net_2_bias_to_fp16, dilations = var_57986, groups = var_30355, pad = var_57988_pad_0, pad_type = var_57988_pad_type_0, strides = var_57984, weight = up_blocks_0_attentions_2_transformer_blocks_9_ff_net_2_weight_to_fp16, x = input_713_cast_fp16)[name = tensor("op_57988_cast_fp16")]; + tensor hidden_states_495_cast_fp16 = add(x = var_57988_cast_fp16, y = inputs_383_cast_fp16)[name = tensor("hidden_states_495_cast_fp16")]; + tensor var_57990 = const()[name = tensor("op_57990"), val = tensor([2, 1280, 32, 32])]; + tensor input_715_cast_fp16 = reshape(shape = var_57990, x = hidden_states_495_cast_fp16)[name = tensor("input_715_cast_fp16")]; + tensor var_57994 = const()[name = tensor("op_57994"), val = tensor([1, 1])]; + tensor var_57996 = const()[name = tensor("op_57996"), val = tensor([1, 1])]; + tensor hidden_states_497_pad_type_0 = const()[name = tensor("hidden_states_497_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_497_pad_0 = const()[name = tensor("hidden_states_497_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_attentions_2_proj_out_weight_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_proj_out_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4867220608)))]; + tensor up_blocks_0_attentions_2_proj_out_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_2_proj_out_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4870497472)))]; + tensor hidden_states_497_cast_fp16 = conv(bias = up_blocks_0_attentions_2_proj_out_bias_to_fp16, dilations = var_57996, groups = var_30355, pad = hidden_states_497_pad_0, pad_type = hidden_states_497_pad_type_0, strides = var_57994, weight = up_blocks_0_attentions_2_proj_out_weight_to_fp16, x = input_715_cast_fp16)[name = tensor("hidden_states_497_cast_fp16")]; + tensor input_717_cast_fp16 = add(x = hidden_states_497_cast_fp16, y = hidden_states_431_cast_fp16)[name = tensor("input_717_cast_fp16")]; + tensor input_719_scale_factor_height_0 = const()[name = tensor("input_719_scale_factor_height_0"), val = tensor(0x1p+1)]; + tensor input_719_scale_factor_width_0 = const()[name = tensor("input_719_scale_factor_width_0"), val = tensor(0x1p+1)]; + tensor input_719_cast_fp16 = upsample_nearest_neighbor(scale_factor_height = input_719_scale_factor_height_0, scale_factor_width = input_719_scale_factor_width_0, x = input_717_cast_fp16)[name = tensor("input_719_cast_fp16")]; + tensor var_58005 = const()[name = tensor("op_58005"), val = tensor([1, 1])]; + tensor var_58007 = const()[name = tensor("op_58007"), val = tensor([1, 1])]; + tensor hidden_states_499_pad_type_0 = const()[name = tensor("hidden_states_499_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_499_pad_0 = const()[name = tensor("hidden_states_499_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor up_blocks_0_upsamplers_0_conv_weight_to_fp16 = const()[name = tensor("up_blocks_0_upsamplers_0_conv_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4870500096)))]; + tensor up_blocks_0_upsamplers_0_conv_bias_to_fp16 = const()[name = tensor("up_blocks_0_upsamplers_0_conv_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4899991360)))]; + tensor hidden_states_499_cast_fp16 = conv(bias = up_blocks_0_upsamplers_0_conv_bias_to_fp16, dilations = var_58007, groups = var_30355, pad = hidden_states_499_pad_0, pad_type = hidden_states_499_pad_type_0, strides = var_58005, weight = up_blocks_0_upsamplers_0_conv_weight_to_fp16, x = input_719_cast_fp16)[name = tensor("hidden_states_499_cast_fp16")]; + tensor var_58034 = const()[name = tensor("op_58034"), val = tensor(1)]; + tensor input_721_interleave_0 = const()[name = tensor("input_721_interleave_0"), val = tensor(false)]; + tensor input_721_cast_fp16 = concat(axis = var_58034, interleave = input_721_interleave_0, values = (hidden_states_499_cast_fp16, input_113_cast_fp16))[name = tensor("input_721_cast_fp16")]; + tensor reshape_120_shape_0 = const()[name = tensor("reshape_120_shape_0"), val = tensor([2, 32, 60, 64, 64])]; + tensor reshape_120_cast_fp16 = reshape(shape = reshape_120_shape_0, x = input_721_cast_fp16)[name = tensor("reshape_120_cast_fp16")]; + tensor reduce_mean_90_axes_0 = const()[name = tensor("reduce_mean_90_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_90_keep_dims_0 = const()[name = tensor("reduce_mean_90_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_90_cast_fp16 = reduce_mean(axes = reduce_mean_90_axes_0, keep_dims = reduce_mean_90_keep_dims_0, x = reshape_120_cast_fp16)[name = tensor("reduce_mean_90_cast_fp16")]; + tensor sub_60_cast_fp16 = sub(x = reshape_120_cast_fp16, y = reduce_mean_90_cast_fp16)[name = tensor("sub_60_cast_fp16")]; + tensor square_30_cast_fp16 = square(x = sub_60_cast_fp16)[name = tensor("square_30_cast_fp16")]; + tensor reduce_mean_92_axes_0 = const()[name = tensor("reduce_mean_92_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_92_keep_dims_0 = const()[name = tensor("reduce_mean_92_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_92_cast_fp16 = reduce_mean(axes = reduce_mean_92_axes_0, keep_dims = reduce_mean_92_keep_dims_0, x = square_30_cast_fp16)[name = tensor("reduce_mean_92_cast_fp16")]; + tensor add_60_y_0_to_fp16 = const()[name = tensor("add_60_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_60_cast_fp16 = add(x = reduce_mean_92_cast_fp16, y = add_60_y_0_to_fp16)[name = tensor("add_60_cast_fp16")]; + tensor sqrt_30_cast_fp16 = sqrt(x = add_60_cast_fp16)[name = tensor("sqrt_30_cast_fp16")]; + tensor real_div_30_cast_fp16 = real_div(x = sub_60_cast_fp16, y = sqrt_30_cast_fp16)[name = tensor("real_div_30_cast_fp16")]; + tensor reshape_121_shape_0 = const()[name = tensor("reshape_121_shape_0"), val = tensor([2, 1920, 64, 64])]; + tensor reshape_121_cast_fp16 = reshape(shape = reshape_121_shape_0, x = real_div_30_cast_fp16)[name = tensor("reshape_121_cast_fp16")]; + tensor add_61_gamma_0_to_fp16 = const()[name = tensor("add_61_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4899993984)))]; + tensor add_61_beta_0_to_fp16 = const()[name = tensor("add_61_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4899997888)))]; + tensor add_61_epsilon_0_to_fp16 = const()[name = tensor("add_61_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_61_cast_fp16 = batch_norm(beta = add_61_beta_0_to_fp16, epsilon = add_61_epsilon_0_to_fp16, gamma = add_61_gamma_0_to_fp16, mean = add_55_mean_0_to_fp16, variance = add_55_variance_0_to_fp16, x = reshape_121_cast_fp16)[name = tensor("add_61_cast_fp16")]; + tensor input_725_cast_fp16 = silu(x = add_61_cast_fp16)[name = tensor("input_725_cast_fp16")]; + tensor var_58063 = const()[name = tensor("op_58063"), val = tensor([1, 1])]; + tensor var_58065 = const()[name = tensor("op_58065"), val = tensor([1, 1])]; + tensor hidden_states_501_pad_type_0 = const()[name = tensor("hidden_states_501_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_501_pad_0 = const()[name = tensor("hidden_states_501_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor up_blocks_1_resnets_0_conv1_weight_to_fp16 = const()[name = tensor("up_blocks_1_resnets_0_conv1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4900001792)))]; + tensor up_blocks_1_resnets_0_conv1_bias_to_fp16 = const()[name = tensor("up_blocks_1_resnets_0_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4922120256)))]; + tensor hidden_states_501_cast_fp16 = conv(bias = up_blocks_1_resnets_0_conv1_bias_to_fp16, dilations = var_58065, groups = var_58034, pad = hidden_states_501_pad_0, pad_type = hidden_states_501_pad_type_0, strides = var_58063, weight = up_blocks_1_resnets_0_conv1_weight_to_fp16, x = input_725_cast_fp16)[name = tensor("hidden_states_501_cast_fp16")]; + tensor var_58071 = const()[name = tensor("op_58071"), val = tensor([1, 1])]; + tensor var_58073 = const()[name = tensor("op_58073"), val = tensor([1, 1])]; + tensor temb_23_pad_type_0 = const()[name = tensor("temb_23_pad_type_0"), val = tensor("custom")]; + tensor temb_23_pad_0 = const()[name = tensor("temb_23_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_resnets_0_time_emb_proj_weight_to_fp16 = const()[name = tensor("up_blocks_1_resnets_0_time_emb_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4922121600)))]; + tensor up_blocks_1_resnets_0_time_emb_proj_bias_to_fp16 = const()[name = tensor("up_blocks_1_resnets_0_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4923760064)))]; + tensor temb_23_cast_fp16 = conv(bias = up_blocks_1_resnets_0_time_emb_proj_bias_to_fp16, dilations = var_58073, groups = var_58034, pad = temb_23_pad_0, pad_type = temb_23_pad_type_0, strides = var_58071, weight = up_blocks_1_resnets_0_time_emb_proj_weight_to_fp16, x = input_21_cast_fp16)[name = tensor("temb_23_cast_fp16")]; + tensor input_729_cast_fp16 = add(x = hidden_states_501_cast_fp16, y = temb_23_cast_fp16)[name = tensor("input_729_cast_fp16")]; + tensor reshape_124_shape_0 = const()[name = tensor("reshape_124_shape_0"), val = tensor([2, 32, 20, 64, 64])]; + tensor reshape_124_cast_fp16 = reshape(shape = reshape_124_shape_0, x = input_729_cast_fp16)[name = tensor("reshape_124_cast_fp16")]; + tensor reduce_mean_93_axes_0 = const()[name = tensor("reduce_mean_93_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_93_keep_dims_0 = const()[name = tensor("reduce_mean_93_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_93_cast_fp16 = reduce_mean(axes = reduce_mean_93_axes_0, keep_dims = reduce_mean_93_keep_dims_0, x = reshape_124_cast_fp16)[name = tensor("reduce_mean_93_cast_fp16")]; + tensor sub_62_cast_fp16 = sub(x = reshape_124_cast_fp16, y = reduce_mean_93_cast_fp16)[name = tensor("sub_62_cast_fp16")]; + tensor square_31_cast_fp16 = square(x = sub_62_cast_fp16)[name = tensor("square_31_cast_fp16")]; + tensor reduce_mean_95_axes_0 = const()[name = tensor("reduce_mean_95_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_95_keep_dims_0 = const()[name = tensor("reduce_mean_95_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_95_cast_fp16 = reduce_mean(axes = reduce_mean_95_axes_0, keep_dims = reduce_mean_95_keep_dims_0, x = square_31_cast_fp16)[name = tensor("reduce_mean_95_cast_fp16")]; + tensor add_62_y_0_to_fp16 = const()[name = tensor("add_62_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_62_cast_fp16 = add(x = reduce_mean_95_cast_fp16, y = add_62_y_0_to_fp16)[name = tensor("add_62_cast_fp16")]; + tensor sqrt_31_cast_fp16 = sqrt(x = add_62_cast_fp16)[name = tensor("sqrt_31_cast_fp16")]; + tensor real_div_31_cast_fp16 = real_div(x = sub_62_cast_fp16, y = sqrt_31_cast_fp16)[name = tensor("real_div_31_cast_fp16")]; + tensor reshape_125_shape_0 = const()[name = tensor("reshape_125_shape_0"), val = tensor([2, 640, 64, 64])]; + tensor reshape_125_cast_fp16 = reshape(shape = reshape_125_shape_0, x = real_div_31_cast_fp16)[name = tensor("reshape_125_cast_fp16")]; + tensor add_63_gamma_0_to_fp16 = const()[name = tensor("add_63_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4923761408)))]; + tensor add_63_beta_0_to_fp16 = const()[name = tensor("add_63_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4923762752)))]; + tensor add_63_epsilon_0_to_fp16 = const()[name = tensor("add_63_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_63_cast_fp16 = batch_norm(beta = add_63_beta_0_to_fp16, epsilon = add_63_epsilon_0_to_fp16, gamma = add_63_gamma_0_to_fp16, mean = add_11_mean_0_to_fp16, variance = add_11_variance_0_to_fp16, x = reshape_125_cast_fp16)[name = tensor("add_63_cast_fp16")]; + tensor input_733_cast_fp16 = silu(x = add_63_cast_fp16)[name = tensor("input_733_cast_fp16")]; + tensor var_58083 = const()[name = tensor("op_58083"), val = tensor([1, 1])]; + tensor var_58085 = const()[name = tensor("op_58085"), val = tensor([1, 1])]; + tensor hidden_states_503_pad_type_0 = const()[name = tensor("hidden_states_503_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_503_pad_0 = const()[name = tensor("hidden_states_503_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor up_blocks_1_resnets_0_conv2_weight_to_fp16 = const()[name = tensor("up_blocks_1_resnets_0_conv2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4923764096)))]; + tensor up_blocks_1_resnets_0_conv2_bias_to_fp16 = const()[name = tensor("up_blocks_1_resnets_0_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4931136960)))]; + tensor hidden_states_503_cast_fp16 = conv(bias = up_blocks_1_resnets_0_conv2_bias_to_fp16, dilations = var_58085, groups = var_58034, pad = hidden_states_503_pad_0, pad_type = hidden_states_503_pad_type_0, strides = var_58083, weight = up_blocks_1_resnets_0_conv2_weight_to_fp16, x = input_733_cast_fp16)[name = tensor("hidden_states_503_cast_fp16")]; + tensor var_58090 = const()[name = tensor("op_58090"), val = tensor([1, 1])]; + tensor var_58092 = const()[name = tensor("op_58092"), val = tensor([1, 1])]; + tensor x_11_pad_type_0 = const()[name = tensor("x_11_pad_type_0"), val = tensor("custom")]; + tensor x_11_pad_0 = const()[name = tensor("x_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_resnets_0_conv_shortcut_weight_to_fp16 = const()[name = tensor("up_blocks_1_resnets_0_conv_shortcut_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4931138304)))]; + tensor up_blocks_1_resnets_0_conv_shortcut_bias_to_fp16 = const()[name = tensor("up_blocks_1_resnets_0_conv_shortcut_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4933595968)))]; + tensor x_11_cast_fp16 = conv(bias = up_blocks_1_resnets_0_conv_shortcut_bias_to_fp16, dilations = var_58092, groups = var_58034, pad = x_11_pad_0, pad_type = x_11_pad_type_0, strides = var_58090, weight = up_blocks_1_resnets_0_conv_shortcut_weight_to_fp16, x = input_721_cast_fp16)[name = tensor("x_11_cast_fp16")]; + tensor hidden_states_505_cast_fp16 = add(x = x_11_cast_fp16, y = hidden_states_503_cast_fp16)[name = tensor("hidden_states_505_cast_fp16")]; + tensor reshape_128_shape_0 = const()[name = tensor("reshape_128_shape_0"), val = tensor([2, 32, 20, 64, 64])]; + tensor reshape_128_cast_fp16 = reshape(shape = reshape_128_shape_0, x = hidden_states_505_cast_fp16)[name = tensor("reshape_128_cast_fp16")]; + tensor reduce_mean_96_axes_0 = const()[name = tensor("reduce_mean_96_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_96_keep_dims_0 = const()[name = tensor("reduce_mean_96_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_96_cast_fp16 = reduce_mean(axes = reduce_mean_96_axes_0, keep_dims = reduce_mean_96_keep_dims_0, x = reshape_128_cast_fp16)[name = tensor("reduce_mean_96_cast_fp16")]; + tensor sub_64_cast_fp16 = sub(x = reshape_128_cast_fp16, y = reduce_mean_96_cast_fp16)[name = tensor("sub_64_cast_fp16")]; + tensor square_32_cast_fp16 = square(x = sub_64_cast_fp16)[name = tensor("square_32_cast_fp16")]; + tensor reduce_mean_98_axes_0 = const()[name = tensor("reduce_mean_98_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_98_keep_dims_0 = const()[name = tensor("reduce_mean_98_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_98_cast_fp16 = reduce_mean(axes = reduce_mean_98_axes_0, keep_dims = reduce_mean_98_keep_dims_0, x = square_32_cast_fp16)[name = tensor("reduce_mean_98_cast_fp16")]; + tensor add_64_y_0_to_fp16 = const()[name = tensor("add_64_y_0_to_fp16"), val = tensor(0x1.1p-20)]; + tensor add_64_cast_fp16 = add(x = reduce_mean_98_cast_fp16, y = add_64_y_0_to_fp16)[name = tensor("add_64_cast_fp16")]; + tensor sqrt_32_cast_fp16 = sqrt(x = add_64_cast_fp16)[name = tensor("sqrt_32_cast_fp16")]; + tensor real_div_32_cast_fp16 = real_div(x = sub_64_cast_fp16, y = sqrt_32_cast_fp16)[name = tensor("real_div_32_cast_fp16")]; + tensor reshape_129_shape_0 = const()[name = tensor("reshape_129_shape_0"), val = tensor([2, 640, 64, 64])]; + tensor reshape_129_cast_fp16 = reshape(shape = reshape_129_shape_0, x = real_div_32_cast_fp16)[name = tensor("reshape_129_cast_fp16")]; + tensor add_65_gamma_0_to_fp16 = const()[name = tensor("add_65_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4933597312)))]; + tensor add_65_beta_0_to_fp16 = const()[name = tensor("add_65_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4933598656)))]; + tensor add_65_epsilon_0_to_fp16 = const()[name = tensor("add_65_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_65_cast_fp16 = batch_norm(beta = add_65_beta_0_to_fp16, epsilon = add_65_epsilon_0_to_fp16, gamma = add_65_gamma_0_to_fp16, mean = add_11_mean_0_to_fp16, variance = add_11_variance_0_to_fp16, x = reshape_129_cast_fp16)[name = tensor("add_65_cast_fp16")]; + tensor var_58114 = const()[name = tensor("op_58114"), val = tensor([1, 1])]; + tensor var_58116 = const()[name = tensor("op_58116"), val = tensor([1, 1])]; + tensor hidden_states_507_pad_type_0 = const()[name = tensor("hidden_states_507_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_507_pad_0 = const()[name = tensor("hidden_states_507_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_0_proj_in_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_proj_in_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4933600000)))]; + tensor up_blocks_1_attentions_0_proj_in_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_proj_in_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4934419264)))]; + tensor hidden_states_507_cast_fp16 = conv(bias = up_blocks_1_attentions_0_proj_in_bias_to_fp16, dilations = var_58116, groups = var_58034, pad = hidden_states_507_pad_0, pad_type = hidden_states_507_pad_type_0, strides = var_58114, weight = up_blocks_1_attentions_0_proj_in_weight_to_fp16, x = add_65_cast_fp16)[name = tensor("hidden_states_507_cast_fp16")]; + tensor var_58121 = const()[name = tensor("op_58121"), val = tensor([2, 640, 1, 4096])]; + tensor inputs_385_cast_fp16 = reshape(shape = var_58121, x = hidden_states_507_cast_fp16)[name = tensor("inputs_385_cast_fp16")]; + tensor hidden_states_509_axes_0 = const()[name = tensor("hidden_states_509_axes_0"), val = tensor([1])]; + tensor hidden_states_509_gamma_0_to_fp16 = const()[name = tensor("hidden_states_509_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4934420608)))]; + tensor hidden_states_509_beta_0_to_fp16 = const()[name = tensor("hidden_states_509_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4934421952)))]; + tensor var_58137_to_fp16 = const()[name = tensor("op_58137_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_509_cast_fp16 = layer_norm(axes = hidden_states_509_axes_0, beta = hidden_states_509_beta_0_to_fp16, epsilon = var_58137_to_fp16, gamma = hidden_states_509_gamma_0_to_fp16, x = inputs_385_cast_fp16)[name = tensor("hidden_states_509_cast_fp16")]; + tensor var_58152 = const()[name = tensor("op_58152"), val = tensor([1, 1])]; + tensor var_58154 = const()[name = tensor("op_58154"), val = tensor([1, 1])]; + tensor q_257_pad_type_0 = const()[name = tensor("q_257_pad_type_0"), val = tensor("custom")]; + tensor q_257_pad_0 = const()[name = tensor("q_257_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4934423296)))]; + tensor q_257_cast_fp16 = conv(dilations = var_58154, groups = var_58034, pad = q_257_pad_0, pad_type = q_257_pad_type_0, strides = var_58152, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16, x = hidden_states_509_cast_fp16)[name = tensor("q_257_cast_fp16")]; + tensor var_58158 = const()[name = tensor("op_58158"), val = tensor([1, 1])]; + tensor var_58160 = const()[name = tensor("op_58160"), val = tensor([1, 1])]; + tensor k_513_pad_type_0 = const()[name = tensor("k_513_pad_type_0"), val = tensor("custom")]; + tensor k_513_pad_0 = const()[name = tensor("k_513_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4935242560)))]; + tensor k_513_cast_fp16 = conv(dilations = var_58160, groups = var_58034, pad = k_513_pad_0, pad_type = k_513_pad_type_0, strides = var_58158, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16, x = hidden_states_509_cast_fp16)[name = tensor("k_513_cast_fp16")]; + tensor var_58164 = const()[name = tensor("op_58164"), val = tensor([1, 1])]; + tensor var_58166 = const()[name = tensor("op_58166"), val = tensor([1, 1])]; + tensor v_257_pad_type_0 = const()[name = tensor("v_257_pad_type_0"), val = tensor("custom")]; + tensor v_257_pad_0 = const()[name = tensor("v_257_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4936061824)))]; + tensor v_257_cast_fp16 = conv(dilations = var_58166, groups = var_58034, pad = v_257_pad_0, pad_type = v_257_pad_type_0, strides = var_58164, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16, x = hidden_states_509_cast_fp16)[name = tensor("v_257_cast_fp16")]; + tensor var_58170_begin_0 = const()[name = tensor("op_58170_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_58170_end_0 = const()[name = tensor("op_58170_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_58170_end_mask_0 = const()[name = tensor("op_58170_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58170_cast_fp16 = slice_by_index(begin = var_58170_begin_0, end = var_58170_end_0, end_mask = var_58170_end_mask_0, x = q_257_cast_fp16)[name = tensor("op_58170_cast_fp16")]; + tensor var_58174_begin_0 = const()[name = tensor("op_58174_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_58174_end_0 = const()[name = tensor("op_58174_end_0"), val = tensor([2, 128, 1, 4096])]; + tensor var_58174_end_mask_0 = const()[name = tensor("op_58174_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58174_cast_fp16 = slice_by_index(begin = var_58174_begin_0, end = var_58174_end_0, end_mask = var_58174_end_mask_0, x = q_257_cast_fp16)[name = tensor("op_58174_cast_fp16")]; + tensor var_58178_begin_0 = const()[name = tensor("op_58178_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_58178_end_0 = const()[name = tensor("op_58178_end_0"), val = tensor([2, 192, 1, 4096])]; + tensor var_58178_end_mask_0 = const()[name = tensor("op_58178_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58178_cast_fp16 = slice_by_index(begin = var_58178_begin_0, end = var_58178_end_0, end_mask = var_58178_end_mask_0, x = q_257_cast_fp16)[name = tensor("op_58178_cast_fp16")]; + tensor var_58182_begin_0 = const()[name = tensor("op_58182_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_58182_end_0 = const()[name = tensor("op_58182_end_0"), val = tensor([2, 256, 1, 4096])]; + tensor var_58182_end_mask_0 = const()[name = tensor("op_58182_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58182_cast_fp16 = slice_by_index(begin = var_58182_begin_0, end = var_58182_end_0, end_mask = var_58182_end_mask_0, x = q_257_cast_fp16)[name = tensor("op_58182_cast_fp16")]; + tensor var_58186_begin_0 = const()[name = tensor("op_58186_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_58186_end_0 = const()[name = tensor("op_58186_end_0"), val = tensor([2, 320, 1, 4096])]; + tensor var_58186_end_mask_0 = const()[name = tensor("op_58186_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58186_cast_fp16 = slice_by_index(begin = var_58186_begin_0, end = var_58186_end_0, end_mask = var_58186_end_mask_0, x = q_257_cast_fp16)[name = tensor("op_58186_cast_fp16")]; + tensor var_58190_begin_0 = const()[name = tensor("op_58190_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_58190_end_0 = const()[name = tensor("op_58190_end_0"), val = tensor([2, 384, 1, 4096])]; + tensor var_58190_end_mask_0 = const()[name = tensor("op_58190_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58190_cast_fp16 = slice_by_index(begin = var_58190_begin_0, end = var_58190_end_0, end_mask = var_58190_end_mask_0, x = q_257_cast_fp16)[name = tensor("op_58190_cast_fp16")]; + tensor var_58194_begin_0 = const()[name = tensor("op_58194_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_58194_end_0 = const()[name = tensor("op_58194_end_0"), val = tensor([2, 448, 1, 4096])]; + tensor var_58194_end_mask_0 = const()[name = tensor("op_58194_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58194_cast_fp16 = slice_by_index(begin = var_58194_begin_0, end = var_58194_end_0, end_mask = var_58194_end_mask_0, x = q_257_cast_fp16)[name = tensor("op_58194_cast_fp16")]; + tensor var_58198_begin_0 = const()[name = tensor("op_58198_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_58198_end_0 = const()[name = tensor("op_58198_end_0"), val = tensor([2, 512, 1, 4096])]; + tensor var_58198_end_mask_0 = const()[name = tensor("op_58198_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58198_cast_fp16 = slice_by_index(begin = var_58198_begin_0, end = var_58198_end_0, end_mask = var_58198_end_mask_0, x = q_257_cast_fp16)[name = tensor("op_58198_cast_fp16")]; + tensor var_58202_begin_0 = const()[name = tensor("op_58202_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_58202_end_0 = const()[name = tensor("op_58202_end_0"), val = tensor([2, 576, 1, 4096])]; + tensor var_58202_end_mask_0 = const()[name = tensor("op_58202_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58202_cast_fp16 = slice_by_index(begin = var_58202_begin_0, end = var_58202_end_0, end_mask = var_58202_end_mask_0, x = q_257_cast_fp16)[name = tensor("op_58202_cast_fp16")]; + tensor var_58206_begin_0 = const()[name = tensor("op_58206_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_58206_end_0 = const()[name = tensor("op_58206_end_0"), val = tensor([2, 640, 1, 4096])]; + tensor var_58206_end_mask_0 = const()[name = tensor("op_58206_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58206_cast_fp16 = slice_by_index(begin = var_58206_begin_0, end = var_58206_end_0, end_mask = var_58206_end_mask_0, x = q_257_cast_fp16)[name = tensor("op_58206_cast_fp16")]; + tensor k_515_perm_0 = const()[name = tensor("k_515_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_58213_begin_0 = const()[name = tensor("op_58213_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_58213_end_0 = const()[name = tensor("op_58213_end_0"), val = tensor([2, 4096, 1, 64])]; + tensor var_58213_end_mask_0 = const()[name = tensor("op_58213_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_515_cast_fp16 = transpose(perm = k_515_perm_0, x = k_513_cast_fp16)[name = tensor("transpose_11")]; + tensor var_58213_cast_fp16 = slice_by_index(begin = var_58213_begin_0, end = var_58213_end_0, end_mask = var_58213_end_mask_0, x = k_515_cast_fp16)[name = tensor("op_58213_cast_fp16")]; + tensor var_58217_begin_0 = const()[name = tensor("op_58217_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_58217_end_0 = const()[name = tensor("op_58217_end_0"), val = tensor([2, 4096, 1, 128])]; + tensor var_58217_end_mask_0 = const()[name = tensor("op_58217_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_58217_cast_fp16 = slice_by_index(begin = var_58217_begin_0, end = var_58217_end_0, end_mask = var_58217_end_mask_0, x = k_515_cast_fp16)[name = tensor("op_58217_cast_fp16")]; + tensor var_58221_begin_0 = const()[name = tensor("op_58221_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_58221_end_0 = const()[name = tensor("op_58221_end_0"), val = tensor([2, 4096, 1, 192])]; + tensor var_58221_end_mask_0 = const()[name = tensor("op_58221_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_58221_cast_fp16 = slice_by_index(begin = var_58221_begin_0, end = var_58221_end_0, end_mask = var_58221_end_mask_0, x = k_515_cast_fp16)[name = tensor("op_58221_cast_fp16")]; + tensor var_58225_begin_0 = const()[name = tensor("op_58225_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_58225_end_0 = const()[name = tensor("op_58225_end_0"), val = tensor([2, 4096, 1, 256])]; + tensor var_58225_end_mask_0 = const()[name = tensor("op_58225_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_58225_cast_fp16 = slice_by_index(begin = var_58225_begin_0, end = var_58225_end_0, end_mask = var_58225_end_mask_0, x = k_515_cast_fp16)[name = tensor("op_58225_cast_fp16")]; + tensor var_58229_begin_0 = const()[name = tensor("op_58229_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_58229_end_0 = const()[name = tensor("op_58229_end_0"), val = tensor([2, 4096, 1, 320])]; + tensor var_58229_end_mask_0 = const()[name = tensor("op_58229_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_58229_cast_fp16 = slice_by_index(begin = var_58229_begin_0, end = var_58229_end_0, end_mask = var_58229_end_mask_0, x = k_515_cast_fp16)[name = tensor("op_58229_cast_fp16")]; + tensor var_58233_begin_0 = const()[name = tensor("op_58233_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_58233_end_0 = const()[name = tensor("op_58233_end_0"), val = tensor([2, 4096, 1, 384])]; + tensor var_58233_end_mask_0 = const()[name = tensor("op_58233_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_58233_cast_fp16 = slice_by_index(begin = var_58233_begin_0, end = var_58233_end_0, end_mask = var_58233_end_mask_0, x = k_515_cast_fp16)[name = tensor("op_58233_cast_fp16")]; + tensor var_58237_begin_0 = const()[name = tensor("op_58237_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_58237_end_0 = const()[name = tensor("op_58237_end_0"), val = tensor([2, 4096, 1, 448])]; + tensor var_58237_end_mask_0 = const()[name = tensor("op_58237_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_58237_cast_fp16 = slice_by_index(begin = var_58237_begin_0, end = var_58237_end_0, end_mask = var_58237_end_mask_0, x = k_515_cast_fp16)[name = tensor("op_58237_cast_fp16")]; + tensor var_58241_begin_0 = const()[name = tensor("op_58241_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_58241_end_0 = const()[name = tensor("op_58241_end_0"), val = tensor([2, 4096, 1, 512])]; + tensor var_58241_end_mask_0 = const()[name = tensor("op_58241_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_58241_cast_fp16 = slice_by_index(begin = var_58241_begin_0, end = var_58241_end_0, end_mask = var_58241_end_mask_0, x = k_515_cast_fp16)[name = tensor("op_58241_cast_fp16")]; + tensor var_58245_begin_0 = const()[name = tensor("op_58245_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_58245_end_0 = const()[name = tensor("op_58245_end_0"), val = tensor([2, 4096, 1, 576])]; + tensor var_58245_end_mask_0 = const()[name = tensor("op_58245_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_58245_cast_fp16 = slice_by_index(begin = var_58245_begin_0, end = var_58245_end_0, end_mask = var_58245_end_mask_0, x = k_515_cast_fp16)[name = tensor("op_58245_cast_fp16")]; + tensor var_58249_begin_0 = const()[name = tensor("op_58249_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_58249_end_0 = const()[name = tensor("op_58249_end_0"), val = tensor([2, 4096, 1, 640])]; + tensor var_58249_end_mask_0 = const()[name = tensor("op_58249_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_58249_cast_fp16 = slice_by_index(begin = var_58249_begin_0, end = var_58249_end_0, end_mask = var_58249_end_mask_0, x = k_515_cast_fp16)[name = tensor("op_58249_cast_fp16")]; + tensor var_58251_begin_0 = const()[name = tensor("op_58251_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_58251_end_0 = const()[name = tensor("op_58251_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_58251_end_mask_0 = const()[name = tensor("op_58251_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58251_cast_fp16 = slice_by_index(begin = var_58251_begin_0, end = var_58251_end_0, end_mask = var_58251_end_mask_0, x = v_257_cast_fp16)[name = tensor("op_58251_cast_fp16")]; + tensor var_58255_begin_0 = const()[name = tensor("op_58255_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_58255_end_0 = const()[name = tensor("op_58255_end_0"), val = tensor([2, 128, 1, 4096])]; + tensor var_58255_end_mask_0 = const()[name = tensor("op_58255_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58255_cast_fp16 = slice_by_index(begin = var_58255_begin_0, end = var_58255_end_0, end_mask = var_58255_end_mask_0, x = v_257_cast_fp16)[name = tensor("op_58255_cast_fp16")]; + tensor var_58259_begin_0 = const()[name = tensor("op_58259_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_58259_end_0 = const()[name = tensor("op_58259_end_0"), val = tensor([2, 192, 1, 4096])]; + tensor var_58259_end_mask_0 = const()[name = tensor("op_58259_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58259_cast_fp16 = slice_by_index(begin = var_58259_begin_0, end = var_58259_end_0, end_mask = var_58259_end_mask_0, x = v_257_cast_fp16)[name = tensor("op_58259_cast_fp16")]; + tensor var_58263_begin_0 = const()[name = tensor("op_58263_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_58263_end_0 = const()[name = tensor("op_58263_end_0"), val = tensor([2, 256, 1, 4096])]; + tensor var_58263_end_mask_0 = const()[name = tensor("op_58263_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58263_cast_fp16 = slice_by_index(begin = var_58263_begin_0, end = var_58263_end_0, end_mask = var_58263_end_mask_0, x = v_257_cast_fp16)[name = tensor("op_58263_cast_fp16")]; + tensor var_58267_begin_0 = const()[name = tensor("op_58267_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_58267_end_0 = const()[name = tensor("op_58267_end_0"), val = tensor([2, 320, 1, 4096])]; + tensor var_58267_end_mask_0 = const()[name = tensor("op_58267_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58267_cast_fp16 = slice_by_index(begin = var_58267_begin_0, end = var_58267_end_0, end_mask = var_58267_end_mask_0, x = v_257_cast_fp16)[name = tensor("op_58267_cast_fp16")]; + tensor var_58271_begin_0 = const()[name = tensor("op_58271_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_58271_end_0 = const()[name = tensor("op_58271_end_0"), val = tensor([2, 384, 1, 4096])]; + tensor var_58271_end_mask_0 = const()[name = tensor("op_58271_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58271_cast_fp16 = slice_by_index(begin = var_58271_begin_0, end = var_58271_end_0, end_mask = var_58271_end_mask_0, x = v_257_cast_fp16)[name = tensor("op_58271_cast_fp16")]; + tensor var_58275_begin_0 = const()[name = tensor("op_58275_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_58275_end_0 = const()[name = tensor("op_58275_end_0"), val = tensor([2, 448, 1, 4096])]; + tensor var_58275_end_mask_0 = const()[name = tensor("op_58275_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58275_cast_fp16 = slice_by_index(begin = var_58275_begin_0, end = var_58275_end_0, end_mask = var_58275_end_mask_0, x = v_257_cast_fp16)[name = tensor("op_58275_cast_fp16")]; + tensor var_58279_begin_0 = const()[name = tensor("op_58279_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_58279_end_0 = const()[name = tensor("op_58279_end_0"), val = tensor([2, 512, 1, 4096])]; + tensor var_58279_end_mask_0 = const()[name = tensor("op_58279_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58279_cast_fp16 = slice_by_index(begin = var_58279_begin_0, end = var_58279_end_0, end_mask = var_58279_end_mask_0, x = v_257_cast_fp16)[name = tensor("op_58279_cast_fp16")]; + tensor var_58283_begin_0 = const()[name = tensor("op_58283_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_58283_end_0 = const()[name = tensor("op_58283_end_0"), val = tensor([2, 576, 1, 4096])]; + tensor var_58283_end_mask_0 = const()[name = tensor("op_58283_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58283_cast_fp16 = slice_by_index(begin = var_58283_begin_0, end = var_58283_end_0, end_mask = var_58283_end_mask_0, x = v_257_cast_fp16)[name = tensor("op_58283_cast_fp16")]; + tensor var_58287_begin_0 = const()[name = tensor("op_58287_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_58287_end_0 = const()[name = tensor("op_58287_end_0"), val = tensor([2, 640, 1, 4096])]; + tensor var_58287_end_mask_0 = const()[name = tensor("op_58287_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58287_cast_fp16 = slice_by_index(begin = var_58287_begin_0, end = var_58287_end_0, end_mask = var_58287_end_mask_0, x = v_257_cast_fp16)[name = tensor("op_58287_cast_fp16")]; + tensor var_58291_equation_0 = const()[name = tensor("op_58291_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_58291_cast_fp16 = einsum(equation = var_58291_equation_0, values = (var_58213_cast_fp16, var_58170_cast_fp16))[name = tensor("op_58291_cast_fp16")]; + tensor var_58292_to_fp16 = const()[name = tensor("op_58292_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4961_cast_fp16 = mul(x = var_58291_cast_fp16, y = var_58292_to_fp16)[name = tensor("aw_4961_cast_fp16")]; + tensor var_58295_equation_0 = const()[name = tensor("op_58295_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_58295_cast_fp16 = einsum(equation = var_58295_equation_0, values = (var_58217_cast_fp16, var_58174_cast_fp16))[name = tensor("op_58295_cast_fp16")]; + tensor var_58296_to_fp16 = const()[name = tensor("op_58296_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4963_cast_fp16 = mul(x = var_58295_cast_fp16, y = var_58296_to_fp16)[name = tensor("aw_4963_cast_fp16")]; + tensor var_58299_equation_0 = const()[name = tensor("op_58299_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_58299_cast_fp16 = einsum(equation = var_58299_equation_0, values = (var_58221_cast_fp16, var_58178_cast_fp16))[name = tensor("op_58299_cast_fp16")]; + tensor var_58300_to_fp16 = const()[name = tensor("op_58300_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4965_cast_fp16 = mul(x = var_58299_cast_fp16, y = var_58300_to_fp16)[name = tensor("aw_4965_cast_fp16")]; + tensor var_58303_equation_0 = const()[name = tensor("op_58303_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_58303_cast_fp16 = einsum(equation = var_58303_equation_0, values = (var_58225_cast_fp16, var_58182_cast_fp16))[name = tensor("op_58303_cast_fp16")]; + tensor var_58304_to_fp16 = const()[name = tensor("op_58304_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4967_cast_fp16 = mul(x = var_58303_cast_fp16, y = var_58304_to_fp16)[name = tensor("aw_4967_cast_fp16")]; + tensor var_58307_equation_0 = const()[name = tensor("op_58307_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_58307_cast_fp16 = einsum(equation = var_58307_equation_0, values = (var_58229_cast_fp16, var_58186_cast_fp16))[name = tensor("op_58307_cast_fp16")]; + tensor var_58308_to_fp16 = const()[name = tensor("op_58308_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4969_cast_fp16 = mul(x = var_58307_cast_fp16, y = var_58308_to_fp16)[name = tensor("aw_4969_cast_fp16")]; + tensor var_58311_equation_0 = const()[name = tensor("op_58311_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_58311_cast_fp16 = einsum(equation = var_58311_equation_0, values = (var_58233_cast_fp16, var_58190_cast_fp16))[name = tensor("op_58311_cast_fp16")]; + tensor var_58312_to_fp16 = const()[name = tensor("op_58312_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4971_cast_fp16 = mul(x = var_58311_cast_fp16, y = var_58312_to_fp16)[name = tensor("aw_4971_cast_fp16")]; + tensor var_58315_equation_0 = const()[name = tensor("op_58315_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_58315_cast_fp16 = einsum(equation = var_58315_equation_0, values = (var_58237_cast_fp16, var_58194_cast_fp16))[name = tensor("op_58315_cast_fp16")]; + tensor var_58316_to_fp16 = const()[name = tensor("op_58316_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4973_cast_fp16 = mul(x = var_58315_cast_fp16, y = var_58316_to_fp16)[name = tensor("aw_4973_cast_fp16")]; + tensor var_58319_equation_0 = const()[name = tensor("op_58319_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_58319_cast_fp16 = einsum(equation = var_58319_equation_0, values = (var_58241_cast_fp16, var_58198_cast_fp16))[name = tensor("op_58319_cast_fp16")]; + tensor var_58320_to_fp16 = const()[name = tensor("op_58320_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4975_cast_fp16 = mul(x = var_58319_cast_fp16, y = var_58320_to_fp16)[name = tensor("aw_4975_cast_fp16")]; + tensor var_58323_equation_0 = const()[name = tensor("op_58323_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_58323_cast_fp16 = einsum(equation = var_58323_equation_0, values = (var_58245_cast_fp16, var_58202_cast_fp16))[name = tensor("op_58323_cast_fp16")]; + tensor var_58324_to_fp16 = const()[name = tensor("op_58324_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4977_cast_fp16 = mul(x = var_58323_cast_fp16, y = var_58324_to_fp16)[name = tensor("aw_4977_cast_fp16")]; + tensor var_58327_equation_0 = const()[name = tensor("op_58327_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_58327_cast_fp16 = einsum(equation = var_58327_equation_0, values = (var_58249_cast_fp16, var_58206_cast_fp16))[name = tensor("op_58327_cast_fp16")]; + tensor var_58328_to_fp16 = const()[name = tensor("op_58328_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4979_cast_fp16 = mul(x = var_58327_cast_fp16, y = var_58328_to_fp16)[name = tensor("aw_4979_cast_fp16")]; + tensor var_58330_cast_fp16 = softmax(axis = var_58034, x = aw_4961_cast_fp16)[name = tensor("op_58330_cast_fp16")]; + tensor var_58331_cast_fp16 = softmax(axis = var_58034, x = aw_4963_cast_fp16)[name = tensor("op_58331_cast_fp16")]; + tensor var_58332_cast_fp16 = softmax(axis = var_58034, x = aw_4965_cast_fp16)[name = tensor("op_58332_cast_fp16")]; + tensor var_58333_cast_fp16 = softmax(axis = var_58034, x = aw_4967_cast_fp16)[name = tensor("op_58333_cast_fp16")]; + tensor var_58334_cast_fp16 = softmax(axis = var_58034, x = aw_4969_cast_fp16)[name = tensor("op_58334_cast_fp16")]; + tensor var_58335_cast_fp16 = softmax(axis = var_58034, x = aw_4971_cast_fp16)[name = tensor("op_58335_cast_fp16")]; + tensor var_58336_cast_fp16 = softmax(axis = var_58034, x = aw_4973_cast_fp16)[name = tensor("op_58336_cast_fp16")]; + tensor var_58337_cast_fp16 = softmax(axis = var_58034, x = aw_4975_cast_fp16)[name = tensor("op_58337_cast_fp16")]; + tensor var_58338_cast_fp16 = softmax(axis = var_58034, x = aw_4977_cast_fp16)[name = tensor("op_58338_cast_fp16")]; + tensor var_58339_cast_fp16 = softmax(axis = var_58034, x = aw_4979_cast_fp16)[name = tensor("op_58339_cast_fp16")]; + tensor var_58341_equation_0 = const()[name = tensor("op_58341_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_58341_cast_fp16 = einsum(equation = var_58341_equation_0, values = (var_58251_cast_fp16, var_58330_cast_fp16))[name = tensor("op_58341_cast_fp16")]; + tensor var_58343_equation_0 = const()[name = tensor("op_58343_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_58343_cast_fp16 = einsum(equation = var_58343_equation_0, values = (var_58255_cast_fp16, var_58331_cast_fp16))[name = tensor("op_58343_cast_fp16")]; + tensor var_58345_equation_0 = const()[name = tensor("op_58345_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_58345_cast_fp16 = einsum(equation = var_58345_equation_0, values = (var_58259_cast_fp16, var_58332_cast_fp16))[name = tensor("op_58345_cast_fp16")]; + tensor var_58347_equation_0 = const()[name = tensor("op_58347_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_58347_cast_fp16 = einsum(equation = var_58347_equation_0, values = (var_58263_cast_fp16, var_58333_cast_fp16))[name = tensor("op_58347_cast_fp16")]; + tensor var_58349_equation_0 = const()[name = tensor("op_58349_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_58349_cast_fp16 = einsum(equation = var_58349_equation_0, values = (var_58267_cast_fp16, var_58334_cast_fp16))[name = tensor("op_58349_cast_fp16")]; + tensor var_58351_equation_0 = const()[name = tensor("op_58351_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_58351_cast_fp16 = einsum(equation = var_58351_equation_0, values = (var_58271_cast_fp16, var_58335_cast_fp16))[name = tensor("op_58351_cast_fp16")]; + tensor var_58353_equation_0 = const()[name = tensor("op_58353_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_58353_cast_fp16 = einsum(equation = var_58353_equation_0, values = (var_58275_cast_fp16, var_58336_cast_fp16))[name = tensor("op_58353_cast_fp16")]; + tensor var_58355_equation_0 = const()[name = tensor("op_58355_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_58355_cast_fp16 = einsum(equation = var_58355_equation_0, values = (var_58279_cast_fp16, var_58337_cast_fp16))[name = tensor("op_58355_cast_fp16")]; + tensor var_58357_equation_0 = const()[name = tensor("op_58357_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_58357_cast_fp16 = einsum(equation = var_58357_equation_0, values = (var_58283_cast_fp16, var_58338_cast_fp16))[name = tensor("op_58357_cast_fp16")]; + tensor var_58359_equation_0 = const()[name = tensor("op_58359_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_58359_cast_fp16 = einsum(equation = var_58359_equation_0, values = (var_58287_cast_fp16, var_58339_cast_fp16))[name = tensor("op_58359_cast_fp16")]; + tensor input_737_interleave_0 = const()[name = tensor("input_737_interleave_0"), val = tensor(false)]; + tensor input_737_cast_fp16 = concat(axis = var_58034, interleave = input_737_interleave_0, values = (var_58341_cast_fp16, var_58343_cast_fp16, var_58345_cast_fp16, var_58347_cast_fp16, var_58349_cast_fp16, var_58351_cast_fp16, var_58353_cast_fp16, var_58355_cast_fp16, var_58357_cast_fp16, var_58359_cast_fp16))[name = tensor("input_737_cast_fp16")]; + tensor var_58365 = const()[name = tensor("op_58365"), val = tensor([1, 1])]; + tensor var_58367 = const()[name = tensor("op_58367"), val = tensor([1, 1])]; + tensor var_58369_pad_type_0 = const()[name = tensor("op_58369_pad_type_0"), val = tensor("custom")]; + tensor var_58369_pad_0 = const()[name = tensor("op_58369_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4936881088)))]; + tensor up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4937700352)))]; + tensor var_58369_cast_fp16 = conv(bias = up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_58367, groups = var_58034, pad = var_58369_pad_0, pad_type = var_58369_pad_type_0, strides = var_58365, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16, x = input_737_cast_fp16)[name = tensor("op_58369_cast_fp16")]; + tensor inputs_387_cast_fp16 = add(x = var_58369_cast_fp16, y = inputs_385_cast_fp16)[name = tensor("inputs_387_cast_fp16")]; + tensor hidden_states_511_axes_0 = const()[name = tensor("hidden_states_511_axes_0"), val = tensor([1])]; + tensor hidden_states_511_gamma_0_to_fp16 = const()[name = tensor("hidden_states_511_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4937701696)))]; + tensor hidden_states_511_beta_0_to_fp16 = const()[name = tensor("hidden_states_511_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4937703040)))]; + tensor var_58379_to_fp16 = const()[name = tensor("op_58379_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_511_cast_fp16 = layer_norm(axes = hidden_states_511_axes_0, beta = hidden_states_511_beta_0_to_fp16, epsilon = var_58379_to_fp16, gamma = hidden_states_511_gamma_0_to_fp16, x = inputs_387_cast_fp16)[name = tensor("hidden_states_511_cast_fp16")]; + tensor var_58394 = const()[name = tensor("op_58394"), val = tensor([1, 1])]; + tensor var_58396 = const()[name = tensor("op_58396"), val = tensor([1, 1])]; + tensor q_259_pad_type_0 = const()[name = tensor("q_259_pad_type_0"), val = tensor("custom")]; + tensor q_259_pad_0 = const()[name = tensor("q_259_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4937704384)))]; + tensor q_259_cast_fp16 = conv(dilations = var_58396, groups = var_58034, pad = q_259_pad_0, pad_type = q_259_pad_type_0, strides = var_58394, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16, x = hidden_states_511_cast_fp16)[name = tensor("q_259_cast_fp16")]; + tensor var_58400 = const()[name = tensor("op_58400"), val = tensor([1, 1])]; + tensor var_58402 = const()[name = tensor("op_58402"), val = tensor([1, 1])]; + tensor k_517_pad_type_0 = const()[name = tensor("k_517_pad_type_0"), val = tensor("custom")]; + tensor k_517_pad_0 = const()[name = tensor("k_517_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4938523648)))]; + tensor k_517_cast_fp16 = conv(dilations = var_58402, groups = var_58034, pad = k_517_pad_0, pad_type = k_517_pad_type_0, strides = var_58400, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_517_cast_fp16")]; + tensor var_58406 = const()[name = tensor("op_58406"), val = tensor([1, 1])]; + tensor var_58408 = const()[name = tensor("op_58408"), val = tensor([1, 1])]; + tensor v_259_pad_type_0 = const()[name = tensor("v_259_pad_type_0"), val = tensor("custom")]; + tensor v_259_pad_0 = const()[name = tensor("v_259_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4941145152)))]; + tensor v_259_cast_fp16 = conv(dilations = var_58408, groups = var_58034, pad = v_259_pad_0, pad_type = v_259_pad_type_0, strides = var_58406, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_259_cast_fp16")]; + tensor var_58412_begin_0 = const()[name = tensor("op_58412_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_58412_end_0 = const()[name = tensor("op_58412_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_58412_end_mask_0 = const()[name = tensor("op_58412_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58412_cast_fp16 = slice_by_index(begin = var_58412_begin_0, end = var_58412_end_0, end_mask = var_58412_end_mask_0, x = q_259_cast_fp16)[name = tensor("op_58412_cast_fp16")]; + tensor var_58416_begin_0 = const()[name = tensor("op_58416_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_58416_end_0 = const()[name = tensor("op_58416_end_0"), val = tensor([2, 128, 1, 4096])]; + tensor var_58416_end_mask_0 = const()[name = tensor("op_58416_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58416_cast_fp16 = slice_by_index(begin = var_58416_begin_0, end = var_58416_end_0, end_mask = var_58416_end_mask_0, x = q_259_cast_fp16)[name = tensor("op_58416_cast_fp16")]; + tensor var_58420_begin_0 = const()[name = tensor("op_58420_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_58420_end_0 = const()[name = tensor("op_58420_end_0"), val = tensor([2, 192, 1, 4096])]; + tensor var_58420_end_mask_0 = const()[name = tensor("op_58420_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58420_cast_fp16 = slice_by_index(begin = var_58420_begin_0, end = var_58420_end_0, end_mask = var_58420_end_mask_0, x = q_259_cast_fp16)[name = tensor("op_58420_cast_fp16")]; + tensor var_58424_begin_0 = const()[name = tensor("op_58424_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_58424_end_0 = const()[name = tensor("op_58424_end_0"), val = tensor([2, 256, 1, 4096])]; + tensor var_58424_end_mask_0 = const()[name = tensor("op_58424_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58424_cast_fp16 = slice_by_index(begin = var_58424_begin_0, end = var_58424_end_0, end_mask = var_58424_end_mask_0, x = q_259_cast_fp16)[name = tensor("op_58424_cast_fp16")]; + tensor var_58428_begin_0 = const()[name = tensor("op_58428_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_58428_end_0 = const()[name = tensor("op_58428_end_0"), val = tensor([2, 320, 1, 4096])]; + tensor var_58428_end_mask_0 = const()[name = tensor("op_58428_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58428_cast_fp16 = slice_by_index(begin = var_58428_begin_0, end = var_58428_end_0, end_mask = var_58428_end_mask_0, x = q_259_cast_fp16)[name = tensor("op_58428_cast_fp16")]; + tensor var_58432_begin_0 = const()[name = tensor("op_58432_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_58432_end_0 = const()[name = tensor("op_58432_end_0"), val = tensor([2, 384, 1, 4096])]; + tensor var_58432_end_mask_0 = const()[name = tensor("op_58432_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58432_cast_fp16 = slice_by_index(begin = var_58432_begin_0, end = var_58432_end_0, end_mask = var_58432_end_mask_0, x = q_259_cast_fp16)[name = tensor("op_58432_cast_fp16")]; + tensor var_58436_begin_0 = const()[name = tensor("op_58436_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_58436_end_0 = const()[name = tensor("op_58436_end_0"), val = tensor([2, 448, 1, 4096])]; + tensor var_58436_end_mask_0 = const()[name = tensor("op_58436_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58436_cast_fp16 = slice_by_index(begin = var_58436_begin_0, end = var_58436_end_0, end_mask = var_58436_end_mask_0, x = q_259_cast_fp16)[name = tensor("op_58436_cast_fp16")]; + tensor var_58440_begin_0 = const()[name = tensor("op_58440_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_58440_end_0 = const()[name = tensor("op_58440_end_0"), val = tensor([2, 512, 1, 4096])]; + tensor var_58440_end_mask_0 = const()[name = tensor("op_58440_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58440_cast_fp16 = slice_by_index(begin = var_58440_begin_0, end = var_58440_end_0, end_mask = var_58440_end_mask_0, x = q_259_cast_fp16)[name = tensor("op_58440_cast_fp16")]; + tensor var_58444_begin_0 = const()[name = tensor("op_58444_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_58444_end_0 = const()[name = tensor("op_58444_end_0"), val = tensor([2, 576, 1, 4096])]; + tensor var_58444_end_mask_0 = const()[name = tensor("op_58444_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58444_cast_fp16 = slice_by_index(begin = var_58444_begin_0, end = var_58444_end_0, end_mask = var_58444_end_mask_0, x = q_259_cast_fp16)[name = tensor("op_58444_cast_fp16")]; + tensor var_58448_begin_0 = const()[name = tensor("op_58448_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_58448_end_0 = const()[name = tensor("op_58448_end_0"), val = tensor([2, 640, 1, 4096])]; + tensor var_58448_end_mask_0 = const()[name = tensor("op_58448_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58448_cast_fp16 = slice_by_index(begin = var_58448_begin_0, end = var_58448_end_0, end_mask = var_58448_end_mask_0, x = q_259_cast_fp16)[name = tensor("op_58448_cast_fp16")]; + tensor k_519_perm_0 = const()[name = tensor("k_519_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_58455_begin_0 = const()[name = tensor("op_58455_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_58455_end_0 = const()[name = tensor("op_58455_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_58455_end_mask_0 = const()[name = tensor("op_58455_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_519_cast_fp16 = transpose(perm = k_519_perm_0, x = k_517_cast_fp16)[name = tensor("transpose_10")]; + tensor var_58455_cast_fp16 = slice_by_index(begin = var_58455_begin_0, end = var_58455_end_0, end_mask = var_58455_end_mask_0, x = k_519_cast_fp16)[name = tensor("op_58455_cast_fp16")]; + tensor var_58459_begin_0 = const()[name = tensor("op_58459_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_58459_end_0 = const()[name = tensor("op_58459_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_58459_end_mask_0 = const()[name = tensor("op_58459_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_58459_cast_fp16 = slice_by_index(begin = var_58459_begin_0, end = var_58459_end_0, end_mask = var_58459_end_mask_0, x = k_519_cast_fp16)[name = tensor("op_58459_cast_fp16")]; + tensor var_58463_begin_0 = const()[name = tensor("op_58463_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_58463_end_0 = const()[name = tensor("op_58463_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_58463_end_mask_0 = const()[name = tensor("op_58463_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_58463_cast_fp16 = slice_by_index(begin = var_58463_begin_0, end = var_58463_end_0, end_mask = var_58463_end_mask_0, x = k_519_cast_fp16)[name = tensor("op_58463_cast_fp16")]; + tensor var_58467_begin_0 = const()[name = tensor("op_58467_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_58467_end_0 = const()[name = tensor("op_58467_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_58467_end_mask_0 = const()[name = tensor("op_58467_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_58467_cast_fp16 = slice_by_index(begin = var_58467_begin_0, end = var_58467_end_0, end_mask = var_58467_end_mask_0, x = k_519_cast_fp16)[name = tensor("op_58467_cast_fp16")]; + tensor var_58471_begin_0 = const()[name = tensor("op_58471_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_58471_end_0 = const()[name = tensor("op_58471_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_58471_end_mask_0 = const()[name = tensor("op_58471_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_58471_cast_fp16 = slice_by_index(begin = var_58471_begin_0, end = var_58471_end_0, end_mask = var_58471_end_mask_0, x = k_519_cast_fp16)[name = tensor("op_58471_cast_fp16")]; + tensor var_58475_begin_0 = const()[name = tensor("op_58475_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_58475_end_0 = const()[name = tensor("op_58475_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_58475_end_mask_0 = const()[name = tensor("op_58475_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_58475_cast_fp16 = slice_by_index(begin = var_58475_begin_0, end = var_58475_end_0, end_mask = var_58475_end_mask_0, x = k_519_cast_fp16)[name = tensor("op_58475_cast_fp16")]; + tensor var_58479_begin_0 = const()[name = tensor("op_58479_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_58479_end_0 = const()[name = tensor("op_58479_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_58479_end_mask_0 = const()[name = tensor("op_58479_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_58479_cast_fp16 = slice_by_index(begin = var_58479_begin_0, end = var_58479_end_0, end_mask = var_58479_end_mask_0, x = k_519_cast_fp16)[name = tensor("op_58479_cast_fp16")]; + tensor var_58483_begin_0 = const()[name = tensor("op_58483_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_58483_end_0 = const()[name = tensor("op_58483_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_58483_end_mask_0 = const()[name = tensor("op_58483_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_58483_cast_fp16 = slice_by_index(begin = var_58483_begin_0, end = var_58483_end_0, end_mask = var_58483_end_mask_0, x = k_519_cast_fp16)[name = tensor("op_58483_cast_fp16")]; + tensor var_58487_begin_0 = const()[name = tensor("op_58487_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_58487_end_0 = const()[name = tensor("op_58487_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_58487_end_mask_0 = const()[name = tensor("op_58487_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_58487_cast_fp16 = slice_by_index(begin = var_58487_begin_0, end = var_58487_end_0, end_mask = var_58487_end_mask_0, x = k_519_cast_fp16)[name = tensor("op_58487_cast_fp16")]; + tensor var_58491_begin_0 = const()[name = tensor("op_58491_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_58491_end_0 = const()[name = tensor("op_58491_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_58491_end_mask_0 = const()[name = tensor("op_58491_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_58491_cast_fp16 = slice_by_index(begin = var_58491_begin_0, end = var_58491_end_0, end_mask = var_58491_end_mask_0, x = k_519_cast_fp16)[name = tensor("op_58491_cast_fp16")]; + tensor var_58493_begin_0 = const()[name = tensor("op_58493_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_58493_end_0 = const()[name = tensor("op_58493_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_58493_end_mask_0 = const()[name = tensor("op_58493_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58493_cast_fp16 = slice_by_index(begin = var_58493_begin_0, end = var_58493_end_0, end_mask = var_58493_end_mask_0, x = v_259_cast_fp16)[name = tensor("op_58493_cast_fp16")]; + tensor var_58497_begin_0 = const()[name = tensor("op_58497_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_58497_end_0 = const()[name = tensor("op_58497_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_58497_end_mask_0 = const()[name = tensor("op_58497_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58497_cast_fp16 = slice_by_index(begin = var_58497_begin_0, end = var_58497_end_0, end_mask = var_58497_end_mask_0, x = v_259_cast_fp16)[name = tensor("op_58497_cast_fp16")]; + tensor var_58501_begin_0 = const()[name = tensor("op_58501_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_58501_end_0 = const()[name = tensor("op_58501_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_58501_end_mask_0 = const()[name = tensor("op_58501_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58501_cast_fp16 = slice_by_index(begin = var_58501_begin_0, end = var_58501_end_0, end_mask = var_58501_end_mask_0, x = v_259_cast_fp16)[name = tensor("op_58501_cast_fp16")]; + tensor var_58505_begin_0 = const()[name = tensor("op_58505_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_58505_end_0 = const()[name = tensor("op_58505_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_58505_end_mask_0 = const()[name = tensor("op_58505_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58505_cast_fp16 = slice_by_index(begin = var_58505_begin_0, end = var_58505_end_0, end_mask = var_58505_end_mask_0, x = v_259_cast_fp16)[name = tensor("op_58505_cast_fp16")]; + tensor var_58509_begin_0 = const()[name = tensor("op_58509_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_58509_end_0 = const()[name = tensor("op_58509_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_58509_end_mask_0 = const()[name = tensor("op_58509_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58509_cast_fp16 = slice_by_index(begin = var_58509_begin_0, end = var_58509_end_0, end_mask = var_58509_end_mask_0, x = v_259_cast_fp16)[name = tensor("op_58509_cast_fp16")]; + tensor var_58513_begin_0 = const()[name = tensor("op_58513_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_58513_end_0 = const()[name = tensor("op_58513_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_58513_end_mask_0 = const()[name = tensor("op_58513_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58513_cast_fp16 = slice_by_index(begin = var_58513_begin_0, end = var_58513_end_0, end_mask = var_58513_end_mask_0, x = v_259_cast_fp16)[name = tensor("op_58513_cast_fp16")]; + tensor var_58517_begin_0 = const()[name = tensor("op_58517_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_58517_end_0 = const()[name = tensor("op_58517_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_58517_end_mask_0 = const()[name = tensor("op_58517_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58517_cast_fp16 = slice_by_index(begin = var_58517_begin_0, end = var_58517_end_0, end_mask = var_58517_end_mask_0, x = v_259_cast_fp16)[name = tensor("op_58517_cast_fp16")]; + tensor var_58521_begin_0 = const()[name = tensor("op_58521_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_58521_end_0 = const()[name = tensor("op_58521_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_58521_end_mask_0 = const()[name = tensor("op_58521_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58521_cast_fp16 = slice_by_index(begin = var_58521_begin_0, end = var_58521_end_0, end_mask = var_58521_end_mask_0, x = v_259_cast_fp16)[name = tensor("op_58521_cast_fp16")]; + tensor var_58525_begin_0 = const()[name = tensor("op_58525_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_58525_end_0 = const()[name = tensor("op_58525_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_58525_end_mask_0 = const()[name = tensor("op_58525_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58525_cast_fp16 = slice_by_index(begin = var_58525_begin_0, end = var_58525_end_0, end_mask = var_58525_end_mask_0, x = v_259_cast_fp16)[name = tensor("op_58525_cast_fp16")]; + tensor var_58529_begin_0 = const()[name = tensor("op_58529_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_58529_end_0 = const()[name = tensor("op_58529_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_58529_end_mask_0 = const()[name = tensor("op_58529_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58529_cast_fp16 = slice_by_index(begin = var_58529_begin_0, end = var_58529_end_0, end_mask = var_58529_end_mask_0, x = v_259_cast_fp16)[name = tensor("op_58529_cast_fp16")]; + tensor var_58533_equation_0 = const()[name = tensor("op_58533_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_58533_cast_fp16 = einsum(equation = var_58533_equation_0, values = (var_58455_cast_fp16, var_58412_cast_fp16))[name = tensor("op_58533_cast_fp16")]; + tensor var_58534_to_fp16 = const()[name = tensor("op_58534_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4981_cast_fp16 = mul(x = var_58533_cast_fp16, y = var_58534_to_fp16)[name = tensor("aw_4981_cast_fp16")]; + tensor var_58537_equation_0 = const()[name = tensor("op_58537_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_58537_cast_fp16 = einsum(equation = var_58537_equation_0, values = (var_58459_cast_fp16, var_58416_cast_fp16))[name = tensor("op_58537_cast_fp16")]; + tensor var_58538_to_fp16 = const()[name = tensor("op_58538_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4983_cast_fp16 = mul(x = var_58537_cast_fp16, y = var_58538_to_fp16)[name = tensor("aw_4983_cast_fp16")]; + tensor var_58541_equation_0 = const()[name = tensor("op_58541_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_58541_cast_fp16 = einsum(equation = var_58541_equation_0, values = (var_58463_cast_fp16, var_58420_cast_fp16))[name = tensor("op_58541_cast_fp16")]; + tensor var_58542_to_fp16 = const()[name = tensor("op_58542_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4985_cast_fp16 = mul(x = var_58541_cast_fp16, y = var_58542_to_fp16)[name = tensor("aw_4985_cast_fp16")]; + tensor var_58545_equation_0 = const()[name = tensor("op_58545_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_58545_cast_fp16 = einsum(equation = var_58545_equation_0, values = (var_58467_cast_fp16, var_58424_cast_fp16))[name = tensor("op_58545_cast_fp16")]; + tensor var_58546_to_fp16 = const()[name = tensor("op_58546_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4987_cast_fp16 = mul(x = var_58545_cast_fp16, y = var_58546_to_fp16)[name = tensor("aw_4987_cast_fp16")]; + tensor var_58549_equation_0 = const()[name = tensor("op_58549_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_58549_cast_fp16 = einsum(equation = var_58549_equation_0, values = (var_58471_cast_fp16, var_58428_cast_fp16))[name = tensor("op_58549_cast_fp16")]; + tensor var_58550_to_fp16 = const()[name = tensor("op_58550_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4989_cast_fp16 = mul(x = var_58549_cast_fp16, y = var_58550_to_fp16)[name = tensor("aw_4989_cast_fp16")]; + tensor var_58553_equation_0 = const()[name = tensor("op_58553_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_58553_cast_fp16 = einsum(equation = var_58553_equation_0, values = (var_58475_cast_fp16, var_58432_cast_fp16))[name = tensor("op_58553_cast_fp16")]; + tensor var_58554_to_fp16 = const()[name = tensor("op_58554_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4991_cast_fp16 = mul(x = var_58553_cast_fp16, y = var_58554_to_fp16)[name = tensor("aw_4991_cast_fp16")]; + tensor var_58557_equation_0 = const()[name = tensor("op_58557_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_58557_cast_fp16 = einsum(equation = var_58557_equation_0, values = (var_58479_cast_fp16, var_58436_cast_fp16))[name = tensor("op_58557_cast_fp16")]; + tensor var_58558_to_fp16 = const()[name = tensor("op_58558_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4993_cast_fp16 = mul(x = var_58557_cast_fp16, y = var_58558_to_fp16)[name = tensor("aw_4993_cast_fp16")]; + tensor var_58561_equation_0 = const()[name = tensor("op_58561_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_58561_cast_fp16 = einsum(equation = var_58561_equation_0, values = (var_58483_cast_fp16, var_58440_cast_fp16))[name = tensor("op_58561_cast_fp16")]; + tensor var_58562_to_fp16 = const()[name = tensor("op_58562_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4995_cast_fp16 = mul(x = var_58561_cast_fp16, y = var_58562_to_fp16)[name = tensor("aw_4995_cast_fp16")]; + tensor var_58565_equation_0 = const()[name = tensor("op_58565_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_58565_cast_fp16 = einsum(equation = var_58565_equation_0, values = (var_58487_cast_fp16, var_58444_cast_fp16))[name = tensor("op_58565_cast_fp16")]; + tensor var_58566_to_fp16 = const()[name = tensor("op_58566_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4997_cast_fp16 = mul(x = var_58565_cast_fp16, y = var_58566_to_fp16)[name = tensor("aw_4997_cast_fp16")]; + tensor var_58569_equation_0 = const()[name = tensor("op_58569_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_58569_cast_fp16 = einsum(equation = var_58569_equation_0, values = (var_58491_cast_fp16, var_58448_cast_fp16))[name = tensor("op_58569_cast_fp16")]; + tensor var_58570_to_fp16 = const()[name = tensor("op_58570_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_4999_cast_fp16 = mul(x = var_58569_cast_fp16, y = var_58570_to_fp16)[name = tensor("aw_4999_cast_fp16")]; + tensor var_58572_cast_fp16 = softmax(axis = var_58034, x = aw_4981_cast_fp16)[name = tensor("op_58572_cast_fp16")]; + tensor var_58573_cast_fp16 = softmax(axis = var_58034, x = aw_4983_cast_fp16)[name = tensor("op_58573_cast_fp16")]; + tensor var_58574_cast_fp16 = softmax(axis = var_58034, x = aw_4985_cast_fp16)[name = tensor("op_58574_cast_fp16")]; + tensor var_58575_cast_fp16 = softmax(axis = var_58034, x = aw_4987_cast_fp16)[name = tensor("op_58575_cast_fp16")]; + tensor var_58576_cast_fp16 = softmax(axis = var_58034, x = aw_4989_cast_fp16)[name = tensor("op_58576_cast_fp16")]; + tensor var_58577_cast_fp16 = softmax(axis = var_58034, x = aw_4991_cast_fp16)[name = tensor("op_58577_cast_fp16")]; + tensor var_58578_cast_fp16 = softmax(axis = var_58034, x = aw_4993_cast_fp16)[name = tensor("op_58578_cast_fp16")]; + tensor var_58579_cast_fp16 = softmax(axis = var_58034, x = aw_4995_cast_fp16)[name = tensor("op_58579_cast_fp16")]; + tensor var_58580_cast_fp16 = softmax(axis = var_58034, x = aw_4997_cast_fp16)[name = tensor("op_58580_cast_fp16")]; + tensor var_58581_cast_fp16 = softmax(axis = var_58034, x = aw_4999_cast_fp16)[name = tensor("op_58581_cast_fp16")]; + tensor var_58583_equation_0 = const()[name = tensor("op_58583_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_58583_cast_fp16 = einsum(equation = var_58583_equation_0, values = (var_58493_cast_fp16, var_58572_cast_fp16))[name = tensor("op_58583_cast_fp16")]; + tensor var_58585_equation_0 = const()[name = tensor("op_58585_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_58585_cast_fp16 = einsum(equation = var_58585_equation_0, values = (var_58497_cast_fp16, var_58573_cast_fp16))[name = tensor("op_58585_cast_fp16")]; + tensor var_58587_equation_0 = const()[name = tensor("op_58587_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_58587_cast_fp16 = einsum(equation = var_58587_equation_0, values = (var_58501_cast_fp16, var_58574_cast_fp16))[name = tensor("op_58587_cast_fp16")]; + tensor var_58589_equation_0 = const()[name = tensor("op_58589_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_58589_cast_fp16 = einsum(equation = var_58589_equation_0, values = (var_58505_cast_fp16, var_58575_cast_fp16))[name = tensor("op_58589_cast_fp16")]; + tensor var_58591_equation_0 = const()[name = tensor("op_58591_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_58591_cast_fp16 = einsum(equation = var_58591_equation_0, values = (var_58509_cast_fp16, var_58576_cast_fp16))[name = tensor("op_58591_cast_fp16")]; + tensor var_58593_equation_0 = const()[name = tensor("op_58593_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_58593_cast_fp16 = einsum(equation = var_58593_equation_0, values = (var_58513_cast_fp16, var_58577_cast_fp16))[name = tensor("op_58593_cast_fp16")]; + tensor var_58595_equation_0 = const()[name = tensor("op_58595_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_58595_cast_fp16 = einsum(equation = var_58595_equation_0, values = (var_58517_cast_fp16, var_58578_cast_fp16))[name = tensor("op_58595_cast_fp16")]; + tensor var_58597_equation_0 = const()[name = tensor("op_58597_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_58597_cast_fp16 = einsum(equation = var_58597_equation_0, values = (var_58521_cast_fp16, var_58579_cast_fp16))[name = tensor("op_58597_cast_fp16")]; + tensor var_58599_equation_0 = const()[name = tensor("op_58599_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_58599_cast_fp16 = einsum(equation = var_58599_equation_0, values = (var_58525_cast_fp16, var_58580_cast_fp16))[name = tensor("op_58599_cast_fp16")]; + tensor var_58601_equation_0 = const()[name = tensor("op_58601_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_58601_cast_fp16 = einsum(equation = var_58601_equation_0, values = (var_58529_cast_fp16, var_58581_cast_fp16))[name = tensor("op_58601_cast_fp16")]; + tensor input_739_interleave_0 = const()[name = tensor("input_739_interleave_0"), val = tensor(false)]; + tensor input_739_cast_fp16 = concat(axis = var_58034, interleave = input_739_interleave_0, values = (var_58583_cast_fp16, var_58585_cast_fp16, var_58587_cast_fp16, var_58589_cast_fp16, var_58591_cast_fp16, var_58593_cast_fp16, var_58595_cast_fp16, var_58597_cast_fp16, var_58599_cast_fp16, var_58601_cast_fp16))[name = tensor("input_739_cast_fp16")]; + tensor var_58607 = const()[name = tensor("op_58607"), val = tensor([1, 1])]; + tensor var_58609 = const()[name = tensor("op_58609"), val = tensor([1, 1])]; + tensor var_58611_pad_type_0 = const()[name = tensor("op_58611_pad_type_0"), val = tensor("custom")]; + tensor var_58611_pad_0 = const()[name = tensor("op_58611_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4943766656)))]; + tensor up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4944585920)))]; + tensor var_58611_cast_fp16 = conv(bias = up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_58609, groups = var_58034, pad = var_58611_pad_0, pad_type = var_58611_pad_type_0, strides = var_58607, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16, x = input_739_cast_fp16)[name = tensor("op_58611_cast_fp16")]; + tensor inputs_389_cast_fp16 = add(x = var_58611_cast_fp16, y = inputs_387_cast_fp16)[name = tensor("inputs_389_cast_fp16")]; + tensor input_741_axes_0 = const()[name = tensor("input_741_axes_0"), val = tensor([1])]; + tensor input_741_gamma_0_to_fp16 = const()[name = tensor("input_741_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4944587264)))]; + tensor input_741_beta_0_to_fp16 = const()[name = tensor("input_741_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4944588608)))]; + tensor var_58621_to_fp16 = const()[name = tensor("op_58621_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_741_cast_fp16 = layer_norm(axes = input_741_axes_0, beta = input_741_beta_0_to_fp16, epsilon = var_58621_to_fp16, gamma = input_741_gamma_0_to_fp16, x = inputs_389_cast_fp16)[name = tensor("input_741_cast_fp16")]; + tensor var_58637 = const()[name = tensor("op_58637"), val = tensor([1, 1])]; + tensor var_58639 = const()[name = tensor("op_58639"), val = tensor([1, 1])]; + tensor var_58641_pad_type_0 = const()[name = tensor("op_58641_pad_type_0"), val = tensor("custom")]; + tensor var_58641_pad_0 = const()[name = tensor("op_58641_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4944589952)))]; + tensor up_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4951143616)))]; + tensor var_58641_cast_fp16 = conv(bias = up_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_58639, groups = var_58034, pad = var_58641_pad_0, pad_type = var_58641_pad_type_0, strides = var_58637, weight = up_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16, x = input_741_cast_fp16)[name = tensor("op_58641_cast_fp16")]; + tensor var_58642_split_sizes_0 = const()[name = tensor("op_58642_split_sizes_0"), val = tensor([2560, 2560])]; + tensor var_58642_axis_0 = const()[name = tensor("op_58642_axis_0"), val = tensor(1)]; + tensor var_58642_cast_fp16_0, tensor var_58642_cast_fp16_1 = split(axis = var_58642_axis_0, split_sizes = var_58642_split_sizes_0, x = var_58641_cast_fp16)[name = tensor("op_58642_cast_fp16")]; + tensor var_58644_mode_0 = const()[name = tensor("op_58644_mode_0"), val = tensor("EXACT")]; + tensor var_58644_cast_fp16 = gelu(mode = var_58644_mode_0, x = var_58642_cast_fp16_1)[name = tensor("op_58644_cast_fp16")]; + tensor input_743_cast_fp16 = mul(x = var_58642_cast_fp16_0, y = var_58644_cast_fp16)[name = tensor("input_743_cast_fp16")]; + tensor var_58648 = const()[name = tensor("op_58648"), val = tensor([1, 1])]; + tensor var_58650 = const()[name = tensor("op_58650"), val = tensor([1, 1])]; + tensor var_58652_pad_type_0 = const()[name = tensor("op_58652_pad_type_0"), val = tensor("custom")]; + tensor var_58652_pad_0 = const()[name = tensor("op_58652_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4951153920)))]; + tensor up_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4954430784)))]; + tensor var_58652_cast_fp16 = conv(bias = up_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_58650, groups = var_58034, pad = var_58652_pad_0, pad_type = var_58652_pad_type_0, strides = var_58648, weight = up_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16, x = input_743_cast_fp16)[name = tensor("op_58652_cast_fp16")]; + tensor inputs_391_cast_fp16 = add(x = var_58652_cast_fp16, y = inputs_389_cast_fp16)[name = tensor("inputs_391_cast_fp16")]; + tensor hidden_states_515_axes_0 = const()[name = tensor("hidden_states_515_axes_0"), val = tensor([1])]; + tensor hidden_states_515_gamma_0_to_fp16 = const()[name = tensor("hidden_states_515_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4954432128)))]; + tensor hidden_states_515_beta_0_to_fp16 = const()[name = tensor("hidden_states_515_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4954433472)))]; + tensor var_58668_to_fp16 = const()[name = tensor("op_58668_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_515_cast_fp16 = layer_norm(axes = hidden_states_515_axes_0, beta = hidden_states_515_beta_0_to_fp16, epsilon = var_58668_to_fp16, gamma = hidden_states_515_gamma_0_to_fp16, x = inputs_391_cast_fp16)[name = tensor("hidden_states_515_cast_fp16")]; + tensor var_58683 = const()[name = tensor("op_58683"), val = tensor([1, 1])]; + tensor var_58685 = const()[name = tensor("op_58685"), val = tensor([1, 1])]; + tensor q_261_pad_type_0 = const()[name = tensor("q_261_pad_type_0"), val = tensor("custom")]; + tensor q_261_pad_0 = const()[name = tensor("q_261_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_0_transformer_blocks_1_attn1_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_transformer_blocks_1_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4954434816)))]; + tensor q_261_cast_fp16 = conv(dilations = var_58685, groups = var_58034, pad = q_261_pad_0, pad_type = q_261_pad_type_0, strides = var_58683, weight = up_blocks_1_attentions_0_transformer_blocks_1_attn1_to_q_weight_to_fp16, x = hidden_states_515_cast_fp16)[name = tensor("q_261_cast_fp16")]; + tensor var_58689 = const()[name = tensor("op_58689"), val = tensor([1, 1])]; + tensor var_58691 = const()[name = tensor("op_58691"), val = tensor([1, 1])]; + tensor k_521_pad_type_0 = const()[name = tensor("k_521_pad_type_0"), val = tensor("custom")]; + tensor k_521_pad_0 = const()[name = tensor("k_521_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_0_transformer_blocks_1_attn1_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_transformer_blocks_1_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4955254080)))]; + tensor k_521_cast_fp16 = conv(dilations = var_58691, groups = var_58034, pad = k_521_pad_0, pad_type = k_521_pad_type_0, strides = var_58689, weight = up_blocks_1_attentions_0_transformer_blocks_1_attn1_to_k_weight_to_fp16, x = hidden_states_515_cast_fp16)[name = tensor("k_521_cast_fp16")]; + tensor var_58695 = const()[name = tensor("op_58695"), val = tensor([1, 1])]; + tensor var_58697 = const()[name = tensor("op_58697"), val = tensor([1, 1])]; + tensor v_261_pad_type_0 = const()[name = tensor("v_261_pad_type_0"), val = tensor("custom")]; + tensor v_261_pad_0 = const()[name = tensor("v_261_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_0_transformer_blocks_1_attn1_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_transformer_blocks_1_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4956073344)))]; + tensor v_261_cast_fp16 = conv(dilations = var_58697, groups = var_58034, pad = v_261_pad_0, pad_type = v_261_pad_type_0, strides = var_58695, weight = up_blocks_1_attentions_0_transformer_blocks_1_attn1_to_v_weight_to_fp16, x = hidden_states_515_cast_fp16)[name = tensor("v_261_cast_fp16")]; + tensor var_58701_begin_0 = const()[name = tensor("op_58701_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_58701_end_0 = const()[name = tensor("op_58701_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_58701_end_mask_0 = const()[name = tensor("op_58701_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58701_cast_fp16 = slice_by_index(begin = var_58701_begin_0, end = var_58701_end_0, end_mask = var_58701_end_mask_0, x = q_261_cast_fp16)[name = tensor("op_58701_cast_fp16")]; + tensor var_58705_begin_0 = const()[name = tensor("op_58705_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_58705_end_0 = const()[name = tensor("op_58705_end_0"), val = tensor([2, 128, 1, 4096])]; + tensor var_58705_end_mask_0 = const()[name = tensor("op_58705_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58705_cast_fp16 = slice_by_index(begin = var_58705_begin_0, end = var_58705_end_0, end_mask = var_58705_end_mask_0, x = q_261_cast_fp16)[name = tensor("op_58705_cast_fp16")]; + tensor var_58709_begin_0 = const()[name = tensor("op_58709_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_58709_end_0 = const()[name = tensor("op_58709_end_0"), val = tensor([2, 192, 1, 4096])]; + tensor var_58709_end_mask_0 = const()[name = tensor("op_58709_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58709_cast_fp16 = slice_by_index(begin = var_58709_begin_0, end = var_58709_end_0, end_mask = var_58709_end_mask_0, x = q_261_cast_fp16)[name = tensor("op_58709_cast_fp16")]; + tensor var_58713_begin_0 = const()[name = tensor("op_58713_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_58713_end_0 = const()[name = tensor("op_58713_end_0"), val = tensor([2, 256, 1, 4096])]; + tensor var_58713_end_mask_0 = const()[name = tensor("op_58713_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58713_cast_fp16 = slice_by_index(begin = var_58713_begin_0, end = var_58713_end_0, end_mask = var_58713_end_mask_0, x = q_261_cast_fp16)[name = tensor("op_58713_cast_fp16")]; + tensor var_58717_begin_0 = const()[name = tensor("op_58717_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_58717_end_0 = const()[name = tensor("op_58717_end_0"), val = tensor([2, 320, 1, 4096])]; + tensor var_58717_end_mask_0 = const()[name = tensor("op_58717_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58717_cast_fp16 = slice_by_index(begin = var_58717_begin_0, end = var_58717_end_0, end_mask = var_58717_end_mask_0, x = q_261_cast_fp16)[name = tensor("op_58717_cast_fp16")]; + tensor var_58721_begin_0 = const()[name = tensor("op_58721_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_58721_end_0 = const()[name = tensor("op_58721_end_0"), val = tensor([2, 384, 1, 4096])]; + tensor var_58721_end_mask_0 = const()[name = tensor("op_58721_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58721_cast_fp16 = slice_by_index(begin = var_58721_begin_0, end = var_58721_end_0, end_mask = var_58721_end_mask_0, x = q_261_cast_fp16)[name = tensor("op_58721_cast_fp16")]; + tensor var_58725_begin_0 = const()[name = tensor("op_58725_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_58725_end_0 = const()[name = tensor("op_58725_end_0"), val = tensor([2, 448, 1, 4096])]; + tensor var_58725_end_mask_0 = const()[name = tensor("op_58725_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58725_cast_fp16 = slice_by_index(begin = var_58725_begin_0, end = var_58725_end_0, end_mask = var_58725_end_mask_0, x = q_261_cast_fp16)[name = tensor("op_58725_cast_fp16")]; + tensor var_58729_begin_0 = const()[name = tensor("op_58729_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_58729_end_0 = const()[name = tensor("op_58729_end_0"), val = tensor([2, 512, 1, 4096])]; + tensor var_58729_end_mask_0 = const()[name = tensor("op_58729_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58729_cast_fp16 = slice_by_index(begin = var_58729_begin_0, end = var_58729_end_0, end_mask = var_58729_end_mask_0, x = q_261_cast_fp16)[name = tensor("op_58729_cast_fp16")]; + tensor var_58733_begin_0 = const()[name = tensor("op_58733_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_58733_end_0 = const()[name = tensor("op_58733_end_0"), val = tensor([2, 576, 1, 4096])]; + tensor var_58733_end_mask_0 = const()[name = tensor("op_58733_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58733_cast_fp16 = slice_by_index(begin = var_58733_begin_0, end = var_58733_end_0, end_mask = var_58733_end_mask_0, x = q_261_cast_fp16)[name = tensor("op_58733_cast_fp16")]; + tensor var_58737_begin_0 = const()[name = tensor("op_58737_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_58737_end_0 = const()[name = tensor("op_58737_end_0"), val = tensor([2, 640, 1, 4096])]; + tensor var_58737_end_mask_0 = const()[name = tensor("op_58737_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58737_cast_fp16 = slice_by_index(begin = var_58737_begin_0, end = var_58737_end_0, end_mask = var_58737_end_mask_0, x = q_261_cast_fp16)[name = tensor("op_58737_cast_fp16")]; + tensor k_523_perm_0 = const()[name = tensor("k_523_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_58744_begin_0 = const()[name = tensor("op_58744_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_58744_end_0 = const()[name = tensor("op_58744_end_0"), val = tensor([2, 4096, 1, 64])]; + tensor var_58744_end_mask_0 = const()[name = tensor("op_58744_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_523_cast_fp16 = transpose(perm = k_523_perm_0, x = k_521_cast_fp16)[name = tensor("transpose_9")]; + tensor var_58744_cast_fp16 = slice_by_index(begin = var_58744_begin_0, end = var_58744_end_0, end_mask = var_58744_end_mask_0, x = k_523_cast_fp16)[name = tensor("op_58744_cast_fp16")]; + tensor var_58748_begin_0 = const()[name = tensor("op_58748_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_58748_end_0 = const()[name = tensor("op_58748_end_0"), val = tensor([2, 4096, 1, 128])]; + tensor var_58748_end_mask_0 = const()[name = tensor("op_58748_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_58748_cast_fp16 = slice_by_index(begin = var_58748_begin_0, end = var_58748_end_0, end_mask = var_58748_end_mask_0, x = k_523_cast_fp16)[name = tensor("op_58748_cast_fp16")]; + tensor var_58752_begin_0 = const()[name = tensor("op_58752_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_58752_end_0 = const()[name = tensor("op_58752_end_0"), val = tensor([2, 4096, 1, 192])]; + tensor var_58752_end_mask_0 = const()[name = tensor("op_58752_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_58752_cast_fp16 = slice_by_index(begin = var_58752_begin_0, end = var_58752_end_0, end_mask = var_58752_end_mask_0, x = k_523_cast_fp16)[name = tensor("op_58752_cast_fp16")]; + tensor var_58756_begin_0 = const()[name = tensor("op_58756_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_58756_end_0 = const()[name = tensor("op_58756_end_0"), val = tensor([2, 4096, 1, 256])]; + tensor var_58756_end_mask_0 = const()[name = tensor("op_58756_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_58756_cast_fp16 = slice_by_index(begin = var_58756_begin_0, end = var_58756_end_0, end_mask = var_58756_end_mask_0, x = k_523_cast_fp16)[name = tensor("op_58756_cast_fp16")]; + tensor var_58760_begin_0 = const()[name = tensor("op_58760_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_58760_end_0 = const()[name = tensor("op_58760_end_0"), val = tensor([2, 4096, 1, 320])]; + tensor var_58760_end_mask_0 = const()[name = tensor("op_58760_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_58760_cast_fp16 = slice_by_index(begin = var_58760_begin_0, end = var_58760_end_0, end_mask = var_58760_end_mask_0, x = k_523_cast_fp16)[name = tensor("op_58760_cast_fp16")]; + tensor var_58764_begin_0 = const()[name = tensor("op_58764_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_58764_end_0 = const()[name = tensor("op_58764_end_0"), val = tensor([2, 4096, 1, 384])]; + tensor var_58764_end_mask_0 = const()[name = tensor("op_58764_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_58764_cast_fp16 = slice_by_index(begin = var_58764_begin_0, end = var_58764_end_0, end_mask = var_58764_end_mask_0, x = k_523_cast_fp16)[name = tensor("op_58764_cast_fp16")]; + tensor var_58768_begin_0 = const()[name = tensor("op_58768_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_58768_end_0 = const()[name = tensor("op_58768_end_0"), val = tensor([2, 4096, 1, 448])]; + tensor var_58768_end_mask_0 = const()[name = tensor("op_58768_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_58768_cast_fp16 = slice_by_index(begin = var_58768_begin_0, end = var_58768_end_0, end_mask = var_58768_end_mask_0, x = k_523_cast_fp16)[name = tensor("op_58768_cast_fp16")]; + tensor var_58772_begin_0 = const()[name = tensor("op_58772_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_58772_end_0 = const()[name = tensor("op_58772_end_0"), val = tensor([2, 4096, 1, 512])]; + tensor var_58772_end_mask_0 = const()[name = tensor("op_58772_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_58772_cast_fp16 = slice_by_index(begin = var_58772_begin_0, end = var_58772_end_0, end_mask = var_58772_end_mask_0, x = k_523_cast_fp16)[name = tensor("op_58772_cast_fp16")]; + tensor var_58776_begin_0 = const()[name = tensor("op_58776_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_58776_end_0 = const()[name = tensor("op_58776_end_0"), val = tensor([2, 4096, 1, 576])]; + tensor var_58776_end_mask_0 = const()[name = tensor("op_58776_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_58776_cast_fp16 = slice_by_index(begin = var_58776_begin_0, end = var_58776_end_0, end_mask = var_58776_end_mask_0, x = k_523_cast_fp16)[name = tensor("op_58776_cast_fp16")]; + tensor var_58780_begin_0 = const()[name = tensor("op_58780_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_58780_end_0 = const()[name = tensor("op_58780_end_0"), val = tensor([2, 4096, 1, 640])]; + tensor var_58780_end_mask_0 = const()[name = tensor("op_58780_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_58780_cast_fp16 = slice_by_index(begin = var_58780_begin_0, end = var_58780_end_0, end_mask = var_58780_end_mask_0, x = k_523_cast_fp16)[name = tensor("op_58780_cast_fp16")]; + tensor var_58782_begin_0 = const()[name = tensor("op_58782_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_58782_end_0 = const()[name = tensor("op_58782_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_58782_end_mask_0 = const()[name = tensor("op_58782_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58782_cast_fp16 = slice_by_index(begin = var_58782_begin_0, end = var_58782_end_0, end_mask = var_58782_end_mask_0, x = v_261_cast_fp16)[name = tensor("op_58782_cast_fp16")]; + tensor var_58786_begin_0 = const()[name = tensor("op_58786_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_58786_end_0 = const()[name = tensor("op_58786_end_0"), val = tensor([2, 128, 1, 4096])]; + tensor var_58786_end_mask_0 = const()[name = tensor("op_58786_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58786_cast_fp16 = slice_by_index(begin = var_58786_begin_0, end = var_58786_end_0, end_mask = var_58786_end_mask_0, x = v_261_cast_fp16)[name = tensor("op_58786_cast_fp16")]; + tensor var_58790_begin_0 = const()[name = tensor("op_58790_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_58790_end_0 = const()[name = tensor("op_58790_end_0"), val = tensor([2, 192, 1, 4096])]; + tensor var_58790_end_mask_0 = const()[name = tensor("op_58790_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58790_cast_fp16 = slice_by_index(begin = var_58790_begin_0, end = var_58790_end_0, end_mask = var_58790_end_mask_0, x = v_261_cast_fp16)[name = tensor("op_58790_cast_fp16")]; + tensor var_58794_begin_0 = const()[name = tensor("op_58794_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_58794_end_0 = const()[name = tensor("op_58794_end_0"), val = tensor([2, 256, 1, 4096])]; + tensor var_58794_end_mask_0 = const()[name = tensor("op_58794_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58794_cast_fp16 = slice_by_index(begin = var_58794_begin_0, end = var_58794_end_0, end_mask = var_58794_end_mask_0, x = v_261_cast_fp16)[name = tensor("op_58794_cast_fp16")]; + tensor var_58798_begin_0 = const()[name = tensor("op_58798_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_58798_end_0 = const()[name = tensor("op_58798_end_0"), val = tensor([2, 320, 1, 4096])]; + tensor var_58798_end_mask_0 = const()[name = tensor("op_58798_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58798_cast_fp16 = slice_by_index(begin = var_58798_begin_0, end = var_58798_end_0, end_mask = var_58798_end_mask_0, x = v_261_cast_fp16)[name = tensor("op_58798_cast_fp16")]; + tensor var_58802_begin_0 = const()[name = tensor("op_58802_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_58802_end_0 = const()[name = tensor("op_58802_end_0"), val = tensor([2, 384, 1, 4096])]; + tensor var_58802_end_mask_0 = const()[name = tensor("op_58802_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58802_cast_fp16 = slice_by_index(begin = var_58802_begin_0, end = var_58802_end_0, end_mask = var_58802_end_mask_0, x = v_261_cast_fp16)[name = tensor("op_58802_cast_fp16")]; + tensor var_58806_begin_0 = const()[name = tensor("op_58806_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_58806_end_0 = const()[name = tensor("op_58806_end_0"), val = tensor([2, 448, 1, 4096])]; + tensor var_58806_end_mask_0 = const()[name = tensor("op_58806_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58806_cast_fp16 = slice_by_index(begin = var_58806_begin_0, end = var_58806_end_0, end_mask = var_58806_end_mask_0, x = v_261_cast_fp16)[name = tensor("op_58806_cast_fp16")]; + tensor var_58810_begin_0 = const()[name = tensor("op_58810_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_58810_end_0 = const()[name = tensor("op_58810_end_0"), val = tensor([2, 512, 1, 4096])]; + tensor var_58810_end_mask_0 = const()[name = tensor("op_58810_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58810_cast_fp16 = slice_by_index(begin = var_58810_begin_0, end = var_58810_end_0, end_mask = var_58810_end_mask_0, x = v_261_cast_fp16)[name = tensor("op_58810_cast_fp16")]; + tensor var_58814_begin_0 = const()[name = tensor("op_58814_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_58814_end_0 = const()[name = tensor("op_58814_end_0"), val = tensor([2, 576, 1, 4096])]; + tensor var_58814_end_mask_0 = const()[name = tensor("op_58814_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58814_cast_fp16 = slice_by_index(begin = var_58814_begin_0, end = var_58814_end_0, end_mask = var_58814_end_mask_0, x = v_261_cast_fp16)[name = tensor("op_58814_cast_fp16")]; + tensor var_58818_begin_0 = const()[name = tensor("op_58818_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_58818_end_0 = const()[name = tensor("op_58818_end_0"), val = tensor([2, 640, 1, 4096])]; + tensor var_58818_end_mask_0 = const()[name = tensor("op_58818_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58818_cast_fp16 = slice_by_index(begin = var_58818_begin_0, end = var_58818_end_0, end_mask = var_58818_end_mask_0, x = v_261_cast_fp16)[name = tensor("op_58818_cast_fp16")]; + tensor var_58822_equation_0 = const()[name = tensor("op_58822_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_58822_cast_fp16 = einsum(equation = var_58822_equation_0, values = (var_58744_cast_fp16, var_58701_cast_fp16))[name = tensor("op_58822_cast_fp16")]; + tensor var_58823_to_fp16 = const()[name = tensor("op_58823_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5001_cast_fp16 = mul(x = var_58822_cast_fp16, y = var_58823_to_fp16)[name = tensor("aw_5001_cast_fp16")]; + tensor var_58826_equation_0 = const()[name = tensor("op_58826_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_58826_cast_fp16 = einsum(equation = var_58826_equation_0, values = (var_58748_cast_fp16, var_58705_cast_fp16))[name = tensor("op_58826_cast_fp16")]; + tensor var_58827_to_fp16 = const()[name = tensor("op_58827_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5003_cast_fp16 = mul(x = var_58826_cast_fp16, y = var_58827_to_fp16)[name = tensor("aw_5003_cast_fp16")]; + tensor var_58830_equation_0 = const()[name = tensor("op_58830_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_58830_cast_fp16 = einsum(equation = var_58830_equation_0, values = (var_58752_cast_fp16, var_58709_cast_fp16))[name = tensor("op_58830_cast_fp16")]; + tensor var_58831_to_fp16 = const()[name = tensor("op_58831_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5005_cast_fp16 = mul(x = var_58830_cast_fp16, y = var_58831_to_fp16)[name = tensor("aw_5005_cast_fp16")]; + tensor var_58834_equation_0 = const()[name = tensor("op_58834_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_58834_cast_fp16 = einsum(equation = var_58834_equation_0, values = (var_58756_cast_fp16, var_58713_cast_fp16))[name = tensor("op_58834_cast_fp16")]; + tensor var_58835_to_fp16 = const()[name = tensor("op_58835_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5007_cast_fp16 = mul(x = var_58834_cast_fp16, y = var_58835_to_fp16)[name = tensor("aw_5007_cast_fp16")]; + tensor var_58838_equation_0 = const()[name = tensor("op_58838_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_58838_cast_fp16 = einsum(equation = var_58838_equation_0, values = (var_58760_cast_fp16, var_58717_cast_fp16))[name = tensor("op_58838_cast_fp16")]; + tensor var_58839_to_fp16 = const()[name = tensor("op_58839_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5009_cast_fp16 = mul(x = var_58838_cast_fp16, y = var_58839_to_fp16)[name = tensor("aw_5009_cast_fp16")]; + tensor var_58842_equation_0 = const()[name = tensor("op_58842_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_58842_cast_fp16 = einsum(equation = var_58842_equation_0, values = (var_58764_cast_fp16, var_58721_cast_fp16))[name = tensor("op_58842_cast_fp16")]; + tensor var_58843_to_fp16 = const()[name = tensor("op_58843_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5011_cast_fp16 = mul(x = var_58842_cast_fp16, y = var_58843_to_fp16)[name = tensor("aw_5011_cast_fp16")]; + tensor var_58846_equation_0 = const()[name = tensor("op_58846_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_58846_cast_fp16 = einsum(equation = var_58846_equation_0, values = (var_58768_cast_fp16, var_58725_cast_fp16))[name = tensor("op_58846_cast_fp16")]; + tensor var_58847_to_fp16 = const()[name = tensor("op_58847_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5013_cast_fp16 = mul(x = var_58846_cast_fp16, y = var_58847_to_fp16)[name = tensor("aw_5013_cast_fp16")]; + tensor var_58850_equation_0 = const()[name = tensor("op_58850_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_58850_cast_fp16 = einsum(equation = var_58850_equation_0, values = (var_58772_cast_fp16, var_58729_cast_fp16))[name = tensor("op_58850_cast_fp16")]; + tensor var_58851_to_fp16 = const()[name = tensor("op_58851_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5015_cast_fp16 = mul(x = var_58850_cast_fp16, y = var_58851_to_fp16)[name = tensor("aw_5015_cast_fp16")]; + tensor var_58854_equation_0 = const()[name = tensor("op_58854_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_58854_cast_fp16 = einsum(equation = var_58854_equation_0, values = (var_58776_cast_fp16, var_58733_cast_fp16))[name = tensor("op_58854_cast_fp16")]; + tensor var_58855_to_fp16 = const()[name = tensor("op_58855_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5017_cast_fp16 = mul(x = var_58854_cast_fp16, y = var_58855_to_fp16)[name = tensor("aw_5017_cast_fp16")]; + tensor var_58858_equation_0 = const()[name = tensor("op_58858_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_58858_cast_fp16 = einsum(equation = var_58858_equation_0, values = (var_58780_cast_fp16, var_58737_cast_fp16))[name = tensor("op_58858_cast_fp16")]; + tensor var_58859_to_fp16 = const()[name = tensor("op_58859_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5019_cast_fp16 = mul(x = var_58858_cast_fp16, y = var_58859_to_fp16)[name = tensor("aw_5019_cast_fp16")]; + tensor var_58861_cast_fp16 = softmax(axis = var_58034, x = aw_5001_cast_fp16)[name = tensor("op_58861_cast_fp16")]; + tensor var_58862_cast_fp16 = softmax(axis = var_58034, x = aw_5003_cast_fp16)[name = tensor("op_58862_cast_fp16")]; + tensor var_58863_cast_fp16 = softmax(axis = var_58034, x = aw_5005_cast_fp16)[name = tensor("op_58863_cast_fp16")]; + tensor var_58864_cast_fp16 = softmax(axis = var_58034, x = aw_5007_cast_fp16)[name = tensor("op_58864_cast_fp16")]; + tensor var_58865_cast_fp16 = softmax(axis = var_58034, x = aw_5009_cast_fp16)[name = tensor("op_58865_cast_fp16")]; + tensor var_58866_cast_fp16 = softmax(axis = var_58034, x = aw_5011_cast_fp16)[name = tensor("op_58866_cast_fp16")]; + tensor var_58867_cast_fp16 = softmax(axis = var_58034, x = aw_5013_cast_fp16)[name = tensor("op_58867_cast_fp16")]; + tensor var_58868_cast_fp16 = softmax(axis = var_58034, x = aw_5015_cast_fp16)[name = tensor("op_58868_cast_fp16")]; + tensor var_58869_cast_fp16 = softmax(axis = var_58034, x = aw_5017_cast_fp16)[name = tensor("op_58869_cast_fp16")]; + tensor var_58870_cast_fp16 = softmax(axis = var_58034, x = aw_5019_cast_fp16)[name = tensor("op_58870_cast_fp16")]; + tensor var_58872_equation_0 = const()[name = tensor("op_58872_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_58872_cast_fp16 = einsum(equation = var_58872_equation_0, values = (var_58782_cast_fp16, var_58861_cast_fp16))[name = tensor("op_58872_cast_fp16")]; + tensor var_58874_equation_0 = const()[name = tensor("op_58874_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_58874_cast_fp16 = einsum(equation = var_58874_equation_0, values = (var_58786_cast_fp16, var_58862_cast_fp16))[name = tensor("op_58874_cast_fp16")]; + tensor var_58876_equation_0 = const()[name = tensor("op_58876_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_58876_cast_fp16 = einsum(equation = var_58876_equation_0, values = (var_58790_cast_fp16, var_58863_cast_fp16))[name = tensor("op_58876_cast_fp16")]; + tensor var_58878_equation_0 = const()[name = tensor("op_58878_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_58878_cast_fp16 = einsum(equation = var_58878_equation_0, values = (var_58794_cast_fp16, var_58864_cast_fp16))[name = tensor("op_58878_cast_fp16")]; + tensor var_58880_equation_0 = const()[name = tensor("op_58880_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_58880_cast_fp16 = einsum(equation = var_58880_equation_0, values = (var_58798_cast_fp16, var_58865_cast_fp16))[name = tensor("op_58880_cast_fp16")]; + tensor var_58882_equation_0 = const()[name = tensor("op_58882_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_58882_cast_fp16 = einsum(equation = var_58882_equation_0, values = (var_58802_cast_fp16, var_58866_cast_fp16))[name = tensor("op_58882_cast_fp16")]; + tensor var_58884_equation_0 = const()[name = tensor("op_58884_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_58884_cast_fp16 = einsum(equation = var_58884_equation_0, values = (var_58806_cast_fp16, var_58867_cast_fp16))[name = tensor("op_58884_cast_fp16")]; + tensor var_58886_equation_0 = const()[name = tensor("op_58886_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_58886_cast_fp16 = einsum(equation = var_58886_equation_0, values = (var_58810_cast_fp16, var_58868_cast_fp16))[name = tensor("op_58886_cast_fp16")]; + tensor var_58888_equation_0 = const()[name = tensor("op_58888_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_58888_cast_fp16 = einsum(equation = var_58888_equation_0, values = (var_58814_cast_fp16, var_58869_cast_fp16))[name = tensor("op_58888_cast_fp16")]; + tensor var_58890_equation_0 = const()[name = tensor("op_58890_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_58890_cast_fp16 = einsum(equation = var_58890_equation_0, values = (var_58818_cast_fp16, var_58870_cast_fp16))[name = tensor("op_58890_cast_fp16")]; + tensor input_745_interleave_0 = const()[name = tensor("input_745_interleave_0"), val = tensor(false)]; + tensor input_745_cast_fp16 = concat(axis = var_58034, interleave = input_745_interleave_0, values = (var_58872_cast_fp16, var_58874_cast_fp16, var_58876_cast_fp16, var_58878_cast_fp16, var_58880_cast_fp16, var_58882_cast_fp16, var_58884_cast_fp16, var_58886_cast_fp16, var_58888_cast_fp16, var_58890_cast_fp16))[name = tensor("input_745_cast_fp16")]; + tensor var_58896 = const()[name = tensor("op_58896"), val = tensor([1, 1])]; + tensor var_58898 = const()[name = tensor("op_58898"), val = tensor([1, 1])]; + tensor var_58900_pad_type_0 = const()[name = tensor("op_58900_pad_type_0"), val = tensor("custom")]; + tensor var_58900_pad_0 = const()[name = tensor("op_58900_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_0_transformer_blocks_1_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_transformer_blocks_1_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4956892608)))]; + tensor up_blocks_1_attentions_0_transformer_blocks_1_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_transformer_blocks_1_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4957711872)))]; + tensor var_58900_cast_fp16 = conv(bias = up_blocks_1_attentions_0_transformer_blocks_1_attn1_to_out_0_bias_to_fp16, dilations = var_58898, groups = var_58034, pad = var_58900_pad_0, pad_type = var_58900_pad_type_0, strides = var_58896, weight = up_blocks_1_attentions_0_transformer_blocks_1_attn1_to_out_0_weight_to_fp16, x = input_745_cast_fp16)[name = tensor("op_58900_cast_fp16")]; + tensor inputs_393_cast_fp16 = add(x = var_58900_cast_fp16, y = inputs_391_cast_fp16)[name = tensor("inputs_393_cast_fp16")]; + tensor hidden_states_517_axes_0 = const()[name = tensor("hidden_states_517_axes_0"), val = tensor([1])]; + tensor hidden_states_517_gamma_0_to_fp16 = const()[name = tensor("hidden_states_517_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4957713216)))]; + tensor hidden_states_517_beta_0_to_fp16 = const()[name = tensor("hidden_states_517_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4957714560)))]; + tensor var_58910_to_fp16 = const()[name = tensor("op_58910_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_517_cast_fp16 = layer_norm(axes = hidden_states_517_axes_0, beta = hidden_states_517_beta_0_to_fp16, epsilon = var_58910_to_fp16, gamma = hidden_states_517_gamma_0_to_fp16, x = inputs_393_cast_fp16)[name = tensor("hidden_states_517_cast_fp16")]; + tensor var_58925 = const()[name = tensor("op_58925"), val = tensor([1, 1])]; + tensor var_58927 = const()[name = tensor("op_58927"), val = tensor([1, 1])]; + tensor q_263_pad_type_0 = const()[name = tensor("q_263_pad_type_0"), val = tensor("custom")]; + tensor q_263_pad_0 = const()[name = tensor("q_263_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_0_transformer_blocks_1_attn2_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_transformer_blocks_1_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4957715904)))]; + tensor q_263_cast_fp16 = conv(dilations = var_58927, groups = var_58034, pad = q_263_pad_0, pad_type = q_263_pad_type_0, strides = var_58925, weight = up_blocks_1_attentions_0_transformer_blocks_1_attn2_to_q_weight_to_fp16, x = hidden_states_517_cast_fp16)[name = tensor("q_263_cast_fp16")]; + tensor var_58931 = const()[name = tensor("op_58931"), val = tensor([1, 1])]; + tensor var_58933 = const()[name = tensor("op_58933"), val = tensor([1, 1])]; + tensor k_525_pad_type_0 = const()[name = tensor("k_525_pad_type_0"), val = tensor("custom")]; + tensor k_525_pad_0 = const()[name = tensor("k_525_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_0_transformer_blocks_1_attn2_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_transformer_blocks_1_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4958535168)))]; + tensor k_525_cast_fp16 = conv(dilations = var_58933, groups = var_58034, pad = k_525_pad_0, pad_type = k_525_pad_type_0, strides = var_58931, weight = up_blocks_1_attentions_0_transformer_blocks_1_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_525_cast_fp16")]; + tensor var_58937 = const()[name = tensor("op_58937"), val = tensor([1, 1])]; + tensor var_58939 = const()[name = tensor("op_58939"), val = tensor([1, 1])]; + tensor v_263_pad_type_0 = const()[name = tensor("v_263_pad_type_0"), val = tensor("custom")]; + tensor v_263_pad_0 = const()[name = tensor("v_263_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_0_transformer_blocks_1_attn2_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_transformer_blocks_1_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4961156672)))]; + tensor v_263_cast_fp16 = conv(dilations = var_58939, groups = var_58034, pad = v_263_pad_0, pad_type = v_263_pad_type_0, strides = var_58937, weight = up_blocks_1_attentions_0_transformer_blocks_1_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_263_cast_fp16")]; + tensor var_58943_begin_0 = const()[name = tensor("op_58943_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_58943_end_0 = const()[name = tensor("op_58943_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_58943_end_mask_0 = const()[name = tensor("op_58943_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58943_cast_fp16 = slice_by_index(begin = var_58943_begin_0, end = var_58943_end_0, end_mask = var_58943_end_mask_0, x = q_263_cast_fp16)[name = tensor("op_58943_cast_fp16")]; + tensor var_58947_begin_0 = const()[name = tensor("op_58947_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_58947_end_0 = const()[name = tensor("op_58947_end_0"), val = tensor([2, 128, 1, 4096])]; + tensor var_58947_end_mask_0 = const()[name = tensor("op_58947_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58947_cast_fp16 = slice_by_index(begin = var_58947_begin_0, end = var_58947_end_0, end_mask = var_58947_end_mask_0, x = q_263_cast_fp16)[name = tensor("op_58947_cast_fp16")]; + tensor var_58951_begin_0 = const()[name = tensor("op_58951_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_58951_end_0 = const()[name = tensor("op_58951_end_0"), val = tensor([2, 192, 1, 4096])]; + tensor var_58951_end_mask_0 = const()[name = tensor("op_58951_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58951_cast_fp16 = slice_by_index(begin = var_58951_begin_0, end = var_58951_end_0, end_mask = var_58951_end_mask_0, x = q_263_cast_fp16)[name = tensor("op_58951_cast_fp16")]; + tensor var_58955_begin_0 = const()[name = tensor("op_58955_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_58955_end_0 = const()[name = tensor("op_58955_end_0"), val = tensor([2, 256, 1, 4096])]; + tensor var_58955_end_mask_0 = const()[name = tensor("op_58955_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58955_cast_fp16 = slice_by_index(begin = var_58955_begin_0, end = var_58955_end_0, end_mask = var_58955_end_mask_0, x = q_263_cast_fp16)[name = tensor("op_58955_cast_fp16")]; + tensor var_58959_begin_0 = const()[name = tensor("op_58959_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_58959_end_0 = const()[name = tensor("op_58959_end_0"), val = tensor([2, 320, 1, 4096])]; + tensor var_58959_end_mask_0 = const()[name = tensor("op_58959_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58959_cast_fp16 = slice_by_index(begin = var_58959_begin_0, end = var_58959_end_0, end_mask = var_58959_end_mask_0, x = q_263_cast_fp16)[name = tensor("op_58959_cast_fp16")]; + tensor var_58963_begin_0 = const()[name = tensor("op_58963_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_58963_end_0 = const()[name = tensor("op_58963_end_0"), val = tensor([2, 384, 1, 4096])]; + tensor var_58963_end_mask_0 = const()[name = tensor("op_58963_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58963_cast_fp16 = slice_by_index(begin = var_58963_begin_0, end = var_58963_end_0, end_mask = var_58963_end_mask_0, x = q_263_cast_fp16)[name = tensor("op_58963_cast_fp16")]; + tensor var_58967_begin_0 = const()[name = tensor("op_58967_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_58967_end_0 = const()[name = tensor("op_58967_end_0"), val = tensor([2, 448, 1, 4096])]; + tensor var_58967_end_mask_0 = const()[name = tensor("op_58967_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58967_cast_fp16 = slice_by_index(begin = var_58967_begin_0, end = var_58967_end_0, end_mask = var_58967_end_mask_0, x = q_263_cast_fp16)[name = tensor("op_58967_cast_fp16")]; + tensor var_58971_begin_0 = const()[name = tensor("op_58971_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_58971_end_0 = const()[name = tensor("op_58971_end_0"), val = tensor([2, 512, 1, 4096])]; + tensor var_58971_end_mask_0 = const()[name = tensor("op_58971_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58971_cast_fp16 = slice_by_index(begin = var_58971_begin_0, end = var_58971_end_0, end_mask = var_58971_end_mask_0, x = q_263_cast_fp16)[name = tensor("op_58971_cast_fp16")]; + tensor var_58975_begin_0 = const()[name = tensor("op_58975_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_58975_end_0 = const()[name = tensor("op_58975_end_0"), val = tensor([2, 576, 1, 4096])]; + tensor var_58975_end_mask_0 = const()[name = tensor("op_58975_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58975_cast_fp16 = slice_by_index(begin = var_58975_begin_0, end = var_58975_end_0, end_mask = var_58975_end_mask_0, x = q_263_cast_fp16)[name = tensor("op_58975_cast_fp16")]; + tensor var_58979_begin_0 = const()[name = tensor("op_58979_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_58979_end_0 = const()[name = tensor("op_58979_end_0"), val = tensor([2, 640, 1, 4096])]; + tensor var_58979_end_mask_0 = const()[name = tensor("op_58979_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_58979_cast_fp16 = slice_by_index(begin = var_58979_begin_0, end = var_58979_end_0, end_mask = var_58979_end_mask_0, x = q_263_cast_fp16)[name = tensor("op_58979_cast_fp16")]; + tensor k_527_perm_0 = const()[name = tensor("k_527_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_58986_begin_0 = const()[name = tensor("op_58986_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_58986_end_0 = const()[name = tensor("op_58986_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_58986_end_mask_0 = const()[name = tensor("op_58986_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_527_cast_fp16 = transpose(perm = k_527_perm_0, x = k_525_cast_fp16)[name = tensor("transpose_8")]; + tensor var_58986_cast_fp16 = slice_by_index(begin = var_58986_begin_0, end = var_58986_end_0, end_mask = var_58986_end_mask_0, x = k_527_cast_fp16)[name = tensor("op_58986_cast_fp16")]; + tensor var_58990_begin_0 = const()[name = tensor("op_58990_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_58990_end_0 = const()[name = tensor("op_58990_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_58990_end_mask_0 = const()[name = tensor("op_58990_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_58990_cast_fp16 = slice_by_index(begin = var_58990_begin_0, end = var_58990_end_0, end_mask = var_58990_end_mask_0, x = k_527_cast_fp16)[name = tensor("op_58990_cast_fp16")]; + tensor var_58994_begin_0 = const()[name = tensor("op_58994_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_58994_end_0 = const()[name = tensor("op_58994_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_58994_end_mask_0 = const()[name = tensor("op_58994_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_58994_cast_fp16 = slice_by_index(begin = var_58994_begin_0, end = var_58994_end_0, end_mask = var_58994_end_mask_0, x = k_527_cast_fp16)[name = tensor("op_58994_cast_fp16")]; + tensor var_58998_begin_0 = const()[name = tensor("op_58998_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_58998_end_0 = const()[name = tensor("op_58998_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_58998_end_mask_0 = const()[name = tensor("op_58998_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_58998_cast_fp16 = slice_by_index(begin = var_58998_begin_0, end = var_58998_end_0, end_mask = var_58998_end_mask_0, x = k_527_cast_fp16)[name = tensor("op_58998_cast_fp16")]; + tensor var_59002_begin_0 = const()[name = tensor("op_59002_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_59002_end_0 = const()[name = tensor("op_59002_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_59002_end_mask_0 = const()[name = tensor("op_59002_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_59002_cast_fp16 = slice_by_index(begin = var_59002_begin_0, end = var_59002_end_0, end_mask = var_59002_end_mask_0, x = k_527_cast_fp16)[name = tensor("op_59002_cast_fp16")]; + tensor var_59006_begin_0 = const()[name = tensor("op_59006_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_59006_end_0 = const()[name = tensor("op_59006_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_59006_end_mask_0 = const()[name = tensor("op_59006_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_59006_cast_fp16 = slice_by_index(begin = var_59006_begin_0, end = var_59006_end_0, end_mask = var_59006_end_mask_0, x = k_527_cast_fp16)[name = tensor("op_59006_cast_fp16")]; + tensor var_59010_begin_0 = const()[name = tensor("op_59010_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_59010_end_0 = const()[name = tensor("op_59010_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_59010_end_mask_0 = const()[name = tensor("op_59010_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_59010_cast_fp16 = slice_by_index(begin = var_59010_begin_0, end = var_59010_end_0, end_mask = var_59010_end_mask_0, x = k_527_cast_fp16)[name = tensor("op_59010_cast_fp16")]; + tensor var_59014_begin_0 = const()[name = tensor("op_59014_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_59014_end_0 = const()[name = tensor("op_59014_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_59014_end_mask_0 = const()[name = tensor("op_59014_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_59014_cast_fp16 = slice_by_index(begin = var_59014_begin_0, end = var_59014_end_0, end_mask = var_59014_end_mask_0, x = k_527_cast_fp16)[name = tensor("op_59014_cast_fp16")]; + tensor var_59018_begin_0 = const()[name = tensor("op_59018_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_59018_end_0 = const()[name = tensor("op_59018_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_59018_end_mask_0 = const()[name = tensor("op_59018_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_59018_cast_fp16 = slice_by_index(begin = var_59018_begin_0, end = var_59018_end_0, end_mask = var_59018_end_mask_0, x = k_527_cast_fp16)[name = tensor("op_59018_cast_fp16")]; + tensor var_59022_begin_0 = const()[name = tensor("op_59022_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_59022_end_0 = const()[name = tensor("op_59022_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_59022_end_mask_0 = const()[name = tensor("op_59022_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_59022_cast_fp16 = slice_by_index(begin = var_59022_begin_0, end = var_59022_end_0, end_mask = var_59022_end_mask_0, x = k_527_cast_fp16)[name = tensor("op_59022_cast_fp16")]; + tensor var_59024_begin_0 = const()[name = tensor("op_59024_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_59024_end_0 = const()[name = tensor("op_59024_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_59024_end_mask_0 = const()[name = tensor("op_59024_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59024_cast_fp16 = slice_by_index(begin = var_59024_begin_0, end = var_59024_end_0, end_mask = var_59024_end_mask_0, x = v_263_cast_fp16)[name = tensor("op_59024_cast_fp16")]; + tensor var_59028_begin_0 = const()[name = tensor("op_59028_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_59028_end_0 = const()[name = tensor("op_59028_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_59028_end_mask_0 = const()[name = tensor("op_59028_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59028_cast_fp16 = slice_by_index(begin = var_59028_begin_0, end = var_59028_end_0, end_mask = var_59028_end_mask_0, x = v_263_cast_fp16)[name = tensor("op_59028_cast_fp16")]; + tensor var_59032_begin_0 = const()[name = tensor("op_59032_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_59032_end_0 = const()[name = tensor("op_59032_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_59032_end_mask_0 = const()[name = tensor("op_59032_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59032_cast_fp16 = slice_by_index(begin = var_59032_begin_0, end = var_59032_end_0, end_mask = var_59032_end_mask_0, x = v_263_cast_fp16)[name = tensor("op_59032_cast_fp16")]; + tensor var_59036_begin_0 = const()[name = tensor("op_59036_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_59036_end_0 = const()[name = tensor("op_59036_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_59036_end_mask_0 = const()[name = tensor("op_59036_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59036_cast_fp16 = slice_by_index(begin = var_59036_begin_0, end = var_59036_end_0, end_mask = var_59036_end_mask_0, x = v_263_cast_fp16)[name = tensor("op_59036_cast_fp16")]; + tensor var_59040_begin_0 = const()[name = tensor("op_59040_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_59040_end_0 = const()[name = tensor("op_59040_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_59040_end_mask_0 = const()[name = tensor("op_59040_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59040_cast_fp16 = slice_by_index(begin = var_59040_begin_0, end = var_59040_end_0, end_mask = var_59040_end_mask_0, x = v_263_cast_fp16)[name = tensor("op_59040_cast_fp16")]; + tensor var_59044_begin_0 = const()[name = tensor("op_59044_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_59044_end_0 = const()[name = tensor("op_59044_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_59044_end_mask_0 = const()[name = tensor("op_59044_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59044_cast_fp16 = slice_by_index(begin = var_59044_begin_0, end = var_59044_end_0, end_mask = var_59044_end_mask_0, x = v_263_cast_fp16)[name = tensor("op_59044_cast_fp16")]; + tensor var_59048_begin_0 = const()[name = tensor("op_59048_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_59048_end_0 = const()[name = tensor("op_59048_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_59048_end_mask_0 = const()[name = tensor("op_59048_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59048_cast_fp16 = slice_by_index(begin = var_59048_begin_0, end = var_59048_end_0, end_mask = var_59048_end_mask_0, x = v_263_cast_fp16)[name = tensor("op_59048_cast_fp16")]; + tensor var_59052_begin_0 = const()[name = tensor("op_59052_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_59052_end_0 = const()[name = tensor("op_59052_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_59052_end_mask_0 = const()[name = tensor("op_59052_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59052_cast_fp16 = slice_by_index(begin = var_59052_begin_0, end = var_59052_end_0, end_mask = var_59052_end_mask_0, x = v_263_cast_fp16)[name = tensor("op_59052_cast_fp16")]; + tensor var_59056_begin_0 = const()[name = tensor("op_59056_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_59056_end_0 = const()[name = tensor("op_59056_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_59056_end_mask_0 = const()[name = tensor("op_59056_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59056_cast_fp16 = slice_by_index(begin = var_59056_begin_0, end = var_59056_end_0, end_mask = var_59056_end_mask_0, x = v_263_cast_fp16)[name = tensor("op_59056_cast_fp16")]; + tensor var_59060_begin_0 = const()[name = tensor("op_59060_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_59060_end_0 = const()[name = tensor("op_59060_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_59060_end_mask_0 = const()[name = tensor("op_59060_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59060_cast_fp16 = slice_by_index(begin = var_59060_begin_0, end = var_59060_end_0, end_mask = var_59060_end_mask_0, x = v_263_cast_fp16)[name = tensor("op_59060_cast_fp16")]; + tensor var_59064_equation_0 = const()[name = tensor("op_59064_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_59064_cast_fp16 = einsum(equation = var_59064_equation_0, values = (var_58986_cast_fp16, var_58943_cast_fp16))[name = tensor("op_59064_cast_fp16")]; + tensor var_59065_to_fp16 = const()[name = tensor("op_59065_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5021_cast_fp16 = mul(x = var_59064_cast_fp16, y = var_59065_to_fp16)[name = tensor("aw_5021_cast_fp16")]; + tensor var_59068_equation_0 = const()[name = tensor("op_59068_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_59068_cast_fp16 = einsum(equation = var_59068_equation_0, values = (var_58990_cast_fp16, var_58947_cast_fp16))[name = tensor("op_59068_cast_fp16")]; + tensor var_59069_to_fp16 = const()[name = tensor("op_59069_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5023_cast_fp16 = mul(x = var_59068_cast_fp16, y = var_59069_to_fp16)[name = tensor("aw_5023_cast_fp16")]; + tensor var_59072_equation_0 = const()[name = tensor("op_59072_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_59072_cast_fp16 = einsum(equation = var_59072_equation_0, values = (var_58994_cast_fp16, var_58951_cast_fp16))[name = tensor("op_59072_cast_fp16")]; + tensor var_59073_to_fp16 = const()[name = tensor("op_59073_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5025_cast_fp16 = mul(x = var_59072_cast_fp16, y = var_59073_to_fp16)[name = tensor("aw_5025_cast_fp16")]; + tensor var_59076_equation_0 = const()[name = tensor("op_59076_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_59076_cast_fp16 = einsum(equation = var_59076_equation_0, values = (var_58998_cast_fp16, var_58955_cast_fp16))[name = tensor("op_59076_cast_fp16")]; + tensor var_59077_to_fp16 = const()[name = tensor("op_59077_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5027_cast_fp16 = mul(x = var_59076_cast_fp16, y = var_59077_to_fp16)[name = tensor("aw_5027_cast_fp16")]; + tensor var_59080_equation_0 = const()[name = tensor("op_59080_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_59080_cast_fp16 = einsum(equation = var_59080_equation_0, values = (var_59002_cast_fp16, var_58959_cast_fp16))[name = tensor("op_59080_cast_fp16")]; + tensor var_59081_to_fp16 = const()[name = tensor("op_59081_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5029_cast_fp16 = mul(x = var_59080_cast_fp16, y = var_59081_to_fp16)[name = tensor("aw_5029_cast_fp16")]; + tensor var_59084_equation_0 = const()[name = tensor("op_59084_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_59084_cast_fp16 = einsum(equation = var_59084_equation_0, values = (var_59006_cast_fp16, var_58963_cast_fp16))[name = tensor("op_59084_cast_fp16")]; + tensor var_59085_to_fp16 = const()[name = tensor("op_59085_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5031_cast_fp16 = mul(x = var_59084_cast_fp16, y = var_59085_to_fp16)[name = tensor("aw_5031_cast_fp16")]; + tensor var_59088_equation_0 = const()[name = tensor("op_59088_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_59088_cast_fp16 = einsum(equation = var_59088_equation_0, values = (var_59010_cast_fp16, var_58967_cast_fp16))[name = tensor("op_59088_cast_fp16")]; + tensor var_59089_to_fp16 = const()[name = tensor("op_59089_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5033_cast_fp16 = mul(x = var_59088_cast_fp16, y = var_59089_to_fp16)[name = tensor("aw_5033_cast_fp16")]; + tensor var_59092_equation_0 = const()[name = tensor("op_59092_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_59092_cast_fp16 = einsum(equation = var_59092_equation_0, values = (var_59014_cast_fp16, var_58971_cast_fp16))[name = tensor("op_59092_cast_fp16")]; + tensor var_59093_to_fp16 = const()[name = tensor("op_59093_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5035_cast_fp16 = mul(x = var_59092_cast_fp16, y = var_59093_to_fp16)[name = tensor("aw_5035_cast_fp16")]; + tensor var_59096_equation_0 = const()[name = tensor("op_59096_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_59096_cast_fp16 = einsum(equation = var_59096_equation_0, values = (var_59018_cast_fp16, var_58975_cast_fp16))[name = tensor("op_59096_cast_fp16")]; + tensor var_59097_to_fp16 = const()[name = tensor("op_59097_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5037_cast_fp16 = mul(x = var_59096_cast_fp16, y = var_59097_to_fp16)[name = tensor("aw_5037_cast_fp16")]; + tensor var_59100_equation_0 = const()[name = tensor("op_59100_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_59100_cast_fp16 = einsum(equation = var_59100_equation_0, values = (var_59022_cast_fp16, var_58979_cast_fp16))[name = tensor("op_59100_cast_fp16")]; + tensor var_59101_to_fp16 = const()[name = tensor("op_59101_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5039_cast_fp16 = mul(x = var_59100_cast_fp16, y = var_59101_to_fp16)[name = tensor("aw_5039_cast_fp16")]; + tensor var_59103_cast_fp16 = softmax(axis = var_58034, x = aw_5021_cast_fp16)[name = tensor("op_59103_cast_fp16")]; + tensor var_59104_cast_fp16 = softmax(axis = var_58034, x = aw_5023_cast_fp16)[name = tensor("op_59104_cast_fp16")]; + tensor var_59105_cast_fp16 = softmax(axis = var_58034, x = aw_5025_cast_fp16)[name = tensor("op_59105_cast_fp16")]; + tensor var_59106_cast_fp16 = softmax(axis = var_58034, x = aw_5027_cast_fp16)[name = tensor("op_59106_cast_fp16")]; + tensor var_59107_cast_fp16 = softmax(axis = var_58034, x = aw_5029_cast_fp16)[name = tensor("op_59107_cast_fp16")]; + tensor var_59108_cast_fp16 = softmax(axis = var_58034, x = aw_5031_cast_fp16)[name = tensor("op_59108_cast_fp16")]; + tensor var_59109_cast_fp16 = softmax(axis = var_58034, x = aw_5033_cast_fp16)[name = tensor("op_59109_cast_fp16")]; + tensor var_59110_cast_fp16 = softmax(axis = var_58034, x = aw_5035_cast_fp16)[name = tensor("op_59110_cast_fp16")]; + tensor var_59111_cast_fp16 = softmax(axis = var_58034, x = aw_5037_cast_fp16)[name = tensor("op_59111_cast_fp16")]; + tensor var_59112_cast_fp16 = softmax(axis = var_58034, x = aw_5039_cast_fp16)[name = tensor("op_59112_cast_fp16")]; + tensor var_59114_equation_0 = const()[name = tensor("op_59114_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_59114_cast_fp16 = einsum(equation = var_59114_equation_0, values = (var_59024_cast_fp16, var_59103_cast_fp16))[name = tensor("op_59114_cast_fp16")]; + tensor var_59116_equation_0 = const()[name = tensor("op_59116_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_59116_cast_fp16 = einsum(equation = var_59116_equation_0, values = (var_59028_cast_fp16, var_59104_cast_fp16))[name = tensor("op_59116_cast_fp16")]; + tensor var_59118_equation_0 = const()[name = tensor("op_59118_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_59118_cast_fp16 = einsum(equation = var_59118_equation_0, values = (var_59032_cast_fp16, var_59105_cast_fp16))[name = tensor("op_59118_cast_fp16")]; + tensor var_59120_equation_0 = const()[name = tensor("op_59120_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_59120_cast_fp16 = einsum(equation = var_59120_equation_0, values = (var_59036_cast_fp16, var_59106_cast_fp16))[name = tensor("op_59120_cast_fp16")]; + tensor var_59122_equation_0 = const()[name = tensor("op_59122_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_59122_cast_fp16 = einsum(equation = var_59122_equation_0, values = (var_59040_cast_fp16, var_59107_cast_fp16))[name = tensor("op_59122_cast_fp16")]; + tensor var_59124_equation_0 = const()[name = tensor("op_59124_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_59124_cast_fp16 = einsum(equation = var_59124_equation_0, values = (var_59044_cast_fp16, var_59108_cast_fp16))[name = tensor("op_59124_cast_fp16")]; + tensor var_59126_equation_0 = const()[name = tensor("op_59126_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_59126_cast_fp16 = einsum(equation = var_59126_equation_0, values = (var_59048_cast_fp16, var_59109_cast_fp16))[name = tensor("op_59126_cast_fp16")]; + tensor var_59128_equation_0 = const()[name = tensor("op_59128_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_59128_cast_fp16 = einsum(equation = var_59128_equation_0, values = (var_59052_cast_fp16, var_59110_cast_fp16))[name = tensor("op_59128_cast_fp16")]; + tensor var_59130_equation_0 = const()[name = tensor("op_59130_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_59130_cast_fp16 = einsum(equation = var_59130_equation_0, values = (var_59056_cast_fp16, var_59111_cast_fp16))[name = tensor("op_59130_cast_fp16")]; + tensor var_59132_equation_0 = const()[name = tensor("op_59132_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_59132_cast_fp16 = einsum(equation = var_59132_equation_0, values = (var_59060_cast_fp16, var_59112_cast_fp16))[name = tensor("op_59132_cast_fp16")]; + tensor input_747_interleave_0 = const()[name = tensor("input_747_interleave_0"), val = tensor(false)]; + tensor input_747_cast_fp16 = concat(axis = var_58034, interleave = input_747_interleave_0, values = (var_59114_cast_fp16, var_59116_cast_fp16, var_59118_cast_fp16, var_59120_cast_fp16, var_59122_cast_fp16, var_59124_cast_fp16, var_59126_cast_fp16, var_59128_cast_fp16, var_59130_cast_fp16, var_59132_cast_fp16))[name = tensor("input_747_cast_fp16")]; + tensor var_59138 = const()[name = tensor("op_59138"), val = tensor([1, 1])]; + tensor var_59140 = const()[name = tensor("op_59140"), val = tensor([1, 1])]; + tensor var_59142_pad_type_0 = const()[name = tensor("op_59142_pad_type_0"), val = tensor("custom")]; + tensor var_59142_pad_0 = const()[name = tensor("op_59142_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_0_transformer_blocks_1_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_transformer_blocks_1_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4963778176)))]; + tensor up_blocks_1_attentions_0_transformer_blocks_1_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_transformer_blocks_1_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4964597440)))]; + tensor var_59142_cast_fp16 = conv(bias = up_blocks_1_attentions_0_transformer_blocks_1_attn2_to_out_0_bias_to_fp16, dilations = var_59140, groups = var_58034, pad = var_59142_pad_0, pad_type = var_59142_pad_type_0, strides = var_59138, weight = up_blocks_1_attentions_0_transformer_blocks_1_attn2_to_out_0_weight_to_fp16, x = input_747_cast_fp16)[name = tensor("op_59142_cast_fp16")]; + tensor inputs_395_cast_fp16 = add(x = var_59142_cast_fp16, y = inputs_393_cast_fp16)[name = tensor("inputs_395_cast_fp16")]; + tensor input_749_axes_0 = const()[name = tensor("input_749_axes_0"), val = tensor([1])]; + tensor input_749_gamma_0_to_fp16 = const()[name = tensor("input_749_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4964598784)))]; + tensor input_749_beta_0_to_fp16 = const()[name = tensor("input_749_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4964600128)))]; + tensor var_59152_to_fp16 = const()[name = tensor("op_59152_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_749_cast_fp16 = layer_norm(axes = input_749_axes_0, beta = input_749_beta_0_to_fp16, epsilon = var_59152_to_fp16, gamma = input_749_gamma_0_to_fp16, x = inputs_395_cast_fp16)[name = tensor("input_749_cast_fp16")]; + tensor var_59168 = const()[name = tensor("op_59168"), val = tensor([1, 1])]; + tensor var_59170 = const()[name = tensor("op_59170"), val = tensor([1, 1])]; + tensor var_59172_pad_type_0 = const()[name = tensor("op_59172_pad_type_0"), val = tensor("custom")]; + tensor var_59172_pad_0 = const()[name = tensor("op_59172_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_0_transformer_blocks_1_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_transformer_blocks_1_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4964601472)))]; + tensor up_blocks_1_attentions_0_transformer_blocks_1_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_transformer_blocks_1_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4971155136)))]; + tensor var_59172_cast_fp16 = conv(bias = up_blocks_1_attentions_0_transformer_blocks_1_ff_net_0_proj_bias_to_fp16, dilations = var_59170, groups = var_58034, pad = var_59172_pad_0, pad_type = var_59172_pad_type_0, strides = var_59168, weight = up_blocks_1_attentions_0_transformer_blocks_1_ff_net_0_proj_weight_to_fp16, x = input_749_cast_fp16)[name = tensor("op_59172_cast_fp16")]; + tensor var_59173_split_sizes_0 = const()[name = tensor("op_59173_split_sizes_0"), val = tensor([2560, 2560])]; + tensor var_59173_axis_0 = const()[name = tensor("op_59173_axis_0"), val = tensor(1)]; + tensor var_59173_cast_fp16_0, tensor var_59173_cast_fp16_1 = split(axis = var_59173_axis_0, split_sizes = var_59173_split_sizes_0, x = var_59172_cast_fp16)[name = tensor("op_59173_cast_fp16")]; + tensor var_59175_mode_0 = const()[name = tensor("op_59175_mode_0"), val = tensor("EXACT")]; + tensor var_59175_cast_fp16 = gelu(mode = var_59175_mode_0, x = var_59173_cast_fp16_1)[name = tensor("op_59175_cast_fp16")]; + tensor input_751_cast_fp16 = mul(x = var_59173_cast_fp16_0, y = var_59175_cast_fp16)[name = tensor("input_751_cast_fp16")]; + tensor var_59179 = const()[name = tensor("op_59179"), val = tensor([1, 1])]; + tensor var_59181 = const()[name = tensor("op_59181"), val = tensor([1, 1])]; + tensor var_59183_pad_type_0 = const()[name = tensor("op_59183_pad_type_0"), val = tensor("custom")]; + tensor var_59183_pad_0 = const()[name = tensor("op_59183_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_0_transformer_blocks_1_ff_net_2_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_transformer_blocks_1_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4971165440)))]; + tensor up_blocks_1_attentions_0_transformer_blocks_1_ff_net_2_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_transformer_blocks_1_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4974442304)))]; + tensor var_59183_cast_fp16 = conv(bias = up_blocks_1_attentions_0_transformer_blocks_1_ff_net_2_bias_to_fp16, dilations = var_59181, groups = var_58034, pad = var_59183_pad_0, pad_type = var_59183_pad_type_0, strides = var_59179, weight = up_blocks_1_attentions_0_transformer_blocks_1_ff_net_2_weight_to_fp16, x = input_751_cast_fp16)[name = tensor("op_59183_cast_fp16")]; + tensor hidden_states_521_cast_fp16 = add(x = var_59183_cast_fp16, y = inputs_395_cast_fp16)[name = tensor("hidden_states_521_cast_fp16")]; + tensor var_59185 = const()[name = tensor("op_59185"), val = tensor([2, 640, 64, 64])]; + tensor input_753_cast_fp16 = reshape(shape = var_59185, x = hidden_states_521_cast_fp16)[name = tensor("input_753_cast_fp16")]; + tensor var_59189 = const()[name = tensor("op_59189"), val = tensor([1, 1])]; + tensor var_59191 = const()[name = tensor("op_59191"), val = tensor([1, 1])]; + tensor hidden_states_523_pad_type_0 = const()[name = tensor("hidden_states_523_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_523_pad_0 = const()[name = tensor("hidden_states_523_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_0_proj_out_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_proj_out_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4974443648)))]; + tensor up_blocks_1_attentions_0_proj_out_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_proj_out_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4975262912)))]; + tensor hidden_states_523_cast_fp16 = conv(bias = up_blocks_1_attentions_0_proj_out_bias_to_fp16, dilations = var_59191, groups = var_58034, pad = hidden_states_523_pad_0, pad_type = hidden_states_523_pad_type_0, strides = var_59189, weight = up_blocks_1_attentions_0_proj_out_weight_to_fp16, x = input_753_cast_fp16)[name = tensor("hidden_states_523_cast_fp16")]; + tensor hidden_states_525_cast_fp16 = add(x = hidden_states_523_cast_fp16, y = hidden_states_505_cast_fp16)[name = tensor("hidden_states_525_cast_fp16")]; + tensor input_755_interleave_0 = const()[name = tensor("input_755_interleave_0"), val = tensor(false)]; + tensor input_755_cast_fp16 = concat(axis = var_58034, interleave = input_755_interleave_0, values = (hidden_states_525_cast_fp16, input_79_cast_fp16))[name = tensor("input_755_cast_fp16")]; + tensor reshape_132_shape_0 = const()[name = tensor("reshape_132_shape_0"), val = tensor([2, 32, 40, 64, 64])]; + tensor reshape_132_cast_fp16 = reshape(shape = reshape_132_shape_0, x = input_755_cast_fp16)[name = tensor("reshape_132_cast_fp16")]; + tensor reduce_mean_99_axes_0 = const()[name = tensor("reduce_mean_99_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_99_keep_dims_0 = const()[name = tensor("reduce_mean_99_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_99_cast_fp16 = reduce_mean(axes = reduce_mean_99_axes_0, keep_dims = reduce_mean_99_keep_dims_0, x = reshape_132_cast_fp16)[name = tensor("reduce_mean_99_cast_fp16")]; + tensor sub_66_cast_fp16 = sub(x = reshape_132_cast_fp16, y = reduce_mean_99_cast_fp16)[name = tensor("sub_66_cast_fp16")]; + tensor square_33_cast_fp16 = square(x = sub_66_cast_fp16)[name = tensor("square_33_cast_fp16")]; + tensor reduce_mean_101_axes_0 = const()[name = tensor("reduce_mean_101_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_101_keep_dims_0 = const()[name = tensor("reduce_mean_101_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_101_cast_fp16 = reduce_mean(axes = reduce_mean_101_axes_0, keep_dims = reduce_mean_101_keep_dims_0, x = square_33_cast_fp16)[name = tensor("reduce_mean_101_cast_fp16")]; + tensor add_66_y_0_to_fp16 = const()[name = tensor("add_66_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_66_cast_fp16 = add(x = reduce_mean_101_cast_fp16, y = add_66_y_0_to_fp16)[name = tensor("add_66_cast_fp16")]; + tensor sqrt_33_cast_fp16 = sqrt(x = add_66_cast_fp16)[name = tensor("sqrt_33_cast_fp16")]; + tensor real_div_33_cast_fp16 = real_div(x = sub_66_cast_fp16, y = sqrt_33_cast_fp16)[name = tensor("real_div_33_cast_fp16")]; + tensor reshape_133_shape_0 = const()[name = tensor("reshape_133_shape_0"), val = tensor([2, 1280, 64, 64])]; + tensor reshape_133_cast_fp16 = reshape(shape = reshape_133_shape_0, x = real_div_33_cast_fp16)[name = tensor("reshape_133_cast_fp16")]; + tensor add_67_gamma_0_to_fp16 = const()[name = tensor("add_67_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4975264256)))]; + tensor add_67_beta_0_to_fp16 = const()[name = tensor("add_67_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4975266880)))]; + tensor add_67_epsilon_0_to_fp16 = const()[name = tensor("add_67_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_67_cast_fp16 = batch_norm(beta = add_67_beta_0_to_fp16, epsilon = add_67_epsilon_0_to_fp16, gamma = add_67_gamma_0_to_fp16, mean = add_23_mean_0_to_fp16, variance = add_23_variance_0_to_fp16, x = reshape_133_cast_fp16)[name = tensor("add_67_cast_fp16")]; + tensor input_759_cast_fp16 = silu(x = add_67_cast_fp16)[name = tensor("input_759_cast_fp16")]; + tensor var_59209 = const()[name = tensor("op_59209"), val = tensor([1, 1])]; + tensor var_59211 = const()[name = tensor("op_59211"), val = tensor([1, 1])]; + tensor hidden_states_527_pad_type_0 = const()[name = tensor("hidden_states_527_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_527_pad_0 = const()[name = tensor("hidden_states_527_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor up_blocks_1_resnets_1_conv1_weight_to_fp16 = const()[name = tensor("up_blocks_1_resnets_1_conv1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4975269504)))]; + tensor up_blocks_1_resnets_1_conv1_bias_to_fp16 = const()[name = tensor("up_blocks_1_resnets_1_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4990015168)))]; + tensor hidden_states_527_cast_fp16 = conv(bias = up_blocks_1_resnets_1_conv1_bias_to_fp16, dilations = var_59211, groups = var_58034, pad = hidden_states_527_pad_0, pad_type = hidden_states_527_pad_type_0, strides = var_59209, weight = up_blocks_1_resnets_1_conv1_weight_to_fp16, x = input_759_cast_fp16)[name = tensor("hidden_states_527_cast_fp16")]; + tensor var_59217 = const()[name = tensor("op_59217"), val = tensor([1, 1])]; + tensor var_59219 = const()[name = tensor("op_59219"), val = tensor([1, 1])]; + tensor temb_25_pad_type_0 = const()[name = tensor("temb_25_pad_type_0"), val = tensor("custom")]; + tensor temb_25_pad_0 = const()[name = tensor("temb_25_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_resnets_1_time_emb_proj_weight_to_fp16 = const()[name = tensor("up_blocks_1_resnets_1_time_emb_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4990016512)))]; + tensor up_blocks_1_resnets_1_time_emb_proj_bias_to_fp16 = const()[name = tensor("up_blocks_1_resnets_1_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4991654976)))]; + tensor temb_25_cast_fp16 = conv(bias = up_blocks_1_resnets_1_time_emb_proj_bias_to_fp16, dilations = var_59219, groups = var_58034, pad = temb_25_pad_0, pad_type = temb_25_pad_type_0, strides = var_59217, weight = up_blocks_1_resnets_1_time_emb_proj_weight_to_fp16, x = input_21_cast_fp16)[name = tensor("temb_25_cast_fp16")]; + tensor input_763_cast_fp16 = add(x = hidden_states_527_cast_fp16, y = temb_25_cast_fp16)[name = tensor("input_763_cast_fp16")]; + tensor reshape_136_shape_0 = const()[name = tensor("reshape_136_shape_0"), val = tensor([2, 32, 20, 64, 64])]; + tensor reshape_136_cast_fp16 = reshape(shape = reshape_136_shape_0, x = input_763_cast_fp16)[name = tensor("reshape_136_cast_fp16")]; + tensor reduce_mean_102_axes_0 = const()[name = tensor("reduce_mean_102_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_102_keep_dims_0 = const()[name = tensor("reduce_mean_102_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_102_cast_fp16 = reduce_mean(axes = reduce_mean_102_axes_0, keep_dims = reduce_mean_102_keep_dims_0, x = reshape_136_cast_fp16)[name = tensor("reduce_mean_102_cast_fp16")]; + tensor sub_68_cast_fp16 = sub(x = reshape_136_cast_fp16, y = reduce_mean_102_cast_fp16)[name = tensor("sub_68_cast_fp16")]; + tensor square_34_cast_fp16 = square(x = sub_68_cast_fp16)[name = tensor("square_34_cast_fp16")]; + tensor reduce_mean_104_axes_0 = const()[name = tensor("reduce_mean_104_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_104_keep_dims_0 = const()[name = tensor("reduce_mean_104_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_104_cast_fp16 = reduce_mean(axes = reduce_mean_104_axes_0, keep_dims = reduce_mean_104_keep_dims_0, x = square_34_cast_fp16)[name = tensor("reduce_mean_104_cast_fp16")]; + tensor add_68_y_0_to_fp16 = const()[name = tensor("add_68_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_68_cast_fp16 = add(x = reduce_mean_104_cast_fp16, y = add_68_y_0_to_fp16)[name = tensor("add_68_cast_fp16")]; + tensor sqrt_34_cast_fp16 = sqrt(x = add_68_cast_fp16)[name = tensor("sqrt_34_cast_fp16")]; + tensor real_div_34_cast_fp16 = real_div(x = sub_68_cast_fp16, y = sqrt_34_cast_fp16)[name = tensor("real_div_34_cast_fp16")]; + tensor reshape_137_shape_0 = const()[name = tensor("reshape_137_shape_0"), val = tensor([2, 640, 64, 64])]; + tensor reshape_137_cast_fp16 = reshape(shape = reshape_137_shape_0, x = real_div_34_cast_fp16)[name = tensor("reshape_137_cast_fp16")]; + tensor add_69_gamma_0_to_fp16 = const()[name = tensor("add_69_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4991656320)))]; + tensor add_69_beta_0_to_fp16 = const()[name = tensor("add_69_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4991657664)))]; + tensor add_69_epsilon_0_to_fp16 = const()[name = tensor("add_69_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_69_cast_fp16 = batch_norm(beta = add_69_beta_0_to_fp16, epsilon = add_69_epsilon_0_to_fp16, gamma = add_69_gamma_0_to_fp16, mean = add_11_mean_0_to_fp16, variance = add_11_variance_0_to_fp16, x = reshape_137_cast_fp16)[name = tensor("add_69_cast_fp16")]; + tensor input_767_cast_fp16 = silu(x = add_69_cast_fp16)[name = tensor("input_767_cast_fp16")]; + tensor var_59229 = const()[name = tensor("op_59229"), val = tensor([1, 1])]; + tensor var_59231 = const()[name = tensor("op_59231"), val = tensor([1, 1])]; + tensor hidden_states_529_pad_type_0 = const()[name = tensor("hidden_states_529_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_529_pad_0 = const()[name = tensor("hidden_states_529_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor up_blocks_1_resnets_1_conv2_weight_to_fp16 = const()[name = tensor("up_blocks_1_resnets_1_conv2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4991659008)))]; + tensor up_blocks_1_resnets_1_conv2_bias_to_fp16 = const()[name = tensor("up_blocks_1_resnets_1_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4999031872)))]; + tensor hidden_states_529_cast_fp16 = conv(bias = up_blocks_1_resnets_1_conv2_bias_to_fp16, dilations = var_59231, groups = var_58034, pad = hidden_states_529_pad_0, pad_type = hidden_states_529_pad_type_0, strides = var_59229, weight = up_blocks_1_resnets_1_conv2_weight_to_fp16, x = input_767_cast_fp16)[name = tensor("hidden_states_529_cast_fp16")]; + tensor var_59236 = const()[name = tensor("op_59236"), val = tensor([1, 1])]; + tensor var_59238 = const()[name = tensor("op_59238"), val = tensor([1, 1])]; + tensor x_13_pad_type_0 = const()[name = tensor("x_13_pad_type_0"), val = tensor("custom")]; + tensor x_13_pad_0 = const()[name = tensor("x_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_resnets_1_conv_shortcut_weight_to_fp16 = const()[name = tensor("up_blocks_1_resnets_1_conv_shortcut_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4999033216)))]; + tensor up_blocks_1_resnets_1_conv_shortcut_bias_to_fp16 = const()[name = tensor("up_blocks_1_resnets_1_conv_shortcut_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5000671680)))]; + tensor x_13_cast_fp16 = conv(bias = up_blocks_1_resnets_1_conv_shortcut_bias_to_fp16, dilations = var_59238, groups = var_58034, pad = x_13_pad_0, pad_type = x_13_pad_type_0, strides = var_59236, weight = up_blocks_1_resnets_1_conv_shortcut_weight_to_fp16, x = input_755_cast_fp16)[name = tensor("x_13_cast_fp16")]; + tensor hidden_states_531_cast_fp16 = add(x = x_13_cast_fp16, y = hidden_states_529_cast_fp16)[name = tensor("hidden_states_531_cast_fp16")]; + tensor reshape_140_shape_0 = const()[name = tensor("reshape_140_shape_0"), val = tensor([2, 32, 20, 64, 64])]; + tensor reshape_140_cast_fp16 = reshape(shape = reshape_140_shape_0, x = hidden_states_531_cast_fp16)[name = tensor("reshape_140_cast_fp16")]; + tensor reduce_mean_105_axes_0 = const()[name = tensor("reduce_mean_105_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_105_keep_dims_0 = const()[name = tensor("reduce_mean_105_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_105_cast_fp16 = reduce_mean(axes = reduce_mean_105_axes_0, keep_dims = reduce_mean_105_keep_dims_0, x = reshape_140_cast_fp16)[name = tensor("reduce_mean_105_cast_fp16")]; + tensor sub_70_cast_fp16 = sub(x = reshape_140_cast_fp16, y = reduce_mean_105_cast_fp16)[name = tensor("sub_70_cast_fp16")]; + tensor square_35_cast_fp16 = square(x = sub_70_cast_fp16)[name = tensor("square_35_cast_fp16")]; + tensor reduce_mean_107_axes_0 = const()[name = tensor("reduce_mean_107_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_107_keep_dims_0 = const()[name = tensor("reduce_mean_107_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_107_cast_fp16 = reduce_mean(axes = reduce_mean_107_axes_0, keep_dims = reduce_mean_107_keep_dims_0, x = square_35_cast_fp16)[name = tensor("reduce_mean_107_cast_fp16")]; + tensor add_70_y_0_to_fp16 = const()[name = tensor("add_70_y_0_to_fp16"), val = tensor(0x1.1p-20)]; + tensor add_70_cast_fp16 = add(x = reduce_mean_107_cast_fp16, y = add_70_y_0_to_fp16)[name = tensor("add_70_cast_fp16")]; + tensor sqrt_35_cast_fp16 = sqrt(x = add_70_cast_fp16)[name = tensor("sqrt_35_cast_fp16")]; + tensor real_div_35_cast_fp16 = real_div(x = sub_70_cast_fp16, y = sqrt_35_cast_fp16)[name = tensor("real_div_35_cast_fp16")]; + tensor reshape_141_shape_0 = const()[name = tensor("reshape_141_shape_0"), val = tensor([2, 640, 64, 64])]; + tensor reshape_141_cast_fp16 = reshape(shape = reshape_141_shape_0, x = real_div_35_cast_fp16)[name = tensor("reshape_141_cast_fp16")]; + tensor add_71_gamma_0_to_fp16 = const()[name = tensor("add_71_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5000673024)))]; + tensor add_71_beta_0_to_fp16 = const()[name = tensor("add_71_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5000674368)))]; + tensor add_71_epsilon_0_to_fp16 = const()[name = tensor("add_71_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_71_cast_fp16 = batch_norm(beta = add_71_beta_0_to_fp16, epsilon = add_71_epsilon_0_to_fp16, gamma = add_71_gamma_0_to_fp16, mean = add_11_mean_0_to_fp16, variance = add_11_variance_0_to_fp16, x = reshape_141_cast_fp16)[name = tensor("add_71_cast_fp16")]; + tensor var_59260 = const()[name = tensor("op_59260"), val = tensor([1, 1])]; + tensor var_59262 = const()[name = tensor("op_59262"), val = tensor([1, 1])]; + tensor hidden_states_533_pad_type_0 = const()[name = tensor("hidden_states_533_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_533_pad_0 = const()[name = tensor("hidden_states_533_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_1_proj_in_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_proj_in_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5000675712)))]; + tensor up_blocks_1_attentions_1_proj_in_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_proj_in_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5001494976)))]; + tensor hidden_states_533_cast_fp16 = conv(bias = up_blocks_1_attentions_1_proj_in_bias_to_fp16, dilations = var_59262, groups = var_58034, pad = hidden_states_533_pad_0, pad_type = hidden_states_533_pad_type_0, strides = var_59260, weight = up_blocks_1_attentions_1_proj_in_weight_to_fp16, x = add_71_cast_fp16)[name = tensor("hidden_states_533_cast_fp16")]; + tensor var_59267 = const()[name = tensor("op_59267"), val = tensor([2, 640, 1, 4096])]; + tensor inputs_397_cast_fp16 = reshape(shape = var_59267, x = hidden_states_533_cast_fp16)[name = tensor("inputs_397_cast_fp16")]; + tensor hidden_states_535_axes_0 = const()[name = tensor("hidden_states_535_axes_0"), val = tensor([1])]; + tensor hidden_states_535_gamma_0_to_fp16 = const()[name = tensor("hidden_states_535_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5001496320)))]; + tensor hidden_states_535_beta_0_to_fp16 = const()[name = tensor("hidden_states_535_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5001497664)))]; + tensor var_59283_to_fp16 = const()[name = tensor("op_59283_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_535_cast_fp16 = layer_norm(axes = hidden_states_535_axes_0, beta = hidden_states_535_beta_0_to_fp16, epsilon = var_59283_to_fp16, gamma = hidden_states_535_gamma_0_to_fp16, x = inputs_397_cast_fp16)[name = tensor("hidden_states_535_cast_fp16")]; + tensor var_59298 = const()[name = tensor("op_59298"), val = tensor([1, 1])]; + tensor var_59300 = const()[name = tensor("op_59300"), val = tensor([1, 1])]; + tensor q_265_pad_type_0 = const()[name = tensor("q_265_pad_type_0"), val = tensor("custom")]; + tensor q_265_pad_0 = const()[name = tensor("q_265_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5001499008)))]; + tensor q_265_cast_fp16 = conv(dilations = var_59300, groups = var_58034, pad = q_265_pad_0, pad_type = q_265_pad_type_0, strides = var_59298, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16, x = hidden_states_535_cast_fp16)[name = tensor("q_265_cast_fp16")]; + tensor var_59304 = const()[name = tensor("op_59304"), val = tensor([1, 1])]; + tensor var_59306 = const()[name = tensor("op_59306"), val = tensor([1, 1])]; + tensor k_529_pad_type_0 = const()[name = tensor("k_529_pad_type_0"), val = tensor("custom")]; + tensor k_529_pad_0 = const()[name = tensor("k_529_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5002318272)))]; + tensor k_529_cast_fp16 = conv(dilations = var_59306, groups = var_58034, pad = k_529_pad_0, pad_type = k_529_pad_type_0, strides = var_59304, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16, x = hidden_states_535_cast_fp16)[name = tensor("k_529_cast_fp16")]; + tensor var_59310 = const()[name = tensor("op_59310"), val = tensor([1, 1])]; + tensor var_59312 = const()[name = tensor("op_59312"), val = tensor([1, 1])]; + tensor v_265_pad_type_0 = const()[name = tensor("v_265_pad_type_0"), val = tensor("custom")]; + tensor v_265_pad_0 = const()[name = tensor("v_265_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5003137536)))]; + tensor v_265_cast_fp16 = conv(dilations = var_59312, groups = var_58034, pad = v_265_pad_0, pad_type = v_265_pad_type_0, strides = var_59310, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16, x = hidden_states_535_cast_fp16)[name = tensor("v_265_cast_fp16")]; + tensor var_59316_begin_0 = const()[name = tensor("op_59316_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_59316_end_0 = const()[name = tensor("op_59316_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_59316_end_mask_0 = const()[name = tensor("op_59316_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59316_cast_fp16 = slice_by_index(begin = var_59316_begin_0, end = var_59316_end_0, end_mask = var_59316_end_mask_0, x = q_265_cast_fp16)[name = tensor("op_59316_cast_fp16")]; + tensor var_59320_begin_0 = const()[name = tensor("op_59320_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_59320_end_0 = const()[name = tensor("op_59320_end_0"), val = tensor([2, 128, 1, 4096])]; + tensor var_59320_end_mask_0 = const()[name = tensor("op_59320_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59320_cast_fp16 = slice_by_index(begin = var_59320_begin_0, end = var_59320_end_0, end_mask = var_59320_end_mask_0, x = q_265_cast_fp16)[name = tensor("op_59320_cast_fp16")]; + tensor var_59324_begin_0 = const()[name = tensor("op_59324_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_59324_end_0 = const()[name = tensor("op_59324_end_0"), val = tensor([2, 192, 1, 4096])]; + tensor var_59324_end_mask_0 = const()[name = tensor("op_59324_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59324_cast_fp16 = slice_by_index(begin = var_59324_begin_0, end = var_59324_end_0, end_mask = var_59324_end_mask_0, x = q_265_cast_fp16)[name = tensor("op_59324_cast_fp16")]; + tensor var_59328_begin_0 = const()[name = tensor("op_59328_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_59328_end_0 = const()[name = tensor("op_59328_end_0"), val = tensor([2, 256, 1, 4096])]; + tensor var_59328_end_mask_0 = const()[name = tensor("op_59328_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59328_cast_fp16 = slice_by_index(begin = var_59328_begin_0, end = var_59328_end_0, end_mask = var_59328_end_mask_0, x = q_265_cast_fp16)[name = tensor("op_59328_cast_fp16")]; + tensor var_59332_begin_0 = const()[name = tensor("op_59332_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_59332_end_0 = const()[name = tensor("op_59332_end_0"), val = tensor([2, 320, 1, 4096])]; + tensor var_59332_end_mask_0 = const()[name = tensor("op_59332_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59332_cast_fp16 = slice_by_index(begin = var_59332_begin_0, end = var_59332_end_0, end_mask = var_59332_end_mask_0, x = q_265_cast_fp16)[name = tensor("op_59332_cast_fp16")]; + tensor var_59336_begin_0 = const()[name = tensor("op_59336_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_59336_end_0 = const()[name = tensor("op_59336_end_0"), val = tensor([2, 384, 1, 4096])]; + tensor var_59336_end_mask_0 = const()[name = tensor("op_59336_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59336_cast_fp16 = slice_by_index(begin = var_59336_begin_0, end = var_59336_end_0, end_mask = var_59336_end_mask_0, x = q_265_cast_fp16)[name = tensor("op_59336_cast_fp16")]; + tensor var_59340_begin_0 = const()[name = tensor("op_59340_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_59340_end_0 = const()[name = tensor("op_59340_end_0"), val = tensor([2, 448, 1, 4096])]; + tensor var_59340_end_mask_0 = const()[name = tensor("op_59340_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59340_cast_fp16 = slice_by_index(begin = var_59340_begin_0, end = var_59340_end_0, end_mask = var_59340_end_mask_0, x = q_265_cast_fp16)[name = tensor("op_59340_cast_fp16")]; + tensor var_59344_begin_0 = const()[name = tensor("op_59344_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_59344_end_0 = const()[name = tensor("op_59344_end_0"), val = tensor([2, 512, 1, 4096])]; + tensor var_59344_end_mask_0 = const()[name = tensor("op_59344_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59344_cast_fp16 = slice_by_index(begin = var_59344_begin_0, end = var_59344_end_0, end_mask = var_59344_end_mask_0, x = q_265_cast_fp16)[name = tensor("op_59344_cast_fp16")]; + tensor var_59348_begin_0 = const()[name = tensor("op_59348_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_59348_end_0 = const()[name = tensor("op_59348_end_0"), val = tensor([2, 576, 1, 4096])]; + tensor var_59348_end_mask_0 = const()[name = tensor("op_59348_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59348_cast_fp16 = slice_by_index(begin = var_59348_begin_0, end = var_59348_end_0, end_mask = var_59348_end_mask_0, x = q_265_cast_fp16)[name = tensor("op_59348_cast_fp16")]; + tensor var_59352_begin_0 = const()[name = tensor("op_59352_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_59352_end_0 = const()[name = tensor("op_59352_end_0"), val = tensor([2, 640, 1, 4096])]; + tensor var_59352_end_mask_0 = const()[name = tensor("op_59352_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59352_cast_fp16 = slice_by_index(begin = var_59352_begin_0, end = var_59352_end_0, end_mask = var_59352_end_mask_0, x = q_265_cast_fp16)[name = tensor("op_59352_cast_fp16")]; + tensor k_531_perm_0 = const()[name = tensor("k_531_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_59359_begin_0 = const()[name = tensor("op_59359_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_59359_end_0 = const()[name = tensor("op_59359_end_0"), val = tensor([2, 4096, 1, 64])]; + tensor var_59359_end_mask_0 = const()[name = tensor("op_59359_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_531_cast_fp16 = transpose(perm = k_531_perm_0, x = k_529_cast_fp16)[name = tensor("transpose_7")]; + tensor var_59359_cast_fp16 = slice_by_index(begin = var_59359_begin_0, end = var_59359_end_0, end_mask = var_59359_end_mask_0, x = k_531_cast_fp16)[name = tensor("op_59359_cast_fp16")]; + tensor var_59363_begin_0 = const()[name = tensor("op_59363_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_59363_end_0 = const()[name = tensor("op_59363_end_0"), val = tensor([2, 4096, 1, 128])]; + tensor var_59363_end_mask_0 = const()[name = tensor("op_59363_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_59363_cast_fp16 = slice_by_index(begin = var_59363_begin_0, end = var_59363_end_0, end_mask = var_59363_end_mask_0, x = k_531_cast_fp16)[name = tensor("op_59363_cast_fp16")]; + tensor var_59367_begin_0 = const()[name = tensor("op_59367_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_59367_end_0 = const()[name = tensor("op_59367_end_0"), val = tensor([2, 4096, 1, 192])]; + tensor var_59367_end_mask_0 = const()[name = tensor("op_59367_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_59367_cast_fp16 = slice_by_index(begin = var_59367_begin_0, end = var_59367_end_0, end_mask = var_59367_end_mask_0, x = k_531_cast_fp16)[name = tensor("op_59367_cast_fp16")]; + tensor var_59371_begin_0 = const()[name = tensor("op_59371_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_59371_end_0 = const()[name = tensor("op_59371_end_0"), val = tensor([2, 4096, 1, 256])]; + tensor var_59371_end_mask_0 = const()[name = tensor("op_59371_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_59371_cast_fp16 = slice_by_index(begin = var_59371_begin_0, end = var_59371_end_0, end_mask = var_59371_end_mask_0, x = k_531_cast_fp16)[name = tensor("op_59371_cast_fp16")]; + tensor var_59375_begin_0 = const()[name = tensor("op_59375_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_59375_end_0 = const()[name = tensor("op_59375_end_0"), val = tensor([2, 4096, 1, 320])]; + tensor var_59375_end_mask_0 = const()[name = tensor("op_59375_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_59375_cast_fp16 = slice_by_index(begin = var_59375_begin_0, end = var_59375_end_0, end_mask = var_59375_end_mask_0, x = k_531_cast_fp16)[name = tensor("op_59375_cast_fp16")]; + tensor var_59379_begin_0 = const()[name = tensor("op_59379_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_59379_end_0 = const()[name = tensor("op_59379_end_0"), val = tensor([2, 4096, 1, 384])]; + tensor var_59379_end_mask_0 = const()[name = tensor("op_59379_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_59379_cast_fp16 = slice_by_index(begin = var_59379_begin_0, end = var_59379_end_0, end_mask = var_59379_end_mask_0, x = k_531_cast_fp16)[name = tensor("op_59379_cast_fp16")]; + tensor var_59383_begin_0 = const()[name = tensor("op_59383_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_59383_end_0 = const()[name = tensor("op_59383_end_0"), val = tensor([2, 4096, 1, 448])]; + tensor var_59383_end_mask_0 = const()[name = tensor("op_59383_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_59383_cast_fp16 = slice_by_index(begin = var_59383_begin_0, end = var_59383_end_0, end_mask = var_59383_end_mask_0, x = k_531_cast_fp16)[name = tensor("op_59383_cast_fp16")]; + tensor var_59387_begin_0 = const()[name = tensor("op_59387_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_59387_end_0 = const()[name = tensor("op_59387_end_0"), val = tensor([2, 4096, 1, 512])]; + tensor var_59387_end_mask_0 = const()[name = tensor("op_59387_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_59387_cast_fp16 = slice_by_index(begin = var_59387_begin_0, end = var_59387_end_0, end_mask = var_59387_end_mask_0, x = k_531_cast_fp16)[name = tensor("op_59387_cast_fp16")]; + tensor var_59391_begin_0 = const()[name = tensor("op_59391_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_59391_end_0 = const()[name = tensor("op_59391_end_0"), val = tensor([2, 4096, 1, 576])]; + tensor var_59391_end_mask_0 = const()[name = tensor("op_59391_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_59391_cast_fp16 = slice_by_index(begin = var_59391_begin_0, end = var_59391_end_0, end_mask = var_59391_end_mask_0, x = k_531_cast_fp16)[name = tensor("op_59391_cast_fp16")]; + tensor var_59395_begin_0 = const()[name = tensor("op_59395_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_59395_end_0 = const()[name = tensor("op_59395_end_0"), val = tensor([2, 4096, 1, 640])]; + tensor var_59395_end_mask_0 = const()[name = tensor("op_59395_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_59395_cast_fp16 = slice_by_index(begin = var_59395_begin_0, end = var_59395_end_0, end_mask = var_59395_end_mask_0, x = k_531_cast_fp16)[name = tensor("op_59395_cast_fp16")]; + tensor var_59397_begin_0 = const()[name = tensor("op_59397_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_59397_end_0 = const()[name = tensor("op_59397_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_59397_end_mask_0 = const()[name = tensor("op_59397_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59397_cast_fp16 = slice_by_index(begin = var_59397_begin_0, end = var_59397_end_0, end_mask = var_59397_end_mask_0, x = v_265_cast_fp16)[name = tensor("op_59397_cast_fp16")]; + tensor var_59401_begin_0 = const()[name = tensor("op_59401_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_59401_end_0 = const()[name = tensor("op_59401_end_0"), val = tensor([2, 128, 1, 4096])]; + tensor var_59401_end_mask_0 = const()[name = tensor("op_59401_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59401_cast_fp16 = slice_by_index(begin = var_59401_begin_0, end = var_59401_end_0, end_mask = var_59401_end_mask_0, x = v_265_cast_fp16)[name = tensor("op_59401_cast_fp16")]; + tensor var_59405_begin_0 = const()[name = tensor("op_59405_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_59405_end_0 = const()[name = tensor("op_59405_end_0"), val = tensor([2, 192, 1, 4096])]; + tensor var_59405_end_mask_0 = const()[name = tensor("op_59405_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59405_cast_fp16 = slice_by_index(begin = var_59405_begin_0, end = var_59405_end_0, end_mask = var_59405_end_mask_0, x = v_265_cast_fp16)[name = tensor("op_59405_cast_fp16")]; + tensor var_59409_begin_0 = const()[name = tensor("op_59409_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_59409_end_0 = const()[name = tensor("op_59409_end_0"), val = tensor([2, 256, 1, 4096])]; + tensor var_59409_end_mask_0 = const()[name = tensor("op_59409_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59409_cast_fp16 = slice_by_index(begin = var_59409_begin_0, end = var_59409_end_0, end_mask = var_59409_end_mask_0, x = v_265_cast_fp16)[name = tensor("op_59409_cast_fp16")]; + tensor var_59413_begin_0 = const()[name = tensor("op_59413_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_59413_end_0 = const()[name = tensor("op_59413_end_0"), val = tensor([2, 320, 1, 4096])]; + tensor var_59413_end_mask_0 = const()[name = tensor("op_59413_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59413_cast_fp16 = slice_by_index(begin = var_59413_begin_0, end = var_59413_end_0, end_mask = var_59413_end_mask_0, x = v_265_cast_fp16)[name = tensor("op_59413_cast_fp16")]; + tensor var_59417_begin_0 = const()[name = tensor("op_59417_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_59417_end_0 = const()[name = tensor("op_59417_end_0"), val = tensor([2, 384, 1, 4096])]; + tensor var_59417_end_mask_0 = const()[name = tensor("op_59417_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59417_cast_fp16 = slice_by_index(begin = var_59417_begin_0, end = var_59417_end_0, end_mask = var_59417_end_mask_0, x = v_265_cast_fp16)[name = tensor("op_59417_cast_fp16")]; + tensor var_59421_begin_0 = const()[name = tensor("op_59421_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_59421_end_0 = const()[name = tensor("op_59421_end_0"), val = tensor([2, 448, 1, 4096])]; + tensor var_59421_end_mask_0 = const()[name = tensor("op_59421_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59421_cast_fp16 = slice_by_index(begin = var_59421_begin_0, end = var_59421_end_0, end_mask = var_59421_end_mask_0, x = v_265_cast_fp16)[name = tensor("op_59421_cast_fp16")]; + tensor var_59425_begin_0 = const()[name = tensor("op_59425_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_59425_end_0 = const()[name = tensor("op_59425_end_0"), val = tensor([2, 512, 1, 4096])]; + tensor var_59425_end_mask_0 = const()[name = tensor("op_59425_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59425_cast_fp16 = slice_by_index(begin = var_59425_begin_0, end = var_59425_end_0, end_mask = var_59425_end_mask_0, x = v_265_cast_fp16)[name = tensor("op_59425_cast_fp16")]; + tensor var_59429_begin_0 = const()[name = tensor("op_59429_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_59429_end_0 = const()[name = tensor("op_59429_end_0"), val = tensor([2, 576, 1, 4096])]; + tensor var_59429_end_mask_0 = const()[name = tensor("op_59429_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59429_cast_fp16 = slice_by_index(begin = var_59429_begin_0, end = var_59429_end_0, end_mask = var_59429_end_mask_0, x = v_265_cast_fp16)[name = tensor("op_59429_cast_fp16")]; + tensor var_59433_begin_0 = const()[name = tensor("op_59433_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_59433_end_0 = const()[name = tensor("op_59433_end_0"), val = tensor([2, 640, 1, 4096])]; + tensor var_59433_end_mask_0 = const()[name = tensor("op_59433_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59433_cast_fp16 = slice_by_index(begin = var_59433_begin_0, end = var_59433_end_0, end_mask = var_59433_end_mask_0, x = v_265_cast_fp16)[name = tensor("op_59433_cast_fp16")]; + tensor var_59437_equation_0 = const()[name = tensor("op_59437_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_59437_cast_fp16 = einsum(equation = var_59437_equation_0, values = (var_59359_cast_fp16, var_59316_cast_fp16))[name = tensor("op_59437_cast_fp16")]; + tensor var_59438_to_fp16 = const()[name = tensor("op_59438_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5041_cast_fp16 = mul(x = var_59437_cast_fp16, y = var_59438_to_fp16)[name = tensor("aw_5041_cast_fp16")]; + tensor var_59441_equation_0 = const()[name = tensor("op_59441_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_59441_cast_fp16 = einsum(equation = var_59441_equation_0, values = (var_59363_cast_fp16, var_59320_cast_fp16))[name = tensor("op_59441_cast_fp16")]; + tensor var_59442_to_fp16 = const()[name = tensor("op_59442_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5043_cast_fp16 = mul(x = var_59441_cast_fp16, y = var_59442_to_fp16)[name = tensor("aw_5043_cast_fp16")]; + tensor var_59445_equation_0 = const()[name = tensor("op_59445_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_59445_cast_fp16 = einsum(equation = var_59445_equation_0, values = (var_59367_cast_fp16, var_59324_cast_fp16))[name = tensor("op_59445_cast_fp16")]; + tensor var_59446_to_fp16 = const()[name = tensor("op_59446_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5045_cast_fp16 = mul(x = var_59445_cast_fp16, y = var_59446_to_fp16)[name = tensor("aw_5045_cast_fp16")]; + tensor var_59449_equation_0 = const()[name = tensor("op_59449_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_59449_cast_fp16 = einsum(equation = var_59449_equation_0, values = (var_59371_cast_fp16, var_59328_cast_fp16))[name = tensor("op_59449_cast_fp16")]; + tensor var_59450_to_fp16 = const()[name = tensor("op_59450_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5047_cast_fp16 = mul(x = var_59449_cast_fp16, y = var_59450_to_fp16)[name = tensor("aw_5047_cast_fp16")]; + tensor var_59453_equation_0 = const()[name = tensor("op_59453_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_59453_cast_fp16 = einsum(equation = var_59453_equation_0, values = (var_59375_cast_fp16, var_59332_cast_fp16))[name = tensor("op_59453_cast_fp16")]; + tensor var_59454_to_fp16 = const()[name = tensor("op_59454_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5049_cast_fp16 = mul(x = var_59453_cast_fp16, y = var_59454_to_fp16)[name = tensor("aw_5049_cast_fp16")]; + tensor var_59457_equation_0 = const()[name = tensor("op_59457_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_59457_cast_fp16 = einsum(equation = var_59457_equation_0, values = (var_59379_cast_fp16, var_59336_cast_fp16))[name = tensor("op_59457_cast_fp16")]; + tensor var_59458_to_fp16 = const()[name = tensor("op_59458_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5051_cast_fp16 = mul(x = var_59457_cast_fp16, y = var_59458_to_fp16)[name = tensor("aw_5051_cast_fp16")]; + tensor var_59461_equation_0 = const()[name = tensor("op_59461_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_59461_cast_fp16 = einsum(equation = var_59461_equation_0, values = (var_59383_cast_fp16, var_59340_cast_fp16))[name = tensor("op_59461_cast_fp16")]; + tensor var_59462_to_fp16 = const()[name = tensor("op_59462_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5053_cast_fp16 = mul(x = var_59461_cast_fp16, y = var_59462_to_fp16)[name = tensor("aw_5053_cast_fp16")]; + tensor var_59465_equation_0 = const()[name = tensor("op_59465_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_59465_cast_fp16 = einsum(equation = var_59465_equation_0, values = (var_59387_cast_fp16, var_59344_cast_fp16))[name = tensor("op_59465_cast_fp16")]; + tensor var_59466_to_fp16 = const()[name = tensor("op_59466_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5055_cast_fp16 = mul(x = var_59465_cast_fp16, y = var_59466_to_fp16)[name = tensor("aw_5055_cast_fp16")]; + tensor var_59469_equation_0 = const()[name = tensor("op_59469_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_59469_cast_fp16 = einsum(equation = var_59469_equation_0, values = (var_59391_cast_fp16, var_59348_cast_fp16))[name = tensor("op_59469_cast_fp16")]; + tensor var_59470_to_fp16 = const()[name = tensor("op_59470_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5057_cast_fp16 = mul(x = var_59469_cast_fp16, y = var_59470_to_fp16)[name = tensor("aw_5057_cast_fp16")]; + tensor var_59473_equation_0 = const()[name = tensor("op_59473_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_59473_cast_fp16 = einsum(equation = var_59473_equation_0, values = (var_59395_cast_fp16, var_59352_cast_fp16))[name = tensor("op_59473_cast_fp16")]; + tensor var_59474_to_fp16 = const()[name = tensor("op_59474_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5059_cast_fp16 = mul(x = var_59473_cast_fp16, y = var_59474_to_fp16)[name = tensor("aw_5059_cast_fp16")]; + tensor var_59476_cast_fp16 = softmax(axis = var_58034, x = aw_5041_cast_fp16)[name = tensor("op_59476_cast_fp16")]; + tensor var_59477_cast_fp16 = softmax(axis = var_58034, x = aw_5043_cast_fp16)[name = tensor("op_59477_cast_fp16")]; + tensor var_59478_cast_fp16 = softmax(axis = var_58034, x = aw_5045_cast_fp16)[name = tensor("op_59478_cast_fp16")]; + tensor var_59479_cast_fp16 = softmax(axis = var_58034, x = aw_5047_cast_fp16)[name = tensor("op_59479_cast_fp16")]; + tensor var_59480_cast_fp16 = softmax(axis = var_58034, x = aw_5049_cast_fp16)[name = tensor("op_59480_cast_fp16")]; + tensor var_59481_cast_fp16 = softmax(axis = var_58034, x = aw_5051_cast_fp16)[name = tensor("op_59481_cast_fp16")]; + tensor var_59482_cast_fp16 = softmax(axis = var_58034, x = aw_5053_cast_fp16)[name = tensor("op_59482_cast_fp16")]; + tensor var_59483_cast_fp16 = softmax(axis = var_58034, x = aw_5055_cast_fp16)[name = tensor("op_59483_cast_fp16")]; + tensor var_59484_cast_fp16 = softmax(axis = var_58034, x = aw_5057_cast_fp16)[name = tensor("op_59484_cast_fp16")]; + tensor var_59485_cast_fp16 = softmax(axis = var_58034, x = aw_5059_cast_fp16)[name = tensor("op_59485_cast_fp16")]; + tensor var_59487_equation_0 = const()[name = tensor("op_59487_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_59487_cast_fp16 = einsum(equation = var_59487_equation_0, values = (var_59397_cast_fp16, var_59476_cast_fp16))[name = tensor("op_59487_cast_fp16")]; + tensor var_59489_equation_0 = const()[name = tensor("op_59489_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_59489_cast_fp16 = einsum(equation = var_59489_equation_0, values = (var_59401_cast_fp16, var_59477_cast_fp16))[name = tensor("op_59489_cast_fp16")]; + tensor var_59491_equation_0 = const()[name = tensor("op_59491_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_59491_cast_fp16 = einsum(equation = var_59491_equation_0, values = (var_59405_cast_fp16, var_59478_cast_fp16))[name = tensor("op_59491_cast_fp16")]; + tensor var_59493_equation_0 = const()[name = tensor("op_59493_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_59493_cast_fp16 = einsum(equation = var_59493_equation_0, values = (var_59409_cast_fp16, var_59479_cast_fp16))[name = tensor("op_59493_cast_fp16")]; + tensor var_59495_equation_0 = const()[name = tensor("op_59495_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_59495_cast_fp16 = einsum(equation = var_59495_equation_0, values = (var_59413_cast_fp16, var_59480_cast_fp16))[name = tensor("op_59495_cast_fp16")]; + tensor var_59497_equation_0 = const()[name = tensor("op_59497_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_59497_cast_fp16 = einsum(equation = var_59497_equation_0, values = (var_59417_cast_fp16, var_59481_cast_fp16))[name = tensor("op_59497_cast_fp16")]; + tensor var_59499_equation_0 = const()[name = tensor("op_59499_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_59499_cast_fp16 = einsum(equation = var_59499_equation_0, values = (var_59421_cast_fp16, var_59482_cast_fp16))[name = tensor("op_59499_cast_fp16")]; + tensor var_59501_equation_0 = const()[name = tensor("op_59501_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_59501_cast_fp16 = einsum(equation = var_59501_equation_0, values = (var_59425_cast_fp16, var_59483_cast_fp16))[name = tensor("op_59501_cast_fp16")]; + tensor var_59503_equation_0 = const()[name = tensor("op_59503_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_59503_cast_fp16 = einsum(equation = var_59503_equation_0, values = (var_59429_cast_fp16, var_59484_cast_fp16))[name = tensor("op_59503_cast_fp16")]; + tensor var_59505_equation_0 = const()[name = tensor("op_59505_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_59505_cast_fp16 = einsum(equation = var_59505_equation_0, values = (var_59433_cast_fp16, var_59485_cast_fp16))[name = tensor("op_59505_cast_fp16")]; + tensor input_771_interleave_0 = const()[name = tensor("input_771_interleave_0"), val = tensor(false)]; + tensor input_771_cast_fp16 = concat(axis = var_58034, interleave = input_771_interleave_0, values = (var_59487_cast_fp16, var_59489_cast_fp16, var_59491_cast_fp16, var_59493_cast_fp16, var_59495_cast_fp16, var_59497_cast_fp16, var_59499_cast_fp16, var_59501_cast_fp16, var_59503_cast_fp16, var_59505_cast_fp16))[name = tensor("input_771_cast_fp16")]; + tensor var_59511 = const()[name = tensor("op_59511"), val = tensor([1, 1])]; + tensor var_59513 = const()[name = tensor("op_59513"), val = tensor([1, 1])]; + tensor var_59515_pad_type_0 = const()[name = tensor("op_59515_pad_type_0"), val = tensor("custom")]; + tensor var_59515_pad_0 = const()[name = tensor("op_59515_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5003956800)))]; + tensor up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5004776064)))]; + tensor var_59515_cast_fp16 = conv(bias = up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_59513, groups = var_58034, pad = var_59515_pad_0, pad_type = var_59515_pad_type_0, strides = var_59511, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16, x = input_771_cast_fp16)[name = tensor("op_59515_cast_fp16")]; + tensor inputs_399_cast_fp16 = add(x = var_59515_cast_fp16, y = inputs_397_cast_fp16)[name = tensor("inputs_399_cast_fp16")]; + tensor hidden_states_537_axes_0 = const()[name = tensor("hidden_states_537_axes_0"), val = tensor([1])]; + tensor hidden_states_537_gamma_0_to_fp16 = const()[name = tensor("hidden_states_537_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5004777408)))]; + tensor hidden_states_537_beta_0_to_fp16 = const()[name = tensor("hidden_states_537_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5004778752)))]; + tensor var_59525_to_fp16 = const()[name = tensor("op_59525_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_537_cast_fp16 = layer_norm(axes = hidden_states_537_axes_0, beta = hidden_states_537_beta_0_to_fp16, epsilon = var_59525_to_fp16, gamma = hidden_states_537_gamma_0_to_fp16, x = inputs_399_cast_fp16)[name = tensor("hidden_states_537_cast_fp16")]; + tensor var_59540 = const()[name = tensor("op_59540"), val = tensor([1, 1])]; + tensor var_59542 = const()[name = tensor("op_59542"), val = tensor([1, 1])]; + tensor q_267_pad_type_0 = const()[name = tensor("q_267_pad_type_0"), val = tensor("custom")]; + tensor q_267_pad_0 = const()[name = tensor("q_267_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5004780096)))]; + tensor q_267_cast_fp16 = conv(dilations = var_59542, groups = var_58034, pad = q_267_pad_0, pad_type = q_267_pad_type_0, strides = var_59540, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16, x = hidden_states_537_cast_fp16)[name = tensor("q_267_cast_fp16")]; + tensor var_59546 = const()[name = tensor("op_59546"), val = tensor([1, 1])]; + tensor var_59548 = const()[name = tensor("op_59548"), val = tensor([1, 1])]; + tensor k_533_pad_type_0 = const()[name = tensor("k_533_pad_type_0"), val = tensor("custom")]; + tensor k_533_pad_0 = const()[name = tensor("k_533_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5005599360)))]; + tensor k_533_cast_fp16 = conv(dilations = var_59548, groups = var_58034, pad = k_533_pad_0, pad_type = k_533_pad_type_0, strides = var_59546, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_533_cast_fp16")]; + tensor var_59552 = const()[name = tensor("op_59552"), val = tensor([1, 1])]; + tensor var_59554 = const()[name = tensor("op_59554"), val = tensor([1, 1])]; + tensor v_267_pad_type_0 = const()[name = tensor("v_267_pad_type_0"), val = tensor("custom")]; + tensor v_267_pad_0 = const()[name = tensor("v_267_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5008220864)))]; + tensor v_267_cast_fp16 = conv(dilations = var_59554, groups = var_58034, pad = v_267_pad_0, pad_type = v_267_pad_type_0, strides = var_59552, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_267_cast_fp16")]; + tensor var_59558_begin_0 = const()[name = tensor("op_59558_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_59558_end_0 = const()[name = tensor("op_59558_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_59558_end_mask_0 = const()[name = tensor("op_59558_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59558_cast_fp16 = slice_by_index(begin = var_59558_begin_0, end = var_59558_end_0, end_mask = var_59558_end_mask_0, x = q_267_cast_fp16)[name = tensor("op_59558_cast_fp16")]; + tensor var_59562_begin_0 = const()[name = tensor("op_59562_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_59562_end_0 = const()[name = tensor("op_59562_end_0"), val = tensor([2, 128, 1, 4096])]; + tensor var_59562_end_mask_0 = const()[name = tensor("op_59562_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59562_cast_fp16 = slice_by_index(begin = var_59562_begin_0, end = var_59562_end_0, end_mask = var_59562_end_mask_0, x = q_267_cast_fp16)[name = tensor("op_59562_cast_fp16")]; + tensor var_59566_begin_0 = const()[name = tensor("op_59566_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_59566_end_0 = const()[name = tensor("op_59566_end_0"), val = tensor([2, 192, 1, 4096])]; + tensor var_59566_end_mask_0 = const()[name = tensor("op_59566_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59566_cast_fp16 = slice_by_index(begin = var_59566_begin_0, end = var_59566_end_0, end_mask = var_59566_end_mask_0, x = q_267_cast_fp16)[name = tensor("op_59566_cast_fp16")]; + tensor var_59570_begin_0 = const()[name = tensor("op_59570_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_59570_end_0 = const()[name = tensor("op_59570_end_0"), val = tensor([2, 256, 1, 4096])]; + tensor var_59570_end_mask_0 = const()[name = tensor("op_59570_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59570_cast_fp16 = slice_by_index(begin = var_59570_begin_0, end = var_59570_end_0, end_mask = var_59570_end_mask_0, x = q_267_cast_fp16)[name = tensor("op_59570_cast_fp16")]; + tensor var_59574_begin_0 = const()[name = tensor("op_59574_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_59574_end_0 = const()[name = tensor("op_59574_end_0"), val = tensor([2, 320, 1, 4096])]; + tensor var_59574_end_mask_0 = const()[name = tensor("op_59574_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59574_cast_fp16 = slice_by_index(begin = var_59574_begin_0, end = var_59574_end_0, end_mask = var_59574_end_mask_0, x = q_267_cast_fp16)[name = tensor("op_59574_cast_fp16")]; + tensor var_59578_begin_0 = const()[name = tensor("op_59578_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_59578_end_0 = const()[name = tensor("op_59578_end_0"), val = tensor([2, 384, 1, 4096])]; + tensor var_59578_end_mask_0 = const()[name = tensor("op_59578_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59578_cast_fp16 = slice_by_index(begin = var_59578_begin_0, end = var_59578_end_0, end_mask = var_59578_end_mask_0, x = q_267_cast_fp16)[name = tensor("op_59578_cast_fp16")]; + tensor var_59582_begin_0 = const()[name = tensor("op_59582_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_59582_end_0 = const()[name = tensor("op_59582_end_0"), val = tensor([2, 448, 1, 4096])]; + tensor var_59582_end_mask_0 = const()[name = tensor("op_59582_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59582_cast_fp16 = slice_by_index(begin = var_59582_begin_0, end = var_59582_end_0, end_mask = var_59582_end_mask_0, x = q_267_cast_fp16)[name = tensor("op_59582_cast_fp16")]; + tensor var_59586_begin_0 = const()[name = tensor("op_59586_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_59586_end_0 = const()[name = tensor("op_59586_end_0"), val = tensor([2, 512, 1, 4096])]; + tensor var_59586_end_mask_0 = const()[name = tensor("op_59586_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59586_cast_fp16 = slice_by_index(begin = var_59586_begin_0, end = var_59586_end_0, end_mask = var_59586_end_mask_0, x = q_267_cast_fp16)[name = tensor("op_59586_cast_fp16")]; + tensor var_59590_begin_0 = const()[name = tensor("op_59590_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_59590_end_0 = const()[name = tensor("op_59590_end_0"), val = tensor([2, 576, 1, 4096])]; + tensor var_59590_end_mask_0 = const()[name = tensor("op_59590_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59590_cast_fp16 = slice_by_index(begin = var_59590_begin_0, end = var_59590_end_0, end_mask = var_59590_end_mask_0, x = q_267_cast_fp16)[name = tensor("op_59590_cast_fp16")]; + tensor var_59594_begin_0 = const()[name = tensor("op_59594_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_59594_end_0 = const()[name = tensor("op_59594_end_0"), val = tensor([2, 640, 1, 4096])]; + tensor var_59594_end_mask_0 = const()[name = tensor("op_59594_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59594_cast_fp16 = slice_by_index(begin = var_59594_begin_0, end = var_59594_end_0, end_mask = var_59594_end_mask_0, x = q_267_cast_fp16)[name = tensor("op_59594_cast_fp16")]; + tensor k_535_perm_0 = const()[name = tensor("k_535_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_59601_begin_0 = const()[name = tensor("op_59601_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_59601_end_0 = const()[name = tensor("op_59601_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_59601_end_mask_0 = const()[name = tensor("op_59601_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_535_cast_fp16 = transpose(perm = k_535_perm_0, x = k_533_cast_fp16)[name = tensor("transpose_6")]; + tensor var_59601_cast_fp16 = slice_by_index(begin = var_59601_begin_0, end = var_59601_end_0, end_mask = var_59601_end_mask_0, x = k_535_cast_fp16)[name = tensor("op_59601_cast_fp16")]; + tensor var_59605_begin_0 = const()[name = tensor("op_59605_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_59605_end_0 = const()[name = tensor("op_59605_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_59605_end_mask_0 = const()[name = tensor("op_59605_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_59605_cast_fp16 = slice_by_index(begin = var_59605_begin_0, end = var_59605_end_0, end_mask = var_59605_end_mask_0, x = k_535_cast_fp16)[name = tensor("op_59605_cast_fp16")]; + tensor var_59609_begin_0 = const()[name = tensor("op_59609_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_59609_end_0 = const()[name = tensor("op_59609_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_59609_end_mask_0 = const()[name = tensor("op_59609_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_59609_cast_fp16 = slice_by_index(begin = var_59609_begin_0, end = var_59609_end_0, end_mask = var_59609_end_mask_0, x = k_535_cast_fp16)[name = tensor("op_59609_cast_fp16")]; + tensor var_59613_begin_0 = const()[name = tensor("op_59613_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_59613_end_0 = const()[name = tensor("op_59613_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_59613_end_mask_0 = const()[name = tensor("op_59613_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_59613_cast_fp16 = slice_by_index(begin = var_59613_begin_0, end = var_59613_end_0, end_mask = var_59613_end_mask_0, x = k_535_cast_fp16)[name = tensor("op_59613_cast_fp16")]; + tensor var_59617_begin_0 = const()[name = tensor("op_59617_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_59617_end_0 = const()[name = tensor("op_59617_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_59617_end_mask_0 = const()[name = tensor("op_59617_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_59617_cast_fp16 = slice_by_index(begin = var_59617_begin_0, end = var_59617_end_0, end_mask = var_59617_end_mask_0, x = k_535_cast_fp16)[name = tensor("op_59617_cast_fp16")]; + tensor var_59621_begin_0 = const()[name = tensor("op_59621_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_59621_end_0 = const()[name = tensor("op_59621_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_59621_end_mask_0 = const()[name = tensor("op_59621_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_59621_cast_fp16 = slice_by_index(begin = var_59621_begin_0, end = var_59621_end_0, end_mask = var_59621_end_mask_0, x = k_535_cast_fp16)[name = tensor("op_59621_cast_fp16")]; + tensor var_59625_begin_0 = const()[name = tensor("op_59625_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_59625_end_0 = const()[name = tensor("op_59625_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_59625_end_mask_0 = const()[name = tensor("op_59625_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_59625_cast_fp16 = slice_by_index(begin = var_59625_begin_0, end = var_59625_end_0, end_mask = var_59625_end_mask_0, x = k_535_cast_fp16)[name = tensor("op_59625_cast_fp16")]; + tensor var_59629_begin_0 = const()[name = tensor("op_59629_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_59629_end_0 = const()[name = tensor("op_59629_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_59629_end_mask_0 = const()[name = tensor("op_59629_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_59629_cast_fp16 = slice_by_index(begin = var_59629_begin_0, end = var_59629_end_0, end_mask = var_59629_end_mask_0, x = k_535_cast_fp16)[name = tensor("op_59629_cast_fp16")]; + tensor var_59633_begin_0 = const()[name = tensor("op_59633_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_59633_end_0 = const()[name = tensor("op_59633_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_59633_end_mask_0 = const()[name = tensor("op_59633_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_59633_cast_fp16 = slice_by_index(begin = var_59633_begin_0, end = var_59633_end_0, end_mask = var_59633_end_mask_0, x = k_535_cast_fp16)[name = tensor("op_59633_cast_fp16")]; + tensor var_59637_begin_0 = const()[name = tensor("op_59637_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_59637_end_0 = const()[name = tensor("op_59637_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_59637_end_mask_0 = const()[name = tensor("op_59637_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_59637_cast_fp16 = slice_by_index(begin = var_59637_begin_0, end = var_59637_end_0, end_mask = var_59637_end_mask_0, x = k_535_cast_fp16)[name = tensor("op_59637_cast_fp16")]; + tensor var_59639_begin_0 = const()[name = tensor("op_59639_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_59639_end_0 = const()[name = tensor("op_59639_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_59639_end_mask_0 = const()[name = tensor("op_59639_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59639_cast_fp16 = slice_by_index(begin = var_59639_begin_0, end = var_59639_end_0, end_mask = var_59639_end_mask_0, x = v_267_cast_fp16)[name = tensor("op_59639_cast_fp16")]; + tensor var_59643_begin_0 = const()[name = tensor("op_59643_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_59643_end_0 = const()[name = tensor("op_59643_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_59643_end_mask_0 = const()[name = tensor("op_59643_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59643_cast_fp16 = slice_by_index(begin = var_59643_begin_0, end = var_59643_end_0, end_mask = var_59643_end_mask_0, x = v_267_cast_fp16)[name = tensor("op_59643_cast_fp16")]; + tensor var_59647_begin_0 = const()[name = tensor("op_59647_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_59647_end_0 = const()[name = tensor("op_59647_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_59647_end_mask_0 = const()[name = tensor("op_59647_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59647_cast_fp16 = slice_by_index(begin = var_59647_begin_0, end = var_59647_end_0, end_mask = var_59647_end_mask_0, x = v_267_cast_fp16)[name = tensor("op_59647_cast_fp16")]; + tensor var_59651_begin_0 = const()[name = tensor("op_59651_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_59651_end_0 = const()[name = tensor("op_59651_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_59651_end_mask_0 = const()[name = tensor("op_59651_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59651_cast_fp16 = slice_by_index(begin = var_59651_begin_0, end = var_59651_end_0, end_mask = var_59651_end_mask_0, x = v_267_cast_fp16)[name = tensor("op_59651_cast_fp16")]; + tensor var_59655_begin_0 = const()[name = tensor("op_59655_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_59655_end_0 = const()[name = tensor("op_59655_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_59655_end_mask_0 = const()[name = tensor("op_59655_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59655_cast_fp16 = slice_by_index(begin = var_59655_begin_0, end = var_59655_end_0, end_mask = var_59655_end_mask_0, x = v_267_cast_fp16)[name = tensor("op_59655_cast_fp16")]; + tensor var_59659_begin_0 = const()[name = tensor("op_59659_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_59659_end_0 = const()[name = tensor("op_59659_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_59659_end_mask_0 = const()[name = tensor("op_59659_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59659_cast_fp16 = slice_by_index(begin = var_59659_begin_0, end = var_59659_end_0, end_mask = var_59659_end_mask_0, x = v_267_cast_fp16)[name = tensor("op_59659_cast_fp16")]; + tensor var_59663_begin_0 = const()[name = tensor("op_59663_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_59663_end_0 = const()[name = tensor("op_59663_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_59663_end_mask_0 = const()[name = tensor("op_59663_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59663_cast_fp16 = slice_by_index(begin = var_59663_begin_0, end = var_59663_end_0, end_mask = var_59663_end_mask_0, x = v_267_cast_fp16)[name = tensor("op_59663_cast_fp16")]; + tensor var_59667_begin_0 = const()[name = tensor("op_59667_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_59667_end_0 = const()[name = tensor("op_59667_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_59667_end_mask_0 = const()[name = tensor("op_59667_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59667_cast_fp16 = slice_by_index(begin = var_59667_begin_0, end = var_59667_end_0, end_mask = var_59667_end_mask_0, x = v_267_cast_fp16)[name = tensor("op_59667_cast_fp16")]; + tensor var_59671_begin_0 = const()[name = tensor("op_59671_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_59671_end_0 = const()[name = tensor("op_59671_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_59671_end_mask_0 = const()[name = tensor("op_59671_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59671_cast_fp16 = slice_by_index(begin = var_59671_begin_0, end = var_59671_end_0, end_mask = var_59671_end_mask_0, x = v_267_cast_fp16)[name = tensor("op_59671_cast_fp16")]; + tensor var_59675_begin_0 = const()[name = tensor("op_59675_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_59675_end_0 = const()[name = tensor("op_59675_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_59675_end_mask_0 = const()[name = tensor("op_59675_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59675_cast_fp16 = slice_by_index(begin = var_59675_begin_0, end = var_59675_end_0, end_mask = var_59675_end_mask_0, x = v_267_cast_fp16)[name = tensor("op_59675_cast_fp16")]; + tensor var_59679_equation_0 = const()[name = tensor("op_59679_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_59679_cast_fp16 = einsum(equation = var_59679_equation_0, values = (var_59601_cast_fp16, var_59558_cast_fp16))[name = tensor("op_59679_cast_fp16")]; + tensor var_59680_to_fp16 = const()[name = tensor("op_59680_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5061_cast_fp16 = mul(x = var_59679_cast_fp16, y = var_59680_to_fp16)[name = tensor("aw_5061_cast_fp16")]; + tensor var_59683_equation_0 = const()[name = tensor("op_59683_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_59683_cast_fp16 = einsum(equation = var_59683_equation_0, values = (var_59605_cast_fp16, var_59562_cast_fp16))[name = tensor("op_59683_cast_fp16")]; + tensor var_59684_to_fp16 = const()[name = tensor("op_59684_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5063_cast_fp16 = mul(x = var_59683_cast_fp16, y = var_59684_to_fp16)[name = tensor("aw_5063_cast_fp16")]; + tensor var_59687_equation_0 = const()[name = tensor("op_59687_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_59687_cast_fp16 = einsum(equation = var_59687_equation_0, values = (var_59609_cast_fp16, var_59566_cast_fp16))[name = tensor("op_59687_cast_fp16")]; + tensor var_59688_to_fp16 = const()[name = tensor("op_59688_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5065_cast_fp16 = mul(x = var_59687_cast_fp16, y = var_59688_to_fp16)[name = tensor("aw_5065_cast_fp16")]; + tensor var_59691_equation_0 = const()[name = tensor("op_59691_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_59691_cast_fp16 = einsum(equation = var_59691_equation_0, values = (var_59613_cast_fp16, var_59570_cast_fp16))[name = tensor("op_59691_cast_fp16")]; + tensor var_59692_to_fp16 = const()[name = tensor("op_59692_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5067_cast_fp16 = mul(x = var_59691_cast_fp16, y = var_59692_to_fp16)[name = tensor("aw_5067_cast_fp16")]; + tensor var_59695_equation_0 = const()[name = tensor("op_59695_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_59695_cast_fp16 = einsum(equation = var_59695_equation_0, values = (var_59617_cast_fp16, var_59574_cast_fp16))[name = tensor("op_59695_cast_fp16")]; + tensor var_59696_to_fp16 = const()[name = tensor("op_59696_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5069_cast_fp16 = mul(x = var_59695_cast_fp16, y = var_59696_to_fp16)[name = tensor("aw_5069_cast_fp16")]; + tensor var_59699_equation_0 = const()[name = tensor("op_59699_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_59699_cast_fp16 = einsum(equation = var_59699_equation_0, values = (var_59621_cast_fp16, var_59578_cast_fp16))[name = tensor("op_59699_cast_fp16")]; + tensor var_59700_to_fp16 = const()[name = tensor("op_59700_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5071_cast_fp16 = mul(x = var_59699_cast_fp16, y = var_59700_to_fp16)[name = tensor("aw_5071_cast_fp16")]; + tensor var_59703_equation_0 = const()[name = tensor("op_59703_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_59703_cast_fp16 = einsum(equation = var_59703_equation_0, values = (var_59625_cast_fp16, var_59582_cast_fp16))[name = tensor("op_59703_cast_fp16")]; + tensor var_59704_to_fp16 = const()[name = tensor("op_59704_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5073_cast_fp16 = mul(x = var_59703_cast_fp16, y = var_59704_to_fp16)[name = tensor("aw_5073_cast_fp16")]; + tensor var_59707_equation_0 = const()[name = tensor("op_59707_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_59707_cast_fp16 = einsum(equation = var_59707_equation_0, values = (var_59629_cast_fp16, var_59586_cast_fp16))[name = tensor("op_59707_cast_fp16")]; + tensor var_59708_to_fp16 = const()[name = tensor("op_59708_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5075_cast_fp16 = mul(x = var_59707_cast_fp16, y = var_59708_to_fp16)[name = tensor("aw_5075_cast_fp16")]; + tensor var_59711_equation_0 = const()[name = tensor("op_59711_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_59711_cast_fp16 = einsum(equation = var_59711_equation_0, values = (var_59633_cast_fp16, var_59590_cast_fp16))[name = tensor("op_59711_cast_fp16")]; + tensor var_59712_to_fp16 = const()[name = tensor("op_59712_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5077_cast_fp16 = mul(x = var_59711_cast_fp16, y = var_59712_to_fp16)[name = tensor("aw_5077_cast_fp16")]; + tensor var_59715_equation_0 = const()[name = tensor("op_59715_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_59715_cast_fp16 = einsum(equation = var_59715_equation_0, values = (var_59637_cast_fp16, var_59594_cast_fp16))[name = tensor("op_59715_cast_fp16")]; + tensor var_59716_to_fp16 = const()[name = tensor("op_59716_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5079_cast_fp16 = mul(x = var_59715_cast_fp16, y = var_59716_to_fp16)[name = tensor("aw_5079_cast_fp16")]; + tensor var_59718_cast_fp16 = softmax(axis = var_58034, x = aw_5061_cast_fp16)[name = tensor("op_59718_cast_fp16")]; + tensor var_59719_cast_fp16 = softmax(axis = var_58034, x = aw_5063_cast_fp16)[name = tensor("op_59719_cast_fp16")]; + tensor var_59720_cast_fp16 = softmax(axis = var_58034, x = aw_5065_cast_fp16)[name = tensor("op_59720_cast_fp16")]; + tensor var_59721_cast_fp16 = softmax(axis = var_58034, x = aw_5067_cast_fp16)[name = tensor("op_59721_cast_fp16")]; + tensor var_59722_cast_fp16 = softmax(axis = var_58034, x = aw_5069_cast_fp16)[name = tensor("op_59722_cast_fp16")]; + tensor var_59723_cast_fp16 = softmax(axis = var_58034, x = aw_5071_cast_fp16)[name = tensor("op_59723_cast_fp16")]; + tensor var_59724_cast_fp16 = softmax(axis = var_58034, x = aw_5073_cast_fp16)[name = tensor("op_59724_cast_fp16")]; + tensor var_59725_cast_fp16 = softmax(axis = var_58034, x = aw_5075_cast_fp16)[name = tensor("op_59725_cast_fp16")]; + tensor var_59726_cast_fp16 = softmax(axis = var_58034, x = aw_5077_cast_fp16)[name = tensor("op_59726_cast_fp16")]; + tensor var_59727_cast_fp16 = softmax(axis = var_58034, x = aw_5079_cast_fp16)[name = tensor("op_59727_cast_fp16")]; + tensor var_59729_equation_0 = const()[name = tensor("op_59729_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_59729_cast_fp16 = einsum(equation = var_59729_equation_0, values = (var_59639_cast_fp16, var_59718_cast_fp16))[name = tensor("op_59729_cast_fp16")]; + tensor var_59731_equation_0 = const()[name = tensor("op_59731_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_59731_cast_fp16 = einsum(equation = var_59731_equation_0, values = (var_59643_cast_fp16, var_59719_cast_fp16))[name = tensor("op_59731_cast_fp16")]; + tensor var_59733_equation_0 = const()[name = tensor("op_59733_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_59733_cast_fp16 = einsum(equation = var_59733_equation_0, values = (var_59647_cast_fp16, var_59720_cast_fp16))[name = tensor("op_59733_cast_fp16")]; + tensor var_59735_equation_0 = const()[name = tensor("op_59735_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_59735_cast_fp16 = einsum(equation = var_59735_equation_0, values = (var_59651_cast_fp16, var_59721_cast_fp16))[name = tensor("op_59735_cast_fp16")]; + tensor var_59737_equation_0 = const()[name = tensor("op_59737_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_59737_cast_fp16 = einsum(equation = var_59737_equation_0, values = (var_59655_cast_fp16, var_59722_cast_fp16))[name = tensor("op_59737_cast_fp16")]; + tensor var_59739_equation_0 = const()[name = tensor("op_59739_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_59739_cast_fp16 = einsum(equation = var_59739_equation_0, values = (var_59659_cast_fp16, var_59723_cast_fp16))[name = tensor("op_59739_cast_fp16")]; + tensor var_59741_equation_0 = const()[name = tensor("op_59741_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_59741_cast_fp16 = einsum(equation = var_59741_equation_0, values = (var_59663_cast_fp16, var_59724_cast_fp16))[name = tensor("op_59741_cast_fp16")]; + tensor var_59743_equation_0 = const()[name = tensor("op_59743_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_59743_cast_fp16 = einsum(equation = var_59743_equation_0, values = (var_59667_cast_fp16, var_59725_cast_fp16))[name = tensor("op_59743_cast_fp16")]; + tensor var_59745_equation_0 = const()[name = tensor("op_59745_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_59745_cast_fp16 = einsum(equation = var_59745_equation_0, values = (var_59671_cast_fp16, var_59726_cast_fp16))[name = tensor("op_59745_cast_fp16")]; + tensor var_59747_equation_0 = const()[name = tensor("op_59747_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_59747_cast_fp16 = einsum(equation = var_59747_equation_0, values = (var_59675_cast_fp16, var_59727_cast_fp16))[name = tensor("op_59747_cast_fp16")]; + tensor input_773_interleave_0 = const()[name = tensor("input_773_interleave_0"), val = tensor(false)]; + tensor input_773_cast_fp16 = concat(axis = var_58034, interleave = input_773_interleave_0, values = (var_59729_cast_fp16, var_59731_cast_fp16, var_59733_cast_fp16, var_59735_cast_fp16, var_59737_cast_fp16, var_59739_cast_fp16, var_59741_cast_fp16, var_59743_cast_fp16, var_59745_cast_fp16, var_59747_cast_fp16))[name = tensor("input_773_cast_fp16")]; + tensor var_59753 = const()[name = tensor("op_59753"), val = tensor([1, 1])]; + tensor var_59755 = const()[name = tensor("op_59755"), val = tensor([1, 1])]; + tensor var_59757_pad_type_0 = const()[name = tensor("op_59757_pad_type_0"), val = tensor("custom")]; + tensor var_59757_pad_0 = const()[name = tensor("op_59757_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5010842368)))]; + tensor up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5011661632)))]; + tensor var_59757_cast_fp16 = conv(bias = up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_59755, groups = var_58034, pad = var_59757_pad_0, pad_type = var_59757_pad_type_0, strides = var_59753, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16, x = input_773_cast_fp16)[name = tensor("op_59757_cast_fp16")]; + tensor inputs_401_cast_fp16 = add(x = var_59757_cast_fp16, y = inputs_399_cast_fp16)[name = tensor("inputs_401_cast_fp16")]; + tensor input_775_axes_0 = const()[name = tensor("input_775_axes_0"), val = tensor([1])]; + tensor input_775_gamma_0_to_fp16 = const()[name = tensor("input_775_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5011662976)))]; + tensor input_775_beta_0_to_fp16 = const()[name = tensor("input_775_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5011664320)))]; + tensor var_59767_to_fp16 = const()[name = tensor("op_59767_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_775_cast_fp16 = layer_norm(axes = input_775_axes_0, beta = input_775_beta_0_to_fp16, epsilon = var_59767_to_fp16, gamma = input_775_gamma_0_to_fp16, x = inputs_401_cast_fp16)[name = tensor("input_775_cast_fp16")]; + tensor var_59783 = const()[name = tensor("op_59783"), val = tensor([1, 1])]; + tensor var_59785 = const()[name = tensor("op_59785"), val = tensor([1, 1])]; + tensor var_59787_pad_type_0 = const()[name = tensor("op_59787_pad_type_0"), val = tensor("custom")]; + tensor var_59787_pad_0 = const()[name = tensor("op_59787_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5011665664)))]; + tensor up_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5018219328)))]; + tensor var_59787_cast_fp16 = conv(bias = up_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_59785, groups = var_58034, pad = var_59787_pad_0, pad_type = var_59787_pad_type_0, strides = var_59783, weight = up_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16, x = input_775_cast_fp16)[name = tensor("op_59787_cast_fp16")]; + tensor var_59788_split_sizes_0 = const()[name = tensor("op_59788_split_sizes_0"), val = tensor([2560, 2560])]; + tensor var_59788_axis_0 = const()[name = tensor("op_59788_axis_0"), val = tensor(1)]; + tensor var_59788_cast_fp16_0, tensor var_59788_cast_fp16_1 = split(axis = var_59788_axis_0, split_sizes = var_59788_split_sizes_0, x = var_59787_cast_fp16)[name = tensor("op_59788_cast_fp16")]; + tensor var_59790_mode_0 = const()[name = tensor("op_59790_mode_0"), val = tensor("EXACT")]; + tensor var_59790_cast_fp16 = gelu(mode = var_59790_mode_0, x = var_59788_cast_fp16_1)[name = tensor("op_59790_cast_fp16")]; + tensor input_777_cast_fp16 = mul(x = var_59788_cast_fp16_0, y = var_59790_cast_fp16)[name = tensor("input_777_cast_fp16")]; + tensor var_59794 = const()[name = tensor("op_59794"), val = tensor([1, 1])]; + tensor var_59796 = const()[name = tensor("op_59796"), val = tensor([1, 1])]; + tensor var_59798_pad_type_0 = const()[name = tensor("op_59798_pad_type_0"), val = tensor("custom")]; + tensor var_59798_pad_0 = const()[name = tensor("op_59798_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5018229632)))]; + tensor up_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5021506496)))]; + tensor var_59798_cast_fp16 = conv(bias = up_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_59796, groups = var_58034, pad = var_59798_pad_0, pad_type = var_59798_pad_type_0, strides = var_59794, weight = up_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16, x = input_777_cast_fp16)[name = tensor("op_59798_cast_fp16")]; + tensor inputs_403_cast_fp16 = add(x = var_59798_cast_fp16, y = inputs_401_cast_fp16)[name = tensor("inputs_403_cast_fp16")]; + tensor hidden_states_541_axes_0 = const()[name = tensor("hidden_states_541_axes_0"), val = tensor([1])]; + tensor hidden_states_541_gamma_0_to_fp16 = const()[name = tensor("hidden_states_541_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5021507840)))]; + tensor hidden_states_541_beta_0_to_fp16 = const()[name = tensor("hidden_states_541_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5021509184)))]; + tensor var_59814_to_fp16 = const()[name = tensor("op_59814_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_541_cast_fp16 = layer_norm(axes = hidden_states_541_axes_0, beta = hidden_states_541_beta_0_to_fp16, epsilon = var_59814_to_fp16, gamma = hidden_states_541_gamma_0_to_fp16, x = inputs_403_cast_fp16)[name = tensor("hidden_states_541_cast_fp16")]; + tensor var_59829 = const()[name = tensor("op_59829"), val = tensor([1, 1])]; + tensor var_59831 = const()[name = tensor("op_59831"), val = tensor([1, 1])]; + tensor q_269_pad_type_0 = const()[name = tensor("q_269_pad_type_0"), val = tensor("custom")]; + tensor q_269_pad_0 = const()[name = tensor("q_269_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_1_transformer_blocks_1_attn1_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_transformer_blocks_1_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5021510528)))]; + tensor q_269_cast_fp16 = conv(dilations = var_59831, groups = var_58034, pad = q_269_pad_0, pad_type = q_269_pad_type_0, strides = var_59829, weight = up_blocks_1_attentions_1_transformer_blocks_1_attn1_to_q_weight_to_fp16, x = hidden_states_541_cast_fp16)[name = tensor("q_269_cast_fp16")]; + tensor var_59835 = const()[name = tensor("op_59835"), val = tensor([1, 1])]; + tensor var_59837 = const()[name = tensor("op_59837"), val = tensor([1, 1])]; + tensor k_537_pad_type_0 = const()[name = tensor("k_537_pad_type_0"), val = tensor("custom")]; + tensor k_537_pad_0 = const()[name = tensor("k_537_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_1_transformer_blocks_1_attn1_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_transformer_blocks_1_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5022329792)))]; + tensor k_537_cast_fp16 = conv(dilations = var_59837, groups = var_58034, pad = k_537_pad_0, pad_type = k_537_pad_type_0, strides = var_59835, weight = up_blocks_1_attentions_1_transformer_blocks_1_attn1_to_k_weight_to_fp16, x = hidden_states_541_cast_fp16)[name = tensor("k_537_cast_fp16")]; + tensor var_59841 = const()[name = tensor("op_59841"), val = tensor([1, 1])]; + tensor var_59843 = const()[name = tensor("op_59843"), val = tensor([1, 1])]; + tensor v_269_pad_type_0 = const()[name = tensor("v_269_pad_type_0"), val = tensor("custom")]; + tensor v_269_pad_0 = const()[name = tensor("v_269_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_1_transformer_blocks_1_attn1_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_transformer_blocks_1_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5023149056)))]; + tensor v_269_cast_fp16 = conv(dilations = var_59843, groups = var_58034, pad = v_269_pad_0, pad_type = v_269_pad_type_0, strides = var_59841, weight = up_blocks_1_attentions_1_transformer_blocks_1_attn1_to_v_weight_to_fp16, x = hidden_states_541_cast_fp16)[name = tensor("v_269_cast_fp16")]; + tensor var_59847_begin_0 = const()[name = tensor("op_59847_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_59847_end_0 = const()[name = tensor("op_59847_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_59847_end_mask_0 = const()[name = tensor("op_59847_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59847_cast_fp16 = slice_by_index(begin = var_59847_begin_0, end = var_59847_end_0, end_mask = var_59847_end_mask_0, x = q_269_cast_fp16)[name = tensor("op_59847_cast_fp16")]; + tensor var_59851_begin_0 = const()[name = tensor("op_59851_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_59851_end_0 = const()[name = tensor("op_59851_end_0"), val = tensor([2, 128, 1, 4096])]; + tensor var_59851_end_mask_0 = const()[name = tensor("op_59851_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59851_cast_fp16 = slice_by_index(begin = var_59851_begin_0, end = var_59851_end_0, end_mask = var_59851_end_mask_0, x = q_269_cast_fp16)[name = tensor("op_59851_cast_fp16")]; + tensor var_59855_begin_0 = const()[name = tensor("op_59855_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_59855_end_0 = const()[name = tensor("op_59855_end_0"), val = tensor([2, 192, 1, 4096])]; + tensor var_59855_end_mask_0 = const()[name = tensor("op_59855_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59855_cast_fp16 = slice_by_index(begin = var_59855_begin_0, end = var_59855_end_0, end_mask = var_59855_end_mask_0, x = q_269_cast_fp16)[name = tensor("op_59855_cast_fp16")]; + tensor var_59859_begin_0 = const()[name = tensor("op_59859_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_59859_end_0 = const()[name = tensor("op_59859_end_0"), val = tensor([2, 256, 1, 4096])]; + tensor var_59859_end_mask_0 = const()[name = tensor("op_59859_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59859_cast_fp16 = slice_by_index(begin = var_59859_begin_0, end = var_59859_end_0, end_mask = var_59859_end_mask_0, x = q_269_cast_fp16)[name = tensor("op_59859_cast_fp16")]; + tensor var_59863_begin_0 = const()[name = tensor("op_59863_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_59863_end_0 = const()[name = tensor("op_59863_end_0"), val = tensor([2, 320, 1, 4096])]; + tensor var_59863_end_mask_0 = const()[name = tensor("op_59863_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59863_cast_fp16 = slice_by_index(begin = var_59863_begin_0, end = var_59863_end_0, end_mask = var_59863_end_mask_0, x = q_269_cast_fp16)[name = tensor("op_59863_cast_fp16")]; + tensor var_59867_begin_0 = const()[name = tensor("op_59867_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_59867_end_0 = const()[name = tensor("op_59867_end_0"), val = tensor([2, 384, 1, 4096])]; + tensor var_59867_end_mask_0 = const()[name = tensor("op_59867_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59867_cast_fp16 = slice_by_index(begin = var_59867_begin_0, end = var_59867_end_0, end_mask = var_59867_end_mask_0, x = q_269_cast_fp16)[name = tensor("op_59867_cast_fp16")]; + tensor var_59871_begin_0 = const()[name = tensor("op_59871_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_59871_end_0 = const()[name = tensor("op_59871_end_0"), val = tensor([2, 448, 1, 4096])]; + tensor var_59871_end_mask_0 = const()[name = tensor("op_59871_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59871_cast_fp16 = slice_by_index(begin = var_59871_begin_0, end = var_59871_end_0, end_mask = var_59871_end_mask_0, x = q_269_cast_fp16)[name = tensor("op_59871_cast_fp16")]; + tensor var_59875_begin_0 = const()[name = tensor("op_59875_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_59875_end_0 = const()[name = tensor("op_59875_end_0"), val = tensor([2, 512, 1, 4096])]; + tensor var_59875_end_mask_0 = const()[name = tensor("op_59875_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59875_cast_fp16 = slice_by_index(begin = var_59875_begin_0, end = var_59875_end_0, end_mask = var_59875_end_mask_0, x = q_269_cast_fp16)[name = tensor("op_59875_cast_fp16")]; + tensor var_59879_begin_0 = const()[name = tensor("op_59879_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_59879_end_0 = const()[name = tensor("op_59879_end_0"), val = tensor([2, 576, 1, 4096])]; + tensor var_59879_end_mask_0 = const()[name = tensor("op_59879_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59879_cast_fp16 = slice_by_index(begin = var_59879_begin_0, end = var_59879_end_0, end_mask = var_59879_end_mask_0, x = q_269_cast_fp16)[name = tensor("op_59879_cast_fp16")]; + tensor var_59883_begin_0 = const()[name = tensor("op_59883_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_59883_end_0 = const()[name = tensor("op_59883_end_0"), val = tensor([2, 640, 1, 4096])]; + tensor var_59883_end_mask_0 = const()[name = tensor("op_59883_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59883_cast_fp16 = slice_by_index(begin = var_59883_begin_0, end = var_59883_end_0, end_mask = var_59883_end_mask_0, x = q_269_cast_fp16)[name = tensor("op_59883_cast_fp16")]; + tensor k_539_perm_0 = const()[name = tensor("k_539_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_59890_begin_0 = const()[name = tensor("op_59890_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_59890_end_0 = const()[name = tensor("op_59890_end_0"), val = tensor([2, 4096, 1, 64])]; + tensor var_59890_end_mask_0 = const()[name = tensor("op_59890_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_539_cast_fp16 = transpose(perm = k_539_perm_0, x = k_537_cast_fp16)[name = tensor("transpose_5")]; + tensor var_59890_cast_fp16 = slice_by_index(begin = var_59890_begin_0, end = var_59890_end_0, end_mask = var_59890_end_mask_0, x = k_539_cast_fp16)[name = tensor("op_59890_cast_fp16")]; + tensor var_59894_begin_0 = const()[name = tensor("op_59894_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_59894_end_0 = const()[name = tensor("op_59894_end_0"), val = tensor([2, 4096, 1, 128])]; + tensor var_59894_end_mask_0 = const()[name = tensor("op_59894_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_59894_cast_fp16 = slice_by_index(begin = var_59894_begin_0, end = var_59894_end_0, end_mask = var_59894_end_mask_0, x = k_539_cast_fp16)[name = tensor("op_59894_cast_fp16")]; + tensor var_59898_begin_0 = const()[name = tensor("op_59898_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_59898_end_0 = const()[name = tensor("op_59898_end_0"), val = tensor([2, 4096, 1, 192])]; + tensor var_59898_end_mask_0 = const()[name = tensor("op_59898_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_59898_cast_fp16 = slice_by_index(begin = var_59898_begin_0, end = var_59898_end_0, end_mask = var_59898_end_mask_0, x = k_539_cast_fp16)[name = tensor("op_59898_cast_fp16")]; + tensor var_59902_begin_0 = const()[name = tensor("op_59902_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_59902_end_0 = const()[name = tensor("op_59902_end_0"), val = tensor([2, 4096, 1, 256])]; + tensor var_59902_end_mask_0 = const()[name = tensor("op_59902_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_59902_cast_fp16 = slice_by_index(begin = var_59902_begin_0, end = var_59902_end_0, end_mask = var_59902_end_mask_0, x = k_539_cast_fp16)[name = tensor("op_59902_cast_fp16")]; + tensor var_59906_begin_0 = const()[name = tensor("op_59906_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_59906_end_0 = const()[name = tensor("op_59906_end_0"), val = tensor([2, 4096, 1, 320])]; + tensor var_59906_end_mask_0 = const()[name = tensor("op_59906_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_59906_cast_fp16 = slice_by_index(begin = var_59906_begin_0, end = var_59906_end_0, end_mask = var_59906_end_mask_0, x = k_539_cast_fp16)[name = tensor("op_59906_cast_fp16")]; + tensor var_59910_begin_0 = const()[name = tensor("op_59910_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_59910_end_0 = const()[name = tensor("op_59910_end_0"), val = tensor([2, 4096, 1, 384])]; + tensor var_59910_end_mask_0 = const()[name = tensor("op_59910_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_59910_cast_fp16 = slice_by_index(begin = var_59910_begin_0, end = var_59910_end_0, end_mask = var_59910_end_mask_0, x = k_539_cast_fp16)[name = tensor("op_59910_cast_fp16")]; + tensor var_59914_begin_0 = const()[name = tensor("op_59914_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_59914_end_0 = const()[name = tensor("op_59914_end_0"), val = tensor([2, 4096, 1, 448])]; + tensor var_59914_end_mask_0 = const()[name = tensor("op_59914_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_59914_cast_fp16 = slice_by_index(begin = var_59914_begin_0, end = var_59914_end_0, end_mask = var_59914_end_mask_0, x = k_539_cast_fp16)[name = tensor("op_59914_cast_fp16")]; + tensor var_59918_begin_0 = const()[name = tensor("op_59918_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_59918_end_0 = const()[name = tensor("op_59918_end_0"), val = tensor([2, 4096, 1, 512])]; + tensor var_59918_end_mask_0 = const()[name = tensor("op_59918_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_59918_cast_fp16 = slice_by_index(begin = var_59918_begin_0, end = var_59918_end_0, end_mask = var_59918_end_mask_0, x = k_539_cast_fp16)[name = tensor("op_59918_cast_fp16")]; + tensor var_59922_begin_0 = const()[name = tensor("op_59922_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_59922_end_0 = const()[name = tensor("op_59922_end_0"), val = tensor([2, 4096, 1, 576])]; + tensor var_59922_end_mask_0 = const()[name = tensor("op_59922_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_59922_cast_fp16 = slice_by_index(begin = var_59922_begin_0, end = var_59922_end_0, end_mask = var_59922_end_mask_0, x = k_539_cast_fp16)[name = tensor("op_59922_cast_fp16")]; + tensor var_59926_begin_0 = const()[name = tensor("op_59926_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_59926_end_0 = const()[name = tensor("op_59926_end_0"), val = tensor([2, 4096, 1, 640])]; + tensor var_59926_end_mask_0 = const()[name = tensor("op_59926_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_59926_cast_fp16 = slice_by_index(begin = var_59926_begin_0, end = var_59926_end_0, end_mask = var_59926_end_mask_0, x = k_539_cast_fp16)[name = tensor("op_59926_cast_fp16")]; + tensor var_59928_begin_0 = const()[name = tensor("op_59928_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_59928_end_0 = const()[name = tensor("op_59928_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_59928_end_mask_0 = const()[name = tensor("op_59928_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59928_cast_fp16 = slice_by_index(begin = var_59928_begin_0, end = var_59928_end_0, end_mask = var_59928_end_mask_0, x = v_269_cast_fp16)[name = tensor("op_59928_cast_fp16")]; + tensor var_59932_begin_0 = const()[name = tensor("op_59932_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_59932_end_0 = const()[name = tensor("op_59932_end_0"), val = tensor([2, 128, 1, 4096])]; + tensor var_59932_end_mask_0 = const()[name = tensor("op_59932_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59932_cast_fp16 = slice_by_index(begin = var_59932_begin_0, end = var_59932_end_0, end_mask = var_59932_end_mask_0, x = v_269_cast_fp16)[name = tensor("op_59932_cast_fp16")]; + tensor var_59936_begin_0 = const()[name = tensor("op_59936_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_59936_end_0 = const()[name = tensor("op_59936_end_0"), val = tensor([2, 192, 1, 4096])]; + tensor var_59936_end_mask_0 = const()[name = tensor("op_59936_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59936_cast_fp16 = slice_by_index(begin = var_59936_begin_0, end = var_59936_end_0, end_mask = var_59936_end_mask_0, x = v_269_cast_fp16)[name = tensor("op_59936_cast_fp16")]; + tensor var_59940_begin_0 = const()[name = tensor("op_59940_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_59940_end_0 = const()[name = tensor("op_59940_end_0"), val = tensor([2, 256, 1, 4096])]; + tensor var_59940_end_mask_0 = const()[name = tensor("op_59940_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59940_cast_fp16 = slice_by_index(begin = var_59940_begin_0, end = var_59940_end_0, end_mask = var_59940_end_mask_0, x = v_269_cast_fp16)[name = tensor("op_59940_cast_fp16")]; + tensor var_59944_begin_0 = const()[name = tensor("op_59944_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_59944_end_0 = const()[name = tensor("op_59944_end_0"), val = tensor([2, 320, 1, 4096])]; + tensor var_59944_end_mask_0 = const()[name = tensor("op_59944_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59944_cast_fp16 = slice_by_index(begin = var_59944_begin_0, end = var_59944_end_0, end_mask = var_59944_end_mask_0, x = v_269_cast_fp16)[name = tensor("op_59944_cast_fp16")]; + tensor var_59948_begin_0 = const()[name = tensor("op_59948_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_59948_end_0 = const()[name = tensor("op_59948_end_0"), val = tensor([2, 384, 1, 4096])]; + tensor var_59948_end_mask_0 = const()[name = tensor("op_59948_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59948_cast_fp16 = slice_by_index(begin = var_59948_begin_0, end = var_59948_end_0, end_mask = var_59948_end_mask_0, x = v_269_cast_fp16)[name = tensor("op_59948_cast_fp16")]; + tensor var_59952_begin_0 = const()[name = tensor("op_59952_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_59952_end_0 = const()[name = tensor("op_59952_end_0"), val = tensor([2, 448, 1, 4096])]; + tensor var_59952_end_mask_0 = const()[name = tensor("op_59952_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59952_cast_fp16 = slice_by_index(begin = var_59952_begin_0, end = var_59952_end_0, end_mask = var_59952_end_mask_0, x = v_269_cast_fp16)[name = tensor("op_59952_cast_fp16")]; + tensor var_59956_begin_0 = const()[name = tensor("op_59956_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_59956_end_0 = const()[name = tensor("op_59956_end_0"), val = tensor([2, 512, 1, 4096])]; + tensor var_59956_end_mask_0 = const()[name = tensor("op_59956_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59956_cast_fp16 = slice_by_index(begin = var_59956_begin_0, end = var_59956_end_0, end_mask = var_59956_end_mask_0, x = v_269_cast_fp16)[name = tensor("op_59956_cast_fp16")]; + tensor var_59960_begin_0 = const()[name = tensor("op_59960_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_59960_end_0 = const()[name = tensor("op_59960_end_0"), val = tensor([2, 576, 1, 4096])]; + tensor var_59960_end_mask_0 = const()[name = tensor("op_59960_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59960_cast_fp16 = slice_by_index(begin = var_59960_begin_0, end = var_59960_end_0, end_mask = var_59960_end_mask_0, x = v_269_cast_fp16)[name = tensor("op_59960_cast_fp16")]; + tensor var_59964_begin_0 = const()[name = tensor("op_59964_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_59964_end_0 = const()[name = tensor("op_59964_end_0"), val = tensor([2, 640, 1, 4096])]; + tensor var_59964_end_mask_0 = const()[name = tensor("op_59964_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_59964_cast_fp16 = slice_by_index(begin = var_59964_begin_0, end = var_59964_end_0, end_mask = var_59964_end_mask_0, x = v_269_cast_fp16)[name = tensor("op_59964_cast_fp16")]; + tensor var_59968_equation_0 = const()[name = tensor("op_59968_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_59968_cast_fp16 = einsum(equation = var_59968_equation_0, values = (var_59890_cast_fp16, var_59847_cast_fp16))[name = tensor("op_59968_cast_fp16")]; + tensor var_59969_to_fp16 = const()[name = tensor("op_59969_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5081_cast_fp16 = mul(x = var_59968_cast_fp16, y = var_59969_to_fp16)[name = tensor("aw_5081_cast_fp16")]; + tensor var_59972_equation_0 = const()[name = tensor("op_59972_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_59972_cast_fp16 = einsum(equation = var_59972_equation_0, values = (var_59894_cast_fp16, var_59851_cast_fp16))[name = tensor("op_59972_cast_fp16")]; + tensor var_59973_to_fp16 = const()[name = tensor("op_59973_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5083_cast_fp16 = mul(x = var_59972_cast_fp16, y = var_59973_to_fp16)[name = tensor("aw_5083_cast_fp16")]; + tensor var_59976_equation_0 = const()[name = tensor("op_59976_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_59976_cast_fp16 = einsum(equation = var_59976_equation_0, values = (var_59898_cast_fp16, var_59855_cast_fp16))[name = tensor("op_59976_cast_fp16")]; + tensor var_59977_to_fp16 = const()[name = tensor("op_59977_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5085_cast_fp16 = mul(x = var_59976_cast_fp16, y = var_59977_to_fp16)[name = tensor("aw_5085_cast_fp16")]; + tensor var_59980_equation_0 = const()[name = tensor("op_59980_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_59980_cast_fp16 = einsum(equation = var_59980_equation_0, values = (var_59902_cast_fp16, var_59859_cast_fp16))[name = tensor("op_59980_cast_fp16")]; + tensor var_59981_to_fp16 = const()[name = tensor("op_59981_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5087_cast_fp16 = mul(x = var_59980_cast_fp16, y = var_59981_to_fp16)[name = tensor("aw_5087_cast_fp16")]; + tensor var_59984_equation_0 = const()[name = tensor("op_59984_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_59984_cast_fp16 = einsum(equation = var_59984_equation_0, values = (var_59906_cast_fp16, var_59863_cast_fp16))[name = tensor("op_59984_cast_fp16")]; + tensor var_59985_to_fp16 = const()[name = tensor("op_59985_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5089_cast_fp16 = mul(x = var_59984_cast_fp16, y = var_59985_to_fp16)[name = tensor("aw_5089_cast_fp16")]; + tensor var_59988_equation_0 = const()[name = tensor("op_59988_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_59988_cast_fp16 = einsum(equation = var_59988_equation_0, values = (var_59910_cast_fp16, var_59867_cast_fp16))[name = tensor("op_59988_cast_fp16")]; + tensor var_59989_to_fp16 = const()[name = tensor("op_59989_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5091_cast_fp16 = mul(x = var_59988_cast_fp16, y = var_59989_to_fp16)[name = tensor("aw_5091_cast_fp16")]; + tensor var_59992_equation_0 = const()[name = tensor("op_59992_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_59992_cast_fp16 = einsum(equation = var_59992_equation_0, values = (var_59914_cast_fp16, var_59871_cast_fp16))[name = tensor("op_59992_cast_fp16")]; + tensor var_59993_to_fp16 = const()[name = tensor("op_59993_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5093_cast_fp16 = mul(x = var_59992_cast_fp16, y = var_59993_to_fp16)[name = tensor("aw_5093_cast_fp16")]; + tensor var_59996_equation_0 = const()[name = tensor("op_59996_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_59996_cast_fp16 = einsum(equation = var_59996_equation_0, values = (var_59918_cast_fp16, var_59875_cast_fp16))[name = tensor("op_59996_cast_fp16")]; + tensor var_59997_to_fp16 = const()[name = tensor("op_59997_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5095_cast_fp16 = mul(x = var_59996_cast_fp16, y = var_59997_to_fp16)[name = tensor("aw_5095_cast_fp16")]; + tensor var_60000_equation_0 = const()[name = tensor("op_60000_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_60000_cast_fp16 = einsum(equation = var_60000_equation_0, values = (var_59922_cast_fp16, var_59879_cast_fp16))[name = tensor("op_60000_cast_fp16")]; + tensor var_60001_to_fp16 = const()[name = tensor("op_60001_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5097_cast_fp16 = mul(x = var_60000_cast_fp16, y = var_60001_to_fp16)[name = tensor("aw_5097_cast_fp16")]; + tensor var_60004_equation_0 = const()[name = tensor("op_60004_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_60004_cast_fp16 = einsum(equation = var_60004_equation_0, values = (var_59926_cast_fp16, var_59883_cast_fp16))[name = tensor("op_60004_cast_fp16")]; + tensor var_60005_to_fp16 = const()[name = tensor("op_60005_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5099_cast_fp16 = mul(x = var_60004_cast_fp16, y = var_60005_to_fp16)[name = tensor("aw_5099_cast_fp16")]; + tensor var_60007_cast_fp16 = softmax(axis = var_58034, x = aw_5081_cast_fp16)[name = tensor("op_60007_cast_fp16")]; + tensor var_60008_cast_fp16 = softmax(axis = var_58034, x = aw_5083_cast_fp16)[name = tensor("op_60008_cast_fp16")]; + tensor var_60009_cast_fp16 = softmax(axis = var_58034, x = aw_5085_cast_fp16)[name = tensor("op_60009_cast_fp16")]; + tensor var_60010_cast_fp16 = softmax(axis = var_58034, x = aw_5087_cast_fp16)[name = tensor("op_60010_cast_fp16")]; + tensor var_60011_cast_fp16 = softmax(axis = var_58034, x = aw_5089_cast_fp16)[name = tensor("op_60011_cast_fp16")]; + tensor var_60012_cast_fp16 = softmax(axis = var_58034, x = aw_5091_cast_fp16)[name = tensor("op_60012_cast_fp16")]; + tensor var_60013_cast_fp16 = softmax(axis = var_58034, x = aw_5093_cast_fp16)[name = tensor("op_60013_cast_fp16")]; + tensor var_60014_cast_fp16 = softmax(axis = var_58034, x = aw_5095_cast_fp16)[name = tensor("op_60014_cast_fp16")]; + tensor var_60015_cast_fp16 = softmax(axis = var_58034, x = aw_5097_cast_fp16)[name = tensor("op_60015_cast_fp16")]; + tensor var_60016_cast_fp16 = softmax(axis = var_58034, x = aw_5099_cast_fp16)[name = tensor("op_60016_cast_fp16")]; + tensor var_60018_equation_0 = const()[name = tensor("op_60018_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_60018_cast_fp16 = einsum(equation = var_60018_equation_0, values = (var_59928_cast_fp16, var_60007_cast_fp16))[name = tensor("op_60018_cast_fp16")]; + tensor var_60020_equation_0 = const()[name = tensor("op_60020_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_60020_cast_fp16 = einsum(equation = var_60020_equation_0, values = (var_59932_cast_fp16, var_60008_cast_fp16))[name = tensor("op_60020_cast_fp16")]; + tensor var_60022_equation_0 = const()[name = tensor("op_60022_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_60022_cast_fp16 = einsum(equation = var_60022_equation_0, values = (var_59936_cast_fp16, var_60009_cast_fp16))[name = tensor("op_60022_cast_fp16")]; + tensor var_60024_equation_0 = const()[name = tensor("op_60024_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_60024_cast_fp16 = einsum(equation = var_60024_equation_0, values = (var_59940_cast_fp16, var_60010_cast_fp16))[name = tensor("op_60024_cast_fp16")]; + tensor var_60026_equation_0 = const()[name = tensor("op_60026_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_60026_cast_fp16 = einsum(equation = var_60026_equation_0, values = (var_59944_cast_fp16, var_60011_cast_fp16))[name = tensor("op_60026_cast_fp16")]; + tensor var_60028_equation_0 = const()[name = tensor("op_60028_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_60028_cast_fp16 = einsum(equation = var_60028_equation_0, values = (var_59948_cast_fp16, var_60012_cast_fp16))[name = tensor("op_60028_cast_fp16")]; + tensor var_60030_equation_0 = const()[name = tensor("op_60030_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_60030_cast_fp16 = einsum(equation = var_60030_equation_0, values = (var_59952_cast_fp16, var_60013_cast_fp16))[name = tensor("op_60030_cast_fp16")]; + tensor var_60032_equation_0 = const()[name = tensor("op_60032_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_60032_cast_fp16 = einsum(equation = var_60032_equation_0, values = (var_59956_cast_fp16, var_60014_cast_fp16))[name = tensor("op_60032_cast_fp16")]; + tensor var_60034_equation_0 = const()[name = tensor("op_60034_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_60034_cast_fp16 = einsum(equation = var_60034_equation_0, values = (var_59960_cast_fp16, var_60015_cast_fp16))[name = tensor("op_60034_cast_fp16")]; + tensor var_60036_equation_0 = const()[name = tensor("op_60036_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_60036_cast_fp16 = einsum(equation = var_60036_equation_0, values = (var_59964_cast_fp16, var_60016_cast_fp16))[name = tensor("op_60036_cast_fp16")]; + tensor input_779_interleave_0 = const()[name = tensor("input_779_interleave_0"), val = tensor(false)]; + tensor input_779_cast_fp16 = concat(axis = var_58034, interleave = input_779_interleave_0, values = (var_60018_cast_fp16, var_60020_cast_fp16, var_60022_cast_fp16, var_60024_cast_fp16, var_60026_cast_fp16, var_60028_cast_fp16, var_60030_cast_fp16, var_60032_cast_fp16, var_60034_cast_fp16, var_60036_cast_fp16))[name = tensor("input_779_cast_fp16")]; + tensor var_60042 = const()[name = tensor("op_60042"), val = tensor([1, 1])]; + tensor var_60044 = const()[name = tensor("op_60044"), val = tensor([1, 1])]; + tensor var_60046_pad_type_0 = const()[name = tensor("op_60046_pad_type_0"), val = tensor("custom")]; + tensor var_60046_pad_0 = const()[name = tensor("op_60046_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_1_transformer_blocks_1_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_transformer_blocks_1_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5023968320)))]; + tensor up_blocks_1_attentions_1_transformer_blocks_1_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_transformer_blocks_1_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5024787584)))]; + tensor var_60046_cast_fp16 = conv(bias = up_blocks_1_attentions_1_transformer_blocks_1_attn1_to_out_0_bias_to_fp16, dilations = var_60044, groups = var_58034, pad = var_60046_pad_0, pad_type = var_60046_pad_type_0, strides = var_60042, weight = up_blocks_1_attentions_1_transformer_blocks_1_attn1_to_out_0_weight_to_fp16, x = input_779_cast_fp16)[name = tensor("op_60046_cast_fp16")]; + tensor inputs_405_cast_fp16 = add(x = var_60046_cast_fp16, y = inputs_403_cast_fp16)[name = tensor("inputs_405_cast_fp16")]; + tensor hidden_states_543_axes_0 = const()[name = tensor("hidden_states_543_axes_0"), val = tensor([1])]; + tensor hidden_states_543_gamma_0_to_fp16 = const()[name = tensor("hidden_states_543_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5024788928)))]; + tensor hidden_states_543_beta_0_to_fp16 = const()[name = tensor("hidden_states_543_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5024790272)))]; + tensor var_60056_to_fp16 = const()[name = tensor("op_60056_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_543_cast_fp16 = layer_norm(axes = hidden_states_543_axes_0, beta = hidden_states_543_beta_0_to_fp16, epsilon = var_60056_to_fp16, gamma = hidden_states_543_gamma_0_to_fp16, x = inputs_405_cast_fp16)[name = tensor("hidden_states_543_cast_fp16")]; + tensor var_60071 = const()[name = tensor("op_60071"), val = tensor([1, 1])]; + tensor var_60073 = const()[name = tensor("op_60073"), val = tensor([1, 1])]; + tensor q_271_pad_type_0 = const()[name = tensor("q_271_pad_type_0"), val = tensor("custom")]; + tensor q_271_pad_0 = const()[name = tensor("q_271_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_1_transformer_blocks_1_attn2_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_transformer_blocks_1_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5024791616)))]; + tensor q_271_cast_fp16 = conv(dilations = var_60073, groups = var_58034, pad = q_271_pad_0, pad_type = q_271_pad_type_0, strides = var_60071, weight = up_blocks_1_attentions_1_transformer_blocks_1_attn2_to_q_weight_to_fp16, x = hidden_states_543_cast_fp16)[name = tensor("q_271_cast_fp16")]; + tensor var_60077 = const()[name = tensor("op_60077"), val = tensor([1, 1])]; + tensor var_60079 = const()[name = tensor("op_60079"), val = tensor([1, 1])]; + tensor k_541_pad_type_0 = const()[name = tensor("k_541_pad_type_0"), val = tensor("custom")]; + tensor k_541_pad_0 = const()[name = tensor("k_541_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_1_transformer_blocks_1_attn2_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_transformer_blocks_1_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5025610880)))]; + tensor k_541_cast_fp16 = conv(dilations = var_60079, groups = var_58034, pad = k_541_pad_0, pad_type = k_541_pad_type_0, strides = var_60077, weight = up_blocks_1_attentions_1_transformer_blocks_1_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_541_cast_fp16")]; + tensor var_60083 = const()[name = tensor("op_60083"), val = tensor([1, 1])]; + tensor var_60085 = const()[name = tensor("op_60085"), val = tensor([1, 1])]; + tensor v_271_pad_type_0 = const()[name = tensor("v_271_pad_type_0"), val = tensor("custom")]; + tensor v_271_pad_0 = const()[name = tensor("v_271_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_1_transformer_blocks_1_attn2_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_transformer_blocks_1_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5028232384)))]; + tensor v_271_cast_fp16 = conv(dilations = var_60085, groups = var_58034, pad = v_271_pad_0, pad_type = v_271_pad_type_0, strides = var_60083, weight = up_blocks_1_attentions_1_transformer_blocks_1_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_271_cast_fp16")]; + tensor var_60089_begin_0 = const()[name = tensor("op_60089_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_60089_end_0 = const()[name = tensor("op_60089_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_60089_end_mask_0 = const()[name = tensor("op_60089_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_60089_cast_fp16 = slice_by_index(begin = var_60089_begin_0, end = var_60089_end_0, end_mask = var_60089_end_mask_0, x = q_271_cast_fp16)[name = tensor("op_60089_cast_fp16")]; + tensor var_60093_begin_0 = const()[name = tensor("op_60093_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_60093_end_0 = const()[name = tensor("op_60093_end_0"), val = tensor([2, 128, 1, 4096])]; + tensor var_60093_end_mask_0 = const()[name = tensor("op_60093_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_60093_cast_fp16 = slice_by_index(begin = var_60093_begin_0, end = var_60093_end_0, end_mask = var_60093_end_mask_0, x = q_271_cast_fp16)[name = tensor("op_60093_cast_fp16")]; + tensor var_60097_begin_0 = const()[name = tensor("op_60097_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_60097_end_0 = const()[name = tensor("op_60097_end_0"), val = tensor([2, 192, 1, 4096])]; + tensor var_60097_end_mask_0 = const()[name = tensor("op_60097_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_60097_cast_fp16 = slice_by_index(begin = var_60097_begin_0, end = var_60097_end_0, end_mask = var_60097_end_mask_0, x = q_271_cast_fp16)[name = tensor("op_60097_cast_fp16")]; + tensor var_60101_begin_0 = const()[name = tensor("op_60101_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_60101_end_0 = const()[name = tensor("op_60101_end_0"), val = tensor([2, 256, 1, 4096])]; + tensor var_60101_end_mask_0 = const()[name = tensor("op_60101_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_60101_cast_fp16 = slice_by_index(begin = var_60101_begin_0, end = var_60101_end_0, end_mask = var_60101_end_mask_0, x = q_271_cast_fp16)[name = tensor("op_60101_cast_fp16")]; + tensor var_60105_begin_0 = const()[name = tensor("op_60105_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_60105_end_0 = const()[name = tensor("op_60105_end_0"), val = tensor([2, 320, 1, 4096])]; + tensor var_60105_end_mask_0 = const()[name = tensor("op_60105_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_60105_cast_fp16 = slice_by_index(begin = var_60105_begin_0, end = var_60105_end_0, end_mask = var_60105_end_mask_0, x = q_271_cast_fp16)[name = tensor("op_60105_cast_fp16")]; + tensor var_60109_begin_0 = const()[name = tensor("op_60109_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_60109_end_0 = const()[name = tensor("op_60109_end_0"), val = tensor([2, 384, 1, 4096])]; + tensor var_60109_end_mask_0 = const()[name = tensor("op_60109_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_60109_cast_fp16 = slice_by_index(begin = var_60109_begin_0, end = var_60109_end_0, end_mask = var_60109_end_mask_0, x = q_271_cast_fp16)[name = tensor("op_60109_cast_fp16")]; + tensor var_60113_begin_0 = const()[name = tensor("op_60113_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_60113_end_0 = const()[name = tensor("op_60113_end_0"), val = tensor([2, 448, 1, 4096])]; + tensor var_60113_end_mask_0 = const()[name = tensor("op_60113_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_60113_cast_fp16 = slice_by_index(begin = var_60113_begin_0, end = var_60113_end_0, end_mask = var_60113_end_mask_0, x = q_271_cast_fp16)[name = tensor("op_60113_cast_fp16")]; + tensor var_60117_begin_0 = const()[name = tensor("op_60117_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_60117_end_0 = const()[name = tensor("op_60117_end_0"), val = tensor([2, 512, 1, 4096])]; + tensor var_60117_end_mask_0 = const()[name = tensor("op_60117_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_60117_cast_fp16 = slice_by_index(begin = var_60117_begin_0, end = var_60117_end_0, end_mask = var_60117_end_mask_0, x = q_271_cast_fp16)[name = tensor("op_60117_cast_fp16")]; + tensor var_60121_begin_0 = const()[name = tensor("op_60121_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_60121_end_0 = const()[name = tensor("op_60121_end_0"), val = tensor([2, 576, 1, 4096])]; + tensor var_60121_end_mask_0 = const()[name = tensor("op_60121_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_60121_cast_fp16 = slice_by_index(begin = var_60121_begin_0, end = var_60121_end_0, end_mask = var_60121_end_mask_0, x = q_271_cast_fp16)[name = tensor("op_60121_cast_fp16")]; + tensor var_60125_begin_0 = const()[name = tensor("op_60125_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_60125_end_0 = const()[name = tensor("op_60125_end_0"), val = tensor([2, 640, 1, 4096])]; + tensor var_60125_end_mask_0 = const()[name = tensor("op_60125_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_60125_cast_fp16 = slice_by_index(begin = var_60125_begin_0, end = var_60125_end_0, end_mask = var_60125_end_mask_0, x = q_271_cast_fp16)[name = tensor("op_60125_cast_fp16")]; + tensor k_543_perm_0 = const()[name = tensor("k_543_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_60132_begin_0 = const()[name = tensor("op_60132_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_60132_end_0 = const()[name = tensor("op_60132_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_60132_end_mask_0 = const()[name = tensor("op_60132_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_543_cast_fp16 = transpose(perm = k_543_perm_0, x = k_541_cast_fp16)[name = tensor("transpose_4")]; + tensor var_60132_cast_fp16 = slice_by_index(begin = var_60132_begin_0, end = var_60132_end_0, end_mask = var_60132_end_mask_0, x = k_543_cast_fp16)[name = tensor("op_60132_cast_fp16")]; + tensor var_60136_begin_0 = const()[name = tensor("op_60136_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_60136_end_0 = const()[name = tensor("op_60136_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_60136_end_mask_0 = const()[name = tensor("op_60136_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_60136_cast_fp16 = slice_by_index(begin = var_60136_begin_0, end = var_60136_end_0, end_mask = var_60136_end_mask_0, x = k_543_cast_fp16)[name = tensor("op_60136_cast_fp16")]; + tensor var_60140_begin_0 = const()[name = tensor("op_60140_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_60140_end_0 = const()[name = tensor("op_60140_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_60140_end_mask_0 = const()[name = tensor("op_60140_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_60140_cast_fp16 = slice_by_index(begin = var_60140_begin_0, end = var_60140_end_0, end_mask = var_60140_end_mask_0, x = k_543_cast_fp16)[name = tensor("op_60140_cast_fp16")]; + tensor var_60144_begin_0 = const()[name = tensor("op_60144_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_60144_end_0 = const()[name = tensor("op_60144_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_60144_end_mask_0 = const()[name = tensor("op_60144_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_60144_cast_fp16 = slice_by_index(begin = var_60144_begin_0, end = var_60144_end_0, end_mask = var_60144_end_mask_0, x = k_543_cast_fp16)[name = tensor("op_60144_cast_fp16")]; + tensor var_60148_begin_0 = const()[name = tensor("op_60148_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_60148_end_0 = const()[name = tensor("op_60148_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_60148_end_mask_0 = const()[name = tensor("op_60148_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_60148_cast_fp16 = slice_by_index(begin = var_60148_begin_0, end = var_60148_end_0, end_mask = var_60148_end_mask_0, x = k_543_cast_fp16)[name = tensor("op_60148_cast_fp16")]; + tensor var_60152_begin_0 = const()[name = tensor("op_60152_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_60152_end_0 = const()[name = tensor("op_60152_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_60152_end_mask_0 = const()[name = tensor("op_60152_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_60152_cast_fp16 = slice_by_index(begin = var_60152_begin_0, end = var_60152_end_0, end_mask = var_60152_end_mask_0, x = k_543_cast_fp16)[name = tensor("op_60152_cast_fp16")]; + tensor var_60156_begin_0 = const()[name = tensor("op_60156_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_60156_end_0 = const()[name = tensor("op_60156_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_60156_end_mask_0 = const()[name = tensor("op_60156_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_60156_cast_fp16 = slice_by_index(begin = var_60156_begin_0, end = var_60156_end_0, end_mask = var_60156_end_mask_0, x = k_543_cast_fp16)[name = tensor("op_60156_cast_fp16")]; + tensor var_60160_begin_0 = const()[name = tensor("op_60160_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_60160_end_0 = const()[name = tensor("op_60160_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_60160_end_mask_0 = const()[name = tensor("op_60160_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_60160_cast_fp16 = slice_by_index(begin = var_60160_begin_0, end = var_60160_end_0, end_mask = var_60160_end_mask_0, x = k_543_cast_fp16)[name = tensor("op_60160_cast_fp16")]; + tensor var_60164_begin_0 = const()[name = tensor("op_60164_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_60164_end_0 = const()[name = tensor("op_60164_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_60164_end_mask_0 = const()[name = tensor("op_60164_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_60164_cast_fp16 = slice_by_index(begin = var_60164_begin_0, end = var_60164_end_0, end_mask = var_60164_end_mask_0, x = k_543_cast_fp16)[name = tensor("op_60164_cast_fp16")]; + tensor var_60168_begin_0 = const()[name = tensor("op_60168_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_60168_end_0 = const()[name = tensor("op_60168_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_60168_end_mask_0 = const()[name = tensor("op_60168_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_60168_cast_fp16 = slice_by_index(begin = var_60168_begin_0, end = var_60168_end_0, end_mask = var_60168_end_mask_0, x = k_543_cast_fp16)[name = tensor("op_60168_cast_fp16")]; + tensor var_60170_begin_0 = const()[name = tensor("op_60170_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_60170_end_0 = const()[name = tensor("op_60170_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_60170_end_mask_0 = const()[name = tensor("op_60170_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_60170_cast_fp16 = slice_by_index(begin = var_60170_begin_0, end = var_60170_end_0, end_mask = var_60170_end_mask_0, x = v_271_cast_fp16)[name = tensor("op_60170_cast_fp16")]; + tensor var_60174_begin_0 = const()[name = tensor("op_60174_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_60174_end_0 = const()[name = tensor("op_60174_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_60174_end_mask_0 = const()[name = tensor("op_60174_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_60174_cast_fp16 = slice_by_index(begin = var_60174_begin_0, end = var_60174_end_0, end_mask = var_60174_end_mask_0, x = v_271_cast_fp16)[name = tensor("op_60174_cast_fp16")]; + tensor var_60178_begin_0 = const()[name = tensor("op_60178_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_60178_end_0 = const()[name = tensor("op_60178_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_60178_end_mask_0 = const()[name = tensor("op_60178_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_60178_cast_fp16 = slice_by_index(begin = var_60178_begin_0, end = var_60178_end_0, end_mask = var_60178_end_mask_0, x = v_271_cast_fp16)[name = tensor("op_60178_cast_fp16")]; + tensor var_60182_begin_0 = const()[name = tensor("op_60182_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_60182_end_0 = const()[name = tensor("op_60182_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_60182_end_mask_0 = const()[name = tensor("op_60182_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_60182_cast_fp16 = slice_by_index(begin = var_60182_begin_0, end = var_60182_end_0, end_mask = var_60182_end_mask_0, x = v_271_cast_fp16)[name = tensor("op_60182_cast_fp16")]; + tensor var_60186_begin_0 = const()[name = tensor("op_60186_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_60186_end_0 = const()[name = tensor("op_60186_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_60186_end_mask_0 = const()[name = tensor("op_60186_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_60186_cast_fp16 = slice_by_index(begin = var_60186_begin_0, end = var_60186_end_0, end_mask = var_60186_end_mask_0, x = v_271_cast_fp16)[name = tensor("op_60186_cast_fp16")]; + tensor var_60190_begin_0 = const()[name = tensor("op_60190_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_60190_end_0 = const()[name = tensor("op_60190_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_60190_end_mask_0 = const()[name = tensor("op_60190_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_60190_cast_fp16 = slice_by_index(begin = var_60190_begin_0, end = var_60190_end_0, end_mask = var_60190_end_mask_0, x = v_271_cast_fp16)[name = tensor("op_60190_cast_fp16")]; + tensor var_60194_begin_0 = const()[name = tensor("op_60194_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_60194_end_0 = const()[name = tensor("op_60194_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_60194_end_mask_0 = const()[name = tensor("op_60194_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_60194_cast_fp16 = slice_by_index(begin = var_60194_begin_0, end = var_60194_end_0, end_mask = var_60194_end_mask_0, x = v_271_cast_fp16)[name = tensor("op_60194_cast_fp16")]; + tensor var_60198_begin_0 = const()[name = tensor("op_60198_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_60198_end_0 = const()[name = tensor("op_60198_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_60198_end_mask_0 = const()[name = tensor("op_60198_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_60198_cast_fp16 = slice_by_index(begin = var_60198_begin_0, end = var_60198_end_0, end_mask = var_60198_end_mask_0, x = v_271_cast_fp16)[name = tensor("op_60198_cast_fp16")]; + tensor var_60202_begin_0 = const()[name = tensor("op_60202_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_60202_end_0 = const()[name = tensor("op_60202_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_60202_end_mask_0 = const()[name = tensor("op_60202_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_60202_cast_fp16 = slice_by_index(begin = var_60202_begin_0, end = var_60202_end_0, end_mask = var_60202_end_mask_0, x = v_271_cast_fp16)[name = tensor("op_60202_cast_fp16")]; + tensor var_60206_begin_0 = const()[name = tensor("op_60206_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_60206_end_0 = const()[name = tensor("op_60206_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_60206_end_mask_0 = const()[name = tensor("op_60206_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_60206_cast_fp16 = slice_by_index(begin = var_60206_begin_0, end = var_60206_end_0, end_mask = var_60206_end_mask_0, x = v_271_cast_fp16)[name = tensor("op_60206_cast_fp16")]; + tensor var_60210_equation_0 = const()[name = tensor("op_60210_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_60210_cast_fp16 = einsum(equation = var_60210_equation_0, values = (var_60132_cast_fp16, var_60089_cast_fp16))[name = tensor("op_60210_cast_fp16")]; + tensor var_60211_to_fp16 = const()[name = tensor("op_60211_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5101_cast_fp16 = mul(x = var_60210_cast_fp16, y = var_60211_to_fp16)[name = tensor("aw_5101_cast_fp16")]; + tensor var_60214_equation_0 = const()[name = tensor("op_60214_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_60214_cast_fp16 = einsum(equation = var_60214_equation_0, values = (var_60136_cast_fp16, var_60093_cast_fp16))[name = tensor("op_60214_cast_fp16")]; + tensor var_60215_to_fp16 = const()[name = tensor("op_60215_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5103_cast_fp16 = mul(x = var_60214_cast_fp16, y = var_60215_to_fp16)[name = tensor("aw_5103_cast_fp16")]; + tensor var_60218_equation_0 = const()[name = tensor("op_60218_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_60218_cast_fp16 = einsum(equation = var_60218_equation_0, values = (var_60140_cast_fp16, var_60097_cast_fp16))[name = tensor("op_60218_cast_fp16")]; + tensor var_60219_to_fp16 = const()[name = tensor("op_60219_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5105_cast_fp16 = mul(x = var_60218_cast_fp16, y = var_60219_to_fp16)[name = tensor("aw_5105_cast_fp16")]; + tensor var_60222_equation_0 = const()[name = tensor("op_60222_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_60222_cast_fp16 = einsum(equation = var_60222_equation_0, values = (var_60144_cast_fp16, var_60101_cast_fp16))[name = tensor("op_60222_cast_fp16")]; + tensor var_60223_to_fp16 = const()[name = tensor("op_60223_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5107_cast_fp16 = mul(x = var_60222_cast_fp16, y = var_60223_to_fp16)[name = tensor("aw_5107_cast_fp16")]; + tensor var_60226_equation_0 = const()[name = tensor("op_60226_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_60226_cast_fp16 = einsum(equation = var_60226_equation_0, values = (var_60148_cast_fp16, var_60105_cast_fp16))[name = tensor("op_60226_cast_fp16")]; + tensor var_60227_to_fp16 = const()[name = tensor("op_60227_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5109_cast_fp16 = mul(x = var_60226_cast_fp16, y = var_60227_to_fp16)[name = tensor("aw_5109_cast_fp16")]; + tensor var_60230_equation_0 = const()[name = tensor("op_60230_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_60230_cast_fp16 = einsum(equation = var_60230_equation_0, values = (var_60152_cast_fp16, var_60109_cast_fp16))[name = tensor("op_60230_cast_fp16")]; + tensor var_60231_to_fp16 = const()[name = tensor("op_60231_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5111_cast_fp16 = mul(x = var_60230_cast_fp16, y = var_60231_to_fp16)[name = tensor("aw_5111_cast_fp16")]; + tensor var_60234_equation_0 = const()[name = tensor("op_60234_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_60234_cast_fp16 = einsum(equation = var_60234_equation_0, values = (var_60156_cast_fp16, var_60113_cast_fp16))[name = tensor("op_60234_cast_fp16")]; + tensor var_60235_to_fp16 = const()[name = tensor("op_60235_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5113_cast_fp16 = mul(x = var_60234_cast_fp16, y = var_60235_to_fp16)[name = tensor("aw_5113_cast_fp16")]; + tensor var_60238_equation_0 = const()[name = tensor("op_60238_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_60238_cast_fp16 = einsum(equation = var_60238_equation_0, values = (var_60160_cast_fp16, var_60117_cast_fp16))[name = tensor("op_60238_cast_fp16")]; + tensor var_60239_to_fp16 = const()[name = tensor("op_60239_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5115_cast_fp16 = mul(x = var_60238_cast_fp16, y = var_60239_to_fp16)[name = tensor("aw_5115_cast_fp16")]; + tensor var_60242_equation_0 = const()[name = tensor("op_60242_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_60242_cast_fp16 = einsum(equation = var_60242_equation_0, values = (var_60164_cast_fp16, var_60121_cast_fp16))[name = tensor("op_60242_cast_fp16")]; + tensor var_60243_to_fp16 = const()[name = tensor("op_60243_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5117_cast_fp16 = mul(x = var_60242_cast_fp16, y = var_60243_to_fp16)[name = tensor("aw_5117_cast_fp16")]; + tensor var_60246_equation_0 = const()[name = tensor("op_60246_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_60246_cast_fp16 = einsum(equation = var_60246_equation_0, values = (var_60168_cast_fp16, var_60125_cast_fp16))[name = tensor("op_60246_cast_fp16")]; + tensor var_60247_to_fp16 = const()[name = tensor("op_60247_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5119_cast_fp16 = mul(x = var_60246_cast_fp16, y = var_60247_to_fp16)[name = tensor("aw_5119_cast_fp16")]; + tensor var_60249_cast_fp16 = softmax(axis = var_58034, x = aw_5101_cast_fp16)[name = tensor("op_60249_cast_fp16")]; + tensor var_60250_cast_fp16 = softmax(axis = var_58034, x = aw_5103_cast_fp16)[name = tensor("op_60250_cast_fp16")]; + tensor var_60251_cast_fp16 = softmax(axis = var_58034, x = aw_5105_cast_fp16)[name = tensor("op_60251_cast_fp16")]; + tensor var_60252_cast_fp16 = softmax(axis = var_58034, x = aw_5107_cast_fp16)[name = tensor("op_60252_cast_fp16")]; + tensor var_60253_cast_fp16 = softmax(axis = var_58034, x = aw_5109_cast_fp16)[name = tensor("op_60253_cast_fp16")]; + tensor var_60254_cast_fp16 = softmax(axis = var_58034, x = aw_5111_cast_fp16)[name = tensor("op_60254_cast_fp16")]; + tensor var_60255_cast_fp16 = softmax(axis = var_58034, x = aw_5113_cast_fp16)[name = tensor("op_60255_cast_fp16")]; + tensor var_60256_cast_fp16 = softmax(axis = var_58034, x = aw_5115_cast_fp16)[name = tensor("op_60256_cast_fp16")]; + tensor var_60257_cast_fp16 = softmax(axis = var_58034, x = aw_5117_cast_fp16)[name = tensor("op_60257_cast_fp16")]; + tensor var_60258_cast_fp16 = softmax(axis = var_58034, x = aw_5119_cast_fp16)[name = tensor("op_60258_cast_fp16")]; + tensor var_60260_equation_0 = const()[name = tensor("op_60260_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_60260_cast_fp16 = einsum(equation = var_60260_equation_0, values = (var_60170_cast_fp16, var_60249_cast_fp16))[name = tensor("op_60260_cast_fp16")]; + tensor var_60262_equation_0 = const()[name = tensor("op_60262_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_60262_cast_fp16 = einsum(equation = var_60262_equation_0, values = (var_60174_cast_fp16, var_60250_cast_fp16))[name = tensor("op_60262_cast_fp16")]; + tensor var_60264_equation_0 = const()[name = tensor("op_60264_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_60264_cast_fp16 = einsum(equation = var_60264_equation_0, values = (var_60178_cast_fp16, var_60251_cast_fp16))[name = tensor("op_60264_cast_fp16")]; + tensor var_60266_equation_0 = const()[name = tensor("op_60266_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_60266_cast_fp16 = einsum(equation = var_60266_equation_0, values = (var_60182_cast_fp16, var_60252_cast_fp16))[name = tensor("op_60266_cast_fp16")]; + tensor var_60268_equation_0 = const()[name = tensor("op_60268_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_60268_cast_fp16 = einsum(equation = var_60268_equation_0, values = (var_60186_cast_fp16, var_60253_cast_fp16))[name = tensor("op_60268_cast_fp16")]; + tensor var_60270_equation_0 = const()[name = tensor("op_60270_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_60270_cast_fp16 = einsum(equation = var_60270_equation_0, values = (var_60190_cast_fp16, var_60254_cast_fp16))[name = tensor("op_60270_cast_fp16")]; + tensor var_60272_equation_0 = const()[name = tensor("op_60272_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_60272_cast_fp16 = einsum(equation = var_60272_equation_0, values = (var_60194_cast_fp16, var_60255_cast_fp16))[name = tensor("op_60272_cast_fp16")]; + tensor var_60274_equation_0 = const()[name = tensor("op_60274_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_60274_cast_fp16 = einsum(equation = var_60274_equation_0, values = (var_60198_cast_fp16, var_60256_cast_fp16))[name = tensor("op_60274_cast_fp16")]; + tensor var_60276_equation_0 = const()[name = tensor("op_60276_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_60276_cast_fp16 = einsum(equation = var_60276_equation_0, values = (var_60202_cast_fp16, var_60257_cast_fp16))[name = tensor("op_60276_cast_fp16")]; + tensor var_60278_equation_0 = const()[name = tensor("op_60278_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_60278_cast_fp16 = einsum(equation = var_60278_equation_0, values = (var_60206_cast_fp16, var_60258_cast_fp16))[name = tensor("op_60278_cast_fp16")]; + tensor input_781_interleave_0 = const()[name = tensor("input_781_interleave_0"), val = tensor(false)]; + tensor input_781_cast_fp16 = concat(axis = var_58034, interleave = input_781_interleave_0, values = (var_60260_cast_fp16, var_60262_cast_fp16, var_60264_cast_fp16, var_60266_cast_fp16, var_60268_cast_fp16, var_60270_cast_fp16, var_60272_cast_fp16, var_60274_cast_fp16, var_60276_cast_fp16, var_60278_cast_fp16))[name = tensor("input_781_cast_fp16")]; + tensor var_60284 = const()[name = tensor("op_60284"), val = tensor([1, 1])]; + tensor var_60286 = const()[name = tensor("op_60286"), val = tensor([1, 1])]; + tensor var_60288_pad_type_0 = const()[name = tensor("op_60288_pad_type_0"), val = tensor("custom")]; + tensor var_60288_pad_0 = const()[name = tensor("op_60288_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_1_transformer_blocks_1_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_transformer_blocks_1_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5030853888)))]; + tensor up_blocks_1_attentions_1_transformer_blocks_1_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_transformer_blocks_1_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5031673152)))]; + tensor var_60288_cast_fp16 = conv(bias = up_blocks_1_attentions_1_transformer_blocks_1_attn2_to_out_0_bias_to_fp16, dilations = var_60286, groups = var_58034, pad = var_60288_pad_0, pad_type = var_60288_pad_type_0, strides = var_60284, weight = up_blocks_1_attentions_1_transformer_blocks_1_attn2_to_out_0_weight_to_fp16, x = input_781_cast_fp16)[name = tensor("op_60288_cast_fp16")]; + tensor inputs_407_cast_fp16 = add(x = var_60288_cast_fp16, y = inputs_405_cast_fp16)[name = tensor("inputs_407_cast_fp16")]; + tensor input_783_axes_0 = const()[name = tensor("input_783_axes_0"), val = tensor([1])]; + tensor input_783_gamma_0_to_fp16 = const()[name = tensor("input_783_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5031674496)))]; + tensor input_783_beta_0_to_fp16 = const()[name = tensor("input_783_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5031675840)))]; + tensor var_60298_to_fp16 = const()[name = tensor("op_60298_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_783_cast_fp16 = layer_norm(axes = input_783_axes_0, beta = input_783_beta_0_to_fp16, epsilon = var_60298_to_fp16, gamma = input_783_gamma_0_to_fp16, x = inputs_407_cast_fp16)[name = tensor("input_783_cast_fp16")]; + tensor var_60314 = const()[name = tensor("op_60314"), val = tensor([1, 1])]; + tensor var_60316 = const()[name = tensor("op_60316"), val = tensor([1, 1])]; + tensor var_60318_pad_type_0 = const()[name = tensor("op_60318_pad_type_0"), val = tensor("custom")]; + tensor var_60318_pad_0 = const()[name = tensor("op_60318_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_1_transformer_blocks_1_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_transformer_blocks_1_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5031677184)))]; + tensor up_blocks_1_attentions_1_transformer_blocks_1_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_transformer_blocks_1_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5038230848)))]; + tensor var_60318_cast_fp16 = conv(bias = up_blocks_1_attentions_1_transformer_blocks_1_ff_net_0_proj_bias_to_fp16, dilations = var_60316, groups = var_58034, pad = var_60318_pad_0, pad_type = var_60318_pad_type_0, strides = var_60314, weight = up_blocks_1_attentions_1_transformer_blocks_1_ff_net_0_proj_weight_to_fp16, x = input_783_cast_fp16)[name = tensor("op_60318_cast_fp16")]; + tensor var_60319_split_sizes_0 = const()[name = tensor("op_60319_split_sizes_0"), val = tensor([2560, 2560])]; + tensor var_60319_axis_0 = const()[name = tensor("op_60319_axis_0"), val = tensor(1)]; + tensor var_60319_cast_fp16_0, tensor var_60319_cast_fp16_1 = split(axis = var_60319_axis_0, split_sizes = var_60319_split_sizes_0, x = var_60318_cast_fp16)[name = tensor("op_60319_cast_fp16")]; + tensor var_60321_mode_0 = const()[name = tensor("op_60321_mode_0"), val = tensor("EXACT")]; + tensor var_60321_cast_fp16 = gelu(mode = var_60321_mode_0, x = var_60319_cast_fp16_1)[name = tensor("op_60321_cast_fp16")]; + tensor input_785_cast_fp16 = mul(x = var_60319_cast_fp16_0, y = var_60321_cast_fp16)[name = tensor("input_785_cast_fp16")]; + tensor var_60325 = const()[name = tensor("op_60325"), val = tensor([1, 1])]; + tensor var_60327 = const()[name = tensor("op_60327"), val = tensor([1, 1])]; + tensor var_60329_pad_type_0 = const()[name = tensor("op_60329_pad_type_0"), val = tensor("custom")]; + tensor var_60329_pad_0 = const()[name = tensor("op_60329_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_1_transformer_blocks_1_ff_net_2_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_transformer_blocks_1_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5038241152)))]; + tensor up_blocks_1_attentions_1_transformer_blocks_1_ff_net_2_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_transformer_blocks_1_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5041518016)))]; + tensor var_60329_cast_fp16 = conv(bias = up_blocks_1_attentions_1_transformer_blocks_1_ff_net_2_bias_to_fp16, dilations = var_60327, groups = var_58034, pad = var_60329_pad_0, pad_type = var_60329_pad_type_0, strides = var_60325, weight = up_blocks_1_attentions_1_transformer_blocks_1_ff_net_2_weight_to_fp16, x = input_785_cast_fp16)[name = tensor("op_60329_cast_fp16")]; + tensor hidden_states_547_cast_fp16 = add(x = var_60329_cast_fp16, y = inputs_407_cast_fp16)[name = tensor("hidden_states_547_cast_fp16")]; + tensor var_60331 = const()[name = tensor("op_60331"), val = tensor([2, 640, 64, 64])]; + tensor input_787_cast_fp16 = reshape(shape = var_60331, x = hidden_states_547_cast_fp16)[name = tensor("input_787_cast_fp16")]; + tensor var_60335 = const()[name = tensor("op_60335"), val = tensor([1, 1])]; + tensor var_60337 = const()[name = tensor("op_60337"), val = tensor([1, 1])]; + tensor hidden_states_549_pad_type_0 = const()[name = tensor("hidden_states_549_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_549_pad_0 = const()[name = tensor("hidden_states_549_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_1_proj_out_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_proj_out_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5041519360)))]; + tensor up_blocks_1_attentions_1_proj_out_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_proj_out_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5042338624)))]; + tensor hidden_states_549_cast_fp16 = conv(bias = up_blocks_1_attentions_1_proj_out_bias_to_fp16, dilations = var_60337, groups = var_58034, pad = hidden_states_549_pad_0, pad_type = hidden_states_549_pad_type_0, strides = var_60335, weight = up_blocks_1_attentions_1_proj_out_weight_to_fp16, x = input_787_cast_fp16)[name = tensor("hidden_states_549_cast_fp16")]; + tensor hidden_states_551_cast_fp16 = add(x = hidden_states_549_cast_fp16, y = hidden_states_531_cast_fp16)[name = tensor("hidden_states_551_cast_fp16")]; + tensor input_789_interleave_0 = const()[name = tensor("input_789_interleave_0"), val = tensor(false)]; + tensor input_789_cast_fp16 = concat(axis = var_58034, interleave = input_789_interleave_0, values = (hidden_states_551_cast_fp16, input_45_cast_fp16))[name = tensor("input_789_cast_fp16")]; + tensor reshape_144_shape_0 = const()[name = tensor("reshape_144_shape_0"), val = tensor([2, 32, 30, 64, 64])]; + tensor reshape_144_cast_fp16 = reshape(shape = reshape_144_shape_0, x = input_789_cast_fp16)[name = tensor("reshape_144_cast_fp16")]; + tensor reduce_mean_108_axes_0 = const()[name = tensor("reduce_mean_108_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_108_keep_dims_0 = const()[name = tensor("reduce_mean_108_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_108_cast_fp16 = reduce_mean(axes = reduce_mean_108_axes_0, keep_dims = reduce_mean_108_keep_dims_0, x = reshape_144_cast_fp16)[name = tensor("reduce_mean_108_cast_fp16")]; + tensor sub_72_cast_fp16 = sub(x = reshape_144_cast_fp16, y = reduce_mean_108_cast_fp16)[name = tensor("sub_72_cast_fp16")]; + tensor square_36_cast_fp16 = square(x = sub_72_cast_fp16)[name = tensor("square_36_cast_fp16")]; + tensor reduce_mean_110_axes_0 = const()[name = tensor("reduce_mean_110_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_110_keep_dims_0 = const()[name = tensor("reduce_mean_110_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_110_cast_fp16 = reduce_mean(axes = reduce_mean_110_axes_0, keep_dims = reduce_mean_110_keep_dims_0, x = square_36_cast_fp16)[name = tensor("reduce_mean_110_cast_fp16")]; + tensor add_72_y_0_to_fp16 = const()[name = tensor("add_72_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_72_cast_fp16 = add(x = reduce_mean_110_cast_fp16, y = add_72_y_0_to_fp16)[name = tensor("add_72_cast_fp16")]; + tensor sqrt_36_cast_fp16 = sqrt(x = add_72_cast_fp16)[name = tensor("sqrt_36_cast_fp16")]; + tensor real_div_36_cast_fp16 = real_div(x = sub_72_cast_fp16, y = sqrt_36_cast_fp16)[name = tensor("real_div_36_cast_fp16")]; + tensor reshape_145_shape_0 = const()[name = tensor("reshape_145_shape_0"), val = tensor([2, 960, 64, 64])]; + tensor reshape_145_cast_fp16 = reshape(shape = reshape_145_shape_0, x = real_div_36_cast_fp16)[name = tensor("reshape_145_cast_fp16")]; + tensor add_73_mean_0_to_fp16 = const()[name = tensor("add_73_mean_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5042339968)))]; + tensor add_73_variance_0_to_fp16 = const()[name = tensor("add_73_variance_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5042341952)))]; + tensor add_73_gamma_0_to_fp16 = const()[name = tensor("add_73_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5042343936)))]; + tensor add_73_beta_0_to_fp16 = const()[name = tensor("add_73_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5042345920)))]; + tensor add_73_epsilon_0_to_fp16 = const()[name = tensor("add_73_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_73_cast_fp16 = batch_norm(beta = add_73_beta_0_to_fp16, epsilon = add_73_epsilon_0_to_fp16, gamma = add_73_gamma_0_to_fp16, mean = add_73_mean_0_to_fp16, variance = add_73_variance_0_to_fp16, x = reshape_145_cast_fp16)[name = tensor("add_73_cast_fp16")]; + tensor input_793_cast_fp16 = silu(x = add_73_cast_fp16)[name = tensor("input_793_cast_fp16")]; + tensor var_60355 = const()[name = tensor("op_60355"), val = tensor([1, 1])]; + tensor var_60357 = const()[name = tensor("op_60357"), val = tensor([1, 1])]; + tensor hidden_states_553_pad_type_0 = const()[name = tensor("hidden_states_553_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_553_pad_0 = const()[name = tensor("hidden_states_553_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor up_blocks_1_resnets_2_conv1_weight_to_fp16 = const()[name = tensor("up_blocks_1_resnets_2_conv1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5042347904)))]; + tensor up_blocks_1_resnets_2_conv1_bias_to_fp16 = const()[name = tensor("up_blocks_1_resnets_2_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5053407168)))]; + tensor hidden_states_553_cast_fp16 = conv(bias = up_blocks_1_resnets_2_conv1_bias_to_fp16, dilations = var_60357, groups = var_58034, pad = hidden_states_553_pad_0, pad_type = hidden_states_553_pad_type_0, strides = var_60355, weight = up_blocks_1_resnets_2_conv1_weight_to_fp16, x = input_793_cast_fp16)[name = tensor("hidden_states_553_cast_fp16")]; + tensor var_60363 = const()[name = tensor("op_60363"), val = tensor([1, 1])]; + tensor var_60365 = const()[name = tensor("op_60365"), val = tensor([1, 1])]; + tensor temb_27_pad_type_0 = const()[name = tensor("temb_27_pad_type_0"), val = tensor("custom")]; + tensor temb_27_pad_0 = const()[name = tensor("temb_27_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_resnets_2_time_emb_proj_weight_to_fp16 = const()[name = tensor("up_blocks_1_resnets_2_time_emb_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5053408512)))]; + tensor up_blocks_1_resnets_2_time_emb_proj_bias_to_fp16 = const()[name = tensor("up_blocks_1_resnets_2_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5055046976)))]; + tensor temb_27_cast_fp16 = conv(bias = up_blocks_1_resnets_2_time_emb_proj_bias_to_fp16, dilations = var_60365, groups = var_58034, pad = temb_27_pad_0, pad_type = temb_27_pad_type_0, strides = var_60363, weight = up_blocks_1_resnets_2_time_emb_proj_weight_to_fp16, x = input_21_cast_fp16)[name = tensor("temb_27_cast_fp16")]; + tensor input_797_cast_fp16 = add(x = hidden_states_553_cast_fp16, y = temb_27_cast_fp16)[name = tensor("input_797_cast_fp16")]; + tensor reshape_148_shape_0 = const()[name = tensor("reshape_148_shape_0"), val = tensor([2, 32, 20, 64, 64])]; + tensor reshape_148_cast_fp16 = reshape(shape = reshape_148_shape_0, x = input_797_cast_fp16)[name = tensor("reshape_148_cast_fp16")]; + tensor reduce_mean_111_axes_0 = const()[name = tensor("reduce_mean_111_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_111_keep_dims_0 = const()[name = tensor("reduce_mean_111_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_111_cast_fp16 = reduce_mean(axes = reduce_mean_111_axes_0, keep_dims = reduce_mean_111_keep_dims_0, x = reshape_148_cast_fp16)[name = tensor("reduce_mean_111_cast_fp16")]; + tensor sub_74_cast_fp16 = sub(x = reshape_148_cast_fp16, y = reduce_mean_111_cast_fp16)[name = tensor("sub_74_cast_fp16")]; + tensor square_37_cast_fp16 = square(x = sub_74_cast_fp16)[name = tensor("square_37_cast_fp16")]; + tensor reduce_mean_113_axes_0 = const()[name = tensor("reduce_mean_113_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_113_keep_dims_0 = const()[name = tensor("reduce_mean_113_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_113_cast_fp16 = reduce_mean(axes = reduce_mean_113_axes_0, keep_dims = reduce_mean_113_keep_dims_0, x = square_37_cast_fp16)[name = tensor("reduce_mean_113_cast_fp16")]; + tensor add_74_y_0_to_fp16 = const()[name = tensor("add_74_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_74_cast_fp16 = add(x = reduce_mean_113_cast_fp16, y = add_74_y_0_to_fp16)[name = tensor("add_74_cast_fp16")]; + tensor sqrt_37_cast_fp16 = sqrt(x = add_74_cast_fp16)[name = tensor("sqrt_37_cast_fp16")]; + tensor real_div_37_cast_fp16 = real_div(x = sub_74_cast_fp16, y = sqrt_37_cast_fp16)[name = tensor("real_div_37_cast_fp16")]; + tensor reshape_149_shape_0 = const()[name = tensor("reshape_149_shape_0"), val = tensor([2, 640, 64, 64])]; + tensor reshape_149_cast_fp16 = reshape(shape = reshape_149_shape_0, x = real_div_37_cast_fp16)[name = tensor("reshape_149_cast_fp16")]; + tensor add_75_gamma_0_to_fp16 = const()[name = tensor("add_75_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5055048320)))]; + tensor add_75_beta_0_to_fp16 = const()[name = tensor("add_75_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5055049664)))]; + tensor add_75_epsilon_0_to_fp16 = const()[name = tensor("add_75_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_75_cast_fp16 = batch_norm(beta = add_75_beta_0_to_fp16, epsilon = add_75_epsilon_0_to_fp16, gamma = add_75_gamma_0_to_fp16, mean = add_11_mean_0_to_fp16, variance = add_11_variance_0_to_fp16, x = reshape_149_cast_fp16)[name = tensor("add_75_cast_fp16")]; + tensor input_801_cast_fp16 = silu(x = add_75_cast_fp16)[name = tensor("input_801_cast_fp16")]; + tensor var_60375 = const()[name = tensor("op_60375"), val = tensor([1, 1])]; + tensor var_60377 = const()[name = tensor("op_60377"), val = tensor([1, 1])]; + tensor hidden_states_555_pad_type_0 = const()[name = tensor("hidden_states_555_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_555_pad_0 = const()[name = tensor("hidden_states_555_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor up_blocks_1_resnets_2_conv2_weight_to_fp16 = const()[name = tensor("up_blocks_1_resnets_2_conv2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5055051008)))]; + tensor up_blocks_1_resnets_2_conv2_bias_to_fp16 = const()[name = tensor("up_blocks_1_resnets_2_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5062423872)))]; + tensor hidden_states_555_cast_fp16 = conv(bias = up_blocks_1_resnets_2_conv2_bias_to_fp16, dilations = var_60377, groups = var_58034, pad = hidden_states_555_pad_0, pad_type = hidden_states_555_pad_type_0, strides = var_60375, weight = up_blocks_1_resnets_2_conv2_weight_to_fp16, x = input_801_cast_fp16)[name = tensor("hidden_states_555_cast_fp16")]; + tensor var_60382 = const()[name = tensor("op_60382"), val = tensor([1, 1])]; + tensor var_60384 = const()[name = tensor("op_60384"), val = tensor([1, 1])]; + tensor x_15_pad_type_0 = const()[name = tensor("x_15_pad_type_0"), val = tensor("custom")]; + tensor x_15_pad_0 = const()[name = tensor("x_15_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_resnets_2_conv_shortcut_weight_to_fp16 = const()[name = tensor("up_blocks_1_resnets_2_conv_shortcut_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5062425216)))]; + tensor up_blocks_1_resnets_2_conv_shortcut_bias_to_fp16 = const()[name = tensor("up_blocks_1_resnets_2_conv_shortcut_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5063654080)))]; + tensor x_15_cast_fp16 = conv(bias = up_blocks_1_resnets_2_conv_shortcut_bias_to_fp16, dilations = var_60384, groups = var_58034, pad = x_15_pad_0, pad_type = x_15_pad_type_0, strides = var_60382, weight = up_blocks_1_resnets_2_conv_shortcut_weight_to_fp16, x = input_789_cast_fp16)[name = tensor("x_15_cast_fp16")]; + tensor hidden_states_557_cast_fp16 = add(x = x_15_cast_fp16, y = hidden_states_555_cast_fp16)[name = tensor("hidden_states_557_cast_fp16")]; + tensor reshape_152_shape_0 = const()[name = tensor("reshape_152_shape_0"), val = tensor([2, 32, 20, 64, 64])]; + tensor reshape_152_cast_fp16 = reshape(shape = reshape_152_shape_0, x = hidden_states_557_cast_fp16)[name = tensor("reshape_152_cast_fp16")]; + tensor reduce_mean_114_axes_0 = const()[name = tensor("reduce_mean_114_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_114_keep_dims_0 = const()[name = tensor("reduce_mean_114_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_114_cast_fp16 = reduce_mean(axes = reduce_mean_114_axes_0, keep_dims = reduce_mean_114_keep_dims_0, x = reshape_152_cast_fp16)[name = tensor("reduce_mean_114_cast_fp16")]; + tensor sub_76_cast_fp16 = sub(x = reshape_152_cast_fp16, y = reduce_mean_114_cast_fp16)[name = tensor("sub_76_cast_fp16")]; + tensor square_38_cast_fp16 = square(x = sub_76_cast_fp16)[name = tensor("square_38_cast_fp16")]; + tensor reduce_mean_116_axes_0 = const()[name = tensor("reduce_mean_116_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_116_keep_dims_0 = const()[name = tensor("reduce_mean_116_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_116_cast_fp16 = reduce_mean(axes = reduce_mean_116_axes_0, keep_dims = reduce_mean_116_keep_dims_0, x = square_38_cast_fp16)[name = tensor("reduce_mean_116_cast_fp16")]; + tensor add_76_y_0_to_fp16 = const()[name = tensor("add_76_y_0_to_fp16"), val = tensor(0x1.1p-20)]; + tensor add_76_cast_fp16 = add(x = reduce_mean_116_cast_fp16, y = add_76_y_0_to_fp16)[name = tensor("add_76_cast_fp16")]; + tensor sqrt_38_cast_fp16 = sqrt(x = add_76_cast_fp16)[name = tensor("sqrt_38_cast_fp16")]; + tensor real_div_38_cast_fp16 = real_div(x = sub_76_cast_fp16, y = sqrt_38_cast_fp16)[name = tensor("real_div_38_cast_fp16")]; + tensor reshape_153_shape_0 = const()[name = tensor("reshape_153_shape_0"), val = tensor([2, 640, 64, 64])]; + tensor reshape_153_cast_fp16 = reshape(shape = reshape_153_shape_0, x = real_div_38_cast_fp16)[name = tensor("reshape_153_cast_fp16")]; + tensor add_77_gamma_0_to_fp16 = const()[name = tensor("add_77_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5063655424)))]; + tensor add_77_beta_0_to_fp16 = const()[name = tensor("add_77_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5063656768)))]; + tensor add_77_epsilon_0_to_fp16 = const()[name = tensor("add_77_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_77_cast_fp16 = batch_norm(beta = add_77_beta_0_to_fp16, epsilon = add_77_epsilon_0_to_fp16, gamma = add_77_gamma_0_to_fp16, mean = add_11_mean_0_to_fp16, variance = add_11_variance_0_to_fp16, x = reshape_153_cast_fp16)[name = tensor("add_77_cast_fp16")]; + tensor var_60406 = const()[name = tensor("op_60406"), val = tensor([1, 1])]; + tensor var_60408 = const()[name = tensor("op_60408"), val = tensor([1, 1])]; + tensor hidden_states_559_pad_type_0 = const()[name = tensor("hidden_states_559_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_559_pad_0 = const()[name = tensor("hidden_states_559_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_2_proj_in_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_proj_in_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5063658112)))]; + tensor up_blocks_1_attentions_2_proj_in_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_proj_in_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5064477376)))]; + tensor hidden_states_559_cast_fp16 = conv(bias = up_blocks_1_attentions_2_proj_in_bias_to_fp16, dilations = var_60408, groups = var_58034, pad = hidden_states_559_pad_0, pad_type = hidden_states_559_pad_type_0, strides = var_60406, weight = up_blocks_1_attentions_2_proj_in_weight_to_fp16, x = add_77_cast_fp16)[name = tensor("hidden_states_559_cast_fp16")]; + tensor var_60413 = const()[name = tensor("op_60413"), val = tensor([2, 640, 1, 4096])]; + tensor inputs_409_cast_fp16 = reshape(shape = var_60413, x = hidden_states_559_cast_fp16)[name = tensor("inputs_409_cast_fp16")]; + tensor hidden_states_561_axes_0 = const()[name = tensor("hidden_states_561_axes_0"), val = tensor([1])]; + tensor hidden_states_561_gamma_0_to_fp16 = const()[name = tensor("hidden_states_561_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5064478720)))]; + tensor hidden_states_561_beta_0_to_fp16 = const()[name = tensor("hidden_states_561_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5064480064)))]; + tensor var_60429_to_fp16 = const()[name = tensor("op_60429_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_561_cast_fp16 = layer_norm(axes = hidden_states_561_axes_0, beta = hidden_states_561_beta_0_to_fp16, epsilon = var_60429_to_fp16, gamma = hidden_states_561_gamma_0_to_fp16, x = inputs_409_cast_fp16)[name = tensor("hidden_states_561_cast_fp16")]; + tensor var_60444 = const()[name = tensor("op_60444"), val = tensor([1, 1])]; + tensor var_60446 = const()[name = tensor("op_60446"), val = tensor([1, 1])]; + tensor q_273_pad_type_0 = const()[name = tensor("q_273_pad_type_0"), val = tensor("custom")]; + tensor q_273_pad_0 = const()[name = tensor("q_273_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5064481408)))]; + tensor q_273_cast_fp16 = conv(dilations = var_60446, groups = var_58034, pad = q_273_pad_0, pad_type = q_273_pad_type_0, strides = var_60444, weight = up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_q_weight_to_fp16, x = hidden_states_561_cast_fp16)[name = tensor("q_273_cast_fp16")]; + tensor var_60450 = const()[name = tensor("op_60450"), val = tensor([1, 1])]; + tensor var_60452 = const()[name = tensor("op_60452"), val = tensor([1, 1])]; + tensor k_545_pad_type_0 = const()[name = tensor("k_545_pad_type_0"), val = tensor("custom")]; + tensor k_545_pad_0 = const()[name = tensor("k_545_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5065300672)))]; + tensor k_545_cast_fp16 = conv(dilations = var_60452, groups = var_58034, pad = k_545_pad_0, pad_type = k_545_pad_type_0, strides = var_60450, weight = up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_k_weight_to_fp16, x = hidden_states_561_cast_fp16)[name = tensor("k_545_cast_fp16")]; + tensor var_60456 = const()[name = tensor("op_60456"), val = tensor([1, 1])]; + tensor var_60458 = const()[name = tensor("op_60458"), val = tensor([1, 1])]; + tensor v_273_pad_type_0 = const()[name = tensor("v_273_pad_type_0"), val = tensor("custom")]; + tensor v_273_pad_0 = const()[name = tensor("v_273_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5066119936)))]; + tensor v_273_cast_fp16 = conv(dilations = var_60458, groups = var_58034, pad = v_273_pad_0, pad_type = v_273_pad_type_0, strides = var_60456, weight = up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_v_weight_to_fp16, x = hidden_states_561_cast_fp16)[name = tensor("v_273_cast_fp16")]; + tensor var_60462_begin_0 = const()[name = tensor("op_60462_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_60462_end_0 = const()[name = tensor("op_60462_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_60462_end_mask_0 = const()[name = tensor("op_60462_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_60462_cast_fp16 = slice_by_index(begin = var_60462_begin_0, end = var_60462_end_0, end_mask = var_60462_end_mask_0, x = q_273_cast_fp16)[name = tensor("op_60462_cast_fp16")]; + tensor var_60466_begin_0 = const()[name = tensor("op_60466_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_60466_end_0 = const()[name = tensor("op_60466_end_0"), val = tensor([2, 128, 1, 4096])]; + tensor var_60466_end_mask_0 = const()[name = tensor("op_60466_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_60466_cast_fp16 = slice_by_index(begin = var_60466_begin_0, end = var_60466_end_0, end_mask = var_60466_end_mask_0, x = q_273_cast_fp16)[name = tensor("op_60466_cast_fp16")]; + tensor var_60470_begin_0 = const()[name = tensor("op_60470_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_60470_end_0 = const()[name = tensor("op_60470_end_0"), val = tensor([2, 192, 1, 4096])]; + tensor var_60470_end_mask_0 = const()[name = tensor("op_60470_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_60470_cast_fp16 = slice_by_index(begin = var_60470_begin_0, end = var_60470_end_0, end_mask = var_60470_end_mask_0, x = q_273_cast_fp16)[name = tensor("op_60470_cast_fp16")]; + tensor var_60474_begin_0 = const()[name = tensor("op_60474_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_60474_end_0 = const()[name = tensor("op_60474_end_0"), val = tensor([2, 256, 1, 4096])]; + tensor var_60474_end_mask_0 = const()[name = tensor("op_60474_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_60474_cast_fp16 = slice_by_index(begin = var_60474_begin_0, end = var_60474_end_0, end_mask = var_60474_end_mask_0, x = q_273_cast_fp16)[name = tensor("op_60474_cast_fp16")]; + tensor var_60478_begin_0 = const()[name = tensor("op_60478_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_60478_end_0 = const()[name = tensor("op_60478_end_0"), val = tensor([2, 320, 1, 4096])]; + tensor var_60478_end_mask_0 = const()[name = tensor("op_60478_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_60478_cast_fp16 = slice_by_index(begin = var_60478_begin_0, end = var_60478_end_0, end_mask = var_60478_end_mask_0, x = q_273_cast_fp16)[name = tensor("op_60478_cast_fp16")]; + tensor var_60482_begin_0 = const()[name = tensor("op_60482_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_60482_end_0 = const()[name = tensor("op_60482_end_0"), val = tensor([2, 384, 1, 4096])]; + tensor var_60482_end_mask_0 = const()[name = tensor("op_60482_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_60482_cast_fp16 = slice_by_index(begin = var_60482_begin_0, end = var_60482_end_0, end_mask = var_60482_end_mask_0, x = q_273_cast_fp16)[name = tensor("op_60482_cast_fp16")]; + tensor var_60486_begin_0 = const()[name = tensor("op_60486_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_60486_end_0 = const()[name = tensor("op_60486_end_0"), val = tensor([2, 448, 1, 4096])]; + tensor var_60486_end_mask_0 = const()[name = tensor("op_60486_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_60486_cast_fp16 = slice_by_index(begin = var_60486_begin_0, end = var_60486_end_0, end_mask = var_60486_end_mask_0, x = q_273_cast_fp16)[name = tensor("op_60486_cast_fp16")]; + tensor var_60490_begin_0 = const()[name = tensor("op_60490_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_60490_end_0 = const()[name = tensor("op_60490_end_0"), val = tensor([2, 512, 1, 4096])]; + tensor var_60490_end_mask_0 = const()[name = tensor("op_60490_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_60490_cast_fp16 = slice_by_index(begin = var_60490_begin_0, end = var_60490_end_0, end_mask = var_60490_end_mask_0, x = q_273_cast_fp16)[name = tensor("op_60490_cast_fp16")]; + tensor var_60494_begin_0 = const()[name = tensor("op_60494_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_60494_end_0 = const()[name = tensor("op_60494_end_0"), val = tensor([2, 576, 1, 4096])]; + tensor var_60494_end_mask_0 = const()[name = tensor("op_60494_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_60494_cast_fp16 = slice_by_index(begin = var_60494_begin_0, end = var_60494_end_0, end_mask = var_60494_end_mask_0, x = q_273_cast_fp16)[name = tensor("op_60494_cast_fp16")]; + tensor var_60498_begin_0 = const()[name = tensor("op_60498_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_60498_end_0 = const()[name = tensor("op_60498_end_0"), val = tensor([2, 640, 1, 4096])]; + tensor var_60498_end_mask_0 = const()[name = tensor("op_60498_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_60498_cast_fp16 = slice_by_index(begin = var_60498_begin_0, end = var_60498_end_0, end_mask = var_60498_end_mask_0, x = q_273_cast_fp16)[name = tensor("op_60498_cast_fp16")]; + tensor k_547_perm_0 = const()[name = tensor("k_547_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_60505_begin_0 = const()[name = tensor("op_60505_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_60505_end_0 = const()[name = tensor("op_60505_end_0"), val = tensor([2, 4096, 1, 64])]; + tensor var_60505_end_mask_0 = const()[name = tensor("op_60505_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_547_cast_fp16 = transpose(perm = k_547_perm_0, x = k_545_cast_fp16)[name = tensor("transpose_3")]; + tensor var_60505_cast_fp16 = slice_by_index(begin = var_60505_begin_0, end = var_60505_end_0, end_mask = var_60505_end_mask_0, x = k_547_cast_fp16)[name = tensor("op_60505_cast_fp16")]; + tensor var_60509_begin_0 = const()[name = tensor("op_60509_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_60509_end_0 = const()[name = tensor("op_60509_end_0"), val = tensor([2, 4096, 1, 128])]; + tensor var_60509_end_mask_0 = const()[name = tensor("op_60509_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_60509_cast_fp16 = slice_by_index(begin = var_60509_begin_0, end = var_60509_end_0, end_mask = var_60509_end_mask_0, x = k_547_cast_fp16)[name = tensor("op_60509_cast_fp16")]; + tensor var_60513_begin_0 = const()[name = tensor("op_60513_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_60513_end_0 = const()[name = tensor("op_60513_end_0"), val = tensor([2, 4096, 1, 192])]; + tensor var_60513_end_mask_0 = const()[name = tensor("op_60513_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_60513_cast_fp16 = slice_by_index(begin = var_60513_begin_0, end = var_60513_end_0, end_mask = var_60513_end_mask_0, x = k_547_cast_fp16)[name = tensor("op_60513_cast_fp16")]; + tensor var_60517_begin_0 = const()[name = tensor("op_60517_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_60517_end_0 = const()[name = tensor("op_60517_end_0"), val = tensor([2, 4096, 1, 256])]; + tensor var_60517_end_mask_0 = const()[name = tensor("op_60517_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_60517_cast_fp16 = slice_by_index(begin = var_60517_begin_0, end = var_60517_end_0, end_mask = var_60517_end_mask_0, x = k_547_cast_fp16)[name = tensor("op_60517_cast_fp16")]; + tensor var_60521_begin_0 = const()[name = tensor("op_60521_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_60521_end_0 = const()[name = tensor("op_60521_end_0"), val = tensor([2, 4096, 1, 320])]; + tensor var_60521_end_mask_0 = const()[name = tensor("op_60521_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_60521_cast_fp16 = slice_by_index(begin = var_60521_begin_0, end = var_60521_end_0, end_mask = var_60521_end_mask_0, x = k_547_cast_fp16)[name = tensor("op_60521_cast_fp16")]; + tensor var_60525_begin_0 = const()[name = tensor("op_60525_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_60525_end_0 = const()[name = tensor("op_60525_end_0"), val = tensor([2, 4096, 1, 384])]; + tensor var_60525_end_mask_0 = const()[name = tensor("op_60525_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_60525_cast_fp16 = slice_by_index(begin = var_60525_begin_0, end = var_60525_end_0, end_mask = var_60525_end_mask_0, x = k_547_cast_fp16)[name = tensor("op_60525_cast_fp16")]; + tensor var_60529_begin_0 = const()[name = tensor("op_60529_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_60529_end_0 = const()[name = tensor("op_60529_end_0"), val = tensor([2, 4096, 1, 448])]; + tensor var_60529_end_mask_0 = const()[name = tensor("op_60529_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_60529_cast_fp16 = slice_by_index(begin = var_60529_begin_0, end = var_60529_end_0, end_mask = var_60529_end_mask_0, x = k_547_cast_fp16)[name = tensor("op_60529_cast_fp16")]; + tensor var_60533_begin_0 = const()[name = tensor("op_60533_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_60533_end_0 = const()[name = tensor("op_60533_end_0"), val = tensor([2, 4096, 1, 512])]; + tensor var_60533_end_mask_0 = const()[name = tensor("op_60533_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_60533_cast_fp16 = slice_by_index(begin = var_60533_begin_0, end = var_60533_end_0, end_mask = var_60533_end_mask_0, x = k_547_cast_fp16)[name = tensor("op_60533_cast_fp16")]; + tensor var_60537_begin_0 = const()[name = tensor("op_60537_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_60537_end_0 = const()[name = tensor("op_60537_end_0"), val = tensor([2, 4096, 1, 576])]; + tensor var_60537_end_mask_0 = const()[name = tensor("op_60537_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_60537_cast_fp16 = slice_by_index(begin = var_60537_begin_0, end = var_60537_end_0, end_mask = var_60537_end_mask_0, x = k_547_cast_fp16)[name = tensor("op_60537_cast_fp16")]; + tensor var_60541_begin_0 = const()[name = tensor("op_60541_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_60541_end_0 = const()[name = tensor("op_60541_end_0"), val = tensor([2, 4096, 1, 640])]; + tensor var_60541_end_mask_0 = const()[name = tensor("op_60541_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_60541_cast_fp16 = slice_by_index(begin = var_60541_begin_0, end = var_60541_end_0, end_mask = var_60541_end_mask_0, x = k_547_cast_fp16)[name = tensor("op_60541_cast_fp16")]; + tensor var_60543_begin_0 = const()[name = tensor("op_60543_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_60543_end_0 = const()[name = tensor("op_60543_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_60543_end_mask_0 = const()[name = tensor("op_60543_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_60543_cast_fp16 = slice_by_index(begin = var_60543_begin_0, end = var_60543_end_0, end_mask = var_60543_end_mask_0, x = v_273_cast_fp16)[name = tensor("op_60543_cast_fp16")]; + tensor var_60547_begin_0 = const()[name = tensor("op_60547_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_60547_end_0 = const()[name = tensor("op_60547_end_0"), val = tensor([2, 128, 1, 4096])]; + tensor var_60547_end_mask_0 = const()[name = tensor("op_60547_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_60547_cast_fp16 = slice_by_index(begin = var_60547_begin_0, end = var_60547_end_0, end_mask = var_60547_end_mask_0, x = v_273_cast_fp16)[name = tensor("op_60547_cast_fp16")]; + tensor var_60551_begin_0 = const()[name = tensor("op_60551_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_60551_end_0 = const()[name = tensor("op_60551_end_0"), val = tensor([2, 192, 1, 4096])]; + tensor var_60551_end_mask_0 = const()[name = tensor("op_60551_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_60551_cast_fp16 = slice_by_index(begin = var_60551_begin_0, end = var_60551_end_0, end_mask = var_60551_end_mask_0, x = v_273_cast_fp16)[name = tensor("op_60551_cast_fp16")]; + tensor var_60555_begin_0 = const()[name = tensor("op_60555_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_60555_end_0 = const()[name = tensor("op_60555_end_0"), val = tensor([2, 256, 1, 4096])]; + tensor var_60555_end_mask_0 = const()[name = tensor("op_60555_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_60555_cast_fp16 = slice_by_index(begin = var_60555_begin_0, end = var_60555_end_0, end_mask = var_60555_end_mask_0, x = v_273_cast_fp16)[name = tensor("op_60555_cast_fp16")]; + tensor var_60559_begin_0 = const()[name = tensor("op_60559_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_60559_end_0 = const()[name = tensor("op_60559_end_0"), val = tensor([2, 320, 1, 4096])]; + tensor var_60559_end_mask_0 = const()[name = tensor("op_60559_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_60559_cast_fp16 = slice_by_index(begin = var_60559_begin_0, end = var_60559_end_0, end_mask = var_60559_end_mask_0, x = v_273_cast_fp16)[name = tensor("op_60559_cast_fp16")]; + tensor var_60563_begin_0 = const()[name = tensor("op_60563_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_60563_end_0 = const()[name = tensor("op_60563_end_0"), val = tensor([2, 384, 1, 4096])]; + tensor var_60563_end_mask_0 = const()[name = tensor("op_60563_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_60563_cast_fp16 = slice_by_index(begin = var_60563_begin_0, end = var_60563_end_0, end_mask = var_60563_end_mask_0, x = v_273_cast_fp16)[name = tensor("op_60563_cast_fp16")]; + tensor var_60567_begin_0 = const()[name = tensor("op_60567_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_60567_end_0 = const()[name = tensor("op_60567_end_0"), val = tensor([2, 448, 1, 4096])]; + tensor var_60567_end_mask_0 = const()[name = tensor("op_60567_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_60567_cast_fp16 = slice_by_index(begin = var_60567_begin_0, end = var_60567_end_0, end_mask = var_60567_end_mask_0, x = v_273_cast_fp16)[name = tensor("op_60567_cast_fp16")]; + tensor var_60571_begin_0 = const()[name = tensor("op_60571_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_60571_end_0 = const()[name = tensor("op_60571_end_0"), val = tensor([2, 512, 1, 4096])]; + tensor var_60571_end_mask_0 = const()[name = tensor("op_60571_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_60571_cast_fp16 = slice_by_index(begin = var_60571_begin_0, end = var_60571_end_0, end_mask = var_60571_end_mask_0, x = v_273_cast_fp16)[name = tensor("op_60571_cast_fp16")]; + tensor var_60575_begin_0 = const()[name = tensor("op_60575_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_60575_end_0 = const()[name = tensor("op_60575_end_0"), val = tensor([2, 576, 1, 4096])]; + tensor var_60575_end_mask_0 = const()[name = tensor("op_60575_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_60575_cast_fp16 = slice_by_index(begin = var_60575_begin_0, end = var_60575_end_0, end_mask = var_60575_end_mask_0, x = v_273_cast_fp16)[name = tensor("op_60575_cast_fp16")]; + tensor var_60579_begin_0 = const()[name = tensor("op_60579_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_60579_end_0 = const()[name = tensor("op_60579_end_0"), val = tensor([2, 640, 1, 4096])]; + tensor var_60579_end_mask_0 = const()[name = tensor("op_60579_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_60579_cast_fp16 = slice_by_index(begin = var_60579_begin_0, end = var_60579_end_0, end_mask = var_60579_end_mask_0, x = v_273_cast_fp16)[name = tensor("op_60579_cast_fp16")]; + tensor var_60583_equation_0 = const()[name = tensor("op_60583_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_60583_cast_fp16 = einsum(equation = var_60583_equation_0, values = (var_60505_cast_fp16, var_60462_cast_fp16))[name = tensor("op_60583_cast_fp16")]; + tensor var_60584_to_fp16 = const()[name = tensor("op_60584_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5121_cast_fp16 = mul(x = var_60583_cast_fp16, y = var_60584_to_fp16)[name = tensor("aw_5121_cast_fp16")]; + tensor var_60587_equation_0 = const()[name = tensor("op_60587_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_60587_cast_fp16 = einsum(equation = var_60587_equation_0, values = (var_60509_cast_fp16, var_60466_cast_fp16))[name = tensor("op_60587_cast_fp16")]; + tensor var_60588_to_fp16 = const()[name = tensor("op_60588_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5123_cast_fp16 = mul(x = var_60587_cast_fp16, y = var_60588_to_fp16)[name = tensor("aw_5123_cast_fp16")]; + tensor var_60591_equation_0 = const()[name = tensor("op_60591_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_60591_cast_fp16 = einsum(equation = var_60591_equation_0, values = (var_60513_cast_fp16, var_60470_cast_fp16))[name = tensor("op_60591_cast_fp16")]; + tensor var_60592_to_fp16 = const()[name = tensor("op_60592_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5125_cast_fp16 = mul(x = var_60591_cast_fp16, y = var_60592_to_fp16)[name = tensor("aw_5125_cast_fp16")]; + tensor var_60595_equation_0 = const()[name = tensor("op_60595_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_60595_cast_fp16 = einsum(equation = var_60595_equation_0, values = (var_60517_cast_fp16, var_60474_cast_fp16))[name = tensor("op_60595_cast_fp16")]; + tensor var_60596_to_fp16 = const()[name = tensor("op_60596_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5127_cast_fp16 = mul(x = var_60595_cast_fp16, y = var_60596_to_fp16)[name = tensor("aw_5127_cast_fp16")]; + tensor var_60599_equation_0 = const()[name = tensor("op_60599_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_60599_cast_fp16 = einsum(equation = var_60599_equation_0, values = (var_60521_cast_fp16, var_60478_cast_fp16))[name = tensor("op_60599_cast_fp16")]; + tensor var_60600_to_fp16 = const()[name = tensor("op_60600_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5129_cast_fp16 = mul(x = var_60599_cast_fp16, y = var_60600_to_fp16)[name = tensor("aw_5129_cast_fp16")]; + tensor var_60603_equation_0 = const()[name = tensor("op_60603_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_60603_cast_fp16 = einsum(equation = var_60603_equation_0, values = (var_60525_cast_fp16, var_60482_cast_fp16))[name = tensor("op_60603_cast_fp16")]; + tensor var_60604_to_fp16 = const()[name = tensor("op_60604_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5131_cast_fp16 = mul(x = var_60603_cast_fp16, y = var_60604_to_fp16)[name = tensor("aw_5131_cast_fp16")]; + tensor var_60607_equation_0 = const()[name = tensor("op_60607_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_60607_cast_fp16 = einsum(equation = var_60607_equation_0, values = (var_60529_cast_fp16, var_60486_cast_fp16))[name = tensor("op_60607_cast_fp16")]; + tensor var_60608_to_fp16 = const()[name = tensor("op_60608_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5133_cast_fp16 = mul(x = var_60607_cast_fp16, y = var_60608_to_fp16)[name = tensor("aw_5133_cast_fp16")]; + tensor var_60611_equation_0 = const()[name = tensor("op_60611_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_60611_cast_fp16 = einsum(equation = var_60611_equation_0, values = (var_60533_cast_fp16, var_60490_cast_fp16))[name = tensor("op_60611_cast_fp16")]; + tensor var_60612_to_fp16 = const()[name = tensor("op_60612_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5135_cast_fp16 = mul(x = var_60611_cast_fp16, y = var_60612_to_fp16)[name = tensor("aw_5135_cast_fp16")]; + tensor var_60615_equation_0 = const()[name = tensor("op_60615_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_60615_cast_fp16 = einsum(equation = var_60615_equation_0, values = (var_60537_cast_fp16, var_60494_cast_fp16))[name = tensor("op_60615_cast_fp16")]; + tensor var_60616_to_fp16 = const()[name = tensor("op_60616_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5137_cast_fp16 = mul(x = var_60615_cast_fp16, y = var_60616_to_fp16)[name = tensor("aw_5137_cast_fp16")]; + tensor var_60619_equation_0 = const()[name = tensor("op_60619_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_60619_cast_fp16 = einsum(equation = var_60619_equation_0, values = (var_60541_cast_fp16, var_60498_cast_fp16))[name = tensor("op_60619_cast_fp16")]; + tensor var_60620_to_fp16 = const()[name = tensor("op_60620_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5139_cast_fp16 = mul(x = var_60619_cast_fp16, y = var_60620_to_fp16)[name = tensor("aw_5139_cast_fp16")]; + tensor var_60622_cast_fp16 = softmax(axis = var_58034, x = aw_5121_cast_fp16)[name = tensor("op_60622_cast_fp16")]; + tensor var_60623_cast_fp16 = softmax(axis = var_58034, x = aw_5123_cast_fp16)[name = tensor("op_60623_cast_fp16")]; + tensor var_60624_cast_fp16 = softmax(axis = var_58034, x = aw_5125_cast_fp16)[name = tensor("op_60624_cast_fp16")]; + tensor var_60625_cast_fp16 = softmax(axis = var_58034, x = aw_5127_cast_fp16)[name = tensor("op_60625_cast_fp16")]; + tensor var_60626_cast_fp16 = softmax(axis = var_58034, x = aw_5129_cast_fp16)[name = tensor("op_60626_cast_fp16")]; + tensor var_60627_cast_fp16 = softmax(axis = var_58034, x = aw_5131_cast_fp16)[name = tensor("op_60627_cast_fp16")]; + tensor var_60628_cast_fp16 = softmax(axis = var_58034, x = aw_5133_cast_fp16)[name = tensor("op_60628_cast_fp16")]; + tensor var_60629_cast_fp16 = softmax(axis = var_58034, x = aw_5135_cast_fp16)[name = tensor("op_60629_cast_fp16")]; + tensor var_60630_cast_fp16 = softmax(axis = var_58034, x = aw_5137_cast_fp16)[name = tensor("op_60630_cast_fp16")]; + tensor var_60631_cast_fp16 = softmax(axis = var_58034, x = aw_5139_cast_fp16)[name = tensor("op_60631_cast_fp16")]; + tensor var_60633_equation_0 = const()[name = tensor("op_60633_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_60633_cast_fp16 = einsum(equation = var_60633_equation_0, values = (var_60543_cast_fp16, var_60622_cast_fp16))[name = tensor("op_60633_cast_fp16")]; + tensor var_60635_equation_0 = const()[name = tensor("op_60635_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_60635_cast_fp16 = einsum(equation = var_60635_equation_0, values = (var_60547_cast_fp16, var_60623_cast_fp16))[name = tensor("op_60635_cast_fp16")]; + tensor var_60637_equation_0 = const()[name = tensor("op_60637_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_60637_cast_fp16 = einsum(equation = var_60637_equation_0, values = (var_60551_cast_fp16, var_60624_cast_fp16))[name = tensor("op_60637_cast_fp16")]; + tensor var_60639_equation_0 = const()[name = tensor("op_60639_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_60639_cast_fp16 = einsum(equation = var_60639_equation_0, values = (var_60555_cast_fp16, var_60625_cast_fp16))[name = tensor("op_60639_cast_fp16")]; + tensor var_60641_equation_0 = const()[name = tensor("op_60641_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_60641_cast_fp16 = einsum(equation = var_60641_equation_0, values = (var_60559_cast_fp16, var_60626_cast_fp16))[name = tensor("op_60641_cast_fp16")]; + tensor var_60643_equation_0 = const()[name = tensor("op_60643_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_60643_cast_fp16 = einsum(equation = var_60643_equation_0, values = (var_60563_cast_fp16, var_60627_cast_fp16))[name = tensor("op_60643_cast_fp16")]; + tensor var_60645_equation_0 = const()[name = tensor("op_60645_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_60645_cast_fp16 = einsum(equation = var_60645_equation_0, values = (var_60567_cast_fp16, var_60628_cast_fp16))[name = tensor("op_60645_cast_fp16")]; + tensor var_60647_equation_0 = const()[name = tensor("op_60647_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_60647_cast_fp16 = einsum(equation = var_60647_equation_0, values = (var_60571_cast_fp16, var_60629_cast_fp16))[name = tensor("op_60647_cast_fp16")]; + tensor var_60649_equation_0 = const()[name = tensor("op_60649_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_60649_cast_fp16 = einsum(equation = var_60649_equation_0, values = (var_60575_cast_fp16, var_60630_cast_fp16))[name = tensor("op_60649_cast_fp16")]; + tensor var_60651_equation_0 = const()[name = tensor("op_60651_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_60651_cast_fp16 = einsum(equation = var_60651_equation_0, values = (var_60579_cast_fp16, var_60631_cast_fp16))[name = tensor("op_60651_cast_fp16")]; + tensor input_805_interleave_0 = const()[name = tensor("input_805_interleave_0"), val = tensor(false)]; + tensor input_805_cast_fp16 = concat(axis = var_58034, interleave = input_805_interleave_0, values = (var_60633_cast_fp16, var_60635_cast_fp16, var_60637_cast_fp16, var_60639_cast_fp16, var_60641_cast_fp16, var_60643_cast_fp16, var_60645_cast_fp16, var_60647_cast_fp16, var_60649_cast_fp16, var_60651_cast_fp16))[name = tensor("input_805_cast_fp16")]; + tensor var_60657 = const()[name = tensor("op_60657"), val = tensor([1, 1])]; + tensor var_60659 = const()[name = tensor("op_60659"), val = tensor([1, 1])]; + tensor var_60661_pad_type_0 = const()[name = tensor("op_60661_pad_type_0"), val = tensor("custom")]; + tensor var_60661_pad_0 = const()[name = tensor("op_60661_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5066939200)))]; + tensor up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5067758464)))]; + tensor var_60661_cast_fp16 = conv(bias = up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_60659, groups = var_58034, pad = var_60661_pad_0, pad_type = var_60661_pad_type_0, strides = var_60657, weight = up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_out_0_weight_to_fp16, x = input_805_cast_fp16)[name = tensor("op_60661_cast_fp16")]; + tensor inputs_411_cast_fp16 = add(x = var_60661_cast_fp16, y = inputs_409_cast_fp16)[name = tensor("inputs_411_cast_fp16")]; + tensor hidden_states_563_axes_0 = const()[name = tensor("hidden_states_563_axes_0"), val = tensor([1])]; + tensor hidden_states_563_gamma_0_to_fp16 = const()[name = tensor("hidden_states_563_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5067759808)))]; + tensor hidden_states_563_beta_0_to_fp16 = const()[name = tensor("hidden_states_563_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5067761152)))]; + tensor var_60671_to_fp16 = const()[name = tensor("op_60671_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_563_cast_fp16 = layer_norm(axes = hidden_states_563_axes_0, beta = hidden_states_563_beta_0_to_fp16, epsilon = var_60671_to_fp16, gamma = hidden_states_563_gamma_0_to_fp16, x = inputs_411_cast_fp16)[name = tensor("hidden_states_563_cast_fp16")]; + tensor var_60686 = const()[name = tensor("op_60686"), val = tensor([1, 1])]; + tensor var_60688 = const()[name = tensor("op_60688"), val = tensor([1, 1])]; + tensor q_275_pad_type_0 = const()[name = tensor("q_275_pad_type_0"), val = tensor("custom")]; + tensor q_275_pad_0 = const()[name = tensor("q_275_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5067762496)))]; + tensor q_275_cast_fp16 = conv(dilations = var_60688, groups = var_58034, pad = q_275_pad_0, pad_type = q_275_pad_type_0, strides = var_60686, weight = up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_q_weight_to_fp16, x = hidden_states_563_cast_fp16)[name = tensor("q_275_cast_fp16")]; + tensor var_60692 = const()[name = tensor("op_60692"), val = tensor([1, 1])]; + tensor var_60694 = const()[name = tensor("op_60694"), val = tensor([1, 1])]; + tensor k_549_pad_type_0 = const()[name = tensor("k_549_pad_type_0"), val = tensor("custom")]; + tensor k_549_pad_0 = const()[name = tensor("k_549_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5068581760)))]; + tensor k_549_cast_fp16 = conv(dilations = var_60694, groups = var_58034, pad = k_549_pad_0, pad_type = k_549_pad_type_0, strides = var_60692, weight = up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_549_cast_fp16")]; + tensor var_60698 = const()[name = tensor("op_60698"), val = tensor([1, 1])]; + tensor var_60700 = const()[name = tensor("op_60700"), val = tensor([1, 1])]; + tensor v_275_pad_type_0 = const()[name = tensor("v_275_pad_type_0"), val = tensor("custom")]; + tensor v_275_pad_0 = const()[name = tensor("v_275_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5071203264)))]; + tensor v_275_cast_fp16 = conv(dilations = var_60700, groups = var_58034, pad = v_275_pad_0, pad_type = v_275_pad_type_0, strides = var_60698, weight = up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_275_cast_fp16")]; + tensor var_60704_begin_0 = const()[name = tensor("op_60704_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_60704_end_0 = const()[name = tensor("op_60704_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_60704_end_mask_0 = const()[name = tensor("op_60704_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_60704_cast_fp16 = slice_by_index(begin = var_60704_begin_0, end = var_60704_end_0, end_mask = var_60704_end_mask_0, x = q_275_cast_fp16)[name = tensor("op_60704_cast_fp16")]; + tensor var_60708_begin_0 = const()[name = tensor("op_60708_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_60708_end_0 = const()[name = tensor("op_60708_end_0"), val = tensor([2, 128, 1, 4096])]; + tensor var_60708_end_mask_0 = const()[name = tensor("op_60708_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_60708_cast_fp16 = slice_by_index(begin = var_60708_begin_0, end = var_60708_end_0, end_mask = var_60708_end_mask_0, x = q_275_cast_fp16)[name = tensor("op_60708_cast_fp16")]; + tensor var_60712_begin_0 = const()[name = tensor("op_60712_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_60712_end_0 = const()[name = tensor("op_60712_end_0"), val = tensor([2, 192, 1, 4096])]; + tensor var_60712_end_mask_0 = const()[name = tensor("op_60712_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_60712_cast_fp16 = slice_by_index(begin = var_60712_begin_0, end = var_60712_end_0, end_mask = var_60712_end_mask_0, x = q_275_cast_fp16)[name = tensor("op_60712_cast_fp16")]; + tensor var_60716_begin_0 = const()[name = tensor("op_60716_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_60716_end_0 = const()[name = tensor("op_60716_end_0"), val = tensor([2, 256, 1, 4096])]; + tensor var_60716_end_mask_0 = const()[name = tensor("op_60716_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_60716_cast_fp16 = slice_by_index(begin = var_60716_begin_0, end = var_60716_end_0, end_mask = var_60716_end_mask_0, x = q_275_cast_fp16)[name = tensor("op_60716_cast_fp16")]; + tensor var_60720_begin_0 = const()[name = tensor("op_60720_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_60720_end_0 = const()[name = tensor("op_60720_end_0"), val = tensor([2, 320, 1, 4096])]; + tensor var_60720_end_mask_0 = const()[name = tensor("op_60720_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_60720_cast_fp16 = slice_by_index(begin = var_60720_begin_0, end = var_60720_end_0, end_mask = var_60720_end_mask_0, x = q_275_cast_fp16)[name = tensor("op_60720_cast_fp16")]; + tensor var_60724_begin_0 = const()[name = tensor("op_60724_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_60724_end_0 = const()[name = tensor("op_60724_end_0"), val = tensor([2, 384, 1, 4096])]; + tensor var_60724_end_mask_0 = const()[name = tensor("op_60724_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_60724_cast_fp16 = slice_by_index(begin = var_60724_begin_0, end = var_60724_end_0, end_mask = var_60724_end_mask_0, x = q_275_cast_fp16)[name = tensor("op_60724_cast_fp16")]; + tensor var_60728_begin_0 = const()[name = tensor("op_60728_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_60728_end_0 = const()[name = tensor("op_60728_end_0"), val = tensor([2, 448, 1, 4096])]; + tensor var_60728_end_mask_0 = const()[name = tensor("op_60728_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_60728_cast_fp16 = slice_by_index(begin = var_60728_begin_0, end = var_60728_end_0, end_mask = var_60728_end_mask_0, x = q_275_cast_fp16)[name = tensor("op_60728_cast_fp16")]; + tensor var_60732_begin_0 = const()[name = tensor("op_60732_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_60732_end_0 = const()[name = tensor("op_60732_end_0"), val = tensor([2, 512, 1, 4096])]; + tensor var_60732_end_mask_0 = const()[name = tensor("op_60732_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_60732_cast_fp16 = slice_by_index(begin = var_60732_begin_0, end = var_60732_end_0, end_mask = var_60732_end_mask_0, x = q_275_cast_fp16)[name = tensor("op_60732_cast_fp16")]; + tensor var_60736_begin_0 = const()[name = tensor("op_60736_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_60736_end_0 = const()[name = tensor("op_60736_end_0"), val = tensor([2, 576, 1, 4096])]; + tensor var_60736_end_mask_0 = const()[name = tensor("op_60736_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_60736_cast_fp16 = slice_by_index(begin = var_60736_begin_0, end = var_60736_end_0, end_mask = var_60736_end_mask_0, x = q_275_cast_fp16)[name = tensor("op_60736_cast_fp16")]; + tensor var_60740_begin_0 = const()[name = tensor("op_60740_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_60740_end_0 = const()[name = tensor("op_60740_end_0"), val = tensor([2, 640, 1, 4096])]; + tensor var_60740_end_mask_0 = const()[name = tensor("op_60740_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_60740_cast_fp16 = slice_by_index(begin = var_60740_begin_0, end = var_60740_end_0, end_mask = var_60740_end_mask_0, x = q_275_cast_fp16)[name = tensor("op_60740_cast_fp16")]; + tensor k_551_perm_0 = const()[name = tensor("k_551_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_60747_begin_0 = const()[name = tensor("op_60747_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_60747_end_0 = const()[name = tensor("op_60747_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_60747_end_mask_0 = const()[name = tensor("op_60747_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_551_cast_fp16 = transpose(perm = k_551_perm_0, x = k_549_cast_fp16)[name = tensor("transpose_2")]; + tensor var_60747_cast_fp16 = slice_by_index(begin = var_60747_begin_0, end = var_60747_end_0, end_mask = var_60747_end_mask_0, x = k_551_cast_fp16)[name = tensor("op_60747_cast_fp16")]; + tensor var_60751_begin_0 = const()[name = tensor("op_60751_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_60751_end_0 = const()[name = tensor("op_60751_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_60751_end_mask_0 = const()[name = tensor("op_60751_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_60751_cast_fp16 = slice_by_index(begin = var_60751_begin_0, end = var_60751_end_0, end_mask = var_60751_end_mask_0, x = k_551_cast_fp16)[name = tensor("op_60751_cast_fp16")]; + tensor var_60755_begin_0 = const()[name = tensor("op_60755_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_60755_end_0 = const()[name = tensor("op_60755_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_60755_end_mask_0 = const()[name = tensor("op_60755_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_60755_cast_fp16 = slice_by_index(begin = var_60755_begin_0, end = var_60755_end_0, end_mask = var_60755_end_mask_0, x = k_551_cast_fp16)[name = tensor("op_60755_cast_fp16")]; + tensor var_60759_begin_0 = const()[name = tensor("op_60759_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_60759_end_0 = const()[name = tensor("op_60759_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_60759_end_mask_0 = const()[name = tensor("op_60759_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_60759_cast_fp16 = slice_by_index(begin = var_60759_begin_0, end = var_60759_end_0, end_mask = var_60759_end_mask_0, x = k_551_cast_fp16)[name = tensor("op_60759_cast_fp16")]; + tensor var_60763_begin_0 = const()[name = tensor("op_60763_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_60763_end_0 = const()[name = tensor("op_60763_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_60763_end_mask_0 = const()[name = tensor("op_60763_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_60763_cast_fp16 = slice_by_index(begin = var_60763_begin_0, end = var_60763_end_0, end_mask = var_60763_end_mask_0, x = k_551_cast_fp16)[name = tensor("op_60763_cast_fp16")]; + tensor var_60767_begin_0 = const()[name = tensor("op_60767_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_60767_end_0 = const()[name = tensor("op_60767_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_60767_end_mask_0 = const()[name = tensor("op_60767_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_60767_cast_fp16 = slice_by_index(begin = var_60767_begin_0, end = var_60767_end_0, end_mask = var_60767_end_mask_0, x = k_551_cast_fp16)[name = tensor("op_60767_cast_fp16")]; + tensor var_60771_begin_0 = const()[name = tensor("op_60771_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_60771_end_0 = const()[name = tensor("op_60771_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_60771_end_mask_0 = const()[name = tensor("op_60771_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_60771_cast_fp16 = slice_by_index(begin = var_60771_begin_0, end = var_60771_end_0, end_mask = var_60771_end_mask_0, x = k_551_cast_fp16)[name = tensor("op_60771_cast_fp16")]; + tensor var_60775_begin_0 = const()[name = tensor("op_60775_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_60775_end_0 = const()[name = tensor("op_60775_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_60775_end_mask_0 = const()[name = tensor("op_60775_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_60775_cast_fp16 = slice_by_index(begin = var_60775_begin_0, end = var_60775_end_0, end_mask = var_60775_end_mask_0, x = k_551_cast_fp16)[name = tensor("op_60775_cast_fp16")]; + tensor var_60779_begin_0 = const()[name = tensor("op_60779_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_60779_end_0 = const()[name = tensor("op_60779_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_60779_end_mask_0 = const()[name = tensor("op_60779_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_60779_cast_fp16 = slice_by_index(begin = var_60779_begin_0, end = var_60779_end_0, end_mask = var_60779_end_mask_0, x = k_551_cast_fp16)[name = tensor("op_60779_cast_fp16")]; + tensor var_60783_begin_0 = const()[name = tensor("op_60783_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_60783_end_0 = const()[name = tensor("op_60783_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_60783_end_mask_0 = const()[name = tensor("op_60783_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_60783_cast_fp16 = slice_by_index(begin = var_60783_begin_0, end = var_60783_end_0, end_mask = var_60783_end_mask_0, x = k_551_cast_fp16)[name = tensor("op_60783_cast_fp16")]; + tensor var_60785_begin_0 = const()[name = tensor("op_60785_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_60785_end_0 = const()[name = tensor("op_60785_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_60785_end_mask_0 = const()[name = tensor("op_60785_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_60785_cast_fp16 = slice_by_index(begin = var_60785_begin_0, end = var_60785_end_0, end_mask = var_60785_end_mask_0, x = v_275_cast_fp16)[name = tensor("op_60785_cast_fp16")]; + tensor var_60789_begin_0 = const()[name = tensor("op_60789_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_60789_end_0 = const()[name = tensor("op_60789_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_60789_end_mask_0 = const()[name = tensor("op_60789_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_60789_cast_fp16 = slice_by_index(begin = var_60789_begin_0, end = var_60789_end_0, end_mask = var_60789_end_mask_0, x = v_275_cast_fp16)[name = tensor("op_60789_cast_fp16")]; + tensor var_60793_begin_0 = const()[name = tensor("op_60793_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_60793_end_0 = const()[name = tensor("op_60793_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_60793_end_mask_0 = const()[name = tensor("op_60793_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_60793_cast_fp16 = slice_by_index(begin = var_60793_begin_0, end = var_60793_end_0, end_mask = var_60793_end_mask_0, x = v_275_cast_fp16)[name = tensor("op_60793_cast_fp16")]; + tensor var_60797_begin_0 = const()[name = tensor("op_60797_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_60797_end_0 = const()[name = tensor("op_60797_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_60797_end_mask_0 = const()[name = tensor("op_60797_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_60797_cast_fp16 = slice_by_index(begin = var_60797_begin_0, end = var_60797_end_0, end_mask = var_60797_end_mask_0, x = v_275_cast_fp16)[name = tensor("op_60797_cast_fp16")]; + tensor var_60801_begin_0 = const()[name = tensor("op_60801_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_60801_end_0 = const()[name = tensor("op_60801_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_60801_end_mask_0 = const()[name = tensor("op_60801_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_60801_cast_fp16 = slice_by_index(begin = var_60801_begin_0, end = var_60801_end_0, end_mask = var_60801_end_mask_0, x = v_275_cast_fp16)[name = tensor("op_60801_cast_fp16")]; + tensor var_60805_begin_0 = const()[name = tensor("op_60805_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_60805_end_0 = const()[name = tensor("op_60805_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_60805_end_mask_0 = const()[name = tensor("op_60805_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_60805_cast_fp16 = slice_by_index(begin = var_60805_begin_0, end = var_60805_end_0, end_mask = var_60805_end_mask_0, x = v_275_cast_fp16)[name = tensor("op_60805_cast_fp16")]; + tensor var_60809_begin_0 = const()[name = tensor("op_60809_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_60809_end_0 = const()[name = tensor("op_60809_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_60809_end_mask_0 = const()[name = tensor("op_60809_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_60809_cast_fp16 = slice_by_index(begin = var_60809_begin_0, end = var_60809_end_0, end_mask = var_60809_end_mask_0, x = v_275_cast_fp16)[name = tensor("op_60809_cast_fp16")]; + tensor var_60813_begin_0 = const()[name = tensor("op_60813_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_60813_end_0 = const()[name = tensor("op_60813_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_60813_end_mask_0 = const()[name = tensor("op_60813_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_60813_cast_fp16 = slice_by_index(begin = var_60813_begin_0, end = var_60813_end_0, end_mask = var_60813_end_mask_0, x = v_275_cast_fp16)[name = tensor("op_60813_cast_fp16")]; + tensor var_60817_begin_0 = const()[name = tensor("op_60817_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_60817_end_0 = const()[name = tensor("op_60817_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_60817_end_mask_0 = const()[name = tensor("op_60817_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_60817_cast_fp16 = slice_by_index(begin = var_60817_begin_0, end = var_60817_end_0, end_mask = var_60817_end_mask_0, x = v_275_cast_fp16)[name = tensor("op_60817_cast_fp16")]; + tensor var_60821_begin_0 = const()[name = tensor("op_60821_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_60821_end_0 = const()[name = tensor("op_60821_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_60821_end_mask_0 = const()[name = tensor("op_60821_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_60821_cast_fp16 = slice_by_index(begin = var_60821_begin_0, end = var_60821_end_0, end_mask = var_60821_end_mask_0, x = v_275_cast_fp16)[name = tensor("op_60821_cast_fp16")]; + tensor var_60825_equation_0 = const()[name = tensor("op_60825_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_60825_cast_fp16 = einsum(equation = var_60825_equation_0, values = (var_60747_cast_fp16, var_60704_cast_fp16))[name = tensor("op_60825_cast_fp16")]; + tensor var_60826_to_fp16 = const()[name = tensor("op_60826_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5141_cast_fp16 = mul(x = var_60825_cast_fp16, y = var_60826_to_fp16)[name = tensor("aw_5141_cast_fp16")]; + tensor var_60829_equation_0 = const()[name = tensor("op_60829_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_60829_cast_fp16 = einsum(equation = var_60829_equation_0, values = (var_60751_cast_fp16, var_60708_cast_fp16))[name = tensor("op_60829_cast_fp16")]; + tensor var_60830_to_fp16 = const()[name = tensor("op_60830_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5143_cast_fp16 = mul(x = var_60829_cast_fp16, y = var_60830_to_fp16)[name = tensor("aw_5143_cast_fp16")]; + tensor var_60833_equation_0 = const()[name = tensor("op_60833_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_60833_cast_fp16 = einsum(equation = var_60833_equation_0, values = (var_60755_cast_fp16, var_60712_cast_fp16))[name = tensor("op_60833_cast_fp16")]; + tensor var_60834_to_fp16 = const()[name = tensor("op_60834_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5145_cast_fp16 = mul(x = var_60833_cast_fp16, y = var_60834_to_fp16)[name = tensor("aw_5145_cast_fp16")]; + tensor var_60837_equation_0 = const()[name = tensor("op_60837_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_60837_cast_fp16 = einsum(equation = var_60837_equation_0, values = (var_60759_cast_fp16, var_60716_cast_fp16))[name = tensor("op_60837_cast_fp16")]; + tensor var_60838_to_fp16 = const()[name = tensor("op_60838_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5147_cast_fp16 = mul(x = var_60837_cast_fp16, y = var_60838_to_fp16)[name = tensor("aw_5147_cast_fp16")]; + tensor var_60841_equation_0 = const()[name = tensor("op_60841_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_60841_cast_fp16 = einsum(equation = var_60841_equation_0, values = (var_60763_cast_fp16, var_60720_cast_fp16))[name = tensor("op_60841_cast_fp16")]; + tensor var_60842_to_fp16 = const()[name = tensor("op_60842_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5149_cast_fp16 = mul(x = var_60841_cast_fp16, y = var_60842_to_fp16)[name = tensor("aw_5149_cast_fp16")]; + tensor var_60845_equation_0 = const()[name = tensor("op_60845_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_60845_cast_fp16 = einsum(equation = var_60845_equation_0, values = (var_60767_cast_fp16, var_60724_cast_fp16))[name = tensor("op_60845_cast_fp16")]; + tensor var_60846_to_fp16 = const()[name = tensor("op_60846_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5151_cast_fp16 = mul(x = var_60845_cast_fp16, y = var_60846_to_fp16)[name = tensor("aw_5151_cast_fp16")]; + tensor var_60849_equation_0 = const()[name = tensor("op_60849_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_60849_cast_fp16 = einsum(equation = var_60849_equation_0, values = (var_60771_cast_fp16, var_60728_cast_fp16))[name = tensor("op_60849_cast_fp16")]; + tensor var_60850_to_fp16 = const()[name = tensor("op_60850_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5153_cast_fp16 = mul(x = var_60849_cast_fp16, y = var_60850_to_fp16)[name = tensor("aw_5153_cast_fp16")]; + tensor var_60853_equation_0 = const()[name = tensor("op_60853_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_60853_cast_fp16 = einsum(equation = var_60853_equation_0, values = (var_60775_cast_fp16, var_60732_cast_fp16))[name = tensor("op_60853_cast_fp16")]; + tensor var_60854_to_fp16 = const()[name = tensor("op_60854_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5155_cast_fp16 = mul(x = var_60853_cast_fp16, y = var_60854_to_fp16)[name = tensor("aw_5155_cast_fp16")]; + tensor var_60857_equation_0 = const()[name = tensor("op_60857_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_60857_cast_fp16 = einsum(equation = var_60857_equation_0, values = (var_60779_cast_fp16, var_60736_cast_fp16))[name = tensor("op_60857_cast_fp16")]; + tensor var_60858_to_fp16 = const()[name = tensor("op_60858_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5157_cast_fp16 = mul(x = var_60857_cast_fp16, y = var_60858_to_fp16)[name = tensor("aw_5157_cast_fp16")]; + tensor var_60861_equation_0 = const()[name = tensor("op_60861_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_60861_cast_fp16 = einsum(equation = var_60861_equation_0, values = (var_60783_cast_fp16, var_60740_cast_fp16))[name = tensor("op_60861_cast_fp16")]; + tensor var_60862_to_fp16 = const()[name = tensor("op_60862_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5159_cast_fp16 = mul(x = var_60861_cast_fp16, y = var_60862_to_fp16)[name = tensor("aw_5159_cast_fp16")]; + tensor var_60864_cast_fp16 = softmax(axis = var_58034, x = aw_5141_cast_fp16)[name = tensor("op_60864_cast_fp16")]; + tensor var_60865_cast_fp16 = softmax(axis = var_58034, x = aw_5143_cast_fp16)[name = tensor("op_60865_cast_fp16")]; + tensor var_60866_cast_fp16 = softmax(axis = var_58034, x = aw_5145_cast_fp16)[name = tensor("op_60866_cast_fp16")]; + tensor var_60867_cast_fp16 = softmax(axis = var_58034, x = aw_5147_cast_fp16)[name = tensor("op_60867_cast_fp16")]; + tensor var_60868_cast_fp16 = softmax(axis = var_58034, x = aw_5149_cast_fp16)[name = tensor("op_60868_cast_fp16")]; + tensor var_60869_cast_fp16 = softmax(axis = var_58034, x = aw_5151_cast_fp16)[name = tensor("op_60869_cast_fp16")]; + tensor var_60870_cast_fp16 = softmax(axis = var_58034, x = aw_5153_cast_fp16)[name = tensor("op_60870_cast_fp16")]; + tensor var_60871_cast_fp16 = softmax(axis = var_58034, x = aw_5155_cast_fp16)[name = tensor("op_60871_cast_fp16")]; + tensor var_60872_cast_fp16 = softmax(axis = var_58034, x = aw_5157_cast_fp16)[name = tensor("op_60872_cast_fp16")]; + tensor var_60873_cast_fp16 = softmax(axis = var_58034, x = aw_5159_cast_fp16)[name = tensor("op_60873_cast_fp16")]; + tensor var_60875_equation_0 = const()[name = tensor("op_60875_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_60875_cast_fp16 = einsum(equation = var_60875_equation_0, values = (var_60785_cast_fp16, var_60864_cast_fp16))[name = tensor("op_60875_cast_fp16")]; + tensor var_60877_equation_0 = const()[name = tensor("op_60877_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_60877_cast_fp16 = einsum(equation = var_60877_equation_0, values = (var_60789_cast_fp16, var_60865_cast_fp16))[name = tensor("op_60877_cast_fp16")]; + tensor var_60879_equation_0 = const()[name = tensor("op_60879_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_60879_cast_fp16 = einsum(equation = var_60879_equation_0, values = (var_60793_cast_fp16, var_60866_cast_fp16))[name = tensor("op_60879_cast_fp16")]; + tensor var_60881_equation_0 = const()[name = tensor("op_60881_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_60881_cast_fp16 = einsum(equation = var_60881_equation_0, values = (var_60797_cast_fp16, var_60867_cast_fp16))[name = tensor("op_60881_cast_fp16")]; + tensor var_60883_equation_0 = const()[name = tensor("op_60883_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_60883_cast_fp16 = einsum(equation = var_60883_equation_0, values = (var_60801_cast_fp16, var_60868_cast_fp16))[name = tensor("op_60883_cast_fp16")]; + tensor var_60885_equation_0 = const()[name = tensor("op_60885_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_60885_cast_fp16 = einsum(equation = var_60885_equation_0, values = (var_60805_cast_fp16, var_60869_cast_fp16))[name = tensor("op_60885_cast_fp16")]; + tensor var_60887_equation_0 = const()[name = tensor("op_60887_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_60887_cast_fp16 = einsum(equation = var_60887_equation_0, values = (var_60809_cast_fp16, var_60870_cast_fp16))[name = tensor("op_60887_cast_fp16")]; + tensor var_60889_equation_0 = const()[name = tensor("op_60889_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_60889_cast_fp16 = einsum(equation = var_60889_equation_0, values = (var_60813_cast_fp16, var_60871_cast_fp16))[name = tensor("op_60889_cast_fp16")]; + tensor var_60891_equation_0 = const()[name = tensor("op_60891_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_60891_cast_fp16 = einsum(equation = var_60891_equation_0, values = (var_60817_cast_fp16, var_60872_cast_fp16))[name = tensor("op_60891_cast_fp16")]; + tensor var_60893_equation_0 = const()[name = tensor("op_60893_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_60893_cast_fp16 = einsum(equation = var_60893_equation_0, values = (var_60821_cast_fp16, var_60873_cast_fp16))[name = tensor("op_60893_cast_fp16")]; + tensor input_807_interleave_0 = const()[name = tensor("input_807_interleave_0"), val = tensor(false)]; + tensor input_807_cast_fp16 = concat(axis = var_58034, interleave = input_807_interleave_0, values = (var_60875_cast_fp16, var_60877_cast_fp16, var_60879_cast_fp16, var_60881_cast_fp16, var_60883_cast_fp16, var_60885_cast_fp16, var_60887_cast_fp16, var_60889_cast_fp16, var_60891_cast_fp16, var_60893_cast_fp16))[name = tensor("input_807_cast_fp16")]; + tensor var_60899 = const()[name = tensor("op_60899"), val = tensor([1, 1])]; + tensor var_60901 = const()[name = tensor("op_60901"), val = tensor([1, 1])]; + tensor var_60903_pad_type_0 = const()[name = tensor("op_60903_pad_type_0"), val = tensor("custom")]; + tensor var_60903_pad_0 = const()[name = tensor("op_60903_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5073824768)))]; + tensor up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5074644032)))]; + tensor var_60903_cast_fp16 = conv(bias = up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_60901, groups = var_58034, pad = var_60903_pad_0, pad_type = var_60903_pad_type_0, strides = var_60899, weight = up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_out_0_weight_to_fp16, x = input_807_cast_fp16)[name = tensor("op_60903_cast_fp16")]; + tensor inputs_413_cast_fp16 = add(x = var_60903_cast_fp16, y = inputs_411_cast_fp16)[name = tensor("inputs_413_cast_fp16")]; + tensor input_809_axes_0 = const()[name = tensor("input_809_axes_0"), val = tensor([1])]; + tensor input_809_gamma_0_to_fp16 = const()[name = tensor("input_809_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5074645376)))]; + tensor input_809_beta_0_to_fp16 = const()[name = tensor("input_809_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5074646720)))]; + tensor var_60913_to_fp16 = const()[name = tensor("op_60913_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_809_cast_fp16 = layer_norm(axes = input_809_axes_0, beta = input_809_beta_0_to_fp16, epsilon = var_60913_to_fp16, gamma = input_809_gamma_0_to_fp16, x = inputs_413_cast_fp16)[name = tensor("input_809_cast_fp16")]; + tensor var_60929 = const()[name = tensor("op_60929"), val = tensor([1, 1])]; + tensor var_60931 = const()[name = tensor("op_60931"), val = tensor([1, 1])]; + tensor var_60933_pad_type_0 = const()[name = tensor("op_60933_pad_type_0"), val = tensor("custom")]; + tensor var_60933_pad_0 = const()[name = tensor("op_60933_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_2_transformer_blocks_0_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_transformer_blocks_0_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5074648064)))]; + tensor up_blocks_1_attentions_2_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5081201728)))]; + tensor var_60933_cast_fp16 = conv(bias = up_blocks_1_attentions_2_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_60931, groups = var_58034, pad = var_60933_pad_0, pad_type = var_60933_pad_type_0, strides = var_60929, weight = up_blocks_1_attentions_2_transformer_blocks_0_ff_net_0_proj_weight_to_fp16, x = input_809_cast_fp16)[name = tensor("op_60933_cast_fp16")]; + tensor var_60934_split_sizes_0 = const()[name = tensor("op_60934_split_sizes_0"), val = tensor([2560, 2560])]; + tensor var_60934_axis_0 = const()[name = tensor("op_60934_axis_0"), val = tensor(1)]; + tensor var_60934_cast_fp16_0, tensor var_60934_cast_fp16_1 = split(axis = var_60934_axis_0, split_sizes = var_60934_split_sizes_0, x = var_60933_cast_fp16)[name = tensor("op_60934_cast_fp16")]; + tensor var_60936_mode_0 = const()[name = tensor("op_60936_mode_0"), val = tensor("EXACT")]; + tensor var_60936_cast_fp16 = gelu(mode = var_60936_mode_0, x = var_60934_cast_fp16_1)[name = tensor("op_60936_cast_fp16")]; + tensor input_811_cast_fp16 = mul(x = var_60934_cast_fp16_0, y = var_60936_cast_fp16)[name = tensor("input_811_cast_fp16")]; + tensor var_60940 = const()[name = tensor("op_60940"), val = tensor([1, 1])]; + tensor var_60942 = const()[name = tensor("op_60942"), val = tensor([1, 1])]; + tensor var_60944_pad_type_0 = const()[name = tensor("op_60944_pad_type_0"), val = tensor("custom")]; + tensor var_60944_pad_0 = const()[name = tensor("op_60944_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_2_transformer_blocks_0_ff_net_2_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_transformer_blocks_0_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5081212032)))]; + tensor up_blocks_1_attentions_2_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5084488896)))]; + tensor var_60944_cast_fp16 = conv(bias = up_blocks_1_attentions_2_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_60942, groups = var_58034, pad = var_60944_pad_0, pad_type = var_60944_pad_type_0, strides = var_60940, weight = up_blocks_1_attentions_2_transformer_blocks_0_ff_net_2_weight_to_fp16, x = input_811_cast_fp16)[name = tensor("op_60944_cast_fp16")]; + tensor inputs_415_cast_fp16 = add(x = var_60944_cast_fp16, y = inputs_413_cast_fp16)[name = tensor("inputs_415_cast_fp16")]; + tensor hidden_states_567_axes_0 = const()[name = tensor("hidden_states_567_axes_0"), val = tensor([1])]; + tensor hidden_states_567_gamma_0_to_fp16 = const()[name = tensor("hidden_states_567_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5084490240)))]; + tensor hidden_states_567_beta_0_to_fp16 = const()[name = tensor("hidden_states_567_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5084491584)))]; + tensor var_60960_to_fp16 = const()[name = tensor("op_60960_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_567_cast_fp16 = layer_norm(axes = hidden_states_567_axes_0, beta = hidden_states_567_beta_0_to_fp16, epsilon = var_60960_to_fp16, gamma = hidden_states_567_gamma_0_to_fp16, x = inputs_415_cast_fp16)[name = tensor("hidden_states_567_cast_fp16")]; + tensor var_60975 = const()[name = tensor("op_60975"), val = tensor([1, 1])]; + tensor var_60977 = const()[name = tensor("op_60977"), val = tensor([1, 1])]; + tensor q_277_pad_type_0 = const()[name = tensor("q_277_pad_type_0"), val = tensor("custom")]; + tensor q_277_pad_0 = const()[name = tensor("q_277_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_2_transformer_blocks_1_attn1_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_transformer_blocks_1_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5084492928)))]; + tensor q_277_cast_fp16 = conv(dilations = var_60977, groups = var_58034, pad = q_277_pad_0, pad_type = q_277_pad_type_0, strides = var_60975, weight = up_blocks_1_attentions_2_transformer_blocks_1_attn1_to_q_weight_to_fp16, x = hidden_states_567_cast_fp16)[name = tensor("q_277_cast_fp16")]; + tensor var_60981 = const()[name = tensor("op_60981"), val = tensor([1, 1])]; + tensor var_60983 = const()[name = tensor("op_60983"), val = tensor([1, 1])]; + tensor k_553_pad_type_0 = const()[name = tensor("k_553_pad_type_0"), val = tensor("custom")]; + tensor k_553_pad_0 = const()[name = tensor("k_553_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_2_transformer_blocks_1_attn1_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_transformer_blocks_1_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5085312192)))]; + tensor k_553_cast_fp16 = conv(dilations = var_60983, groups = var_58034, pad = k_553_pad_0, pad_type = k_553_pad_type_0, strides = var_60981, weight = up_blocks_1_attentions_2_transformer_blocks_1_attn1_to_k_weight_to_fp16, x = hidden_states_567_cast_fp16)[name = tensor("k_553_cast_fp16")]; + tensor var_60987 = const()[name = tensor("op_60987"), val = tensor([1, 1])]; + tensor var_60989 = const()[name = tensor("op_60989"), val = tensor([1, 1])]; + tensor v_277_pad_type_0 = const()[name = tensor("v_277_pad_type_0"), val = tensor("custom")]; + tensor v_277_pad_0 = const()[name = tensor("v_277_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_2_transformer_blocks_1_attn1_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_transformer_blocks_1_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5086131456)))]; + tensor v_277_cast_fp16 = conv(dilations = var_60989, groups = var_58034, pad = v_277_pad_0, pad_type = v_277_pad_type_0, strides = var_60987, weight = up_blocks_1_attentions_2_transformer_blocks_1_attn1_to_v_weight_to_fp16, x = hidden_states_567_cast_fp16)[name = tensor("v_277_cast_fp16")]; + tensor var_60993_begin_0 = const()[name = tensor("op_60993_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_60993_end_0 = const()[name = tensor("op_60993_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_60993_end_mask_0 = const()[name = tensor("op_60993_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_60993_cast_fp16 = slice_by_index(begin = var_60993_begin_0, end = var_60993_end_0, end_mask = var_60993_end_mask_0, x = q_277_cast_fp16)[name = tensor("op_60993_cast_fp16")]; + tensor var_60997_begin_0 = const()[name = tensor("op_60997_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_60997_end_0 = const()[name = tensor("op_60997_end_0"), val = tensor([2, 128, 1, 4096])]; + tensor var_60997_end_mask_0 = const()[name = tensor("op_60997_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_60997_cast_fp16 = slice_by_index(begin = var_60997_begin_0, end = var_60997_end_0, end_mask = var_60997_end_mask_0, x = q_277_cast_fp16)[name = tensor("op_60997_cast_fp16")]; + tensor var_61001_begin_0 = const()[name = tensor("op_61001_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_61001_end_0 = const()[name = tensor("op_61001_end_0"), val = tensor([2, 192, 1, 4096])]; + tensor var_61001_end_mask_0 = const()[name = tensor("op_61001_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_61001_cast_fp16 = slice_by_index(begin = var_61001_begin_0, end = var_61001_end_0, end_mask = var_61001_end_mask_0, x = q_277_cast_fp16)[name = tensor("op_61001_cast_fp16")]; + tensor var_61005_begin_0 = const()[name = tensor("op_61005_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_61005_end_0 = const()[name = tensor("op_61005_end_0"), val = tensor([2, 256, 1, 4096])]; + tensor var_61005_end_mask_0 = const()[name = tensor("op_61005_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_61005_cast_fp16 = slice_by_index(begin = var_61005_begin_0, end = var_61005_end_0, end_mask = var_61005_end_mask_0, x = q_277_cast_fp16)[name = tensor("op_61005_cast_fp16")]; + tensor var_61009_begin_0 = const()[name = tensor("op_61009_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_61009_end_0 = const()[name = tensor("op_61009_end_0"), val = tensor([2, 320, 1, 4096])]; + tensor var_61009_end_mask_0 = const()[name = tensor("op_61009_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_61009_cast_fp16 = slice_by_index(begin = var_61009_begin_0, end = var_61009_end_0, end_mask = var_61009_end_mask_0, x = q_277_cast_fp16)[name = tensor("op_61009_cast_fp16")]; + tensor var_61013_begin_0 = const()[name = tensor("op_61013_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_61013_end_0 = const()[name = tensor("op_61013_end_0"), val = tensor([2, 384, 1, 4096])]; + tensor var_61013_end_mask_0 = const()[name = tensor("op_61013_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_61013_cast_fp16 = slice_by_index(begin = var_61013_begin_0, end = var_61013_end_0, end_mask = var_61013_end_mask_0, x = q_277_cast_fp16)[name = tensor("op_61013_cast_fp16")]; + tensor var_61017_begin_0 = const()[name = tensor("op_61017_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_61017_end_0 = const()[name = tensor("op_61017_end_0"), val = tensor([2, 448, 1, 4096])]; + tensor var_61017_end_mask_0 = const()[name = tensor("op_61017_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_61017_cast_fp16 = slice_by_index(begin = var_61017_begin_0, end = var_61017_end_0, end_mask = var_61017_end_mask_0, x = q_277_cast_fp16)[name = tensor("op_61017_cast_fp16")]; + tensor var_61021_begin_0 = const()[name = tensor("op_61021_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_61021_end_0 = const()[name = tensor("op_61021_end_0"), val = tensor([2, 512, 1, 4096])]; + tensor var_61021_end_mask_0 = const()[name = tensor("op_61021_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_61021_cast_fp16 = slice_by_index(begin = var_61021_begin_0, end = var_61021_end_0, end_mask = var_61021_end_mask_0, x = q_277_cast_fp16)[name = tensor("op_61021_cast_fp16")]; + tensor var_61025_begin_0 = const()[name = tensor("op_61025_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_61025_end_0 = const()[name = tensor("op_61025_end_0"), val = tensor([2, 576, 1, 4096])]; + tensor var_61025_end_mask_0 = const()[name = tensor("op_61025_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_61025_cast_fp16 = slice_by_index(begin = var_61025_begin_0, end = var_61025_end_0, end_mask = var_61025_end_mask_0, x = q_277_cast_fp16)[name = tensor("op_61025_cast_fp16")]; + tensor var_61029_begin_0 = const()[name = tensor("op_61029_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_61029_end_0 = const()[name = tensor("op_61029_end_0"), val = tensor([2, 640, 1, 4096])]; + tensor var_61029_end_mask_0 = const()[name = tensor("op_61029_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_61029_cast_fp16 = slice_by_index(begin = var_61029_begin_0, end = var_61029_end_0, end_mask = var_61029_end_mask_0, x = q_277_cast_fp16)[name = tensor("op_61029_cast_fp16")]; + tensor k_555_perm_0 = const()[name = tensor("k_555_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_61036_begin_0 = const()[name = tensor("op_61036_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_61036_end_0 = const()[name = tensor("op_61036_end_0"), val = tensor([2, 4096, 1, 64])]; + tensor var_61036_end_mask_0 = const()[name = tensor("op_61036_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_555_cast_fp16 = transpose(perm = k_555_perm_0, x = k_553_cast_fp16)[name = tensor("transpose_1")]; + tensor var_61036_cast_fp16 = slice_by_index(begin = var_61036_begin_0, end = var_61036_end_0, end_mask = var_61036_end_mask_0, x = k_555_cast_fp16)[name = tensor("op_61036_cast_fp16")]; + tensor var_61040_begin_0 = const()[name = tensor("op_61040_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_61040_end_0 = const()[name = tensor("op_61040_end_0"), val = tensor([2, 4096, 1, 128])]; + tensor var_61040_end_mask_0 = const()[name = tensor("op_61040_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_61040_cast_fp16 = slice_by_index(begin = var_61040_begin_0, end = var_61040_end_0, end_mask = var_61040_end_mask_0, x = k_555_cast_fp16)[name = tensor("op_61040_cast_fp16")]; + tensor var_61044_begin_0 = const()[name = tensor("op_61044_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_61044_end_0 = const()[name = tensor("op_61044_end_0"), val = tensor([2, 4096, 1, 192])]; + tensor var_61044_end_mask_0 = const()[name = tensor("op_61044_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_61044_cast_fp16 = slice_by_index(begin = var_61044_begin_0, end = var_61044_end_0, end_mask = var_61044_end_mask_0, x = k_555_cast_fp16)[name = tensor("op_61044_cast_fp16")]; + tensor var_61048_begin_0 = const()[name = tensor("op_61048_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_61048_end_0 = const()[name = tensor("op_61048_end_0"), val = tensor([2, 4096, 1, 256])]; + tensor var_61048_end_mask_0 = const()[name = tensor("op_61048_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_61048_cast_fp16 = slice_by_index(begin = var_61048_begin_0, end = var_61048_end_0, end_mask = var_61048_end_mask_0, x = k_555_cast_fp16)[name = tensor("op_61048_cast_fp16")]; + tensor var_61052_begin_0 = const()[name = tensor("op_61052_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_61052_end_0 = const()[name = tensor("op_61052_end_0"), val = tensor([2, 4096, 1, 320])]; + tensor var_61052_end_mask_0 = const()[name = tensor("op_61052_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_61052_cast_fp16 = slice_by_index(begin = var_61052_begin_0, end = var_61052_end_0, end_mask = var_61052_end_mask_0, x = k_555_cast_fp16)[name = tensor("op_61052_cast_fp16")]; + tensor var_61056_begin_0 = const()[name = tensor("op_61056_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_61056_end_0 = const()[name = tensor("op_61056_end_0"), val = tensor([2, 4096, 1, 384])]; + tensor var_61056_end_mask_0 = const()[name = tensor("op_61056_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_61056_cast_fp16 = slice_by_index(begin = var_61056_begin_0, end = var_61056_end_0, end_mask = var_61056_end_mask_0, x = k_555_cast_fp16)[name = tensor("op_61056_cast_fp16")]; + tensor var_61060_begin_0 = const()[name = tensor("op_61060_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_61060_end_0 = const()[name = tensor("op_61060_end_0"), val = tensor([2, 4096, 1, 448])]; + tensor var_61060_end_mask_0 = const()[name = tensor("op_61060_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_61060_cast_fp16 = slice_by_index(begin = var_61060_begin_0, end = var_61060_end_0, end_mask = var_61060_end_mask_0, x = k_555_cast_fp16)[name = tensor("op_61060_cast_fp16")]; + tensor var_61064_begin_0 = const()[name = tensor("op_61064_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_61064_end_0 = const()[name = tensor("op_61064_end_0"), val = tensor([2, 4096, 1, 512])]; + tensor var_61064_end_mask_0 = const()[name = tensor("op_61064_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_61064_cast_fp16 = slice_by_index(begin = var_61064_begin_0, end = var_61064_end_0, end_mask = var_61064_end_mask_0, x = k_555_cast_fp16)[name = tensor("op_61064_cast_fp16")]; + tensor var_61068_begin_0 = const()[name = tensor("op_61068_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_61068_end_0 = const()[name = tensor("op_61068_end_0"), val = tensor([2, 4096, 1, 576])]; + tensor var_61068_end_mask_0 = const()[name = tensor("op_61068_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_61068_cast_fp16 = slice_by_index(begin = var_61068_begin_0, end = var_61068_end_0, end_mask = var_61068_end_mask_0, x = k_555_cast_fp16)[name = tensor("op_61068_cast_fp16")]; + tensor var_61072_begin_0 = const()[name = tensor("op_61072_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_61072_end_0 = const()[name = tensor("op_61072_end_0"), val = tensor([2, 4096, 1, 640])]; + tensor var_61072_end_mask_0 = const()[name = tensor("op_61072_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_61072_cast_fp16 = slice_by_index(begin = var_61072_begin_0, end = var_61072_end_0, end_mask = var_61072_end_mask_0, x = k_555_cast_fp16)[name = tensor("op_61072_cast_fp16")]; + tensor var_61074_begin_0 = const()[name = tensor("op_61074_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_61074_end_0 = const()[name = tensor("op_61074_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_61074_end_mask_0 = const()[name = tensor("op_61074_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_61074_cast_fp16 = slice_by_index(begin = var_61074_begin_0, end = var_61074_end_0, end_mask = var_61074_end_mask_0, x = v_277_cast_fp16)[name = tensor("op_61074_cast_fp16")]; + tensor var_61078_begin_0 = const()[name = tensor("op_61078_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_61078_end_0 = const()[name = tensor("op_61078_end_0"), val = tensor([2, 128, 1, 4096])]; + tensor var_61078_end_mask_0 = const()[name = tensor("op_61078_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_61078_cast_fp16 = slice_by_index(begin = var_61078_begin_0, end = var_61078_end_0, end_mask = var_61078_end_mask_0, x = v_277_cast_fp16)[name = tensor("op_61078_cast_fp16")]; + tensor var_61082_begin_0 = const()[name = tensor("op_61082_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_61082_end_0 = const()[name = tensor("op_61082_end_0"), val = tensor([2, 192, 1, 4096])]; + tensor var_61082_end_mask_0 = const()[name = tensor("op_61082_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_61082_cast_fp16 = slice_by_index(begin = var_61082_begin_0, end = var_61082_end_0, end_mask = var_61082_end_mask_0, x = v_277_cast_fp16)[name = tensor("op_61082_cast_fp16")]; + tensor var_61086_begin_0 = const()[name = tensor("op_61086_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_61086_end_0 = const()[name = tensor("op_61086_end_0"), val = tensor([2, 256, 1, 4096])]; + tensor var_61086_end_mask_0 = const()[name = tensor("op_61086_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_61086_cast_fp16 = slice_by_index(begin = var_61086_begin_0, end = var_61086_end_0, end_mask = var_61086_end_mask_0, x = v_277_cast_fp16)[name = tensor("op_61086_cast_fp16")]; + tensor var_61090_begin_0 = const()[name = tensor("op_61090_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_61090_end_0 = const()[name = tensor("op_61090_end_0"), val = tensor([2, 320, 1, 4096])]; + tensor var_61090_end_mask_0 = const()[name = tensor("op_61090_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_61090_cast_fp16 = slice_by_index(begin = var_61090_begin_0, end = var_61090_end_0, end_mask = var_61090_end_mask_0, x = v_277_cast_fp16)[name = tensor("op_61090_cast_fp16")]; + tensor var_61094_begin_0 = const()[name = tensor("op_61094_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_61094_end_0 = const()[name = tensor("op_61094_end_0"), val = tensor([2, 384, 1, 4096])]; + tensor var_61094_end_mask_0 = const()[name = tensor("op_61094_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_61094_cast_fp16 = slice_by_index(begin = var_61094_begin_0, end = var_61094_end_0, end_mask = var_61094_end_mask_0, x = v_277_cast_fp16)[name = tensor("op_61094_cast_fp16")]; + tensor var_61098_begin_0 = const()[name = tensor("op_61098_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_61098_end_0 = const()[name = tensor("op_61098_end_0"), val = tensor([2, 448, 1, 4096])]; + tensor var_61098_end_mask_0 = const()[name = tensor("op_61098_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_61098_cast_fp16 = slice_by_index(begin = var_61098_begin_0, end = var_61098_end_0, end_mask = var_61098_end_mask_0, x = v_277_cast_fp16)[name = tensor("op_61098_cast_fp16")]; + tensor var_61102_begin_0 = const()[name = tensor("op_61102_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_61102_end_0 = const()[name = tensor("op_61102_end_0"), val = tensor([2, 512, 1, 4096])]; + tensor var_61102_end_mask_0 = const()[name = tensor("op_61102_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_61102_cast_fp16 = slice_by_index(begin = var_61102_begin_0, end = var_61102_end_0, end_mask = var_61102_end_mask_0, x = v_277_cast_fp16)[name = tensor("op_61102_cast_fp16")]; + tensor var_61106_begin_0 = const()[name = tensor("op_61106_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_61106_end_0 = const()[name = tensor("op_61106_end_0"), val = tensor([2, 576, 1, 4096])]; + tensor var_61106_end_mask_0 = const()[name = tensor("op_61106_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_61106_cast_fp16 = slice_by_index(begin = var_61106_begin_0, end = var_61106_end_0, end_mask = var_61106_end_mask_0, x = v_277_cast_fp16)[name = tensor("op_61106_cast_fp16")]; + tensor var_61110_begin_0 = const()[name = tensor("op_61110_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_61110_end_0 = const()[name = tensor("op_61110_end_0"), val = tensor([2, 640, 1, 4096])]; + tensor var_61110_end_mask_0 = const()[name = tensor("op_61110_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_61110_cast_fp16 = slice_by_index(begin = var_61110_begin_0, end = var_61110_end_0, end_mask = var_61110_end_mask_0, x = v_277_cast_fp16)[name = tensor("op_61110_cast_fp16")]; + tensor var_61114_equation_0 = const()[name = tensor("op_61114_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_61114_cast_fp16 = einsum(equation = var_61114_equation_0, values = (var_61036_cast_fp16, var_60993_cast_fp16))[name = tensor("op_61114_cast_fp16")]; + tensor var_61115_to_fp16 = const()[name = tensor("op_61115_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5161_cast_fp16 = mul(x = var_61114_cast_fp16, y = var_61115_to_fp16)[name = tensor("aw_5161_cast_fp16")]; + tensor var_61118_equation_0 = const()[name = tensor("op_61118_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_61118_cast_fp16 = einsum(equation = var_61118_equation_0, values = (var_61040_cast_fp16, var_60997_cast_fp16))[name = tensor("op_61118_cast_fp16")]; + tensor var_61119_to_fp16 = const()[name = tensor("op_61119_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5163_cast_fp16 = mul(x = var_61118_cast_fp16, y = var_61119_to_fp16)[name = tensor("aw_5163_cast_fp16")]; + tensor var_61122_equation_0 = const()[name = tensor("op_61122_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_61122_cast_fp16 = einsum(equation = var_61122_equation_0, values = (var_61044_cast_fp16, var_61001_cast_fp16))[name = tensor("op_61122_cast_fp16")]; + tensor var_61123_to_fp16 = const()[name = tensor("op_61123_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5165_cast_fp16 = mul(x = var_61122_cast_fp16, y = var_61123_to_fp16)[name = tensor("aw_5165_cast_fp16")]; + tensor var_61126_equation_0 = const()[name = tensor("op_61126_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_61126_cast_fp16 = einsum(equation = var_61126_equation_0, values = (var_61048_cast_fp16, var_61005_cast_fp16))[name = tensor("op_61126_cast_fp16")]; + tensor var_61127_to_fp16 = const()[name = tensor("op_61127_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5167_cast_fp16 = mul(x = var_61126_cast_fp16, y = var_61127_to_fp16)[name = tensor("aw_5167_cast_fp16")]; + tensor var_61130_equation_0 = const()[name = tensor("op_61130_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_61130_cast_fp16 = einsum(equation = var_61130_equation_0, values = (var_61052_cast_fp16, var_61009_cast_fp16))[name = tensor("op_61130_cast_fp16")]; + tensor var_61131_to_fp16 = const()[name = tensor("op_61131_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5169_cast_fp16 = mul(x = var_61130_cast_fp16, y = var_61131_to_fp16)[name = tensor("aw_5169_cast_fp16")]; + tensor var_61134_equation_0 = const()[name = tensor("op_61134_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_61134_cast_fp16 = einsum(equation = var_61134_equation_0, values = (var_61056_cast_fp16, var_61013_cast_fp16))[name = tensor("op_61134_cast_fp16")]; + tensor var_61135_to_fp16 = const()[name = tensor("op_61135_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5171_cast_fp16 = mul(x = var_61134_cast_fp16, y = var_61135_to_fp16)[name = tensor("aw_5171_cast_fp16")]; + tensor var_61138_equation_0 = const()[name = tensor("op_61138_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_61138_cast_fp16 = einsum(equation = var_61138_equation_0, values = (var_61060_cast_fp16, var_61017_cast_fp16))[name = tensor("op_61138_cast_fp16")]; + tensor var_61139_to_fp16 = const()[name = tensor("op_61139_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5173_cast_fp16 = mul(x = var_61138_cast_fp16, y = var_61139_to_fp16)[name = tensor("aw_5173_cast_fp16")]; + tensor var_61142_equation_0 = const()[name = tensor("op_61142_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_61142_cast_fp16 = einsum(equation = var_61142_equation_0, values = (var_61064_cast_fp16, var_61021_cast_fp16))[name = tensor("op_61142_cast_fp16")]; + tensor var_61143_to_fp16 = const()[name = tensor("op_61143_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5175_cast_fp16 = mul(x = var_61142_cast_fp16, y = var_61143_to_fp16)[name = tensor("aw_5175_cast_fp16")]; + tensor var_61146_equation_0 = const()[name = tensor("op_61146_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_61146_cast_fp16 = einsum(equation = var_61146_equation_0, values = (var_61068_cast_fp16, var_61025_cast_fp16))[name = tensor("op_61146_cast_fp16")]; + tensor var_61147_to_fp16 = const()[name = tensor("op_61147_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5177_cast_fp16 = mul(x = var_61146_cast_fp16, y = var_61147_to_fp16)[name = tensor("aw_5177_cast_fp16")]; + tensor var_61150_equation_0 = const()[name = tensor("op_61150_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_61150_cast_fp16 = einsum(equation = var_61150_equation_0, values = (var_61072_cast_fp16, var_61029_cast_fp16))[name = tensor("op_61150_cast_fp16")]; + tensor var_61151_to_fp16 = const()[name = tensor("op_61151_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5179_cast_fp16 = mul(x = var_61150_cast_fp16, y = var_61151_to_fp16)[name = tensor("aw_5179_cast_fp16")]; + tensor var_61153_cast_fp16 = softmax(axis = var_58034, x = aw_5161_cast_fp16)[name = tensor("op_61153_cast_fp16")]; + tensor var_61154_cast_fp16 = softmax(axis = var_58034, x = aw_5163_cast_fp16)[name = tensor("op_61154_cast_fp16")]; + tensor var_61155_cast_fp16 = softmax(axis = var_58034, x = aw_5165_cast_fp16)[name = tensor("op_61155_cast_fp16")]; + tensor var_61156_cast_fp16 = softmax(axis = var_58034, x = aw_5167_cast_fp16)[name = tensor("op_61156_cast_fp16")]; + tensor var_61157_cast_fp16 = softmax(axis = var_58034, x = aw_5169_cast_fp16)[name = tensor("op_61157_cast_fp16")]; + tensor var_61158_cast_fp16 = softmax(axis = var_58034, x = aw_5171_cast_fp16)[name = tensor("op_61158_cast_fp16")]; + tensor var_61159_cast_fp16 = softmax(axis = var_58034, x = aw_5173_cast_fp16)[name = tensor("op_61159_cast_fp16")]; + tensor var_61160_cast_fp16 = softmax(axis = var_58034, x = aw_5175_cast_fp16)[name = tensor("op_61160_cast_fp16")]; + tensor var_61161_cast_fp16 = softmax(axis = var_58034, x = aw_5177_cast_fp16)[name = tensor("op_61161_cast_fp16")]; + tensor var_61162_cast_fp16 = softmax(axis = var_58034, x = aw_5179_cast_fp16)[name = tensor("op_61162_cast_fp16")]; + tensor var_61164_equation_0 = const()[name = tensor("op_61164_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_61164_cast_fp16 = einsum(equation = var_61164_equation_0, values = (var_61074_cast_fp16, var_61153_cast_fp16))[name = tensor("op_61164_cast_fp16")]; + tensor var_61166_equation_0 = const()[name = tensor("op_61166_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_61166_cast_fp16 = einsum(equation = var_61166_equation_0, values = (var_61078_cast_fp16, var_61154_cast_fp16))[name = tensor("op_61166_cast_fp16")]; + tensor var_61168_equation_0 = const()[name = tensor("op_61168_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_61168_cast_fp16 = einsum(equation = var_61168_equation_0, values = (var_61082_cast_fp16, var_61155_cast_fp16))[name = tensor("op_61168_cast_fp16")]; + tensor var_61170_equation_0 = const()[name = tensor("op_61170_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_61170_cast_fp16 = einsum(equation = var_61170_equation_0, values = (var_61086_cast_fp16, var_61156_cast_fp16))[name = tensor("op_61170_cast_fp16")]; + tensor var_61172_equation_0 = const()[name = tensor("op_61172_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_61172_cast_fp16 = einsum(equation = var_61172_equation_0, values = (var_61090_cast_fp16, var_61157_cast_fp16))[name = tensor("op_61172_cast_fp16")]; + tensor var_61174_equation_0 = const()[name = tensor("op_61174_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_61174_cast_fp16 = einsum(equation = var_61174_equation_0, values = (var_61094_cast_fp16, var_61158_cast_fp16))[name = tensor("op_61174_cast_fp16")]; + tensor var_61176_equation_0 = const()[name = tensor("op_61176_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_61176_cast_fp16 = einsum(equation = var_61176_equation_0, values = (var_61098_cast_fp16, var_61159_cast_fp16))[name = tensor("op_61176_cast_fp16")]; + tensor var_61178_equation_0 = const()[name = tensor("op_61178_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_61178_cast_fp16 = einsum(equation = var_61178_equation_0, values = (var_61102_cast_fp16, var_61160_cast_fp16))[name = tensor("op_61178_cast_fp16")]; + tensor var_61180_equation_0 = const()[name = tensor("op_61180_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_61180_cast_fp16 = einsum(equation = var_61180_equation_0, values = (var_61106_cast_fp16, var_61161_cast_fp16))[name = tensor("op_61180_cast_fp16")]; + tensor var_61182_equation_0 = const()[name = tensor("op_61182_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_61182_cast_fp16 = einsum(equation = var_61182_equation_0, values = (var_61110_cast_fp16, var_61162_cast_fp16))[name = tensor("op_61182_cast_fp16")]; + tensor input_813_interleave_0 = const()[name = tensor("input_813_interleave_0"), val = tensor(false)]; + tensor input_813_cast_fp16 = concat(axis = var_58034, interleave = input_813_interleave_0, values = (var_61164_cast_fp16, var_61166_cast_fp16, var_61168_cast_fp16, var_61170_cast_fp16, var_61172_cast_fp16, var_61174_cast_fp16, var_61176_cast_fp16, var_61178_cast_fp16, var_61180_cast_fp16, var_61182_cast_fp16))[name = tensor("input_813_cast_fp16")]; + tensor var_61188 = const()[name = tensor("op_61188"), val = tensor([1, 1])]; + tensor var_61190 = const()[name = tensor("op_61190"), val = tensor([1, 1])]; + tensor var_61192_pad_type_0 = const()[name = tensor("op_61192_pad_type_0"), val = tensor("custom")]; + tensor var_61192_pad_0 = const()[name = tensor("op_61192_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_2_transformer_blocks_1_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_transformer_blocks_1_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5086950720)))]; + tensor up_blocks_1_attentions_2_transformer_blocks_1_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_transformer_blocks_1_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5087769984)))]; + tensor var_61192_cast_fp16 = conv(bias = up_blocks_1_attentions_2_transformer_blocks_1_attn1_to_out_0_bias_to_fp16, dilations = var_61190, groups = var_58034, pad = var_61192_pad_0, pad_type = var_61192_pad_type_0, strides = var_61188, weight = up_blocks_1_attentions_2_transformer_blocks_1_attn1_to_out_0_weight_to_fp16, x = input_813_cast_fp16)[name = tensor("op_61192_cast_fp16")]; + tensor inputs_417_cast_fp16 = add(x = var_61192_cast_fp16, y = inputs_415_cast_fp16)[name = tensor("inputs_417_cast_fp16")]; + tensor hidden_states_569_axes_0 = const()[name = tensor("hidden_states_569_axes_0"), val = tensor([1])]; + tensor hidden_states_569_gamma_0_to_fp16 = const()[name = tensor("hidden_states_569_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5087771328)))]; + tensor hidden_states_569_beta_0_to_fp16 = const()[name = tensor("hidden_states_569_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5087772672)))]; + tensor var_61202_to_fp16 = const()[name = tensor("op_61202_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_569_cast_fp16 = layer_norm(axes = hidden_states_569_axes_0, beta = hidden_states_569_beta_0_to_fp16, epsilon = var_61202_to_fp16, gamma = hidden_states_569_gamma_0_to_fp16, x = inputs_417_cast_fp16)[name = tensor("hidden_states_569_cast_fp16")]; + tensor var_61217 = const()[name = tensor("op_61217"), val = tensor([1, 1])]; + tensor var_61219 = const()[name = tensor("op_61219"), val = tensor([1, 1])]; + tensor q_pad_type_0 = const()[name = tensor("q_pad_type_0"), val = tensor("custom")]; + tensor q_pad_0 = const()[name = tensor("q_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_2_transformer_blocks_1_attn2_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_transformer_blocks_1_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5087774016)))]; + tensor q_cast_fp16 = conv(dilations = var_61219, groups = var_58034, pad = q_pad_0, pad_type = q_pad_type_0, strides = var_61217, weight = up_blocks_1_attentions_2_transformer_blocks_1_attn2_to_q_weight_to_fp16, x = hidden_states_569_cast_fp16)[name = tensor("q_cast_fp16")]; + tensor var_61223 = const()[name = tensor("op_61223"), val = tensor([1, 1])]; + tensor var_61225 = const()[name = tensor("op_61225"), val = tensor([1, 1])]; + tensor k_557_pad_type_0 = const()[name = tensor("k_557_pad_type_0"), val = tensor("custom")]; + tensor k_557_pad_0 = const()[name = tensor("k_557_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_2_transformer_blocks_1_attn2_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_transformer_blocks_1_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5088593280)))]; + tensor k_557_cast_fp16 = conv(dilations = var_61225, groups = var_58034, pad = k_557_pad_0, pad_type = k_557_pad_type_0, strides = var_61223, weight = up_blocks_1_attentions_2_transformer_blocks_1_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_557_cast_fp16")]; + tensor var_61229 = const()[name = tensor("op_61229"), val = tensor([1, 1])]; + tensor var_61231 = const()[name = tensor("op_61231"), val = tensor([1, 1])]; + tensor v_pad_type_0 = const()[name = tensor("v_pad_type_0"), val = tensor("custom")]; + tensor v_pad_0 = const()[name = tensor("v_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_2_transformer_blocks_1_attn2_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_transformer_blocks_1_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5091214784)))]; + tensor v_cast_fp16 = conv(dilations = var_61231, groups = var_58034, pad = v_pad_0, pad_type = v_pad_type_0, strides = var_61229, weight = up_blocks_1_attentions_2_transformer_blocks_1_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_cast_fp16")]; + tensor var_61235_begin_0 = const()[name = tensor("op_61235_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_61235_end_0 = const()[name = tensor("op_61235_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_61235_end_mask_0 = const()[name = tensor("op_61235_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_61235_cast_fp16 = slice_by_index(begin = var_61235_begin_0, end = var_61235_end_0, end_mask = var_61235_end_mask_0, x = q_cast_fp16)[name = tensor("op_61235_cast_fp16")]; + tensor var_61239_begin_0 = const()[name = tensor("op_61239_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_61239_end_0 = const()[name = tensor("op_61239_end_0"), val = tensor([2, 128, 1, 4096])]; + tensor var_61239_end_mask_0 = const()[name = tensor("op_61239_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_61239_cast_fp16 = slice_by_index(begin = var_61239_begin_0, end = var_61239_end_0, end_mask = var_61239_end_mask_0, x = q_cast_fp16)[name = tensor("op_61239_cast_fp16")]; + tensor var_61243_begin_0 = const()[name = tensor("op_61243_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_61243_end_0 = const()[name = tensor("op_61243_end_0"), val = tensor([2, 192, 1, 4096])]; + tensor var_61243_end_mask_0 = const()[name = tensor("op_61243_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_61243_cast_fp16 = slice_by_index(begin = var_61243_begin_0, end = var_61243_end_0, end_mask = var_61243_end_mask_0, x = q_cast_fp16)[name = tensor("op_61243_cast_fp16")]; + tensor var_61247_begin_0 = const()[name = tensor("op_61247_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_61247_end_0 = const()[name = tensor("op_61247_end_0"), val = tensor([2, 256, 1, 4096])]; + tensor var_61247_end_mask_0 = const()[name = tensor("op_61247_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_61247_cast_fp16 = slice_by_index(begin = var_61247_begin_0, end = var_61247_end_0, end_mask = var_61247_end_mask_0, x = q_cast_fp16)[name = tensor("op_61247_cast_fp16")]; + tensor var_61251_begin_0 = const()[name = tensor("op_61251_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_61251_end_0 = const()[name = tensor("op_61251_end_0"), val = tensor([2, 320, 1, 4096])]; + tensor var_61251_end_mask_0 = const()[name = tensor("op_61251_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_61251_cast_fp16 = slice_by_index(begin = var_61251_begin_0, end = var_61251_end_0, end_mask = var_61251_end_mask_0, x = q_cast_fp16)[name = tensor("op_61251_cast_fp16")]; + tensor var_61255_begin_0 = const()[name = tensor("op_61255_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_61255_end_0 = const()[name = tensor("op_61255_end_0"), val = tensor([2, 384, 1, 4096])]; + tensor var_61255_end_mask_0 = const()[name = tensor("op_61255_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_61255_cast_fp16 = slice_by_index(begin = var_61255_begin_0, end = var_61255_end_0, end_mask = var_61255_end_mask_0, x = q_cast_fp16)[name = tensor("op_61255_cast_fp16")]; + tensor var_61259_begin_0 = const()[name = tensor("op_61259_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_61259_end_0 = const()[name = tensor("op_61259_end_0"), val = tensor([2, 448, 1, 4096])]; + tensor var_61259_end_mask_0 = const()[name = tensor("op_61259_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_61259_cast_fp16 = slice_by_index(begin = var_61259_begin_0, end = var_61259_end_0, end_mask = var_61259_end_mask_0, x = q_cast_fp16)[name = tensor("op_61259_cast_fp16")]; + tensor var_61263_begin_0 = const()[name = tensor("op_61263_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_61263_end_0 = const()[name = tensor("op_61263_end_0"), val = tensor([2, 512, 1, 4096])]; + tensor var_61263_end_mask_0 = const()[name = tensor("op_61263_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_61263_cast_fp16 = slice_by_index(begin = var_61263_begin_0, end = var_61263_end_0, end_mask = var_61263_end_mask_0, x = q_cast_fp16)[name = tensor("op_61263_cast_fp16")]; + tensor var_61267_begin_0 = const()[name = tensor("op_61267_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_61267_end_0 = const()[name = tensor("op_61267_end_0"), val = tensor([2, 576, 1, 4096])]; + tensor var_61267_end_mask_0 = const()[name = tensor("op_61267_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_61267_cast_fp16 = slice_by_index(begin = var_61267_begin_0, end = var_61267_end_0, end_mask = var_61267_end_mask_0, x = q_cast_fp16)[name = tensor("op_61267_cast_fp16")]; + tensor var_61271_begin_0 = const()[name = tensor("op_61271_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_61271_end_0 = const()[name = tensor("op_61271_end_0"), val = tensor([2, 640, 1, 4096])]; + tensor var_61271_end_mask_0 = const()[name = tensor("op_61271_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_61271_cast_fp16 = slice_by_index(begin = var_61271_begin_0, end = var_61271_end_0, end_mask = var_61271_end_mask_0, x = q_cast_fp16)[name = tensor("op_61271_cast_fp16")]; + tensor k_perm_0 = const()[name = tensor("k_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_61278_begin_0 = const()[name = tensor("op_61278_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_61278_end_0 = const()[name = tensor("op_61278_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_61278_end_mask_0 = const()[name = tensor("op_61278_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_cast_fp16 = transpose(perm = k_perm_0, x = k_557_cast_fp16)[name = tensor("transpose_0")]; + tensor var_61278_cast_fp16 = slice_by_index(begin = var_61278_begin_0, end = var_61278_end_0, end_mask = var_61278_end_mask_0, x = k_cast_fp16)[name = tensor("op_61278_cast_fp16")]; + tensor var_61282_begin_0 = const()[name = tensor("op_61282_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_61282_end_0 = const()[name = tensor("op_61282_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_61282_end_mask_0 = const()[name = tensor("op_61282_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_61282_cast_fp16 = slice_by_index(begin = var_61282_begin_0, end = var_61282_end_0, end_mask = var_61282_end_mask_0, x = k_cast_fp16)[name = tensor("op_61282_cast_fp16")]; + tensor var_61286_begin_0 = const()[name = tensor("op_61286_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_61286_end_0 = const()[name = tensor("op_61286_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_61286_end_mask_0 = const()[name = tensor("op_61286_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_61286_cast_fp16 = slice_by_index(begin = var_61286_begin_0, end = var_61286_end_0, end_mask = var_61286_end_mask_0, x = k_cast_fp16)[name = tensor("op_61286_cast_fp16")]; + tensor var_61290_begin_0 = const()[name = tensor("op_61290_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_61290_end_0 = const()[name = tensor("op_61290_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_61290_end_mask_0 = const()[name = tensor("op_61290_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_61290_cast_fp16 = slice_by_index(begin = var_61290_begin_0, end = var_61290_end_0, end_mask = var_61290_end_mask_0, x = k_cast_fp16)[name = tensor("op_61290_cast_fp16")]; + tensor var_61294_begin_0 = const()[name = tensor("op_61294_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_61294_end_0 = const()[name = tensor("op_61294_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_61294_end_mask_0 = const()[name = tensor("op_61294_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_61294_cast_fp16 = slice_by_index(begin = var_61294_begin_0, end = var_61294_end_0, end_mask = var_61294_end_mask_0, x = k_cast_fp16)[name = tensor("op_61294_cast_fp16")]; + tensor var_61298_begin_0 = const()[name = tensor("op_61298_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_61298_end_0 = const()[name = tensor("op_61298_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_61298_end_mask_0 = const()[name = tensor("op_61298_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_61298_cast_fp16 = slice_by_index(begin = var_61298_begin_0, end = var_61298_end_0, end_mask = var_61298_end_mask_0, x = k_cast_fp16)[name = tensor("op_61298_cast_fp16")]; + tensor var_61302_begin_0 = const()[name = tensor("op_61302_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_61302_end_0 = const()[name = tensor("op_61302_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_61302_end_mask_0 = const()[name = tensor("op_61302_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_61302_cast_fp16 = slice_by_index(begin = var_61302_begin_0, end = var_61302_end_0, end_mask = var_61302_end_mask_0, x = k_cast_fp16)[name = tensor("op_61302_cast_fp16")]; + tensor var_61306_begin_0 = const()[name = tensor("op_61306_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_61306_end_0 = const()[name = tensor("op_61306_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_61306_end_mask_0 = const()[name = tensor("op_61306_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_61306_cast_fp16 = slice_by_index(begin = var_61306_begin_0, end = var_61306_end_0, end_mask = var_61306_end_mask_0, x = k_cast_fp16)[name = tensor("op_61306_cast_fp16")]; + tensor var_61310_begin_0 = const()[name = tensor("op_61310_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_61310_end_0 = const()[name = tensor("op_61310_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_61310_end_mask_0 = const()[name = tensor("op_61310_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_61310_cast_fp16 = slice_by_index(begin = var_61310_begin_0, end = var_61310_end_0, end_mask = var_61310_end_mask_0, x = k_cast_fp16)[name = tensor("op_61310_cast_fp16")]; + tensor var_61314_begin_0 = const()[name = tensor("op_61314_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_61314_end_0 = const()[name = tensor("op_61314_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_61314_end_mask_0 = const()[name = tensor("op_61314_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_61314_cast_fp16 = slice_by_index(begin = var_61314_begin_0, end = var_61314_end_0, end_mask = var_61314_end_mask_0, x = k_cast_fp16)[name = tensor("op_61314_cast_fp16")]; + tensor var_61316_begin_0 = const()[name = tensor("op_61316_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_61316_end_0 = const()[name = tensor("op_61316_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_61316_end_mask_0 = const()[name = tensor("op_61316_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_61316_cast_fp16 = slice_by_index(begin = var_61316_begin_0, end = var_61316_end_0, end_mask = var_61316_end_mask_0, x = v_cast_fp16)[name = tensor("op_61316_cast_fp16")]; + tensor var_61320_begin_0 = const()[name = tensor("op_61320_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_61320_end_0 = const()[name = tensor("op_61320_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_61320_end_mask_0 = const()[name = tensor("op_61320_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_61320_cast_fp16 = slice_by_index(begin = var_61320_begin_0, end = var_61320_end_0, end_mask = var_61320_end_mask_0, x = v_cast_fp16)[name = tensor("op_61320_cast_fp16")]; + tensor var_61324_begin_0 = const()[name = tensor("op_61324_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_61324_end_0 = const()[name = tensor("op_61324_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_61324_end_mask_0 = const()[name = tensor("op_61324_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_61324_cast_fp16 = slice_by_index(begin = var_61324_begin_0, end = var_61324_end_0, end_mask = var_61324_end_mask_0, x = v_cast_fp16)[name = tensor("op_61324_cast_fp16")]; + tensor var_61328_begin_0 = const()[name = tensor("op_61328_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_61328_end_0 = const()[name = tensor("op_61328_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_61328_end_mask_0 = const()[name = tensor("op_61328_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_61328_cast_fp16 = slice_by_index(begin = var_61328_begin_0, end = var_61328_end_0, end_mask = var_61328_end_mask_0, x = v_cast_fp16)[name = tensor("op_61328_cast_fp16")]; + tensor var_61332_begin_0 = const()[name = tensor("op_61332_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_61332_end_0 = const()[name = tensor("op_61332_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_61332_end_mask_0 = const()[name = tensor("op_61332_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_61332_cast_fp16 = slice_by_index(begin = var_61332_begin_0, end = var_61332_end_0, end_mask = var_61332_end_mask_0, x = v_cast_fp16)[name = tensor("op_61332_cast_fp16")]; + tensor var_61336_begin_0 = const()[name = tensor("op_61336_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_61336_end_0 = const()[name = tensor("op_61336_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_61336_end_mask_0 = const()[name = tensor("op_61336_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_61336_cast_fp16 = slice_by_index(begin = var_61336_begin_0, end = var_61336_end_0, end_mask = var_61336_end_mask_0, x = v_cast_fp16)[name = tensor("op_61336_cast_fp16")]; + tensor var_61340_begin_0 = const()[name = tensor("op_61340_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_61340_end_0 = const()[name = tensor("op_61340_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_61340_end_mask_0 = const()[name = tensor("op_61340_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_61340_cast_fp16 = slice_by_index(begin = var_61340_begin_0, end = var_61340_end_0, end_mask = var_61340_end_mask_0, x = v_cast_fp16)[name = tensor("op_61340_cast_fp16")]; + tensor var_61344_begin_0 = const()[name = tensor("op_61344_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_61344_end_0 = const()[name = tensor("op_61344_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_61344_end_mask_0 = const()[name = tensor("op_61344_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_61344_cast_fp16 = slice_by_index(begin = var_61344_begin_0, end = var_61344_end_0, end_mask = var_61344_end_mask_0, x = v_cast_fp16)[name = tensor("op_61344_cast_fp16")]; + tensor var_61348_begin_0 = const()[name = tensor("op_61348_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_61348_end_0 = const()[name = tensor("op_61348_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_61348_end_mask_0 = const()[name = tensor("op_61348_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_61348_cast_fp16 = slice_by_index(begin = var_61348_begin_0, end = var_61348_end_0, end_mask = var_61348_end_mask_0, x = v_cast_fp16)[name = tensor("op_61348_cast_fp16")]; + tensor var_61352_begin_0 = const()[name = tensor("op_61352_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_61352_end_0 = const()[name = tensor("op_61352_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_61352_end_mask_0 = const()[name = tensor("op_61352_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_61352_cast_fp16 = slice_by_index(begin = var_61352_begin_0, end = var_61352_end_0, end_mask = var_61352_end_mask_0, x = v_cast_fp16)[name = tensor("op_61352_cast_fp16")]; + tensor var_61356_equation_0 = const()[name = tensor("op_61356_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_61356_cast_fp16 = einsum(equation = var_61356_equation_0, values = (var_61278_cast_fp16, var_61235_cast_fp16))[name = tensor("op_61356_cast_fp16")]; + tensor var_61357_to_fp16 = const()[name = tensor("op_61357_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5181_cast_fp16 = mul(x = var_61356_cast_fp16, y = var_61357_to_fp16)[name = tensor("aw_5181_cast_fp16")]; + tensor var_61360_equation_0 = const()[name = tensor("op_61360_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_61360_cast_fp16 = einsum(equation = var_61360_equation_0, values = (var_61282_cast_fp16, var_61239_cast_fp16))[name = tensor("op_61360_cast_fp16")]; + tensor var_61361_to_fp16 = const()[name = tensor("op_61361_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5183_cast_fp16 = mul(x = var_61360_cast_fp16, y = var_61361_to_fp16)[name = tensor("aw_5183_cast_fp16")]; + tensor var_61364_equation_0 = const()[name = tensor("op_61364_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_61364_cast_fp16 = einsum(equation = var_61364_equation_0, values = (var_61286_cast_fp16, var_61243_cast_fp16))[name = tensor("op_61364_cast_fp16")]; + tensor var_61365_to_fp16 = const()[name = tensor("op_61365_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5185_cast_fp16 = mul(x = var_61364_cast_fp16, y = var_61365_to_fp16)[name = tensor("aw_5185_cast_fp16")]; + tensor var_61368_equation_0 = const()[name = tensor("op_61368_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_61368_cast_fp16 = einsum(equation = var_61368_equation_0, values = (var_61290_cast_fp16, var_61247_cast_fp16))[name = tensor("op_61368_cast_fp16")]; + tensor var_61369_to_fp16 = const()[name = tensor("op_61369_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5187_cast_fp16 = mul(x = var_61368_cast_fp16, y = var_61369_to_fp16)[name = tensor("aw_5187_cast_fp16")]; + tensor var_61372_equation_0 = const()[name = tensor("op_61372_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_61372_cast_fp16 = einsum(equation = var_61372_equation_0, values = (var_61294_cast_fp16, var_61251_cast_fp16))[name = tensor("op_61372_cast_fp16")]; + tensor var_61373_to_fp16 = const()[name = tensor("op_61373_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5189_cast_fp16 = mul(x = var_61372_cast_fp16, y = var_61373_to_fp16)[name = tensor("aw_5189_cast_fp16")]; + tensor var_61376_equation_0 = const()[name = tensor("op_61376_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_61376_cast_fp16 = einsum(equation = var_61376_equation_0, values = (var_61298_cast_fp16, var_61255_cast_fp16))[name = tensor("op_61376_cast_fp16")]; + tensor var_61377_to_fp16 = const()[name = tensor("op_61377_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5191_cast_fp16 = mul(x = var_61376_cast_fp16, y = var_61377_to_fp16)[name = tensor("aw_5191_cast_fp16")]; + tensor var_61380_equation_0 = const()[name = tensor("op_61380_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_61380_cast_fp16 = einsum(equation = var_61380_equation_0, values = (var_61302_cast_fp16, var_61259_cast_fp16))[name = tensor("op_61380_cast_fp16")]; + tensor var_61381_to_fp16 = const()[name = tensor("op_61381_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5193_cast_fp16 = mul(x = var_61380_cast_fp16, y = var_61381_to_fp16)[name = tensor("aw_5193_cast_fp16")]; + tensor var_61384_equation_0 = const()[name = tensor("op_61384_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_61384_cast_fp16 = einsum(equation = var_61384_equation_0, values = (var_61306_cast_fp16, var_61263_cast_fp16))[name = tensor("op_61384_cast_fp16")]; + tensor var_61385_to_fp16 = const()[name = tensor("op_61385_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5195_cast_fp16 = mul(x = var_61384_cast_fp16, y = var_61385_to_fp16)[name = tensor("aw_5195_cast_fp16")]; + tensor var_61388_equation_0 = const()[name = tensor("op_61388_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_61388_cast_fp16 = einsum(equation = var_61388_equation_0, values = (var_61310_cast_fp16, var_61267_cast_fp16))[name = tensor("op_61388_cast_fp16")]; + tensor var_61389_to_fp16 = const()[name = tensor("op_61389_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5197_cast_fp16 = mul(x = var_61388_cast_fp16, y = var_61389_to_fp16)[name = tensor("aw_5197_cast_fp16")]; + tensor var_61392_equation_0 = const()[name = tensor("op_61392_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_61392_cast_fp16 = einsum(equation = var_61392_equation_0, values = (var_61314_cast_fp16, var_61271_cast_fp16))[name = tensor("op_61392_cast_fp16")]; + tensor var_61393_to_fp16 = const()[name = tensor("op_61393_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_cast_fp16 = mul(x = var_61392_cast_fp16, y = var_61393_to_fp16)[name = tensor("aw_cast_fp16")]; + tensor var_61395_cast_fp16 = softmax(axis = var_58034, x = aw_5181_cast_fp16)[name = tensor("op_61395_cast_fp16")]; + tensor var_61396_cast_fp16 = softmax(axis = var_58034, x = aw_5183_cast_fp16)[name = tensor("op_61396_cast_fp16")]; + tensor var_61397_cast_fp16 = softmax(axis = var_58034, x = aw_5185_cast_fp16)[name = tensor("op_61397_cast_fp16")]; + tensor var_61398_cast_fp16 = softmax(axis = var_58034, x = aw_5187_cast_fp16)[name = tensor("op_61398_cast_fp16")]; + tensor var_61399_cast_fp16 = softmax(axis = var_58034, x = aw_5189_cast_fp16)[name = tensor("op_61399_cast_fp16")]; + tensor var_61400_cast_fp16 = softmax(axis = var_58034, x = aw_5191_cast_fp16)[name = tensor("op_61400_cast_fp16")]; + tensor var_61401_cast_fp16 = softmax(axis = var_58034, x = aw_5193_cast_fp16)[name = tensor("op_61401_cast_fp16")]; + tensor var_61402_cast_fp16 = softmax(axis = var_58034, x = aw_5195_cast_fp16)[name = tensor("op_61402_cast_fp16")]; + tensor var_61403_cast_fp16 = softmax(axis = var_58034, x = aw_5197_cast_fp16)[name = tensor("op_61403_cast_fp16")]; + tensor var_61404_cast_fp16 = softmax(axis = var_58034, x = aw_cast_fp16)[name = tensor("op_61404_cast_fp16")]; + tensor var_61406_equation_0 = const()[name = tensor("op_61406_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_61406_cast_fp16 = einsum(equation = var_61406_equation_0, values = (var_61316_cast_fp16, var_61395_cast_fp16))[name = tensor("op_61406_cast_fp16")]; + tensor var_61408_equation_0 = const()[name = tensor("op_61408_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_61408_cast_fp16 = einsum(equation = var_61408_equation_0, values = (var_61320_cast_fp16, var_61396_cast_fp16))[name = tensor("op_61408_cast_fp16")]; + tensor var_61410_equation_0 = const()[name = tensor("op_61410_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_61410_cast_fp16 = einsum(equation = var_61410_equation_0, values = (var_61324_cast_fp16, var_61397_cast_fp16))[name = tensor("op_61410_cast_fp16")]; + tensor var_61412_equation_0 = const()[name = tensor("op_61412_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_61412_cast_fp16 = einsum(equation = var_61412_equation_0, values = (var_61328_cast_fp16, var_61398_cast_fp16))[name = tensor("op_61412_cast_fp16")]; + tensor var_61414_equation_0 = const()[name = tensor("op_61414_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_61414_cast_fp16 = einsum(equation = var_61414_equation_0, values = (var_61332_cast_fp16, var_61399_cast_fp16))[name = tensor("op_61414_cast_fp16")]; + tensor var_61416_equation_0 = const()[name = tensor("op_61416_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_61416_cast_fp16 = einsum(equation = var_61416_equation_0, values = (var_61336_cast_fp16, var_61400_cast_fp16))[name = tensor("op_61416_cast_fp16")]; + tensor var_61418_equation_0 = const()[name = tensor("op_61418_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_61418_cast_fp16 = einsum(equation = var_61418_equation_0, values = (var_61340_cast_fp16, var_61401_cast_fp16))[name = tensor("op_61418_cast_fp16")]; + tensor var_61420_equation_0 = const()[name = tensor("op_61420_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_61420_cast_fp16 = einsum(equation = var_61420_equation_0, values = (var_61344_cast_fp16, var_61402_cast_fp16))[name = tensor("op_61420_cast_fp16")]; + tensor var_61422_equation_0 = const()[name = tensor("op_61422_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_61422_cast_fp16 = einsum(equation = var_61422_equation_0, values = (var_61348_cast_fp16, var_61403_cast_fp16))[name = tensor("op_61422_cast_fp16")]; + tensor var_61424_equation_0 = const()[name = tensor("op_61424_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_61424_cast_fp16 = einsum(equation = var_61424_equation_0, values = (var_61352_cast_fp16, var_61404_cast_fp16))[name = tensor("op_61424_cast_fp16")]; + tensor input_815_interleave_0 = const()[name = tensor("input_815_interleave_0"), val = tensor(false)]; + tensor input_815_cast_fp16 = concat(axis = var_58034, interleave = input_815_interleave_0, values = (var_61406_cast_fp16, var_61408_cast_fp16, var_61410_cast_fp16, var_61412_cast_fp16, var_61414_cast_fp16, var_61416_cast_fp16, var_61418_cast_fp16, var_61420_cast_fp16, var_61422_cast_fp16, var_61424_cast_fp16))[name = tensor("input_815_cast_fp16")]; + tensor var_61430 = const()[name = tensor("op_61430"), val = tensor([1, 1])]; + tensor var_61432 = const()[name = tensor("op_61432"), val = tensor([1, 1])]; + tensor var_61434_pad_type_0 = const()[name = tensor("op_61434_pad_type_0"), val = tensor("custom")]; + tensor var_61434_pad_0 = const()[name = tensor("op_61434_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_2_transformer_blocks_1_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_transformer_blocks_1_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5093836288)))]; + tensor up_blocks_1_attentions_2_transformer_blocks_1_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_transformer_blocks_1_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5094655552)))]; + tensor var_61434_cast_fp16 = conv(bias = up_blocks_1_attentions_2_transformer_blocks_1_attn2_to_out_0_bias_to_fp16, dilations = var_61432, groups = var_58034, pad = var_61434_pad_0, pad_type = var_61434_pad_type_0, strides = var_61430, weight = up_blocks_1_attentions_2_transformer_blocks_1_attn2_to_out_0_weight_to_fp16, x = input_815_cast_fp16)[name = tensor("op_61434_cast_fp16")]; + tensor inputs_cast_fp16 = add(x = var_61434_cast_fp16, y = inputs_417_cast_fp16)[name = tensor("inputs_cast_fp16")]; + tensor input_817_axes_0 = const()[name = tensor("input_817_axes_0"), val = tensor([1])]; + tensor input_817_gamma_0_to_fp16 = const()[name = tensor("input_817_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5094656896)))]; + tensor input_817_beta_0_to_fp16 = const()[name = tensor("input_817_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5094658240)))]; + tensor var_61444_to_fp16 = const()[name = tensor("op_61444_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_817_cast_fp16 = layer_norm(axes = input_817_axes_0, beta = input_817_beta_0_to_fp16, epsilon = var_61444_to_fp16, gamma = input_817_gamma_0_to_fp16, x = inputs_cast_fp16)[name = tensor("input_817_cast_fp16")]; + tensor var_61460 = const()[name = tensor("op_61460"), val = tensor([1, 1])]; + tensor var_61462 = const()[name = tensor("op_61462"), val = tensor([1, 1])]; + tensor var_61464_pad_type_0 = const()[name = tensor("op_61464_pad_type_0"), val = tensor("custom")]; + tensor var_61464_pad_0 = const()[name = tensor("op_61464_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_2_transformer_blocks_1_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_transformer_blocks_1_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5094659584)))]; + tensor up_blocks_1_attentions_2_transformer_blocks_1_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_transformer_blocks_1_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5101213248)))]; + tensor var_61464_cast_fp16 = conv(bias = up_blocks_1_attentions_2_transformer_blocks_1_ff_net_0_proj_bias_to_fp16, dilations = var_61462, groups = var_58034, pad = var_61464_pad_0, pad_type = var_61464_pad_type_0, strides = var_61460, weight = up_blocks_1_attentions_2_transformer_blocks_1_ff_net_0_proj_weight_to_fp16, x = input_817_cast_fp16)[name = tensor("op_61464_cast_fp16")]; + tensor var_61465_split_sizes_0 = const()[name = tensor("op_61465_split_sizes_0"), val = tensor([2560, 2560])]; + tensor var_61465_axis_0 = const()[name = tensor("op_61465_axis_0"), val = tensor(1)]; + tensor var_61465_cast_fp16_0, tensor var_61465_cast_fp16_1 = split(axis = var_61465_axis_0, split_sizes = var_61465_split_sizes_0, x = var_61464_cast_fp16)[name = tensor("op_61465_cast_fp16")]; + tensor var_61467_mode_0 = const()[name = tensor("op_61467_mode_0"), val = tensor("EXACT")]; + tensor var_61467_cast_fp16 = gelu(mode = var_61467_mode_0, x = var_61465_cast_fp16_1)[name = tensor("op_61467_cast_fp16")]; + tensor input_819_cast_fp16 = mul(x = var_61465_cast_fp16_0, y = var_61467_cast_fp16)[name = tensor("input_819_cast_fp16")]; + tensor var_61471 = const()[name = tensor("op_61471"), val = tensor([1, 1])]; + tensor var_61473 = const()[name = tensor("op_61473"), val = tensor([1, 1])]; + tensor var_61475_pad_type_0 = const()[name = tensor("op_61475_pad_type_0"), val = tensor("custom")]; + tensor var_61475_pad_0 = const()[name = tensor("op_61475_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_2_transformer_blocks_1_ff_net_2_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_transformer_blocks_1_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5101223552)))]; + tensor up_blocks_1_attentions_2_transformer_blocks_1_ff_net_2_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_transformer_blocks_1_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5104500416)))]; + tensor var_61475_cast_fp16 = conv(bias = up_blocks_1_attentions_2_transformer_blocks_1_ff_net_2_bias_to_fp16, dilations = var_61473, groups = var_58034, pad = var_61475_pad_0, pad_type = var_61475_pad_type_0, strides = var_61471, weight = up_blocks_1_attentions_2_transformer_blocks_1_ff_net_2_weight_to_fp16, x = input_819_cast_fp16)[name = tensor("op_61475_cast_fp16")]; + tensor hidden_states_573_cast_fp16 = add(x = var_61475_cast_fp16, y = inputs_cast_fp16)[name = tensor("hidden_states_573_cast_fp16")]; + tensor var_61477 = const()[name = tensor("op_61477"), val = tensor([2, 640, 64, 64])]; + tensor input_821_cast_fp16 = reshape(shape = var_61477, x = hidden_states_573_cast_fp16)[name = tensor("input_821_cast_fp16")]; + tensor var_61481 = const()[name = tensor("op_61481"), val = tensor([1, 1])]; + tensor var_61483 = const()[name = tensor("op_61483"), val = tensor([1, 1])]; + tensor hidden_states_575_pad_type_0 = const()[name = tensor("hidden_states_575_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_575_pad_0 = const()[name = tensor("hidden_states_575_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_2_proj_out_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_proj_out_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5104501760)))]; + tensor up_blocks_1_attentions_2_proj_out_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_proj_out_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5105321024)))]; + tensor hidden_states_575_cast_fp16 = conv(bias = up_blocks_1_attentions_2_proj_out_bias_to_fp16, dilations = var_61483, groups = var_58034, pad = hidden_states_575_pad_0, pad_type = hidden_states_575_pad_type_0, strides = var_61481, weight = up_blocks_1_attentions_2_proj_out_weight_to_fp16, x = input_821_cast_fp16)[name = tensor("hidden_states_575_cast_fp16")]; + tensor input_823_cast_fp16 = add(x = hidden_states_575_cast_fp16, y = hidden_states_557_cast_fp16)[name = tensor("input_823_cast_fp16")]; + tensor input_825_scale_factor_height_0 = const()[name = tensor("input_825_scale_factor_height_0"), val = tensor(0x1p+1)]; + tensor input_825_scale_factor_width_0 = const()[name = tensor("input_825_scale_factor_width_0"), val = tensor(0x1p+1)]; + tensor input_825_cast_fp16 = upsample_nearest_neighbor(scale_factor_height = input_825_scale_factor_height_0, scale_factor_width = input_825_scale_factor_width_0, x = input_823_cast_fp16)[name = tensor("input_825_cast_fp16")]; + tensor var_61492 = const()[name = tensor("op_61492"), val = tensor([1, 1])]; + tensor var_61494 = const()[name = tensor("op_61494"), val = tensor([1, 1])]; + tensor hidden_states_577_pad_type_0 = const()[name = tensor("hidden_states_577_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_577_pad_0 = const()[name = tensor("hidden_states_577_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor up_blocks_1_upsamplers_0_conv_weight_to_fp16 = const()[name = tensor("up_blocks_1_upsamplers_0_conv_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5105322368)))]; + tensor up_blocks_1_upsamplers_0_conv_bias_to_fp16 = const()[name = tensor("up_blocks_1_upsamplers_0_conv_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5112695232)))]; + tensor hidden_states_577_cast_fp16 = conv(bias = up_blocks_1_upsamplers_0_conv_bias_to_fp16, dilations = var_61494, groups = var_58034, pad = hidden_states_577_pad_0, pad_type = hidden_states_577_pad_type_0, strides = var_61492, weight = up_blocks_1_upsamplers_0_conv_weight_to_fp16, x = input_825_cast_fp16)[name = tensor("hidden_states_577_cast_fp16")]; + tensor var_61502 = const()[name = tensor("op_61502"), val = tensor(1)]; + tensor input_827_interleave_0 = const()[name = tensor("input_827_interleave_0"), val = tensor(false)]; + tensor input_827_cast_fp16 = concat(axis = var_61502, interleave = input_827_interleave_0, values = (hidden_states_577_cast_fp16, input_43_cast_fp16))[name = tensor("input_827_cast_fp16")]; + tensor reshape_156_shape_0 = const()[name = tensor("reshape_156_shape_0"), val = tensor([2, 32, 30, 128, 128])]; + tensor reshape_156_cast_fp16 = reshape(shape = reshape_156_shape_0, x = input_827_cast_fp16)[name = tensor("reshape_156_cast_fp16")]; + tensor reduce_mean_117_axes_0 = const()[name = tensor("reduce_mean_117_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_117_keep_dims_0 = const()[name = tensor("reduce_mean_117_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_117_cast_fp16 = reduce_mean(axes = reduce_mean_117_axes_0, keep_dims = reduce_mean_117_keep_dims_0, x = reshape_156_cast_fp16)[name = tensor("reduce_mean_117_cast_fp16")]; + tensor sub_78_cast_fp16 = sub(x = reshape_156_cast_fp16, y = reduce_mean_117_cast_fp16)[name = tensor("sub_78_cast_fp16")]; + tensor square_39_cast_fp16 = square(x = sub_78_cast_fp16)[name = tensor("square_39_cast_fp16")]; + tensor reduce_mean_119_axes_0 = const()[name = tensor("reduce_mean_119_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_119_keep_dims_0 = const()[name = tensor("reduce_mean_119_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_119_cast_fp16 = reduce_mean(axes = reduce_mean_119_axes_0, keep_dims = reduce_mean_119_keep_dims_0, x = square_39_cast_fp16)[name = tensor("reduce_mean_119_cast_fp16")]; + tensor add_78_y_0_to_fp16 = const()[name = tensor("add_78_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_78_cast_fp16 = add(x = reduce_mean_119_cast_fp16, y = add_78_y_0_to_fp16)[name = tensor("add_78_cast_fp16")]; + tensor sqrt_39_cast_fp16 = sqrt(x = add_78_cast_fp16)[name = tensor("sqrt_39_cast_fp16")]; + tensor real_div_39_cast_fp16 = real_div(x = sub_78_cast_fp16, y = sqrt_39_cast_fp16)[name = tensor("real_div_39_cast_fp16")]; + tensor reshape_157_shape_0 = const()[name = tensor("reshape_157_shape_0"), val = tensor([2, 960, 128, 128])]; + tensor reshape_157_cast_fp16 = reshape(shape = reshape_157_shape_0, x = real_div_39_cast_fp16)[name = tensor("reshape_157_cast_fp16")]; + tensor add_79_gamma_0_to_fp16 = const()[name = tensor("add_79_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5112696576)))]; + tensor add_79_beta_0_to_fp16 = const()[name = tensor("add_79_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5112698560)))]; + tensor add_79_epsilon_0_to_fp16 = const()[name = tensor("add_79_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_79_cast_fp16 = batch_norm(beta = add_79_beta_0_to_fp16, epsilon = add_79_epsilon_0_to_fp16, gamma = add_79_gamma_0_to_fp16, mean = add_73_mean_0_to_fp16, variance = add_73_variance_0_to_fp16, x = reshape_157_cast_fp16)[name = tensor("add_79_cast_fp16")]; + tensor input_831_cast_fp16 = silu(x = add_79_cast_fp16)[name = tensor("input_831_cast_fp16")]; + tensor var_61523 = const()[name = tensor("op_61523"), val = tensor([1, 1])]; + tensor var_61525 = const()[name = tensor("op_61525"), val = tensor([1, 1])]; + tensor hidden_states_579_pad_type_0 = const()[name = tensor("hidden_states_579_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_579_pad_0 = const()[name = tensor("hidden_states_579_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor up_blocks_2_resnets_0_conv1_weight_to_fp16 = const()[name = tensor("up_blocks_2_resnets_0_conv1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5112700544)))]; + tensor up_blocks_2_resnets_0_conv1_bias_to_fp16 = const()[name = tensor("up_blocks_2_resnets_0_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5118230208)))]; + tensor hidden_states_579_cast_fp16 = conv(bias = up_blocks_2_resnets_0_conv1_bias_to_fp16, dilations = var_61525, groups = var_61502, pad = hidden_states_579_pad_0, pad_type = hidden_states_579_pad_type_0, strides = var_61523, weight = up_blocks_2_resnets_0_conv1_weight_to_fp16, x = input_831_cast_fp16)[name = tensor("hidden_states_579_cast_fp16")]; + tensor var_61531 = const()[name = tensor("op_61531"), val = tensor([1, 1])]; + tensor var_61533 = const()[name = tensor("op_61533"), val = tensor([1, 1])]; + tensor temb_29_pad_type_0 = const()[name = tensor("temb_29_pad_type_0"), val = tensor("custom")]; + tensor temb_29_pad_0 = const()[name = tensor("temb_29_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_2_resnets_0_time_emb_proj_weight_to_fp16 = const()[name = tensor("up_blocks_2_resnets_0_time_emb_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5118230912)))]; + tensor up_blocks_2_resnets_0_time_emb_proj_bias_to_fp16 = const()[name = tensor("up_blocks_2_resnets_0_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5119050176)))]; + tensor temb_29_cast_fp16 = conv(bias = up_blocks_2_resnets_0_time_emb_proj_bias_to_fp16, dilations = var_61533, groups = var_61502, pad = temb_29_pad_0, pad_type = temb_29_pad_type_0, strides = var_61531, weight = up_blocks_2_resnets_0_time_emb_proj_weight_to_fp16, x = input_21_cast_fp16)[name = tensor("temb_29_cast_fp16")]; + tensor input_835_cast_fp16 = add(x = hidden_states_579_cast_fp16, y = temb_29_cast_fp16)[name = tensor("input_835_cast_fp16")]; + tensor reshape_160_shape_0 = const()[name = tensor("reshape_160_shape_0"), val = tensor([2, 32, 10, 128, 128])]; + tensor reshape_160_cast_fp16 = reshape(shape = reshape_160_shape_0, x = input_835_cast_fp16)[name = tensor("reshape_160_cast_fp16")]; + tensor reduce_mean_120_axes_0 = const()[name = tensor("reduce_mean_120_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_120_keep_dims_0 = const()[name = tensor("reduce_mean_120_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_120_cast_fp16 = reduce_mean(axes = reduce_mean_120_axes_0, keep_dims = reduce_mean_120_keep_dims_0, x = reshape_160_cast_fp16)[name = tensor("reduce_mean_120_cast_fp16")]; + tensor sub_80_cast_fp16 = sub(x = reshape_160_cast_fp16, y = reduce_mean_120_cast_fp16)[name = tensor("sub_80_cast_fp16")]; + tensor square_40_cast_fp16 = square(x = sub_80_cast_fp16)[name = tensor("square_40_cast_fp16")]; + tensor reduce_mean_122_axes_0 = const()[name = tensor("reduce_mean_122_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_122_keep_dims_0 = const()[name = tensor("reduce_mean_122_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_122_cast_fp16 = reduce_mean(axes = reduce_mean_122_axes_0, keep_dims = reduce_mean_122_keep_dims_0, x = square_40_cast_fp16)[name = tensor("reduce_mean_122_cast_fp16")]; + tensor add_80_y_0_to_fp16 = const()[name = tensor("add_80_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_80_cast_fp16 = add(x = reduce_mean_122_cast_fp16, y = add_80_y_0_to_fp16)[name = tensor("add_80_cast_fp16")]; + tensor sqrt_40_cast_fp16 = sqrt(x = add_80_cast_fp16)[name = tensor("sqrt_40_cast_fp16")]; + tensor real_div_40_cast_fp16 = real_div(x = sub_80_cast_fp16, y = sqrt_40_cast_fp16)[name = tensor("real_div_40_cast_fp16")]; + tensor reshape_161_shape_0 = const()[name = tensor("reshape_161_shape_0"), val = tensor([2, 320, 128, 128])]; + tensor reshape_161_cast_fp16 = reshape(shape = reshape_161_shape_0, x = real_div_40_cast_fp16)[name = tensor("reshape_161_cast_fp16")]; + tensor add_81_gamma_0_to_fp16 = const()[name = tensor("add_81_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5119050880)))]; + tensor add_81_beta_0_to_fp16 = const()[name = tensor("add_81_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5119051584)))]; + tensor add_81_epsilon_0_to_fp16 = const()[name = tensor("add_81_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_81_cast_fp16 = batch_norm(beta = add_81_beta_0_to_fp16, epsilon = add_81_epsilon_0_to_fp16, gamma = add_81_gamma_0_to_fp16, mean = add_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_161_cast_fp16)[name = tensor("add_81_cast_fp16")]; + tensor input_839_cast_fp16 = silu(x = add_81_cast_fp16)[name = tensor("input_839_cast_fp16")]; + tensor var_61543 = const()[name = tensor("op_61543"), val = tensor([1, 1])]; + tensor var_61545 = const()[name = tensor("op_61545"), val = tensor([1, 1])]; + tensor hidden_states_581_pad_type_0 = const()[name = tensor("hidden_states_581_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_581_pad_0 = const()[name = tensor("hidden_states_581_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor up_blocks_2_resnets_0_conv2_weight_to_fp16 = const()[name = tensor("up_blocks_2_resnets_0_conv2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5119052288)))]; + tensor up_blocks_2_resnets_0_conv2_bias_to_fp16 = const()[name = tensor("up_blocks_2_resnets_0_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5120895552)))]; + tensor hidden_states_581_cast_fp16 = conv(bias = up_blocks_2_resnets_0_conv2_bias_to_fp16, dilations = var_61545, groups = var_61502, pad = hidden_states_581_pad_0, pad_type = hidden_states_581_pad_type_0, strides = var_61543, weight = up_blocks_2_resnets_0_conv2_weight_to_fp16, x = input_839_cast_fp16)[name = tensor("hidden_states_581_cast_fp16")]; + tensor var_61550 = const()[name = tensor("op_61550"), val = tensor([1, 1])]; + tensor var_61552 = const()[name = tensor("op_61552"), val = tensor([1, 1])]; + tensor x_17_pad_type_0 = const()[name = tensor("x_17_pad_type_0"), val = tensor("custom")]; + tensor x_17_pad_0 = const()[name = tensor("x_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_2_resnets_0_conv_shortcut_weight_to_fp16 = const()[name = tensor("up_blocks_2_resnets_0_conv_shortcut_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5120896256)))]; + tensor up_blocks_2_resnets_0_conv_shortcut_bias_to_fp16 = const()[name = tensor("up_blocks_2_resnets_0_conv_shortcut_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5121510720)))]; + tensor x_17_cast_fp16 = conv(bias = up_blocks_2_resnets_0_conv_shortcut_bias_to_fp16, dilations = var_61552, groups = var_61502, pad = x_17_pad_0, pad_type = x_17_pad_type_0, strides = var_61550, weight = up_blocks_2_resnets_0_conv_shortcut_weight_to_fp16, x = input_827_cast_fp16)[name = tensor("x_17_cast_fp16")]; + tensor hidden_states_583_cast_fp16 = add(x = x_17_cast_fp16, y = hidden_states_581_cast_fp16)[name = tensor("hidden_states_583_cast_fp16")]; + tensor input_841_interleave_0 = const()[name = tensor("input_841_interleave_0"), val = tensor(false)]; + tensor input_841_cast_fp16 = concat(axis = var_61502, interleave = input_841_interleave_0, values = (hidden_states_583_cast_fp16, input_29_cast_fp16))[name = tensor("input_841_cast_fp16")]; + tensor reshape_164_shape_0 = const()[name = tensor("reshape_164_shape_0"), val = tensor([2, 32, 20, 128, 128])]; + tensor reshape_164_cast_fp16 = reshape(shape = reshape_164_shape_0, x = input_841_cast_fp16)[name = tensor("reshape_164_cast_fp16")]; + tensor reduce_mean_123_axes_0 = const()[name = tensor("reduce_mean_123_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_123_keep_dims_0 = const()[name = tensor("reduce_mean_123_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_123_cast_fp16 = reduce_mean(axes = reduce_mean_123_axes_0, keep_dims = reduce_mean_123_keep_dims_0, x = reshape_164_cast_fp16)[name = tensor("reduce_mean_123_cast_fp16")]; + tensor sub_82_cast_fp16 = sub(x = reshape_164_cast_fp16, y = reduce_mean_123_cast_fp16)[name = tensor("sub_82_cast_fp16")]; + tensor square_41_cast_fp16 = square(x = sub_82_cast_fp16)[name = tensor("square_41_cast_fp16")]; + tensor reduce_mean_125_axes_0 = const()[name = tensor("reduce_mean_125_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_125_keep_dims_0 = const()[name = tensor("reduce_mean_125_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_125_cast_fp16 = reduce_mean(axes = reduce_mean_125_axes_0, keep_dims = reduce_mean_125_keep_dims_0, x = square_41_cast_fp16)[name = tensor("reduce_mean_125_cast_fp16")]; + tensor add_82_y_0_to_fp16 = const()[name = tensor("add_82_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_82_cast_fp16 = add(x = reduce_mean_125_cast_fp16, y = add_82_y_0_to_fp16)[name = tensor("add_82_cast_fp16")]; + tensor sqrt_41_cast_fp16 = sqrt(x = add_82_cast_fp16)[name = tensor("sqrt_41_cast_fp16")]; + tensor real_div_41_cast_fp16 = real_div(x = sub_82_cast_fp16, y = sqrt_41_cast_fp16)[name = tensor("real_div_41_cast_fp16")]; + tensor reshape_165_shape_0 = const()[name = tensor("reshape_165_shape_0"), val = tensor([2, 640, 128, 128])]; + tensor reshape_165_cast_fp16 = reshape(shape = reshape_165_shape_0, x = real_div_41_cast_fp16)[name = tensor("reshape_165_cast_fp16")]; + tensor add_83_gamma_0_to_fp16 = const()[name = tensor("add_83_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5121511424)))]; + tensor add_83_beta_0_to_fp16 = const()[name = tensor("add_83_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5121512768)))]; + tensor add_83_epsilon_0_to_fp16 = const()[name = tensor("add_83_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_83_cast_fp16 = batch_norm(beta = add_83_beta_0_to_fp16, epsilon = add_83_epsilon_0_to_fp16, gamma = add_83_gamma_0_to_fp16, mean = add_11_mean_0_to_fp16, variance = add_11_variance_0_to_fp16, x = reshape_165_cast_fp16)[name = tensor("add_83_cast_fp16")]; + tensor input_845_cast_fp16 = silu(x = add_83_cast_fp16)[name = tensor("input_845_cast_fp16")]; + tensor var_61570 = const()[name = tensor("op_61570"), val = tensor([1, 1])]; + tensor var_61572 = const()[name = tensor("op_61572"), val = tensor([1, 1])]; + tensor hidden_states_585_pad_type_0 = const()[name = tensor("hidden_states_585_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_585_pad_0 = const()[name = tensor("hidden_states_585_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor up_blocks_2_resnets_1_conv1_weight_to_fp16 = const()[name = tensor("up_blocks_2_resnets_1_conv1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5121514112)))]; + tensor up_blocks_2_resnets_1_conv1_bias_to_fp16 = const()[name = tensor("up_blocks_2_resnets_1_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5125200576)))]; + tensor hidden_states_585_cast_fp16 = conv(bias = up_blocks_2_resnets_1_conv1_bias_to_fp16, dilations = var_61572, groups = var_61502, pad = hidden_states_585_pad_0, pad_type = hidden_states_585_pad_type_0, strides = var_61570, weight = up_blocks_2_resnets_1_conv1_weight_to_fp16, x = input_845_cast_fp16)[name = tensor("hidden_states_585_cast_fp16")]; + tensor var_61578 = const()[name = tensor("op_61578"), val = tensor([1, 1])]; + tensor var_61580 = const()[name = tensor("op_61580"), val = tensor([1, 1])]; + tensor temb_31_pad_type_0 = const()[name = tensor("temb_31_pad_type_0"), val = tensor("custom")]; + tensor temb_31_pad_0 = const()[name = tensor("temb_31_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_2_resnets_1_time_emb_proj_weight_to_fp16 = const()[name = tensor("up_blocks_2_resnets_1_time_emb_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5125201280)))]; + tensor up_blocks_2_resnets_1_time_emb_proj_bias_to_fp16 = const()[name = tensor("up_blocks_2_resnets_1_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5126020544)))]; + tensor temb_31_cast_fp16 = conv(bias = up_blocks_2_resnets_1_time_emb_proj_bias_to_fp16, dilations = var_61580, groups = var_61502, pad = temb_31_pad_0, pad_type = temb_31_pad_type_0, strides = var_61578, weight = up_blocks_2_resnets_1_time_emb_proj_weight_to_fp16, x = input_21_cast_fp16)[name = tensor("temb_31_cast_fp16")]; + tensor input_849_cast_fp16 = add(x = hidden_states_585_cast_fp16, y = temb_31_cast_fp16)[name = tensor("input_849_cast_fp16")]; + tensor reshape_168_shape_0 = const()[name = tensor("reshape_168_shape_0"), val = tensor([2, 32, 10, 128, 128])]; + tensor reshape_168_cast_fp16 = reshape(shape = reshape_168_shape_0, x = input_849_cast_fp16)[name = tensor("reshape_168_cast_fp16")]; + tensor reduce_mean_126_axes_0 = const()[name = tensor("reduce_mean_126_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_126_keep_dims_0 = const()[name = tensor("reduce_mean_126_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_126_cast_fp16 = reduce_mean(axes = reduce_mean_126_axes_0, keep_dims = reduce_mean_126_keep_dims_0, x = reshape_168_cast_fp16)[name = tensor("reduce_mean_126_cast_fp16")]; + tensor sub_84_cast_fp16 = sub(x = reshape_168_cast_fp16, y = reduce_mean_126_cast_fp16)[name = tensor("sub_84_cast_fp16")]; + tensor square_42_cast_fp16 = square(x = sub_84_cast_fp16)[name = tensor("square_42_cast_fp16")]; + tensor reduce_mean_128_axes_0 = const()[name = tensor("reduce_mean_128_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_128_keep_dims_0 = const()[name = tensor("reduce_mean_128_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_128_cast_fp16 = reduce_mean(axes = reduce_mean_128_axes_0, keep_dims = reduce_mean_128_keep_dims_0, x = square_42_cast_fp16)[name = tensor("reduce_mean_128_cast_fp16")]; + tensor add_84_y_0_to_fp16 = const()[name = tensor("add_84_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_84_cast_fp16 = add(x = reduce_mean_128_cast_fp16, y = add_84_y_0_to_fp16)[name = tensor("add_84_cast_fp16")]; + tensor sqrt_42_cast_fp16 = sqrt(x = add_84_cast_fp16)[name = tensor("sqrt_42_cast_fp16")]; + tensor real_div_42_cast_fp16 = real_div(x = sub_84_cast_fp16, y = sqrt_42_cast_fp16)[name = tensor("real_div_42_cast_fp16")]; + tensor reshape_169_shape_0 = const()[name = tensor("reshape_169_shape_0"), val = tensor([2, 320, 128, 128])]; + tensor reshape_169_cast_fp16 = reshape(shape = reshape_169_shape_0, x = real_div_42_cast_fp16)[name = tensor("reshape_169_cast_fp16")]; + tensor add_85_gamma_0_to_fp16 = const()[name = tensor("add_85_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5126021248)))]; + tensor add_85_beta_0_to_fp16 = const()[name = tensor("add_85_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5126021952)))]; + tensor add_85_epsilon_0_to_fp16 = const()[name = tensor("add_85_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_85_cast_fp16 = batch_norm(beta = add_85_beta_0_to_fp16, epsilon = add_85_epsilon_0_to_fp16, gamma = add_85_gamma_0_to_fp16, mean = add_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_169_cast_fp16)[name = tensor("add_85_cast_fp16")]; + tensor input_853_cast_fp16 = silu(x = add_85_cast_fp16)[name = tensor("input_853_cast_fp16")]; + tensor var_61590 = const()[name = tensor("op_61590"), val = tensor([1, 1])]; + tensor var_61592 = const()[name = tensor("op_61592"), val = tensor([1, 1])]; + tensor hidden_states_587_pad_type_0 = const()[name = tensor("hidden_states_587_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_587_pad_0 = const()[name = tensor("hidden_states_587_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor up_blocks_2_resnets_1_conv2_weight_to_fp16 = const()[name = tensor("up_blocks_2_resnets_1_conv2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5126022656)))]; + tensor up_blocks_2_resnets_1_conv2_bias_to_fp16 = const()[name = tensor("up_blocks_2_resnets_1_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5127865920)))]; + tensor hidden_states_587_cast_fp16 = conv(bias = up_blocks_2_resnets_1_conv2_bias_to_fp16, dilations = var_61592, groups = var_61502, pad = hidden_states_587_pad_0, pad_type = hidden_states_587_pad_type_0, strides = var_61590, weight = up_blocks_2_resnets_1_conv2_weight_to_fp16, x = input_853_cast_fp16)[name = tensor("hidden_states_587_cast_fp16")]; + tensor var_61597 = const()[name = tensor("op_61597"), val = tensor([1, 1])]; + tensor var_61599 = const()[name = tensor("op_61599"), val = tensor([1, 1])]; + tensor x_19_pad_type_0 = const()[name = tensor("x_19_pad_type_0"), val = tensor("custom")]; + tensor x_19_pad_0 = const()[name = tensor("x_19_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_2_resnets_1_conv_shortcut_weight_to_fp16 = const()[name = tensor("up_blocks_2_resnets_1_conv_shortcut_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5127866624)))]; + tensor up_blocks_2_resnets_1_conv_shortcut_bias_to_fp16 = const()[name = tensor("up_blocks_2_resnets_1_conv_shortcut_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5128276288)))]; + tensor x_19_cast_fp16 = conv(bias = up_blocks_2_resnets_1_conv_shortcut_bias_to_fp16, dilations = var_61599, groups = var_61502, pad = x_19_pad_0, pad_type = x_19_pad_type_0, strides = var_61597, weight = up_blocks_2_resnets_1_conv_shortcut_weight_to_fp16, x = input_841_cast_fp16)[name = tensor("x_19_cast_fp16")]; + tensor hidden_states_589_cast_fp16 = add(x = x_19_cast_fp16, y = hidden_states_587_cast_fp16)[name = tensor("hidden_states_589_cast_fp16")]; + tensor input_855_interleave_0 = const()[name = tensor("input_855_interleave_0"), val = tensor(false)]; + tensor input_855_cast_fp16 = concat(axis = var_61502, interleave = input_855_interleave_0, values = (hidden_states_589_cast_fp16, input_13_cast_fp16))[name = tensor("input_855_cast_fp16")]; + tensor reshape_172_shape_0 = const()[name = tensor("reshape_172_shape_0"), val = tensor([2, 32, 20, 128, 128])]; + tensor reshape_172_cast_fp16 = reshape(shape = reshape_172_shape_0, x = input_855_cast_fp16)[name = tensor("reshape_172_cast_fp16")]; + tensor reduce_mean_129_axes_0 = const()[name = tensor("reduce_mean_129_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_129_keep_dims_0 = const()[name = tensor("reduce_mean_129_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_129_cast_fp16 = reduce_mean(axes = reduce_mean_129_axes_0, keep_dims = reduce_mean_129_keep_dims_0, x = reshape_172_cast_fp16)[name = tensor("reduce_mean_129_cast_fp16")]; + tensor sub_86_cast_fp16 = sub(x = reshape_172_cast_fp16, y = reduce_mean_129_cast_fp16)[name = tensor("sub_86_cast_fp16")]; + tensor square_43_cast_fp16 = square(x = sub_86_cast_fp16)[name = tensor("square_43_cast_fp16")]; + tensor reduce_mean_131_axes_0 = const()[name = tensor("reduce_mean_131_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_131_keep_dims_0 = const()[name = tensor("reduce_mean_131_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_131_cast_fp16 = reduce_mean(axes = reduce_mean_131_axes_0, keep_dims = reduce_mean_131_keep_dims_0, x = square_43_cast_fp16)[name = tensor("reduce_mean_131_cast_fp16")]; + tensor add_86_y_0_to_fp16 = const()[name = tensor("add_86_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_86_cast_fp16 = add(x = reduce_mean_131_cast_fp16, y = add_86_y_0_to_fp16)[name = tensor("add_86_cast_fp16")]; + tensor sqrt_43_cast_fp16 = sqrt(x = add_86_cast_fp16)[name = tensor("sqrt_43_cast_fp16")]; + tensor real_div_43_cast_fp16 = real_div(x = sub_86_cast_fp16, y = sqrt_43_cast_fp16)[name = tensor("real_div_43_cast_fp16")]; + tensor reshape_173_shape_0 = const()[name = tensor("reshape_173_shape_0"), val = tensor([2, 640, 128, 128])]; + tensor reshape_173_cast_fp16 = reshape(shape = reshape_173_shape_0, x = real_div_43_cast_fp16)[name = tensor("reshape_173_cast_fp16")]; + tensor add_87_gamma_0_to_fp16 = const()[name = tensor("add_87_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5128276992)))]; + tensor add_87_beta_0_to_fp16 = const()[name = tensor("add_87_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5128278336)))]; + tensor add_87_epsilon_0_to_fp16 = const()[name = tensor("add_87_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_87_cast_fp16 = batch_norm(beta = add_87_beta_0_to_fp16, epsilon = add_87_epsilon_0_to_fp16, gamma = add_87_gamma_0_to_fp16, mean = add_11_mean_0_to_fp16, variance = add_11_variance_0_to_fp16, x = reshape_173_cast_fp16)[name = tensor("add_87_cast_fp16")]; + tensor input_859_cast_fp16 = silu(x = add_87_cast_fp16)[name = tensor("input_859_cast_fp16")]; + tensor var_61617 = const()[name = tensor("op_61617"), val = tensor([1, 1])]; + tensor var_61619 = const()[name = tensor("op_61619"), val = tensor([1, 1])]; + tensor hidden_states_591_pad_type_0 = const()[name = tensor("hidden_states_591_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_591_pad_0 = const()[name = tensor("hidden_states_591_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor up_blocks_2_resnets_2_conv1_weight_to_fp16 = const()[name = tensor("up_blocks_2_resnets_2_conv1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5128279680)))]; + tensor up_blocks_2_resnets_2_conv1_bias_to_fp16 = const()[name = tensor("up_blocks_2_resnets_2_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5131966144)))]; + tensor hidden_states_591_cast_fp16 = conv(bias = up_blocks_2_resnets_2_conv1_bias_to_fp16, dilations = var_61619, groups = var_61502, pad = hidden_states_591_pad_0, pad_type = hidden_states_591_pad_type_0, strides = var_61617, weight = up_blocks_2_resnets_2_conv1_weight_to_fp16, x = input_859_cast_fp16)[name = tensor("hidden_states_591_cast_fp16")]; + tensor var_61625 = const()[name = tensor("op_61625"), val = tensor([1, 1])]; + tensor var_61627 = const()[name = tensor("op_61627"), val = tensor([1, 1])]; + tensor temb_pad_type_0 = const()[name = tensor("temb_pad_type_0"), val = tensor("custom")]; + tensor temb_pad_0 = const()[name = tensor("temb_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_2_resnets_2_time_emb_proj_weight_to_fp16 = const()[name = tensor("up_blocks_2_resnets_2_time_emb_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5131966848)))]; + tensor up_blocks_2_resnets_2_time_emb_proj_bias_to_fp16 = const()[name = tensor("up_blocks_2_resnets_2_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5132786112)))]; + tensor temb_cast_fp16 = conv(bias = up_blocks_2_resnets_2_time_emb_proj_bias_to_fp16, dilations = var_61627, groups = var_61502, pad = temb_pad_0, pad_type = temb_pad_type_0, strides = var_61625, weight = up_blocks_2_resnets_2_time_emb_proj_weight_to_fp16, x = input_21_cast_fp16)[name = tensor("temb_cast_fp16")]; + tensor input_863_cast_fp16 = add(x = hidden_states_591_cast_fp16, y = temb_cast_fp16)[name = tensor("input_863_cast_fp16")]; + tensor reshape_176_shape_0 = const()[name = tensor("reshape_176_shape_0"), val = tensor([2, 32, 10, 128, 128])]; + tensor reshape_176_cast_fp16 = reshape(shape = reshape_176_shape_0, x = input_863_cast_fp16)[name = tensor("reshape_176_cast_fp16")]; + tensor reduce_mean_132_axes_0 = const()[name = tensor("reduce_mean_132_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_132_keep_dims_0 = const()[name = tensor("reduce_mean_132_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_132_cast_fp16 = reduce_mean(axes = reduce_mean_132_axes_0, keep_dims = reduce_mean_132_keep_dims_0, x = reshape_176_cast_fp16)[name = tensor("reduce_mean_132_cast_fp16")]; + tensor sub_88_cast_fp16 = sub(x = reshape_176_cast_fp16, y = reduce_mean_132_cast_fp16)[name = tensor("sub_88_cast_fp16")]; + tensor square_44_cast_fp16 = square(x = sub_88_cast_fp16)[name = tensor("square_44_cast_fp16")]; + tensor reduce_mean_134_axes_0 = const()[name = tensor("reduce_mean_134_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_134_keep_dims_0 = const()[name = tensor("reduce_mean_134_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_134_cast_fp16 = reduce_mean(axes = reduce_mean_134_axes_0, keep_dims = reduce_mean_134_keep_dims_0, x = square_44_cast_fp16)[name = tensor("reduce_mean_134_cast_fp16")]; + tensor add_88_y_0_to_fp16 = const()[name = tensor("add_88_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_88_cast_fp16 = add(x = reduce_mean_134_cast_fp16, y = add_88_y_0_to_fp16)[name = tensor("add_88_cast_fp16")]; + tensor sqrt_44_cast_fp16 = sqrt(x = add_88_cast_fp16)[name = tensor("sqrt_44_cast_fp16")]; + tensor real_div_44_cast_fp16 = real_div(x = sub_88_cast_fp16, y = sqrt_44_cast_fp16)[name = tensor("real_div_44_cast_fp16")]; + tensor reshape_177_shape_0 = const()[name = tensor("reshape_177_shape_0"), val = tensor([2, 320, 128, 128])]; + tensor reshape_177_cast_fp16 = reshape(shape = reshape_177_shape_0, x = real_div_44_cast_fp16)[name = tensor("reshape_177_cast_fp16")]; + tensor add_89_gamma_0_to_fp16 = const()[name = tensor("add_89_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5132786816)))]; + tensor add_89_beta_0_to_fp16 = const()[name = tensor("add_89_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5132787520)))]; + tensor add_89_epsilon_0_to_fp16 = const()[name = tensor("add_89_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_89_cast_fp16 = batch_norm(beta = add_89_beta_0_to_fp16, epsilon = add_89_epsilon_0_to_fp16, gamma = add_89_gamma_0_to_fp16, mean = add_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_177_cast_fp16)[name = tensor("add_89_cast_fp16")]; + tensor input_867_cast_fp16 = silu(x = add_89_cast_fp16)[name = tensor("input_867_cast_fp16")]; + tensor var_61637 = const()[name = tensor("op_61637"), val = tensor([1, 1])]; + tensor var_61639 = const()[name = tensor("op_61639"), val = tensor([1, 1])]; + tensor hidden_states_pad_type_0 = const()[name = tensor("hidden_states_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_pad_0 = const()[name = tensor("hidden_states_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor up_blocks_2_resnets_2_conv2_weight_to_fp16 = const()[name = tensor("up_blocks_2_resnets_2_conv2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5132788224)))]; + tensor up_blocks_2_resnets_2_conv2_bias_to_fp16 = const()[name = tensor("up_blocks_2_resnets_2_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5134631488)))]; + tensor hidden_states_cast_fp16 = conv(bias = up_blocks_2_resnets_2_conv2_bias_to_fp16, dilations = var_61639, groups = var_61502, pad = hidden_states_pad_0, pad_type = hidden_states_pad_type_0, strides = var_61637, weight = up_blocks_2_resnets_2_conv2_weight_to_fp16, x = input_867_cast_fp16)[name = tensor("hidden_states_cast_fp16")]; + tensor var_61644 = const()[name = tensor("op_61644"), val = tensor([1, 1])]; + tensor var_61646 = const()[name = tensor("op_61646"), val = tensor([1, 1])]; + tensor x_pad_type_0 = const()[name = tensor("x_pad_type_0"), val = tensor("custom")]; + tensor x_pad_0 = const()[name = tensor("x_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_2_resnets_2_conv_shortcut_weight_to_fp16 = const()[name = tensor("up_blocks_2_resnets_2_conv_shortcut_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5134632192)))]; + tensor up_blocks_2_resnets_2_conv_shortcut_bias_to_fp16 = const()[name = tensor("up_blocks_2_resnets_2_conv_shortcut_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5135041856)))]; + tensor x_cast_fp16 = conv(bias = up_blocks_2_resnets_2_conv_shortcut_bias_to_fp16, dilations = var_61646, groups = var_61502, pad = x_pad_0, pad_type = x_pad_type_0, strides = var_61644, weight = up_blocks_2_resnets_2_conv_shortcut_weight_to_fp16, x = input_855_cast_fp16)[name = tensor("x_cast_fp16")]; + tensor input_869_cast_fp16 = add(x = x_cast_fp16, y = hidden_states_cast_fp16)[name = tensor("input_869_cast_fp16")]; + tensor reshape_180_shape_0 = const()[name = tensor("reshape_180_shape_0"), val = tensor([2, 32, 10, 128, 128])]; + tensor reshape_180_cast_fp16 = reshape(shape = reshape_180_shape_0, x = input_869_cast_fp16)[name = tensor("reshape_180_cast_fp16")]; + tensor reduce_mean_135_axes_0 = const()[name = tensor("reduce_mean_135_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_135_keep_dims_0 = const()[name = tensor("reduce_mean_135_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_135_cast_fp16 = reduce_mean(axes = reduce_mean_135_axes_0, keep_dims = reduce_mean_135_keep_dims_0, x = reshape_180_cast_fp16)[name = tensor("reduce_mean_135_cast_fp16")]; + tensor sub_90_cast_fp16 = sub(x = reshape_180_cast_fp16, y = reduce_mean_135_cast_fp16)[name = tensor("sub_90_cast_fp16")]; + tensor square_45_cast_fp16 = square(x = sub_90_cast_fp16)[name = tensor("square_45_cast_fp16")]; + tensor reduce_mean_137_axes_0 = const()[name = tensor("reduce_mean_137_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_137_keep_dims_0 = const()[name = tensor("reduce_mean_137_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_137_cast_fp16 = reduce_mean(axes = reduce_mean_137_axes_0, keep_dims = reduce_mean_137_keep_dims_0, x = square_45_cast_fp16)[name = tensor("reduce_mean_137_cast_fp16")]; + tensor add_90_y_0_to_fp16 = const()[name = tensor("add_90_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_90_cast_fp16 = add(x = reduce_mean_137_cast_fp16, y = add_90_y_0_to_fp16)[name = tensor("add_90_cast_fp16")]; + tensor sqrt_45_cast_fp16 = sqrt(x = add_90_cast_fp16)[name = tensor("sqrt_45_cast_fp16")]; + tensor real_div_45_cast_fp16 = real_div(x = sub_90_cast_fp16, y = sqrt_45_cast_fp16)[name = tensor("real_div_45_cast_fp16")]; + tensor reshape_181_shape_0 = const()[name = tensor("reshape_181_shape_0"), val = tensor([2, 320, 128, 128])]; + tensor reshape_181_cast_fp16 = reshape(shape = reshape_181_shape_0, x = real_div_45_cast_fp16)[name = tensor("reshape_181_cast_fp16")]; + tensor add_91_gamma_0_to_fp16 = const()[name = tensor("add_91_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5135042560)))]; + tensor add_91_beta_0_to_fp16 = const()[name = tensor("add_91_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5135043264)))]; + tensor add_91_epsilon_0_to_fp16 = const()[name = tensor("add_91_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_91_cast_fp16 = batch_norm(beta = add_91_beta_0_to_fp16, epsilon = add_91_epsilon_0_to_fp16, gamma = add_91_gamma_0_to_fp16, mean = add_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_181_cast_fp16)[name = tensor("add_91_cast_fp16")]; + tensor input_cast_fp16 = silu(x = add_91_cast_fp16)[name = tensor("input_cast_fp16")]; + tensor var_61660 = const()[name = tensor("op_61660"), val = tensor(1)]; + tensor var_61663 = const()[name = tensor("op_61663"), val = tensor([1, 1])]; + tensor var_61665 = const()[name = tensor("op_61665"), val = tensor([1, 1])]; + tensor var_61667_pad_type_0 = const()[name = tensor("op_61667_pad_type_0"), val = tensor("custom")]; + tensor var_61667_pad_0 = const()[name = tensor("op_61667_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor conv_out_weight_to_fp16 = const()[name = tensor("conv_out_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5135043968)))]; + tensor conv_out_bias_to_fp16 = const()[name = tensor("conv_out_bias_to_fp16"), val = tensor([0x1.96p-8, -0x1.74p-10, 0x1.14p-9, -0x1.8ap-9])]; + tensor var_61667_cast_fp16 = conv(bias = conv_out_bias_to_fp16, dilations = var_61665, groups = var_61660, pad = var_61667_pad_0, pad_type = var_61667_pad_type_0, strides = var_61663, weight = conv_out_weight_to_fp16, x = input_cast_fp16)[name = tensor("op_61667_cast_fp16")]; + tensor var_61667_cast_fp16_to_fp32_dtype_0 = const()[name = tensor("op_61667_cast_fp16_to_fp32_dtype_0"), val = tensor("fp32")]; + tensor noise_pred = cast(dtype = var_61667_cast_fp16_to_fp32_dtype_0, x = var_61667_cast_fp16)[name = tensor("cast_453")]; + } -> (noise_pred); +} \ No newline at end of file